2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
103 #include <linux/if_bridge.h>
104 #include <linux/if_macvlan.h>
106 #include <net/pkt_sched.h>
107 #include <net/checksum.h>
108 #include <net/xfrm.h>
109 #include <linux/highmem.h>
110 #include <linux/init.h>
111 #include <linux/kmod.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/wext.h>
117 #include <net/iw_handler.h>
118 #include <asm/current.h>
119 #include <linux/audit.h>
120 #include <linux/dmaengine.h>
121 #include <linux/err.h>
122 #include <linux/ctype.h>
123 #include <linux/if_arp.h>
124 #include <linux/if_vlan.h>
125 #include <linux/ip.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
133 #include "net-sysfs.h"
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
142 * The list of packet types we will receive (as opposed to discard)
143 * and the routines to invoke.
145 * Why 16. Because with 16 the only overlap we get on a hash of the
146 * low nibble of the protocol value is RARP/SNAP/X.25.
148 * NOTE: That is no longer true with the addition of VLAN tags. Not
149 * sure which should go first, but I bet it won't make much
150 * difference if we are running VLANs. The good news is that
151 * this protocol won't be in the list unless compiled in, so
152 * the average user (w/out VLANs) will not be adversely affected.
169 #define PTYPE_HASH_SIZE (16)
170 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
172 static DEFINE_SPINLOCK(ptype_lock);
173 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
174 static struct list_head ptype_all __read_mostly; /* Taps */
177 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
180 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
182 * Writers must hold the rtnl semaphore while they loop through the
183 * dev_base_head list, and hold dev_base_lock for writing when they do the
184 * actual updates. This allows pure readers to access the list even
185 * while a writer is preparing to update it.
187 * To put it another way, dev_base_lock is held for writing only to
188 * protect against pure readers; the rtnl semaphore provides the
189 * protection against other writers.
191 * See, for example usages, register_netdevice() and
192 * unregister_netdevice(), which must be called with the rtnl
195 DEFINE_RWLOCK(dev_base_lock);
196 EXPORT_SYMBOL(dev_base_lock);
198 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
200 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
201 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
204 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
206 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
209 /* Device list insertion */
210 static int list_netdevice(struct net_device *dev)
212 struct net *net = dev_net(dev);
216 write_lock_bh(&dev_base_lock);
217 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
218 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
219 hlist_add_head_rcu(&dev->index_hlist,
220 dev_index_hash(net, dev->ifindex));
221 write_unlock_bh(&dev_base_lock);
225 /* Device list removal
226 * caller must respect a RCU grace period before freeing/reusing dev
228 static void unlist_netdevice(struct net_device *dev)
232 /* Unlink dev from the device chain */
233 write_lock_bh(&dev_base_lock);
234 list_del_rcu(&dev->dev_list);
235 hlist_del_rcu(&dev->name_hlist);
236 hlist_del_rcu(&dev->index_hlist);
237 write_unlock_bh(&dev_base_lock);
244 static RAW_NOTIFIER_HEAD(netdev_chain);
247 * Device drivers call our routines to queue packets here. We empty the
248 * queue in the local softnet handler.
251 DEFINE_PER_CPU(struct softnet_data, softnet_data);
252 EXPORT_PER_CPU_SYMBOL(softnet_data);
254 #ifdef CONFIG_LOCKDEP
256 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
257 * according to dev->type
259 static const unsigned short netdev_lock_type[] =
260 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
261 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
262 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
263 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
264 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
265 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
266 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
267 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
268 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
269 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
270 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
271 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
272 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
273 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
274 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
275 ARPHRD_VOID, ARPHRD_NONE};
277 static const char *const netdev_lock_name[] =
278 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
279 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
280 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
281 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
282 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
283 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
284 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
285 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
286 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
287 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
288 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
289 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
290 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
291 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
292 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
293 "_xmit_VOID", "_xmit_NONE"};
295 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
296 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
298 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
302 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
303 if (netdev_lock_type[i] == dev_type)
305 /* the last key is used by default */
306 return ARRAY_SIZE(netdev_lock_type) - 1;
309 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
310 unsigned short dev_type)
314 i = netdev_lock_pos(dev_type);
315 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
316 netdev_lock_name[i]);
319 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
323 i = netdev_lock_pos(dev->type);
324 lockdep_set_class_and_name(&dev->addr_list_lock,
325 &netdev_addr_lock_key[i],
326 netdev_lock_name[i]);
329 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
330 unsigned short dev_type)
333 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
338 /*******************************************************************************
340 Protocol management and registration routines
342 *******************************************************************************/
345 * Add a protocol ID to the list. Now that the input handler is
346 * smarter we can dispense with all the messy stuff that used to be
349 * BEWARE!!! Protocol handlers, mangling input packets,
350 * MUST BE last in hash buckets and checking protocol handlers
351 * MUST start from promiscuous ptype_all chain in net_bh.
352 * It is true now, do not change it.
353 * Explanation follows: if protocol handler, mangling packet, will
354 * be the first on list, it is not able to sense, that packet
355 * is cloned and should be copied-on-write, so that it will
356 * change it and subsequent readers will get broken packet.
361 * dev_add_pack - add packet handler
362 * @pt: packet type declaration
364 * Add a protocol handler to the networking stack. The passed &packet_type
365 * is linked into kernel lists and may not be freed until it has been
366 * removed from the kernel lists.
368 * This call does not sleep therefore it can not
369 * guarantee all CPU's that are in middle of receiving packets
370 * will see the new packet type (until the next received packet).
373 void dev_add_pack(struct packet_type *pt)
377 spin_lock_bh(&ptype_lock);
378 if (pt->type == htons(ETH_P_ALL))
379 list_add_rcu(&pt->list, &ptype_all);
381 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
382 list_add_rcu(&pt->list, &ptype_base[hash]);
384 spin_unlock_bh(&ptype_lock);
386 EXPORT_SYMBOL(dev_add_pack);
389 * __dev_remove_pack - remove packet handler
390 * @pt: packet type declaration
392 * Remove a protocol handler that was previously added to the kernel
393 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
394 * from the kernel lists and can be freed or reused once this function
397 * The packet type might still be in use by receivers
398 * and must not be freed until after all the CPU's have gone
399 * through a quiescent state.
401 void __dev_remove_pack(struct packet_type *pt)
403 struct list_head *head;
404 struct packet_type *pt1;
406 spin_lock_bh(&ptype_lock);
408 if (pt->type == htons(ETH_P_ALL))
411 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
413 list_for_each_entry(pt1, head, list) {
415 list_del_rcu(&pt->list);
420 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
422 spin_unlock_bh(&ptype_lock);
424 EXPORT_SYMBOL(__dev_remove_pack);
427 * dev_remove_pack - remove packet handler
428 * @pt: packet type declaration
430 * Remove a protocol handler that was previously added to the kernel
431 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
432 * from the kernel lists and can be freed or reused once this function
435 * This call sleeps to guarantee that no CPU is looking at the packet
438 void dev_remove_pack(struct packet_type *pt)
440 __dev_remove_pack(pt);
444 EXPORT_SYMBOL(dev_remove_pack);
446 /******************************************************************************
448 Device Boot-time Settings Routines
450 *******************************************************************************/
452 /* Boot time configuration table */
453 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
456 * netdev_boot_setup_add - add new setup entry
457 * @name: name of the device
458 * @map: configured settings for the device
460 * Adds new setup entry to the dev_boot_setup list. The function
461 * returns 0 on error and 1 on success. This is a generic routine to
464 static int netdev_boot_setup_add(char *name, struct ifmap *map)
466 struct netdev_boot_setup *s;
470 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
471 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
472 memset(s[i].name, 0, sizeof(s[i].name));
473 strlcpy(s[i].name, name, IFNAMSIZ);
474 memcpy(&s[i].map, map, sizeof(s[i].map));
479 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
483 * netdev_boot_setup_check - check boot time settings
484 * @dev: the netdevice
486 * Check boot time settings for the device.
487 * The found settings are set for the device to be used
488 * later in the device probing.
489 * Returns 0 if no settings found, 1 if they are.
491 int netdev_boot_setup_check(struct net_device *dev)
493 struct netdev_boot_setup *s = dev_boot_setup;
496 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
497 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
498 !strcmp(dev->name, s[i].name)) {
499 dev->irq = s[i].map.irq;
500 dev->base_addr = s[i].map.base_addr;
501 dev->mem_start = s[i].map.mem_start;
502 dev->mem_end = s[i].map.mem_end;
508 EXPORT_SYMBOL(netdev_boot_setup_check);
512 * netdev_boot_base - get address from boot time settings
513 * @prefix: prefix for network device
514 * @unit: id for network device
516 * Check boot time settings for the base address of device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found.
521 unsigned long netdev_boot_base(const char *prefix, int unit)
523 const struct netdev_boot_setup *s = dev_boot_setup;
527 sprintf(name, "%s%d", prefix, unit);
530 * If device already registered then return base of 1
531 * to indicate not to probe for this interface
533 if (__dev_get_by_name(&init_net, name))
536 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
537 if (!strcmp(name, s[i].name))
538 return s[i].map.base_addr;
543 * Saves at boot time configured settings for any netdevice.
545 int __init netdev_boot_setup(char *str)
550 str = get_options(str, ARRAY_SIZE(ints), ints);
555 memset(&map, 0, sizeof(map));
559 map.base_addr = ints[2];
561 map.mem_start = ints[3];
563 map.mem_end = ints[4];
565 /* Add new entry to the list */
566 return netdev_boot_setup_add(str, &map);
569 __setup("netdev=", netdev_boot_setup);
571 /*******************************************************************************
573 Device Interface Subroutines
575 *******************************************************************************/
578 * __dev_get_by_name - find a device by its name
579 * @net: the applicable net namespace
580 * @name: name to find
582 * Find an interface by name. Must be called under RTNL semaphore
583 * or @dev_base_lock. If the name is found a pointer to the device
584 * is returned. If the name is not found then %NULL is returned. The
585 * reference counters are not incremented so the caller must be
586 * careful with locks.
589 struct net_device *__dev_get_by_name(struct net *net, const char *name)
591 struct hlist_node *p;
592 struct net_device *dev;
593 struct hlist_head *head = dev_name_hash(net, name);
595 hlist_for_each_entry(dev, p, head, name_hlist)
596 if (!strncmp(dev->name, name, IFNAMSIZ))
601 EXPORT_SYMBOL(__dev_get_by_name);
604 * dev_get_by_name_rcu - find a device by its name
605 * @net: the applicable net namespace
606 * @name: name to find
608 * Find an interface by name.
609 * If the name is found a pointer to the device is returned.
610 * If the name is not found then %NULL is returned.
611 * The reference counters are not incremented so the caller must be
612 * careful with locks. The caller must hold RCU lock.
615 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
617 struct hlist_node *p;
618 struct net_device *dev;
619 struct hlist_head *head = dev_name_hash(net, name);
621 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
622 if (!strncmp(dev->name, name, IFNAMSIZ))
627 EXPORT_SYMBOL(dev_get_by_name_rcu);
630 * dev_get_by_name - find a device by its name
631 * @net: the applicable net namespace
632 * @name: name to find
634 * Find an interface by name. This can be called from any
635 * context and does its own locking. The returned handle has
636 * the usage count incremented and the caller must use dev_put() to
637 * release it when it is no longer needed. %NULL is returned if no
638 * matching device is found.
641 struct net_device *dev_get_by_name(struct net *net, const char *name)
643 struct net_device *dev;
646 dev = dev_get_by_name_rcu(net, name);
652 EXPORT_SYMBOL(dev_get_by_name);
655 * __dev_get_by_index - find a device by its ifindex
656 * @net: the applicable net namespace
657 * @ifindex: index of device
659 * Search for an interface by index. Returns %NULL if the device
660 * is not found or a pointer to the device. The device has not
661 * had its reference counter increased so the caller must be careful
662 * about locking. The caller must hold either the RTNL semaphore
666 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
668 struct hlist_node *p;
669 struct net_device *dev;
670 struct hlist_head *head = dev_index_hash(net, ifindex);
672 hlist_for_each_entry(dev, p, head, index_hlist)
673 if (dev->ifindex == ifindex)
678 EXPORT_SYMBOL(__dev_get_by_index);
681 * dev_get_by_index_rcu - find a device by its ifindex
682 * @net: the applicable net namespace
683 * @ifindex: index of device
685 * Search for an interface by index. Returns %NULL if the device
686 * is not found or a pointer to the device. The device has not
687 * had its reference counter increased so the caller must be careful
688 * about locking. The caller must hold RCU lock.
691 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
693 struct hlist_node *p;
694 struct net_device *dev;
695 struct hlist_head *head = dev_index_hash(net, ifindex);
697 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
698 if (dev->ifindex == ifindex)
703 EXPORT_SYMBOL(dev_get_by_index_rcu);
707 * dev_get_by_index - find a device by its ifindex
708 * @net: the applicable net namespace
709 * @ifindex: index of device
711 * Search for an interface by index. Returns NULL if the device
712 * is not found or a pointer to the device. The device returned has
713 * had a reference added and the pointer is safe until the user calls
714 * dev_put to indicate they have finished with it.
717 struct net_device *dev_get_by_index(struct net *net, int ifindex)
719 struct net_device *dev;
722 dev = dev_get_by_index_rcu(net, ifindex);
728 EXPORT_SYMBOL(dev_get_by_index);
731 * dev_getbyhwaddr - find a device by its hardware address
732 * @net: the applicable net namespace
733 * @type: media type of device
734 * @ha: hardware address
736 * Search for an interface by MAC address. Returns NULL if the device
737 * is not found or a pointer to the device. The caller must hold the
738 * rtnl semaphore. The returned device has not had its ref count increased
739 * and the caller must therefore be careful about locking
742 * If the API was consistent this would be __dev_get_by_hwaddr
745 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
747 struct net_device *dev;
751 for_each_netdev(net, dev)
752 if (dev->type == type &&
753 !memcmp(dev->dev_addr, ha, dev->addr_len))
758 EXPORT_SYMBOL(dev_getbyhwaddr);
760 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
762 struct net_device *dev;
765 for_each_netdev(net, dev)
766 if (dev->type == type)
771 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
773 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
775 struct net_device *dev;
778 dev = __dev_getfirstbyhwtype(net, type);
784 EXPORT_SYMBOL(dev_getfirstbyhwtype);
787 * dev_get_by_flags - find any device with given flags
788 * @net: the applicable net namespace
789 * @if_flags: IFF_* values
790 * @mask: bitmask of bits in if_flags to check
792 * Search for any interface with the given flags. Returns NULL if a device
793 * is not found or a pointer to the device. The device returned has
794 * had a reference added and the pointer is safe until the user calls
795 * dev_put to indicate they have finished with it.
798 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
801 struct net_device *dev, *ret;
805 for_each_netdev_rcu(net, dev) {
806 if (((dev->flags ^ if_flags) & mask) == 0) {
815 EXPORT_SYMBOL(dev_get_by_flags);
818 * dev_valid_name - check if name is okay for network device
821 * Network device names need to be valid file names to
822 * to allow sysfs to work. We also disallow any kind of
825 int dev_valid_name(const char *name)
829 if (strlen(name) >= IFNAMSIZ)
831 if (!strcmp(name, ".") || !strcmp(name, ".."))
