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/sched.h>
83 #include <linux/mutex.h>
84 #include <linux/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/ethtool.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
96 #include <net/net_namespace.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/stat.h>
102 #include <linux/if_bridge.h>
103 #include <linux/if_macvlan.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
131 #include "net-sysfs.h"
133 /* Instead of increasing this, you should create a hash table. */
134 #define MAX_GRO_SKBS 8
136 /* This should be increased if a protocol with a bigger head is added. */
137 #define GRO_MAX_HEAD (MAX_HEADER + 128)
140 * The list of packet types we will receive (as opposed to discard)
141 * and the routines to invoke.
143 * Why 16. Because with 16 the only overlap we get on a hash of the
144 * low nibble of the protocol value is RARP/SNAP/X.25.
146 * NOTE: That is no longer true with the addition of VLAN tags. Not
147 * sure which should go first, but I bet it won't make much
148 * difference if we are running VLANs. The good news is that
149 * this protocol won't be in the list unless compiled in, so
150 * the average user (w/out VLANs) will not be adversely affected.
167 #define PTYPE_HASH_SIZE (16)
168 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
170 static DEFINE_SPINLOCK(ptype_lock);
171 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
172 static struct list_head ptype_all __read_mostly; /* Taps */
175 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
178 * Pure readers hold dev_base_lock for reading.
180 * Writers must hold the rtnl semaphore while they loop through the
181 * dev_base_head list, and hold dev_base_lock for writing when they do the
182 * actual updates. This allows pure readers to access the list even
183 * while a writer is preparing to update it.
185 * To put it another way, dev_base_lock is held for writing only to
186 * protect against pure readers; the rtnl semaphore provides the
187 * protection against other writers.
189 * See, for example usages, register_netdevice() and
190 * unregister_netdevice(), which must be called with the rtnl
193 DEFINE_RWLOCK(dev_base_lock);
194 EXPORT_SYMBOL(dev_base_lock);
196 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
198 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
199 return &net->dev_name_head[hash & (NETDEV_HASHENTRIES - 1)];
202 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
204 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
207 /* Device list insertion */
208 static int list_netdevice(struct net_device *dev)
210 struct net *net = dev_net(dev);
214 write_lock_bh(&dev_base_lock);
215 list_add_tail(&dev->dev_list, &net->dev_base_head);
216 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
217 hlist_add_head_rcu(&dev->index_hlist,
218 dev_index_hash(net, dev->ifindex));
219 write_unlock_bh(&dev_base_lock);
223 /* Device list removal
224 * caller must respect a RCU grace period before freeing/reusing dev
226 static void unlist_netdevice(struct net_device *dev)
230 /* Unlink dev from the device chain */
231 write_lock_bh(&dev_base_lock);
232 list_del(&dev->dev_list);
233 hlist_del(&dev->name_hlist);
234 hlist_del_rcu(&dev->index_hlist);
235 write_unlock_bh(&dev_base_lock);
242 static RAW_NOTIFIER_HEAD(netdev_chain);
245 * Device drivers call our routines to queue packets here. We empty the
246 * queue in the local softnet handler.
249 DEFINE_PER_CPU(struct softnet_data, softnet_data);
250 EXPORT_PER_CPU_SYMBOL(softnet_data);
252 #ifdef CONFIG_LOCKDEP
254 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
255 * according to dev->type
257 static const unsigned short netdev_lock_type[] =
258 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
259 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
260 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
261 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
262 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
263 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
264 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
265 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
266 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
267 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
268 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
269 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
270 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
271 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
272 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
273 ARPHRD_VOID, ARPHRD_NONE};
275 static const char *const netdev_lock_name[] =
276 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
277 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
278 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
279 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
280 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
281 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
282 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
283 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
284 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
285 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
286 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
287 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
288 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
289 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
290 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
291 "_xmit_VOID", "_xmit_NONE"};
293 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
294 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
296 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
300 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
301 if (netdev_lock_type[i] == dev_type)
303 /* the last key is used by default */
304 return ARRAY_SIZE(netdev_lock_type) - 1;
307 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
308 unsigned short dev_type)
312 i = netdev_lock_pos(dev_type);
313 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
314 netdev_lock_name[i]);
317 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
321 i = netdev_lock_pos(dev->type);
322 lockdep_set_class_and_name(&dev->addr_list_lock,
323 &netdev_addr_lock_key[i],
324 netdev_lock_name[i]);
327 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
328 unsigned short dev_type)
331 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
336 /*******************************************************************************
338 Protocol management and registration routines
340 *******************************************************************************/
343 * Add a protocol ID to the list. Now that the input handler is
344 * smarter we can dispense with all the messy stuff that used to be
347 * BEWARE!!! Protocol handlers, mangling input packets,
348 * MUST BE last in hash buckets and checking protocol handlers
349 * MUST start from promiscuous ptype_all chain in net_bh.
350 * It is true now, do not change it.
351 * Explanation follows: if protocol handler, mangling packet, will
352 * be the first on list, it is not able to sense, that packet
353 * is cloned and should be copied-on-write, so that it will
354 * change it and subsequent readers will get broken packet.
359 * dev_add_pack - add packet handler
360 * @pt: packet type declaration
362 * Add a protocol handler to the networking stack. The passed &packet_type
363 * is linked into kernel lists and may not be freed until it has been
364 * removed from the kernel lists.
366 * This call does not sleep therefore it can not
367 * guarantee all CPU's that are in middle of receiving packets
368 * will see the new packet type (until the next received packet).
371 void dev_add_pack(struct packet_type *pt)
375 spin_lock_bh(&ptype_lock);
376 if (pt->type == htons(ETH_P_ALL))
377 list_add_rcu(&pt->list, &ptype_all);
379 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
380 list_add_rcu(&pt->list, &ptype_base[hash]);
382 spin_unlock_bh(&ptype_lock);
384 EXPORT_SYMBOL(dev_add_pack);
387 * __dev_remove_pack - remove packet handler
388 * @pt: packet type declaration
390 * Remove a protocol handler that was previously added to the kernel
391 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
392 * from the kernel lists and can be freed or reused once this function
395 * The packet type might still be in use by receivers
396 * and must not be freed until after all the CPU's have gone
397 * through a quiescent state.
399 void __dev_remove_pack(struct packet_type *pt)
401 struct list_head *head;
402 struct packet_type *pt1;
404 spin_lock_bh(&ptype_lock);
406 if (pt->type == htons(ETH_P_ALL))
409 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
411 list_for_each_entry(pt1, head, list) {
413 list_del_rcu(&pt->list);
418 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
420 spin_unlock_bh(&ptype_lock);
422 EXPORT_SYMBOL(__dev_remove_pack);
425 * dev_remove_pack - remove packet handler
426 * @pt: packet type declaration
428 * Remove a protocol handler that was previously added to the kernel
429 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
430 * from the kernel lists and can be freed or reused once this function
433 * This call sleeps to guarantee that no CPU is looking at the packet
436 void dev_remove_pack(struct packet_type *pt)
438 __dev_remove_pack(pt);
442 EXPORT_SYMBOL(dev_remove_pack);
444 /******************************************************************************
446 Device Boot-time Settings Routines
448 *******************************************************************************/
450 /* Boot time configuration table */
451 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
454 * netdev_boot_setup_add - add new setup entry
455 * @name: name of the device
456 * @map: configured settings for the device
458 * Adds new setup entry to the dev_boot_setup list. The function
459 * returns 0 on error and 1 on success. This is a generic routine to
462 static int netdev_boot_setup_add(char *name, struct ifmap *map)
464 struct netdev_boot_setup *s;
468 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
469 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
470 memset(s[i].name, 0, sizeof(s[i].name));
471 strlcpy(s[i].name, name, IFNAMSIZ);
472 memcpy(&s[i].map, map, sizeof(s[i].map));
477 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
481 * netdev_boot_setup_check - check boot time settings
482 * @dev: the netdevice
484 * Check boot time settings for the device.
485 * The found settings are set for the device to be used
486 * later in the device probing.
487 * Returns 0 if no settings found, 1 if they are.
489 int netdev_boot_setup_check(struct net_device *dev)
491 struct netdev_boot_setup *s = dev_boot_setup;
494 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
495 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
496 !strcmp(dev->name, s[i].name)) {
497 dev->irq = s[i].map.irq;
498 dev->base_addr = s[i].map.base_addr;
499 dev->mem_start = s[i].map.mem_start;
500 dev->mem_end = s[i].map.mem_end;
506 EXPORT_SYMBOL(netdev_boot_setup_check);
510 * netdev_boot_base - get address from boot time settings
511 * @prefix: prefix for network device
512 * @unit: id for network device
514 * Check boot time settings for the base address of device.
515 * The found settings are set for the device to be used
516 * later in the device probing.
517 * Returns 0 if no settings found.
519 unsigned long netdev_boot_base(const char *prefix, int unit)
521 const struct netdev_boot_setup *s = dev_boot_setup;
525 sprintf(name, "%s%d", prefix, unit);
528 * If device already registered then return base of 1
529 * to indicate not to probe for this interface
531 if (__dev_get_by_name(&init_net, name))
534 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
535 if (!strcmp(name, s[i].name))
536 return s[i].map.base_addr;
541 * Saves at boot time configured settings for any netdevice.
543 int __init netdev_boot_setup(char *str)
548 str = get_options(str, ARRAY_SIZE(ints), ints);
553 memset(&map, 0, sizeof(map));
557 map.base_addr = ints[2];
559 map.mem_start = ints[3];
561 map.mem_end = ints[4];
563 /* Add new entry to the list */
564 return netdev_boot_setup_add(str, &map);
567 __setup("netdev=", netdev_boot_setup);
569 /*******************************************************************************
571 Device Interface Subroutines
573 *******************************************************************************/
576 * __dev_get_by_name - find a device by its name
577 * @net: the applicable net namespace
578 * @name: name to find
580 * Find an interface by name. Must be called under RTNL semaphore
581 * or @dev_base_lock. If the name is found a pointer to the device
582 * is returned. If the name is not found then %NULL is returned. The
583 * reference counters are not incremented so the caller must be
584 * careful with locks.
587 struct net_device *__dev_get_by_name(struct net *net, const char *name)
589 struct hlist_node *p;
591 hlist_for_each(p, dev_name_hash(net, name)) {
592 struct net_device *dev
593 = hlist_entry(p, struct net_device, name_hlist);
594 if (!strncmp(dev->name, name, IFNAMSIZ))
599 EXPORT_SYMBOL(__dev_get_by_name);
602 * dev_get_by_name - find a device by its name
603 * @net: the applicable net namespace
604 * @name: name to find
606 * Find an interface by name. This can be called from any
607 * context and does its own locking. The returned handle has
608 * the usage count incremented and the caller must use dev_put() to
609 * release it when it is no longer needed. %NULL is returned if no
610 * matching device is found.
613 struct net_device *dev_get_by_name(struct net *net, const char *name)
615 struct net_device *dev;
617 read_lock(&dev_base_lock);
618 dev = __dev_get_by_name(net, name);
621 read_unlock(&dev_base_lock);
624 EXPORT_SYMBOL(dev_get_by_name);
627 * __dev_get_by_index - find a device by its ifindex
628 * @net: the applicable net namespace
629 * @ifindex: index of device
631 * Search for an interface by index. Returns %NULL if the device
632 * is not found or a pointer to the device. The device has not
633 * had its reference counter increased so the caller must be careful
634 * about locking. The caller must hold either the RTNL semaphore
638 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
640 struct hlist_node *p;
642 hlist_for_each(p, dev_index_hash(net, ifindex)) {
643 struct net_device *dev
644 = hlist_entry(p, struct net_device, index_hlist);
645 if (dev->ifindex == ifindex)
650 EXPORT_SYMBOL(__dev_get_by_index);
653 * dev_get_by_index_rcu - find a device by its ifindex
654 * @net: the applicable net namespace
655 * @ifindex: index of device
657 * Search for an interface by index. Returns %NULL if the device
658 * is not found or a pointer to the device. The device has not
659 * had its reference counter increased so the caller must be careful
660 * about locking. The caller must hold RCU lock.
663 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
665 struct hlist_node *p;
666 struct net_device *dev;
667 struct hlist_head *head = dev_index_hash(net, ifindex);
669 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
670 if (dev->ifindex == ifindex)
675 EXPORT_SYMBOL(dev_get_by_index_rcu);
679 * dev_get_by_index - find a device by its ifindex
680 * @net: the applicable net namespace
681 * @ifindex: index of device
683 * Search for an interface by index. Returns NULL if the device
684 * is not found or a pointer to the device. The device returned has
685 * had a reference added and the pointer is safe until the user calls
686 * dev_put to indicate they have finished with it.
689 struct net_device *dev_get_by_index(struct net *net, int ifindex)
691 struct net_device *dev;
694 dev = dev_get_by_index_rcu(net, ifindex);
700 EXPORT_SYMBOL(dev_get_by_index);
703 * dev_getbyhwaddr - find a device by its hardware address
704 * @net: the applicable net namespace
705 * @type: media type of device
706 * @ha: hardware address
708 * Search for an interface by MAC address. Returns NULL if the device
709 * is not found or a pointer to the device. The caller must hold the
710 * rtnl semaphore. The returned device has not had its ref count increased
711 * and the caller must therefore be careful about locking
714 * If the API was consistent this would be __dev_get_by_hwaddr
717 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
719 struct net_device *dev;
723 for_each_netdev(net, dev)
724 if (dev->type == type &&
725 !memcmp(dev->dev_addr, ha, dev->addr_len))
730 EXPORT_SYMBOL(dev_getbyhwaddr);
732 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
734 struct net_device *dev;
737 for_each_netdev(net, dev)
738 if (dev->type == type)
743 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
745 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
747 struct net_device *dev;
750 dev = __dev_getfirstbyhwtype(net, type);
756 EXPORT_SYMBOL(dev_getfirstbyhwtype);
759 * dev_get_by_flags - find any device with given flags
760 * @net: the applicable net namespace
761 * @if_flags: IFF_* values
762 * @mask: bitmask of bits in if_flags to check
764 * Search for any interface with the given flags. Returns NULL if a device
765 * is not found or a pointer to the device. The device returned has
766 * had a reference added and the pointer is safe until the user calls
767 * dev_put to indicate they have finished with it.
770 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
773 struct net_device *dev, *ret;
776 read_lock(&dev_base_lock);
777 for_each_netdev(net, dev) {
778 if (((dev->flags ^ if_flags) & mask) == 0) {
784 read_unlock(&dev_base_lock);
787 EXPORT_SYMBOL(dev_get_by_flags);
790 * dev_valid_name - check if name is okay for network device
793 * Network device names need to be valid file names to
794 * to allow sysfs to work. We also disallow any kind of
797 int dev_valid_name(const char *name)
801 if (strlen(name) >= IFNAMSIZ)
803 if (!strcmp(name, ".") || !strcmp(name, ".."))
