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 int 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];
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);
2571 static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2575 if (netpoll_rx_on(skb))
2578 for (p = napi->gro_list; p; p = p->next) {
2579 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2580 && !compare_ether_header(skb_mac_header(p),
2581 skb_gro_mac_header(skb));
2582 NAPI_GRO_CB(p)->flush = 0;
2585 return dev_gro_receive(napi, skb);
2588 int napi_skb_finish(int ret, struct sk_buff *skb)
2590 int err = NET_RX_SUCCESS;
2594 return netif_receive_skb(skb);
2600 case GRO_MERGED_FREE:
2607 EXPORT_SYMBOL(napi_skb_finish);
2609 void skb_gro_reset_offset(struct sk_buff *skb)
2611 NAPI_GRO_CB(skb)->data_offset = 0;
2612 NAPI_GRO_CB(skb)->frag0 = NULL;
2613 NAPI_GRO_CB(skb)->frag0_len = 0;
2615 if (skb->mac_header == skb->tail &&
2616 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2617 NAPI_GRO_CB(skb)->frag0 =
2618 page_address(skb_shinfo(skb)->frags[0].page) +
2619 skb_shinfo(skb)->frags[0].page_offset;
2620 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2623 EXPORT_SYMBOL(skb_gro_reset_offset);
2625 int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2627 skb_gro_reset_offset(skb);
2629 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2631 EXPORT_SYMBOL(napi_gro_receive);
2633 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2635 __skb_pull(skb, skb_headlen(skb));
2636 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2640 EXPORT_SYMBOL(napi_reuse_skb);
2642 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2644 struct sk_buff *skb = napi->skb;
2647 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2653 EXPORT_SYMBOL(napi_get_frags);
2655 int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret)
2657 int err = NET_RX_SUCCESS;
2662 skb->protocol = eth_type_trans(skb, napi->dev);
2664 if (ret == GRO_NORMAL)
2665 return netif_receive_skb(skb);
2667 skb_gro_pull(skb, -ETH_HLEN);
2674 case GRO_MERGED_FREE:
2675 napi_reuse_skb(napi, skb);
2681 EXPORT_SYMBOL(napi_frags_finish);
2683 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2685 struct sk_buff *skb = napi->skb;
2692 skb_reset_mac_header(skb);
2693 skb_gro_reset_offset(skb);
2695 off = skb_gro_offset(skb);
2696 hlen = off + sizeof(*eth);
2697 eth = skb_gro_header_fast(skb, off);
2698 if (skb_gro_header_hard(skb, hlen)) {
2699 eth = skb_gro_header_slow(skb, hlen, off);
2700 if (unlikely(!eth)) {
2701 napi_reuse_skb(napi, skb);
2707 skb_gro_pull(skb, sizeof(*eth));
2710 * This works because the only protocols we care about don't require
2711 * special handling. We'll fix it up properly at the end.
2713 skb->protocol = eth->h_proto;
2718 EXPORT_SYMBOL(napi_frags_skb);
2720 int napi_gro_frags(struct napi_struct *napi)
2722 struct sk_buff *skb = napi_frags_skb(napi);
2727 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2729 EXPORT_SYMBOL(napi_gro_frags);
2731 static int process_backlog(struct napi_struct *napi, int quota)
2734 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2735 unsigned long start_time = jiffies;
2737 napi->weight = weight_p;
2739 struct sk_buff *skb;
2741 local_irq_disable();
2742 skb = __skb_dequeue(&queue->input_pkt_queue);
2744 __napi_complete(napi);
2750 netif_receive_skb(skb);
2751 } while (++work < quota && jiffies == start_time);
2757 * __napi_schedule - schedule for receive
2758 * @n: entry to schedule
2760 * The entry's receive function will be scheduled to run
2762 void __napi_schedule(struct napi_struct *n)
2764 unsigned long flags;
2766 local_irq_save(flags);
2767 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2768 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2769 local_irq_restore(flags);
2771 EXPORT_SYMBOL(__napi_schedule);
2773 void __napi_complete(struct napi_struct *n)
2775 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2776 BUG_ON(n->gro_list);
2778 list_del(&n->poll_list);
2779 smp_mb__before_clear_bit();
2780 clear_bit(NAPI_STATE_SCHED, &n->state);
2782 EXPORT_SYMBOL(__napi_complete);
2784 void napi_complete(struct napi_struct *n)
2786 unsigned long flags;
2789 * don't let napi dequeue from the cpu poll list
2790 * just in case its running on a different cpu
2792 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2796 local_irq_save(flags);
2798 local_irq_restore(flags);
2800 EXPORT_SYMBOL(napi_complete);
2802 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2803 int (*poll)(struct napi_struct *, int), int weight)
2805 INIT_LIST_HEAD(&napi->poll_list);
2806 napi->gro_count = 0;
2807 napi->gro_list = NULL;
2810 napi->weight = weight;
2811 list_add(&napi->dev_list, &dev->napi_list);
2813 #ifdef CONFIG_NETPOLL
2814 spin_lock_init(&napi->poll_lock);
2815 napi->poll_owner = -1;
2817 set_bit(NAPI_STATE_SCHED, &napi->state);
2819 EXPORT_SYMBOL(netif_napi_add);
2821 void netif_napi_del(struct napi_struct *napi)
2823 struct sk_buff *skb, *next;
2825 list_del_init(&napi->dev_list);
2826 napi_free_frags(napi);
2828 for (skb = napi->gro_list; skb; skb = next) {
2834 napi->gro_list = NULL;
2835 napi->gro_count = 0;
2837 EXPORT_SYMBOL(netif_napi_del);
2840 static void net_rx_action(struct softirq_action *h)
2842 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2843 unsigned long time_limit = jiffies + 2;
2844 int budget = netdev_budget;
2847 local_irq_disable();
2849 while (!list_empty(list)) {
2850 struct napi_struct *n;
2853 /* If softirq window is exhuasted then punt.
2854 * Allow this to run for 2 jiffies since which will allow
2855 * an average latency of 1.5/HZ.
2857 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2862 /* Even though interrupts have been re-enabled, this
2863 * access is safe because interrupts can only add new
2864 * entries to the tail of this list, and only ->poll()
2865 * calls can remove this head entry from the list.
2867 n = list_entry(list->next, struct napi_struct, poll_list);
2869 have = netpoll_poll_lock(n);
2873 /* This NAPI_STATE_SCHED test is for avoiding a race
2874 * with netpoll's poll_napi(). Only the entity which
2875 * obtains the lock and sees NAPI_STATE_SCHED set will
2876 * actually make the ->poll() call. Therefore we avoid
2877 * accidently calling ->poll() when NAPI is not scheduled.
2880 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2881 work = n->poll(n, weight);
2885 WARN_ON_ONCE(work > weight);
2889 local_irq_disable();
2891 /* Drivers must not modify the NAPI state if they
2892 * consume the entire weight. In such cases this code
2893 * still "owns" the NAPI instance and therefore can
2894 * move the instance around on the list at-will.
2896 if (unlikely(work == weight)) {
2897 if (unlikely(napi_disable_pending(n))) {
2900 local_irq_disable();
2902 list_move_tail(&n->poll_list, list);
2905 netpoll_poll_unlock(have);
2910 #ifdef CONFIG_NET_DMA
2912 * There may not be any more sk_buffs coming right now, so push
2913 * any pending DMA copies to hardware
2915 dma_issue_pending_all();
2921 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2922 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2926 static gifconf_func_t *gifconf_list[NPROTO];
2929 * register_gifconf - register a SIOCGIF handler
2930 * @family: Address family
2931 * @gifconf: Function handler
2933 * Register protocol dependent address dumping routines. The handler
2934 * that is passed must not be freed or reused until it has been replaced
2935 * by another handler.