835 if (*name == '/' || isspace(*name))
841 EXPORT_SYMBOL(dev_valid_name);
844 * __dev_alloc_name - allocate a name for a device
845 * @net: network namespace to allocate the device name in
846 * @name: name format string
847 * @buf: scratch buffer and result name string
849 * Passed a format string - eg "lt%d" it will try and find a suitable
850 * id. It scans list of devices to build up a free map, then chooses
851 * the first empty slot. The caller must hold the dev_base or rtnl lock
852 * while allocating the name and adding the device in order to avoid
854 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
855 * Returns the number of the unit assigned or a negative errno code.
858 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
862 const int max_netdevices = 8*PAGE_SIZE;
863 unsigned long *inuse;
864 struct net_device *d;
866 p = strnchr(name, IFNAMSIZ-1, '%');
869 * Verify the string as this thing may have come from
870 * the user. There must be either one "%d" and no other "%"
873 if (p[1] != 'd' || strchr(p + 2, '%'))
876 /* Use one page as a bit array of possible slots */
877 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
881 for_each_netdev(net, d) {
882 if (!sscanf(d->name, name, &i))
884 if (i < 0 || i >= max_netdevices)
887 /* avoid cases where sscanf is not exact inverse of printf */
888 snprintf(buf, IFNAMSIZ, name, i);
889 if (!strncmp(buf, d->name, IFNAMSIZ))
893 i = find_first_zero_bit(inuse, max_netdevices);
894 free_page((unsigned long) inuse);
898 snprintf(buf, IFNAMSIZ, name, i);
899 if (!__dev_get_by_name(net, buf))
902 /* It is possible to run out of possible slots
903 * when the name is long and there isn't enough space left
904 * for the digits, or if all bits are used.
910 * dev_alloc_name - allocate a name for a device
912 * @name: name format string
914 * Passed a format string - eg "lt%d" it will try and find a suitable
915 * id. It scans list of devices to build up a free map, then chooses
916 * the first empty slot. The caller must hold the dev_base or rtnl lock
917 * while allocating the name and adding the device in order to avoid
919 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
920 * Returns the number of the unit assigned or a negative errno code.
923 int dev_alloc_name(struct net_device *dev, const char *name)
929 BUG_ON(!dev_net(dev));
931 ret = __dev_alloc_name(net, name, buf);
933 strlcpy(dev->name, buf, IFNAMSIZ);
936 EXPORT_SYMBOL(dev_alloc_name);
938 static int dev_get_valid_name(struct net *net, const char *name, char *buf,
941 if (!dev_valid_name(name))
944 if (fmt && strchr(name, '%'))
945 return __dev_alloc_name(net, name, buf);
946 else if (__dev_get_by_name(net, name))
948 else if (buf != name)
949 strlcpy(buf, name, IFNAMSIZ);
955 * dev_change_name - change name of a device
957 * @newname: name (or format string) must be at least IFNAMSIZ
959 * Change name of a device, can pass format strings "eth%d".
962 int dev_change_name(struct net_device *dev, const char *newname)
964 char oldname[IFNAMSIZ];
970 BUG_ON(!dev_net(dev));
973 if (dev->flags & IFF_UP)
976 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
979 memcpy(oldname, dev->name, IFNAMSIZ);
981 err = dev_get_valid_name(net, newname, dev->name, 1);
986 /* For now only devices in the initial network namespace
989 if (net_eq(net, &init_net)) {
990 ret = device_rename(&dev->dev, dev->name);
992 memcpy(dev->name, oldname, IFNAMSIZ);
997 write_lock_bh(&dev_base_lock);
998 hlist_del(&dev->name_hlist);
999 write_unlock_bh(&dev_base_lock);
1003 write_lock_bh(&dev_base_lock);
1004 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1005 write_unlock_bh(&dev_base_lock);
1007 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1008 ret = notifier_to_errno(ret);
1011 /* err >= 0 after dev_alloc_name() or stores the first errno */
1014 memcpy(dev->name, oldname, IFNAMSIZ);
1018 "%s: name change rollback failed: %d.\n",
1027 * dev_set_alias - change ifalias of a device
1029 * @alias: name up to IFALIASZ
1030 * @len: limit of bytes to copy from info
1032 * Set ifalias for a device,
1034 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1038 if (len >= IFALIASZ)
1043 kfree(dev->ifalias);
1044 dev->ifalias = NULL;
1049 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1053 strlcpy(dev->ifalias, alias, len+1);
1059 * netdev_features_change - device changes features
1060 * @dev: device to cause notification
1062 * Called to indicate a device has changed features.
1064 void netdev_features_change(struct net_device *dev)
1066 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1068 EXPORT_SYMBOL(netdev_features_change);
1071 * netdev_state_change - device changes state
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed state. This function calls
1075 * the notifier chains for netdev_chain and sends a NEWLINK message
1076 * to the routing socket.
1078 void netdev_state_change(struct net_device *dev)
1080 if (dev->flags & IFF_UP) {
1081 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1082 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1085 EXPORT_SYMBOL(netdev_state_change);
1087 void netdev_bonding_change(struct net_device *dev, unsigned long event)
1089 call_netdevice_notifiers(event, dev);
1091 EXPORT_SYMBOL(netdev_bonding_change);
1094 * dev_load - load a network module
1095 * @net: the applicable net namespace
1096 * @name: name of interface
1098 * If a network interface is not present and the process has suitable
1099 * privileges this function loads the module. If module loading is not
1100 * available in this kernel then it becomes a nop.
1103 void dev_load(struct net *net, const char *name)
1105 struct net_device *dev;
1108 dev = dev_get_by_name_rcu(net, name);
1111 if (!dev && capable(CAP_NET_ADMIN))
1112 request_module("%s", name);
1114 EXPORT_SYMBOL(dev_load);
1117 * dev_open - prepare an interface for use.
1118 * @dev: device to open
1120 * Takes a device from down to up state. The device's private open
1121 * function is invoked and then the multicast lists are loaded. Finally
1122 * the device is moved into the up state and a %NETDEV_UP message is
1123 * sent to the netdev notifier chain.
1125 * Calling this function on an active interface is a nop. On a failure
1126 * a negative errno code is returned.
1128 int dev_open(struct net_device *dev)
1130 const struct net_device_ops *ops = dev->netdev_ops;
1139 if (dev->flags & IFF_UP)
1143 * Is it even present?
1145 if (!netif_device_present(dev))
1148 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1149 ret = notifier_to_errno(ret);
1154 * Call device private open method
1156 set_bit(__LINK_STATE_START, &dev->state);
1158 if (ops->ndo_validate_addr)
1159 ret = ops->ndo_validate_addr(dev);
1161 if (!ret && ops->ndo_open)
1162 ret = ops->ndo_open(dev);
1165 * If it went open OK then:
1169 clear_bit(__LINK_STATE_START, &dev->state);
1174 dev->flags |= IFF_UP;
1179 net_dmaengine_get();
1182 * Initialize multicasting status
1184 dev_set_rx_mode(dev);
1187 * Wakeup transmit queue engine
1192 * ... and announce new interface.
1194 call_netdevice_notifiers(NETDEV_UP, dev);
1199 EXPORT_SYMBOL(dev_open);
1202 * dev_close - shutdown an interface.
1203 * @dev: device to shutdown
1205 * This function moves an active device into down state. A
1206 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1207 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1210 int dev_close(struct net_device *dev)
1212 const struct net_device_ops *ops = dev->netdev_ops;
1217 if (!(dev->flags & IFF_UP))
1221 * Tell people we are going down, so that they can
1222 * prepare to death, when device is still operating.
1224 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1226 clear_bit(__LINK_STATE_START, &dev->state);
1228 /* Synchronize to scheduled poll. We cannot touch poll list,
1229 * it can be even on different cpu. So just clear netif_running().
1231 * dev->stop() will invoke napi_disable() on all of it's
1232 * napi_struct instances on this device.
1234 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1236 dev_deactivate(dev);
1239 * Call the device specific close. This cannot fail.
1240 * Only if device is UP
1242 * We allow it to be called even after a DETACH hot-plug
1249 * Device is now down.
1252 dev->flags &= ~IFF_UP;
1255 * Tell people we are down
1257 call_netdevice_notifiers(NETDEV_DOWN, dev);
1262 net_dmaengine_put();
1266 EXPORT_SYMBOL(dev_close);
1270 * dev_disable_lro - disable Large Receive Offload on a device
1273 * Disable Large Receive Offload (LRO) on a net device. Must be
1274 * called under RTNL. This is needed if received packets may be
1275 * forwarded to another interface.
1277 void dev_disable_lro(struct net_device *dev)
1279 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1280 dev->ethtool_ops->set_flags) {
1281 u32 flags = dev->ethtool_ops->get_flags(dev);
1282 if (flags & ETH_FLAG_LRO) {
1283 flags &= ~ETH_FLAG_LRO;
1284 dev->ethtool_ops->set_flags(dev, flags);
1287 WARN_ON(dev->features & NETIF_F_LRO);
1289 EXPORT_SYMBOL(dev_disable_lro);
1292 static int dev_boot_phase = 1;
1295 * Device change register/unregister. These are not inline or static
1296 * as we export them to the world.
1300 * register_netdevice_notifier - register a network notifier block
1303 * Register a notifier to be called when network device events occur.
1304 * The notifier passed is linked into the kernel structures and must
1305 * not be reused until it has been unregistered. A negative errno code
1306 * is returned on a failure.
1308 * When registered all registration and up events are replayed
1309 * to the new notifier to allow device to have a race free
1310 * view of the network device list.
1313 int register_netdevice_notifier(struct notifier_block *nb)
1315 struct net_device *dev;
1316 struct net_device *last;
1321 err = raw_notifier_chain_register(&netdev_chain, nb);
1327 for_each_netdev(net, dev) {
1328 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1329 err = notifier_to_errno(err);
1333 if (!(dev->flags & IFF_UP))
1336 nb->notifier_call(nb, NETDEV_UP, dev);
1347 for_each_netdev(net, dev) {
1351 if (dev->flags & IFF_UP) {
1352 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1353 nb->notifier_call(nb, NETDEV_DOWN, dev);
1355 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1356 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1360 raw_notifier_chain_unregister(&netdev_chain, nb);
1363 EXPORT_SYMBOL(register_netdevice_notifier);
1366 * unregister_netdevice_notifier - unregister a network notifier block
1369 * Unregister a notifier previously registered by
1370 * register_netdevice_notifier(). The notifier is unlinked into the
1371 * kernel structures and may then be reused. A negative errno code
1372 * is returned on a failure.
1375 int unregister_netdevice_notifier(struct notifier_block *nb)
1380 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1384 EXPORT_SYMBOL(unregister_netdevice_notifier);
1387 * call_netdevice_notifiers - call all network notifier blocks
1388 * @val: value passed unmodified to notifier function
1389 * @dev: net_device pointer passed unmodified to notifier function
1391 * Call all network notifier blocks. Parameters and return value
1392 * are as for raw_notifier_call_chain().
1395 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1397 return raw_notifier_call_chain(&netdev_chain, val, dev);
1400 /* When > 0 there are consumers of rx skb time stamps */
1401 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1403 void net_enable_timestamp(void)
1405 atomic_inc(&netstamp_needed);
1407 EXPORT_SYMBOL(net_enable_timestamp);
1409 void net_disable_timestamp(void)
1411 atomic_dec(&netstamp_needed);
1413 EXPORT_SYMBOL(net_disable_timestamp);
1415 static inline void net_timestamp(struct sk_buff *skb)
1417 if (atomic_read(&netstamp_needed))
1418 __net_timestamp(skb);
1420 skb->tstamp.tv64 = 0;
1424 * dev_forward_skb - loopback an skb to another netif
1426 * @dev: destination network device
1427 * @skb: buffer to forward
1430 * NET_RX_SUCCESS (no congestion)
1431 * NET_RX_DROP (packet was dropped)
1433 * dev_forward_skb can be used for injecting an skb from the
1434 * start_xmit function of one device into the receive queue
1435 * of another device.
1437 * The receiving device may be in another namespace, so
1438 * we have to clear all information in the skb that could
1439 * impact namespace isolation.
1441 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1445 if (!(dev->flags & IFF_UP))
1448 if (skb->len > (dev->mtu + dev->hard_header_len))
1452 skb->tstamp.tv64 = 0;
1453 skb->pkt_type = PACKET_HOST;
1454 skb->protocol = eth_type_trans(skb, dev);
1458 return netif_rx(skb);
1460 EXPORT_SYMBOL_GPL(dev_forward_skb);
1463 * Support routine. Sends outgoing frames to any network
1464 * taps currently in use.
1467 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1469 struct packet_type *ptype;
1471 #ifdef CONFIG_NET_CLS_ACT
1472 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1479 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1480 /* Never send packets back to the socket
1481 * they originated from - MvS (miquels@drinkel.ow.org)
1483 if ((ptype->dev == dev || !ptype->dev) &&
1484 (ptype->af_packet_priv == NULL ||
1485 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1486 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1490 /* skb->nh should be correctly
1491 set by sender, so that the second statement is
1492 just protection against buggy protocols.
1494 skb_reset_mac_header(skb2);
1496 if (skb_network_header(skb2) < skb2->data ||
1497 skb2->network_header > skb2->tail) {
1498 if (net_ratelimit())
1499 printk(KERN_CRIT "protocol %04x is "
1501 skb2->protocol, dev->name);
1502 skb_reset_network_header(skb2);
1505 skb2->transport_header = skb2->network_header;
1506 skb2->pkt_type = PACKET_OUTGOING;
1507 ptype->func(skb2, skb->dev, ptype, skb->dev);
1514 static inline void __netif_reschedule(struct Qdisc *q)
1516 struct softnet_data *sd;
1517 unsigned long flags;
1519 local_irq_save(flags);
1520 sd = &__get_cpu_var(softnet_data);
1521 q->next_sched = sd->output_queue;
1522 sd->output_queue = q;
1523 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1524 local_irq_restore(flags);
1527 void __netif_schedule(struct Qdisc *q)
1529 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1530 __netif_reschedule(q);
1532 EXPORT_SYMBOL(__netif_schedule);
1534 void dev_kfree_skb_irq(struct sk_buff *skb)
1536 if (atomic_dec_and_test(&skb->users)) {
1537 struct softnet_data *sd;
1538 unsigned long flags;
1540 local_irq_save(flags);
1541 sd = &__get_cpu_var(softnet_data);
1542 skb->next = sd->completion_queue;
1543 sd->completion_queue = skb;
1544 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1545 local_irq_restore(flags);
1548 EXPORT_SYMBOL(dev_kfree_skb_irq);
1550 void dev_kfree_skb_any(struct sk_buff *skb)
1552 if (in_irq() || irqs_disabled())
1553 dev_kfree_skb_irq(skb);
1557 EXPORT_SYMBOL(dev_kfree_skb_any);
1561 * netif_device_detach - mark device as removed
1562 * @dev: network device
1564 * Mark device as removed from system and therefore no longer available.
1566 void netif_device_detach(struct net_device *dev)
1568 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1569 netif_running(dev)) {
1570 netif_tx_stop_all_queues(dev);
1573 EXPORT_SYMBOL(netif_device_detach);
1576 * netif_device_attach - mark device as attached
1577 * @dev: network device
1579 * Mark device as attached from system and restart if needed.