807 if (*name == '/' || isspace(*name))
813 EXPORT_SYMBOL(dev_valid_name);
816 * __dev_alloc_name - allocate a name for a device
817 * @net: network namespace to allocate the device name in
818 * @name: name format string
819 * @buf: scratch buffer and result name string
821 * Passed a format string - eg "lt%d" it will try and find a suitable
822 * id. It scans list of devices to build up a free map, then chooses
823 * the first empty slot. The caller must hold the dev_base or rtnl lock
824 * while allocating the name and adding the device in order to avoid
826 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
827 * Returns the number of the unit assigned or a negative errno code.
830 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
834 const int max_netdevices = 8*PAGE_SIZE;
835 unsigned long *inuse;
836 struct net_device *d;
838 p = strnchr(name, IFNAMSIZ-1, '%');
841 * Verify the string as this thing may have come from
842 * the user. There must be either one "%d" and no other "%"
845 if (p[1] != 'd' || strchr(p + 2, '%'))
848 /* Use one page as a bit array of possible slots */
849 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
853 for_each_netdev(net, d) {
854 if (!sscanf(d->name, name, &i))
856 if (i < 0 || i >= max_netdevices)
859 /* avoid cases where sscanf is not exact inverse of printf */
860 snprintf(buf, IFNAMSIZ, name, i);
861 if (!strncmp(buf, d->name, IFNAMSIZ))
865 i = find_first_zero_bit(inuse, max_netdevices);
866 free_page((unsigned long) inuse);
869 snprintf(buf, IFNAMSIZ, name, i);
870 if (!__dev_get_by_name(net, buf))
873 /* It is possible to run out of possible slots
874 * when the name is long and there isn't enough space left
875 * for the digits, or if all bits are used.
881 * dev_alloc_name - allocate a name for a device
883 * @name: name format string
885 * Passed a format string - eg "lt%d" it will try and find a suitable
886 * id. It scans list of devices to build up a free map, then chooses
887 * the first empty slot. The caller must hold the dev_base or rtnl lock
888 * while allocating the name and adding the device in order to avoid
890 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
891 * Returns the number of the unit assigned or a negative errno code.
894 int dev_alloc_name(struct net_device *dev, const char *name)
900 BUG_ON(!dev_net(dev));
902 ret = __dev_alloc_name(net, name, buf);
904 strlcpy(dev->name, buf, IFNAMSIZ);
907 EXPORT_SYMBOL(dev_alloc_name);
911 * dev_change_name - change name of a device
913 * @newname: name (or format string) must be at least IFNAMSIZ
915 * Change name of a device, can pass format strings "eth%d".
918 int dev_change_name(struct net_device *dev, const char *newname)
920 char oldname[IFNAMSIZ];
926 BUG_ON(!dev_net(dev));
929 if (dev->flags & IFF_UP)
932 if (!dev_valid_name(newname))
935 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
938 memcpy(oldname, dev->name, IFNAMSIZ);
940 if (strchr(newname, '%')) {
941 err = dev_alloc_name(dev, newname);
944 } else if (__dev_get_by_name(net, newname))
947 strlcpy(dev->name, newname, IFNAMSIZ);
950 /* For now only devices in the initial network namespace
953 if (net == &init_net) {
954 ret = device_rename(&dev->dev, dev->name);
956 memcpy(dev->name, oldname, IFNAMSIZ);
961 write_lock_bh(&dev_base_lock);
962 hlist_del(&dev->name_hlist);
963 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
964 write_unlock_bh(&dev_base_lock);
966 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
967 ret = notifier_to_errno(ret);
972 "%s: name change rollback failed: %d.\n",
976 memcpy(dev->name, oldname, IFNAMSIZ);
985 * dev_set_alias - change ifalias of a device
987 * @alias: name up to IFALIASZ
988 * @len: limit of bytes to copy from info
990 * Set ifalias for a device,
992 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1001 kfree(dev->ifalias);
1002 dev->ifalias = NULL;
1007 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1011 strlcpy(dev->ifalias, alias, len+1);
1017 * netdev_features_change - device changes features
1018 * @dev: device to cause notification
1020 * Called to indicate a device has changed features.
1022 void netdev_features_change(struct net_device *dev)
1024 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1026 EXPORT_SYMBOL(netdev_features_change);
1029 * netdev_state_change - device changes state
1030 * @dev: device to cause notification
1032 * Called to indicate a device has changed state. This function calls
1033 * the notifier chains for netdev_chain and sends a NEWLINK message
1034 * to the routing socket.
1036 void netdev_state_change(struct net_device *dev)
1038 if (dev->flags & IFF_UP) {
1039 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1040 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1043 EXPORT_SYMBOL(netdev_state_change);
1045 void netdev_bonding_change(struct net_device *dev, unsigned long event)
1047 call_netdevice_notifiers(event, dev);
1049 EXPORT_SYMBOL(netdev_bonding_change);
1052 * dev_load - load a network module
1053 * @net: the applicable net namespace
1054 * @name: name of interface
1056 * If a network interface is not present and the process has suitable
1057 * privileges this function loads the module. If module loading is not
1058 * available in this kernel then it becomes a nop.
1061 void dev_load(struct net *net, const char *name)
1063 struct net_device *dev;
1065 read_lock(&dev_base_lock);
1066 dev = __dev_get_by_name(net, name);
1067 read_unlock(&dev_base_lock);
1069 if (!dev && capable(CAP_NET_ADMIN))
1070 request_module("%s", name);
1072 EXPORT_SYMBOL(dev_load);
1075 * dev_open - prepare an interface for use.
1076 * @dev: device to open
1078 * Takes a device from down to up state. The device's private open
1079 * function is invoked and then the multicast lists are loaded. Finally
1080 * the device is moved into the up state and a %NETDEV_UP message is
1081 * sent to the netdev notifier chain.
1083 * Calling this function on an active interface is a nop. On a failure
1084 * a negative errno code is returned.
1086 int dev_open(struct net_device *dev)
1088 const struct net_device_ops *ops = dev->netdev_ops;
1097 if (dev->flags & IFF_UP)
1101 * Is it even present?
1103 if (!netif_device_present(dev))
1106 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1107 ret = notifier_to_errno(ret);
1112 * Call device private open method
1114 set_bit(__LINK_STATE_START, &dev->state);
1116 if (ops->ndo_validate_addr)
1117 ret = ops->ndo_validate_addr(dev);
1119 if (!ret && ops->ndo_open)
1120 ret = ops->ndo_open(dev);
1123 * If it went open OK then:
1127 clear_bit(__LINK_STATE_START, &dev->state);
1132 dev->flags |= IFF_UP;
1137 net_dmaengine_get();
1140 * Initialize multicasting status
1142 dev_set_rx_mode(dev);
1145 * Wakeup transmit queue engine
1150 * ... and announce new interface.
1152 call_netdevice_notifiers(NETDEV_UP, dev);
1157 EXPORT_SYMBOL(dev_open);
1160 * dev_close - shutdown an interface.
1161 * @dev: device to shutdown
1163 * This function moves an active device into down state. A
1164 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1165 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1168 int dev_close(struct net_device *dev)
1170 const struct net_device_ops *ops = dev->netdev_ops;
1175 if (!(dev->flags & IFF_UP))
1179 * Tell people we are going down, so that they can
1180 * prepare to death, when device is still operating.
1182 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1184 clear_bit(__LINK_STATE_START, &dev->state);
1186 /* Synchronize to scheduled poll. We cannot touch poll list,
1187 * it can be even on different cpu. So just clear netif_running().
1189 * dev->stop() will invoke napi_disable() on all of it's
1190 * napi_struct instances on this device.
1192 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1194 dev_deactivate(dev);
1197 * Call the device specific close. This cannot fail.
1198 * Only if device is UP
1200 * We allow it to be called even after a DETACH hot-plug
1207 * Device is now down.
1210 dev->flags &= ~IFF_UP;
1213 * Tell people we are down
1215 call_netdevice_notifiers(NETDEV_DOWN, dev);
1220 net_dmaengine_put();
1224 EXPORT_SYMBOL(dev_close);
1228 * dev_disable_lro - disable Large Receive Offload on a device
1231 * Disable Large Receive Offload (LRO) on a net device. Must be
1232 * called under RTNL. This is needed if received packets may be
1233 * forwarded to another interface.
1235 void dev_disable_lro(struct net_device *dev)
1237 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1238 dev->ethtool_ops->set_flags) {
1239 u32 flags = dev->ethtool_ops->get_flags(dev);
1240 if (flags & ETH_FLAG_LRO) {
1241 flags &= ~ETH_FLAG_LRO;
1242 dev->ethtool_ops->set_flags(dev, flags);
1245 WARN_ON(dev->features & NETIF_F_LRO);
1247 EXPORT_SYMBOL(dev_disable_lro);
1250 static int dev_boot_phase = 1;
1253 * Device change register/unregister. These are not inline or static
1254 * as we export them to the world.
1258 * register_netdevice_notifier - register a network notifier block
1261 * Register a notifier to be called when network device events occur.
1262 * The notifier passed is linked into the kernel structures and must
1263 * not be reused until it has been unregistered. A negative errno code
1264 * is returned on a failure.
1266 * When registered all registration and up events are replayed
1267 * to the new notifier to allow device to have a race free
1268 * view of the network device list.
1271 int register_netdevice_notifier(struct notifier_block *nb)
1273 struct net_device *dev;
1274 struct net_device *last;
1279 err = raw_notifier_chain_register(&netdev_chain, nb);
1285 for_each_netdev(net, dev) {
1286 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1287 err = notifier_to_errno(err);
1291 if (!(dev->flags & IFF_UP))
1294 nb->notifier_call(nb, NETDEV_UP, dev);
1305 for_each_netdev(net, dev) {
1309 if (dev->flags & IFF_UP) {
1310 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1311 nb->notifier_call(nb, NETDEV_DOWN, dev);
1313 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1317 raw_notifier_chain_unregister(&netdev_chain, nb);
1320 EXPORT_SYMBOL(register_netdevice_notifier);
1323 * unregister_netdevice_notifier - unregister a network notifier block
1326 * Unregister a notifier previously registered by
1327 * register_netdevice_notifier(). The notifier is unlinked into the
1328 * kernel structures and may then be reused. A negative errno code
1329 * is returned on a failure.
1332 int unregister_netdevice_notifier(struct notifier_block *nb)
1337 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1341 EXPORT_SYMBOL(unregister_netdevice_notifier);
1344 * call_netdevice_notifiers - call all network notifier blocks
1345 * @val: value passed unmodified to notifier function
1346 * @dev: net_device pointer passed unmodified to notifier function
1348 * Call all network notifier blocks. Parameters and return value
1349 * are as for raw_notifier_call_chain().
1352 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1354 return raw_notifier_call_chain(&netdev_chain, val, dev);
1357 /* When > 0 there are consumers of rx skb time stamps */
1358 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1360 void net_enable_timestamp(void)
1362 atomic_inc(&netstamp_needed);
1364 EXPORT_SYMBOL(net_enable_timestamp);
1366 void net_disable_timestamp(void)
1368 atomic_dec(&netstamp_needed);
1370 EXPORT_SYMBOL(net_disable_timestamp);
1372 static inline void net_timestamp(struct sk_buff *skb)
1374 if (atomic_read(&netstamp_needed))
1375 __net_timestamp(skb);
1377 skb->tstamp.tv64 = 0;
1381 * Support routine. Sends outgoing frames to any network
1382 * taps currently in use.
1385 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1387 struct packet_type *ptype;
1389 #ifdef CONFIG_NET_CLS_ACT
1390 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1397 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1398 /* Never send packets back to the socket
1399 * they originated from - MvS (miquels@drinkel.ow.org)
1401 if ((ptype->dev == dev || !ptype->dev) &&
1402 (ptype->af_packet_priv == NULL ||
1403 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1404 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1408 /* skb->nh should be correctly
1409 set by sender, so that the second statement is
1410 just protection against buggy protocols.
1412 skb_reset_mac_header(skb2);
1414 if (skb_network_header(skb2) < skb2->data ||
1415 skb2->network_header > skb2->tail) {
1416 if (net_ratelimit())
1417 printk(KERN_CRIT "protocol %04x is "
1419 skb2->protocol, dev->name);
1420 skb_reset_network_header(skb2);
1423 skb2->transport_header = skb2->network_header;
1424 skb2->pkt_type = PACKET_OUTGOING;
1425 ptype->func(skb2, skb->dev, ptype, skb->dev);
1432 static inline void __netif_reschedule(struct Qdisc *q)
1434 struct softnet_data *sd;
1435 unsigned long flags;
1437 local_irq_save(flags);
1438 sd = &__get_cpu_var(softnet_data);
1439 q->next_sched = sd->output_queue;
1440 sd->output_queue = q;
1441 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1442 local_irq_restore(flags);
1445 void __netif_schedule(struct Qdisc *q)
1447 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1448 __netif_reschedule(q);
1450 EXPORT_SYMBOL(__netif_schedule);
1452 void dev_kfree_skb_irq(struct sk_buff *skb)
1454 if (atomic_dec_and_test(&skb->users)) {
1455 struct softnet_data *sd;
1456 unsigned long flags;
1458 local_irq_save(flags);
1459 sd = &__get_cpu_var(softnet_data);
1460 skb->next = sd->completion_queue;
1461 sd->completion_queue = skb;
1462 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1463 local_irq_restore(flags);
1466 EXPORT_SYMBOL(dev_kfree_skb_irq);
1468 void dev_kfree_skb_any(struct sk_buff *skb)
1470 if (in_irq() || irqs_disabled())
1471 dev_kfree_skb_irq(skb);
1475 EXPORT_SYMBOL(dev_kfree_skb_any);
1479 * netif_device_detach - mark device as removed
1480 * @dev: network device
1482 * Mark device as removed from system and therefore no longer available.
1484 void netif_device_detach(struct net_device *dev)
1486 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1487 netif_running(dev)) {
1488 netif_tx_stop_all_queues(dev);
1491 EXPORT_SYMBOL(netif_device_detach);
1494 * netif_device_attach - mark device as attached
1495 * @dev: network device
1497 * Mark device as attached from system and restart if needed.
1499 void netif_device_attach(struct net_device *dev)
1501 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1502 netif_running(dev)) {
1503 netif_tx_wake_all_queues(dev);
1504 __netdev_watchdog_up(dev);
1507 EXPORT_SYMBOL(netif_device_attach);
1509 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1511 return ((features & NETIF_F_GEN_CSUM) ||
1512 ((features & NETIF_F_IP_CSUM) &&
1513 protocol == htons(ETH_P_IP)) ||
1514 ((features & NETIF_F_IPV6_CSUM) &&
1515 protocol == htons(ETH_P_IPV6)) ||
1516 ((features & NETIF_F_FCOE_CRC) &&
1517 protocol == htons(ETH_P_FCOE)));
1520 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1522 if (can_checksum_protocol(dev->features, skb->protocol))
1525 if (skb->protocol == htons(ETH_P_8021Q)) {
1526 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1527 if (can_checksum_protocol(dev->features & dev->vlan_features,
1528 veh->h_vlan_encapsulated_proto))
1536 * Invalidate hardware checksum when packet is to be mangled, and
1537 * complete checksum manually on outgoing path.
