2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
104 #include <linux/if_bridge.h>
105 #include <linux/if_macvlan.h>
107 #include <net/pkt_sched.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/kmod.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/wext.h>
118 #include <net/iw_handler.h>
119 #include <asm/current.h>
120 #include <linux/audit.h>
121 #include <linux/dmaengine.h>
122 #include <linux/err.h>
123 #include <linux/ctype.h>
124 #include <linux/if_arp.h>
125 #include <linux/if_vlan.h>
126 #include <linux/ip.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
134 #include "net-sysfs.h"
136 /* Instead of increasing this, you should create a hash table. */
137 #define MAX_GRO_SKBS 8
139 /* This should be increased if a protocol with a bigger head is added. */
140 #define GRO_MAX_HEAD (MAX_HEADER + 128)
143 * The list of packet types we will receive (as opposed to discard)
144 * and the routines to invoke.
146 * Why 16. Because with 16 the only overlap we get on a hash of the
147 * low nibble of the protocol value is RARP/SNAP/X.25.
149 * NOTE: That is no longer true with the addition of VLAN tags. Not
150 * sure which should go first, but I bet it won't make much
151 * difference if we are running VLANs. The good news is that
152 * this protocol won't be in the list unless compiled in, so
153 * the average user (w/out VLANs) will not be adversely affected.
170 #define PTYPE_HASH_SIZE (16)
171 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
173 static DEFINE_SPINLOCK(ptype_lock);
174 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
175 static struct list_head ptype_all __read_mostly; /* Taps */
178 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
181 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
183 * Writers must hold the rtnl semaphore while they loop through the
184 * dev_base_head list, and hold dev_base_lock for writing when they do the
185 * actual updates. This allows pure readers to access the list even
186 * while a writer is preparing to update it.
188 * To put it another way, dev_base_lock is held for writing only to
189 * protect against pure readers; the rtnl semaphore provides the
190 * protection against other writers.
192 * See, for example usages, register_netdevice() and
193 * unregister_netdevice(), which must be called with the rtnl
196 DEFINE_RWLOCK(dev_base_lock);
197 EXPORT_SYMBOL(dev_base_lock);
199 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
201 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
202 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
210 /* Device list insertion */
211 static int list_netdevice(struct net_device *dev)
213 struct net *net = dev_net(dev);
217 write_lock_bh(&dev_base_lock);
218 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
219 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
220 hlist_add_head_rcu(&dev->index_hlist,
221 dev_index_hash(net, dev->ifindex));
222 write_unlock_bh(&dev_base_lock);
226 /* Device list removal
227 * caller must respect a RCU grace period before freeing/reusing dev
229 static void unlist_netdevice(struct net_device *dev)
233 /* Unlink dev from the device chain */
234 write_lock_bh(&dev_base_lock);
235 list_del_rcu(&dev->dev_list);
236 hlist_del_rcu(&dev->name_hlist);
237 hlist_del_rcu(&dev->index_hlist);
238 write_unlock_bh(&dev_base_lock);
245 static RAW_NOTIFIER_HEAD(netdev_chain);
248 * Device drivers call our routines to queue packets here. We empty the
249 * queue in the local softnet handler.
252 DEFINE_PER_CPU(struct softnet_data, softnet_data);
253 EXPORT_PER_CPU_SYMBOL(softnet_data);
255 #ifdef CONFIG_LOCKDEP
257 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
258 * according to dev->type
260 static const unsigned short netdev_lock_type[] =
261 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
262 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
263 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
264 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
265 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
266 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
267 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
268 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
269 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
270 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
271 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
272 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
273 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
274 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
275 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
276 ARPHRD_VOID, ARPHRD_NONE};
278 static const char *const netdev_lock_name[] =
279 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
280 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
281 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
282 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
283 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
284 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
285 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
286 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
287 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
288 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
289 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
290 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
291 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
292 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
293 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
294 "_xmit_VOID", "_xmit_NONE"};
296 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
297 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
299 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
303 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
304 if (netdev_lock_type[i] == dev_type)
306 /* the last key is used by default */
307 return ARRAY_SIZE(netdev_lock_type) - 1;
310 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
311 unsigned short dev_type)
315 i = netdev_lock_pos(dev_type);
316 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
317 netdev_lock_name[i]);
320 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
324 i = netdev_lock_pos(dev->type);
325 lockdep_set_class_and_name(&dev->addr_list_lock,
326 &netdev_addr_lock_key[i],
327 netdev_lock_name[i]);
330 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
331 unsigned short dev_type)
334 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 /*******************************************************************************
341 Protocol management and registration routines
343 *******************************************************************************/
346 * Add a protocol ID to the list. Now that the input handler is
347 * smarter we can dispense with all the messy stuff that used to be
350 * BEWARE!!! Protocol handlers, mangling input packets,
351 * MUST BE last in hash buckets and checking protocol handlers
352 * MUST start from promiscuous ptype_all chain in net_bh.
353 * It is true now, do not change it.
354 * Explanation follows: if protocol handler, mangling packet, will
355 * be the first on list, it is not able to sense, that packet
356 * is cloned and should be copied-on-write, so that it will
357 * change it and subsequent readers will get broken packet.
362 * dev_add_pack - add packet handler
363 * @pt: packet type declaration
365 * Add a protocol handler to the networking stack. The passed &packet_type
366 * is linked into kernel lists and may not be freed until it has been
367 * removed from the kernel lists.
369 * This call does not sleep therefore it can not
370 * guarantee all CPU's that are in middle of receiving packets
371 * will see the new packet type (until the next received packet).
374 void dev_add_pack(struct packet_type *pt)
378 spin_lock_bh(&ptype_lock);
379 if (pt->type == htons(ETH_P_ALL))
380 list_add_rcu(&pt->list, &ptype_all);
382 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
383 list_add_rcu(&pt->list, &ptype_base[hash]);
385 spin_unlock_bh(&ptype_lock);
387 EXPORT_SYMBOL(dev_add_pack);
390 * __dev_remove_pack - remove packet handler
391 * @pt: packet type declaration
393 * Remove a protocol handler that was previously added to the kernel
394 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
395 * from the kernel lists and can be freed or reused once this function
398 * The packet type might still be in use by receivers
399 * and must not be freed until after all the CPU's have gone
400 * through a quiescent state.
402 void __dev_remove_pack(struct packet_type *pt)
404 struct list_head *head;
405 struct packet_type *pt1;
407 spin_lock_bh(&ptype_lock);
409 if (pt->type == htons(ETH_P_ALL))
412 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
414 list_for_each_entry(pt1, head, list) {
416 list_del_rcu(&pt->list);
421 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
423 spin_unlock_bh(&ptype_lock);
425 EXPORT_SYMBOL(__dev_remove_pack);
428 * dev_remove_pack - remove packet handler
429 * @pt: packet type declaration
431 * Remove a protocol handler that was previously added to the kernel
432 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
433 * from the kernel lists and can be freed or reused once this function
436 * This call sleeps to guarantee that no CPU is looking at the packet
439 void dev_remove_pack(struct packet_type *pt)
441 __dev_remove_pack(pt);
445 EXPORT_SYMBOL(dev_remove_pack);
447 /******************************************************************************
449 Device Boot-time Settings Routines
451 *******************************************************************************/
453 /* Boot time configuration table */
454 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
457 * netdev_boot_setup_add - add new setup entry
458 * @name: name of the device
459 * @map: configured settings for the device
461 * Adds new setup entry to the dev_boot_setup list. The function
462 * returns 0 on error and 1 on success. This is a generic routine to
465 static int netdev_boot_setup_add(char *name, struct ifmap *map)
467 struct netdev_boot_setup *s;
471 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
472 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
473 memset(s[i].name, 0, sizeof(s[i].name));
474 strlcpy(s[i].name, name, IFNAMSIZ);
475 memcpy(&s[i].map, map, sizeof(s[i].map));
480 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
484 * netdev_boot_setup_check - check boot time settings
485 * @dev: the netdevice
487 * Check boot time settings for the device.
488 * The found settings are set for the device to be used
489 * later in the device probing.
490 * Returns 0 if no settings found, 1 if they are.
492 int netdev_boot_setup_check(struct net_device *dev)
494 struct netdev_boot_setup *s = dev_boot_setup;
497 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
498 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
499 !strcmp(dev->name, s[i].name)) {
500 dev->irq = s[i].map.irq;
501 dev->base_addr = s[i].map.base_addr;
502 dev->mem_start = s[i].map.mem_start;
503 dev->mem_end = s[i].map.mem_end;
509 EXPORT_SYMBOL(netdev_boot_setup_check);
513 * netdev_boot_base - get address from boot time settings
514 * @prefix: prefix for network device
515 * @unit: id for network device
517 * Check boot time settings for the base address of device.
518 * The found settings are set for the device to be used
519 * later in the device probing.
520 * Returns 0 if no settings found.
522 unsigned long netdev_boot_base(const char *prefix, int unit)
524 const struct netdev_boot_setup *s = dev_boot_setup;
528 sprintf(name, "%s%d", prefix, unit);
531 * If device already registered then return base of 1
532 * to indicate not to probe for this interface
534 if (__dev_get_by_name(&init_net, name))
537 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
538 if (!strcmp(name, s[i].name))
539 return s[i].map.base_addr;
544 * Saves at boot time configured settings for any netdevice.
546 int __init netdev_boot_setup(char *str)
551 str = get_options(str, ARRAY_SIZE(ints), ints);
556 memset(&map, 0, sizeof(map));
560 map.base_addr = ints[2];
562 map.mem_start = ints[3];
564 map.mem_end = ints[4];
566 /* Add new entry to the list */
567 return netdev_boot_setup_add(str, &map);
570 __setup("netdev=", netdev_boot_setup);
572 /*******************************************************************************
574 Device Interface Subroutines
576 *******************************************************************************/
579 * __dev_get_by_name - find a device by its name
580 * @net: the applicable net namespace
581 * @name: name to find
583 * Find an interface by name. Must be called under RTNL semaphore
584 * or @dev_base_lock. If the name is found a pointer to the device
585 * is returned. If the name is not found then %NULL is returned. The
586 * reference counters are not incremented so the caller must be
587 * careful with locks.
590 struct net_device *__dev_get_by_name(struct net *net, const char *name)
592 struct hlist_node *p;
593 struct net_device *dev;
594 struct hlist_head *head = dev_name_hash(net, name);
596 hlist_for_each_entry(dev, p, head, name_hlist)
597 if (!strncmp(dev->name, name, IFNAMSIZ))
602 EXPORT_SYMBOL(__dev_get_by_name);
605 * dev_get_by_name_rcu - find a device by its name
606 * @net: the applicable net namespace
607 * @name: name to find
609 * Find an interface by name.
610 * If the name is found a pointer to the device is returned.
611 * If the name is not found then %NULL is returned.
612 * The reference counters are not incremented so the caller must be
613 * careful with locks. The caller must hold RCU lock.
616 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
618 struct hlist_node *p;
619 struct net_device *dev;
620 struct hlist_head *head = dev_name_hash(net, name);
622 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
623 if (!strncmp(dev->name, name, IFNAMSIZ))
628 EXPORT_SYMBOL(dev_get_by_name_rcu);
631 * dev_get_by_name - find a device by its name
632 * @net: the applicable net namespace
633 * @name: name to find
635 * Find an interface by name. This can be called from any
636 * context and does its own locking. The returned handle has
637 * the usage count incremented and the caller must use dev_put() to
638 * release it when it is no longer needed. %NULL is returned if no
639 * matching device is found.
642 struct net_device *dev_get_by_name(struct net *net, const char *name)
644 struct net_device *dev;
647 dev = dev_get_by_name_rcu(net, name);
653 EXPORT_SYMBOL(dev_get_by_name);
656 * __dev_get_by_index - find a device by its ifindex
657 * @net: the applicable net namespace
658 * @ifindex: index of device
660 * Search for an interface by index. Returns %NULL if the device
661 * is not found or a pointer to the device. The device has not
662 * had its reference counter increased so the caller must be careful
663 * about locking. The caller must hold either the RTNL semaphore
667 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
669 struct hlist_node *p;
670 struct net_device *dev;
671 struct hlist_head *head = dev_index_hash(net, ifindex);
673 hlist_for_each_entry(dev, p, head, index_hlist)
674 if (dev->ifindex == ifindex)
679 EXPORT_SYMBOL(__dev_get_by_index);
682 * dev_get_by_index_rcu - find a device by its ifindex
683 * @net: the applicable net namespace
684 * @ifindex: index of device
686 * Search for an interface by index. Returns %NULL if the device
687 * is not found or a pointer to the device. The device has not
688 * had its reference counter increased so the caller must be careful
689 * about locking. The caller must hold RCU lock.
692 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
694 struct hlist_node *p;
695 struct net_device *dev;
696 struct hlist_head *head = dev_index_hash(net, ifindex);
698 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
699 if (dev->ifindex == ifindex)
704 EXPORT_SYMBOL(dev_get_by_index_rcu);
708 * dev_get_by_index - find a device by its ifindex
709 * @net: the applicable net namespace
710 * @ifindex: index of device
712 * Search for an interface by index. Returns NULL if the device
713 * is not found or a pointer to the device. The device returned has
714 * had a reference added and the pointer is safe until the user calls
715 * dev_put to indicate they have finished with it.
718 struct net_device *dev_get_by_index(struct net *net, int ifindex)
720 struct net_device *dev;
723 dev = dev_get_by_index_rcu(net, ifindex);
729 EXPORT_SYMBOL(dev_get_by_index);
732 * dev_getbyhwaddr - find a device by its hardware address
733 * @net: the applicable net namespace
734 * @type: media type of device
735 * @ha: hardware address
737 * Search for an interface by MAC address. Returns NULL if the device
738 * is not found or a pointer to the device. The caller must hold the
739 * rtnl semaphore. The returned device has not had its ref count increased
740 * and the caller must therefore be careful about locking
743 * If the API was consistent this would be __dev_get_by_hwaddr
746 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
748 struct net_device *dev;
752 for_each_netdev(net, dev)
753 if (dev->type == type &&
754 !memcmp(dev->dev_addr, ha, dev->addr_len))
759 EXPORT_SYMBOL(dev_getbyhwaddr);
761 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
763 struct net_device *dev;
766 for_each_netdev(net, dev)
767 if (dev->type == type)
772 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
774 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
776 struct net_device *dev;
779 dev = __dev_getfirstbyhwtype(net, type);
785 EXPORT_SYMBOL(dev_getfirstbyhwtype);
788 * dev_get_by_flags - find any device with given flags
789 * @net: the applicable net namespace
790 * @if_flags: IFF_* values
791 * @mask: bitmask of bits in if_flags to check
793 * Search for any interface with the given flags. Returns NULL if a device
794 * is not found or a pointer to the device. The device returned has
795 * had a reference added and the pointer is safe until the user calls
796 * dev_put to indicate they have finished with it.
799 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
802 struct net_device *dev, *ret;
806 for_each_netdev_rcu(net, dev) {
807 if (((dev->flags ^ if_flags) & mask) == 0) {
816 EXPORT_SYMBOL(dev_get_by_flags);
819 * dev_valid_name - check if name is okay for network device
822 * Network device names need to be valid file names to
823 * to allow sysfs to work. We also disallow any kind of
826 int dev_valid_name(const char *name)
830 if (strlen(name) >= IFNAMSIZ)
832 if (!strcmp(name, ".") || !strcmp(name, ".."))