1581 void netif_device_attach(struct net_device *dev)
1583 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1584 netif_running(dev)) {
1585 netif_tx_wake_all_queues(dev);
1586 __netdev_watchdog_up(dev);
1589 EXPORT_SYMBOL(netif_device_attach);
1591 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1593 return ((features & NETIF_F_GEN_CSUM) ||
1594 ((features & NETIF_F_IP_CSUM) &&
1595 protocol == htons(ETH_P_IP)) ||
1596 ((features & NETIF_F_IPV6_CSUM) &&
1597 protocol == htons(ETH_P_IPV6)) ||
1598 ((features & NETIF_F_FCOE_CRC) &&
1599 protocol == htons(ETH_P_FCOE)));
1602 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1604 if (can_checksum_protocol(dev->features, skb->protocol))
1607 if (skb->protocol == htons(ETH_P_8021Q)) {
1608 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1609 if (can_checksum_protocol(dev->features & dev->vlan_features,
1610 veh->h_vlan_encapsulated_proto))
1618 * Invalidate hardware checksum when packet is to be mangled, and
1619 * complete checksum manually on outgoing path.
1621 int skb_checksum_help(struct sk_buff *skb)
1624 int ret = 0, offset;
1626 if (skb->ip_summed == CHECKSUM_COMPLETE)
1627 goto out_set_summed;
1629 if (unlikely(skb_shinfo(skb)->gso_size)) {
1630 /* Let GSO fix up the checksum. */
1631 goto out_set_summed;
1634 offset = skb->csum_start - skb_headroom(skb);
1635 BUG_ON(offset >= skb_headlen(skb));
1636 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1638 offset += skb->csum_offset;
1639 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1641 if (skb_cloned(skb) &&
1642 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1643 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1648 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1650 skb->ip_summed = CHECKSUM_NONE;
1654 EXPORT_SYMBOL(skb_checksum_help);
1657 * skb_gso_segment - Perform segmentation on skb.
1658 * @skb: buffer to segment
1659 * @features: features for the output path (see dev->features)
1661 * This function segments the given skb and returns a list of segments.
1663 * It may return NULL if the skb requires no segmentation. This is
1664 * only possible when GSO is used for verifying header integrity.
1666 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1668 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1669 struct packet_type *ptype;
1670 __be16 type = skb->protocol;
1673 skb_reset_mac_header(skb);
1674 skb->mac_len = skb->network_header - skb->mac_header;
1675 __skb_pull(skb, skb->mac_len);
1677 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1678 struct net_device *dev = skb->dev;
1679 struct ethtool_drvinfo info = {};
1681 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1682 dev->ethtool_ops->get_drvinfo(dev, &info);
1684 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1686 info.driver, dev ? dev->features : 0L,
1687 skb->sk ? skb->sk->sk_route_caps : 0L,
1688 skb->len, skb->data_len, skb->ip_summed);
1690 if (skb_header_cloned(skb) &&
1691 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1692 return ERR_PTR(err);
1696 list_for_each_entry_rcu(ptype,
1697 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1698 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1699 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1700 err = ptype->gso_send_check(skb);
1701 segs = ERR_PTR(err);
1702 if (err || skb_gso_ok(skb, features))
1704 __skb_push(skb, (skb->data -
1705 skb_network_header(skb)));
1707 segs = ptype->gso_segment(skb, features);
1713 __skb_push(skb, skb->data - skb_mac_header(skb));
1717 EXPORT_SYMBOL(skb_gso_segment);
1719 /* Take action when hardware reception checksum errors are detected. */
1721 void netdev_rx_csum_fault(struct net_device *dev)
1723 if (net_ratelimit()) {
1724 printk(KERN_ERR "%s: hw csum failure.\n",
1725 dev ? dev->name : "<unknown>");
1729 EXPORT_SYMBOL(netdev_rx_csum_fault);
1732 /* Actually, we should eliminate this check as soon as we know, that:
1733 * 1. IOMMU is present and allows to map all the memory.
1734 * 2. No high memory really exists on this machine.
1737 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1739 #ifdef CONFIG_HIGHMEM
1742 if (dev->features & NETIF_F_HIGHDMA)
1745 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1746 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1754 void (*destructor)(struct sk_buff *skb);
1757 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1759 static void dev_gso_skb_destructor(struct sk_buff *skb)
1761 struct dev_gso_cb *cb;
1764 struct sk_buff *nskb = skb->next;
1766 skb->next = nskb->next;
1769 } while (skb->next);
1771 cb = DEV_GSO_CB(skb);
1773 cb->destructor(skb);
1777 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1778 * @skb: buffer to segment
1780 * This function segments the given skb and stores the list of segments
1783 static int dev_gso_segment(struct sk_buff *skb)
1785 struct net_device *dev = skb->dev;
1786 struct sk_buff *segs;
1787 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1790 segs = skb_gso_segment(skb, features);
1792 /* Verifying header integrity only. */
1797 return PTR_ERR(segs);
1800 DEV_GSO_CB(skb)->destructor = skb->destructor;
1801 skb->destructor = dev_gso_skb_destructor;
1806 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1807 struct netdev_queue *txq)
1809 const struct net_device_ops *ops = dev->netdev_ops;
1810 int rc = NETDEV_TX_OK;
1812 if (likely(!skb->next)) {
1813 if (!list_empty(&ptype_all))
1814 dev_queue_xmit_nit(skb, dev);
1816 if (netif_needs_gso(dev, skb)) {
1817 if (unlikely(dev_gso_segment(skb)))
1824 * If device doesnt need skb->dst, release it right now while
1825 * its hot in this cpu cache
1827 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1830 rc = ops->ndo_start_xmit(skb, dev);
1831 if (rc == NETDEV_TX_OK)
1832 txq_trans_update(txq);
1834 * TODO: if skb_orphan() was called by
1835 * dev->hard_start_xmit() (for example, the unmodified
1836 * igb driver does that; bnx2 doesn't), then
1837 * skb_tx_software_timestamp() will be unable to send
1838 * back the time stamp.
1840 * How can this be prevented? Always create another
1841 * reference to the socket before calling
1842 * dev->hard_start_xmit()? Prevent that skb_orphan()
1843 * does anything in dev->hard_start_xmit() by clearing
1844 * the skb destructor before the call and restoring it
1845 * afterwards, then doing the skb_orphan() ourselves?
1852 struct sk_buff *nskb = skb->next;
1854 skb->next = nskb->next;
1858 * If device doesnt need nskb->dst, release it right now while
1859 * its hot in this cpu cache
1861 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1864 rc = ops->ndo_start_xmit(nskb, dev);
1865 if (unlikely(rc != NETDEV_TX_OK)) {
1866 if (rc & ~NETDEV_TX_MASK)
1867 goto out_kfree_gso_skb;
1868 nskb->next = skb->next;
1872 txq_trans_update(txq);
1873 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1874 return NETDEV_TX_BUSY;
1875 } while (skb->next);
1878 if (likely(skb->next == NULL))
1879 skb->destructor = DEV_GSO_CB(skb)->destructor;
1885 static u32 skb_tx_hashrnd;
1887 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1891 if (skb_rx_queue_recorded(skb)) {
1892 hash = skb_get_rx_queue(skb);
1893 while (unlikely(hash >= dev->real_num_tx_queues))
1894 hash -= dev->real_num_tx_queues;
1898 if (skb->sk && skb->sk->sk_hash)
1899 hash = skb->sk->sk_hash;
1901 hash = skb->protocol;
1903 hash = jhash_1word(hash, skb_tx_hashrnd);
1905 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1907 EXPORT_SYMBOL(skb_tx_hash);
1909 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1911 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1912 if (net_ratelimit()) {
1913 WARN(1, "%s selects TX queue %d, but "
1914 "real number of TX queues is %d\n",
1915 dev->name, queue_index,
1916 dev->real_num_tx_queues);
1923 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1924 struct sk_buff *skb)
1927 struct sock *sk = skb->sk;
1929 if (sk_tx_queue_recorded(sk)) {
1930 queue_index = sk_tx_queue_get(sk);
1932 const struct net_device_ops *ops = dev->netdev_ops;
1934 if (ops->ndo_select_queue) {
1935 queue_index = ops->ndo_select_queue(dev, skb);
1936 queue_index = dev_cap_txqueue(dev, queue_index);
1939 if (dev->real_num_tx_queues > 1)
1940 queue_index = skb_tx_hash(dev, skb);
1942 if (sk && sk->sk_dst_cache)
1943 sk_tx_queue_set(sk, queue_index);
1947 skb_set_queue_mapping(skb, queue_index);
1948 return netdev_get_tx_queue(dev, queue_index);
1951 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1952 struct net_device *dev,
1953 struct netdev_queue *txq)
1955 spinlock_t *root_lock = qdisc_lock(q);
1958 spin_lock(root_lock);
1959 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1962 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1963 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1965 * This is a work-conserving queue; there are no old skbs
1966 * waiting to be sent out; and the qdisc is not running -
1967 * xmit the skb directly.
1969 __qdisc_update_bstats(q, skb->len);
1970 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1973 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1975 rc = NET_XMIT_SUCCESS;
1977 rc = qdisc_enqueue_root(skb, q);
1980 spin_unlock(root_lock);
1986 * dev_queue_xmit - transmit a buffer
1987 * @skb: buffer to transmit
1989 * Queue a buffer for transmission to a network device. The caller must
1990 * have set the device and priority and built the buffer before calling
1991 * this function. The function can be called from an interrupt.
1993 * A negative errno code is returned on a failure. A success does not
1994 * guarantee the frame will be transmitted as it may be dropped due
1995 * to congestion or traffic shaping.
1997 * -----------------------------------------------------------------------------------
1998 * I notice this method can also return errors from the queue disciplines,
1999 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2002 * Regardless of the return value, the skb is consumed, so it is currently
2003 * difficult to retry a send to this method. (You can bump the ref count
2004 * before sending to hold a reference for retry if you are careful.)
2006 * When calling this method, interrupts MUST be enabled. This is because
2007 * the BH enable code must have IRQs enabled so that it will not deadlock.
2010 int dev_queue_xmit(struct sk_buff *skb)
2012 struct net_device *dev = skb->dev;
2013 struct netdev_queue *txq;
2017 /* GSO will handle the following emulations directly. */
2018 if (netif_needs_gso(dev, skb))
2021 if (skb_has_frags(skb) &&
2022 !(dev->features & NETIF_F_FRAGLIST) &&
2023 __skb_linearize(skb))
2026 /* Fragmented skb is linearized if device does not support SG,
2027 * or if at least one of fragments is in highmem and device
2028 * does not support DMA from it.
2030 if (skb_shinfo(skb)->nr_frags &&
2031 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
2032 __skb_linearize(skb))
2035 /* If packet is not checksummed and device does not support
2036 * checksumming for this protocol, complete checksumming here.
2038 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2039 skb_set_transport_header(skb, skb->csum_start -
2041 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2046 /* Disable soft irqs for various locks below. Also
2047 * stops preemption for RCU.
2051 txq = dev_pick_tx(dev, skb);
2052 q = rcu_dereference(txq->qdisc);
2054 #ifdef CONFIG_NET_CLS_ACT
2055 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2058 rc = __dev_xmit_skb(skb, q, dev, txq);
2062 /* The device has no queue. Common case for software devices:
2063 loopback, all the sorts of tunnels...
2065 Really, it is unlikely that netif_tx_lock protection is necessary
2066 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2068 However, it is possible, that they rely on protection
2071 Check this and shot the lock. It is not prone from deadlocks.
2072 Either shot noqueue qdisc, it is even simpler 8)
2074 if (dev->flags & IFF_UP) {
2075 int cpu = smp_processor_id(); /* ok because BHs are off */
2077 if (txq->xmit_lock_owner != cpu) {
2079 HARD_TX_LOCK(dev, txq, cpu);
2081 if (!netif_tx_queue_stopped(txq)) {
2082 rc = dev_hard_start_xmit(skb, dev, txq);
2083 if (dev_xmit_complete(rc)) {
2084 HARD_TX_UNLOCK(dev, txq);
2088 HARD_TX_UNLOCK(dev, txq);
2089 if (net_ratelimit())
2090 printk(KERN_CRIT "Virtual device %s asks to "
2091 "queue packet!\n", dev->name);
2093 /* Recursion is detected! It is possible,
2095 if (net_ratelimit())
2096 printk(KERN_CRIT "Dead loop on virtual device "
2097 "%s, fix it urgently!\n", dev->name);
2102 rcu_read_unlock_bh();
2108 rcu_read_unlock_bh();
2111 EXPORT_SYMBOL(dev_queue_xmit);
2114 /*=======================================================================
2116 =======================================================================*/
2118 int netdev_max_backlog __read_mostly = 1000;
2119 int netdev_budget __read_mostly = 300;
2120 int weight_p __read_mostly = 64; /* old backlog weight */
2122 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2126 * netif_rx - post buffer to the network code
2127 * @skb: buffer to post
2129 * This function receives a packet from a device driver and queues it for
2130 * the upper (protocol) levels to process. It always succeeds. The buffer
2131 * may be dropped during processing for congestion control or by the
2135 * NET_RX_SUCCESS (no congestion)
2136 * NET_RX_DROP (packet was dropped)
2140 int netif_rx(struct sk_buff *skb)
2142 struct softnet_data *queue;
2143 unsigned long flags;
2145 /* if netpoll wants it, pretend we never saw it */
2146 if (netpoll_rx(skb))
2149 if (!skb->tstamp.tv64)
2153 * The code is rearranged so that the path is the most
2154 * short when CPU is congested, but is still operating.
2156 local_irq_save(flags);
2157 queue = &__get_cpu_var(softnet_data);
2159 __get_cpu_var(netdev_rx_stat).total++;
2160 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2161 if (queue->input_pkt_queue.qlen) {
2163 __skb_queue_tail(&queue->input_pkt_queue, skb);
2164 local_irq_restore(flags);
2165 return NET_RX_SUCCESS;
2168 napi_schedule(&queue->backlog);
2172 __get_cpu_var(netdev_rx_stat).dropped++;
2173 local_irq_restore(flags);
2178 EXPORT_SYMBOL(netif_rx);
2180 int netif_rx_ni(struct sk_buff *skb)
2185 err = netif_rx(skb);
2186 if (local_softirq_pending())
2192 EXPORT_SYMBOL(netif_rx_ni);
2194 static void net_tx_action(struct softirq_action *h)
2196 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2198 if (sd->completion_queue) {
2199 struct sk_buff *clist;
2201 local_irq_disable();
2202 clist = sd->completion_queue;
2203 sd->completion_queue = NULL;
2207 struct sk_buff *skb = clist;
2208 clist = clist->next;
2210 WARN_ON(atomic_read(&skb->users));
2215 if (sd->output_queue) {
2218 local_irq_disable();
2219 head = sd->output_queue;
2220 sd->output_queue = NULL;
2224 struct Qdisc *q = head;
2225 spinlock_t *root_lock;
2227 head = head->next_sched;
2229 root_lock = qdisc_lock(q);
2230 if (spin_trylock(root_lock)) {
2231 smp_mb__before_clear_bit();
2232 clear_bit(__QDISC_STATE_SCHED,
2235 spin_unlock(root_lock);
2237 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2239 __netif_reschedule(q);
2241 smp_mb__before_clear_bit();
2242 clear_bit(__QDISC_STATE_SCHED,
2250 static inline int deliver_skb(struct sk_buff *skb,
2251 struct packet_type *pt_prev,
2252 struct net_device *orig_dev)
2254 atomic_inc(&skb->users);
2255 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2258 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2260 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2261 /* This hook is defined here for ATM LANE */
2262 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2263 unsigned char *addr) __read_mostly;
2264 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2268 * If bridge module is loaded call bridging hook.