1539 int skb_checksum_help(struct sk_buff *skb)
1542 int ret = 0, offset;
1544 if (skb->ip_summed == CHECKSUM_COMPLETE)
1545 goto out_set_summed;
1547 if (unlikely(skb_shinfo(skb)->gso_size)) {
1548 /* Let GSO fix up the checksum. */
1549 goto out_set_summed;
1552 offset = skb->csum_start - skb_headroom(skb);
1553 BUG_ON(offset >= skb_headlen(skb));
1554 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1556 offset += skb->csum_offset;
1557 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1559 if (skb_cloned(skb) &&
1560 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1561 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1566 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1568 skb->ip_summed = CHECKSUM_NONE;
1572 EXPORT_SYMBOL(skb_checksum_help);
1575 * skb_gso_segment - Perform segmentation on skb.
1576 * @skb: buffer to segment
1577 * @features: features for the output path (see dev->features)
1579 * This function segments the given skb and returns a list of segments.
1581 * It may return NULL if the skb requires no segmentation. This is
1582 * only possible when GSO is used for verifying header integrity.
1584 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1586 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1587 struct packet_type *ptype;
1588 __be16 type = skb->protocol;
1591 skb_reset_mac_header(skb);
1592 skb->mac_len = skb->network_header - skb->mac_header;
1593 __skb_pull(skb, skb->mac_len);
1595 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1596 struct net_device *dev = skb->dev;
1597 struct ethtool_drvinfo info = {};
1599 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1600 dev->ethtool_ops->get_drvinfo(dev, &info);
1602 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1604 info.driver, dev ? dev->features : 0L,
1605 skb->sk ? skb->sk->sk_route_caps : 0L,
1606 skb->len, skb->data_len, skb->ip_summed);
1608 if (skb_header_cloned(skb) &&
1609 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1610 return ERR_PTR(err);
1614 list_for_each_entry_rcu(ptype,
1615 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1616 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1617 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1618 err = ptype->gso_send_check(skb);
1619 segs = ERR_PTR(err);
1620 if (err || skb_gso_ok(skb, features))
1622 __skb_push(skb, (skb->data -
1623 skb_network_header(skb)));
1625 segs = ptype->gso_segment(skb, features);
1631 __skb_push(skb, skb->data - skb_mac_header(skb));
1635 EXPORT_SYMBOL(skb_gso_segment);
1637 /* Take action when hardware reception checksum errors are detected. */
1639 void netdev_rx_csum_fault(struct net_device *dev)
1641 if (net_ratelimit()) {
1642 printk(KERN_ERR "%s: hw csum failure.\n",
1643 dev ? dev->name : "<unknown>");
1647 EXPORT_SYMBOL(netdev_rx_csum_fault);
1650 /* Actually, we should eliminate this check as soon as we know, that:
1651 * 1. IOMMU is present and allows to map all the memory.
1652 * 2. No high memory really exists on this machine.
1655 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1657 #ifdef CONFIG_HIGHMEM
1660 if (dev->features & NETIF_F_HIGHDMA)
1663 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1664 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1672 void (*destructor)(struct sk_buff *skb);
1675 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1677 static void dev_gso_skb_destructor(struct sk_buff *skb)
1679 struct dev_gso_cb *cb;
1682 struct sk_buff *nskb = skb->next;
1684 skb->next = nskb->next;
1687 } while (skb->next);
1689 cb = DEV_GSO_CB(skb);
1691 cb->destructor(skb);
1695 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1696 * @skb: buffer to segment
1698 * This function segments the given skb and stores the list of segments
1701 static int dev_gso_segment(struct sk_buff *skb)
1703 struct net_device *dev = skb->dev;
1704 struct sk_buff *segs;
1705 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1708 segs = skb_gso_segment(skb, features);
1710 /* Verifying header integrity only. */
1715 return PTR_ERR(segs);
1718 DEV_GSO_CB(skb)->destructor = skb->destructor;
1719 skb->destructor = dev_gso_skb_destructor;
1724 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1725 struct netdev_queue *txq)
1727 const struct net_device_ops *ops = dev->netdev_ops;
1730 if (likely(!skb->next)) {
1731 if (!list_empty(&ptype_all))
1732 dev_queue_xmit_nit(skb, dev);
1734 if (netif_needs_gso(dev, skb)) {
1735 if (unlikely(dev_gso_segment(skb)))
1742 * If device doesnt need skb->dst, release it right now while
1743 * its hot in this cpu cache
1745 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1748 rc = ops->ndo_start_xmit(skb, dev);
1749 if (rc == NETDEV_TX_OK)
1750 txq_trans_update(txq);
1752 * TODO: if skb_orphan() was called by
1753 * dev->hard_start_xmit() (for example, the unmodified
1754 * igb driver does that; bnx2 doesn't), then
1755 * skb_tx_software_timestamp() will be unable to send
1756 * back the time stamp.
1758 * How can this be prevented? Always create another
1759 * reference to the socket before calling
1760 * dev->hard_start_xmit()? Prevent that skb_orphan()
1761 * does anything in dev->hard_start_xmit() by clearing
1762 * the skb destructor before the call and restoring it
1763 * afterwards, then doing the skb_orphan() ourselves?
1770 struct sk_buff *nskb = skb->next;
1772 skb->next = nskb->next;
1774 rc = ops->ndo_start_xmit(nskb, dev);
1775 if (unlikely(rc != NETDEV_TX_OK)) {
1776 nskb->next = skb->next;
1780 txq_trans_update(txq);
1781 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1782 return NETDEV_TX_BUSY;
1783 } while (skb->next);
1785 skb->destructor = DEV_GSO_CB(skb)->destructor;
1789 return NETDEV_TX_OK;
1792 static u32 skb_tx_hashrnd;
1794 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1798 if (skb_rx_queue_recorded(skb)) {
1799 hash = skb_get_rx_queue(skb);
1800 while (unlikely(hash >= dev->real_num_tx_queues))
1801 hash -= dev->real_num_tx_queues;
1805 if (skb->sk && skb->sk->sk_hash)
1806 hash = skb->sk->sk_hash;
1808 hash = skb->protocol;
1810 hash = jhash_1word(hash, skb_tx_hashrnd);
1812 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1814 EXPORT_SYMBOL(skb_tx_hash);
1816 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1817 struct sk_buff *skb)
1820 struct sock *sk = skb->sk;
1822 if (sk_tx_queue_recorded(sk)) {
1823 queue_index = sk_tx_queue_get(sk);
1825 const struct net_device_ops *ops = dev->netdev_ops;
1827 if (ops->ndo_select_queue) {
1828 queue_index = ops->ndo_select_queue(dev, skb);
1831 if (dev->real_num_tx_queues > 1)
1832 queue_index = skb_tx_hash(dev, skb);
1834 if (sk && sk->sk_dst_cache)
1835 sk_tx_queue_set(sk, queue_index);
1839 skb_set_queue_mapping(skb, queue_index);
1840 return netdev_get_tx_queue(dev, queue_index);
1843 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1844 struct net_device *dev,
1845 struct netdev_queue *txq)
1847 spinlock_t *root_lock = qdisc_lock(q);
1850 spin_lock(root_lock);
1851 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1854 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1855 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1857 * This is a work-conserving queue; there are no old skbs
1858 * waiting to be sent out; and the qdisc is not running -
1859 * xmit the skb directly.
1861 __qdisc_update_bstats(q, skb->len);
1862 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1865 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1867 rc = NET_XMIT_SUCCESS;
1869 rc = qdisc_enqueue_root(skb, q);
1872 spin_unlock(root_lock);
1878 * dev_queue_xmit - transmit a buffer
1879 * @skb: buffer to transmit
1881 * Queue a buffer for transmission to a network device. The caller must
1882 * have set the device and priority and built the buffer before calling
1883 * this function. The function can be called from an interrupt.
1885 * A negative errno code is returned on a failure. A success does not
1886 * guarantee the frame will be transmitted as it may be dropped due
1887 * to congestion or traffic shaping.
1889 * -----------------------------------------------------------------------------------
1890 * I notice this method can also return errors from the queue disciplines,
1891 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1894 * Regardless of the return value, the skb is consumed, so it is currently
1895 * difficult to retry a send to this method. (You can bump the ref count
1896 * before sending to hold a reference for retry if you are careful.)
1898 * When calling this method, interrupts MUST be enabled. This is because
1899 * the BH enable code must have IRQs enabled so that it will not deadlock.
1902 int dev_queue_xmit(struct sk_buff *skb)
1904 struct net_device *dev = skb->dev;
1905 struct netdev_queue *txq;
1909 /* GSO will handle the following emulations directly. */
1910 if (netif_needs_gso(dev, skb))
1913 if (skb_has_frags(skb) &&
1914 !(dev->features & NETIF_F_FRAGLIST) &&
1915 __skb_linearize(skb))
1918 /* Fragmented skb is linearized if device does not support SG,
1919 * or if at least one of fragments is in highmem and device
1920 * does not support DMA from it.
1922 if (skb_shinfo(skb)->nr_frags &&
1923 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1924 __skb_linearize(skb))
1927 /* If packet is not checksummed and device does not support
1928 * checksumming for this protocol, complete checksumming here.
1930 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1931 skb_set_transport_header(skb, skb->csum_start -
1933 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1938 /* Disable soft irqs for various locks below. Also
1939 * stops preemption for RCU.
1943 txq = dev_pick_tx(dev, skb);
1944 q = rcu_dereference(txq->qdisc);
1946 #ifdef CONFIG_NET_CLS_ACT
1947 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
1950 rc = __dev_xmit_skb(skb, q, dev, txq);
1954 /* The device has no queue. Common case for software devices:
1955 loopback, all the sorts of tunnels...
1957 Really, it is unlikely that netif_tx_lock protection is necessary
1958 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1960 However, it is possible, that they rely on protection
1963 Check this and shot the lock. It is not prone from deadlocks.
1964 Either shot noqueue qdisc, it is even simpler 8)
1966 if (dev->flags & IFF_UP) {
1967 int cpu = smp_processor_id(); /* ok because BHs are off */
1969 if (txq->xmit_lock_owner != cpu) {
1971 HARD_TX_LOCK(dev, txq, cpu);
1973 if (!netif_tx_queue_stopped(txq)) {
1974 rc = NET_XMIT_SUCCESS;
1975 if (!dev_hard_start_xmit(skb, dev, txq)) {
1976 HARD_TX_UNLOCK(dev, txq);
1980 HARD_TX_UNLOCK(dev, txq);
1981 if (net_ratelimit())
1982 printk(KERN_CRIT "Virtual device %s asks to "
1983 "queue packet!\n", dev->name);
1985 /* Recursion is detected! It is possible,
1987 if (net_ratelimit())
1988 printk(KERN_CRIT "Dead loop on virtual device "
1989 "%s, fix it urgently!\n", dev->name);
1994 rcu_read_unlock_bh();
2000 rcu_read_unlock_bh();
2003 EXPORT_SYMBOL(dev_queue_xmit);
2006 /*=======================================================================
2008 =======================================================================*/
2010 int netdev_max_backlog __read_mostly = 1000;
2011 int netdev_budget __read_mostly = 300;
2012 int weight_p __read_mostly = 64; /* old backlog weight */
2014 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2018 * netif_rx - post buffer to the network code
2019 * @skb: buffer to post
2021 * This function receives a packet from a device driver and queues it for
2022 * the upper (protocol) levels to process. It always succeeds. The buffer
2023 * may be dropped during processing for congestion control or by the
2027 * NET_RX_SUCCESS (no congestion)
2028 * NET_RX_DROP (packet was dropped)
2032 int netif_rx(struct sk_buff *skb)
2034 struct softnet_data *queue;
2035 unsigned long flags;
2037 /* if netpoll wants it, pretend we never saw it */
2038 if (netpoll_rx(skb))
2041 if (!skb->tstamp.tv64)
2045 * The code is rearranged so that the path is the most
2046 * short when CPU is congested, but is still operating.
2048 local_irq_save(flags);
2049 queue = &__get_cpu_var(softnet_data);
2051 __get_cpu_var(netdev_rx_stat).total++;
2052 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2053 if (queue->input_pkt_queue.qlen) {
2055 __skb_queue_tail(&queue->input_pkt_queue, skb);
2056 local_irq_restore(flags);
2057 return NET_RX_SUCCESS;
2060 napi_schedule(&queue->backlog);
2064 __get_cpu_var(netdev_rx_stat).dropped++;
2065 local_irq_restore(flags);
2070 EXPORT_SYMBOL(netif_rx);
2072 int netif_rx_ni(struct sk_buff *skb)
2077 err = netif_rx(skb);
2078 if (local_softirq_pending())
2084 EXPORT_SYMBOL(netif_rx_ni);
2086 static void net_tx_action(struct softirq_action *h)
2088 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2090 if (sd->completion_queue) {
2091 struct sk_buff *clist;
2093 local_irq_disable();
2094 clist = sd->completion_queue;
2095 sd->completion_queue = NULL;
2099 struct sk_buff *skb = clist;
2100 clist = clist->next;
2102 WARN_ON(atomic_read(&skb->users));
2107 if (sd->output_queue) {
2110 local_irq_disable();
2111 head = sd->output_queue;
2112 sd->output_queue = NULL;
2116 struct Qdisc *q = head;
2117 spinlock_t *root_lock;
2119 head = head->next_sched;
2121 root_lock = qdisc_lock(q);
2122 if (spin_trylock(root_lock)) {
2123 smp_mb__before_clear_bit();
2124 clear_bit(__QDISC_STATE_SCHED,
2127 spin_unlock(root_lock);
2129 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2131 __netif_reschedule(q);
2133 smp_mb__before_clear_bit();
2134 clear_bit(__QDISC_STATE_SCHED,
2142 static inline int deliver_skb(struct sk_buff *skb,
2143 struct packet_type *pt_prev,
2144 struct net_device *orig_dev)
2146 atomic_inc(&skb->users);
2147 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2150 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2152 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2153 /* This hook is defined here for ATM LANE */
2154 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2155 unsigned char *addr) __read_mostly;
2156 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2160 * If bridge module is loaded call bridging hook.
2161 * returns NULL if packet was consumed.