2937 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
2939 if (family >= NPROTO)
2941 gifconf_list[family] = gifconf;
2944 EXPORT_SYMBOL(register_gifconf);
2948 * Map an interface index to its name (SIOCGIFNAME)
2952 * We need this ioctl for efficient implementation of the
2953 * if_indextoname() function required by the IPv6 API. Without
2954 * it, we would have to search all the interfaces to find a
2958 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2960 struct net_device *dev;
2964 * Fetch the caller's info block.
2967 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2971 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
2977 strcpy(ifr.ifr_name, dev->name);
2980 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2986 * Perform a SIOCGIFCONF call. This structure will change
2987 * size eventually, and there is nothing I can do about it.
2988 * Thus we will need a 'compatibility mode'.
2991 static int dev_ifconf(struct net *net, char __user *arg)
2994 struct net_device *dev;
3001 * Fetch the caller's info block.
3004 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3011 * Loop over the interfaces, and write an info block for each.
3015 for_each_netdev(net, dev) {
3016 for (i = 0; i < NPROTO; i++) {
3017 if (gifconf_list[i]) {
3020 done = gifconf_list[i](dev, NULL, 0);
3022 done = gifconf_list[i](dev, pos + total,
3032 * All done. Write the updated control block back to the caller.
3034 ifc.ifc_len = total;
3037 * Both BSD and Solaris return 0 here, so we do too.
3039 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3042 #ifdef CONFIG_PROC_FS
3044 * This is invoked by the /proc filesystem handler to display a device
3047 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3048 __acquires(dev_base_lock)
3050 struct net *net = seq_file_net(seq);
3052 struct net_device *dev;
3054 read_lock(&dev_base_lock);
3056 return SEQ_START_TOKEN;
3059 for_each_netdev(net, dev)
3066 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3068 struct net *net = seq_file_net(seq);
3070 return v == SEQ_START_TOKEN ?
3071 first_net_device(net) : next_net_device((struct net_device *)v);
3074 void dev_seq_stop(struct seq_file *seq, void *v)
3075 __releases(dev_base_lock)
3077 read_unlock(&dev_base_lock);
3080 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3082 const struct net_device_stats *stats = dev_get_stats(dev);
3084 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3085 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3086 dev->name, stats->rx_bytes, stats->rx_packets,
3088 stats->rx_dropped + stats->rx_missed_errors,
3089 stats->rx_fifo_errors,
3090 stats->rx_length_errors + stats->rx_over_errors +
3091 stats->rx_crc_errors + stats->rx_frame_errors,
3092 stats->rx_compressed, stats->multicast,
3093 stats->tx_bytes, stats->tx_packets,
3094 stats->tx_errors, stats->tx_dropped,
3095 stats->tx_fifo_errors, stats->collisions,
3096 stats->tx_carrier_errors +
3097 stats->tx_aborted_errors +
3098 stats->tx_window_errors +
3099 stats->tx_heartbeat_errors,
3100 stats->tx_compressed);
3104 * Called from the PROCfs module. This now uses the new arbitrary sized
3105 * /proc/net interface to create /proc/net/dev
3107 static int dev_seq_show(struct seq_file *seq, void *v)
3109 if (v == SEQ_START_TOKEN)
3110 seq_puts(seq, "Inter-| Receive "
3112 " face |bytes packets errs drop fifo frame "
3113 "compressed multicast|bytes packets errs "
3114 "drop fifo colls carrier compressed\n");
3116 dev_seq_printf_stats(seq, v);
3120 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3122 struct netif_rx_stats *rc = NULL;
3124 while (*pos < nr_cpu_ids)
3125 if (cpu_online(*pos)) {
3126 rc = &per_cpu(netdev_rx_stat, *pos);
3133 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3135 return softnet_get_online(pos);
3138 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3141 return softnet_get_online(pos);
3144 static void softnet_seq_stop(struct seq_file *seq, void *v)
3148 static int softnet_seq_show(struct seq_file *seq, void *v)
3150 struct netif_rx_stats *s = v;
3152 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3153 s->total, s->dropped, s->time_squeeze, 0,
3154 0, 0, 0, 0, /* was fastroute */
3159 static const struct seq_operations dev_seq_ops = {
3160 .start = dev_seq_start,
3161 .next = dev_seq_next,
3162 .stop = dev_seq_stop,
3163 .show = dev_seq_show,
3166 static int dev_seq_open(struct inode *inode, struct file *file)
3168 return seq_open_net(inode, file, &dev_seq_ops,
3169 sizeof(struct seq_net_private));
3172 static const struct file_operations dev_seq_fops = {
3173 .owner = THIS_MODULE,
3174 .open = dev_seq_open,
3176 .llseek = seq_lseek,
3177 .release = seq_release_net,
3180 static const struct seq_operations softnet_seq_ops = {
3181 .start = softnet_seq_start,
3182 .next = softnet_seq_next,
3183 .stop = softnet_seq_stop,
3184 .show = softnet_seq_show,
3187 static int softnet_seq_open(struct inode *inode, struct file *file)
3189 return seq_open(file, &softnet_seq_ops);
3192 static const struct file_operations softnet_seq_fops = {
3193 .owner = THIS_MODULE,
3194 .open = softnet_seq_open,
3196 .llseek = seq_lseek,
3197 .release = seq_release,
3200 static void *ptype_get_idx(loff_t pos)
3202 struct packet_type *pt = NULL;
3206 list_for_each_entry_rcu(pt, &ptype_all, list) {
3212 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3213 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3222 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3226 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3229 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3231 struct packet_type *pt;
3232 struct list_head *nxt;
3236 if (v == SEQ_START_TOKEN)
3237 return ptype_get_idx(0);
3240 nxt = pt->list.next;
3241 if (pt->type == htons(ETH_P_ALL)) {
3242 if (nxt != &ptype_all)
3245 nxt = ptype_base[0].next;
3247 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3249 while (nxt == &ptype_base[hash]) {
3250 if (++hash >= PTYPE_HASH_SIZE)
3252 nxt = ptype_base[hash].next;
3255 return list_entry(nxt, struct packet_type, list);
3258 static void ptype_seq_stop(struct seq_file *seq, void *v)
3264 static int ptype_seq_show(struct seq_file *seq, void *v)
3266 struct packet_type *pt = v;
3268 if (v == SEQ_START_TOKEN)
3269 seq_puts(seq, "Type Device Function\n");
3270 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3271 if (pt->type == htons(ETH_P_ALL))
3272 seq_puts(seq, "ALL ");
3274 seq_printf(seq, "%04x", ntohs(pt->type));
3276 seq_printf(seq, " %-8s %pF\n",
3277 pt->dev ? pt->dev->name : "", pt->func);
3283 static const struct seq_operations ptype_seq_ops = {
3284 .start = ptype_seq_start,
3285 .next = ptype_seq_next,
3286 .stop = ptype_seq_stop,
3287 .show = ptype_seq_show,
3290 static int ptype_seq_open(struct inode *inode, struct file *file)
3292 return seq_open_net(inode, file, &ptype_seq_ops,
3293 sizeof(struct seq_net_private));
3296 static const struct file_operations ptype_seq_fops = {
3297 .owner = THIS_MODULE,
3298 .open = ptype_seq_open,
3300 .llseek = seq_lseek,
3301 .release = seq_release_net,
3305 static int __net_init dev_proc_net_init(struct net *net)
3309 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3311 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3313 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3316 if (wext_proc_init(net))
3322 proc_net_remove(net, "ptype");
3324 proc_net_remove(net, "softnet_stat");
3326 proc_net_remove(net, "dev");
3330 static void __net_exit dev_proc_net_exit(struct net *net)
3332 wext_proc_exit(net);
3334 proc_net_remove(net, "ptype");
3335 proc_net_remove(net, "softnet_stat");
3336 proc_net_remove(net, "dev");
3339 static struct pernet_operations __net_initdata dev_proc_ops = {
3340 .init = dev_proc_net_init,
3341 .exit = dev_proc_net_exit,
3344 static int __init dev_proc_init(void)
3346 return register_pernet_subsys(&dev_proc_ops);
3349 #define dev_proc_init() 0
3350 #endif /* CONFIG_PROC_FS */
3354 * netdev_set_master - set up master/slave pair
3355 * @slave: slave device
3356 * @master: new master device
3358 * Changes the master device of the slave. Pass %NULL to break the
3359 * bonding. The caller must hold the RTNL semaphore. On a failure
3360 * a negative errno code is returned. On success the reference counts
3361 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3362 * function returns zero.