2269 * returns NULL if packet was consumed.
2271 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2272 struct sk_buff *skb) __read_mostly;
2273 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2275 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2276 struct packet_type **pt_prev, int *ret,
2277 struct net_device *orig_dev)
2279 struct net_bridge_port *port;
2281 if (skb->pkt_type == PACKET_LOOPBACK ||
2282 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2286 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2290 return br_handle_frame_hook(port, skb);
2293 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2296 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2297 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2298 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2300 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2301 struct packet_type **pt_prev,
2303 struct net_device *orig_dev)
2305 if (skb->dev->macvlan_port == NULL)
2309 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2312 return macvlan_handle_frame_hook(skb);
2315 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2318 #ifdef CONFIG_NET_CLS_ACT
2319 /* TODO: Maybe we should just force sch_ingress to be compiled in
2320 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2321 * a compare and 2 stores extra right now if we dont have it on
2322 * but have CONFIG_NET_CLS_ACT
2323 * NOTE: This doesnt stop any functionality; if you dont have
2324 * the ingress scheduler, you just cant add policies on ingress.
2327 static int ing_filter(struct sk_buff *skb)
2329 struct net_device *dev = skb->dev;
2330 u32 ttl = G_TC_RTTL(skb->tc_verd);
2331 struct netdev_queue *rxq;
2332 int result = TC_ACT_OK;
2335 if (MAX_RED_LOOP < ttl++) {
2337 "Redir loop detected Dropping packet (%d->%d)\n",
2338 skb->skb_iif, dev->ifindex);
2342 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2343 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2345 rxq = &dev->rx_queue;
2348 if (q != &noop_qdisc) {
2349 spin_lock(qdisc_lock(q));
2350 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2351 result = qdisc_enqueue_root(skb, q);
2352 spin_unlock(qdisc_lock(q));
2358 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2359 struct packet_type **pt_prev,
2360 int *ret, struct net_device *orig_dev)
2362 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2366 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2369 /* Huh? Why does turning on AF_PACKET affect this? */
2370 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2373 switch (ing_filter(skb)) {
2387 * netif_nit_deliver - deliver received packets to network taps
2390 * This function is used to deliver incoming packets to network
2391 * taps. It should be used when the normal netif_receive_skb path
2392 * is bypassed, for example because of VLAN acceleration.
2394 void netif_nit_deliver(struct sk_buff *skb)
2396 struct packet_type *ptype;
2398 if (list_empty(&ptype_all))
2401 skb_reset_network_header(skb);
2402 skb_reset_transport_header(skb);
2403 skb->mac_len = skb->network_header - skb->mac_header;
2406 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2407 if (!ptype->dev || ptype->dev == skb->dev)
2408 deliver_skb(skb, ptype, skb->dev);
2414 * netif_receive_skb - process receive buffer from network
2415 * @skb: buffer to process
2417 * netif_receive_skb() is the main receive data processing function.
2418 * It always succeeds. The buffer may be dropped during processing
2419 * for congestion control or by the protocol layers.
2421 * This function may only be called from softirq context and interrupts
2422 * should be enabled.
2424 * Return values (usually ignored):
2425 * NET_RX_SUCCESS: no congestion
2426 * NET_RX_DROP: packet was dropped
2428 int netif_receive_skb(struct sk_buff *skb)
2430 struct packet_type *ptype, *pt_prev;
2431 struct net_device *orig_dev;
2432 struct net_device *null_or_orig;
2433 struct net_device *null_or_bond;
2434 int ret = NET_RX_DROP;
2437 if (!skb->tstamp.tv64)
2440 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2441 return NET_RX_SUCCESS;
2443 /* if we've gotten here through NAPI, check netpoll */
2444 if (netpoll_receive_skb(skb))
2448 skb->skb_iif = skb->dev->ifindex;
2450 null_or_orig = NULL;
2451 orig_dev = skb->dev;
2452 if (orig_dev->master) {
2453 if (skb_bond_should_drop(skb))
2454 null_or_orig = orig_dev; /* deliver only exact match */
2456 skb->dev = orig_dev->master;
2459 __get_cpu_var(netdev_rx_stat).total++;
2461 skb_reset_network_header(skb);
2462 skb_reset_transport_header(skb);
2463 skb->mac_len = skb->network_header - skb->mac_header;
2469 #ifdef CONFIG_NET_CLS_ACT
2470 if (skb->tc_verd & TC_NCLS) {
2471 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2476 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2477 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2478 ptype->dev == orig_dev) {
2480 ret = deliver_skb(skb, pt_prev, orig_dev);
2485 #ifdef CONFIG_NET_CLS_ACT
2486 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2492 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2495 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2500 * Make sure frames received on VLAN interfaces stacked on
2501 * bonding interfaces still make their way to any base bonding
2502 * device that may have registered for a specific ptype. The
2503 * handler may have to adjust skb->dev and orig_dev.
2505 null_or_bond = NULL;
2506 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2507 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2508 null_or_bond = vlan_dev_real_dev(skb->dev);
2511 type = skb->protocol;
2512 list_for_each_entry_rcu(ptype,
2513 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2514 if (ptype->type == type && (ptype->dev == null_or_orig ||
2515 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2516 ptype->dev == null_or_bond)) {
2518 ret = deliver_skb(skb, pt_prev, orig_dev);
2524 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2527 /* Jamal, now you will not able to escape explaining
2528 * me how you were going to use this. :-)
2537 EXPORT_SYMBOL(netif_receive_skb);
2539 /* Network device is going away, flush any packets still pending */
2540 static void flush_backlog(void *arg)
2542 struct net_device *dev = arg;
2543 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2544 struct sk_buff *skb, *tmp;
2546 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2547 if (skb->dev == dev) {
2548 __skb_unlink(skb, &queue->input_pkt_queue);
2553 static int napi_gro_complete(struct sk_buff *skb)
2555 struct packet_type *ptype;
2556 __be16 type = skb->protocol;
2557 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2560 if (NAPI_GRO_CB(skb)->count == 1) {
2561 skb_shinfo(skb)->gso_size = 0;
2566 list_for_each_entry_rcu(ptype, head, list) {
2567 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2570 err = ptype->gro_complete(skb);
2576 WARN_ON(&ptype->list == head);
2578 return NET_RX_SUCCESS;
2582 return netif_receive_skb(skb);
2585 void napi_gro_flush(struct napi_struct *napi)
2587 struct sk_buff *skb, *next;
2589 for (skb = napi->gro_list; skb; skb = next) {
2592 napi_gro_complete(skb);
2595 napi->gro_count = 0;
2596 napi->gro_list = NULL;
2598 EXPORT_SYMBOL(napi_gro_flush);
2600 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2602 struct sk_buff **pp = NULL;
2603 struct packet_type *ptype;
2604 __be16 type = skb->protocol;
2605 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2608 enum gro_result ret;
2610 if (!(skb->dev->features & NETIF_F_GRO))
2613 if (skb_is_gso(skb) || skb_has_frags(skb))
2617 list_for_each_entry_rcu(ptype, head, list) {
2618 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2621 skb_set_network_header(skb, skb_gro_offset(skb));
2622 mac_len = skb->network_header - skb->mac_header;
2623 skb->mac_len = mac_len;
2624 NAPI_GRO_CB(skb)->same_flow = 0;
2625 NAPI_GRO_CB(skb)->flush = 0;
2626 NAPI_GRO_CB(skb)->free = 0;
2628 pp = ptype->gro_receive(&napi->gro_list, skb);
2633 if (&ptype->list == head)
2636 same_flow = NAPI_GRO_CB(skb)->same_flow;
2637 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2640 struct sk_buff *nskb = *pp;
2644 napi_gro_complete(nskb);
2651 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2655 NAPI_GRO_CB(skb)->count = 1;
2656 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2657 skb->next = napi->gro_list;
2658 napi->gro_list = skb;
2662 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2663 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2665 BUG_ON(skb->end - skb->tail < grow);
2667 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2670 skb->data_len -= grow;
2672 skb_shinfo(skb)->frags[0].page_offset += grow;
2673 skb_shinfo(skb)->frags[0].size -= grow;
2675 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2676 put_page(skb_shinfo(skb)->frags[0].page);
2677 memmove(skb_shinfo(skb)->frags,
2678 skb_shinfo(skb)->frags + 1,
2679 --skb_shinfo(skb)->nr_frags);
2690 EXPORT_SYMBOL(dev_gro_receive);
2693 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2697 if (netpoll_rx_on(skb))
2700 for (p = napi->gro_list; p; p = p->next) {
2701 NAPI_GRO_CB(p)->same_flow =
2702 (p->dev == skb->dev) &&
2703 !compare_ether_header(skb_mac_header(p),
2704 skb_gro_mac_header(skb));
2705 NAPI_GRO_CB(p)->flush = 0;
2708 return dev_gro_receive(napi, skb);
2711 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2715 if (netif_receive_skb(skb))
2720 case GRO_MERGED_FREE:
2731 EXPORT_SYMBOL(napi_skb_finish);
2733 void skb_gro_reset_offset(struct sk_buff *skb)
2735 NAPI_GRO_CB(skb)->data_offset = 0;
2736 NAPI_GRO_CB(skb)->frag0 = NULL;
2737 NAPI_GRO_CB(skb)->frag0_len = 0;
2739 if (skb->mac_header == skb->tail &&
2740 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2741 NAPI_GRO_CB(skb)->frag0 =
2742 page_address(skb_shinfo(skb)->frags[0].page) +
2743 skb_shinfo(skb)->frags[0].page_offset;
2744 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2747 EXPORT_SYMBOL(skb_gro_reset_offset);
2749 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2751 skb_gro_reset_offset(skb);
2753 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2755 EXPORT_SYMBOL(napi_gro_receive);
2757 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2759 __skb_pull(skb, skb_headlen(skb));
2760 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2764 EXPORT_SYMBOL(napi_reuse_skb);
2766 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2768 struct sk_buff *skb = napi->skb;
2771 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2777 EXPORT_SYMBOL(napi_get_frags);
2779 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
2785 skb->protocol = eth_type_trans(skb, napi->dev);
2787 if (ret == GRO_HELD)
2788 skb_gro_pull(skb, -ETH_HLEN);
2789 else if (netif_receive_skb(skb))
2794 case GRO_MERGED_FREE:
2795 napi_reuse_skb(napi, skb);
2804 EXPORT_SYMBOL(napi_frags_finish);
2806 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2808 struct sk_buff *skb = napi->skb;
2815 skb_reset_mac_header(skb);
2816 skb_gro_reset_offset(skb);
2818 off = skb_gro_offset(skb);
2819 hlen = off + sizeof(*eth);
2820 eth = skb_gro_header_fast(skb, off);
2821 if (skb_gro_header_hard(skb, hlen)) {
2822 eth = skb_gro_header_slow(skb, hlen, off);
2823 if (unlikely(!eth)) {
2824 napi_reuse_skb(napi, skb);
2830 skb_gro_pull(skb, sizeof(*eth));
2833 * This works because the only protocols we care about don't require
2834 * special handling. We'll fix it up properly at the end.
2836 skb->protocol = eth->h_proto;
2841 EXPORT_SYMBOL(napi_frags_skb);
2843 gro_result_t napi_gro_frags(struct napi_struct *napi)
2845 struct sk_buff *skb = napi_frags_skb(napi);
2850 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2852 EXPORT_SYMBOL(napi_gro_frags);
2854 static int process_backlog(struct napi_struct *napi, int quota)
2857 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2858 unsigned long start_time = jiffies;
2860 napi->weight = weight_p;
2862 struct sk_buff *skb;
2864 local_irq_disable();
2865 skb = __skb_dequeue(&queue->input_pkt_queue);
2867 __napi_complete(napi);
2873 netif_receive_skb(skb);
2874 } while (++work < quota && jiffies == start_time);
2880 * __napi_schedule - schedule for receive
2881 * @n: entry to schedule
2883 * The entry's receive function will be scheduled to run
2885 void __napi_schedule(struct napi_struct *n)
2887 unsigned long flags;
2889 local_irq_save(flags);
2890 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2891 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2892 local_irq_restore(flags);
2894 EXPORT_SYMBOL(__napi_schedule);
2896 void __napi_complete(struct napi_struct *n)
2898 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2899 BUG_ON(n->gro_list);
2901 list_del(&n->poll_list);
2902 smp_mb__before_clear_bit();
2903 clear_bit(NAPI_STATE_SCHED, &n->state);
2905 EXPORT_SYMBOL(__napi_complete);
2907 void napi_complete(struct napi_struct *n)
2909 unsigned long flags;
2912 * don't let napi dequeue from the cpu poll list
2913 * just in case its running on a different cpu
2915 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2919 local_irq_save(flags);
2921 local_irq_restore(flags);
2923 EXPORT_SYMBOL(napi_complete);
2925 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2926 int (*poll)(struct napi_struct *, int), int weight)
2928 INIT_LIST_HEAD(&napi->poll_list);
2929 napi->gro_count = 0;
2930 napi->gro_list = NULL;
2933 napi->weight = weight;
2934 list_add(&napi->dev_list, &dev->napi_list);
2936 #ifdef CONFIG_NETPOLL
2937 spin_lock_init(&napi->poll_lock);
2938 napi->poll_owner = -1;
2940 set_bit(NAPI_STATE_SCHED, &napi->state);
2942 EXPORT_SYMBOL(netif_napi_add);
2944 void netif_napi_del(struct napi_struct *napi)
2946 struct sk_buff *skb, *next;
2948 list_del_init(&napi->dev_list);
2949 napi_free_frags(napi);
2951 for (skb = napi->gro_list; skb; skb = next) {
2957 napi->gro_list = NULL;
2958 napi->gro_count = 0;
2960 EXPORT_SYMBOL(netif_napi_del);
2963 static void net_rx_action(struct softirq_action *h)
2965 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2966 unsigned long time_limit = jiffies + 2;
2967 int budget = netdev_budget;
2970 local_irq_disable();
2972 while (!list_empty(list)) {
2973 struct napi_struct *n;
2976 /* If softirq window is exhuasted then punt.
2977 * Allow this to run for 2 jiffies since which will allow
2978 * an average latency of 1.5/HZ.
2980 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2985 /* Even though interrupts have been re-enabled, this
2986 * access is safe because interrupts can only add new
2987 * entries to the tail of this list, and only ->poll()
2988 * calls can remove this head entry from the list.
2990 n = list_entry(list->next, struct napi_struct, poll_list);
2992 have = netpoll_poll_lock(n);
2996 /* This NAPI_STATE_SCHED test is for avoiding a race
2997 * with netpoll's poll_napi(). Only the entity which
2998 * obtains the lock and sees NAPI_STATE_SCHED set will
2999 * actually make the ->poll() call. Therefore we avoid
3000 * accidently calling ->poll() when NAPI is not scheduled.