2163 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2164 struct sk_buff *skb) __read_mostly;
2165 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2167 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2168 struct packet_type **pt_prev, int *ret,
2169 struct net_device *orig_dev)
2171 struct net_bridge_port *port;
2173 if (skb->pkt_type == PACKET_LOOPBACK ||
2174 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2178 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2182 return br_handle_frame_hook(port, skb);
2185 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2188 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2189 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2190 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2192 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2193 struct packet_type **pt_prev,
2195 struct net_device *orig_dev)
2197 if (skb->dev->macvlan_port == NULL)
2201 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2204 return macvlan_handle_frame_hook(skb);
2207 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2210 #ifdef CONFIG_NET_CLS_ACT
2211 /* TODO: Maybe we should just force sch_ingress to be compiled in
2212 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2213 * a compare and 2 stores extra right now if we dont have it on
2214 * but have CONFIG_NET_CLS_ACT
2215 * NOTE: This doesnt stop any functionality; if you dont have
2216 * the ingress scheduler, you just cant add policies on ingress.
2219 static int ing_filter(struct sk_buff *skb)
2221 struct net_device *dev = skb->dev;
2222 u32 ttl = G_TC_RTTL(skb->tc_verd);
2223 struct netdev_queue *rxq;
2224 int result = TC_ACT_OK;
2227 if (MAX_RED_LOOP < ttl++) {
2229 "Redir loop detected Dropping packet (%d->%d)\n",
2230 skb->iif, dev->ifindex);
2234 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2235 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2237 rxq = &dev->rx_queue;
2240 if (q != &noop_qdisc) {
2241 spin_lock(qdisc_lock(q));
2242 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2243 result = qdisc_enqueue_root(skb, q);
2244 spin_unlock(qdisc_lock(q));
2250 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2251 struct packet_type **pt_prev,
2252 int *ret, struct net_device *orig_dev)
2254 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2258 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2261 /* Huh? Why does turning on AF_PACKET affect this? */
2262 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2265 switch (ing_filter(skb)) {
2279 * netif_nit_deliver - deliver received packets to network taps
2282 * This function is used to deliver incoming packets to network
2283 * taps. It should be used when the normal netif_receive_skb path
2284 * is bypassed, for example because of VLAN acceleration.
2286 void netif_nit_deliver(struct sk_buff *skb)
2288 struct packet_type *ptype;
2290 if (list_empty(&ptype_all))
2293 skb_reset_network_header(skb);
2294 skb_reset_transport_header(skb);
2295 skb->mac_len = skb->network_header - skb->mac_header;
2298 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2299 if (!ptype->dev || ptype->dev == skb->dev)
2300 deliver_skb(skb, ptype, skb->dev);
2306 * netif_receive_skb - process receive buffer from network
2307 * @skb: buffer to process
2309 * netif_receive_skb() is the main receive data processing function.
2310 * It always succeeds. The buffer may be dropped during processing
2311 * for congestion control or by the protocol layers.
2313 * This function may only be called from softirq context and interrupts
2314 * should be enabled.
2316 * Return values (usually ignored):
2317 * NET_RX_SUCCESS: no congestion
2318 * NET_RX_DROP: packet was dropped
2320 int netif_receive_skb(struct sk_buff *skb)
2322 struct packet_type *ptype, *pt_prev;
2323 struct net_device *orig_dev;
2324 struct net_device *null_or_orig;
2325 int ret = NET_RX_DROP;
2328 if (!skb->tstamp.tv64)
2331 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2332 return NET_RX_SUCCESS;
2334 /* if we've gotten here through NAPI, check netpoll */
2335 if (netpoll_receive_skb(skb))
2339 skb->iif = skb->dev->ifindex;
2341 null_or_orig = NULL;
2342 orig_dev = skb->dev;
2343 if (orig_dev->master) {
2344 if (skb_bond_should_drop(skb))
2345 null_or_orig = orig_dev; /* deliver only exact match */
2347 skb->dev = orig_dev->master;
2350 __get_cpu_var(netdev_rx_stat).total++;
2352 skb_reset_network_header(skb);
2353 skb_reset_transport_header(skb);
2354 skb->mac_len = skb->network_header - skb->mac_header;
2360 #ifdef CONFIG_NET_CLS_ACT
2361 if (skb->tc_verd & TC_NCLS) {
2362 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2367 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2368 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2369 ptype->dev == orig_dev) {
2371 ret = deliver_skb(skb, pt_prev, orig_dev);
2376 #ifdef CONFIG_NET_CLS_ACT
2377 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2383 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2386 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2390 type = skb->protocol;
2391 list_for_each_entry_rcu(ptype,
2392 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2393 if (ptype->type == type &&
2394 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2395 ptype->dev == orig_dev)) {
2397 ret = deliver_skb(skb, pt_prev, orig_dev);
2403 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2406 /* Jamal, now you will not able to escape explaining
2407 * me how you were going to use this. :-)
2416 EXPORT_SYMBOL(netif_receive_skb);
2418 /* Network device is going away, flush any packets still pending */
2419 static void flush_backlog(void *arg)
2421 struct net_device *dev = arg;
2422 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2423 struct sk_buff *skb, *tmp;
2425 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2426 if (skb->dev == dev) {
2427 __skb_unlink(skb, &queue->input_pkt_queue);
2432 static int napi_gro_complete(struct sk_buff *skb)
2434 struct packet_type *ptype;
2435 __be16 type = skb->protocol;
2436 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2439 if (NAPI_GRO_CB(skb)->count == 1) {
2440 skb_shinfo(skb)->gso_size = 0;
2445 list_for_each_entry_rcu(ptype, head, list) {
2446 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2449 err = ptype->gro_complete(skb);
2455 WARN_ON(&ptype->list == head);
2457 return NET_RX_SUCCESS;
2461 return netif_receive_skb(skb);
2464 void napi_gro_flush(struct napi_struct *napi)
2466 struct sk_buff *skb, *next;
2468 for (skb = napi->gro_list; skb; skb = next) {
2471 napi_gro_complete(skb);
2474 napi->gro_count = 0;
2475 napi->gro_list = NULL;
2477 EXPORT_SYMBOL(napi_gro_flush);
2479 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2481 struct sk_buff **pp = NULL;
2482 struct packet_type *ptype;
2483 __be16 type = skb->protocol;
2484 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2487 enum gro_result ret;
2489 if (!(skb->dev->features & NETIF_F_GRO))
2492 if (skb_is_gso(skb) || skb_has_frags(skb))
2496 list_for_each_entry_rcu(ptype, head, list) {
2497 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2500 skb_set_network_header(skb, skb_gro_offset(skb));
2501 mac_len = skb->network_header - skb->mac_header;
2502 skb->mac_len = mac_len;
2503 NAPI_GRO_CB(skb)->same_flow = 0;
2504 NAPI_GRO_CB(skb)->flush = 0;
2505 NAPI_GRO_CB(skb)->free = 0;
2507 pp = ptype->gro_receive(&napi->gro_list, skb);
2512 if (&ptype->list == head)
2515 same_flow = NAPI_GRO_CB(skb)->same_flow;
2516 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2519 struct sk_buff *nskb = *pp;
2523 napi_gro_complete(nskb);
2530 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2534 NAPI_GRO_CB(skb)->count = 1;
2535 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2536 skb->next = napi->gro_list;
2537 napi->gro_list = skb;
2541 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2542 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2544 BUG_ON(skb->end - skb->tail < grow);
2546 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2549 skb->data_len -= grow;
2551 skb_shinfo(skb)->frags[0].page_offset += grow;
2552 skb_shinfo(skb)->frags[0].size -= grow;
2554 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2555 put_page(skb_shinfo(skb)->frags[0].page);
2556 memmove(skb_shinfo(skb)->frags,
2557 skb_shinfo(skb)->frags + 1,
2558 --skb_shinfo(skb)->nr_frags);
2569 EXPORT_SYMBOL(dev_gro_receive);
2572 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2576 if (netpoll_rx_on(skb))
2579 for (p = napi->gro_list; p; p = p->next) {
2580 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2581 && !compare_ether_header(skb_mac_header(p),
2582 skb_gro_mac_header(skb));
2583 NAPI_GRO_CB(p)->flush = 0;
2586 return dev_gro_receive(napi, skb);
2589 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2593 if (netif_receive_skb(skb))
2598 case GRO_MERGED_FREE:
2609 EXPORT_SYMBOL(napi_skb_finish);
2611 void skb_gro_reset_offset(struct sk_buff *skb)
2613 NAPI_GRO_CB(skb)->data_offset = 0;
2614 NAPI_GRO_CB(skb)->frag0 = NULL;
2615 NAPI_GRO_CB(skb)->frag0_len = 0;
2617 if (skb->mac_header == skb->tail &&
2618 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2619 NAPI_GRO_CB(skb)->frag0 =
2620 page_address(skb_shinfo(skb)->frags[0].page) +
2621 skb_shinfo(skb)->frags[0].page_offset;
2622 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2625 EXPORT_SYMBOL(skb_gro_reset_offset);
2627 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2629 skb_gro_reset_offset(skb);
2631 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2633 EXPORT_SYMBOL(napi_gro_receive);
2635 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2637 __skb_pull(skb, skb_headlen(skb));
2638 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2642 EXPORT_SYMBOL(napi_reuse_skb);
2644 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2646 struct sk_buff *skb = napi->skb;
2649 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2655 EXPORT_SYMBOL(napi_get_frags);
2657 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
2663 skb->protocol = eth_type_trans(skb, napi->dev);
2665 if (ret == GRO_HELD)
2666 skb_gro_pull(skb, -ETH_HLEN);
2667 else if (netif_receive_skb(skb))
2672 case GRO_MERGED_FREE:
2673 napi_reuse_skb(napi, skb);
2682 EXPORT_SYMBOL(napi_frags_finish);
2684 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2686 struct sk_buff *skb = napi->skb;
2693 skb_reset_mac_header(skb);
2694 skb_gro_reset_offset(skb);
2696 off = skb_gro_offset(skb);
2697 hlen = off + sizeof(*eth);
2698 eth = skb_gro_header_fast(skb, off);
2699 if (skb_gro_header_hard(skb, hlen)) {
2700 eth = skb_gro_header_slow(skb, hlen, off);
2701 if (unlikely(!eth)) {
2702 napi_reuse_skb(napi, skb);
2708 skb_gro_pull(skb, sizeof(*eth));
2711 * This works because the only protocols we care about don't require
2712 * special handling. We'll fix it up properly at the end.
2714 skb->protocol = eth->h_proto;
2719 EXPORT_SYMBOL(napi_frags_skb);
2721 gro_result_t napi_gro_frags(struct napi_struct *napi)
2723 struct sk_buff *skb = napi_frags_skb(napi);
2728 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2730 EXPORT_SYMBOL(napi_gro_frags);
2732 static int process_backlog(struct napi_struct *napi, int quota)
2735 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2736 unsigned long start_time = jiffies;
2738 napi->weight = weight_p;
2740 struct sk_buff *skb;
2742 local_irq_disable();
2743 skb = __skb_dequeue(&queue->input_pkt_queue);
2745 __napi_complete(napi);
2751 netif_receive_skb(skb);
2752 } while (++work < quota && jiffies == start_time);
2758 * __napi_schedule - schedule for receive
2759 * @n: entry to schedule
2761 * The entry's receive function will be scheduled to run
2763 void __napi_schedule(struct napi_struct *n)
2765 unsigned long flags;
2767 local_irq_save(flags);
2768 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2769 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2770 local_irq_restore(flags);
2772 EXPORT_SYMBOL(__napi_schedule);
2774 void __napi_complete(struct napi_struct *n)
2776 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2777 BUG_ON(n->gro_list);
2779 list_del(&n->poll_list);
2780 smp_mb__before_clear_bit();
2781 clear_bit(NAPI_STATE_SCHED, &n->state);
2783 EXPORT_SYMBOL(__napi_complete);
2785 void napi_complete(struct napi_struct *n)
2787 unsigned long flags;
2790 * don't let napi dequeue from the cpu poll list
2791 * just in case its running on a different cpu
2793 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2797 local_irq_save(flags);
2799 local_irq_restore(flags);
2801 EXPORT_SYMBOL(napi_complete);
2803 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2804 int (*poll)(struct napi_struct *, int), int weight)
2806 INIT_LIST_HEAD(&napi->poll_list);
2807 napi->gro_count = 0;
2808 napi->gro_list = NULL;
2811 napi->weight = weight;
2812 list_add(&napi->dev_list, &dev->napi_list);
2814 #ifdef CONFIG_NETPOLL
2815 spin_lock_init(&napi->poll_lock);
2816 napi->poll_owner = -1;
2818 set_bit(NAPI_STATE_SCHED, &napi->state);
2820 EXPORT_SYMBOL(netif_napi_add);
2822 void netif_napi_del(struct napi_struct *napi)
2824 struct sk_buff *skb, *next;
2826 list_del_init(&napi->dev_list);
2827 napi_free_frags(napi);
2829 for (skb = napi->gro_list; skb; skb = next) {
2835 napi->gro_list = NULL;
2836 napi->gro_count = 0;
2838 EXPORT_SYMBOL(netif_napi_del);
2841 static void net_rx_action(struct softirq_action *h)
2843 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2844 unsigned long time_limit = jiffies + 2;
2845 int budget = netdev_budget;
2848 local_irq_disable();
2850 while (!list_empty(list)) {
2851 struct napi_struct *n;
2854 /* If softirq window is exhuasted then punt.
2855 * Allow this to run for 2 jiffies since which will allow
2856 * an average latency of 1.5/HZ.
2858 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2863 /* Even though interrupts have been re-enabled, this
2864 * access is safe because interrupts can only add new
2865 * entries to the tail of this list, and only ->poll()
2866 * calls can remove this head entry from the list.
2868 n = list_entry(list->next, struct napi_struct, poll_list);
2870 have = netpoll_poll_lock(n);
2874 /* This NAPI_STATE_SCHED test is for avoiding a race
2875 * with netpoll's poll_napi(). Only the entity which
2876 * obtains the lock and sees NAPI_STATE_SCHED set will
2877 * actually make the ->poll() call. Therefore we avoid
2878 * accidently calling ->poll() when NAPI is not scheduled.
2881 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2882 work = n->poll(n, weight);
2886 WARN_ON_ONCE(work > weight);
2890 local_irq_disable();
2892 /* Drivers must not modify the NAPI state if they
2893 * consume the entire weight. In such cases this code
2894 * still "owns" the NAPI instance and therefore can
2895 * move the instance around on the list at-will.