836 if (*name == '/' || isspace(*name))
842 EXPORT_SYMBOL(dev_valid_name);
845 * __dev_alloc_name - allocate a name for a device
846 * @net: network namespace to allocate the device name in
847 * @name: name format string
848 * @buf: scratch buffer and result name string
850 * Passed a format string - eg "lt%d" it will try and find a suitable
851 * id. It scans list of devices to build up a free map, then chooses
852 * the first empty slot. The caller must hold the dev_base or rtnl lock
853 * while allocating the name and adding the device in order to avoid
855 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
856 * Returns the number of the unit assigned or a negative errno code.
859 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
863 const int max_netdevices = 8*PAGE_SIZE;
864 unsigned long *inuse;
865 struct net_device *d;
867 p = strnchr(name, IFNAMSIZ-1, '%');
870 * Verify the string as this thing may have come from
871 * the user. There must be either one "%d" and no other "%"
874 if (p[1] != 'd' || strchr(p + 2, '%'))
877 /* Use one page as a bit array of possible slots */
878 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
882 for_each_netdev(net, d) {
883 if (!sscanf(d->name, name, &i))
885 if (i < 0 || i >= max_netdevices)
888 /* avoid cases where sscanf is not exact inverse of printf */
889 snprintf(buf, IFNAMSIZ, name, i);
890 if (!strncmp(buf, d->name, IFNAMSIZ))
894 i = find_first_zero_bit(inuse, max_netdevices);
895 free_page((unsigned long) inuse);
899 snprintf(buf, IFNAMSIZ, name, i);
900 if (!__dev_get_by_name(net, buf))
903 /* It is possible to run out of possible slots
904 * when the name is long and there isn't enough space left
905 * for the digits, or if all bits are used.
911 * dev_alloc_name - allocate a name for a device
913 * @name: name format string
915 * Passed a format string - eg "lt%d" it will try and find a suitable
916 * id. It scans list of devices to build up a free map, then chooses
917 * the first empty slot. The caller must hold the dev_base or rtnl lock
918 * while allocating the name and adding the device in order to avoid
920 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
921 * Returns the number of the unit assigned or a negative errno code.
924 int dev_alloc_name(struct net_device *dev, const char *name)
930 BUG_ON(!dev_net(dev));
932 ret = __dev_alloc_name(net, name, buf);
934 strlcpy(dev->name, buf, IFNAMSIZ);
937 EXPORT_SYMBOL(dev_alloc_name);
939 static int dev_get_valid_name(struct net *net, const char *name, char *buf,
942 if (!dev_valid_name(name))
945 if (fmt && strchr(name, '%'))
946 return __dev_alloc_name(net, name, buf);
947 else if (__dev_get_by_name(net, name))
949 else if (buf != name)
950 strlcpy(buf, name, IFNAMSIZ);
956 * dev_change_name - change name of a device
958 * @newname: name (or format string) must be at least IFNAMSIZ
960 * Change name of a device, can pass format strings "eth%d".
963 int dev_change_name(struct net_device *dev, const char *newname)
965 char oldname[IFNAMSIZ];
971 BUG_ON(!dev_net(dev));
974 if (dev->flags & IFF_UP)
977 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
980 memcpy(oldname, dev->name, IFNAMSIZ);
982 err = dev_get_valid_name(net, newname, dev->name, 1);
987 /* For now only devices in the initial network namespace
990 if (net_eq(net, &init_net)) {
991 ret = device_rename(&dev->dev, dev->name);
993 memcpy(dev->name, oldname, IFNAMSIZ);
998 write_lock_bh(&dev_base_lock);
999 hlist_del(&dev->name_hlist);
1000 write_unlock_bh(&dev_base_lock);
1004 write_lock_bh(&dev_base_lock);
1005 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1006 write_unlock_bh(&dev_base_lock);
1008 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1009 ret = notifier_to_errno(ret);
1012 /* err >= 0 after dev_alloc_name() or stores the first errno */
1015 memcpy(dev->name, oldname, IFNAMSIZ);
1019 "%s: name change rollback failed: %d.\n",
1028 * dev_set_alias - change ifalias of a device
1030 * @alias: name up to IFALIASZ
1031 * @len: limit of bytes to copy from info
1033 * Set ifalias for a device,
1035 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1039 if (len >= IFALIASZ)
1044 kfree(dev->ifalias);
1045 dev->ifalias = NULL;
1050 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1054 strlcpy(dev->ifalias, alias, len+1);
1060 * netdev_features_change - device changes features
1061 * @dev: device to cause notification
1063 * Called to indicate a device has changed features.
1065 void netdev_features_change(struct net_device *dev)
1067 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1069 EXPORT_SYMBOL(netdev_features_change);
1072 * netdev_state_change - device changes state
1073 * @dev: device to cause notification
1075 * Called to indicate a device has changed state. This function calls
1076 * the notifier chains for netdev_chain and sends a NEWLINK message
1077 * to the routing socket.
1079 void netdev_state_change(struct net_device *dev)
1081 if (dev->flags & IFF_UP) {
1082 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1083 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1086 EXPORT_SYMBOL(netdev_state_change);
1088 void netdev_bonding_change(struct net_device *dev, unsigned long event)
1090 call_netdevice_notifiers(event, dev);
1092 EXPORT_SYMBOL(netdev_bonding_change);
1095 * dev_load - load a network module
1096 * @net: the applicable net namespace
1097 * @name: name of interface
1099 * If a network interface is not present and the process has suitable
1100 * privileges this function loads the module. If module loading is not
1101 * available in this kernel then it becomes a nop.
1104 void dev_load(struct net *net, const char *name)
1106 struct net_device *dev;
1109 dev = dev_get_by_name_rcu(net, name);
1112 if (!dev && capable(CAP_NET_ADMIN))
1113 request_module("%s", name);
1115 EXPORT_SYMBOL(dev_load);
1117 static int __dev_open(struct net_device *dev)
1119 const struct net_device_ops *ops = dev->netdev_ops;
1125 * Is it even present?
1127 if (!netif_device_present(dev))
1130 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1131 ret = notifier_to_errno(ret);
1136 * Call device private open method
1138 set_bit(__LINK_STATE_START, &dev->state);
1140 if (ops->ndo_validate_addr)
1141 ret = ops->ndo_validate_addr(dev);
1143 if (!ret && ops->ndo_open)
1144 ret = ops->ndo_open(dev);
1147 * If it went open OK then:
1151 clear_bit(__LINK_STATE_START, &dev->state);
1156 dev->flags |= IFF_UP;
1161 net_dmaengine_get();
1164 * Initialize multicasting status
1166 dev_set_rx_mode(dev);
1169 * Wakeup transmit queue engine
1178 * dev_open - prepare an interface for use.
1179 * @dev: device to open
1181 * Takes a device from down to up state. The device's private open
1182 * function is invoked and then the multicast lists are loaded. Finally
1183 * the device is moved into the up state and a %NETDEV_UP message is
1184 * sent to the netdev notifier chain.
1186 * Calling this function on an active interface is a nop. On a failure
1187 * a negative errno code is returned.
1189 int dev_open(struct net_device *dev)
1196 if (dev->flags & IFF_UP)
1202 ret = __dev_open(dev);
1207 * ... and announce new interface.
1209 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1210 call_netdevice_notifiers(NETDEV_UP, dev);
1214 EXPORT_SYMBOL(dev_open);
1216 static int __dev_close(struct net_device *dev)
1218 const struct net_device_ops *ops = dev->netdev_ops;
1224 * Tell people we are going down, so that they can
1225 * prepare to death, when device is still operating.
1227 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1229 clear_bit(__LINK_STATE_START, &dev->state);
1231 /* Synchronize to scheduled poll. We cannot touch poll list,
1232 * it can be even on different cpu. So just clear netif_running().
1234 * dev->stop() will invoke napi_disable() on all of it's
1235 * napi_struct instances on this device.
1237 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1239 dev_deactivate(dev);
1242 * Call the device specific close. This cannot fail.
1243 * Only if device is UP
1245 * We allow it to be called even after a DETACH hot-plug
1252 * Device is now down.
1255 dev->flags &= ~IFF_UP;
1260 net_dmaengine_put();
1266 * dev_close - shutdown an interface.
1267 * @dev: device to shutdown
1269 * This function moves an active device into down state. A
1270 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1271 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1274 int dev_close(struct net_device *dev)
1276 if (!(dev->flags & IFF_UP))
1282 * Tell people we are down
1284 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1285 call_netdevice_notifiers(NETDEV_DOWN, dev);
1289 EXPORT_SYMBOL(dev_close);
1293 * dev_disable_lro - disable Large Receive Offload on a device
1296 * Disable Large Receive Offload (LRO) on a net device. Must be
1297 * called under RTNL. This is needed if received packets may be
1298 * forwarded to another interface.
1300 void dev_disable_lro(struct net_device *dev)
1302 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1303 dev->ethtool_ops->set_flags) {
1304 u32 flags = dev->ethtool_ops->get_flags(dev);
1305 if (flags & ETH_FLAG_LRO) {
1306 flags &= ~ETH_FLAG_LRO;
1307 dev->ethtool_ops->set_flags(dev, flags);
1310 WARN_ON(dev->features & NETIF_F_LRO);
1312 EXPORT_SYMBOL(dev_disable_lro);
1315 static int dev_boot_phase = 1;
1318 * Device change register/unregister. These are not inline or static
1319 * as we export them to the world.
1323 * register_netdevice_notifier - register a network notifier block
1326 * Register a notifier to be called when network device events occur.
1327 * The notifier passed is linked into the kernel structures and must
1328 * not be reused until it has been unregistered. A negative errno code
1329 * is returned on a failure.
1331 * When registered all registration and up events are replayed
1332 * to the new notifier to allow device to have a race free
1333 * view of the network device list.
1336 int register_netdevice_notifier(struct notifier_block *nb)
1338 struct net_device *dev;
1339 struct net_device *last;
1344 err = raw_notifier_chain_register(&netdev_chain, nb);
1350 for_each_netdev(net, dev) {
1351 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1352 err = notifier_to_errno(err);
1356 if (!(dev->flags & IFF_UP))
1359 nb->notifier_call(nb, NETDEV_UP, dev);
1370 for_each_netdev(net, dev) {
1374 if (dev->flags & IFF_UP) {
1375 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1376 nb->notifier_call(nb, NETDEV_DOWN, dev);
1378 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1379 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1383 raw_notifier_chain_unregister(&netdev_chain, nb);
1386 EXPORT_SYMBOL(register_netdevice_notifier);
1389 * unregister_netdevice_notifier - unregister a network notifier block
1392 * Unregister a notifier previously registered by
1393 * register_netdevice_notifier(). The notifier is unlinked into the
1394 * kernel structures and may then be reused. A negative errno code
1395 * is returned on a failure.
1398 int unregister_netdevice_notifier(struct notifier_block *nb)
1403 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1407 EXPORT_SYMBOL(unregister_netdevice_notifier);
1410 * call_netdevice_notifiers - call all network notifier blocks
1411 * @val: value passed unmodified to notifier function
1412 * @dev: net_device pointer passed unmodified to notifier function
1414 * Call all network notifier blocks. Parameters and return value
1415 * are as for raw_notifier_call_chain().
1418 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1420 return raw_notifier_call_chain(&netdev_chain, val, dev);
1423 /* When > 0 there are consumers of rx skb time stamps */
1424 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1426 void net_enable_timestamp(void)
1428 atomic_inc(&netstamp_needed);
1430 EXPORT_SYMBOL(net_enable_timestamp);
1432 void net_disable_timestamp(void)
1434 atomic_dec(&netstamp_needed);
1436 EXPORT_SYMBOL(net_disable_timestamp);
1438 static inline void net_timestamp(struct sk_buff *skb)
1440 if (atomic_read(&netstamp_needed))
1441 __net_timestamp(skb);
1443 skb->tstamp.tv64 = 0;
1447 * dev_forward_skb - loopback an skb to another netif
1449 * @dev: destination network device
1450 * @skb: buffer to forward
1453 * NET_RX_SUCCESS (no congestion)
1454 * NET_RX_DROP (packet was dropped)
1456 * dev_forward_skb can be used for injecting an skb from the
1457 * start_xmit function of one device into the receive queue
1458 * of another device.
1460 * The receiving device may be in another namespace, so
1461 * we have to clear all information in the skb that could
1462 * impact namespace isolation.
1464 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1468 if (!(dev->flags & IFF_UP))
1471 if (skb->len > (dev->mtu + dev->hard_header_len))
1474 skb_set_dev(skb, dev);
1475 skb->tstamp.tv64 = 0;
1476 skb->pkt_type = PACKET_HOST;
1477 skb->protocol = eth_type_trans(skb, dev);
1478 return netif_rx(skb);
1480 EXPORT_SYMBOL_GPL(dev_forward_skb);
1483 * Support routine. Sends outgoing frames to any network
1484 * taps currently in use.
1487 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1489 struct packet_type *ptype;
1491 #ifdef CONFIG_NET_CLS_ACT
1492 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1499 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1500 /* Never send packets back to the socket
1501 * they originated from - MvS (miquels@drinkel.ow.org)
1503 if ((ptype->dev == dev || !ptype->dev) &&
1504 (ptype->af_packet_priv == NULL ||
1505 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1506 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1510 /* skb->nh should be correctly
1511 set by sender, so that the second statement is
1512 just protection against buggy protocols.
1514 skb_reset_mac_header(skb2);
1516 if (skb_network_header(skb2) < skb2->data ||
1517 skb2->network_header > skb2->tail) {
1518 if (net_ratelimit())
1519 printk(KERN_CRIT "protocol %04x is "
1521 skb2->protocol, dev->name);
1522 skb_reset_network_header(skb2);
1525 skb2->transport_header = skb2->network_header;
1526 skb2->pkt_type = PACKET_OUTGOING;
1527 ptype->func(skb2, skb->dev, ptype, skb->dev);
1534 static inline void __netif_reschedule(struct Qdisc *q)
1536 struct softnet_data *sd;
1537 unsigned long flags;
1539 local_irq_save(flags);
1540 sd = &__get_cpu_var(softnet_data);
1541 q->next_sched = sd->output_queue;
1542 sd->output_queue = q;
1543 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1544 local_irq_restore(flags);
1547 void __netif_schedule(struct Qdisc *q)
1549 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1550 __netif_reschedule(q);
1552 EXPORT_SYMBOL(__netif_schedule);
1554 void dev_kfree_skb_irq(struct sk_buff *skb)
1556 if (atomic_dec_and_test(&skb->users)) {
1557 struct softnet_data *sd;
1558 unsigned long flags;
1560 local_irq_save(flags);
1561 sd = &__get_cpu_var(softnet_data);
1562 skb->next = sd->completion_queue;
1563 sd->completion_queue = skb;
1564 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1565 local_irq_restore(flags);
1568 EXPORT_SYMBOL(dev_kfree_skb_irq);
1570 void dev_kfree_skb_any(struct sk_buff *skb)
1572 if (in_irq() || irqs_disabled())
1573 dev_kfree_skb_irq(skb);
1577 EXPORT_SYMBOL(dev_kfree_skb_any);
1581 * netif_device_detach - mark device as removed
1582 * @dev: network device
1584 * Mark device as removed from system and therefore no longer available.
1586 void netif_device_detach(struct net_device *dev)
1588 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1589 netif_running(dev)) {
1590 netif_tx_stop_all_queues(dev);
1593 EXPORT_SYMBOL(netif_device_detach);
1596 * netif_device_attach - mark device as attached
1597 * @dev: network device
1599 * Mark device as attached from system and restart if needed.