3003 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3004 work = n->poll(n, weight);
3008 WARN_ON_ONCE(work > weight);
3012 local_irq_disable();
3014 /* Drivers must not modify the NAPI state if they
3015 * consume the entire weight. In such cases this code
3016 * still "owns" the NAPI instance and therefore can
3017 * move the instance around on the list at-will.
3019 if (unlikely(work == weight)) {
3020 if (unlikely(napi_disable_pending(n))) {
3023 local_irq_disable();
3025 list_move_tail(&n->poll_list, list);
3028 netpoll_poll_unlock(have);
3033 #ifdef CONFIG_NET_DMA
3035 * There may not be any more sk_buffs coming right now, so push
3036 * any pending DMA copies to hardware
3038 dma_issue_pending_all();
3044 __get_cpu_var(netdev_rx_stat).time_squeeze++;
3045 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3049 static gifconf_func_t *gifconf_list[NPROTO];
3052 * register_gifconf - register a SIOCGIF handler
3053 * @family: Address family
3054 * @gifconf: Function handler
3056 * Register protocol dependent address dumping routines. The handler
3057 * that is passed must not be freed or reused until it has been replaced
3058 * by another handler.
3060 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3062 if (family >= NPROTO)
3064 gifconf_list[family] = gifconf;
3067 EXPORT_SYMBOL(register_gifconf);
3071 * Map an interface index to its name (SIOCGIFNAME)
3075 * We need this ioctl for efficient implementation of the
3076 * if_indextoname() function required by the IPv6 API. Without
3077 * it, we would have to search all the interfaces to find a
3081 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3083 struct net_device *dev;
3087 * Fetch the caller's info block.
3090 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3094 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3100 strcpy(ifr.ifr_name, dev->name);
3103 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3109 * Perform a SIOCGIFCONF call. This structure will change
3110 * size eventually, and there is nothing I can do about it.
3111 * Thus we will need a 'compatibility mode'.
3114 static int dev_ifconf(struct net *net, char __user *arg)
3117 struct net_device *dev;
3124 * Fetch the caller's info block.
3127 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3134 * Loop over the interfaces, and write an info block for each.
3138 for_each_netdev(net, dev) {
3139 for (i = 0; i < NPROTO; i++) {
3140 if (gifconf_list[i]) {
3143 done = gifconf_list[i](dev, NULL, 0);
3145 done = gifconf_list[i](dev, pos + total,
3155 * All done. Write the updated control block back to the caller.
3157 ifc.ifc_len = total;
3160 * Both BSD and Solaris return 0 here, so we do too.
3162 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3165 #ifdef CONFIG_PROC_FS
3167 * This is invoked by the /proc filesystem handler to display a device
3170 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3173 struct net *net = seq_file_net(seq);
3175 struct net_device *dev;
3179 return SEQ_START_TOKEN;
3182 for_each_netdev_rcu(net, dev)
3189 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3191 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3192 first_net_device(seq_file_net(seq)) :
3193 next_net_device((struct net_device *)v);
3196 return rcu_dereference(dev);
3199 void dev_seq_stop(struct seq_file *seq, void *v)
3205 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3207 const struct net_device_stats *stats = dev_get_stats(dev);
3209 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3210 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3211 dev->name, stats->rx_bytes, stats->rx_packets,
3213 stats->rx_dropped + stats->rx_missed_errors,
3214 stats->rx_fifo_errors,
3215 stats->rx_length_errors + stats->rx_over_errors +
3216 stats->rx_crc_errors + stats->rx_frame_errors,
3217 stats->rx_compressed, stats->multicast,
3218 stats->tx_bytes, stats->tx_packets,
3219 stats->tx_errors, stats->tx_dropped,
3220 stats->tx_fifo_errors, stats->collisions,
3221 stats->tx_carrier_errors +
3222 stats->tx_aborted_errors +
3223 stats->tx_window_errors +
3224 stats->tx_heartbeat_errors,
3225 stats->tx_compressed);
3229 * Called from the PROCfs module. This now uses the new arbitrary sized
3230 * /proc/net interface to create /proc/net/dev
3232 static int dev_seq_show(struct seq_file *seq, void *v)
3234 if (v == SEQ_START_TOKEN)
3235 seq_puts(seq, "Inter-| Receive "
3237 " face |bytes packets errs drop fifo frame "
3238 "compressed multicast|bytes packets errs "
3239 "drop fifo colls carrier compressed\n");
3241 dev_seq_printf_stats(seq, v);
3245 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3247 struct netif_rx_stats *rc = NULL;
3249 while (*pos < nr_cpu_ids)
3250 if (cpu_online(*pos)) {
3251 rc = &per_cpu(netdev_rx_stat, *pos);
3258 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3260 return softnet_get_online(pos);
3263 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3266 return softnet_get_online(pos);
3269 static void softnet_seq_stop(struct seq_file *seq, void *v)
3273 static int softnet_seq_show(struct seq_file *seq, void *v)
3275 struct netif_rx_stats *s = v;
3277 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3278 s->total, s->dropped, s->time_squeeze, 0,
3279 0, 0, 0, 0, /* was fastroute */
3284 static const struct seq_operations dev_seq_ops = {
3285 .start = dev_seq_start,
3286 .next = dev_seq_next,
3287 .stop = dev_seq_stop,
3288 .show = dev_seq_show,
3291 static int dev_seq_open(struct inode *inode, struct file *file)
3293 return seq_open_net(inode, file, &dev_seq_ops,
3294 sizeof(struct seq_net_private));
3297 static const struct file_operations dev_seq_fops = {
3298 .owner = THIS_MODULE,
3299 .open = dev_seq_open,
3301 .llseek = seq_lseek,
3302 .release = seq_release_net,
3305 static const struct seq_operations softnet_seq_ops = {
3306 .start = softnet_seq_start,
3307 .next = softnet_seq_next,
3308 .stop = softnet_seq_stop,
3309 .show = softnet_seq_show,
3312 static int softnet_seq_open(struct inode *inode, struct file *file)
3314 return seq_open(file, &softnet_seq_ops);
3317 static const struct file_operations softnet_seq_fops = {
3318 .owner = THIS_MODULE,
3319 .open = softnet_seq_open,
3321 .llseek = seq_lseek,
3322 .release = seq_release,
3325 static void *ptype_get_idx(loff_t pos)
3327 struct packet_type *pt = NULL;
3331 list_for_each_entry_rcu(pt, &ptype_all, list) {
3337 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3338 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3347 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3351 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3354 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3356 struct packet_type *pt;
3357 struct list_head *nxt;
3361 if (v == SEQ_START_TOKEN)
3362 return ptype_get_idx(0);
3365 nxt = pt->list.next;
3366 if (pt->type == htons(ETH_P_ALL)) {
3367 if (nxt != &ptype_all)
3370 nxt = ptype_base[0].next;
3372 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3374 while (nxt == &ptype_base[hash]) {
3375 if (++hash >= PTYPE_HASH_SIZE)
3377 nxt = ptype_base[hash].next;
3380 return list_entry(nxt, struct packet_type, list);
3383 static void ptype_seq_stop(struct seq_file *seq, void *v)
3389 static int ptype_seq_show(struct seq_file *seq, void *v)
3391 struct packet_type *pt = v;
3393 if (v == SEQ_START_TOKEN)
3394 seq_puts(seq, "Type Device Function\n");
3395 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3396 if (pt->type == htons(ETH_P_ALL))
3397 seq_puts(seq, "ALL ");
3399 seq_printf(seq, "%04x", ntohs(pt->type));
3401 seq_printf(seq, " %-8s %pF\n",
3402 pt->dev ? pt->dev->name : "", pt->func);
3408 static const struct seq_operations ptype_seq_ops = {
3409 .start = ptype_seq_start,
3410 .next = ptype_seq_next,
3411 .stop = ptype_seq_stop,
3412 .show = ptype_seq_show,
3415 static int ptype_seq_open(struct inode *inode, struct file *file)
3417 return seq_open_net(inode, file, &ptype_seq_ops,
3418 sizeof(struct seq_net_private));
3421 static const struct file_operations ptype_seq_fops = {
3422 .owner = THIS_MODULE,
3423 .open = ptype_seq_open,
3425 .llseek = seq_lseek,
3426 .release = seq_release_net,
3430 static int __net_init dev_proc_net_init(struct net *net)
3434 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3436 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3438 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3441 if (wext_proc_init(net))
3447 proc_net_remove(net, "ptype");
3449 proc_net_remove(net, "softnet_stat");
3451 proc_net_remove(net, "dev");
3455 static void __net_exit dev_proc_net_exit(struct net *net)
3457 wext_proc_exit(net);
3459 proc_net_remove(net, "ptype");
3460 proc_net_remove(net, "softnet_stat");
3461 proc_net_remove(net, "dev");
3464 static struct pernet_operations __net_initdata dev_proc_ops = {
3465 .init = dev_proc_net_init,
3466 .exit = dev_proc_net_exit,
3469 static int __init dev_proc_init(void)
3471 return register_pernet_subsys(&dev_proc_ops);
3474 #define dev_proc_init() 0
3475 #endif /* CONFIG_PROC_FS */
3479 * netdev_set_master - set up master/slave pair
3480 * @slave: slave device
3481 * @master: new master device
3483 * Changes the master device of the slave. Pass %NULL to break the
3484 * bonding. The caller must hold the RTNL semaphore. On a failure
3485 * a negative errno code is returned. On success the reference counts
3486 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3487 * function returns zero.
3489 int netdev_set_master(struct net_device *slave, struct net_device *master)
3491 struct net_device *old = slave->master;
3501 slave->master = master;
3509 slave->flags |= IFF_SLAVE;
3511 slave->flags &= ~IFF_SLAVE;
3513 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3516 EXPORT_SYMBOL(netdev_set_master);
3518 static void dev_change_rx_flags(struct net_device *dev, int flags)
3520 const struct net_device_ops *ops = dev->netdev_ops;
3522 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3523 ops->ndo_change_rx_flags(dev, flags);
3526 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3528 unsigned short old_flags = dev->flags;
3534 dev->flags |= IFF_PROMISC;
3535 dev->promiscuity += inc;
3536 if (dev->promiscuity == 0) {
3539 * If inc causes overflow, untouch promisc and return error.
3542 dev->flags &= ~IFF_PROMISC;
3544 dev->promiscuity -= inc;
3545 printk(KERN_WARNING "%s: promiscuity touches roof, "
3546 "set promiscuity failed, promiscuity feature "
3547 "of device might be broken.\n", dev->name);
3551 if (dev->flags != old_flags) {
3552 printk(KERN_INFO "device %s %s promiscuous mode\n",
3553 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3555 if (audit_enabled) {
3556 current_uid_gid(&uid, &gid);
3557 audit_log(current->audit_context, GFP_ATOMIC,
3558 AUDIT_ANOM_PROMISCUOUS,
3559 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3560 dev->name, (dev->flags & IFF_PROMISC),
3561 (old_flags & IFF_PROMISC),
3562 audit_get_loginuid(current),
3564 audit_get_sessionid(current));
3567 dev_change_rx_flags(dev, IFF_PROMISC);
3573 * dev_set_promiscuity - update promiscuity count on a device
3577 * Add or remove promiscuity from a device. While the count in the device
3578 * remains above zero the interface remains promiscuous. Once it hits zero
3579 * the device reverts back to normal filtering operation. A negative inc
3580 * value is used to drop promiscuity on the device.
3581 * Return 0 if successful or a negative errno code on error.
3583 int dev_set_promiscuity(struct net_device *dev, int inc)
3585 unsigned short old_flags = dev->flags;
3588 err = __dev_set_promiscuity(dev, inc);
3591 if (dev->flags != old_flags)
3592 dev_set_rx_mode(dev);
3595 EXPORT_SYMBOL(dev_set_promiscuity);
3598 * dev_set_allmulti - update allmulti count on a device
3602 * Add or remove reception of all multicast frames to a device. While the
3603 * count in the device remains above zero the interface remains listening
3604 * to all interfaces. Once it hits zero the device reverts back to normal
3605 * filtering operation. A negative @inc value is used to drop the counter
3606 * when releasing a resource needing all multicasts.
3607 * Return 0 if successful or a negative errno code on error.
3610 int dev_set_allmulti(struct net_device *dev, int inc)
3612 unsigned short old_flags = dev->flags;
3616 dev->flags |= IFF_ALLMULTI;
3617 dev->allmulti += inc;
3618 if (dev->allmulti == 0) {
3621 * If inc causes overflow, untouch allmulti and return error.
3624 dev->flags &= ~IFF_ALLMULTI;
3626 dev->allmulti -= inc;
3627 printk(KERN_WARNING "%s: allmulti touches roof, "
3628 "set allmulti failed, allmulti feature of "
3629 "device might be broken.\n", dev->name);
3633 if (dev->flags ^ old_flags) {
3634 dev_change_rx_flags(dev, IFF_ALLMULTI);
3635 dev_set_rx_mode(dev);
3639 EXPORT_SYMBOL(dev_set_allmulti);
3642 * Upload unicast and multicast address lists to device and
3643 * configure RX filtering. When the device doesn't support unicast
3644 * filtering it is put in promiscuous mode while unicast addresses
3647 void __dev_set_rx_mode(struct net_device *dev)
3649 const struct net_device_ops *ops = dev->netdev_ops;
3651 /* dev_open will call this function so the list will stay sane. */
3652 if (!(dev->flags&IFF_UP))
3655 if (!netif_device_present(dev))
3658 if (ops->ndo_set_rx_mode)
3659 ops->ndo_set_rx_mode(dev);
3661 /* Unicast addresses changes may only happen under the rtnl,
3662 * therefore calling __dev_set_promiscuity here is safe.