2897 if (unlikely(work == weight)) {
2898 if (unlikely(napi_disable_pending(n))) {
2901 local_irq_disable();
2903 list_move_tail(&n->poll_list, list);
2906 netpoll_poll_unlock(have);
2911 #ifdef CONFIG_NET_DMA
2913 * There may not be any more sk_buffs coming right now, so push
2914 * any pending DMA copies to hardware
2916 dma_issue_pending_all();
2922 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2923 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2927 static gifconf_func_t *gifconf_list[NPROTO];
2930 * register_gifconf - register a SIOCGIF handler
2931 * @family: Address family
2932 * @gifconf: Function handler
2934 * Register protocol dependent address dumping routines. The handler
2935 * that is passed must not be freed or reused until it has been replaced
2936 * by another handler.
2938 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
2940 if (family >= NPROTO)
2942 gifconf_list[family] = gifconf;
2945 EXPORT_SYMBOL(register_gifconf);
2949 * Map an interface index to its name (SIOCGIFNAME)
2953 * We need this ioctl for efficient implementation of the
2954 * if_indextoname() function required by the IPv6 API. Without
2955 * it, we would have to search all the interfaces to find a
2959 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2961 struct net_device *dev;
2965 * Fetch the caller's info block.
2968 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2972 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
2978 strcpy(ifr.ifr_name, dev->name);
2981 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2987 * Perform a SIOCGIFCONF call. This structure will change
2988 * size eventually, and there is nothing I can do about it.
2989 * Thus we will need a 'compatibility mode'.
2992 static int dev_ifconf(struct net *net, char __user *arg)
2995 struct net_device *dev;
3002 * Fetch the caller's info block.
3005 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3012 * Loop over the interfaces, and write an info block for each.
3016 for_each_netdev(net, dev) {
3017 for (i = 0; i < NPROTO; i++) {
3018 if (gifconf_list[i]) {
3021 done = gifconf_list[i](dev, NULL, 0);
3023 done = gifconf_list[i](dev, pos + total,
3033 * All done. Write the updated control block back to the caller.
3035 ifc.ifc_len = total;
3038 * Both BSD and Solaris return 0 here, so we do too.
3040 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3043 #ifdef CONFIG_PROC_FS
3045 * This is invoked by the /proc filesystem handler to display a device
3048 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3049 __acquires(dev_base_lock)
3051 struct net *net = seq_file_net(seq);
3053 struct net_device *dev;
3055 read_lock(&dev_base_lock);
3057 return SEQ_START_TOKEN;
3060 for_each_netdev(net, dev)
3067 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3069 struct net *net = seq_file_net(seq);
3071 return v == SEQ_START_TOKEN ?
3072 first_net_device(net) : next_net_device((struct net_device *)v);
3075 void dev_seq_stop(struct seq_file *seq, void *v)
3076 __releases(dev_base_lock)
3078 read_unlock(&dev_base_lock);
3081 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3083 const struct net_device_stats *stats = dev_get_stats(dev);
3085 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3086 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3087 dev->name, stats->rx_bytes, stats->rx_packets,
3089 stats->rx_dropped + stats->rx_missed_errors,
3090 stats->rx_fifo_errors,
3091 stats->rx_length_errors + stats->rx_over_errors +
3092 stats->rx_crc_errors + stats->rx_frame_errors,
3093 stats->rx_compressed, stats->multicast,
3094 stats->tx_bytes, stats->tx_packets,
3095 stats->tx_errors, stats->tx_dropped,
3096 stats->tx_fifo_errors, stats->collisions,
3097 stats->tx_carrier_errors +
3098 stats->tx_aborted_errors +
3099 stats->tx_window_errors +
3100 stats->tx_heartbeat_errors,
3101 stats->tx_compressed);
3105 * Called from the PROCfs module. This now uses the new arbitrary sized
3106 * /proc/net interface to create /proc/net/dev
3108 static int dev_seq_show(struct seq_file *seq, void *v)
3110 if (v == SEQ_START_TOKEN)
3111 seq_puts(seq, "Inter-| Receive "
3113 " face |bytes packets errs drop fifo frame "
3114 "compressed multicast|bytes packets errs "
3115 "drop fifo colls carrier compressed\n");
3117 dev_seq_printf_stats(seq, v);
3121 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3123 struct netif_rx_stats *rc = NULL;
3125 while (*pos < nr_cpu_ids)
3126 if (cpu_online(*pos)) {
3127 rc = &per_cpu(netdev_rx_stat, *pos);
3134 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3136 return softnet_get_online(pos);
3139 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3142 return softnet_get_online(pos);
3145 static void softnet_seq_stop(struct seq_file *seq, void *v)
3149 static int softnet_seq_show(struct seq_file *seq, void *v)
3151 struct netif_rx_stats *s = v;
3153 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3154 s->total, s->dropped, s->time_squeeze, 0,
3155 0, 0, 0, 0, /* was fastroute */
3160 static const struct seq_operations dev_seq_ops = {
3161 .start = dev_seq_start,
3162 .next = dev_seq_next,
3163 .stop = dev_seq_stop,
3164 .show = dev_seq_show,
3167 static int dev_seq_open(struct inode *inode, struct file *file)
3169 return seq_open_net(inode, file, &dev_seq_ops,
3170 sizeof(struct seq_net_private));
3173 static const struct file_operations dev_seq_fops = {
3174 .owner = THIS_MODULE,
3175 .open = dev_seq_open,
3177 .llseek = seq_lseek,
3178 .release = seq_release_net,
3181 static const struct seq_operations softnet_seq_ops = {
3182 .start = softnet_seq_start,
3183 .next = softnet_seq_next,
3184 .stop = softnet_seq_stop,
3185 .show = softnet_seq_show,
3188 static int softnet_seq_open(struct inode *inode, struct file *file)
3190 return seq_open(file, &softnet_seq_ops);
3193 static const struct file_operations softnet_seq_fops = {
3194 .owner = THIS_MODULE,
3195 .open = softnet_seq_open,
3197 .llseek = seq_lseek,
3198 .release = seq_release,
3201 static void *ptype_get_idx(loff_t pos)
3203 struct packet_type *pt = NULL;
3207 list_for_each_entry_rcu(pt, &ptype_all, list) {
3213 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3214 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3223 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3227 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3230 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3232 struct packet_type *pt;
3233 struct list_head *nxt;
3237 if (v == SEQ_START_TOKEN)
3238 return ptype_get_idx(0);
3241 nxt = pt->list.next;
3242 if (pt->type == htons(ETH_P_ALL)) {
3243 if (nxt != &ptype_all)
3246 nxt = ptype_base[0].next;
3248 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3250 while (nxt == &ptype_base[hash]) {
3251 if (++hash >= PTYPE_HASH_SIZE)
3253 nxt = ptype_base[hash].next;
3256 return list_entry(nxt, struct packet_type, list);
3259 static void ptype_seq_stop(struct seq_file *seq, void *v)
3265 static int ptype_seq_show(struct seq_file *seq, void *v)
3267 struct packet_type *pt = v;
3269 if (v == SEQ_START_TOKEN)
3270 seq_puts(seq, "Type Device Function\n");
3271 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3272 if (pt->type == htons(ETH_P_ALL))
3273 seq_puts(seq, "ALL ");
3275 seq_printf(seq, "%04x", ntohs(pt->type));
3277 seq_printf(seq, " %-8s %pF\n",
3278 pt->dev ? pt->dev->name : "", pt->func);
3284 static const struct seq_operations ptype_seq_ops = {
3285 .start = ptype_seq_start,
3286 .next = ptype_seq_next,
3287 .stop = ptype_seq_stop,
3288 .show = ptype_seq_show,
3291 static int ptype_seq_open(struct inode *inode, struct file *file)
3293 return seq_open_net(inode, file, &ptype_seq_ops,
3294 sizeof(struct seq_net_private));
3297 static const struct file_operations ptype_seq_fops = {
3298 .owner = THIS_MODULE,
3299 .open = ptype_seq_open,
3301 .llseek = seq_lseek,
3302 .release = seq_release_net,
3306 static int __net_init dev_proc_net_init(struct net *net)
3310 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3312 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3314 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3317 if (wext_proc_init(net))
3323 proc_net_remove(net, "ptype");
3325 proc_net_remove(net, "softnet_stat");
3327 proc_net_remove(net, "dev");
3331 static void __net_exit dev_proc_net_exit(struct net *net)
3333 wext_proc_exit(net);
3335 proc_net_remove(net, "ptype");
3336 proc_net_remove(net, "softnet_stat");
3337 proc_net_remove(net, "dev");
3340 static struct pernet_operations __net_initdata dev_proc_ops = {
3341 .init = dev_proc_net_init,
3342 .exit = dev_proc_net_exit,
3345 static int __init dev_proc_init(void)
3347 return register_pernet_subsys(&dev_proc_ops);
3350 #define dev_proc_init() 0
3351 #endif /* CONFIG_PROC_FS */
3355 * netdev_set_master - set up master/slave pair
3356 * @slave: slave device
3357 * @master: new master device
3359 * Changes the master device of the slave. Pass %NULL to break the
3360 * bonding. The caller must hold the RTNL semaphore. On a failure
3361 * a negative errno code is returned. On success the reference counts
3362 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3363 * function returns zero.
3365 int netdev_set_master(struct net_device *slave, struct net_device *master)
3367 struct net_device *old = slave->master;
3377 slave->master = master;
3385 slave->flags |= IFF_SLAVE;
3387 slave->flags &= ~IFF_SLAVE;
3389 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3392 EXPORT_SYMBOL(netdev_set_master);
3394 static void dev_change_rx_flags(struct net_device *dev, int flags)
3396 const struct net_device_ops *ops = dev->netdev_ops;
3398 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3399 ops->ndo_change_rx_flags(dev, flags);
3402 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3404 unsigned short old_flags = dev->flags;
3410 dev->flags |= IFF_PROMISC;
3411 dev->promiscuity += inc;
3412 if (dev->promiscuity == 0) {
3415 * If inc causes overflow, untouch promisc and return error.
3418 dev->flags &= ~IFF_PROMISC;
3420 dev->promiscuity -= inc;
3421 printk(KERN_WARNING "%s: promiscuity touches roof, "
3422 "set promiscuity failed, promiscuity feature "
3423 "of device might be broken.\n", dev->name);
3427 if (dev->flags != old_flags) {
3428 printk(KERN_INFO "device %s %s promiscuous mode\n",
3429 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3431 if (audit_enabled) {
3432 current_uid_gid(&uid, &gid);
3433 audit_log(current->audit_context, GFP_ATOMIC,
3434 AUDIT_ANOM_PROMISCUOUS,
3435 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3436 dev->name, (dev->flags & IFF_PROMISC),
3437 (old_flags & IFF_PROMISC),
3438 audit_get_loginuid(current),
3440 audit_get_sessionid(current));
3443 dev_change_rx_flags(dev, IFF_PROMISC);
3449 * dev_set_promiscuity - update promiscuity count on a device
3453 * Add or remove promiscuity from a device. While the count in the device
3454 * remains above zero the interface remains promiscuous. Once it hits zero
3455 * the device reverts back to normal filtering operation. A negative inc
3456 * value is used to drop promiscuity on the device.
3457 * Return 0 if successful or a negative errno code on error.
3459 int dev_set_promiscuity(struct net_device *dev, int inc)
3461 unsigned short old_flags = dev->flags;
3464 err = __dev_set_promiscuity(dev, inc);
3467 if (dev->flags != old_flags)
3468 dev_set_rx_mode(dev);
3471 EXPORT_SYMBOL(dev_set_promiscuity);
3474 * dev_set_allmulti - update allmulti count on a device
3478 * Add or remove reception of all multicast frames to a device. While the
3479 * count in the device remains above zero the interface remains listening
3480 * to all interfaces. Once it hits zero the device reverts back to normal
3481 * filtering operation. A negative @inc value is used to drop the counter
3482 * when releasing a resource needing all multicasts.
3483 * Return 0 if successful or a negative errno code on error.
3486 int dev_set_allmulti(struct net_device *dev, int inc)
3488 unsigned short old_flags = dev->flags;
3492 dev->flags |= IFF_ALLMULTI;
3493 dev->allmulti += inc;
3494 if (dev->allmulti == 0) {
3497 * If inc causes overflow, untouch allmulti and return error.
3500 dev->flags &= ~IFF_ALLMULTI;
3502 dev->allmulti -= inc;
3503 printk(KERN_WARNING "%s: allmulti touches roof, "
3504 "set allmulti failed, allmulti feature of "
3505 "device might be broken.\n", dev->name);
3509 if (dev->flags ^ old_flags) {
3510 dev_change_rx_flags(dev, IFF_ALLMULTI);
3511 dev_set_rx_mode(dev);
3515 EXPORT_SYMBOL(dev_set_allmulti);
3518 * Upload unicast and multicast address lists to device and
3519 * configure RX filtering. When the device doesn't support unicast
3520 * filtering it is put in promiscuous mode while unicast addresses
3523 void __dev_set_rx_mode(struct net_device *dev)
3525 const struct net_device_ops *ops = dev->netdev_ops;
3527 /* dev_open will call this function so the list will stay sane. */
3528 if (!(dev->flags&IFF_UP))
3531 if (!netif_device_present(dev))
3534 if (ops->ndo_set_rx_mode)
3535 ops->ndo_set_rx_mode(dev);
3537 /* Unicast addresses changes may only happen under the rtnl,
3538 * therefore calling __dev_set_promiscuity here is safe.