3364 int netdev_set_master(struct net_device *slave, struct net_device *master)
3366 struct net_device *old = slave->master;
3376 slave->master = master;
3384 slave->flags |= IFF_SLAVE;
3386 slave->flags &= ~IFF_SLAVE;
3388 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3391 EXPORT_SYMBOL(netdev_set_master);
3393 static void dev_change_rx_flags(struct net_device *dev, int flags)
3395 const struct net_device_ops *ops = dev->netdev_ops;
3397 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3398 ops->ndo_change_rx_flags(dev, flags);
3401 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3403 unsigned short old_flags = dev->flags;
3409 dev->flags |= IFF_PROMISC;
3410 dev->promiscuity += inc;
3411 if (dev->promiscuity == 0) {
3414 * If inc causes overflow, untouch promisc and return error.
3417 dev->flags &= ~IFF_PROMISC;
3419 dev->promiscuity -= inc;
3420 printk(KERN_WARNING "%s: promiscuity touches roof, "
3421 "set promiscuity failed, promiscuity feature "
3422 "of device might be broken.\n", dev->name);
3426 if (dev->flags != old_flags) {
3427 printk(KERN_INFO "device %s %s promiscuous mode\n",
3428 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3430 if (audit_enabled) {
3431 current_uid_gid(&uid, &gid);
3432 audit_log(current->audit_context, GFP_ATOMIC,
3433 AUDIT_ANOM_PROMISCUOUS,
3434 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3435 dev->name, (dev->flags & IFF_PROMISC),
3436 (old_flags & IFF_PROMISC),
3437 audit_get_loginuid(current),
3439 audit_get_sessionid(current));
3442 dev_change_rx_flags(dev, IFF_PROMISC);
3448 * dev_set_promiscuity - update promiscuity count on a device
3452 * Add or remove promiscuity from a device. While the count in the device
3453 * remains above zero the interface remains promiscuous. Once it hits zero
3454 * the device reverts back to normal filtering operation. A negative inc
3455 * value is used to drop promiscuity on the device.
3456 * Return 0 if successful or a negative errno code on error.
3458 int dev_set_promiscuity(struct net_device *dev, int inc)
3460 unsigned short old_flags = dev->flags;
3463 err = __dev_set_promiscuity(dev, inc);
3466 if (dev->flags != old_flags)
3467 dev_set_rx_mode(dev);
3470 EXPORT_SYMBOL(dev_set_promiscuity);
3473 * dev_set_allmulti - update allmulti count on a device
3477 * Add or remove reception of all multicast frames to a device. While the
3478 * count in the device remains above zero the interface remains listening
3479 * to all interfaces. Once it hits zero the device reverts back to normal
3480 * filtering operation. A negative @inc value is used to drop the counter
3481 * when releasing a resource needing all multicasts.
3482 * Return 0 if successful or a negative errno code on error.
3485 int dev_set_allmulti(struct net_device *dev, int inc)
3487 unsigned short old_flags = dev->flags;
3491 dev->flags |= IFF_ALLMULTI;
3492 dev->allmulti += inc;
3493 if (dev->allmulti == 0) {
3496 * If inc causes overflow, untouch allmulti and return error.
3499 dev->flags &= ~IFF_ALLMULTI;
3501 dev->allmulti -= inc;
3502 printk(KERN_WARNING "%s: allmulti touches roof, "
3503 "set allmulti failed, allmulti feature of "
3504 "device might be broken.\n", dev->name);
3508 if (dev->flags ^ old_flags) {
3509 dev_change_rx_flags(dev, IFF_ALLMULTI);
3510 dev_set_rx_mode(dev);
3514 EXPORT_SYMBOL(dev_set_allmulti);
3517 * Upload unicast and multicast address lists to device and
3518 * configure RX filtering. When the device doesn't support unicast
3519 * filtering it is put in promiscuous mode while unicast addresses
3522 void __dev_set_rx_mode(struct net_device *dev)
3524 const struct net_device_ops *ops = dev->netdev_ops;
3526 /* dev_open will call this function so the list will stay sane. */
3527 if (!(dev->flags&IFF_UP))
3530 if (!netif_device_present(dev))
3533 if (ops->ndo_set_rx_mode)
3534 ops->ndo_set_rx_mode(dev);
3536 /* Unicast addresses changes may only happen under the rtnl,
3537 * therefore calling __dev_set_promiscuity here is safe.