1601 void netif_device_attach(struct net_device *dev)
1603 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1604 netif_running(dev)) {
1605 netif_tx_wake_all_queues(dev);
1606 __netdev_watchdog_up(dev);
1609 EXPORT_SYMBOL(netif_device_attach);
1611 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1613 return ((features & NETIF_F_GEN_CSUM) ||
1614 ((features & NETIF_F_IP_CSUM) &&
1615 protocol == htons(ETH_P_IP)) ||
1616 ((features & NETIF_F_IPV6_CSUM) &&
1617 protocol == htons(ETH_P_IPV6)) ||
1618 ((features & NETIF_F_FCOE_CRC) &&
1619 protocol == htons(ETH_P_FCOE)));
1622 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1624 if (can_checksum_protocol(dev->features, skb->protocol))
1627 if (skb->protocol == htons(ETH_P_8021Q)) {
1628 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1629 if (can_checksum_protocol(dev->features & dev->vlan_features,
1630 veh->h_vlan_encapsulated_proto))
1638 * skb_dev_set -- assign a new device to a buffer
1639 * @skb: buffer for the new device
1640 * @dev: network device
1642 * If an skb is owned by a device already, we have to reset
1643 * all data private to the namespace a device belongs to
1644 * before assigning it a new device.
1646 #ifdef CONFIG_NET_NS
1647 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1650 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1653 skb_init_secmark(skb);
1657 skb->ipvs_property = 0;
1658 #ifdef CONFIG_NET_SCHED
1664 EXPORT_SYMBOL(skb_set_dev);
1665 #endif /* CONFIG_NET_NS */
1668 * Invalidate hardware checksum when packet is to be mangled, and
1669 * complete checksum manually on outgoing path.
1671 int skb_checksum_help(struct sk_buff *skb)
1674 int ret = 0, offset;
1676 if (skb->ip_summed == CHECKSUM_COMPLETE)
1677 goto out_set_summed;
1679 if (unlikely(skb_shinfo(skb)->gso_size)) {
1680 /* Let GSO fix up the checksum. */
1681 goto out_set_summed;
1684 offset = skb->csum_start - skb_headroom(skb);
1685 BUG_ON(offset >= skb_headlen(skb));
1686 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1688 offset += skb->csum_offset;
1689 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1691 if (skb_cloned(skb) &&
1692 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1693 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1698 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1700 skb->ip_summed = CHECKSUM_NONE;
1704 EXPORT_SYMBOL(skb_checksum_help);
1707 * skb_gso_segment - Perform segmentation on skb.
1708 * @skb: buffer to segment
1709 * @features: features for the output path (see dev->features)
1711 * This function segments the given skb and returns a list of segments.
1713 * It may return NULL if the skb requires no segmentation. This is
1714 * only possible when GSO is used for verifying header integrity.
1716 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1718 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1719 struct packet_type *ptype;
1720 __be16 type = skb->protocol;
1723 skb_reset_mac_header(skb);
1724 skb->mac_len = skb->network_header - skb->mac_header;
1725 __skb_pull(skb, skb->mac_len);
1727 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1728 struct net_device *dev = skb->dev;
1729 struct ethtool_drvinfo info = {};
1731 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1732 dev->ethtool_ops->get_drvinfo(dev, &info);
1734 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1736 info.driver, dev ? dev->features : 0L,
1737 skb->sk ? skb->sk->sk_route_caps : 0L,
1738 skb->len, skb->data_len, skb->ip_summed);
1740 if (skb_header_cloned(skb) &&
1741 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1742 return ERR_PTR(err);
1746 list_for_each_entry_rcu(ptype,
1747 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1748 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1749 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1750 err = ptype->gso_send_check(skb);
1751 segs = ERR_PTR(err);
1752 if (err || skb_gso_ok(skb, features))
1754 __skb_push(skb, (skb->data -
1755 skb_network_header(skb)));
1757 segs = ptype->gso_segment(skb, features);
1763 __skb_push(skb, skb->data - skb_mac_header(skb));
1767 EXPORT_SYMBOL(skb_gso_segment);
1769 /* Take action when hardware reception checksum errors are detected. */
1771 void netdev_rx_csum_fault(struct net_device *dev)
1773 if (net_ratelimit()) {
1774 printk(KERN_ERR "%s: hw csum failure.\n",
1775 dev ? dev->name : "<unknown>");
1779 EXPORT_SYMBOL(netdev_rx_csum_fault);
1782 /* Actually, we should eliminate this check as soon as we know, that:
1783 * 1. IOMMU is present and allows to map all the memory.
1784 * 2. No high memory really exists on this machine.
1787 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1789 #ifdef CONFIG_HIGHMEM
1792 if (dev->features & NETIF_F_HIGHDMA)
1795 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1796 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1804 void (*destructor)(struct sk_buff *skb);
1807 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1809 static void dev_gso_skb_destructor(struct sk_buff *skb)
1811 struct dev_gso_cb *cb;
1814 struct sk_buff *nskb = skb->next;
1816 skb->next = nskb->next;
1819 } while (skb->next);
1821 cb = DEV_GSO_CB(skb);
1823 cb->destructor(skb);
1827 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1828 * @skb: buffer to segment
1830 * This function segments the given skb and stores the list of segments
1833 static int dev_gso_segment(struct sk_buff *skb)
1835 struct net_device *dev = skb->dev;
1836 struct sk_buff *segs;
1837 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1840 segs = skb_gso_segment(skb, features);
1842 /* Verifying header integrity only. */
1847 return PTR_ERR(segs);
1850 DEV_GSO_CB(skb)->destructor = skb->destructor;
1851 skb->destructor = dev_gso_skb_destructor;
1856 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1857 struct netdev_queue *txq)
1859 const struct net_device_ops *ops = dev->netdev_ops;
1860 int rc = NETDEV_TX_OK;
1862 if (likely(!skb->next)) {
1863 if (!list_empty(&ptype_all))
1864 dev_queue_xmit_nit(skb, dev);
1866 if (netif_needs_gso(dev, skb)) {
1867 if (unlikely(dev_gso_segment(skb)))
1874 * If device doesnt need skb->dst, release it right now while
1875 * its hot in this cpu cache
1877 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1880 rc = ops->ndo_start_xmit(skb, dev);
1881 if (rc == NETDEV_TX_OK)
1882 txq_trans_update(txq);
1884 * TODO: if skb_orphan() was called by
1885 * dev->hard_start_xmit() (for example, the unmodified
1886 * igb driver does that; bnx2 doesn't), then
1887 * skb_tx_software_timestamp() will be unable to send
1888 * back the time stamp.
1890 * How can this be prevented? Always create another
1891 * reference to the socket before calling
1892 * dev->hard_start_xmit()? Prevent that skb_orphan()
1893 * does anything in dev->hard_start_xmit() by clearing
1894 * the skb destructor before the call and restoring it
1895 * afterwards, then doing the skb_orphan() ourselves?
1902 struct sk_buff *nskb = skb->next;
1904 skb->next = nskb->next;
1908 * If device doesnt need nskb->dst, release it right now while
1909 * its hot in this cpu cache
1911 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1914 rc = ops->ndo_start_xmit(nskb, dev);
1915 if (unlikely(rc != NETDEV_TX_OK)) {
1916 if (rc & ~NETDEV_TX_MASK)
1917 goto out_kfree_gso_skb;
1918 nskb->next = skb->next;
1922 txq_trans_update(txq);
1923 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1924 return NETDEV_TX_BUSY;
1925 } while (skb->next);
1928 if (likely(skb->next == NULL))
1929 skb->destructor = DEV_GSO_CB(skb)->destructor;
1935 static u32 skb_tx_hashrnd;
1937 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1941 if (skb_rx_queue_recorded(skb)) {
1942 hash = skb_get_rx_queue(skb);
1943 while (unlikely(hash >= dev->real_num_tx_queues))
1944 hash -= dev->real_num_tx_queues;
1948 if (skb->sk && skb->sk->sk_hash)
1949 hash = skb->sk->sk_hash;
1951 hash = skb->protocol;
1953 hash = jhash_1word(hash, skb_tx_hashrnd);
1955 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1957 EXPORT_SYMBOL(skb_tx_hash);
1959 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1961 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1962 if (net_ratelimit()) {
1963 WARN(1, "%s selects TX queue %d, but "
1964 "real number of TX queues is %d\n",
1965 dev->name, queue_index,
1966 dev->real_num_tx_queues);
1973 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1974 struct sk_buff *skb)
1977 struct sock *sk = skb->sk;
1979 if (sk_tx_queue_recorded(sk)) {
1980 queue_index = sk_tx_queue_get(sk);
1982 const struct net_device_ops *ops = dev->netdev_ops;
1984 if (ops->ndo_select_queue) {
1985 queue_index = ops->ndo_select_queue(dev, skb);
1986 queue_index = dev_cap_txqueue(dev, queue_index);
1989 if (dev->real_num_tx_queues > 1)
1990 queue_index = skb_tx_hash(dev, skb);
1992 if (sk && sk->sk_dst_cache)
1993 sk_tx_queue_set(sk, queue_index);
1997 skb_set_queue_mapping(skb, queue_index);
1998 return netdev_get_tx_queue(dev, queue_index);
2001 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2002 struct net_device *dev,
2003 struct netdev_queue *txq)
2005 spinlock_t *root_lock = qdisc_lock(q);
2008 spin_lock(root_lock);
2009 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2012 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2013 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2015 * This is a work-conserving queue; there are no old skbs
2016 * waiting to be sent out; and the qdisc is not running -
2017 * xmit the skb directly.
2019 __qdisc_update_bstats(q, skb->len);
2020 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2023 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2025 rc = NET_XMIT_SUCCESS;
2027 rc = qdisc_enqueue_root(skb, q);
2030 spin_unlock(root_lock);
2036 * Returns true if either:
2037 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2038 * 2. skb is fragmented and the device does not support SG, or if
2039 * at least one of fragments is in highmem and device does not
2040 * support DMA from it.
2042 static inline int skb_needs_linearize(struct sk_buff *skb,
2043 struct net_device *dev)
2045 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2046 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2047 illegal_highdma(dev, skb)));
2051 * dev_queue_xmit - transmit a buffer
2052 * @skb: buffer to transmit
2054 * Queue a buffer for transmission to a network device. The caller must
2055 * have set the device and priority and built the buffer before calling
2056 * this function. The function can be called from an interrupt.
2058 * A negative errno code is returned on a failure. A success does not
2059 * guarantee the frame will be transmitted as it may be dropped due
2060 * to congestion or traffic shaping.
2062 * -----------------------------------------------------------------------------------
2063 * I notice this method can also return errors from the queue disciplines,
2064 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2067 * Regardless of the return value, the skb is consumed, so it is currently
2068 * difficult to retry a send to this method. (You can bump the ref count
2069 * before sending to hold a reference for retry if you are careful.)
2071 * When calling this method, interrupts MUST be enabled. This is because
2072 * the BH enable code must have IRQs enabled so that it will not deadlock.
2075 int dev_queue_xmit(struct sk_buff *skb)
2077 struct net_device *dev = skb->dev;
2078 struct netdev_queue *txq;
2082 /* GSO will handle the following emulations directly. */
2083 if (netif_needs_gso(dev, skb))
2086 /* Convert a paged skb to linear, if required */
2087 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2090 /* If packet is not checksummed and device does not support
2091 * checksumming for this protocol, complete checksumming here.
2093 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2094 skb_set_transport_header(skb, skb->csum_start -
2096 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2101 /* Disable soft irqs for various locks below. Also
2102 * stops preemption for RCU.
2106 txq = dev_pick_tx(dev, skb);
2107 q = rcu_dereference_bh(txq->qdisc);
2109 #ifdef CONFIG_NET_CLS_ACT
2110 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2113 rc = __dev_xmit_skb(skb, q, dev, txq);
2117 /* The device has no queue. Common case for software devices:
2118 loopback, all the sorts of tunnels...
2120 Really, it is unlikely that netif_tx_lock protection is necessary
2121 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2123 However, it is possible, that they rely on protection
2126 Check this and shot the lock. It is not prone from deadlocks.
2127 Either shot noqueue qdisc, it is even simpler 8)
2129 if (dev->flags & IFF_UP) {
2130 int cpu = smp_processor_id(); /* ok because BHs are off */
2132 if (txq->xmit_lock_owner != cpu) {
2134 HARD_TX_LOCK(dev, txq, cpu);
2136 if (!netif_tx_queue_stopped(txq)) {
2137 rc = dev_hard_start_xmit(skb, dev, txq);
2138 if (dev_xmit_complete(rc)) {
2139 HARD_TX_UNLOCK(dev, txq);
2143 HARD_TX_UNLOCK(dev, txq);
2144 if (net_ratelimit())
2145 printk(KERN_CRIT "Virtual device %s asks to "
2146 "queue packet!\n", dev->name);
2148 /* Recursion is detected! It is possible,
2150 if (net_ratelimit())
2151 printk(KERN_CRIT "Dead loop on virtual device "
2152 "%s, fix it urgently!\n", dev->name);
2157 rcu_read_unlock_bh();
2163 rcu_read_unlock_bh();
2166 EXPORT_SYMBOL(dev_queue_xmit);
2169 /*=======================================================================
2171 =======================================================================*/
2173 int netdev_max_backlog __read_mostly = 1000;
2174 int netdev_budget __read_mostly = 300;
2175 int weight_p __read_mostly = 64; /* old backlog weight */
2177 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2181 * netif_rx - post buffer to the network code
2182 * @skb: buffer to post
2184 * This function receives a packet from a device driver and queues it for
2185 * the upper (protocol) levels to process. It always succeeds. The buffer
2186 * may be dropped during processing for congestion control or by the
2190 * NET_RX_SUCCESS (no congestion)
2191 * NET_RX_DROP (packet was dropped)
2195 int netif_rx(struct sk_buff *skb)
2197 struct softnet_data *queue;
2198 unsigned long flags;
2200 /* if netpoll wants it, pretend we never saw it */
2201 if (netpoll_rx(skb))
2204 if (!skb->tstamp.tv64)
2208 * The code is rearranged so that the path is the most
2209 * short when CPU is congested, but is still operating.
2211 local_irq_save(flags);
2212 queue = &__get_cpu_var(softnet_data);
2214 __get_cpu_var(netdev_rx_stat).total++;
2215 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2216 if (queue->input_pkt_queue.qlen) {
2218 __skb_queue_tail(&queue->input_pkt_queue, skb);
2219 local_irq_restore(flags);
2220 return NET_RX_SUCCESS;
2223 napi_schedule(&queue->backlog);
2227 __get_cpu_var(netdev_rx_stat).dropped++;
2228 local_irq_restore(flags);
2233 EXPORT_SYMBOL(netif_rx);
2235 int netif_rx_ni(struct sk_buff *skb)
2240 err = netif_rx(skb);
2241 if (local_softirq_pending())
2247 EXPORT_SYMBOL(netif_rx_ni);
2249 static void net_tx_action(struct softirq_action *h)
2251 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2253 if (sd->completion_queue) {
2254 struct sk_buff *clist;
2256 local_irq_disable();
2257 clist = sd->completion_queue;
2258 sd->completion_queue = NULL;
2262 struct sk_buff *skb = clist;
2263 clist = clist->next;
2265 WARN_ON(atomic_read(&skb->users));
2270 if (sd->output_queue) {
2273 local_irq_disable();
2274 head = sd->output_queue;
2275 sd->output_queue = NULL;
2279 struct Qdisc *q = head;
2280 spinlock_t *root_lock;
2282 head = head->next_sched;
2284 root_lock = qdisc_lock(q);
2285 if (spin_trylock(root_lock)) {
2286 smp_mb__before_clear_bit();
2287 clear_bit(__QDISC_STATE_SCHED,
2290 spin_unlock(root_lock);
2292 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2294 __netif_reschedule(q);
2296 smp_mb__before_clear_bit();
2297 clear_bit(__QDISC_STATE_SCHED,
2305 static inline int deliver_skb(struct sk_buff *skb,
2306 struct packet_type *pt_prev,
2307 struct net_device *orig_dev)
2309 atomic_inc(&skb->users);
2310 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2313 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2315 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2316 /* This hook is defined here for ATM LANE */
2317 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2318 unsigned char *addr) __read_mostly;
2319 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2323 * If bridge module is loaded call bridging hook.