3540 if (dev->uc.count > 0 && !dev->uc_promisc) {
3541 __dev_set_promiscuity(dev, 1);
3542 dev->uc_promisc = 1;
3543 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3544 __dev_set_promiscuity(dev, -1);
3545 dev->uc_promisc = 0;
3548 if (ops->ndo_set_multicast_list)
3549 ops->ndo_set_multicast_list(dev);
3553 void dev_set_rx_mode(struct net_device *dev)
3555 netif_addr_lock_bh(dev);
3556 __dev_set_rx_mode(dev);
3557 netif_addr_unlock_bh(dev);
3560 /* hw addresses list handling functions */
3562 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3563 int addr_len, unsigned char addr_type)
3565 struct netdev_hw_addr *ha;
3568 if (addr_len > MAX_ADDR_LEN)
3571 list_for_each_entry(ha, &list->list, list) {
3572 if (!memcmp(ha->addr, addr, addr_len) &&
3573 ha->type == addr_type) {
3580 alloc_size = sizeof(*ha);
3581 if (alloc_size < L1_CACHE_BYTES)
3582 alloc_size = L1_CACHE_BYTES;
3583 ha = kmalloc(alloc_size, GFP_ATOMIC);
3586 memcpy(ha->addr, addr, addr_len);
3587 ha->type = addr_type;
3590 list_add_tail_rcu(&ha->list, &list->list);
3595 static void ha_rcu_free(struct rcu_head *head)
3597 struct netdev_hw_addr *ha;
3599 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3603 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3604 int addr_len, unsigned char addr_type)
3606 struct netdev_hw_addr *ha;
3608 list_for_each_entry(ha, &list->list, list) {
3609 if (!memcmp(ha->addr, addr, addr_len) &&
3610 (ha->type == addr_type || !addr_type)) {
3613 list_del_rcu(&ha->list);
3614 call_rcu(&ha->rcu_head, ha_rcu_free);
3622 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3623 struct netdev_hw_addr_list *from_list,
3625 unsigned char addr_type)
3628 struct netdev_hw_addr *ha, *ha2;
3631 list_for_each_entry(ha, &from_list->list, list) {
3632 type = addr_type ? addr_type : ha->type;
3633 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3640 list_for_each_entry(ha2, &from_list->list, list) {
3643 type = addr_type ? addr_type : ha2->type;
3644 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3649 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3650 struct netdev_hw_addr_list *from_list,
3652 unsigned char addr_type)
3654 struct netdev_hw_addr *ha;
3657 list_for_each_entry(ha, &from_list->list, list) {
3658 type = addr_type ? addr_type : ha->type;
3659 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3663 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3664 struct netdev_hw_addr_list *from_list,
3668 struct netdev_hw_addr *ha, *tmp;
3670 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3672 err = __hw_addr_add(to_list, ha->addr,
3673 addr_len, ha->type);
3678 } else if (ha->refcount == 1) {
3679 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3680 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3686 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3687 struct netdev_hw_addr_list *from_list,
3690 struct netdev_hw_addr *ha, *tmp;
3692 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3694 __hw_addr_del(to_list, ha->addr,
3695 addr_len, ha->type);
3697 __hw_addr_del(from_list, ha->addr,
3698 addr_len, ha->type);
3703 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3705 struct netdev_hw_addr *ha, *tmp;
3707 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3708 list_del_rcu(&ha->list);
3709 call_rcu(&ha->rcu_head, ha_rcu_free);
3714 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3716 INIT_LIST_HEAD(&list->list);
3720 /* Device addresses handling functions */
3722 static void dev_addr_flush(struct net_device *dev)
3724 /* rtnl_mutex must be held here */
3726 __hw_addr_flush(&dev->dev_addrs);
3727 dev->dev_addr = NULL;
3730 static int dev_addr_init(struct net_device *dev)
3732 unsigned char addr[MAX_ADDR_LEN];
3733 struct netdev_hw_addr *ha;
3736 /* rtnl_mutex must be held here */
3738 __hw_addr_init(&dev->dev_addrs);
3739 memset(addr, 0, sizeof(addr));
3740 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3741 NETDEV_HW_ADDR_T_LAN);
3744 * Get the first (previously created) address from the list
3745 * and set dev_addr pointer to this location.
3747 ha = list_first_entry(&dev->dev_addrs.list,
3748 struct netdev_hw_addr, list);
3749 dev->dev_addr = ha->addr;
3755 * dev_addr_add - Add a device address
3757 * @addr: address to add
3758 * @addr_type: address type
3760 * Add a device address to the device or increase the reference count if
3761 * it already exists.
3763 * The caller must hold the rtnl_mutex.
3765 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3766 unsigned char addr_type)
3772 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3774 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3777 EXPORT_SYMBOL(dev_addr_add);
3780 * dev_addr_del - Release a device address.
3782 * @addr: address to delete
3783 * @addr_type: address type
3785 * Release reference to a device address and remove it from the device
3786 * if the reference count drops to zero.
3788 * The caller must hold the rtnl_mutex.
3790 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3791 unsigned char addr_type)
3794 struct netdev_hw_addr *ha;
3799 * We can not remove the first address from the list because
3800 * dev->dev_addr points to that.
3802 ha = list_first_entry(&dev->dev_addrs.list,
3803 struct netdev_hw_addr, list);
3804 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3807 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3810 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3813 EXPORT_SYMBOL(dev_addr_del);
3816 * dev_addr_add_multiple - Add device addresses from another device
3817 * @to_dev: device to which addresses will be added
3818 * @from_dev: device from which addresses will be added
3819 * @addr_type: address type - 0 means type will be used from from_dev
3821 * Add device addresses of the one device to another.
3823 * The caller must hold the rtnl_mutex.
3825 int dev_addr_add_multiple(struct net_device *to_dev,
3826 struct net_device *from_dev,
3827 unsigned char addr_type)
3833 if (from_dev->addr_len != to_dev->addr_len)
3835 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3836 to_dev->addr_len, addr_type);
3838 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3841 EXPORT_SYMBOL(dev_addr_add_multiple);
3844 * dev_addr_del_multiple - Delete device addresses by another device
3845 * @to_dev: device where the addresses will be deleted
3846 * @from_dev: device by which addresses the addresses will be deleted
3847 * @addr_type: address type - 0 means type will used from from_dev
3849 * Deletes addresses in to device by the list of addresses in from device.
3851 * The caller must hold the rtnl_mutex.
3853 int dev_addr_del_multiple(struct net_device *to_dev,
3854 struct net_device *from_dev,
3855 unsigned char addr_type)
3859 if (from_dev->addr_len != to_dev->addr_len)
3861 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3862 to_dev->addr_len, addr_type);
3863 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3866 EXPORT_SYMBOL(dev_addr_del_multiple);
3868 /* multicast addresses handling functions */
3870 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3871 void *addr, int alen, int glbl)
3873 struct dev_addr_list *da;
3875 for (; (da = *list) != NULL; list = &da->next) {
3876 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3877 alen == da->da_addrlen) {
3879 int old_glbl = da->da_gusers;
3896 int __dev_addr_add(struct dev_addr_list **list, int *count,
3897 void *addr, int alen, int glbl)
3899 struct dev_addr_list *da;
3901 for (da = *list; da != NULL; da = da->next) {
3902 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3903 da->da_addrlen == alen) {
3905 int old_glbl = da->da_gusers;
3915 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3918 memcpy(da->da_addr, addr, alen);
3919 da->da_addrlen = alen;
3921 da->da_gusers = glbl ? 1 : 0;
3929 * dev_unicast_delete - Release secondary unicast address.
3931 * @addr: address to delete
3933 * Release reference to a secondary unicast address and remove it
3934 * from the device if the reference count drops to zero.
3936 * The caller must hold the rtnl_mutex.
3938 int dev_unicast_delete(struct net_device *dev, void *addr)
3944 netif_addr_lock_bh(dev);
3945 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
3946 NETDEV_HW_ADDR_T_UNICAST);
3948 __dev_set_rx_mode(dev);
3949 netif_addr_unlock_bh(dev);
3952 EXPORT_SYMBOL(dev_unicast_delete);
3955 * dev_unicast_add - add a secondary unicast address
3957 * @addr: address to add
3959 * Add a secondary unicast address to the device or increase
3960 * the reference count if it already exists.
3962 * The caller must hold the rtnl_mutex.
3964 int dev_unicast_add(struct net_device *dev, void *addr)
3970 netif_addr_lock_bh(dev);
3971 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
3972 NETDEV_HW_ADDR_T_UNICAST);
3974 __dev_set_rx_mode(dev);
3975 netif_addr_unlock_bh(dev);
3978 EXPORT_SYMBOL(dev_unicast_add);
3980 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3981 struct dev_addr_list **from, int *from_count)
3983 struct dev_addr_list *da, *next;
3987 while (da != NULL) {
3989 if (!da->da_synced) {
3990 err = __dev_addr_add(to, to_count,
3991 da->da_addr, da->da_addrlen, 0);
3996 } else if (da->da_users == 1) {
3997 __dev_addr_delete(to, to_count,
3998 da->da_addr, da->da_addrlen, 0);
3999 __dev_addr_delete(from, from_count,
4000 da->da_addr, da->da_addrlen, 0);
4006 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4008 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4009 struct dev_addr_list **from, int *from_count)
4011 struct dev_addr_list *da, *next;
4014 while (da != NULL) {
4016 if (da->da_synced) {
4017 __dev_addr_delete(to, to_count,
4018 da->da_addr, da->da_addrlen, 0);
4020 __dev_addr_delete(from, from_count,
4021 da->da_addr, da->da_addrlen, 0);
4026 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4029 * dev_unicast_sync - Synchronize device's unicast list to another device
4030 * @to: destination device
4031 * @from: source device
4033 * Add newly added addresses to the destination device and release
4034 * addresses that have no users left. The source device must be
4035 * locked by netif_tx_lock_bh.
4037 * This function is intended to be called from the dev->set_rx_mode
4038 * function of layered software devices.
4040 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4044 if (to->addr_len != from->addr_len)
4047 netif_addr_lock_bh(to);
4048 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4050 __dev_set_rx_mode(to);
4051 netif_addr_unlock_bh(to);
4054 EXPORT_SYMBOL(dev_unicast_sync);
4057 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4058 * @to: destination device
4059 * @from: source device
4061 * Remove all addresses that were added to the destination device by
4062 * dev_unicast_sync(). This function is intended to be called from the
4063 * dev->stop function of layered software devices.
4065 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4067 if (to->addr_len != from->addr_len)
4070 netif_addr_lock_bh(from);
4071 netif_addr_lock(to);
4072 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4073 __dev_set_rx_mode(to);
4074 netif_addr_unlock(to);
4075 netif_addr_unlock_bh(from);
4077 EXPORT_SYMBOL(dev_unicast_unsync);
4079 static void dev_unicast_flush(struct net_device *dev)
4081 netif_addr_lock_bh(dev);
4082 __hw_addr_flush(&dev->uc);
4083 netif_addr_unlock_bh(dev);
4086 static void dev_unicast_init(struct net_device *dev)
4088 __hw_addr_init(&dev->uc);
4092 static void __dev_addr_discard(struct dev_addr_list **list)
4094 struct dev_addr_list *tmp;
4096 while (*list != NULL) {
4099 if (tmp->da_users > tmp->da_gusers)
4100 printk("__dev_addr_discard: address leakage! "
4101 "da_users=%d\n", tmp->da_users);
4106 static void dev_addr_discard(struct net_device *dev)
4108 netif_addr_lock_bh(dev);
4110 __dev_addr_discard(&dev->mc_list);
4113 netif_addr_unlock_bh(dev);
4117 * dev_get_flags - get flags reported to userspace
4120 * Get the combination of flag bits exported through APIs to userspace.
4122 unsigned dev_get_flags(const struct net_device *dev)
4126 flags = (dev->flags & ~(IFF_PROMISC |
4131 (dev->gflags & (IFF_PROMISC |
4134 if (netif_running(dev)) {
4135 if (netif_oper_up(dev))
4136 flags |= IFF_RUNNING;
4137 if (netif_carrier_ok(dev))
4138 flags |= IFF_LOWER_UP;
4139 if (netif_dormant(dev))
4140 flags |= IFF_DORMANT;
4145 EXPORT_SYMBOL(dev_get_flags);
4148 * dev_change_flags - change device settings
4150 * @flags: device state flags
4152 * Change settings on device based state flags. The flags are
4153 * in the userspace exported format.
4155 int dev_change_flags(struct net_device *dev, unsigned flags)
4158 int old_flags = dev->flags;
4163 * Set the flags on our device.
4166 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4167 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4169 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4173 * Load in the correct multicast list now the flags have changed.
4176 if ((old_flags ^ flags) & IFF_MULTICAST)
4177 dev_change_rx_flags(dev, IFF_MULTICAST);
4179 dev_set_rx_mode(dev);
4182 * Have we downed the interface. We handle IFF_UP ourselves
4183 * according to user attempts to set it, rather than blindly
4188 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4189 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4192 dev_set_rx_mode(dev);
4195 if (dev->flags & IFF_UP &&
4196 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4198 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4200 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4201 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4203 dev->gflags ^= IFF_PROMISC;
4204 dev_set_promiscuity(dev, inc);
4207 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4208 is important. Some (broken) drivers set IFF_PROMISC, when
4209 IFF_ALLMULTI is requested not asking us and not reporting.
4211 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4212 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4214 dev->gflags ^= IFF_ALLMULTI;
4215 dev_set_allmulti(dev, inc);
4218 /* Exclude state transition flags, already notified */
4219 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4221 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4225 EXPORT_SYMBOL(dev_change_flags);
4228 * dev_set_mtu - Change maximum transfer unit
4230 * @new_mtu: new transfer unit
4232 * Change the maximum transfer size of the network device.
4234 int dev_set_mtu(struct net_device *dev, int new_mtu)
4236 const struct net_device_ops *ops = dev->netdev_ops;
4239 if (new_mtu == dev->mtu)
4242 /* MTU must be positive. */
4246 if (!netif_device_present(dev))
4250 if (ops->ndo_change_mtu)
4251 err = ops->ndo_change_mtu(dev, new_mtu);
4255 if (!err && dev->flags & IFF_UP)
4256 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4259 EXPORT_SYMBOL(dev_set_mtu);
4262 * dev_set_mac_address - Change Media Access Control Address
4266 * Change the hardware (MAC) address of the device
4268 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4270 const struct net_device_ops *ops = dev->netdev_ops;
4273 if (!ops->ndo_set_mac_address)
4275 if (sa->sa_family != dev->type)
4277 if (!netif_device_present(dev))
4279 err = ops->ndo_set_mac_address(dev, sa);
4281 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4284 EXPORT_SYMBOL(dev_set_mac_address);
4287 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
4289 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4292 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4298 case SIOCGIFFLAGS: /* Get interface flags */
4299 ifr->ifr_flags = (short) dev_get_flags(dev);
4302 case SIOCGIFMETRIC: /* Get the metric on the interface
4303 (currently unused) */
4304 ifr->ifr_metric = 0;
4307 case SIOCGIFMTU: /* Get the MTU of a device */
4308 ifr->ifr_mtu = dev->mtu;
4313 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4315 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4316 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4317 ifr->ifr_hwaddr.sa_family = dev->type;
4325 ifr->ifr_map.mem_start = dev->mem_start;
4326 ifr->ifr_map.mem_end = dev->mem_end;
4327 ifr->ifr_map.base_addr = dev->base_addr;
4328 ifr->ifr_map.irq = dev->irq;
4329 ifr->ifr_map.dma = dev->dma;
4330 ifr->ifr_map.port = dev->if_port;
4334 ifr->ifr_ifindex = dev->ifindex;
4338 ifr->ifr_qlen = dev->tx_queue_len;
4342 /* dev_ioctl() should ensure this case
4354 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4356 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4359 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4360 const struct net_device_ops *ops;
4365 ops = dev->netdev_ops;
4368 case SIOCSIFFLAGS: /* Set interface flags */
4369 return dev_change_flags(dev, ifr->ifr_flags);
4371 case SIOCSIFMETRIC: /* Set the metric on the interface
4372 (currently unused) */
4375 case SIOCSIFMTU: /* Set the MTU of a device */
4376 return dev_set_mtu(dev, ifr->ifr_mtu);
4379 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4381 case SIOCSIFHWBROADCAST:
4382 if (ifr->ifr_hwaddr.sa_family != dev->type)
4384 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4385 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4386 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4390 if (ops->ndo_set_config) {
4391 if (!netif_device_present(dev))
4393 return ops->ndo_set_config(dev, &ifr->ifr_map);
4398 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4399 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4401 if (!netif_device_present(dev))
4403 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4407 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4408 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4410 if (!netif_device_present(dev))
4412 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4416 if (ifr->ifr_qlen < 0)
4418 dev->tx_queue_len = ifr->ifr_qlen;
4422 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4423 return dev_change_name(dev, ifr->ifr_newname);
4426 * Unknown or private ioctl
4429 if ((cmd >= SIOCDEVPRIVATE &&
4430 cmd <= SIOCDEVPRIVATE + 15) ||
4431 cmd == SIOCBONDENSLAVE ||
4432 cmd == SIOCBONDRELEASE ||
4433 cmd == SIOCBONDSETHWADDR ||
4434 cmd == SIOCBONDSLAVEINFOQUERY ||
4435 cmd == SIOCBONDINFOQUERY ||
4436 cmd == SIOCBONDCHANGEACTIVE ||
4437 cmd == SIOCGMIIPHY ||
4438 cmd == SIOCGMIIREG ||
4439 cmd == SIOCSMIIREG ||
4440 cmd == SIOCBRADDIF ||
4441 cmd == SIOCBRDELIF ||
4442 cmd == SIOCSHWTSTAMP ||
4443 cmd == SIOCWANDEV) {
4445 if (ops->ndo_do_ioctl) {
4446 if (netif_device_present(dev))
4447 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4459 * This function handles all "interface"-type I/O control requests. The actual
4460 * 'doing' part of this is dev_ifsioc above.