3539 if (dev->uc.count > 0 && !dev->uc_promisc) {
3540 __dev_set_promiscuity(dev, 1);
3541 dev->uc_promisc = 1;
3542 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3543 __dev_set_promiscuity(dev, -1);
3544 dev->uc_promisc = 0;
3547 if (ops->ndo_set_multicast_list)
3548 ops->ndo_set_multicast_list(dev);
3552 void dev_set_rx_mode(struct net_device *dev)
3554 netif_addr_lock_bh(dev);
3555 __dev_set_rx_mode(dev);
3556 netif_addr_unlock_bh(dev);
3559 /* hw addresses list handling functions */
3561 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3562 int addr_len, unsigned char addr_type)
3564 struct netdev_hw_addr *ha;
3567 if (addr_len > MAX_ADDR_LEN)
3570 list_for_each_entry(ha, &list->list, list) {
3571 if (!memcmp(ha->addr, addr, addr_len) &&
3572 ha->type == addr_type) {
3579 alloc_size = sizeof(*ha);
3580 if (alloc_size < L1_CACHE_BYTES)
3581 alloc_size = L1_CACHE_BYTES;
3582 ha = kmalloc(alloc_size, GFP_ATOMIC);
3585 memcpy(ha->addr, addr, addr_len);
3586 ha->type = addr_type;
3589 list_add_tail_rcu(&ha->list, &list->list);
3594 static void ha_rcu_free(struct rcu_head *head)
3596 struct netdev_hw_addr *ha;
3598 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3602 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3603 int addr_len, unsigned char addr_type)
3605 struct netdev_hw_addr *ha;
3607 list_for_each_entry(ha, &list->list, list) {
3608 if (!memcmp(ha->addr, addr, addr_len) &&
3609 (ha->type == addr_type || !addr_type)) {
3612 list_del_rcu(&ha->list);
3613 call_rcu(&ha->rcu_head, ha_rcu_free);
3621 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3622 struct netdev_hw_addr_list *from_list,
3624 unsigned char addr_type)
3627 struct netdev_hw_addr *ha, *ha2;
3630 list_for_each_entry(ha, &from_list->list, list) {
3631 type = addr_type ? addr_type : ha->type;
3632 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3639 list_for_each_entry(ha2, &from_list->list, list) {
3642 type = addr_type ? addr_type : ha2->type;
3643 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3648 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3649 struct netdev_hw_addr_list *from_list,
3651 unsigned char addr_type)
3653 struct netdev_hw_addr *ha;
3656 list_for_each_entry(ha, &from_list->list, list) {
3657 type = addr_type ? addr_type : ha->type;
3658 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3662 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3663 struct netdev_hw_addr_list *from_list,
3667 struct netdev_hw_addr *ha, *tmp;
3669 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3671 err = __hw_addr_add(to_list, ha->addr,
3672 addr_len, ha->type);
3677 } else if (ha->refcount == 1) {
3678 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3679 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3685 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3686 struct netdev_hw_addr_list *from_list,
3689 struct netdev_hw_addr *ha, *tmp;
3691 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3693 __hw_addr_del(to_list, ha->addr,
3694 addr_len, ha->type);
3696 __hw_addr_del(from_list, ha->addr,
3697 addr_len, ha->type);
3702 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3704 struct netdev_hw_addr *ha, *tmp;
3706 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3707 list_del_rcu(&ha->list);
3708 call_rcu(&ha->rcu_head, ha_rcu_free);
3713 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3715 INIT_LIST_HEAD(&list->list);
3719 /* Device addresses handling functions */
3721 static void dev_addr_flush(struct net_device *dev)
3723 /* rtnl_mutex must be held here */
3725 __hw_addr_flush(&dev->dev_addrs);
3726 dev->dev_addr = NULL;
3729 static int dev_addr_init(struct net_device *dev)
3731 unsigned char addr[MAX_ADDR_LEN];
3732 struct netdev_hw_addr *ha;
3735 /* rtnl_mutex must be held here */
3737 __hw_addr_init(&dev->dev_addrs);
3738 memset(addr, 0, sizeof(addr));
3739 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3740 NETDEV_HW_ADDR_T_LAN);
3743 * Get the first (previously created) address from the list
3744 * and set dev_addr pointer to this location.
3746 ha = list_first_entry(&dev->dev_addrs.list,
3747 struct netdev_hw_addr, list);
3748 dev->dev_addr = ha->addr;
3754 * dev_addr_add - Add a device address
3756 * @addr: address to add
3757 * @addr_type: address type
3759 * Add a device address to the device or increase the reference count if
3760 * it already exists.
3762 * The caller must hold the rtnl_mutex.
3764 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3765 unsigned char addr_type)
3771 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3773 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3776 EXPORT_SYMBOL(dev_addr_add);
3779 * dev_addr_del - Release a device address.
3781 * @addr: address to delete
3782 * @addr_type: address type
3784 * Release reference to a device address and remove it from the device
3785 * if the reference count drops to zero.
3787 * The caller must hold the rtnl_mutex.
3789 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3790 unsigned char addr_type)
3793 struct netdev_hw_addr *ha;
3798 * We can not remove the first address from the list because
3799 * dev->dev_addr points to that.
3801 ha = list_first_entry(&dev->dev_addrs.list,
3802 struct netdev_hw_addr, list);
3803 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3806 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3809 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3812 EXPORT_SYMBOL(dev_addr_del);
3815 * dev_addr_add_multiple - Add device addresses from another device
3816 * @to_dev: device to which addresses will be added
3817 * @from_dev: device from which addresses will be added
3818 * @addr_type: address type - 0 means type will be used from from_dev
3820 * Add device addresses of the one device to another.
3822 * The caller must hold the rtnl_mutex.
3824 int dev_addr_add_multiple(struct net_device *to_dev,
3825 struct net_device *from_dev,
3826 unsigned char addr_type)
3832 if (from_dev->addr_len != to_dev->addr_len)
3834 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3835 to_dev->addr_len, addr_type);
3837 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3840 EXPORT_SYMBOL(dev_addr_add_multiple);
3843 * dev_addr_del_multiple - Delete device addresses by another device
3844 * @to_dev: device where the addresses will be deleted
3845 * @from_dev: device by which addresses the addresses will be deleted
3846 * @addr_type: address type - 0 means type will used from from_dev
3848 * Deletes addresses in to device by the list of addresses in from device.
3850 * The caller must hold the rtnl_mutex.
3852 int dev_addr_del_multiple(struct net_device *to_dev,
3853 struct net_device *from_dev,
3854 unsigned char addr_type)
3858 if (from_dev->addr_len != to_dev->addr_len)
3860 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3861 to_dev->addr_len, addr_type);
3862 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3865 EXPORT_SYMBOL(dev_addr_del_multiple);
3867 /* multicast addresses handling functions */
3869 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3870 void *addr, int alen, int glbl)
3872 struct dev_addr_list *da;
3874 for (; (da = *list) != NULL; list = &da->next) {
3875 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3876 alen == da->da_addrlen) {
3878 int old_glbl = da->da_gusers;
3895 int __dev_addr_add(struct dev_addr_list **list, int *count,
3896 void *addr, int alen, int glbl)
3898 struct dev_addr_list *da;
3900 for (da = *list; da != NULL; da = da->next) {
3901 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3902 da->da_addrlen == alen) {
3904 int old_glbl = da->da_gusers;
3914 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3917 memcpy(da->da_addr, addr, alen);
3918 da->da_addrlen = alen;
3920 da->da_gusers = glbl ? 1 : 0;
3928 * dev_unicast_delete - Release secondary unicast address.
3930 * @addr: address to delete
3932 * Release reference to a secondary unicast address and remove it
3933 * from the device if the reference count drops to zero.
3935 * The caller must hold the rtnl_mutex.
3937 int dev_unicast_delete(struct net_device *dev, void *addr)
3943 netif_addr_lock_bh(dev);
3944 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
3945 NETDEV_HW_ADDR_T_UNICAST);
3947 __dev_set_rx_mode(dev);
3948 netif_addr_unlock_bh(dev);
3951 EXPORT_SYMBOL(dev_unicast_delete);
3954 * dev_unicast_add - add a secondary unicast address
3956 * @addr: address to add
3958 * Add a secondary unicast address to the device or increase
3959 * the reference count if it already exists.
3961 * The caller must hold the rtnl_mutex.
3963 int dev_unicast_add(struct net_device *dev, void *addr)
3969 netif_addr_lock_bh(dev);
3970 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
3971 NETDEV_HW_ADDR_T_UNICAST);
3973 __dev_set_rx_mode(dev);
3974 netif_addr_unlock_bh(dev);
3977 EXPORT_SYMBOL(dev_unicast_add);
3979 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3980 struct dev_addr_list **from, int *from_count)
3982 struct dev_addr_list *da, *next;
3986 while (da != NULL) {
3988 if (!da->da_synced) {
3989 err = __dev_addr_add(to, to_count,
3990 da->da_addr, da->da_addrlen, 0);
3995 } else if (da->da_users == 1) {
3996 __dev_addr_delete(to, to_count,
3997 da->da_addr, da->da_addrlen, 0);
3998 __dev_addr_delete(from, from_count,
3999 da->da_addr, da->da_addrlen, 0);
4005 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4007 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4008 struct dev_addr_list **from, int *from_count)
4010 struct dev_addr_list *da, *next;
4013 while (da != NULL) {
4015 if (da->da_synced) {
4016 __dev_addr_delete(to, to_count,
4017 da->da_addr, da->da_addrlen, 0);
4019 __dev_addr_delete(from, from_count,
4020 da->da_addr, da->da_addrlen, 0);
4025 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4028 * dev_unicast_sync - Synchronize device's unicast list to another device
4029 * @to: destination device
4030 * @from: source device
4032 * Add newly added addresses to the destination device and release
4033 * addresses that have no users left. The source device must be
4034 * locked by netif_tx_lock_bh.