3664 if (dev->uc.count > 0 && !dev->uc_promisc) {
3665 __dev_set_promiscuity(dev, 1);
3666 dev->uc_promisc = 1;
3667 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3668 __dev_set_promiscuity(dev, -1);
3669 dev->uc_promisc = 0;
3672 if (ops->ndo_set_multicast_list)
3673 ops->ndo_set_multicast_list(dev);
3677 void dev_set_rx_mode(struct net_device *dev)
3679 netif_addr_lock_bh(dev);
3680 __dev_set_rx_mode(dev);
3681 netif_addr_unlock_bh(dev);
3684 /* hw addresses list handling functions */
3686 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3687 int addr_len, unsigned char addr_type)
3689 struct netdev_hw_addr *ha;
3692 if (addr_len > MAX_ADDR_LEN)
3695 list_for_each_entry(ha, &list->list, list) {
3696 if (!memcmp(ha->addr, addr, addr_len) &&
3697 ha->type == addr_type) {
3704 alloc_size = sizeof(*ha);
3705 if (alloc_size < L1_CACHE_BYTES)
3706 alloc_size = L1_CACHE_BYTES;
3707 ha = kmalloc(alloc_size, GFP_ATOMIC);
3710 memcpy(ha->addr, addr, addr_len);
3711 ha->type = addr_type;
3714 list_add_tail_rcu(&ha->list, &list->list);
3719 static void ha_rcu_free(struct rcu_head *head)
3721 struct netdev_hw_addr *ha;
3723 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3727 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3728 int addr_len, unsigned char addr_type)
3730 struct netdev_hw_addr *ha;
3732 list_for_each_entry(ha, &list->list, list) {
3733 if (!memcmp(ha->addr, addr, addr_len) &&
3734 (ha->type == addr_type || !addr_type)) {
3737 list_del_rcu(&ha->list);
3738 call_rcu(&ha->rcu_head, ha_rcu_free);
3746 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3747 struct netdev_hw_addr_list *from_list,
3749 unsigned char addr_type)
3752 struct netdev_hw_addr *ha, *ha2;
3755 list_for_each_entry(ha, &from_list->list, list) {
3756 type = addr_type ? addr_type : ha->type;
3757 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3764 list_for_each_entry(ha2, &from_list->list, list) {
3767 type = addr_type ? addr_type : ha2->type;
3768 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3773 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3774 struct netdev_hw_addr_list *from_list,
3776 unsigned char addr_type)
3778 struct netdev_hw_addr *ha;
3781 list_for_each_entry(ha, &from_list->list, list) {
3782 type = addr_type ? addr_type : ha->type;
3783 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3787 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3788 struct netdev_hw_addr_list *from_list,
3792 struct netdev_hw_addr *ha, *tmp;
3794 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3796 err = __hw_addr_add(to_list, ha->addr,
3797 addr_len, ha->type);
3802 } else if (ha->refcount == 1) {
3803 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3804 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3810 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3811 struct netdev_hw_addr_list *from_list,
3814 struct netdev_hw_addr *ha, *tmp;
3816 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3818 __hw_addr_del(to_list, ha->addr,
3819 addr_len, ha->type);
3821 __hw_addr_del(from_list, ha->addr,
3822 addr_len, ha->type);
3827 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3829 struct netdev_hw_addr *ha, *tmp;
3831 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3832 list_del_rcu(&ha->list);
3833 call_rcu(&ha->rcu_head, ha_rcu_free);
3838 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3840 INIT_LIST_HEAD(&list->list);
3844 /* Device addresses handling functions */
3846 static void dev_addr_flush(struct net_device *dev)
3848 /* rtnl_mutex must be held here */
3850 __hw_addr_flush(&dev->dev_addrs);
3851 dev->dev_addr = NULL;
3854 static int dev_addr_init(struct net_device *dev)
3856 unsigned char addr[MAX_ADDR_LEN];
3857 struct netdev_hw_addr *ha;
3860 /* rtnl_mutex must be held here */
3862 __hw_addr_init(&dev->dev_addrs);
3863 memset(addr, 0, sizeof(addr));
3864 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3865 NETDEV_HW_ADDR_T_LAN);
3868 * Get the first (previously created) address from the list
3869 * and set dev_addr pointer to this location.
3871 ha = list_first_entry(&dev->dev_addrs.list,
3872 struct netdev_hw_addr, list);
3873 dev->dev_addr = ha->addr;
3879 * dev_addr_add - Add a device address
3881 * @addr: address to add
3882 * @addr_type: address type
3884 * Add a device address to the device or increase the reference count if
3885 * it already exists.
3887 * The caller must hold the rtnl_mutex.
3889 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3890 unsigned char addr_type)
3896 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3898 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3901 EXPORT_SYMBOL(dev_addr_add);
3904 * dev_addr_del - Release a device address.
3906 * @addr: address to delete
3907 * @addr_type: address type
3909 * Release reference to a device address and remove it from the device
3910 * if the reference count drops to zero.
3912 * The caller must hold the rtnl_mutex.
3914 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3915 unsigned char addr_type)
3918 struct netdev_hw_addr *ha;
3923 * We can not remove the first address from the list because
3924 * dev->dev_addr points to that.
3926 ha = list_first_entry(&dev->dev_addrs.list,
3927 struct netdev_hw_addr, list);
3928 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3931 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3934 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3937 EXPORT_SYMBOL(dev_addr_del);
3940 * dev_addr_add_multiple - Add device addresses from another device
3941 * @to_dev: device to which addresses will be added
3942 * @from_dev: device from which addresses will be added
3943 * @addr_type: address type - 0 means type will be used from from_dev
3945 * Add device addresses of the one device to another.
3947 * The caller must hold the rtnl_mutex.
3949 int dev_addr_add_multiple(struct net_device *to_dev,
3950 struct net_device *from_dev,
3951 unsigned char addr_type)
3957 if (from_dev->addr_len != to_dev->addr_len)
3959 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3960 to_dev->addr_len, addr_type);
3962 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3965 EXPORT_SYMBOL(dev_addr_add_multiple);
3968 * dev_addr_del_multiple - Delete device addresses by another device
3969 * @to_dev: device where the addresses will be deleted
3970 * @from_dev: device by which addresses the addresses will be deleted
3971 * @addr_type: address type - 0 means type will used from from_dev
3973 * Deletes addresses in to device by the list of addresses in from device.
3975 * The caller must hold the rtnl_mutex.
3977 int dev_addr_del_multiple(struct net_device *to_dev,
3978 struct net_device *from_dev,
3979 unsigned char addr_type)
3983 if (from_dev->addr_len != to_dev->addr_len)
3985 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3986 to_dev->addr_len, addr_type);
3987 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3990 EXPORT_SYMBOL(dev_addr_del_multiple);
3992 /* multicast addresses handling functions */
3994 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3995 void *addr, int alen, int glbl)
3997 struct dev_addr_list *da;
3999 for (; (da = *list) != NULL; list = &da->next) {
4000 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4001 alen == da->da_addrlen) {
4003 int old_glbl = da->da_gusers;
4020 int __dev_addr_add(struct dev_addr_list **list, int *count,
4021 void *addr, int alen, int glbl)
4023 struct dev_addr_list *da;
4025 for (da = *list; da != NULL; da = da->next) {
4026 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4027 da->da_addrlen == alen) {
4029 int old_glbl = da->da_gusers;
4039 da = kzalloc(sizeof(*da), GFP_ATOMIC);
4042 memcpy(da->da_addr, addr, alen);
4043 da->da_addrlen = alen;
4045 da->da_gusers = glbl ? 1 : 0;
4053 * dev_unicast_delete - Release secondary unicast address.
4055 * @addr: address to delete
4057 * Release reference to a secondary unicast address and remove it
4058 * from the device if the reference count drops to zero.
4060 * The caller must hold the rtnl_mutex.
4062 int dev_unicast_delete(struct net_device *dev, void *addr)
4068 netif_addr_lock_bh(dev);
4069 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
4070 NETDEV_HW_ADDR_T_UNICAST);
4072 __dev_set_rx_mode(dev);
4073 netif_addr_unlock_bh(dev);
4076 EXPORT_SYMBOL(dev_unicast_delete);
4079 * dev_unicast_add - add a secondary unicast address
4081 * @addr: address to add
4083 * Add a secondary unicast address to the device or increase
4084 * the reference count if it already exists.
4086 * The caller must hold the rtnl_mutex.
4088 int dev_unicast_add(struct net_device *dev, void *addr)
4094 netif_addr_lock_bh(dev);
4095 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4096 NETDEV_HW_ADDR_T_UNICAST);
4098 __dev_set_rx_mode(dev);
4099 netif_addr_unlock_bh(dev);
4102 EXPORT_SYMBOL(dev_unicast_add);
4104 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4105 struct dev_addr_list **from, int *from_count)
4107 struct dev_addr_list *da, *next;
4111 while (da != NULL) {
4113 if (!da->da_synced) {
4114 err = __dev_addr_add(to, to_count,
4115 da->da_addr, da->da_addrlen, 0);
4120 } else if (da->da_users == 1) {
4121 __dev_addr_delete(to, to_count,
4122 da->da_addr, da->da_addrlen, 0);
4123 __dev_addr_delete(from, from_count,
4124 da->da_addr, da->da_addrlen, 0);
4130 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4132 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4133 struct dev_addr_list **from, int *from_count)
4135 struct dev_addr_list *da, *next;
4138 while (da != NULL) {
4140 if (da->da_synced) {
4141 __dev_addr_delete(to, to_count,
4142 da->da_addr, da->da_addrlen, 0);
4144 __dev_addr_delete(from, from_count,
4145 da->da_addr, da->da_addrlen, 0);
4150 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4153 * dev_unicast_sync - Synchronize device's unicast list to another device
4154 * @to: destination device
4155 * @from: source device
4157 * Add newly added addresses to the destination device and release
4158 * addresses that have no users left. The source device must be
4159 * locked by netif_tx_lock_bh.
4161 * This function is intended to be called from the dev->set_rx_mode
4162 * function of layered software devices.
4164 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4168 if (to->addr_len != from->addr_len)
4171 netif_addr_lock_bh(to);
4172 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4174 __dev_set_rx_mode(to);
4175 netif_addr_unlock_bh(to);
4178 EXPORT_SYMBOL(dev_unicast_sync);
4181 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4182 * @to: destination device
4183 * @from: source device
4185 * Remove all addresses that were added to the destination device by
4186 * dev_unicast_sync(). This function is intended to be called from the
4187 * dev->stop function of layered software devices.
4189 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4191 if (to->addr_len != from->addr_len)
4194 netif_addr_lock_bh(from);
4195 netif_addr_lock(to);
4196 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4197 __dev_set_rx_mode(to);
4198 netif_addr_unlock(to);
4199 netif_addr_unlock_bh(from);
4201 EXPORT_SYMBOL(dev_unicast_unsync);
4203 static void dev_unicast_flush(struct net_device *dev)
4205 netif_addr_lock_bh(dev);
4206 __hw_addr_flush(&dev->uc);
4207 netif_addr_unlock_bh(dev);
4210 static void dev_unicast_init(struct net_device *dev)
4212 __hw_addr_init(&dev->uc);
4216 static void __dev_addr_discard(struct dev_addr_list **list)
4218 struct dev_addr_list *tmp;
4220 while (*list != NULL) {
4223 if (tmp->da_users > tmp->da_gusers)
4224 printk("__dev_addr_discard: address leakage! "
4225 "da_users=%d\n", tmp->da_users);
4230 static void dev_addr_discard(struct net_device *dev)
4232 netif_addr_lock_bh(dev);
4234 __dev_addr_discard(&dev->mc_list);
4237 netif_addr_unlock_bh(dev);
4241 * dev_get_flags - get flags reported to userspace
4244 * Get the combination of flag bits exported through APIs to userspace.
4246 unsigned dev_get_flags(const struct net_device *dev)
4250 flags = (dev->flags & ~(IFF_PROMISC |
4255 (dev->gflags & (IFF_PROMISC |
4258 if (netif_running(dev)) {
4259 if (netif_oper_up(dev))
4260 flags |= IFF_RUNNING;
4261 if (netif_carrier_ok(dev))
4262 flags |= IFF_LOWER_UP;
4263 if (netif_dormant(dev))
4264 flags |= IFF_DORMANT;
4269 EXPORT_SYMBOL(dev_get_flags);
4272 * dev_change_flags - change device settings
4274 * @flags: device state flags
4276 * Change settings on device based state flags. The flags are
4277 * in the userspace exported format.
4279 int dev_change_flags(struct net_device *dev, unsigned flags)
4282 int old_flags = dev->flags;
4287 * Set the flags on our device.
4290 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4291 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4293 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4297 * Load in the correct multicast list now the flags have changed.
4300 if ((old_flags ^ flags) & IFF_MULTICAST)
4301 dev_change_rx_flags(dev, IFF_MULTICAST);
4303 dev_set_rx_mode(dev);
4306 * Have we downed the interface. We handle IFF_UP ourselves
4307 * according to user attempts to set it, rather than blindly
4312 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4313 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4316 dev_set_rx_mode(dev);
4319 if (dev->flags & IFF_UP &&
4320 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4322 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4324 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4325 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4327 dev->gflags ^= IFF_PROMISC;
4328 dev_set_promiscuity(dev, inc);
4331 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4332 is important. Some (broken) drivers set IFF_PROMISC, when
4333 IFF_ALLMULTI is requested not asking us and not reporting.
4335 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4336 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4338 dev->gflags ^= IFF_ALLMULTI;
4339 dev_set_allmulti(dev, inc);
4342 /* Exclude state transition flags, already notified */
4343 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4345 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4349 EXPORT_SYMBOL(dev_change_flags);
4352 * dev_set_mtu - Change maximum transfer unit
4354 * @new_mtu: new transfer unit
4356 * Change the maximum transfer size of the network device.
4358 int dev_set_mtu(struct net_device *dev, int new_mtu)
4360 const struct net_device_ops *ops = dev->netdev_ops;
4363 if (new_mtu == dev->mtu)
4366 /* MTU must be positive. */
4370 if (!netif_device_present(dev))
4374 if (ops->ndo_change_mtu)
4375 err = ops->ndo_change_mtu(dev, new_mtu);
4379 if (!err && dev->flags & IFF_UP)
4380 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4383 EXPORT_SYMBOL(dev_set_mtu);
4386 * dev_set_mac_address - Change Media Access Control Address
4390 * Change the hardware (MAC) address of the device
4392 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4394 const struct net_device_ops *ops = dev->netdev_ops;
4397 if (!ops->ndo_set_mac_address)
4399 if (sa->sa_family != dev->type)
4401 if (!netif_device_present(dev))
4403 err = ops->ndo_set_mac_address(dev, sa);
4405 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4408 EXPORT_SYMBOL(dev_set_mac_address);
4411 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4413 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4416 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4422 case SIOCGIFFLAGS: /* Get interface flags */
4423 ifr->ifr_flags = (short) dev_get_flags(dev);
4426 case SIOCGIFMETRIC: /* Get the metric on the interface
4427 (currently unused) */
4428 ifr->ifr_metric = 0;
4431 case SIOCGIFMTU: /* Get the MTU of a device */
4432 ifr->ifr_mtu = dev->mtu;
4437 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4439 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4440 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4441 ifr->ifr_hwaddr.sa_family = dev->type;
4449 ifr->ifr_map.mem_start = dev->mem_start;
4450 ifr->ifr_map.mem_end = dev->mem_end;
4451 ifr->ifr_map.base_addr = dev->base_addr;
4452 ifr->ifr_map.irq = dev->irq;
4453 ifr->ifr_map.dma = dev->dma;
4454 ifr->ifr_map.port = dev->if_port;
4458 ifr->ifr_ifindex = dev->ifindex;
4462 ifr->ifr_qlen = dev->tx_queue_len;
4466 /* dev_ioctl() should ensure this case
4478 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4480 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4483 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4484 const struct net_device_ops *ops;
4489 ops = dev->netdev_ops;
4492 case SIOCSIFFLAGS: /* Set interface flags */
4493 return dev_change_flags(dev, ifr->ifr_flags);
4495 case SIOCSIFMETRIC: /* Set the metric on the interface
4496 (currently unused) */
4499 case SIOCSIFMTU: /* Set the MTU of a device */
4500 return dev_set_mtu(dev, ifr->ifr_mtu);
4503 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4505 case SIOCSIFHWBROADCAST:
4506 if (ifr->ifr_hwaddr.sa_family != dev->type)
4508 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4509 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4510 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4514 if (ops->ndo_set_config) {
4515 if (!netif_device_present(dev))
4517 return ops->ndo_set_config(dev, &ifr->ifr_map);
4522 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4523 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4525 if (!netif_device_present(dev))
4527 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4531 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4532 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4534 if (!netif_device_present(dev))
4536 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4540 if (ifr->ifr_qlen < 0)
4542 dev->tx_queue_len = ifr->ifr_qlen;
4546 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4547 return dev_change_name(dev, ifr->ifr_newname);
4550 * Unknown or private ioctl
4553 if ((cmd >= SIOCDEVPRIVATE &&
4554 cmd <= SIOCDEVPRIVATE + 15) ||
4555 cmd == SIOCBONDENSLAVE ||
4556 cmd == SIOCBONDRELEASE ||
4557 cmd == SIOCBONDSETHWADDR ||
4558 cmd == SIOCBONDSLAVEINFOQUERY ||
4559 cmd == SIOCBONDINFOQUERY ||
4560 cmd == SIOCBONDCHANGEACTIVE ||
4561 cmd == SIOCGMIIPHY ||
4562 cmd == SIOCGMIIREG ||
4563 cmd == SIOCSMIIREG ||
4564 cmd == SIOCBRADDIF ||
4565 cmd == SIOCBRDELIF ||
4566 cmd == SIOCSHWTSTAMP ||
4567 cmd == SIOCWANDEV) {
4569 if (ops->ndo_do_ioctl) {
4570 if (netif_device_present(dev))
4571 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4583 * This function handles all "interface"-type I/O control requests. The actual
4584 * 'doing' part of this is dev_ifsioc above.