4464 * dev_ioctl - network device ioctl
4465 * @net: the applicable net namespace
4466 * @cmd: command to issue
4467 * @arg: pointer to a struct ifreq in user space
4469 * Issue ioctl functions to devices. This is normally called by the
4470 * user space syscall interfaces but can sometimes be useful for
4471 * other purposes. The return value is the return from the syscall if
4472 * positive or a negative errno code on error.
4475 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4481 /* One special case: SIOCGIFCONF takes ifconf argument
4482 and requires shared lock, because it sleeps writing
4486 if (cmd == SIOCGIFCONF) {
4488 ret = dev_ifconf(net, (char __user *) arg);
4492 if (cmd == SIOCGIFNAME)
4493 return dev_ifname(net, (struct ifreq __user *)arg);
4495 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4498 ifr.ifr_name[IFNAMSIZ-1] = 0;
4500 colon = strchr(ifr.ifr_name, ':');
4505 * See which interface the caller is talking about.
4510 * These ioctl calls:
4511 * - can be done by all.
4512 * - atomic and do not require locking.
4523 dev_load(net, ifr.ifr_name);
4524 read_lock(&dev_base_lock);
4525 ret = dev_ifsioc_locked(net, &ifr, cmd);
4526 read_unlock(&dev_base_lock);
4530 if (copy_to_user(arg, &ifr,
4531 sizeof(struct ifreq)))
4537 dev_load(net, ifr.ifr_name);
4539 ret = dev_ethtool(net, &ifr);
4544 if (copy_to_user(arg, &ifr,
4545 sizeof(struct ifreq)))
4551 * These ioctl calls:
4552 * - require superuser power.
4553 * - require strict serialization.
4559 if (!capable(CAP_NET_ADMIN))
4561 dev_load(net, ifr.ifr_name);
4563 ret = dev_ifsioc(net, &ifr, cmd);
4568 if (copy_to_user(arg, &ifr,
4569 sizeof(struct ifreq)))
4575 * These ioctl calls:
4576 * - require superuser power.
4577 * - require strict serialization.
4578 * - do not return a value
4588 case SIOCSIFHWBROADCAST:
4591 case SIOCBONDENSLAVE:
4592 case SIOCBONDRELEASE:
4593 case SIOCBONDSETHWADDR:
4594 case SIOCBONDCHANGEACTIVE:
4598 if (!capable(CAP_NET_ADMIN))
4601 case SIOCBONDSLAVEINFOQUERY:
4602 case SIOCBONDINFOQUERY:
4603 dev_load(net, ifr.ifr_name);
4605 ret = dev_ifsioc(net, &ifr, cmd);
4610 /* Get the per device memory space. We can add this but
4611 * currently do not support it */
4613 /* Set the per device memory buffer space.
4614 * Not applicable in our case */
4619 * Unknown or private ioctl.
4622 if (cmd == SIOCWANDEV ||
4623 (cmd >= SIOCDEVPRIVATE &&
4624 cmd <= SIOCDEVPRIVATE + 15)) {
4625 dev_load(net, ifr.ifr_name);
4627 ret = dev_ifsioc(net, &ifr, cmd);
4629 if (!ret && copy_to_user(arg, &ifr,
4630 sizeof(struct ifreq)))
4634 /* Take care of Wireless Extensions */
4635 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4636 return wext_handle_ioctl(net, &ifr, cmd, arg);
4643 * dev_new_index - allocate an ifindex
4644 * @net: the applicable net namespace
4646 * Returns a suitable unique value for a new device interface
4647 * number. The caller must hold the rtnl semaphore or the
4648 * dev_base_lock to be sure it remains unique.
4650 static int dev_new_index(struct net *net)
4656 if (!__dev_get_by_index(net, ifindex))
4661 /* Delayed registration/unregisteration */
4662 static LIST_HEAD(net_todo_list);
4664 static void net_set_todo(struct net_device *dev)
4666 list_add_tail(&dev->todo_list, &net_todo_list);
4669 static void rollback_registered_many(struct list_head *head)
4671 struct net_device *dev;
4673 BUG_ON(dev_boot_phase);
4676 list_for_each_entry(dev, head, unreg_list) {
4677 /* Some devices call without registering
4678 * for initialization unwind.
4680 if (dev->reg_state == NETREG_UNINITIALIZED) {
4681 pr_debug("unregister_netdevice: device %s/%p never "
4682 "was registered\n", dev->name, dev);
4688 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4690 /* If device is running, close it first. */
4693 /* And unlink it from device chain. */
4694 unlist_netdevice(dev);
4696 dev->reg_state = NETREG_UNREGISTERING;
4701 list_for_each_entry(dev, head, unreg_list) {
4702 /* Shutdown queueing discipline. */
4706 /* Notify protocols, that we are about to destroy
4707 this device. They should clean all the things.
4709 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4712 * Flush the unicast and multicast chains
4714 dev_unicast_flush(dev);
4715 dev_addr_discard(dev);
4717 if (dev->netdev_ops->ndo_uninit)
4718 dev->netdev_ops->ndo_uninit(dev);
4720 /* Notifier chain MUST detach us from master device. */
4721 WARN_ON(dev->master);
4723 /* Remove entries from kobject tree */
4724 netdev_unregister_kobject(dev);
4729 list_for_each_entry(dev, head, unreg_list)
4733 static void rollback_registered(struct net_device *dev)
4737 list_add(&dev->unreg_list, &single);
4738 rollback_registered_many(&single);
4741 static void __netdev_init_queue_locks_one(struct net_device *dev,
4742 struct netdev_queue *dev_queue,
4745 spin_lock_init(&dev_queue->_xmit_lock);
4746 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4747 dev_queue->xmit_lock_owner = -1;
4750 static void netdev_init_queue_locks(struct net_device *dev)
4752 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4753 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4756 unsigned long netdev_fix_features(unsigned long features, const char *name)
4758 /* Fix illegal SG+CSUM combinations. */
4759 if ((features & NETIF_F_SG) &&
4760 !(features & NETIF_F_ALL_CSUM)) {
4762 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4763 "checksum feature.\n", name);
4764 features &= ~NETIF_F_SG;
4767 /* TSO requires that SG is present as well. */
4768 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4770 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4771 "SG feature.\n", name);
4772 features &= ~NETIF_F_TSO;
4775 if (features & NETIF_F_UFO) {
4776 if (!(features & NETIF_F_GEN_CSUM)) {
4778 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4779 "since no NETIF_F_HW_CSUM feature.\n",
4781 features &= ~NETIF_F_UFO;
4784 if (!(features & NETIF_F_SG)) {
4786 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4787 "since no NETIF_F_SG feature.\n", name);
4788 features &= ~NETIF_F_UFO;
4794 EXPORT_SYMBOL(netdev_fix_features);
4797 * register_netdevice - register a network device
4798 * @dev: device to register
4800 * Take a completed network device structure and add it to the kernel
4801 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4802 * chain. 0 is returned on success. A negative errno code is returned
4803 * on a failure to set up the device, or if the name is a duplicate.
4805 * Callers must hold the rtnl semaphore. You may want
4806 * register_netdev() instead of this.
4809 * The locking appears insufficient to guarantee two parallel registers
4810 * will not get the same name.
4813 int register_netdevice(struct net_device *dev)
4815 struct hlist_head *head;
4816 struct hlist_node *p;
4818 struct net *net = dev_net(dev);
4820 BUG_ON(dev_boot_phase);
4825 /* When net_device's are persistent, this will be fatal. */
4826 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4829 spin_lock_init(&dev->addr_list_lock);
4830 netdev_set_addr_lockdep_class(dev);
4831 netdev_init_queue_locks(dev);
4835 /* Init, if this function is available */
4836 if (dev->netdev_ops->ndo_init) {
4837 ret = dev->netdev_ops->ndo_init(dev);
4845 if (!dev_valid_name(dev->name)) {
4850 dev->ifindex = dev_new_index(net);
4851 if (dev->iflink == -1)
4852 dev->iflink = dev->ifindex;
4854 /* Check for existence of name */
4855 head = dev_name_hash(net, dev->name);
4856 hlist_for_each(p, head) {
4857 struct net_device *d
4858 = hlist_entry(p, struct net_device, name_hlist);
4859 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4865 /* Fix illegal checksum combinations */
4866 if ((dev->features & NETIF_F_HW_CSUM) &&
4867 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4868 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4870 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4873 if ((dev->features & NETIF_F_NO_CSUM) &&
4874 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4875 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4877 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4880 dev->features = netdev_fix_features(dev->features, dev->name);
4882 /* Enable software GSO if SG is supported. */
4883 if (dev->features & NETIF_F_SG)
4884 dev->features |= NETIF_F_GSO;
4886 netdev_initialize_kobject(dev);
4888 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4889 ret = notifier_to_errno(ret);
4893 ret = netdev_register_kobject(dev);
4896 dev->reg_state = NETREG_REGISTERED;
4899 * Default initial state at registry is that the
4900 * device is present.
4903 set_bit(__LINK_STATE_PRESENT, &dev->state);
4905 dev_init_scheduler(dev);
4907 list_netdevice(dev);
4909 /* Notify protocols, that a new device appeared. */
4910 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4911 ret = notifier_to_errno(ret);
4913 rollback_registered(dev);
4914 dev->reg_state = NETREG_UNREGISTERED;
4921 if (dev->netdev_ops->ndo_uninit)
4922 dev->netdev_ops->ndo_uninit(dev);
4925 EXPORT_SYMBOL(register_netdevice);
4928 * init_dummy_netdev - init a dummy network device for NAPI
4929 * @dev: device to init
4931 * This takes a network device structure and initialize the minimum
4932 * amount of fields so it can be used to schedule NAPI polls without
4933 * registering a full blown interface. This is to be used by drivers
4934 * that need to tie several hardware interfaces to a single NAPI
4935 * poll scheduler due to HW limitations.
4937 int init_dummy_netdev(struct net_device *dev)
4939 /* Clear everything. Note we don't initialize spinlocks
4940 * are they aren't supposed to be taken by any of the
4941 * NAPI code and this dummy netdev is supposed to be
4942 * only ever used for NAPI polls
4944 memset(dev, 0, sizeof(struct net_device));
4946 /* make sure we BUG if trying to hit standard
4947 * register/unregister code path
4949 dev->reg_state = NETREG_DUMMY;
4951 /* initialize the ref count */
4952 atomic_set(&dev->refcnt, 1);
4954 /* NAPI wants this */
4955 INIT_LIST_HEAD(&dev->napi_list);
4957 /* a dummy interface is started by default */
4958 set_bit(__LINK_STATE_PRESENT, &dev->state);
4959 set_bit(__LINK_STATE_START, &dev->state);
4963 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4967 * register_netdev - register a network device
4968 * @dev: device to register
4970 * Take a completed network device structure and add it to the kernel
4971 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4972 * chain. 0 is returned on success. A negative errno code is returned
4973 * on a failure to set up the device, or if the name is a duplicate.
4975 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4976 * and expands the device name if you passed a format string to
4979 int register_netdev(struct net_device *dev)
4986 * If the name is a format string the caller wants us to do a
4989 if (strchr(dev->name, '%')) {
4990 err = dev_alloc_name(dev, dev->name);
4995 err = register_netdevice(dev);
5000 EXPORT_SYMBOL(register_netdev);
5003 * netdev_wait_allrefs - wait until all references are gone.
5005 * This is called when unregistering network devices.
5007 * Any protocol or device that holds a reference should register
5008 * for netdevice notification, and cleanup and put back the
5009 * reference if they receive an UNREGISTER event.
5010 * We can get stuck here if buggy protocols don't correctly
5013 static void netdev_wait_allrefs(struct net_device *dev)
5015 unsigned long rebroadcast_time, warning_time;
5017 rebroadcast_time = warning_time = jiffies;
5018 while (atomic_read(&dev->refcnt) != 0) {
5019 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5022 /* Rebroadcast unregister notification */
5023 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5025 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5027 /* We must not have linkwatch events
5028 * pending on unregister. If this
5029 * happens, we simply run the queue
5030 * unscheduled, resulting in a noop
5033 linkwatch_run_queue();
5038 rebroadcast_time = jiffies;
5043 if (time_after(jiffies, warning_time + 10 * HZ)) {
5044 printk(KERN_EMERG "unregister_netdevice: "
5045 "waiting for %s to become free. Usage "
5047 dev->name, atomic_read(&dev->refcnt));
5048 warning_time = jiffies;
5057 * register_netdevice(x1);
5058 * register_netdevice(x2);
5060 * unregister_netdevice(y1);
5061 * unregister_netdevice(y2);
5067 * We are invoked by rtnl_unlock().
5068 * This allows us to deal with problems:
5069 * 1) We can delete sysfs objects which invoke hotplug
5070 * without deadlocking with linkwatch via keventd.
5071 * 2) Since we run with the RTNL semaphore not held, we can sleep
5072 * safely in order to wait for the netdev refcnt to drop to zero.
5074 * We must not return until all unregister events added during
5075 * the interval the lock was held have been completed.