4036 * This function is intended to be called from the dev->set_rx_mode
4037 * function of layered software devices.
4039 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4043 if (to->addr_len != from->addr_len)
4046 netif_addr_lock_bh(to);
4047 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4049 __dev_set_rx_mode(to);
4050 netif_addr_unlock_bh(to);
4053 EXPORT_SYMBOL(dev_unicast_sync);
4056 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4057 * @to: destination device
4058 * @from: source device
4060 * Remove all addresses that were added to the destination device by
4061 * dev_unicast_sync(). This function is intended to be called from the
4062 * dev->stop function of layered software devices.
4064 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4066 if (to->addr_len != from->addr_len)
4069 netif_addr_lock_bh(from);
4070 netif_addr_lock(to);
4071 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4072 __dev_set_rx_mode(to);
4073 netif_addr_unlock(to);
4074 netif_addr_unlock_bh(from);
4076 EXPORT_SYMBOL(dev_unicast_unsync);
4078 static void dev_unicast_flush(struct net_device *dev)
4080 netif_addr_lock_bh(dev);
4081 __hw_addr_flush(&dev->uc);
4082 netif_addr_unlock_bh(dev);
4085 static void dev_unicast_init(struct net_device *dev)
4087 __hw_addr_init(&dev->uc);
4091 static void __dev_addr_discard(struct dev_addr_list **list)
4093 struct dev_addr_list *tmp;
4095 while (*list != NULL) {
4098 if (tmp->da_users > tmp->da_gusers)
4099 printk("__dev_addr_discard: address leakage! "
4100 "da_users=%d\n", tmp->da_users);
4105 static void dev_addr_discard(struct net_device *dev)
4107 netif_addr_lock_bh(dev);
4109 __dev_addr_discard(&dev->mc_list);
4112 netif_addr_unlock_bh(dev);
4116 * dev_get_flags - get flags reported to userspace
4119 * Get the combination of flag bits exported through APIs to userspace.
4121 unsigned dev_get_flags(const struct net_device *dev)
4125 flags = (dev->flags & ~(IFF_PROMISC |
4130 (dev->gflags & (IFF_PROMISC |
4133 if (netif_running(dev)) {
4134 if (netif_oper_up(dev))
4135 flags |= IFF_RUNNING;
4136 if (netif_carrier_ok(dev))
4137 flags |= IFF_LOWER_UP;
4138 if (netif_dormant(dev))
4139 flags |= IFF_DORMANT;
4144 EXPORT_SYMBOL(dev_get_flags);
4147 * dev_change_flags - change device settings
4149 * @flags: device state flags
4151 * Change settings on device based state flags. The flags are
4152 * in the userspace exported format.
4154 int dev_change_flags(struct net_device *dev, unsigned flags)
4157 int old_flags = dev->flags;
4162 * Set the flags on our device.
4165 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4166 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4168 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4172 * Load in the correct multicast list now the flags have changed.
4175 if ((old_flags ^ flags) & IFF_MULTICAST)
4176 dev_change_rx_flags(dev, IFF_MULTICAST);
4178 dev_set_rx_mode(dev);
4181 * Have we downed the interface. We handle IFF_UP ourselves
4182 * according to user attempts to set it, rather than blindly
4187 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4188 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4191 dev_set_rx_mode(dev);
4194 if (dev->flags & IFF_UP &&
4195 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4197 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4199 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4200 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4202 dev->gflags ^= IFF_PROMISC;
4203 dev_set_promiscuity(dev, inc);
4206 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4207 is important. Some (broken) drivers set IFF_PROMISC, when
4208 IFF_ALLMULTI is requested not asking us and not reporting.
4210 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4211 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4213 dev->gflags ^= IFF_ALLMULTI;
4214 dev_set_allmulti(dev, inc);
4217 /* Exclude state transition flags, already notified */
4218 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4220 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4224 EXPORT_SYMBOL(dev_change_flags);
4227 * dev_set_mtu - Change maximum transfer unit
4229 * @new_mtu: new transfer unit
4231 * Change the maximum transfer size of the network device.
4233 int dev_set_mtu(struct net_device *dev, int new_mtu)
4235 const struct net_device_ops *ops = dev->netdev_ops;
4238 if (new_mtu == dev->mtu)
4241 /* MTU must be positive. */
4245 if (!netif_device_present(dev))
4249 if (ops->ndo_change_mtu)
4250 err = ops->ndo_change_mtu(dev, new_mtu);
4254 if (!err && dev->flags & IFF_UP)
4255 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4258 EXPORT_SYMBOL(dev_set_mtu);
4261 * dev_set_mac_address - Change Media Access Control Address
4265 * Change the hardware (MAC) address of the device
4267 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4269 const struct net_device_ops *ops = dev->netdev_ops;
4272 if (!ops->ndo_set_mac_address)
4274 if (sa->sa_family != dev->type)
4276 if (!netif_device_present(dev))
4278 err = ops->ndo_set_mac_address(dev, sa);
4280 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4283 EXPORT_SYMBOL(dev_set_mac_address);
4286 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
4288 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4291 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4297 case SIOCGIFFLAGS: /* Get interface flags */
4298 ifr->ifr_flags = (short) dev_get_flags(dev);
4301 case SIOCGIFMETRIC: /* Get the metric on the interface
4302 (currently unused) */
4303 ifr->ifr_metric = 0;
4306 case SIOCGIFMTU: /* Get the MTU of a device */
4307 ifr->ifr_mtu = dev->mtu;
4312 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4314 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4315 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4316 ifr->ifr_hwaddr.sa_family = dev->type;
4324 ifr->ifr_map.mem_start = dev->mem_start;
4325 ifr->ifr_map.mem_end = dev->mem_end;
4326 ifr->ifr_map.base_addr = dev->base_addr;
4327 ifr->ifr_map.irq = dev->irq;
4328 ifr->ifr_map.dma = dev->dma;
4329 ifr->ifr_map.port = dev->if_port;
4333 ifr->ifr_ifindex = dev->ifindex;
4337 ifr->ifr_qlen = dev->tx_queue_len;
4341 /* dev_ioctl() should ensure this case
4353 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4355 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4358 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4359 const struct net_device_ops *ops;
4364 ops = dev->netdev_ops;
4367 case SIOCSIFFLAGS: /* Set interface flags */
4368 return dev_change_flags(dev, ifr->ifr_flags);
4370 case SIOCSIFMETRIC: /* Set the metric on the interface
4371 (currently unused) */
4374 case SIOCSIFMTU: /* Set the MTU of a device */
4375 return dev_set_mtu(dev, ifr->ifr_mtu);
4378 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4380 case SIOCSIFHWBROADCAST:
4381 if (ifr->ifr_hwaddr.sa_family != dev->type)
4383 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4384 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4385 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4389 if (ops->ndo_set_config) {
4390 if (!netif_device_present(dev))
4392 return ops->ndo_set_config(dev, &ifr->ifr_map);
4397 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4398 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4400 if (!netif_device_present(dev))
4402 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4406 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4407 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4409 if (!netif_device_present(dev))
4411 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4415 if (ifr->ifr_qlen < 0)
4417 dev->tx_queue_len = ifr->ifr_qlen;
4421 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4422 return dev_change_name(dev, ifr->ifr_newname);
4425 * Unknown or private ioctl
4428 if ((cmd >= SIOCDEVPRIVATE &&
4429 cmd <= SIOCDEVPRIVATE + 15) ||
4430 cmd == SIOCBONDENSLAVE ||
4431 cmd == SIOCBONDRELEASE ||
4432 cmd == SIOCBONDSETHWADDR ||
4433 cmd == SIOCBONDSLAVEINFOQUERY ||
4434 cmd == SIOCBONDINFOQUERY ||
4435 cmd == SIOCBONDCHANGEACTIVE ||
4436 cmd == SIOCGMIIPHY ||
4437 cmd == SIOCGMIIREG ||
4438 cmd == SIOCSMIIREG ||
4439 cmd == SIOCBRADDIF ||
4440 cmd == SIOCBRDELIF ||
4441 cmd == SIOCSHWTSTAMP ||
4442 cmd == SIOCWANDEV) {
4444 if (ops->ndo_do_ioctl) {
4445 if (netif_device_present(dev))
4446 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4458 * This function handles all "interface"-type I/O control requests. The actual
4459 * 'doing' part of this is dev_ifsioc above.