2324 * returns NULL if packet was consumed.
2326 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2327 struct sk_buff *skb) __read_mostly;
2328 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2330 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2331 struct packet_type **pt_prev, int *ret,
2332 struct net_device *orig_dev)
2334 struct net_bridge_port *port;
2336 if (skb->pkt_type == PACKET_LOOPBACK ||
2337 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2341 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2345 return br_handle_frame_hook(port, skb);
2348 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2351 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2352 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2353 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2355 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2356 struct packet_type **pt_prev,
2358 struct net_device *orig_dev)
2360 if (skb->dev->macvlan_port == NULL)
2364 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2367 return macvlan_handle_frame_hook(skb);
2370 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2373 #ifdef CONFIG_NET_CLS_ACT
2374 /* TODO: Maybe we should just force sch_ingress to be compiled in
2375 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2376 * a compare and 2 stores extra right now if we dont have it on
2377 * but have CONFIG_NET_CLS_ACT
2378 * NOTE: This doesnt stop any functionality; if you dont have
2379 * the ingress scheduler, you just cant add policies on ingress.
2382 static int ing_filter(struct sk_buff *skb)
2384 struct net_device *dev = skb->dev;
2385 u32 ttl = G_TC_RTTL(skb->tc_verd);
2386 struct netdev_queue *rxq;
2387 int result = TC_ACT_OK;
2390 if (MAX_RED_LOOP < ttl++) {
2392 "Redir loop detected Dropping packet (%d->%d)\n",
2393 skb->skb_iif, dev->ifindex);
2397 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2398 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2400 rxq = &dev->rx_queue;
2403 if (q != &noop_qdisc) {
2404 spin_lock(qdisc_lock(q));
2405 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2406 result = qdisc_enqueue_root(skb, q);
2407 spin_unlock(qdisc_lock(q));
2413 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2414 struct packet_type **pt_prev,
2415 int *ret, struct net_device *orig_dev)
2417 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2421 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2424 /* Huh? Why does turning on AF_PACKET affect this? */
2425 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2428 switch (ing_filter(skb)) {
2442 * netif_nit_deliver - deliver received packets to network taps
2445 * This function is used to deliver incoming packets to network
2446 * taps. It should be used when the normal netif_receive_skb path
2447 * is bypassed, for example because of VLAN acceleration.
2449 void netif_nit_deliver(struct sk_buff *skb)
2451 struct packet_type *ptype;
2453 if (list_empty(&ptype_all))
2456 skb_reset_network_header(skb);
2457 skb_reset_transport_header(skb);
2458 skb->mac_len = skb->network_header - skb->mac_header;
2461 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2462 if (!ptype->dev || ptype->dev == skb->dev)
2463 deliver_skb(skb, ptype, skb->dev);
2469 * netif_receive_skb - process receive buffer from network
2470 * @skb: buffer to process
2472 * netif_receive_skb() is the main receive data processing function.
2473 * It always succeeds. The buffer may be dropped during processing
2474 * for congestion control or by the protocol layers.
2476 * This function may only be called from softirq context and interrupts
2477 * should be enabled.
2479 * Return values (usually ignored):
2480 * NET_RX_SUCCESS: no congestion
2481 * NET_RX_DROP: packet was dropped
2483 int netif_receive_skb(struct sk_buff *skb)
2485 struct packet_type *ptype, *pt_prev;
2486 struct net_device *orig_dev;
2487 struct net_device *master;
2488 struct net_device *null_or_orig;
2489 struct net_device *null_or_bond;
2490 int ret = NET_RX_DROP;
2493 if (!skb->tstamp.tv64)
2496 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2497 return NET_RX_SUCCESS;
2499 /* if we've gotten here through NAPI, check netpoll */
2500 if (netpoll_receive_skb(skb))
2504 skb->skb_iif = skb->dev->ifindex;
2506 null_or_orig = NULL;
2507 orig_dev = skb->dev;
2508 master = ACCESS_ONCE(orig_dev->master);
2510 if (skb_bond_should_drop(skb, master))
2511 null_or_orig = orig_dev; /* deliver only exact match */
2516 __get_cpu_var(netdev_rx_stat).total++;
2518 skb_reset_network_header(skb);
2519 skb_reset_transport_header(skb);
2520 skb->mac_len = skb->network_header - skb->mac_header;
2526 #ifdef CONFIG_NET_CLS_ACT
2527 if (skb->tc_verd & TC_NCLS) {
2528 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2533 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2534 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2535 ptype->dev == orig_dev) {
2537 ret = deliver_skb(skb, pt_prev, orig_dev);
2542 #ifdef CONFIG_NET_CLS_ACT
2543 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2549 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2552 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2557 * Make sure frames received on VLAN interfaces stacked on
2558 * bonding interfaces still make their way to any base bonding
2559 * device that may have registered for a specific ptype. The
2560 * handler may have to adjust skb->dev and orig_dev.
2562 null_or_bond = NULL;
2563 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2564 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2565 null_or_bond = vlan_dev_real_dev(skb->dev);
2568 type = skb->protocol;
2569 list_for_each_entry_rcu(ptype,
2570 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2571 if (ptype->type == type && (ptype->dev == null_or_orig ||
2572 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2573 ptype->dev == null_or_bond)) {
2575 ret = deliver_skb(skb, pt_prev, orig_dev);
2581 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2584 /* Jamal, now you will not able to escape explaining
2585 * me how you were going to use this. :-)
2594 EXPORT_SYMBOL(netif_receive_skb);
2596 /* Network device is going away, flush any packets still pending */
2597 static void flush_backlog(void *arg)
2599 struct net_device *dev = arg;
2600 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2601 struct sk_buff *skb, *tmp;
2603 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2604 if (skb->dev == dev) {
2605 __skb_unlink(skb, &queue->input_pkt_queue);
2610 static int napi_gro_complete(struct sk_buff *skb)
2612 struct packet_type *ptype;
2613 __be16 type = skb->protocol;
2614 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2617 if (NAPI_GRO_CB(skb)->count == 1) {
2618 skb_shinfo(skb)->gso_size = 0;
2623 list_for_each_entry_rcu(ptype, head, list) {
2624 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2627 err = ptype->gro_complete(skb);
2633 WARN_ON(&ptype->list == head);
2635 return NET_RX_SUCCESS;
2639 return netif_receive_skb(skb);
2642 static void napi_gro_flush(struct napi_struct *napi)
2644 struct sk_buff *skb, *next;
2646 for (skb = napi->gro_list; skb; skb = next) {
2649 napi_gro_complete(skb);
2652 napi->gro_count = 0;
2653 napi->gro_list = NULL;
2656 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2658 struct sk_buff **pp = NULL;
2659 struct packet_type *ptype;
2660 __be16 type = skb->protocol;
2661 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2664 enum gro_result ret;
2666 if (!(skb->dev->features & NETIF_F_GRO))
2669 if (skb_is_gso(skb) || skb_has_frags(skb))
2673 list_for_each_entry_rcu(ptype, head, list) {
2674 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2677 skb_set_network_header(skb, skb_gro_offset(skb));
2678 mac_len = skb->network_header - skb->mac_header;
2679 skb->mac_len = mac_len;
2680 NAPI_GRO_CB(skb)->same_flow = 0;
2681 NAPI_GRO_CB(skb)->flush = 0;
2682 NAPI_GRO_CB(skb)->free = 0;
2684 pp = ptype->gro_receive(&napi->gro_list, skb);
2689 if (&ptype->list == head)
2692 same_flow = NAPI_GRO_CB(skb)->same_flow;
2693 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2696 struct sk_buff *nskb = *pp;
2700 napi_gro_complete(nskb);
2707 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2711 NAPI_GRO_CB(skb)->count = 1;
2712 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2713 skb->next = napi->gro_list;
2714 napi->gro_list = skb;
2718 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2719 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2721 BUG_ON(skb->end - skb->tail < grow);
2723 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2726 skb->data_len -= grow;
2728 skb_shinfo(skb)->frags[0].page_offset += grow;
2729 skb_shinfo(skb)->frags[0].size -= grow;
2731 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2732 put_page(skb_shinfo(skb)->frags[0].page);
2733 memmove(skb_shinfo(skb)->frags,
2734 skb_shinfo(skb)->frags + 1,
2735 --skb_shinfo(skb)->nr_frags);
2746 EXPORT_SYMBOL(dev_gro_receive);
2749 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2753 if (netpoll_rx_on(skb))
2756 for (p = napi->gro_list; p; p = p->next) {
2757 NAPI_GRO_CB(p)->same_flow =
2758 (p->dev == skb->dev) &&
2759 !compare_ether_header(skb_mac_header(p),
2760 skb_gro_mac_header(skb));
2761 NAPI_GRO_CB(p)->flush = 0;
2764 return dev_gro_receive(napi, skb);
2767 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2771 if (netif_receive_skb(skb))
2776 case GRO_MERGED_FREE:
2787 EXPORT_SYMBOL(napi_skb_finish);
2789 void skb_gro_reset_offset(struct sk_buff *skb)
2791 NAPI_GRO_CB(skb)->data_offset = 0;
2792 NAPI_GRO_CB(skb)->frag0 = NULL;
2793 NAPI_GRO_CB(skb)->frag0_len = 0;
2795 if (skb->mac_header == skb->tail &&
2796 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2797 NAPI_GRO_CB(skb)->frag0 =
2798 page_address(skb_shinfo(skb)->frags[0].page) +
2799 skb_shinfo(skb)->frags[0].page_offset;
2800 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2803 EXPORT_SYMBOL(skb_gro_reset_offset);
2805 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2807 skb_gro_reset_offset(skb);
2809 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2811 EXPORT_SYMBOL(napi_gro_receive);
2813 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2815 __skb_pull(skb, skb_headlen(skb));
2816 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2820 EXPORT_SYMBOL(napi_reuse_skb);
2822 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2824 struct sk_buff *skb = napi->skb;
2827 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2833 EXPORT_SYMBOL(napi_get_frags);
2835 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
2841 skb->protocol = eth_type_trans(skb, skb->dev);
2843 if (ret == GRO_HELD)
2844 skb_gro_pull(skb, -ETH_HLEN);
2845 else if (netif_receive_skb(skb))
2850 case GRO_MERGED_FREE:
2851 napi_reuse_skb(napi, skb);
2860 EXPORT_SYMBOL(napi_frags_finish);
2862 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2864 struct sk_buff *skb = napi->skb;
2871 skb_reset_mac_header(skb);
2872 skb_gro_reset_offset(skb);
2874 off = skb_gro_offset(skb);
2875 hlen = off + sizeof(*eth);
2876 eth = skb_gro_header_fast(skb, off);
2877 if (skb_gro_header_hard(skb, hlen)) {
2878 eth = skb_gro_header_slow(skb, hlen, off);
2879 if (unlikely(!eth)) {
2880 napi_reuse_skb(napi, skb);
2886 skb_gro_pull(skb, sizeof(*eth));
2889 * This works because the only protocols we care about don't require
2890 * special handling. We'll fix it up properly at the end.
2892 skb->protocol = eth->h_proto;
2897 EXPORT_SYMBOL(napi_frags_skb);
2899 gro_result_t napi_gro_frags(struct napi_struct *napi)
2901 struct sk_buff *skb = napi_frags_skb(napi);
2906 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2908 EXPORT_SYMBOL(napi_gro_frags);
2910 static int process_backlog(struct napi_struct *napi, int quota)
2913 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2914 unsigned long start_time = jiffies;
2916 napi->weight = weight_p;
2918 struct sk_buff *skb;
2920 local_irq_disable();
2921 skb = __skb_dequeue(&queue->input_pkt_queue);
2923 __napi_complete(napi);
2929 netif_receive_skb(skb);
2930 } while (++work < quota && jiffies == start_time);
2936 * __napi_schedule - schedule for receive
2937 * @n: entry to schedule
2939 * The entry's receive function will be scheduled to run
2941 void __napi_schedule(struct napi_struct *n)
2943 unsigned long flags;
2945 local_irq_save(flags);
2946 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2947 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2948 local_irq_restore(flags);
2950 EXPORT_SYMBOL(__napi_schedule);
2952 void __napi_complete(struct napi_struct *n)
2954 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2955 BUG_ON(n->gro_list);
2957 list_del(&n->poll_list);
2958 smp_mb__before_clear_bit();
2959 clear_bit(NAPI_STATE_SCHED, &n->state);
2961 EXPORT_SYMBOL(__napi_complete);
2963 void napi_complete(struct napi_struct *n)
2965 unsigned long flags;
2968 * don't let napi dequeue from the cpu poll list
2969 * just in case its running on a different cpu
2971 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2975 local_irq_save(flags);
2977 local_irq_restore(flags);
2979 EXPORT_SYMBOL(napi_complete);
2981 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2982 int (*poll)(struct napi_struct *, int), int weight)
2984 INIT_LIST_HEAD(&napi->poll_list);
2985 napi->gro_count = 0;
2986 napi->gro_list = NULL;
2989 napi->weight = weight;
2990 list_add(&napi->dev_list, &dev->napi_list);
2992 #ifdef CONFIG_NETPOLL
2993 spin_lock_init(&napi->poll_lock);
2994 napi->poll_owner = -1;
2996 set_bit(NAPI_STATE_SCHED, &napi->state);
2998 EXPORT_SYMBOL(netif_napi_add);
3000 void netif_napi_del(struct napi_struct *napi)
3002 struct sk_buff *skb, *next;
3004 list_del_init(&napi->dev_list);
3005 napi_free_frags(napi);
3007 for (skb = napi->gro_list; skb; skb = next) {
3013 napi->gro_list = NULL;
3014 napi->gro_count = 0;
3016 EXPORT_SYMBOL(netif_napi_del);
3019 static void net_rx_action(struct softirq_action *h)
3021 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
3022 unsigned long time_limit = jiffies + 2;
3023 int budget = netdev_budget;
3026 local_irq_disable();
3028 while (!list_empty(list)) {
3029 struct napi_struct *n;
3032 /* If softirq window is exhuasted then punt.
3033 * Allow this to run for 2 jiffies since which will allow
3034 * an average latency of 1.5/HZ.
3036 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3041 /* Even though interrupts have been re-enabled, this
3042 * access is safe because interrupts can only add new
3043 * entries to the tail of this list, and only ->poll()
3044 * calls can remove this head entry from the list.
3046 n = list_first_entry(list, struct napi_struct, poll_list);
3048 have = netpoll_poll_lock(n);
3052 /* This NAPI_STATE_SCHED test is for avoiding a race
3053 * with netpoll's poll_napi(). Only the entity which
3054 * obtains the lock and sees NAPI_STATE_SCHED set will
3055 * actually make the ->poll() call. Therefore we avoid
3056 * accidently calling ->poll() when NAPI is not scheduled.