5077 void netdev_run_todo(void)
5079 struct list_head list;
5081 /* Snapshot list, allow later requests */
5082 list_replace_init(&net_todo_list, &list);
5086 while (!list_empty(&list)) {
5087 struct net_device *dev
5088 = list_entry(list.next, struct net_device, todo_list);
5089 list_del(&dev->todo_list);
5091 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5092 printk(KERN_ERR "network todo '%s' but state %d\n",
5093 dev->name, dev->reg_state);
5098 dev->reg_state = NETREG_UNREGISTERED;
5100 on_each_cpu(flush_backlog, dev, 1);
5102 netdev_wait_allrefs(dev);
5105 BUG_ON(atomic_read(&dev->refcnt));
5106 WARN_ON(dev->ip_ptr);
5107 WARN_ON(dev->ip6_ptr);
5108 WARN_ON(dev->dn_ptr);
5110 if (dev->destructor)
5111 dev->destructor(dev);
5113 /* Free network device */
5114 kobject_put(&dev->dev.kobj);
5119 * dev_get_stats - get network device statistics
5120 * @dev: device to get statistics from
5122 * Get network statistics from device. The device driver may provide
5123 * its own method by setting dev->netdev_ops->get_stats; otherwise
5124 * the internal statistics structure is used.
5126 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5128 const struct net_device_ops *ops = dev->netdev_ops;
5130 if (ops->ndo_get_stats)
5131 return ops->ndo_get_stats(dev);
5133 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5134 struct net_device_stats *stats = &dev->stats;
5136 struct netdev_queue *txq;
5138 for (i = 0; i < dev->num_tx_queues; i++) {
5139 txq = netdev_get_tx_queue(dev, i);
5140 tx_bytes += txq->tx_bytes;
5141 tx_packets += txq->tx_packets;
5142 tx_dropped += txq->tx_dropped;
5144 if (tx_bytes || tx_packets || tx_dropped) {
5145 stats->tx_bytes = tx_bytes;
5146 stats->tx_packets = tx_packets;
5147 stats->tx_dropped = tx_dropped;
5152 EXPORT_SYMBOL(dev_get_stats);
5154 static void netdev_init_one_queue(struct net_device *dev,
5155 struct netdev_queue *queue,
5161 static void netdev_init_queues(struct net_device *dev)
5163 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5164 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5165 spin_lock_init(&dev->tx_global_lock);
5169 * alloc_netdev_mq - allocate network device
5170 * @sizeof_priv: size of private data to allocate space for
5171 * @name: device name format string
5172 * @setup: callback to initialize device
5173 * @queue_count: the number of subqueues to allocate
5175 * Allocates a struct net_device with private data area for driver use
5176 * and performs basic initialization. Also allocates subquue structs
5177 * for each queue on the device at the end of the netdevice.
5179 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5180 void (*setup)(struct net_device *), unsigned int queue_count)
5182 struct netdev_queue *tx;
5183 struct net_device *dev;
5185 struct net_device *p;
5187 BUG_ON(strlen(name) >= sizeof(dev->name));
5189 alloc_size = sizeof(struct net_device);
5191 /* ensure 32-byte alignment of private area */
5192 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5193 alloc_size += sizeof_priv;
5195 /* ensure 32-byte alignment of whole construct */
5196 alloc_size += NETDEV_ALIGN - 1;
5198 p = kzalloc(alloc_size, GFP_KERNEL);
5200 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5204 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5206 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5211 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5212 dev->padded = (char *)dev - (char *)p;
5214 if (dev_addr_init(dev))
5217 dev_unicast_init(dev);
5219 dev_net_set(dev, &init_net);
5222 dev->num_tx_queues = queue_count;
5223 dev->real_num_tx_queues = queue_count;
5225 dev->gso_max_size = GSO_MAX_SIZE;
5227 netdev_init_queues(dev);
5229 INIT_LIST_HEAD(&dev->napi_list);
5230 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5232 strcpy(dev->name, name);
5242 EXPORT_SYMBOL(alloc_netdev_mq);
5245 * free_netdev - free network device
5248 * This function does the last stage of destroying an allocated device
5249 * interface. The reference to the device object is released.
5250 * If this is the last reference then it will be freed.
5252 void free_netdev(struct net_device *dev)
5254 struct napi_struct *p, *n;
5256 release_net(dev_net(dev));
5260 /* Flush device addresses */
5261 dev_addr_flush(dev);
5263 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5266 /* Compatibility with error handling in drivers */
5267 if (dev->reg_state == NETREG_UNINITIALIZED) {
5268 kfree((char *)dev - dev->padded);
5272 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5273 dev->reg_state = NETREG_RELEASED;
5275 /* will free via device release */
5276 put_device(&dev->dev);
5278 EXPORT_SYMBOL(free_netdev);
5281 * synchronize_net - Synchronize with packet receive processing
5283 * Wait for packets currently being received to be done.
5284 * Does not block later packets from starting.
5286 void synchronize_net(void)
5291 EXPORT_SYMBOL(synchronize_net);
5294 * unregister_netdevice_queue - remove device from the kernel
5298 * This function shuts down a device interface and removes it
5299 * from the kernel tables.
5300 * If head not NULL, device is queued to be unregistered later.
5302 * Callers must hold the rtnl semaphore. You may want
5303 * unregister_netdev() instead of this.
5306 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5311 list_add_tail(&dev->unreg_list, head);
5313 rollback_registered(dev);
5314 /* Finish processing unregister after unlock */
5318 EXPORT_SYMBOL(unregister_netdevice_queue);
5321 * unregister_netdevice_many - unregister many devices
5322 * @head: list of devices
5325 void unregister_netdevice_many(struct list_head *head)
5327 struct net_device *dev;
5329 if (!list_empty(head)) {
5330 rollback_registered_many(head);
5331 list_for_each_entry(dev, head, unreg_list)
5335 EXPORT_SYMBOL(unregister_netdevice_many);
5338 * unregister_netdev - remove device from the kernel
5341 * This function shuts down a device interface and removes it
5342 * from the kernel tables.
5344 * This is just a wrapper for unregister_netdevice that takes
5345 * the rtnl semaphore. In general you want to use this and not
5346 * unregister_netdevice.
5348 void unregister_netdev(struct net_device *dev)
5351 unregister_netdevice(dev);
5354 EXPORT_SYMBOL(unregister_netdev);
5357 * dev_change_net_namespace - move device to different nethost namespace
5359 * @net: network namespace
5360 * @pat: If not NULL name pattern to try if the current device name
5361 * is already taken in the destination network namespace.
5363 * This function shuts down a device interface and moves it
5364 * to a new network namespace. On success 0 is returned, on
5365 * a failure a netagive errno code is returned.
5367 * Callers must hold the rtnl semaphore.
5370 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5373 const char *destname;
5378 /* Don't allow namespace local devices to be moved. */
5380 if (dev->features & NETIF_F_NETNS_LOCAL)
5384 /* Don't allow real devices to be moved when sysfs
5388 if (dev->dev.parent)
5392 /* Ensure the device has been registrered */
5394 if (dev->reg_state != NETREG_REGISTERED)
5397 /* Get out if there is nothing todo */
5399 if (net_eq(dev_net(dev), net))
5402 /* Pick the destination device name, and ensure
5403 * we can use it in the destination network namespace.
5406 destname = dev->name;
5407 if (__dev_get_by_name(net, destname)) {
5408 /* We get here if we can't use the current device name */
5411 if (!dev_valid_name(pat))
5413 if (strchr(pat, '%')) {
5414 if (__dev_alloc_name(net, pat, buf) < 0)
5419 if (__dev_get_by_name(net, destname))
5424 * And now a mini version of register_netdevice unregister_netdevice.
5427 /* If device is running close it first. */
5430 /* And unlink it from device chain */
5432 unlist_netdevice(dev);
5436 /* Shutdown queueing discipline. */
5439 /* Notify protocols, that we are about to destroy
5440 this device. They should clean all the things.
5442 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5445 * Flush the unicast and multicast chains
5447 dev_unicast_flush(dev);
5448 dev_addr_discard(dev);
5450 netdev_unregister_kobject(dev);
5452 /* Actually switch the network namespace */
5453 dev_net_set(dev, net);
5455 /* Assign the new device name */
5456 if (destname != dev->name)
5457 strcpy(dev->name, destname);
5459 /* If there is an ifindex conflict assign a new one */
5460 if (__dev_get_by_index(net, dev->ifindex)) {
5461 int iflink = (dev->iflink == dev->ifindex);
5462 dev->ifindex = dev_new_index(net);
5464 dev->iflink = dev->ifindex;
5467 /* Fixup kobjects */
5468 err = netdev_register_kobject(dev);
5471 /* Add the device back in the hashes */
5472 list_netdevice(dev);
5474 /* Notify protocols, that a new device appeared. */
5475 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5482 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5484 static int dev_cpu_callback(struct notifier_block *nfb,
5485 unsigned long action,
5488 struct sk_buff **list_skb;
5489 struct Qdisc **list_net;
5490 struct sk_buff *skb;
5491 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5492 struct softnet_data *sd, *oldsd;
5494 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5497 local_irq_disable();
5498 cpu = smp_processor_id();
5499 sd = &per_cpu(softnet_data, cpu);
5500 oldsd = &per_cpu(softnet_data, oldcpu);
5502 /* Find end of our completion_queue. */
5503 list_skb = &sd->completion_queue;
5505 list_skb = &(*list_skb)->next;
5506 /* Append completion queue from offline CPU. */
5507 *list_skb = oldsd->completion_queue;
5508 oldsd->completion_queue = NULL;
5510 /* Find end of our output_queue. */
5511 list_net = &sd->output_queue;
5513 list_net = &(*list_net)->next_sched;
5514 /* Append output queue from offline CPU. */
5515 *list_net = oldsd->output_queue;
5516 oldsd->output_queue = NULL;
5518 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5521 /* Process offline CPU's input_pkt_queue */
5522 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5530 * netdev_increment_features - increment feature set by one
5531 * @all: current feature set
5532 * @one: new feature set
5533 * @mask: mask feature set
5535 * Computes a new feature set after adding a device with feature set
5536 * @one to the master device with current feature set @all. Will not
5537 * enable anything that is off in @mask. Returns the new feature set.
5539 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5542 /* If device needs checksumming, downgrade to it. */
5543 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5544 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5545 else if (mask & NETIF_F_ALL_CSUM) {
5546 /* If one device supports v4/v6 checksumming, set for all. */
5547 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5548 !(all & NETIF_F_GEN_CSUM)) {
5549 all &= ~NETIF_F_ALL_CSUM;
5550 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5553 /* If one device supports hw checksumming, set for all. */
5554 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5555 all &= ~NETIF_F_ALL_CSUM;
5556 all |= NETIF_F_HW_CSUM;
5560 one |= NETIF_F_ALL_CSUM;
5562 one |= all & NETIF_F_ONE_FOR_ALL;
5563 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5564 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5568 EXPORT_SYMBOL(netdev_increment_features);
5570 static struct hlist_head *netdev_create_hash(void)
5573 struct hlist_head *hash;
5575 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5577 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5578 INIT_HLIST_HEAD(&hash[i]);
5583 /* Initialize per network namespace state */
5584 static int __net_init netdev_init(struct net *net)
5586 INIT_LIST_HEAD(&net->dev_base_head);
5588 net->dev_name_head = netdev_create_hash();
5589 if (net->dev_name_head == NULL)
5592 net->dev_index_head = netdev_create_hash();
5593 if (net->dev_index_head == NULL)
5599 kfree(net->dev_name_head);
5605 * netdev_drivername - network driver for the device
5606 * @dev: network device
5607 * @buffer: buffer for resulting name
5608 * @len: size of buffer
5610 * Determine network driver for device.
5612 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5614 const struct device_driver *driver;
5615 const struct device *parent;
5617 if (len <= 0 || !buffer)
5621 parent = dev->dev.parent;
5626 driver = parent->driver;
5627 if (driver && driver->name)
5628 strlcpy(buffer, driver->name, len);
5632 static void __net_exit netdev_exit(struct net *net)
5634 kfree(net->dev_name_head);
5635 kfree(net->dev_index_head);
5638 static struct pernet_operations __net_initdata netdev_net_ops = {
5639 .init = netdev_init,
5640 .exit = netdev_exit,
5643 static void __net_exit default_device_exit(struct net *net)
5645 struct net_device *dev;
5647 * Push all migratable of the network devices back to the
5648 * initial network namespace
5652 for_each_netdev(net, dev) {
5654 char fb_name[IFNAMSIZ];
5656 /* Ignore unmoveable devices (i.e. loopback) */
5657 if (dev->features & NETIF_F_NETNS_LOCAL)
5660 /* Delete virtual devices */
5661 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5662 dev->rtnl_link_ops->dellink(dev, NULL);
5666 /* Push remaing network devices to init_net */
5667 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5668 err = dev_change_net_namespace(dev, &init_net, fb_name);
5670 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5671 __func__, dev->name, err);
5679 static struct pernet_operations __net_initdata default_device_ops = {
5680 .exit = default_device_exit,
5684 * Initialize the DEV module. At boot time this walks the device list and
5685 * unhooks any devices that fail to initialise (normally hardware not
5686 * present) and leaves us with a valid list of present and active devices.
5691 * This is called single threaded during boot, so no need
5692 * to take the rtnl semaphore.
5694 static int __init net_dev_init(void)
5696 int i, rc = -ENOMEM;
5698 BUG_ON(!dev_boot_phase);
5700 if (dev_proc_init())
5703 if (netdev_kobject_init())
5706 INIT_LIST_HEAD(&ptype_all);
5707 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5708 INIT_LIST_HEAD(&ptype_base[i]);
5710 if (register_pernet_subsys(&netdev_net_ops))
5714 * Initialise the packet receive queues.
5717 for_each_possible_cpu(i) {
5718 struct softnet_data *queue;
5720 queue = &per_cpu(softnet_data, i);
5721 skb_queue_head_init(&queue->input_pkt_queue);
5722 queue->completion_queue = NULL;
5723 INIT_LIST_HEAD(&queue->poll_list);
5725 queue->backlog.poll = process_backlog;
5726 queue->backlog.weight = weight_p;
5727 queue->backlog.gro_list = NULL;
5728 queue->backlog.gro_count = 0;
5733 /* The loopback device is special if any other network devices
5734 * is present in a network namespace the loopback device must
5735 * be present. Since we now dynamically allocate and free the
5736 * loopback device ensure this invariant is maintained by
5737 * keeping the loopback device as the first device on the
5738 * list of network devices. Ensuring the loopback devices
5739 * is the first device that appears and the last network device
5742 if (register_pernet_device(&loopback_net_ops))
5745 if (register_pernet_device(&default_device_ops))
5748 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5749 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5751 hotcpu_notifier(dev_cpu_callback, 0);
5759 subsys_initcall(net_dev_init);
5761 static int __init initialize_hashrnd(void)
5763 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5767 late_initcall_sync(initialize_hashrnd);