4588 * dev_ioctl - network device ioctl
4589 * @net: the applicable net namespace
4590 * @cmd: command to issue
4591 * @arg: pointer to a struct ifreq in user space
4593 * Issue ioctl functions to devices. This is normally called by the
4594 * user space syscall interfaces but can sometimes be useful for
4595 * other purposes. The return value is the return from the syscall if
4596 * positive or a negative errno code on error.
4599 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4605 /* One special case: SIOCGIFCONF takes ifconf argument
4606 and requires shared lock, because it sleeps writing
4610 if (cmd == SIOCGIFCONF) {
4612 ret = dev_ifconf(net, (char __user *) arg);
4616 if (cmd == SIOCGIFNAME)
4617 return dev_ifname(net, (struct ifreq __user *)arg);
4619 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4622 ifr.ifr_name[IFNAMSIZ-1] = 0;
4624 colon = strchr(ifr.ifr_name, ':');
4629 * See which interface the caller is talking about.
4634 * These ioctl calls:
4635 * - can be done by all.
4636 * - atomic and do not require locking.
4647 dev_load(net, ifr.ifr_name);
4649 ret = dev_ifsioc_locked(net, &ifr, cmd);
4654 if (copy_to_user(arg, &ifr,
4655 sizeof(struct ifreq)))
4661 dev_load(net, ifr.ifr_name);
4663 ret = dev_ethtool(net, &ifr);
4668 if (copy_to_user(arg, &ifr,
4669 sizeof(struct ifreq)))
4675 * These ioctl calls:
4676 * - require superuser power.
4677 * - require strict serialization.
4683 if (!capable(CAP_NET_ADMIN))
4685 dev_load(net, ifr.ifr_name);
4687 ret = dev_ifsioc(net, &ifr, cmd);
4692 if (copy_to_user(arg, &ifr,
4693 sizeof(struct ifreq)))
4699 * These ioctl calls:
4700 * - require superuser power.
4701 * - require strict serialization.
4702 * - do not return a value
4712 case SIOCSIFHWBROADCAST:
4715 case SIOCBONDENSLAVE:
4716 case SIOCBONDRELEASE:
4717 case SIOCBONDSETHWADDR:
4718 case SIOCBONDCHANGEACTIVE:
4722 if (!capable(CAP_NET_ADMIN))
4725 case SIOCBONDSLAVEINFOQUERY:
4726 case SIOCBONDINFOQUERY:
4727 dev_load(net, ifr.ifr_name);
4729 ret = dev_ifsioc(net, &ifr, cmd);
4734 /* Get the per device memory space. We can add this but
4735 * currently do not support it */
4737 /* Set the per device memory buffer space.
4738 * Not applicable in our case */
4743 * Unknown or private ioctl.
4746 if (cmd == SIOCWANDEV ||
4747 (cmd >= SIOCDEVPRIVATE &&
4748 cmd <= SIOCDEVPRIVATE + 15)) {
4749 dev_load(net, ifr.ifr_name);
4751 ret = dev_ifsioc(net, &ifr, cmd);
4753 if (!ret && copy_to_user(arg, &ifr,
4754 sizeof(struct ifreq)))
4758 /* Take care of Wireless Extensions */
4759 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4760 return wext_handle_ioctl(net, &ifr, cmd, arg);
4767 * dev_new_index - allocate an ifindex
4768 * @net: the applicable net namespace
4770 * Returns a suitable unique value for a new device interface
4771 * number. The caller must hold the rtnl semaphore or the
4772 * dev_base_lock to be sure it remains unique.
4774 static int dev_new_index(struct net *net)
4780 if (!__dev_get_by_index(net, ifindex))
4785 /* Delayed registration/unregisteration */
4786 static LIST_HEAD(net_todo_list);
4788 static void net_set_todo(struct net_device *dev)
4790 list_add_tail(&dev->todo_list, &net_todo_list);
4793 static void rollback_registered_many(struct list_head *head)
4795 struct net_device *dev, *tmp;
4797 BUG_ON(dev_boot_phase);
4800 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4801 /* Some devices call without registering
4802 * for initialization unwind. Remove those
4803 * devices and proceed with the remaining.
4805 if (dev->reg_state == NETREG_UNINITIALIZED) {
4806 pr_debug("unregister_netdevice: device %s/%p never "
4807 "was registered\n", dev->name, dev);
4810 list_del(&dev->unreg_list);
4814 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4816 /* If device is running, close it first. */
4819 /* And unlink it from device chain. */
4820 unlist_netdevice(dev);
4822 dev->reg_state = NETREG_UNREGISTERING;
4827 list_for_each_entry(dev, head, unreg_list) {
4828 /* Shutdown queueing discipline. */
4832 /* Notify protocols, that we are about to destroy
4833 this device. They should clean all the things.
4835 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4838 * Flush the unicast and multicast chains
4840 dev_unicast_flush(dev);
4841 dev_addr_discard(dev);
4843 if (dev->netdev_ops->ndo_uninit)
4844 dev->netdev_ops->ndo_uninit(dev);
4846 /* Notifier chain MUST detach us from master device. */
4847 WARN_ON(dev->master);
4849 /* Remove entries from kobject tree */
4850 netdev_unregister_kobject(dev);
4853 /* Process any work delayed until the end of the batch */
4854 dev = list_entry(head->next, struct net_device, unreg_list);
4855 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4859 list_for_each_entry(dev, head, unreg_list)
4863 static void rollback_registered(struct net_device *dev)
4867 list_add(&dev->unreg_list, &single);
4868 rollback_registered_many(&single);
4871 static void __netdev_init_queue_locks_one(struct net_device *dev,
4872 struct netdev_queue *dev_queue,
4875 spin_lock_init(&dev_queue->_xmit_lock);
4876 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4877 dev_queue->xmit_lock_owner = -1;
4880 static void netdev_init_queue_locks(struct net_device *dev)
4882 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4883 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4886 unsigned long netdev_fix_features(unsigned long features, const char *name)
4888 /* Fix illegal SG+CSUM combinations. */
4889 if ((features & NETIF_F_SG) &&
4890 !(features & NETIF_F_ALL_CSUM)) {
4892 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4893 "checksum feature.\n", name);
4894 features &= ~NETIF_F_SG;
4897 /* TSO requires that SG is present as well. */
4898 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4900 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4901 "SG feature.\n", name);
4902 features &= ~NETIF_F_TSO;
4905 if (features & NETIF_F_UFO) {
4906 if (!(features & NETIF_F_GEN_CSUM)) {
4908 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4909 "since no NETIF_F_HW_CSUM feature.\n",
4911 features &= ~NETIF_F_UFO;
4914 if (!(features & NETIF_F_SG)) {
4916 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4917 "since no NETIF_F_SG feature.\n", name);
4918 features &= ~NETIF_F_UFO;
4924 EXPORT_SYMBOL(netdev_fix_features);
4927 * netif_stacked_transfer_operstate - transfer operstate
4928 * @rootdev: the root or lower level device to transfer state from
4929 * @dev: the device to transfer operstate to
4931 * Transfer operational state from root to device. This is normally
4932 * called when a stacking relationship exists between the root
4933 * device and the device(a leaf device).
4935 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4936 struct net_device *dev)
4938 if (rootdev->operstate == IF_OPER_DORMANT)
4939 netif_dormant_on(dev);
4941 netif_dormant_off(dev);
4943 if (netif_carrier_ok(rootdev)) {
4944 if (!netif_carrier_ok(dev))
4945 netif_carrier_on(dev);
4947 if (netif_carrier_ok(dev))
4948 netif_carrier_off(dev);
4951 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4954 * register_netdevice - register a network device
4955 * @dev: device to register
4957 * Take a completed network device structure and add it to the kernel
4958 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4959 * chain. 0 is returned on success. A negative errno code is returned
4960 * on a failure to set up the device, or if the name is a duplicate.
4962 * Callers must hold the rtnl semaphore. You may want
4963 * register_netdev() instead of this.
4966 * The locking appears insufficient to guarantee two parallel registers
4967 * will not get the same name.
4970 int register_netdevice(struct net_device *dev)
4973 struct net *net = dev_net(dev);
4975 BUG_ON(dev_boot_phase);
4980 /* When net_device's are persistent, this will be fatal. */
4981 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4984 spin_lock_init(&dev->addr_list_lock);
4985 netdev_set_addr_lockdep_class(dev);
4986 netdev_init_queue_locks(dev);
4990 /* Init, if this function is available */
4991 if (dev->netdev_ops->ndo_init) {
4992 ret = dev->netdev_ops->ndo_init(dev);
5000 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
5004 dev->ifindex = dev_new_index(net);
5005 if (dev->iflink == -1)
5006 dev->iflink = dev->ifindex;
5008 /* Fix illegal checksum combinations */
5009 if ((dev->features & NETIF_F_HW_CSUM) &&
5010 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5011 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5013 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5016 if ((dev->features & NETIF_F_NO_CSUM) &&
5017 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5018 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5020 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5023 dev->features = netdev_fix_features(dev->features, dev->name);
5025 /* Enable software GSO if SG is supported. */
5026 if (dev->features & NETIF_F_SG)
5027 dev->features |= NETIF_F_GSO;
5029 netdev_initialize_kobject(dev);
5031 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5032 ret = notifier_to_errno(ret);
5036 ret = netdev_register_kobject(dev);
5039 dev->reg_state = NETREG_REGISTERED;
5042 * Default initial state at registry is that the
5043 * device is present.
5046 set_bit(__LINK_STATE_PRESENT, &dev->state);
5048 dev_init_scheduler(dev);
5050 list_netdevice(dev);
5052 /* Notify protocols, that a new device appeared. */
5053 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5054 ret = notifier_to_errno(ret);
5056 rollback_registered(dev);
5057 dev->reg_state = NETREG_UNREGISTERED;
5060 * Prevent userspace races by waiting until the network
5061 * device is fully setup before sending notifications.
5063 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5069 if (dev->netdev_ops->ndo_uninit)
5070 dev->netdev_ops->ndo_uninit(dev);
5073 EXPORT_SYMBOL(register_netdevice);
5076 * init_dummy_netdev - init a dummy network device for NAPI
5077 * @dev: device to init
5079 * This takes a network device structure and initialize the minimum
5080 * amount of fields so it can be used to schedule NAPI polls without
5081 * registering a full blown interface. This is to be used by drivers
5082 * that need to tie several hardware interfaces to a single NAPI
5083 * poll scheduler due to HW limitations.
5085 int init_dummy_netdev(struct net_device *dev)
5087 /* Clear everything. Note we don't initialize spinlocks
5088 * are they aren't supposed to be taken by any of the
5089 * NAPI code and this dummy netdev is supposed to be
5090 * only ever used for NAPI polls
5092 memset(dev, 0, sizeof(struct net_device));
5094 /* make sure we BUG if trying to hit standard
5095 * register/unregister code path
5097 dev->reg_state = NETREG_DUMMY;
5099 /* initialize the ref count */
5100 atomic_set(&dev->refcnt, 1);
5102 /* NAPI wants this */
5103 INIT_LIST_HEAD(&dev->napi_list);
5105 /* a dummy interface is started by default */
5106 set_bit(__LINK_STATE_PRESENT, &dev->state);
5107 set_bit(__LINK_STATE_START, &dev->state);
5111 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5115 * register_netdev - register a network device
5116 * @dev: device to register
5118 * Take a completed network device structure and add it to the kernel
5119 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5120 * chain. 0 is returned on success. A negative errno code is returned
5121 * on a failure to set up the device, or if the name is a duplicate.
5123 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5124 * and expands the device name if you passed a format string to
5127 int register_netdev(struct net_device *dev)
5134 * If the name is a format string the caller wants us to do a
5137 if (strchr(dev->name, '%')) {
5138 err = dev_alloc_name(dev, dev->name);
5143 err = register_netdevice(dev);
5148 EXPORT_SYMBOL(register_netdev);
5151 * netdev_wait_allrefs - wait until all references are gone.
5153 * This is called when unregistering network devices.
5155 * Any protocol or device that holds a reference should register
5156 * for netdevice notification, and cleanup and put back the
5157 * reference if they receive an UNREGISTER event.
5158 * We can get stuck here if buggy protocols don't correctly
5161 static void netdev_wait_allrefs(struct net_device *dev)
5163 unsigned long rebroadcast_time, warning_time;
5165 linkwatch_forget_dev(dev);
5167 rebroadcast_time = warning_time = jiffies;
5168 while (atomic_read(&dev->refcnt) != 0) {
5169 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5172 /* Rebroadcast unregister notification */
5173 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5174 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5175 * should have already handle it the first time */
5177 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5179 /* We must not have linkwatch events
5180 * pending on unregister. If this
5181 * happens, we simply run the queue
5182 * unscheduled, resulting in a noop
5185 linkwatch_run_queue();
5190 rebroadcast_time = jiffies;
5195 if (time_after(jiffies, warning_time + 10 * HZ)) {
5196 printk(KERN_EMERG "unregister_netdevice: "
5197 "waiting for %s to become free. Usage "
5199 dev->name, atomic_read(&dev->refcnt));
5200 warning_time = jiffies;
5209 * register_netdevice(x1);
5210 * register_netdevice(x2);
5212 * unregister_netdevice(y1);
5213 * unregister_netdevice(y2);
5219 * We are invoked by rtnl_unlock().
5220 * This allows us to deal with problems:
5221 * 1) We can delete sysfs objects which invoke hotplug
5222 * without deadlocking with linkwatch via keventd.
5223 * 2) Since we run with the RTNL semaphore not held, we can sleep
5224 * safely in order to wait for the netdev refcnt to drop to zero.
5226 * We must not return until all unregister events added during
5227 * the interval the lock was held have been completed.