4463 * dev_ioctl - network device ioctl
4464 * @net: the applicable net namespace
4465 * @cmd: command to issue
4466 * @arg: pointer to a struct ifreq in user space
4468 * Issue ioctl functions to devices. This is normally called by the
4469 * user space syscall interfaces but can sometimes be useful for
4470 * other purposes. The return value is the return from the syscall if
4471 * positive or a negative errno code on error.
4474 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4480 /* One special case: SIOCGIFCONF takes ifconf argument
4481 and requires shared lock, because it sleeps writing
4485 if (cmd == SIOCGIFCONF) {
4487 ret = dev_ifconf(net, (char __user *) arg);
4491 if (cmd == SIOCGIFNAME)
4492 return dev_ifname(net, (struct ifreq __user *)arg);
4494 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4497 ifr.ifr_name[IFNAMSIZ-1] = 0;
4499 colon = strchr(ifr.ifr_name, ':');
4504 * See which interface the caller is talking about.
4509 * These ioctl calls:
4510 * - can be done by all.
4511 * - atomic and do not require locking.
4522 dev_load(net, ifr.ifr_name);
4523 read_lock(&dev_base_lock);
4524 ret = dev_ifsioc_locked(net, &ifr, cmd);
4525 read_unlock(&dev_base_lock);
4529 if (copy_to_user(arg, &ifr,
4530 sizeof(struct ifreq)))
4536 dev_load(net, ifr.ifr_name);
4538 ret = dev_ethtool(net, &ifr);
4543 if (copy_to_user(arg, &ifr,
4544 sizeof(struct ifreq)))
4550 * These ioctl calls:
4551 * - require superuser power.
4552 * - require strict serialization.
4558 if (!capable(CAP_NET_ADMIN))
4560 dev_load(net, ifr.ifr_name);
4562 ret = dev_ifsioc(net, &ifr, cmd);
4567 if (copy_to_user(arg, &ifr,
4568 sizeof(struct ifreq)))
4574 * These ioctl calls:
4575 * - require superuser power.
4576 * - require strict serialization.
4577 * - do not return a value
4587 case SIOCSIFHWBROADCAST:
4590 case SIOCBONDENSLAVE:
4591 case SIOCBONDRELEASE:
4592 case SIOCBONDSETHWADDR:
4593 case SIOCBONDCHANGEACTIVE:
4597 if (!capable(CAP_NET_ADMIN))
4600 case SIOCBONDSLAVEINFOQUERY:
4601 case SIOCBONDINFOQUERY:
4602 dev_load(net, ifr.ifr_name);
4604 ret = dev_ifsioc(net, &ifr, cmd);
4609 /* Get the per device memory space. We can add this but
4610 * currently do not support it */
4612 /* Set the per device memory buffer space.
4613 * Not applicable in our case */
4618 * Unknown or private ioctl.
4621 if (cmd == SIOCWANDEV ||
4622 (cmd >= SIOCDEVPRIVATE &&
4623 cmd <= SIOCDEVPRIVATE + 15)) {
4624 dev_load(net, ifr.ifr_name);
4626 ret = dev_ifsioc(net, &ifr, cmd);
4628 if (!ret && copy_to_user(arg, &ifr,
4629 sizeof(struct ifreq)))
4633 /* Take care of Wireless Extensions */
4634 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4635 return wext_handle_ioctl(net, &ifr, cmd, arg);
4642 * dev_new_index - allocate an ifindex
4643 * @net: the applicable net namespace
4645 * Returns a suitable unique value for a new device interface
4646 * number. The caller must hold the rtnl semaphore or the
4647 * dev_base_lock to be sure it remains unique.
4649 static int dev_new_index(struct net *net)
4655 if (!__dev_get_by_index(net, ifindex))
4660 /* Delayed registration/unregisteration */
4661 static LIST_HEAD(net_todo_list);
4663 static void net_set_todo(struct net_device *dev)
4665 list_add_tail(&dev->todo_list, &net_todo_list);
4668 static void rollback_registered_many(struct list_head *head)
4670 struct net_device *dev;
4672 BUG_ON(dev_boot_phase);
4675 list_for_each_entry(dev, head, unreg_list) {
4676 /* Some devices call without registering
4677 * for initialization unwind.
4679 if (dev->reg_state == NETREG_UNINITIALIZED) {
4680 pr_debug("unregister_netdevice: device %s/%p never "
4681 "was registered\n", dev->name, dev);
4687 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4689 /* If device is running, close it first. */
4692 /* And unlink it from device chain. */
4693 unlist_netdevice(dev);
4695 dev->reg_state = NETREG_UNREGISTERING;
4700 list_for_each_entry(dev, head, unreg_list) {
4701 /* Shutdown queueing discipline. */
4705 /* Notify protocols, that we are about to destroy
4706 this device. They should clean all the things.
4708 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4711 * Flush the unicast and multicast chains
4713 dev_unicast_flush(dev);
4714 dev_addr_discard(dev);
4716 if (dev->netdev_ops->ndo_uninit)
4717 dev->netdev_ops->ndo_uninit(dev);
4719 /* Notifier chain MUST detach us from master device. */
4720 WARN_ON(dev->master);
4722 /* Remove entries from kobject tree */
4723 netdev_unregister_kobject(dev);
4728 list_for_each_entry(dev, head, unreg_list)
4732 static void rollback_registered(struct net_device *dev)
4736 list_add(&dev->unreg_list, &single);
4737 rollback_registered_many(&single);
4740 static void __netdev_init_queue_locks_one(struct net_device *dev,
4741 struct netdev_queue *dev_queue,
4744 spin_lock_init(&dev_queue->_xmit_lock);
4745 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4746 dev_queue->xmit_lock_owner = -1;
4749 static void netdev_init_queue_locks(struct net_device *dev)
4751 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4752 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4755 unsigned long netdev_fix_features(unsigned long features, const char *name)
4757 /* Fix illegal SG+CSUM combinations. */
4758 if ((features & NETIF_F_SG) &&
4759 !(features & NETIF_F_ALL_CSUM)) {
4761 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4762 "checksum feature.\n", name);
4763 features &= ~NETIF_F_SG;
4766 /* TSO requires that SG is present as well. */
4767 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4769 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4770 "SG feature.\n", name);
4771 features &= ~NETIF_F_TSO;
4774 if (features & NETIF_F_UFO) {
4775 if (!(features & NETIF_F_GEN_CSUM)) {
4777 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4778 "since no NETIF_F_HW_CSUM feature.\n",
4780 features &= ~NETIF_F_UFO;
4783 if (!(features & NETIF_F_SG)) {
4785 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4786 "since no NETIF_F_SG feature.\n", name);
4787 features &= ~NETIF_F_UFO;
4793 EXPORT_SYMBOL(netdev_fix_features);
4796 * register_netdevice - register a network device
4797 * @dev: device to register
4799 * Take a completed network device structure and add it to the kernel
4800 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4801 * chain. 0 is returned on success. A negative errno code is returned
4802 * on a failure to set up the device, or if the name is a duplicate.