3059 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3060 work = n->poll(n, weight);
3064 WARN_ON_ONCE(work > weight);
3068 local_irq_disable();
3070 /* Drivers must not modify the NAPI state if they
3071 * consume the entire weight. In such cases this code
3072 * still "owns" the NAPI instance and therefore can
3073 * move the instance around on the list at-will.
3075 if (unlikely(work == weight)) {
3076 if (unlikely(napi_disable_pending(n))) {
3079 local_irq_disable();
3081 list_move_tail(&n->poll_list, list);
3084 netpoll_poll_unlock(have);
3089 #ifdef CONFIG_NET_DMA
3091 * There may not be any more sk_buffs coming right now, so push
3092 * any pending DMA copies to hardware
3094 dma_issue_pending_all();
3100 __get_cpu_var(netdev_rx_stat).time_squeeze++;
3101 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3105 static gifconf_func_t *gifconf_list[NPROTO];
3108 * register_gifconf - register a SIOCGIF handler
3109 * @family: Address family
3110 * @gifconf: Function handler
3112 * Register protocol dependent address dumping routines. The handler
3113 * that is passed must not be freed or reused until it has been replaced
3114 * by another handler.
3116 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3118 if (family >= NPROTO)
3120 gifconf_list[family] = gifconf;
3123 EXPORT_SYMBOL(register_gifconf);
3127 * Map an interface index to its name (SIOCGIFNAME)
3131 * We need this ioctl for efficient implementation of the
3132 * if_indextoname() function required by the IPv6 API. Without
3133 * it, we would have to search all the interfaces to find a
3137 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3139 struct net_device *dev;
3143 * Fetch the caller's info block.
3146 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3150 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3156 strcpy(ifr.ifr_name, dev->name);
3159 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3165 * Perform a SIOCGIFCONF call. This structure will change
3166 * size eventually, and there is nothing I can do about it.
3167 * Thus we will need a 'compatibility mode'.
3170 static int dev_ifconf(struct net *net, char __user *arg)
3173 struct net_device *dev;
3180 * Fetch the caller's info block.
3183 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3190 * Loop over the interfaces, and write an info block for each.
3194 for_each_netdev(net, dev) {
3195 for (i = 0; i < NPROTO; i++) {
3196 if (gifconf_list[i]) {
3199 done = gifconf_list[i](dev, NULL, 0);
3201 done = gifconf_list[i](dev, pos + total,
3211 * All done. Write the updated control block back to the caller.
3213 ifc.ifc_len = total;
3216 * Both BSD and Solaris return 0 here, so we do too.
3218 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3221 #ifdef CONFIG_PROC_FS
3223 * This is invoked by the /proc filesystem handler to display a device
3226 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3229 struct net *net = seq_file_net(seq);
3231 struct net_device *dev;
3235 return SEQ_START_TOKEN;
3238 for_each_netdev_rcu(net, dev)
3245 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3247 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3248 first_net_device(seq_file_net(seq)) :
3249 next_net_device((struct net_device *)v);
3252 return rcu_dereference(dev);
3255 void dev_seq_stop(struct seq_file *seq, void *v)
3261 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3263 const struct net_device_stats *stats = dev_get_stats(dev);
3265 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3266 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3267 dev->name, stats->rx_bytes, stats->rx_packets,
3269 stats->rx_dropped + stats->rx_missed_errors,
3270 stats->rx_fifo_errors,
3271 stats->rx_length_errors + stats->rx_over_errors +
3272 stats->rx_crc_errors + stats->rx_frame_errors,
3273 stats->rx_compressed, stats->multicast,
3274 stats->tx_bytes, stats->tx_packets,
3275 stats->tx_errors, stats->tx_dropped,
3276 stats->tx_fifo_errors, stats->collisions,
3277 stats->tx_carrier_errors +
3278 stats->tx_aborted_errors +
3279 stats->tx_window_errors +
3280 stats->tx_heartbeat_errors,
3281 stats->tx_compressed);
3285 * Called from the PROCfs module. This now uses the new arbitrary sized
3286 * /proc/net interface to create /proc/net/dev
3288 static int dev_seq_show(struct seq_file *seq, void *v)
3290 if (v == SEQ_START_TOKEN)
3291 seq_puts(seq, "Inter-| Receive "
3293 " face |bytes packets errs drop fifo frame "
3294 "compressed multicast|bytes packets errs "
3295 "drop fifo colls carrier compressed\n");
3297 dev_seq_printf_stats(seq, v);
3301 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3303 struct netif_rx_stats *rc = NULL;
3305 while (*pos < nr_cpu_ids)
3306 if (cpu_online(*pos)) {
3307 rc = &per_cpu(netdev_rx_stat, *pos);
3314 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3316 return softnet_get_online(pos);
3319 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3322 return softnet_get_online(pos);
3325 static void softnet_seq_stop(struct seq_file *seq, void *v)
3329 static int softnet_seq_show(struct seq_file *seq, void *v)
3331 struct netif_rx_stats *s = v;
3333 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3334 s->total, s->dropped, s->time_squeeze, 0,
3335 0, 0, 0, 0, /* was fastroute */
3340 static const struct seq_operations dev_seq_ops = {
3341 .start = dev_seq_start,
3342 .next = dev_seq_next,
3343 .stop = dev_seq_stop,
3344 .show = dev_seq_show,
3347 static int dev_seq_open(struct inode *inode, struct file *file)
3349 return seq_open_net(inode, file, &dev_seq_ops,
3350 sizeof(struct seq_net_private));
3353 static const struct file_operations dev_seq_fops = {
3354 .owner = THIS_MODULE,
3355 .open = dev_seq_open,
3357 .llseek = seq_lseek,
3358 .release = seq_release_net,
3361 static const struct seq_operations softnet_seq_ops = {
3362 .start = softnet_seq_start,
3363 .next = softnet_seq_next,
3364 .stop = softnet_seq_stop,
3365 .show = softnet_seq_show,
3368 static int softnet_seq_open(struct inode *inode, struct file *file)
3370 return seq_open(file, &softnet_seq_ops);
3373 static const struct file_operations softnet_seq_fops = {
3374 .owner = THIS_MODULE,
3375 .open = softnet_seq_open,
3377 .llseek = seq_lseek,
3378 .release = seq_release,
3381 static void *ptype_get_idx(loff_t pos)
3383 struct packet_type *pt = NULL;
3387 list_for_each_entry_rcu(pt, &ptype_all, list) {
3393 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3394 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3403 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3407 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3410 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3412 struct packet_type *pt;
3413 struct list_head *nxt;
3417 if (v == SEQ_START_TOKEN)
3418 return ptype_get_idx(0);
3421 nxt = pt->list.next;
3422 if (pt->type == htons(ETH_P_ALL)) {
3423 if (nxt != &ptype_all)
3426 nxt = ptype_base[0].next;
3428 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3430 while (nxt == &ptype_base[hash]) {
3431 if (++hash >= PTYPE_HASH_SIZE)
3433 nxt = ptype_base[hash].next;
3436 return list_entry(nxt, struct packet_type, list);
3439 static void ptype_seq_stop(struct seq_file *seq, void *v)
3445 static int ptype_seq_show(struct seq_file *seq, void *v)
3447 struct packet_type *pt = v;
3449 if (v == SEQ_START_TOKEN)
3450 seq_puts(seq, "Type Device Function\n");
3451 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3452 if (pt->type == htons(ETH_P_ALL))
3453 seq_puts(seq, "ALL ");
3455 seq_printf(seq, "%04x", ntohs(pt->type));
3457 seq_printf(seq, " %-8s %pF\n",
3458 pt->dev ? pt->dev->name : "", pt->func);
3464 static const struct seq_operations ptype_seq_ops = {
3465 .start = ptype_seq_start,
3466 .next = ptype_seq_next,
3467 .stop = ptype_seq_stop,
3468 .show = ptype_seq_show,
3471 static int ptype_seq_open(struct inode *inode, struct file *file)
3473 return seq_open_net(inode, file, &ptype_seq_ops,
3474 sizeof(struct seq_net_private));
3477 static const struct file_operations ptype_seq_fops = {
3478 .owner = THIS_MODULE,
3479 .open = ptype_seq_open,
3481 .llseek = seq_lseek,
3482 .release = seq_release_net,
3486 static int __net_init dev_proc_net_init(struct net *net)
3490 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3492 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3494 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3497 if (wext_proc_init(net))
3503 proc_net_remove(net, "ptype");
3505 proc_net_remove(net, "softnet_stat");
3507 proc_net_remove(net, "dev");
3511 static void __net_exit dev_proc_net_exit(struct net *net)
3513 wext_proc_exit(net);
3515 proc_net_remove(net, "ptype");
3516 proc_net_remove(net, "softnet_stat");
3517 proc_net_remove(net, "dev");
3520 static struct pernet_operations __net_initdata dev_proc_ops = {
3521 .init = dev_proc_net_init,
3522 .exit = dev_proc_net_exit,
3525 static int __init dev_proc_init(void)
3527 return register_pernet_subsys(&dev_proc_ops);
3530 #define dev_proc_init() 0
3531 #endif /* CONFIG_PROC_FS */
3535 * netdev_set_master - set up master/slave pair
3536 * @slave: slave device
3537 * @master: new master device
3539 * Changes the master device of the slave. Pass %NULL to break the
3540 * bonding. The caller must hold the RTNL semaphore. On a failure
3541 * a negative errno code is returned. On success the reference counts
3542 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3543 * function returns zero.
3545 int netdev_set_master(struct net_device *slave, struct net_device *master)
3547 struct net_device *old = slave->master;
3557 slave->master = master;
3565 slave->flags |= IFF_SLAVE;
3567 slave->flags &= ~IFF_SLAVE;
3569 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3572 EXPORT_SYMBOL(netdev_set_master);
3574 static void dev_change_rx_flags(struct net_device *dev, int flags)
3576 const struct net_device_ops *ops = dev->netdev_ops;
3578 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3579 ops->ndo_change_rx_flags(dev, flags);
3582 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3584 unsigned short old_flags = dev->flags;
3590 dev->flags |= IFF_PROMISC;
3591 dev->promiscuity += inc;
3592 if (dev->promiscuity == 0) {
3595 * If inc causes overflow, untouch promisc and return error.
3598 dev->flags &= ~IFF_PROMISC;
3600 dev->promiscuity -= inc;
3601 printk(KERN_WARNING "%s: promiscuity touches roof, "
3602 "set promiscuity failed, promiscuity feature "
3603 "of device might be broken.\n", dev->name);
3607 if (dev->flags != old_flags) {
3608 printk(KERN_INFO "device %s %s promiscuous mode\n",
3609 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3611 if (audit_enabled) {
3612 current_uid_gid(&uid, &gid);
3613 audit_log(current->audit_context, GFP_ATOMIC,
3614 AUDIT_ANOM_PROMISCUOUS,
3615 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3616 dev->name, (dev->flags & IFF_PROMISC),
3617 (old_flags & IFF_PROMISC),
3618 audit_get_loginuid(current),
3620 audit_get_sessionid(current));
3623 dev_change_rx_flags(dev, IFF_PROMISC);
3629 * dev_set_promiscuity - update promiscuity count on a device
3633 * Add or remove promiscuity from a device. While the count in the device
3634 * remains above zero the interface remains promiscuous. Once it hits zero
3635 * the device reverts back to normal filtering operation. A negative inc
3636 * value is used to drop promiscuity on the device.
3637 * Return 0 if successful or a negative errno code on error.
3639 int dev_set_promiscuity(struct net_device *dev, int inc)
3641 unsigned short old_flags = dev->flags;
3644 err = __dev_set_promiscuity(dev, inc);
3647 if (dev->flags != old_flags)
3648 dev_set_rx_mode(dev);
3651 EXPORT_SYMBOL(dev_set_promiscuity);
3654 * dev_set_allmulti - update allmulti count on a device
3658 * Add or remove reception of all multicast frames to a device. While the
3659 * count in the device remains above zero the interface remains listening
3660 * to all interfaces. Once it hits zero the device reverts back to normal
3661 * filtering operation. A negative @inc value is used to drop the counter
3662 * when releasing a resource needing all multicasts.
3663 * Return 0 if successful or a negative errno code on error.
3666 int dev_set_allmulti(struct net_device *dev, int inc)
3668 unsigned short old_flags = dev->flags;
3672 dev->flags |= IFF_ALLMULTI;
3673 dev->allmulti += inc;
3674 if (dev->allmulti == 0) {
3677 * If inc causes overflow, untouch allmulti and return error.
3680 dev->flags &= ~IFF_ALLMULTI;
3682 dev->allmulti -= inc;
3683 printk(KERN_WARNING "%s: allmulti touches roof, "
3684 "set allmulti failed, allmulti feature of "
3685 "device might be broken.\n", dev->name);
3689 if (dev->flags ^ old_flags) {
3690 dev_change_rx_flags(dev, IFF_ALLMULTI);
3691 dev_set_rx_mode(dev);
3695 EXPORT_SYMBOL(dev_set_allmulti);
3698 * Upload unicast and multicast address lists to device and
3699 * configure RX filtering. When the device doesn't support unicast
3700 * filtering it is put in promiscuous mode while unicast addresses
3703 void __dev_set_rx_mode(struct net_device *dev)
3705 const struct net_device_ops *ops = dev->netdev_ops;
3707 /* dev_open will call this function so the list will stay sane. */
3708 if (!(dev->flags&IFF_UP))
3711 if (!netif_device_present(dev))
3714 if (ops->ndo_set_rx_mode)
3715 ops->ndo_set_rx_mode(dev);
3717 /* Unicast addresses changes may only happen under the rtnl,
3718 * therefore calling __dev_set_promiscuity here is safe.