5229 void netdev_run_todo(void)
5231 struct list_head list;
5233 /* Snapshot list, allow later requests */
5234 list_replace_init(&net_todo_list, &list);
5238 while (!list_empty(&list)) {
5239 struct net_device *dev
5240 = list_entry(list.next, struct net_device, todo_list);
5241 list_del(&dev->todo_list);
5243 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5244 printk(KERN_ERR "network todo '%s' but state %d\n",
5245 dev->name, dev->reg_state);
5250 dev->reg_state = NETREG_UNREGISTERED;
5252 on_each_cpu(flush_backlog, dev, 1);
5254 netdev_wait_allrefs(dev);
5257 BUG_ON(atomic_read(&dev->refcnt));
5258 WARN_ON(dev->ip_ptr);
5259 WARN_ON(dev->ip6_ptr);
5260 WARN_ON(dev->dn_ptr);
5262 if (dev->destructor)
5263 dev->destructor(dev);
5265 /* Free network device */
5266 kobject_put(&dev->dev.kobj);
5271 * dev_txq_stats_fold - fold tx_queues stats
5272 * @dev: device to get statistics from
5273 * @stats: struct net_device_stats to hold results
5275 void dev_txq_stats_fold(const struct net_device *dev,
5276 struct net_device_stats *stats)
5278 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5280 struct netdev_queue *txq;
5282 for (i = 0; i < dev->num_tx_queues; i++) {
5283 txq = netdev_get_tx_queue(dev, i);
5284 tx_bytes += txq->tx_bytes;
5285 tx_packets += txq->tx_packets;
5286 tx_dropped += txq->tx_dropped;
5288 if (tx_bytes || tx_packets || tx_dropped) {
5289 stats->tx_bytes = tx_bytes;
5290 stats->tx_packets = tx_packets;
5291 stats->tx_dropped = tx_dropped;
5294 EXPORT_SYMBOL(dev_txq_stats_fold);
5297 * dev_get_stats - get network device statistics
5298 * @dev: device to get statistics from
5300 * Get network statistics from device. The device driver may provide
5301 * its own method by setting dev->netdev_ops->get_stats; otherwise
5302 * the internal statistics structure is used.
5304 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5306 const struct net_device_ops *ops = dev->netdev_ops;
5308 if (ops->ndo_get_stats)
5309 return ops->ndo_get_stats(dev);
5311 dev_txq_stats_fold(dev, &dev->stats);
5314 EXPORT_SYMBOL(dev_get_stats);
5316 static void netdev_init_one_queue(struct net_device *dev,
5317 struct netdev_queue *queue,
5323 static void netdev_init_queues(struct net_device *dev)
5325 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5326 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5327 spin_lock_init(&dev->tx_global_lock);
5331 * alloc_netdev_mq - allocate network device
5332 * @sizeof_priv: size of private data to allocate space for
5333 * @name: device name format string
5334 * @setup: callback to initialize device
5335 * @queue_count: the number of subqueues to allocate
5337 * Allocates a struct net_device with private data area for driver use
5338 * and performs basic initialization. Also allocates subquue structs
5339 * for each queue on the device at the end of the netdevice.
5341 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5342 void (*setup)(struct net_device *), unsigned int queue_count)
5344 struct netdev_queue *tx;
5345 struct net_device *dev;
5347 struct net_device *p;
5349 BUG_ON(strlen(name) >= sizeof(dev->name));
5351 alloc_size = sizeof(struct net_device);
5353 /* ensure 32-byte alignment of private area */
5354 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5355 alloc_size += sizeof_priv;
5357 /* ensure 32-byte alignment of whole construct */
5358 alloc_size += NETDEV_ALIGN - 1;
5360 p = kzalloc(alloc_size, GFP_KERNEL);
5362 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5366 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5368 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5373 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5374 dev->padded = (char *)dev - (char *)p;
5376 if (dev_addr_init(dev))
5379 dev_unicast_init(dev);
5381 dev_net_set(dev, &init_net);
5384 dev->num_tx_queues = queue_count;
5385 dev->real_num_tx_queues = queue_count;
5387 dev->gso_max_size = GSO_MAX_SIZE;
5389 netdev_init_queues(dev);
5391 INIT_LIST_HEAD(&dev->napi_list);
5392 INIT_LIST_HEAD(&dev->unreg_list);
5393 INIT_LIST_HEAD(&dev->link_watch_list);
5394 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5396 strcpy(dev->name, name);
5406 EXPORT_SYMBOL(alloc_netdev_mq);
5409 * free_netdev - free network device
5412 * This function does the last stage of destroying an allocated device
5413 * interface. The reference to the device object is released.
5414 * If this is the last reference then it will be freed.
5416 void free_netdev(struct net_device *dev)
5418 struct napi_struct *p, *n;
5420 release_net(dev_net(dev));
5424 /* Flush device addresses */
5425 dev_addr_flush(dev);
5427 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5430 /* Compatibility with error handling in drivers */
5431 if (dev->reg_state == NETREG_UNINITIALIZED) {
5432 kfree((char *)dev - dev->padded);
5436 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5437 dev->reg_state = NETREG_RELEASED;
5439 /* will free via device release */
5440 put_device(&dev->dev);
5442 EXPORT_SYMBOL(free_netdev);
5445 * synchronize_net - Synchronize with packet receive processing
5447 * Wait for packets currently being received to be done.
5448 * Does not block later packets from starting.
5450 void synchronize_net(void)
5455 EXPORT_SYMBOL(synchronize_net);
5458 * unregister_netdevice_queue - remove device from the kernel
5462 * This function shuts down a device interface and removes it
5463 * from the kernel tables.
5464 * If head not NULL, device is queued to be unregistered later.
5466 * Callers must hold the rtnl semaphore. You may want
5467 * unregister_netdev() instead of this.
5470 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5475 list_move_tail(&dev->unreg_list, head);
5477 rollback_registered(dev);
5478 /* Finish processing unregister after unlock */
5482 EXPORT_SYMBOL(unregister_netdevice_queue);
5485 * unregister_netdevice_many - unregister many devices
5486 * @head: list of devices
5488 void unregister_netdevice_many(struct list_head *head)
5490 struct net_device *dev;
5492 if (!list_empty(head)) {
5493 rollback_registered_many(head);
5494 list_for_each_entry(dev, head, unreg_list)
5498 EXPORT_SYMBOL(unregister_netdevice_many);
5501 * unregister_netdev - remove device from the kernel
5504 * This function shuts down a device interface and removes it
5505 * from the kernel tables.
5507 * This is just a wrapper for unregister_netdevice that takes
5508 * the rtnl semaphore. In general you want to use this and not
5509 * unregister_netdevice.
5511 void unregister_netdev(struct net_device *dev)
5514 unregister_netdevice(dev);
5517 EXPORT_SYMBOL(unregister_netdev);
5520 * dev_change_net_namespace - move device to different nethost namespace
5522 * @net: network namespace
5523 * @pat: If not NULL name pattern to try if the current device name
5524 * is already taken in the destination network namespace.
5526 * This function shuts down a device interface and moves it
5527 * to a new network namespace. On success 0 is returned, on
5528 * a failure a netagive errno code is returned.
5530 * Callers must hold the rtnl semaphore.
5533 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5539 /* Don't allow namespace local devices to be moved. */
5541 if (dev->features & NETIF_F_NETNS_LOCAL)
5545 /* Don't allow real devices to be moved when sysfs
5549 if (dev->dev.parent)
5553 /* Ensure the device has been registrered */
5555 if (dev->reg_state != NETREG_REGISTERED)
5558 /* Get out if there is nothing todo */
5560 if (net_eq(dev_net(dev), net))
5563 /* Pick the destination device name, and ensure
5564 * we can use it in the destination network namespace.
5567 if (__dev_get_by_name(net, dev->name)) {
5568 /* We get here if we can't use the current device name */
5571 if (dev_get_valid_name(net, pat, dev->name, 1))
5576 * And now a mini version of register_netdevice unregister_netdevice.
5579 /* If device is running close it first. */
5582 /* And unlink it from device chain */
5584 unlist_netdevice(dev);
5588 /* Shutdown queueing discipline. */
5591 /* Notify protocols, that we are about to destroy
5592 this device. They should clean all the things.
5594 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5595 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5598 * Flush the unicast and multicast chains
5600 dev_unicast_flush(dev);
5601 dev_addr_discard(dev);
5603 netdev_unregister_kobject(dev);
5605 /* Actually switch the network namespace */
5606 dev_net_set(dev, net);
5608 /* If there is an ifindex conflict assign a new one */
5609 if (__dev_get_by_index(net, dev->ifindex)) {
5610 int iflink = (dev->iflink == dev->ifindex);
5611 dev->ifindex = dev_new_index(net);
5613 dev->iflink = dev->ifindex;
5616 /* Fixup kobjects */
5617 err = netdev_register_kobject(dev);
5620 /* Add the device back in the hashes */
5621 list_netdevice(dev);
5623 /* Notify protocols, that a new device appeared. */
5624 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5627 * Prevent userspace races by waiting until the network
5628 * device is fully setup before sending notifications.
5630 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5637 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5639 static int dev_cpu_callback(struct notifier_block *nfb,
5640 unsigned long action,
5643 struct sk_buff **list_skb;
5644 struct Qdisc **list_net;
5645 struct sk_buff *skb;
5646 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5647 struct softnet_data *sd, *oldsd;
5649 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5652 local_irq_disable();
5653 cpu = smp_processor_id();
5654 sd = &per_cpu(softnet_data, cpu);
5655 oldsd = &per_cpu(softnet_data, oldcpu);
5657 /* Find end of our completion_queue. */
5658 list_skb = &sd->completion_queue;
5660 list_skb = &(*list_skb)->next;
5661 /* Append completion queue from offline CPU. */
5662 *list_skb = oldsd->completion_queue;
5663 oldsd->completion_queue = NULL;
5665 /* Find end of our output_queue. */
5666 list_net = &sd->output_queue;
5668 list_net = &(*list_net)->next_sched;
5669 /* Append output queue from offline CPU. */
5670 *list_net = oldsd->output_queue;
5671 oldsd->output_queue = NULL;
5673 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5676 /* Process offline CPU's input_pkt_queue */
5677 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5685 * netdev_increment_features - increment feature set by one
5686 * @all: current feature set
5687 * @one: new feature set
5688 * @mask: mask feature set
5690 * Computes a new feature set after adding a device with feature set
5691 * @one to the master device with current feature set @all. Will not
5692 * enable anything that is off in @mask. Returns the new feature set.
5694 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5697 /* If device needs checksumming, downgrade to it. */
5698 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5699 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5700 else if (mask & NETIF_F_ALL_CSUM) {
5701 /* If one device supports v4/v6 checksumming, set for all. */
5702 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5703 !(all & NETIF_F_GEN_CSUM)) {
5704 all &= ~NETIF_F_ALL_CSUM;
5705 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5708 /* If one device supports hw checksumming, set for all. */
5709 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5710 all &= ~NETIF_F_ALL_CSUM;
5711 all |= NETIF_F_HW_CSUM;
5715 one |= NETIF_F_ALL_CSUM;
5717 one |= all & NETIF_F_ONE_FOR_ALL;
5718 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5719 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5723 EXPORT_SYMBOL(netdev_increment_features);
5725 static struct hlist_head *netdev_create_hash(void)
5728 struct hlist_head *hash;
5730 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5732 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5733 INIT_HLIST_HEAD(&hash[i]);
5738 /* Initialize per network namespace state */
5739 static int __net_init netdev_init(struct net *net)
5741 INIT_LIST_HEAD(&net->dev_base_head);
5743 net->dev_name_head = netdev_create_hash();
5744 if (net->dev_name_head == NULL)
5747 net->dev_index_head = netdev_create_hash();
5748 if (net->dev_index_head == NULL)
5754 kfree(net->dev_name_head);
5760 * netdev_drivername - network driver for the device
5761 * @dev: network device
5762 * @buffer: buffer for resulting name
5763 * @len: size of buffer
5765 * Determine network driver for device.
5767 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5769 const struct device_driver *driver;
5770 const struct device *parent;
5772 if (len <= 0 || !buffer)
5776 parent = dev->dev.parent;
5781 driver = parent->driver;
5782 if (driver && driver->name)
5783 strlcpy(buffer, driver->name, len);
5787 static void __net_exit netdev_exit(struct net *net)
5789 kfree(net->dev_name_head);
5790 kfree(net->dev_index_head);
5793 static struct pernet_operations __net_initdata netdev_net_ops = {
5794 .init = netdev_init,
5795 .exit = netdev_exit,
5798 static void __net_exit default_device_exit(struct net *net)
5800 struct net_device *dev, *aux;
5802 * Push all migratable network devices back to the
5803 * initial network namespace
5806 for_each_netdev_safe(net, dev, aux) {
5808 char fb_name[IFNAMSIZ];
5810 /* Ignore unmoveable devices (i.e. loopback) */
5811 if (dev->features & NETIF_F_NETNS_LOCAL)
5814 /* Leave virtual devices for the generic cleanup */
5815 if (dev->rtnl_link_ops)
5818 /* Push remaing network devices to init_net */
5819 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5820 err = dev_change_net_namespace(dev, &init_net, fb_name);
5822 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5823 __func__, dev->name, err);
5830 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5832 /* At exit all network devices most be removed from a network
5833 * namespace. Do this in the reverse order of registeration.
5834 * Do this across as many network namespaces as possible to
5835 * improve batching efficiency.
5837 struct net_device *dev;
5839 LIST_HEAD(dev_kill_list);
5842 list_for_each_entry(net, net_list, exit_list) {
5843 for_each_netdev_reverse(net, dev) {
5844 if (dev->rtnl_link_ops)
5845 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5847 unregister_netdevice_queue(dev, &dev_kill_list);
5850 unregister_netdevice_many(&dev_kill_list);
5854 static struct pernet_operations __net_initdata default_device_ops = {
5855 .exit = default_device_exit,
5856 .exit_batch = default_device_exit_batch,
5860 * Initialize the DEV module. At boot time this walks the device list and
5861 * unhooks any devices that fail to initialise (normally hardware not
5862 * present) and leaves us with a valid list of present and active devices.
5867 * This is called single threaded during boot, so no need
5868 * to take the rtnl semaphore.
5870 static int __init net_dev_init(void)
5872 int i, rc = -ENOMEM;
5874 BUG_ON(!dev_boot_phase);
5876 if (dev_proc_init())
5879 if (netdev_kobject_init())
5882 INIT_LIST_HEAD(&ptype_all);
5883 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5884 INIT_LIST_HEAD(&ptype_base[i]);
5886 if (register_pernet_subsys(&netdev_net_ops))
5890 * Initialise the packet receive queues.
5893 for_each_possible_cpu(i) {
5894 struct softnet_data *queue;
5896 queue = &per_cpu(softnet_data, i);
5897 skb_queue_head_init(&queue->input_pkt_queue);
5898 queue->completion_queue = NULL;
5899 INIT_LIST_HEAD(&queue->poll_list);
5901 queue->backlog.poll = process_backlog;
5902 queue->backlog.weight = weight_p;
5903 queue->backlog.gro_list = NULL;
5904 queue->backlog.gro_count = 0;
5909 /* The loopback device is special if any other network devices
5910 * is present in a network namespace the loopback device must
5911 * be present. Since we now dynamically allocate and free the
5912 * loopback device ensure this invariant is maintained by
5913 * keeping the loopback device as the first device on the
5914 * list of network devices. Ensuring the loopback devices
5915 * is the first device that appears and the last network device
5918 if (register_pernet_device(&loopback_net_ops))
5921 if (register_pernet_device(&default_device_ops))
5924 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5925 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5927 hotcpu_notifier(dev_cpu_callback, 0);
5935 subsys_initcall(net_dev_init);
5937 static int __init initialize_hashrnd(void)
5939 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5943 late_initcall_sync(initialize_hashrnd);