4804 * Callers must hold the rtnl semaphore. You may want
4805 * register_netdev() instead of this.
4808 * The locking appears insufficient to guarantee two parallel registers
4809 * will not get the same name.
4812 int register_netdevice(struct net_device *dev)
4814 struct hlist_head *head;
4815 struct hlist_node *p;
4817 struct net *net = dev_net(dev);
4819 BUG_ON(dev_boot_phase);
4824 /* When net_device's are persistent, this will be fatal. */
4825 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4828 spin_lock_init(&dev->addr_list_lock);
4829 netdev_set_addr_lockdep_class(dev);
4830 netdev_init_queue_locks(dev);
4834 /* Init, if this function is available */
4835 if (dev->netdev_ops->ndo_init) {
4836 ret = dev->netdev_ops->ndo_init(dev);
4844 if (!dev_valid_name(dev->name)) {
4849 dev->ifindex = dev_new_index(net);
4850 if (dev->iflink == -1)
4851 dev->iflink = dev->ifindex;
4853 /* Check for existence of name */
4854 head = dev_name_hash(net, dev->name);
4855 hlist_for_each(p, head) {
4856 struct net_device *d
4857 = hlist_entry(p, struct net_device, name_hlist);
4858 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4864 /* Fix illegal checksum combinations */
4865 if ((dev->features & NETIF_F_HW_CSUM) &&
4866 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4867 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4869 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4872 if ((dev->features & NETIF_F_NO_CSUM) &&
4873 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4874 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4876 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4879 dev->features = netdev_fix_features(dev->features, dev->name);
4881 /* Enable software GSO if SG is supported. */
4882 if (dev->features & NETIF_F_SG)
4883 dev->features |= NETIF_F_GSO;
4885 netdev_initialize_kobject(dev);
4887 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4888 ret = notifier_to_errno(ret);
4892 ret = netdev_register_kobject(dev);
4895 dev->reg_state = NETREG_REGISTERED;
4898 * Default initial state at registry is that the
4899 * device is present.
4902 set_bit(__LINK_STATE_PRESENT, &dev->state);
4904 dev_init_scheduler(dev);
4906 list_netdevice(dev);
4908 /* Notify protocols, that a new device appeared. */
4909 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4910 ret = notifier_to_errno(ret);
4912 rollback_registered(dev);
4913 dev->reg_state = NETREG_UNREGISTERED;
4920 if (dev->netdev_ops->ndo_uninit)
4921 dev->netdev_ops->ndo_uninit(dev);
4924 EXPORT_SYMBOL(register_netdevice);
4927 * init_dummy_netdev - init a dummy network device for NAPI
4928 * @dev: device to init
4930 * This takes a network device structure and initialize the minimum
4931 * amount of fields so it can be used to schedule NAPI polls without
4932 * registering a full blown interface. This is to be used by drivers
4933 * that need to tie several hardware interfaces to a single NAPI
4934 * poll scheduler due to HW limitations.
4936 int init_dummy_netdev(struct net_device *dev)
4938 /* Clear everything. Note we don't initialize spinlocks
4939 * are they aren't supposed to be taken by any of the
4940 * NAPI code and this dummy netdev is supposed to be
4941 * only ever used for NAPI polls
4943 memset(dev, 0, sizeof(struct net_device));
4945 /* make sure we BUG if trying to hit standard
4946 * register/unregister code path
4948 dev->reg_state = NETREG_DUMMY;
4950 /* initialize the ref count */
4951 atomic_set(&dev->refcnt, 1);
4953 /* NAPI wants this */
4954 INIT_LIST_HEAD(&dev->napi_list);
4956 /* a dummy interface is started by default */
4957 set_bit(__LINK_STATE_PRESENT, &dev->state);
4958 set_bit(__LINK_STATE_START, &dev->state);
4962 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4966 * register_netdev - register a network device
4967 * @dev: device to register
4969 * Take a completed network device structure and add it to the kernel
4970 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4971 * chain. 0 is returned on success. A negative errno code is returned
4972 * on a failure to set up the device, or if the name is a duplicate.
4974 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4975 * and expands the device name if you passed a format string to
4978 int register_netdev(struct net_device *dev)
4985 * If the name is a format string the caller wants us to do a
4988 if (strchr(dev->name, '%')) {
4989 err = dev_alloc_name(dev, dev->name);
4994 err = register_netdevice(dev);
4999 EXPORT_SYMBOL(register_netdev);
5002 * netdev_wait_allrefs - wait until all references are gone.
5004 * This is called when unregistering network devices.
5006 * Any protocol or device that holds a reference should register
5007 * for netdevice notification, and cleanup and put back the
5008 * reference if they receive an UNREGISTER event.
5009 * We can get stuck here if buggy protocols don't correctly
5012 static void netdev_wait_allrefs(struct net_device *dev)
5014 unsigned long rebroadcast_time, warning_time;
5016 rebroadcast_time = warning_time = jiffies;
5017 while (atomic_read(&dev->refcnt) != 0) {
5018 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5021 /* Rebroadcast unregister notification */
5022 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5024 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5026 /* We must not have linkwatch events
5027 * pending on unregister. If this
5028 * happens, we simply run the queue
5029 * unscheduled, resulting in a noop
5032 linkwatch_run_queue();
5037 rebroadcast_time = jiffies;
5042 if (time_after(jiffies, warning_time + 10 * HZ)) {
5043 printk(KERN_EMERG "unregister_netdevice: "
5044 "waiting for %s to become free. Usage "
5046 dev->name, atomic_read(&dev->refcnt));
5047 warning_time = jiffies;
5056 * register_netdevice(x1);
5057 * register_netdevice(x2);
5059 * unregister_netdevice(y1);
5060 * unregister_netdevice(y2);
5066 * We are invoked by rtnl_unlock().
5067 * This allows us to deal with problems:
5068 * 1) We can delete sysfs objects which invoke hotplug
5069 * without deadlocking with linkwatch via keventd.
5070 * 2) Since we run with the RTNL semaphore not held, we can sleep
5071 * safely in order to wait for the netdev refcnt to drop to zero.
5073 * We must not return until all unregister events added during
5074 * the interval the lock was held have been completed.