3720 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
3721 __dev_set_promiscuity(dev, 1);
3722 dev->uc_promisc = 1;
3723 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
3724 __dev_set_promiscuity(dev, -1);
3725 dev->uc_promisc = 0;
3728 if (ops->ndo_set_multicast_list)
3729 ops->ndo_set_multicast_list(dev);
3733 void dev_set_rx_mode(struct net_device *dev)
3735 netif_addr_lock_bh(dev);
3736 __dev_set_rx_mode(dev);
3737 netif_addr_unlock_bh(dev);
3740 /* hw addresses list handling functions */
3742 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3743 int addr_len, unsigned char addr_type)
3745 struct netdev_hw_addr *ha;
3748 if (addr_len > MAX_ADDR_LEN)
3751 list_for_each_entry(ha, &list->list, list) {
3752 if (!memcmp(ha->addr, addr, addr_len) &&
3753 ha->type == addr_type) {
3760 alloc_size = sizeof(*ha);
3761 if (alloc_size < L1_CACHE_BYTES)
3762 alloc_size = L1_CACHE_BYTES;
3763 ha = kmalloc(alloc_size, GFP_ATOMIC);
3766 memcpy(ha->addr, addr, addr_len);
3767 ha->type = addr_type;
3770 list_add_tail_rcu(&ha->list, &list->list);
3775 static void ha_rcu_free(struct rcu_head *head)
3777 struct netdev_hw_addr *ha;
3779 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3783 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3784 int addr_len, unsigned char addr_type)
3786 struct netdev_hw_addr *ha;
3788 list_for_each_entry(ha, &list->list, list) {
3789 if (!memcmp(ha->addr, addr, addr_len) &&
3790 (ha->type == addr_type || !addr_type)) {
3793 list_del_rcu(&ha->list);
3794 call_rcu(&ha->rcu_head, ha_rcu_free);
3802 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3803 struct netdev_hw_addr_list *from_list,
3805 unsigned char addr_type)
3808 struct netdev_hw_addr *ha, *ha2;
3811 list_for_each_entry(ha, &from_list->list, list) {
3812 type = addr_type ? addr_type : ha->type;
3813 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3820 list_for_each_entry(ha2, &from_list->list, list) {
3823 type = addr_type ? addr_type : ha2->type;
3824 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3829 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3830 struct netdev_hw_addr_list *from_list,
3832 unsigned char addr_type)
3834 struct netdev_hw_addr *ha;
3837 list_for_each_entry(ha, &from_list->list, list) {
3838 type = addr_type ? addr_type : ha->type;
3839 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3843 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3844 struct netdev_hw_addr_list *from_list,
3848 struct netdev_hw_addr *ha, *tmp;
3850 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3852 err = __hw_addr_add(to_list, ha->addr,
3853 addr_len, ha->type);
3858 } else if (ha->refcount == 1) {
3859 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3860 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3866 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3867 struct netdev_hw_addr_list *from_list,
3870 struct netdev_hw_addr *ha, *tmp;
3872 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3874 __hw_addr_del(to_list, ha->addr,
3875 addr_len, ha->type);
3877 __hw_addr_del(from_list, ha->addr,
3878 addr_len, ha->type);
3883 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3885 struct netdev_hw_addr *ha, *tmp;
3887 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3888 list_del_rcu(&ha->list);
3889 call_rcu(&ha->rcu_head, ha_rcu_free);
3894 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3896 INIT_LIST_HEAD(&list->list);
3900 /* Device addresses handling functions */
3902 static void dev_addr_flush(struct net_device *dev)
3904 /* rtnl_mutex must be held here */
3906 __hw_addr_flush(&dev->dev_addrs);
3907 dev->dev_addr = NULL;
3910 static int dev_addr_init(struct net_device *dev)
3912 unsigned char addr[MAX_ADDR_LEN];
3913 struct netdev_hw_addr *ha;
3916 /* rtnl_mutex must be held here */
3918 __hw_addr_init(&dev->dev_addrs);
3919 memset(addr, 0, sizeof(addr));
3920 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3921 NETDEV_HW_ADDR_T_LAN);
3924 * Get the first (previously created) address from the list
3925 * and set dev_addr pointer to this location.
3927 ha = list_first_entry(&dev->dev_addrs.list,
3928 struct netdev_hw_addr, list);
3929 dev->dev_addr = ha->addr;
3935 * dev_addr_add - Add a device address
3937 * @addr: address to add
3938 * @addr_type: address type
3940 * Add a device address to the device or increase the reference count if
3941 * it already exists.
3943 * The caller must hold the rtnl_mutex.
3945 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3946 unsigned char addr_type)
3952 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3954 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3957 EXPORT_SYMBOL(dev_addr_add);
3960 * dev_addr_del - Release a device address.
3962 * @addr: address to delete
3963 * @addr_type: address type
3965 * Release reference to a device address and remove it from the device
3966 * if the reference count drops to zero.
3968 * The caller must hold the rtnl_mutex.
3970 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3971 unsigned char addr_type)
3974 struct netdev_hw_addr *ha;
3979 * We can not remove the first address from the list because
3980 * dev->dev_addr points to that.
3982 ha = list_first_entry(&dev->dev_addrs.list,
3983 struct netdev_hw_addr, list);
3984 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3987 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3990 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3993 EXPORT_SYMBOL(dev_addr_del);
3996 * dev_addr_add_multiple - Add device addresses from another device
3997 * @to_dev: device to which addresses will be added
3998 * @from_dev: device from which addresses will be added
3999 * @addr_type: address type - 0 means type will be used from from_dev
4001 * Add device addresses of the one device to another.
4003 * The caller must hold the rtnl_mutex.
4005 int dev_addr_add_multiple(struct net_device *to_dev,
4006 struct net_device *from_dev,
4007 unsigned char addr_type)
4013 if (from_dev->addr_len != to_dev->addr_len)
4015 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
4016 to_dev->addr_len, addr_type);
4018 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
4021 EXPORT_SYMBOL(dev_addr_add_multiple);
4024 * dev_addr_del_multiple - Delete device addresses by another device
4025 * @to_dev: device where the addresses will be deleted
4026 * @from_dev: device by which addresses the addresses will be deleted
4027 * @addr_type: address type - 0 means type will used from from_dev
4029 * Deletes addresses in to device by the list of addresses in from device.
4031 * The caller must hold the rtnl_mutex.
4033 int dev_addr_del_multiple(struct net_device *to_dev,
4034 struct net_device *from_dev,
4035 unsigned char addr_type)
4039 if (from_dev->addr_len != to_dev->addr_len)
4041 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
4042 to_dev->addr_len, addr_type);
4043 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
4046 EXPORT_SYMBOL(dev_addr_del_multiple);
4048 /* multicast addresses handling functions */
4050 int __dev_addr_delete(struct dev_addr_list **list, int *count,
4051 void *addr, int alen, int glbl)
4053 struct dev_addr_list *da;
4055 for (; (da = *list) != NULL; list = &da->next) {
4056 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4057 alen == da->da_addrlen) {
4059 int old_glbl = da->da_gusers;
4076 int __dev_addr_add(struct dev_addr_list **list, int *count,
4077 void *addr, int alen, int glbl)
4079 struct dev_addr_list *da;
4081 for (da = *list; da != NULL; da = da->next) {
4082 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4083 da->da_addrlen == alen) {
4085 int old_glbl = da->da_gusers;
4095 da = kzalloc(sizeof(*da), GFP_ATOMIC);
4098 memcpy(da->da_addr, addr, alen);
4099 da->da_addrlen = alen;
4101 da->da_gusers = glbl ? 1 : 0;
4109 * dev_unicast_delete - Release secondary unicast address.
4111 * @addr: address to delete
4113 * Release reference to a secondary unicast address and remove it
4114 * from the device if the reference count drops to zero.
4116 * The caller must hold the rtnl_mutex.
4118 int dev_unicast_delete(struct net_device *dev, void *addr)
4124 netif_addr_lock_bh(dev);
4125 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
4126 NETDEV_HW_ADDR_T_UNICAST);
4128 __dev_set_rx_mode(dev);
4129 netif_addr_unlock_bh(dev);
4132 EXPORT_SYMBOL(dev_unicast_delete);
4135 * dev_unicast_add - add a secondary unicast address
4137 * @addr: address to add
4139 * Add a secondary unicast address to the device or increase
4140 * the reference count if it already exists.
4142 * The caller must hold the rtnl_mutex.
4144 int dev_unicast_add(struct net_device *dev, void *addr)
4150 netif_addr_lock_bh(dev);
4151 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4152 NETDEV_HW_ADDR_T_UNICAST);
4154 __dev_set_rx_mode(dev);
4155 netif_addr_unlock_bh(dev);
4158 EXPORT_SYMBOL(dev_unicast_add);
4160 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4161 struct dev_addr_list **from, int *from_count)
4163 struct dev_addr_list *da, *next;
4167 while (da != NULL) {
4169 if (!da->da_synced) {
4170 err = __dev_addr_add(to, to_count,
4171 da->da_addr, da->da_addrlen, 0);
4176 } else if (da->da_users == 1) {
4177 __dev_addr_delete(to, to_count,
4178 da->da_addr, da->da_addrlen, 0);
4179 __dev_addr_delete(from, from_count,
4180 da->da_addr, da->da_addrlen, 0);
4186 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4188 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4189 struct dev_addr_list **from, int *from_count)
4191 struct dev_addr_list *da, *next;
4194 while (da != NULL) {
4196 if (da->da_synced) {
4197 __dev_addr_delete(to, to_count,
4198 da->da_addr, da->da_addrlen, 0);
4200 __dev_addr_delete(from, from_count,
4201 da->da_addr, da->da_addrlen, 0);
4206 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4209 * dev_unicast_sync - Synchronize device's unicast list to another device
4210 * @to: destination device
4211 * @from: source device
4213 * Add newly added addresses to the destination device and release
4214 * addresses that have no users left. The source device must be
4215 * locked by netif_tx_lock_bh.
4217 * This function is intended to be called from the dev->set_rx_mode
4218 * function of layered software devices.
4220 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4224 if (to->addr_len != from->addr_len)
4227 netif_addr_lock_bh(to);
4228 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4230 __dev_set_rx_mode(to);
4231 netif_addr_unlock_bh(to);
4234 EXPORT_SYMBOL(dev_unicast_sync);
4237 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4238 * @to: destination device
4239 * @from: source device
4241 * Remove all addresses that were added to the destination device by
4242 * dev_unicast_sync(). This function is intended to be called from the
4243 * dev->stop function of layered software devices.
4245 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4247 if (to->addr_len != from->addr_len)
4250 netif_addr_lock_bh(from);
4251 netif_addr_lock(to);
4252 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4253 __dev_set_rx_mode(to);
4254 netif_addr_unlock(to);
4255 netif_addr_unlock_bh(from);
4257 EXPORT_SYMBOL(dev_unicast_unsync);
4259 static void dev_unicast_flush(struct net_device *dev)
4261 netif_addr_lock_bh(dev);
4262 __hw_addr_flush(&dev->uc);
4263 netif_addr_unlock_bh(dev);
4266 static void dev_unicast_init(struct net_device *dev)
4268 __hw_addr_init(&dev->uc);
4272 static void __dev_addr_discard(struct dev_addr_list **list)
4274 struct dev_addr_list *tmp;
4276 while (*list != NULL) {
4279 if (tmp->da_users > tmp->da_gusers)
4280 printk("__dev_addr_discard: address leakage! "
4281 "da_users=%d\n", tmp->da_users);
4286 static void dev_addr_discard(struct net_device *dev)
4288 netif_addr_lock_bh(dev);
4290 __dev_addr_discard(&dev->mc_list);
4291 netdev_mc_count(dev) = 0;
4293 netif_addr_unlock_bh(dev);
4297 * dev_get_flags - get flags reported to userspace
4300 * Get the combination of flag bits exported through APIs to userspace.
4302 unsigned dev_get_flags(const struct net_device *dev)
4306 flags = (dev->flags & ~(IFF_PROMISC |
4311 (dev->gflags & (IFF_PROMISC |
4314 if (netif_running(dev)) {
4315 if (netif_oper_up(dev))
4316 flags |= IFF_RUNNING;
4317 if (netif_carrier_ok(dev))
4318 flags |= IFF_LOWER_UP;
4319 if (netif_dormant(dev))
4320 flags |= IFF_DORMANT;
4325 EXPORT_SYMBOL(dev_get_flags);
4327 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4329 int old_flags = dev->flags;
4335 * Set the flags on our device.
4338 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4339 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4341 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4345 * Load in the correct multicast list now the flags have changed.
4348 if ((old_flags ^ flags) & IFF_MULTICAST)
4349 dev_change_rx_flags(dev, IFF_MULTICAST);
4351 dev_set_rx_mode(dev);
4354 * Have we downed the interface. We handle IFF_UP ourselves
4355 * according to user attempts to set it, rather than blindly
4360 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4361 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4364 dev_set_rx_mode(dev);
4367 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4368 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4370 dev->gflags ^= IFF_PROMISC;
4371 dev_set_promiscuity(dev, inc);
4374 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4375 is important. Some (broken) drivers set IFF_PROMISC, when
4376 IFF_ALLMULTI is requested not asking us and not reporting.
4378 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4379 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4381 dev->gflags ^= IFF_ALLMULTI;
4382 dev_set_allmulti(dev, inc);
4388 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4390 unsigned int changes = dev->flags ^ old_flags;
4392 if (changes & IFF_UP) {
4393 if (dev->flags & IFF_UP)
4394 call_netdevice_notifiers(NETDEV_UP, dev);
4396 call_netdevice_notifiers(NETDEV_DOWN, dev);
4399 if (dev->flags & IFF_UP &&
4400 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4401 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4405 * dev_change_flags - change device settings
4407 * @flags: device state flags
4409 * Change settings on device based state flags. The flags are
4410 * in the userspace exported format.
4412 int dev_change_flags(struct net_device *dev, unsigned flags)
4415 int old_flags = dev->flags;
4417 ret = __dev_change_flags(dev, flags);
4421 changes = old_flags ^ dev->flags;
4423 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4425 __dev_notify_flags(dev, old_flags);
4428 EXPORT_SYMBOL(dev_change_flags);
4431 * dev_set_mtu - Change maximum transfer unit
4433 * @new_mtu: new transfer unit
4435 * Change the maximum transfer size of the network device.
4437 int dev_set_mtu(struct net_device *dev, int new_mtu)
4439 const struct net_device_ops *ops = dev->netdev_ops;
4442 if (new_mtu == dev->mtu)
4445 /* MTU must be positive. */
4449 if (!netif_device_present(dev))
4453 if (ops->ndo_change_mtu)
4454 err = ops->ndo_change_mtu(dev, new_mtu);
4458 if (!err && dev->flags & IFF_UP)
4459 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4462 EXPORT_SYMBOL(dev_set_mtu);
4465 * dev_set_mac_address - Change Media Access Control Address
4469 * Change the hardware (MAC) address of the device
4471 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4473 const struct net_device_ops *ops = dev->netdev_ops;
4476 if (!ops->ndo_set_mac_address)
4478 if (sa->sa_family != dev->type)
4480 if (!netif_device_present(dev))
4482 err = ops->ndo_set_mac_address(dev, sa);
4484 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4487 EXPORT_SYMBOL(dev_set_mac_address);
4490 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4492 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4495 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4501 case SIOCGIFFLAGS: /* Get interface flags */
4502 ifr->ifr_flags = (short) dev_get_flags(dev);
4505 case SIOCGIFMETRIC: /* Get the metric on the interface
4506 (currently unused) */
4507 ifr->ifr_metric = 0;
4510 case SIOCGIFMTU: /* Get the MTU of a device */
4511 ifr->ifr_mtu = dev->mtu;
4516 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4518 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4519 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4520 ifr->ifr_hwaddr.sa_family = dev->type;
4528 ifr->ifr_map.mem_start = dev->mem_start;
4529 ifr->ifr_map.mem_end = dev->mem_end;
4530 ifr->ifr_map.base_addr = dev->base_addr;
4531 ifr->ifr_map.irq = dev->irq;
4532 ifr->ifr_map.dma = dev->dma;
4533 ifr->ifr_map.port = dev->if_port;
4537 ifr->ifr_ifindex = dev->ifindex;
4541 ifr->ifr_qlen = dev->tx_queue_len;
4545 /* dev_ioctl() should ensure this case
4557 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4559 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4562 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4563 const struct net_device_ops *ops;
4568 ops = dev->netdev_ops;
4571 case SIOCSIFFLAGS: /* Set interface flags */
4572 return dev_change_flags(dev, ifr->ifr_flags);
4574 case SIOCSIFMETRIC: /* Set the metric on the interface
4575 (currently unused) */
4578 case SIOCSIFMTU: /* Set the MTU of a device */
4579 return dev_set_mtu(dev, ifr->ifr_mtu);
4582 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4584 case SIOCSIFHWBROADCAST:
4585 if (ifr->ifr_hwaddr.sa_family != dev->type)
4587 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4588 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4589 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4593 if (ops->ndo_set_config) {
4594 if (!netif_device_present(dev))
4596 return ops->ndo_set_config(dev, &ifr->ifr_map);
4601 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4602 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4604 if (!netif_device_present(dev))
4606 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4610 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4611 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4613 if (!netif_device_present(dev))
4615 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4619 if (ifr->ifr_qlen < 0)
4621 dev->tx_queue_len = ifr->ifr_qlen;
4625 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4626 return dev_change_name(dev, ifr->ifr_newname);
4629 * Unknown or private ioctl
4632 if ((cmd >= SIOCDEVPRIVATE &&
4633 cmd <= SIOCDEVPRIVATE + 15) ||
4634 cmd == SIOCBONDENSLAVE ||
4635 cmd == SIOCBONDRELEASE ||
4636 cmd == SIOCBONDSETHWADDR ||
4637 cmd == SIOCBONDSLAVEINFOQUERY ||
4638 cmd == SIOCBONDINFOQUERY ||
4639 cmd == SIOCBONDCHANGEACTIVE ||
4640 cmd == SIOCGMIIPHY ||
4641 cmd == SIOCGMIIREG ||
4642 cmd == SIOCSMIIREG ||
4643 cmd == SIOCBRADDIF ||
4644 cmd == SIOCBRDELIF ||
4645 cmd == SIOCSHWTSTAMP ||
4646 cmd == SIOCWANDEV) {
4648 if (ops->ndo_do_ioctl) {
4649 if (netif_device_present(dev))
4650 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4662 * This function handles all "interface"-type I/O control requests. The actual
4663 * 'doing' part of this is dev_ifsioc above.
4667 * dev_ioctl - network device ioctl
4668 * @net: the applicable net namespace
4669 * @cmd: command to issue
4670 * @arg: pointer to a struct ifreq in user space
4672 * Issue ioctl functions to devices. This is normally called by the
4673 * user space syscall interfaces but can sometimes be useful for
4674 * other purposes. The return value is the return from the syscall if
4675 * positive or a negative errno code on error.
4678 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4684 /* One special case: SIOCGIFCONF takes ifconf argument
4685 and requires shared lock, because it sleeps writing
4689 if (cmd == SIOCGIFCONF) {
4691 ret = dev_ifconf(net, (char __user *) arg);
4695 if (cmd == SIOCGIFNAME)
4696 return dev_ifname(net, (struct ifreq __user *)arg);
4698 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4701 ifr.ifr_name[IFNAMSIZ-1] = 0;
4703 colon = strchr(ifr.ifr_name, ':');
4708 * See which interface the caller is talking about.
4713 * These ioctl calls:
4714 * - can be done by all.
4715 * - atomic and do not require locking.
4726 dev_load(net, ifr.ifr_name);
4728 ret = dev_ifsioc_locked(net, &ifr, cmd);
4733 if (copy_to_user(arg, &ifr,
4734 sizeof(struct ifreq)))
4740 dev_load(net, ifr.ifr_name);
4742 ret = dev_ethtool(net, &ifr);
4747 if (copy_to_user(arg, &ifr,
4748 sizeof(struct ifreq)))
4754 * These ioctl calls:
4755 * - require superuser power.
4756 * - require strict serialization.
4762 if (!capable(CAP_NET_ADMIN))
4764 dev_load(net, ifr.ifr_name);
4766 ret = dev_ifsioc(net, &ifr, cmd);
4771 if (copy_to_user(arg, &ifr,
4772 sizeof(struct ifreq)))
4778 * These ioctl calls:
4779 * - require superuser power.
4780 * - require strict serialization.
4781 * - do not return a value
4791 case SIOCSIFHWBROADCAST:
4794 case SIOCBONDENSLAVE:
4795 case SIOCBONDRELEASE:
4796 case SIOCBONDSETHWADDR:
4797 case SIOCBONDCHANGEACTIVE:
4801 if (!capable(CAP_NET_ADMIN))
4804 case SIOCBONDSLAVEINFOQUERY:
4805 case SIOCBONDINFOQUERY:
4806 dev_load(net, ifr.ifr_name);
4808 ret = dev_ifsioc(net, &ifr, cmd);
4813 /* Get the per device memory space. We can add this but
4814 * currently do not support it */
4816 /* Set the per device memory buffer space.
4817 * Not applicable in our case */
4822 * Unknown or private ioctl.
4825 if (cmd == SIOCWANDEV ||
4826 (cmd >= SIOCDEVPRIVATE &&
4827 cmd <= SIOCDEVPRIVATE + 15)) {
4828 dev_load(net, ifr.ifr_name);
4830 ret = dev_ifsioc(net, &ifr, cmd);
4832 if (!ret && copy_to_user(arg, &ifr,
4833 sizeof(struct ifreq)))
4837 /* Take care of Wireless Extensions */
4838 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4839 return wext_handle_ioctl(net, &ifr, cmd, arg);
4846 * dev_new_index - allocate an ifindex
4847 * @net: the applicable net namespace
4849 * Returns a suitable unique value for a new device interface
4850 * number. The caller must hold the rtnl semaphore or the
4851 * dev_base_lock to be sure it remains unique.
4853 static int dev_new_index(struct net *net)
4859 if (!__dev_get_by_index(net, ifindex))
4864 /* Delayed registration/unregisteration */
4865 static LIST_HEAD(net_todo_list);
4867 static void net_set_todo(struct net_device *dev)
4869 list_add_tail(&dev->todo_list, &net_todo_list);
4872 static void rollback_registered_many(struct list_head *head)
4874 struct net_device *dev, *tmp;
4876 BUG_ON(dev_boot_phase);
4879 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4880 /* Some devices call without registering
4881 * for initialization unwind. Remove those
4882 * devices and proceed with the remaining.
4884 if (dev->reg_state == NETREG_UNINITIALIZED) {
4885 pr_debug("unregister_netdevice: device %s/%p never "
4886 "was registered\n", dev->name, dev);
4889 list_del(&dev->unreg_list);
4893 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4895 /* If device is running, close it first. */
4898 /* And unlink it from device chain. */
4899 unlist_netdevice(dev);
4901 dev->reg_state = NETREG_UNREGISTERING;
4906 list_for_each_entry(dev, head, unreg_list) {
4907 /* Shutdown queueing discipline. */
4911 /* Notify protocols, that we are about to destroy
4912 this device. They should clean all the things.
4914 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4916 if (!dev->rtnl_link_ops ||
4917 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4918 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4921 * Flush the unicast and multicast chains
4923 dev_unicast_flush(dev);
4924 dev_addr_discard(dev);
4926 if (dev->netdev_ops->ndo_uninit)
4927 dev->netdev_ops->ndo_uninit(dev);
4929 /* Notifier chain MUST detach us from master device. */
4930 WARN_ON(dev->master);
4932 /* Remove entries from kobject tree */
4933 netdev_unregister_kobject(dev);
4936 /* Process any work delayed until the end of the batch */
4937 dev = list_first_entry(head, struct net_device, unreg_list);
4938 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4942 list_for_each_entry(dev, head, unreg_list)
4946 static void rollback_registered(struct net_device *dev)
4950 list_add(&dev->unreg_list, &single);
4951 rollback_registered_many(&single);
4954 static void __netdev_init_queue_locks_one(struct net_device *dev,
4955 struct netdev_queue *dev_queue,
4958 spin_lock_init(&dev_queue->_xmit_lock);
4959 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4960 dev_queue->xmit_lock_owner = -1;
4963 static void netdev_init_queue_locks(struct net_device *dev)
4965 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4966 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4969 unsigned long netdev_fix_features(unsigned long features, const char *name)
4971 /* Fix illegal SG+CSUM combinations. */
4972 if ((features & NETIF_F_SG) &&
4973 !(features & NETIF_F_ALL_CSUM)) {
4975 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4976 "checksum feature.\n", name);
4977 features &= ~NETIF_F_SG;
4980 /* TSO requires that SG is present as well. */
4981 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4983 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4984 "SG feature.\n", name);
4985 features &= ~NETIF_F_TSO;
4988 if (features & NETIF_F_UFO) {
4989 if (!(features & NETIF_F_GEN_CSUM)) {
4991 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4992 "since no NETIF_F_HW_CSUM feature.\n",
4994 features &= ~NETIF_F_UFO;
4997 if (!(features & NETIF_F_SG)) {
4999 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
5000 "since no NETIF_F_SG feature.\n", name);
5001 features &= ~NETIF_F_UFO;
5007 EXPORT_SYMBOL(netdev_fix_features);
5010 * netif_stacked_transfer_operstate - transfer operstate
5011 * @rootdev: the root or lower level device to transfer state from
5012 * @dev: the device to transfer operstate to
5014 * Transfer operational state from root to device. This is normally
5015 * called when a stacking relationship exists between the root
5016 * device and the device(a leaf device).
5018 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5019 struct net_device *dev)
5021 if (rootdev->operstate == IF_OPER_DORMANT)
5022 netif_dormant_on(dev);
5024 netif_dormant_off(dev);
5026 if (netif_carrier_ok(rootdev)) {
5027 if (!netif_carrier_ok(dev))
5028 netif_carrier_on(dev);
5030 if (netif_carrier_ok(dev))
5031 netif_carrier_off(dev);
5034 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5037 * register_netdevice - register a network device
5038 * @dev: device to register
5040 * Take a completed network device structure and add it to the kernel
5041 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5042 * chain. 0 is returned on success. A negative errno code is returned
5043 * on a failure to set up the device, or if the name is a duplicate.
5045 * Callers must hold the rtnl semaphore. You may want
5046 * register_netdev() instead of this.
5049 * The locking appears insufficient to guarantee two parallel registers
5050 * will not get the same name.
5053 int register_netdevice(struct net_device *dev)
5056 struct net *net = dev_net(dev);
5058 BUG_ON(dev_boot_phase);
5063 /* When net_device's are persistent, this will be fatal. */
5064 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5067 spin_lock_init(&dev->addr_list_lock);
5068 netdev_set_addr_lockdep_class(dev);
5069 netdev_init_queue_locks(dev);
5073 /* Init, if this function is available */
5074 if (dev->netdev_ops->ndo_init) {
5075 ret = dev->netdev_ops->ndo_init(dev);
5083 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
5087 dev->ifindex = dev_new_index(net);
5088 if (dev->iflink == -1)
5089 dev->iflink = dev->ifindex;
5091 /* Fix illegal checksum combinations */
5092 if ((dev->features & NETIF_F_HW_CSUM) &&
5093 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5094 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5096 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5099 if ((dev->features & NETIF_F_NO_CSUM) &&
5100 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5101 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5103 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5106 dev->features = netdev_fix_features(dev->features, dev->name);
5108 /* Enable software GSO if SG is supported. */
5109 if (dev->features & NETIF_F_SG)
5110 dev->features |= NETIF_F_GSO;
5112 netdev_initialize_kobject(dev);
5114 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5115 ret = notifier_to_errno(ret);
5119 ret = netdev_register_kobject(dev);
5122 dev->reg_state = NETREG_REGISTERED;
5125 * Default initial state at registry is that the
5126 * device is present.
5129 set_bit(__LINK_STATE_PRESENT, &dev->state);
5131 dev_init_scheduler(dev);
5133 list_netdevice(dev);
5135 /* Notify protocols, that a new device appeared. */
5136 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5137 ret = notifier_to_errno(ret);
5139 rollback_registered(dev);
5140 dev->reg_state = NETREG_UNREGISTERED;
5143 * Prevent userspace races by waiting until the network
5144 * device is fully setup before sending notifications.
5146 if (!dev->rtnl_link_ops ||
5147 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5148 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5154 if (dev->netdev_ops->ndo_uninit)
5155 dev->netdev_ops->ndo_uninit(dev);
5158 EXPORT_SYMBOL(register_netdevice);
5161 * init_dummy_netdev - init a dummy network device for NAPI
5162 * @dev: device to init
5164 * This takes a network device structure and initialize the minimum
5165 * amount of fields so it can be used to schedule NAPI polls without
5166 * registering a full blown interface. This is to be used by drivers
5167 * that need to tie several hardware interfaces to a single NAPI
5168 * poll scheduler due to HW limitations.
5170 int init_dummy_netdev(struct net_device *dev)
5172 /* Clear everything. Note we don't initialize spinlocks
5173 * are they aren't supposed to be taken by any of the
5174 * NAPI code and this dummy netdev is supposed to be
5175 * only ever used for NAPI polls
5177 memset(dev, 0, sizeof(struct net_device));
5179 /* make sure we BUG if trying to hit standard
5180 * register/unregister code path
5182 dev->reg_state = NETREG_DUMMY;
5184 /* initialize the ref count */
5185 atomic_set(&dev->refcnt, 1);
5187 /* NAPI wants this */
5188 INIT_LIST_HEAD(&dev->napi_list);
5190 /* a dummy interface is started by default */
5191 set_bit(__LINK_STATE_PRESENT, &dev->state);
5192 set_bit(__LINK_STATE_START, &dev->state);
5196 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5200 * register_netdev - register a network device
5201 * @dev: device to register
5203 * Take a completed network device structure and add it to the kernel
5204 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5205 * chain. 0 is returned on success. A negative errno code is returned
5206 * on a failure to set up the device, or if the name is a duplicate.
5208 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5209 * and expands the device name if you passed a format string to
5212 int register_netdev(struct net_device *dev)
5219 * If the name is a format string the caller wants us to do a
5222 if (strchr(dev->name, '%')) {
5223 err = dev_alloc_name(dev, dev->name);
5228 err = register_netdevice(dev);
5233 EXPORT_SYMBOL(register_netdev);
5236 * netdev_wait_allrefs - wait until all references are gone.
5238 * This is called when unregistering network devices.
5240 * Any protocol or device that holds a reference should register
5241 * for netdevice notification, and cleanup and put back the
5242 * reference if they receive an UNREGISTER event.
5243 * We can get stuck here if buggy protocols don't correctly
5246 static void netdev_wait_allrefs(struct net_device *dev)
5248 unsigned long rebroadcast_time, warning_time;
5250 linkwatch_forget_dev(dev);
5252 rebroadcast_time = warning_time = jiffies;
5253 while (atomic_read(&dev->refcnt) != 0) {
5254 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5257 /* Rebroadcast unregister notification */
5258 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5259 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5260 * should have already handle it the first time */
5262 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5264 /* We must not have linkwatch events
5265 * pending on unregister. If this
5266 * happens, we simply run the queue
5267 * unscheduled, resulting in a noop
5270 linkwatch_run_queue();
5275 rebroadcast_time = jiffies;
5280 if (time_after(jiffies, warning_time + 10 * HZ)) {
5281 printk(KERN_EMERG "unregister_netdevice: "
5282 "waiting for %s to become free. Usage "
5284 dev->name, atomic_read(&dev->refcnt));
5285 warning_time = jiffies;
5294 * register_netdevice(x1);
5295 * register_netdevice(x2);
5297 * unregister_netdevice(y1);
5298 * unregister_netdevice(y2);
5304 * We are invoked by rtnl_unlock().
5305 * This allows us to deal with problems:
5306 * 1) We can delete sysfs objects which invoke hotplug
5307 * without deadlocking with linkwatch via keventd.
5308 * 2) Since we run with the RTNL semaphore not held, we can sleep
5309 * safely in order to wait for the netdev refcnt to drop to zero.
5311 * We must not return until all unregister events added during
5312 * the interval the lock was held have been completed.
5314 void netdev_run_todo(void)
5316 struct list_head list;
5318 /* Snapshot list, allow later requests */
5319 list_replace_init(&net_todo_list, &list);
5323 while (!list_empty(&list)) {
5324 struct net_device *dev
5325 = list_first_entry(&list, struct net_device, todo_list);
5326 list_del(&dev->todo_list);
5328 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5329 printk(KERN_ERR "network todo '%s' but state %d\n",
5330 dev->name, dev->reg_state);
5335 dev->reg_state = NETREG_UNREGISTERED;
5337 on_each_cpu(flush_backlog, dev, 1);
5339 netdev_wait_allrefs(dev);
5342 BUG_ON(atomic_read(&dev->refcnt));
5343 WARN_ON(dev->ip_ptr);
5344 WARN_ON(dev->ip6_ptr);
5345 WARN_ON(dev->dn_ptr);
5347 if (dev->destructor)
5348 dev->destructor(dev);
5350 /* Free network device */
5351 kobject_put(&dev->dev.kobj);
5356 * dev_txq_stats_fold - fold tx_queues stats
5357 * @dev: device to get statistics from
5358 * @stats: struct net_device_stats to hold results
5360 void dev_txq_stats_fold(const struct net_device *dev,
5361 struct net_device_stats *stats)
5363 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5365 struct netdev_queue *txq;
5367 for (i = 0; i < dev->num_tx_queues; i++) {
5368 txq = netdev_get_tx_queue(dev, i);
5369 tx_bytes += txq->tx_bytes;
5370 tx_packets += txq->tx_packets;
5371 tx_dropped += txq->tx_dropped;
5373 if (tx_bytes || tx_packets || tx_dropped) {
5374 stats->tx_bytes = tx_bytes;
5375 stats->tx_packets = tx_packets;
5376 stats->tx_dropped = tx_dropped;
5379 EXPORT_SYMBOL(dev_txq_stats_fold);
5382 * dev_get_stats - get network device statistics
5383 * @dev: device to get statistics from
5385 * Get network statistics from device. The device driver may provide
5386 * its own method by setting dev->netdev_ops->get_stats; otherwise
5387 * the internal statistics structure is used.
5389 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5391 const struct net_device_ops *ops = dev->netdev_ops;
5393 if (ops->ndo_get_stats)
5394 return ops->ndo_get_stats(dev);
5396 dev_txq_stats_fold(dev, &dev->stats);
5399 EXPORT_SYMBOL(dev_get_stats);
5401 static void netdev_init_one_queue(struct net_device *dev,
5402 struct netdev_queue *queue,
5408 static void netdev_init_queues(struct net_device *dev)
5410 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5411 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5412 spin_lock_init(&dev->tx_global_lock);
5416 * alloc_netdev_mq - allocate network device
5417 * @sizeof_priv: size of private data to allocate space for
5418 * @name: device name format string
5419 * @setup: callback to initialize device
5420 * @queue_count: the number of subqueues to allocate
5422 * Allocates a struct net_device with private data area for driver use
5423 * and performs basic initialization. Also allocates subquue structs
5424 * for each queue on the device at the end of the netdevice.
5426 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5427 void (*setup)(struct net_device *), unsigned int queue_count)
5429 struct netdev_queue *tx;
5430 struct net_device *dev;
5432 struct net_device *p;
5434 BUG_ON(strlen(name) >= sizeof(dev->name));
5436 alloc_size = sizeof(struct net_device);
5438 /* ensure 32-byte alignment of private area */
5439 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5440 alloc_size += sizeof_priv;
5442 /* ensure 32-byte alignment of whole construct */
5443 alloc_size += NETDEV_ALIGN - 1;
5445 p = kzalloc(alloc_size, GFP_KERNEL);
5447 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5451 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5453 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5458 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5459 dev->padded = (char *)dev - (char *)p;
5461 if (dev_addr_init(dev))
5464 dev_unicast_init(dev);
5466 dev_net_set(dev, &init_net);
5469 dev->num_tx_queues = queue_count;
5470 dev->real_num_tx_queues = queue_count;
5472 dev->gso_max_size = GSO_MAX_SIZE;
5474 netdev_init_queues(dev);
5476 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5477 dev->ethtool_ntuple_list.count = 0;
5478 INIT_LIST_HEAD(&dev->napi_list);
5479 INIT_LIST_HEAD(&dev->unreg_list);
5480 INIT_LIST_HEAD(&dev->link_watch_list);
5481 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5483 strcpy(dev->name, name);
5493 EXPORT_SYMBOL(alloc_netdev_mq);
5496 * free_netdev - free network device
5499 * This function does the last stage of destroying an allocated device
5500 * interface. The reference to the device object is released.
5501 * If this is the last reference then it will be freed.
5503 void free_netdev(struct net_device *dev)
5505 struct napi_struct *p, *n;
5507 release_net(dev_net(dev));
5511 /* Flush device addresses */
5512 dev_addr_flush(dev);
5514 /* Clear ethtool n-tuple list */
5515 ethtool_ntuple_flush(dev);
5517 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5520 /* Compatibility with error handling in drivers */
5521 if (dev->reg_state == NETREG_UNINITIALIZED) {
5522 kfree((char *)dev - dev->padded);
5526 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5527 dev->reg_state = NETREG_RELEASED;
5529 /* will free via device release */
5530 put_device(&dev->dev);
5532 EXPORT_SYMBOL(free_netdev);
5535 * synchronize_net - Synchronize with packet receive processing
5537 * Wait for packets currently being received to be done.
5538 * Does not block later packets from starting.
5540 void synchronize_net(void)
5545 EXPORT_SYMBOL(synchronize_net);
5548 * unregister_netdevice_queue - remove device from the kernel
5552 * This function shuts down a device interface and removes it
5553 * from the kernel tables.
5554 * If head not NULL, device is queued to be unregistered later.
5556 * Callers must hold the rtnl semaphore. You may want
5557 * unregister_netdev() instead of this.
5560 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5565 list_move_tail(&dev->unreg_list, head);
5567 rollback_registered(dev);
5568 /* Finish processing unregister after unlock */
5572 EXPORT_SYMBOL(unregister_netdevice_queue);
5575 * unregister_netdevice_many - unregister many devices
5576 * @head: list of devices
5578 void unregister_netdevice_many(struct list_head *head)
5580 struct net_device *dev;
5582 if (!list_empty(head)) {
5583 rollback_registered_many(head);
5584 list_for_each_entry(dev, head, unreg_list)
5588 EXPORT_SYMBOL(unregister_netdevice_many);
5591 * unregister_netdev - remove device from the kernel
5594 * This function shuts down a device interface and removes it
5595 * from the kernel tables.
5597 * This is just a wrapper for unregister_netdevice that takes
5598 * the rtnl semaphore. In general you want to use this and not
5599 * unregister_netdevice.
5601 void unregister_netdev(struct net_device *dev)
5604 unregister_netdevice(dev);
5607 EXPORT_SYMBOL(unregister_netdev);
5610 * dev_change_net_namespace - move device to different nethost namespace
5612 * @net: network namespace
5613 * @pat: If not NULL name pattern to try if the current device name
5614 * is already taken in the destination network namespace.
5616 * This function shuts down a device interface and moves it
5617 * to a new network namespace. On success 0 is returned, on
5618 * a failure a netagive errno code is returned.
5620 * Callers must hold the rtnl semaphore.
5623 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5629 /* Don't allow namespace local devices to be moved. */
5631 if (dev->features & NETIF_F_NETNS_LOCAL)
5635 /* Don't allow real devices to be moved when sysfs
5639 if (dev->dev.parent)
5643 /* Ensure the device has been registrered */
5645 if (dev->reg_state != NETREG_REGISTERED)
5648 /* Get out if there is nothing todo */
5650 if (net_eq(dev_net(dev), net))
5653 /* Pick the destination device name, and ensure
5654 * we can use it in the destination network namespace.
5657 if (__dev_get_by_name(net, dev->name)) {
5658 /* We get here if we can't use the current device name */
5661 if (dev_get_valid_name(net, pat, dev->name, 1))
5666 * And now a mini version of register_netdevice unregister_netdevice.
5669 /* If device is running close it first. */
5672 /* And unlink it from device chain */
5674 unlist_netdevice(dev);
5678 /* Shutdown queueing discipline. */
5681 /* Notify protocols, that we are about to destroy
5682 this device. They should clean all the things.
5684 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5685 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5688 * Flush the unicast and multicast chains
5690 dev_unicast_flush(dev);
5691 dev_addr_discard(dev);
5693 netdev_unregister_kobject(dev);
5695 /* Actually switch the network namespace */
5696 dev_net_set(dev, net);
5698 /* If there is an ifindex conflict assign a new one */
5699 if (__dev_get_by_index(net, dev->ifindex)) {
5700 int iflink = (dev->iflink == dev->ifindex);
5701 dev->ifindex = dev_new_index(net);
5703 dev->iflink = dev->ifindex;
5706 /* Fixup kobjects */
5707 err = netdev_register_kobject(dev);
5710 /* Add the device back in the hashes */
5711 list_netdevice(dev);
5713 /* Notify protocols, that a new device appeared. */
5714 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5717 * Prevent userspace races by waiting until the network
5718 * device is fully setup before sending notifications.
5720 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5727 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5729 static int dev_cpu_callback(struct notifier_block *nfb,
5730 unsigned long action,
5733 struct sk_buff **list_skb;
5734 struct Qdisc **list_net;
5735 struct sk_buff *skb;
5736 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5737 struct softnet_data *sd, *oldsd;
5739 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5742 local_irq_disable();
5743 cpu = smp_processor_id();
5744 sd = &per_cpu(softnet_data, cpu);
5745 oldsd = &per_cpu(softnet_data, oldcpu);
5747 /* Find end of our completion_queue. */
5748 list_skb = &sd->completion_queue;
5750 list_skb = &(*list_skb)->next;
5751 /* Append completion queue from offline CPU. */
5752 *list_skb = oldsd->completion_queue;
5753 oldsd->completion_queue = NULL;
5755 /* Find end of our output_queue. */
5756 list_net = &sd->output_queue;
5758 list_net = &(*list_net)->next_sched;
5759 /* Append output queue from offline CPU. */
5760 *list_net = oldsd->output_queue;
5761 oldsd->output_queue = NULL;
5763 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5766 /* Process offline CPU's input_pkt_queue */
5767 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5775 * netdev_increment_features - increment feature set by one
5776 * @all: current feature set
5777 * @one: new feature set
5778 * @mask: mask feature set
5780 * Computes a new feature set after adding a device with feature set
5781 * @one to the master device with current feature set @all. Will not
5782 * enable anything that is off in @mask. Returns the new feature set.
5784 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5787 /* If device needs checksumming, downgrade to it. */
5788 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5789 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5790 else if (mask & NETIF_F_ALL_CSUM) {
5791 /* If one device supports v4/v6 checksumming, set for all. */
5792 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5793 !(all & NETIF_F_GEN_CSUM)) {
5794 all &= ~NETIF_F_ALL_CSUM;
5795 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5798 /* If one device supports hw checksumming, set for all. */
5799 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5800 all &= ~NETIF_F_ALL_CSUM;
5801 all |= NETIF_F_HW_CSUM;
5805 one |= NETIF_F_ALL_CSUM;
5807 one |= all & NETIF_F_ONE_FOR_ALL;
5808 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5809 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5813 EXPORT_SYMBOL(netdev_increment_features);
5815 static struct hlist_head *netdev_create_hash(void)
5818 struct hlist_head *hash;
5820 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5822 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5823 INIT_HLIST_HEAD(&hash[i]);
5828 /* Initialize per network namespace state */
5829 static int __net_init netdev_init(struct net *net)
5831 INIT_LIST_HEAD(&net->dev_base_head);
5833 net->dev_name_head = netdev_create_hash();
5834 if (net->dev_name_head == NULL)
5837 net->dev_index_head = netdev_create_hash();
5838 if (net->dev_index_head == NULL)
5844 kfree(net->dev_name_head);
5850 * netdev_drivername - network driver for the device
5851 * @dev: network device
5852 * @buffer: buffer for resulting name
5853 * @len: size of buffer
5855 * Determine network driver for device.
5857 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5859 const struct device_driver *driver;
5860 const struct device *parent;
5862 if (len <= 0 || !buffer)
5866 parent = dev->dev.parent;
5871 driver = parent->driver;
5872 if (driver && driver->name)
5873 strlcpy(buffer, driver->name, len);
5877 static void __net_exit netdev_exit(struct net *net)
5879 kfree(net->dev_name_head);
5880 kfree(net->dev_index_head);
5883 static struct pernet_operations __net_initdata netdev_net_ops = {
5884 .init = netdev_init,
5885 .exit = netdev_exit,
5888 static void __net_exit default_device_exit(struct net *net)
5890 struct net_device *dev, *aux;
5892 * Push all migratable network devices back to the
5893 * initial network namespace
5896 for_each_netdev_safe(net, dev, aux) {
5898 char fb_name[IFNAMSIZ];
5900 /* Ignore unmoveable devices (i.e. loopback) */
5901 if (dev->features & NETIF_F_NETNS_LOCAL)
5904 /* Leave virtual devices for the generic cleanup */
5905 if (dev->rtnl_link_ops)
5908 /* Push remaing network devices to init_net */
5909 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5910 err = dev_change_net_namespace(dev, &init_net, fb_name);
5912 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5913 __func__, dev->name, err);
5920 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5922 /* At exit all network devices most be removed from a network
5923 * namespace. Do this in the reverse order of registeration.
5924 * Do this across as many network namespaces as possible to
5925 * improve batching efficiency.
5927 struct net_device *dev;
5929 LIST_HEAD(dev_kill_list);
5932 list_for_each_entry(net, net_list, exit_list) {
5933 for_each_netdev_reverse(net, dev) {
5934 if (dev->rtnl_link_ops)
5935 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5937 unregister_netdevice_queue(dev, &dev_kill_list);
5940 unregister_netdevice_many(&dev_kill_list);
5944 static struct pernet_operations __net_initdata default_device_ops = {
5945 .exit = default_device_exit,
5946 .exit_batch = default_device_exit_batch,
5950 * Initialize the DEV module. At boot time this walks the device list and
5951 * unhooks any devices that fail to initialise (normally hardware not
5952 * present) and leaves us with a valid list of present and active devices.
5957 * This is called single threaded during boot, so no need
5958 * to take the rtnl semaphore.
5960 static int __init net_dev_init(void)
5962 int i, rc = -ENOMEM;
5964 BUG_ON(!dev_boot_phase);
5966 if (dev_proc_init())
5969 if (netdev_kobject_init())
5972 INIT_LIST_HEAD(&ptype_all);
5973 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5974 INIT_LIST_HEAD(&ptype_base[i]);
5976 if (register_pernet_subsys(&netdev_net_ops))
5980 * Initialise the packet receive queues.
5983 for_each_possible_cpu(i) {
5984 struct softnet_data *queue;
5986 queue = &per_cpu(softnet_data, i);
5987 skb_queue_head_init(&queue->input_pkt_queue);
5988 queue->completion_queue = NULL;
5989 INIT_LIST_HEAD(&queue->poll_list);
5991 queue->backlog.poll = process_backlog;
5992 queue->backlog.weight = weight_p;
5993 queue->backlog.gro_list = NULL;
5994 queue->backlog.gro_count = 0;
5999 /* The loopback device is special if any other network devices
6000 * is present in a network namespace the loopback device must
6001 * be present. Since we now dynamically allocate and free the
6002 * loopback device ensure this invariant is maintained by
6003 * keeping the loopback device as the first device on the
6004 * list of network devices. Ensuring the loopback devices
6005 * is the first device that appears and the last network device
6008 if (register_pernet_device(&loopback_net_ops))
6011 if (register_pernet_device(&default_device_ops))
6014 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6015 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6017 hotcpu_notifier(dev_cpu_callback, 0);
6025 subsys_initcall(net_dev_init);
6027 static int __init initialize_hashrnd(void)
6029 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
6033 late_initcall_sync(initialize_hashrnd);