5076 void netdev_run_todo(void)
5078 struct list_head list;
5080 /* Snapshot list, allow later requests */
5081 list_replace_init(&net_todo_list, &list);
5085 while (!list_empty(&list)) {
5086 struct net_device *dev
5087 = list_entry(list.next, struct net_device, todo_list);
5088 list_del(&dev->todo_list);
5090 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5091 printk(KERN_ERR "network todo '%s' but state %d\n",
5092 dev->name, dev->reg_state);
5097 dev->reg_state = NETREG_UNREGISTERED;
5099 on_each_cpu(flush_backlog, dev, 1);
5101 netdev_wait_allrefs(dev);
5104 BUG_ON(atomic_read(&dev->refcnt));
5105 WARN_ON(dev->ip_ptr);
5106 WARN_ON(dev->ip6_ptr);
5107 WARN_ON(dev->dn_ptr);
5109 if (dev->destructor)
5110 dev->destructor(dev);
5112 /* Free network device */
5113 kobject_put(&dev->dev.kobj);
5118 * dev_get_stats - get network device statistics
5119 * @dev: device to get statistics from
5121 * Get network statistics from device. The device driver may provide
5122 * its own method by setting dev->netdev_ops->get_stats; otherwise
5123 * the internal statistics structure is used.
5125 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5127 const struct net_device_ops *ops = dev->netdev_ops;
5129 if (ops->ndo_get_stats)
5130 return ops->ndo_get_stats(dev);
5132 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5133 struct net_device_stats *stats = &dev->stats;
5135 struct netdev_queue *txq;
5137 for (i = 0; i < dev->num_tx_queues; i++) {
5138 txq = netdev_get_tx_queue(dev, i);
5139 tx_bytes += txq->tx_bytes;
5140 tx_packets += txq->tx_packets;
5141 tx_dropped += txq->tx_dropped;
5143 if (tx_bytes || tx_packets || tx_dropped) {
5144 stats->tx_bytes = tx_bytes;
5145 stats->tx_packets = tx_packets;
5146 stats->tx_dropped = tx_dropped;
5151 EXPORT_SYMBOL(dev_get_stats);
5153 static void netdev_init_one_queue(struct net_device *dev,
5154 struct netdev_queue *queue,
5160 static void netdev_init_queues(struct net_device *dev)
5162 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5163 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5164 spin_lock_init(&dev->tx_global_lock);
5168 * alloc_netdev_mq - allocate network device
5169 * @sizeof_priv: size of private data to allocate space for
5170 * @name: device name format string
5171 * @setup: callback to initialize device
5172 * @queue_count: the number of subqueues to allocate
5174 * Allocates a struct net_device with private data area for driver use
5175 * and performs basic initialization. Also allocates subquue structs
5176 * for each queue on the device at the end of the netdevice.
5178 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5179 void (*setup)(struct net_device *), unsigned int queue_count)
5181 struct netdev_queue *tx;
5182 struct net_device *dev;
5184 struct net_device *p;
5186 BUG_ON(strlen(name) >= sizeof(dev->name));
5188 alloc_size = sizeof(struct net_device);
5190 /* ensure 32-byte alignment of private area */
5191 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5192 alloc_size += sizeof_priv;
5194 /* ensure 32-byte alignment of whole construct */
5195 alloc_size += NETDEV_ALIGN - 1;
5197 p = kzalloc(alloc_size, GFP_KERNEL);
5199 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5203 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5205 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5210 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5211 dev->padded = (char *)dev - (char *)p;
5213 if (dev_addr_init(dev))
5216 dev_unicast_init(dev);
5218 dev_net_set(dev, &init_net);
5221 dev->num_tx_queues = queue_count;
5222 dev->real_num_tx_queues = queue_count;
5224 dev->gso_max_size = GSO_MAX_SIZE;
5226 netdev_init_queues(dev);
5228 INIT_LIST_HEAD(&dev->napi_list);
5229 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5231 strcpy(dev->name, name);
5241 EXPORT_SYMBOL(alloc_netdev_mq);
5244 * free_netdev - free network device
5247 * This function does the last stage of destroying an allocated device
5248 * interface. The reference to the device object is released.
5249 * If this is the last reference then it will be freed.
5251 void free_netdev(struct net_device *dev)
5253 struct napi_struct *p, *n;
5255 release_net(dev_net(dev));
5259 /* Flush device addresses */
5260 dev_addr_flush(dev);
5262 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5265 /* Compatibility with error handling in drivers */
5266 if (dev->reg_state == NETREG_UNINITIALIZED) {
5267 kfree((char *)dev - dev->padded);
5271 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5272 dev->reg_state = NETREG_RELEASED;
5274 /* will free via device release */
5275 put_device(&dev->dev);
5277 EXPORT_SYMBOL(free_netdev);
5280 * synchronize_net - Synchronize with packet receive processing
5282 * Wait for packets currently being received to be done.
5283 * Does not block later packets from starting.
5285 void synchronize_net(void)
5290 EXPORT_SYMBOL(synchronize_net);
5293 * unregister_netdevice_queue - remove device from the kernel
5297 * This function shuts down a device interface and removes it
5298 * from the kernel tables.
5299 * If head not NULL, device is queued to be unregistered later.
5301 * Callers must hold the rtnl semaphore. You may want
5302 * unregister_netdev() instead of this.
5305 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5310 list_add_tail(&dev->unreg_list, head);
5312 rollback_registered(dev);
5313 /* Finish processing unregister after unlock */
5317 EXPORT_SYMBOL(unregister_netdevice_queue);
5320 * unregister_netdevice_many - unregister many devices
5321 * @head: list of devices
5324 void unregister_netdevice_many(struct list_head *head)
5326 struct net_device *dev;
5328 if (!list_empty(head)) {
5329 rollback_registered_many(head);
5330 list_for_each_entry(dev, head, unreg_list)
5334 EXPORT_SYMBOL(unregister_netdevice_many);
5337 * unregister_netdev - remove device from the kernel
5340 * This function shuts down a device interface and removes it
5341 * from the kernel tables.
5343 * This is just a wrapper for unregister_netdevice that takes
5344 * the rtnl semaphore. In general you want to use this and not
5345 * unregister_netdevice.
5347 void unregister_netdev(struct net_device *dev)
5350 unregister_netdevice(dev);
5353 EXPORT_SYMBOL(unregister_netdev);
5356 * dev_change_net_namespace - move device to different nethost namespace
5358 * @net: network namespace
5359 * @pat: If not NULL name pattern to try if the current device name
5360 * is already taken in the destination network namespace.
5362 * This function shuts down a device interface and moves it
5363 * to a new network namespace. On success 0 is returned, on
5364 * a failure a netagive errno code is returned.
5366 * Callers must hold the rtnl semaphore.
5369 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5372 const char *destname;
5377 /* Don't allow namespace local devices to be moved. */
5379 if (dev->features & NETIF_F_NETNS_LOCAL)
5383 /* Don't allow real devices to be moved when sysfs
5387 if (dev->dev.parent)
5391 /* Ensure the device has been registrered */
5393 if (dev->reg_state != NETREG_REGISTERED)
5396 /* Get out if there is nothing todo */
5398 if (net_eq(dev_net(dev), net))
5401 /* Pick the destination device name, and ensure
5402 * we can use it in the destination network namespace.
5405 destname = dev->name;
5406 if (__dev_get_by_name(net, destname)) {
5407 /* We get here if we can't use the current device name */
5410 if (!dev_valid_name(pat))
5412 if (strchr(pat, '%')) {
5413 if (__dev_alloc_name(net, pat, buf) < 0)
5418 if (__dev_get_by_name(net, destname))
5423 * And now a mini version of register_netdevice unregister_netdevice.
5426 /* If device is running close it first. */
5429 /* And unlink it from device chain */
5431 unlist_netdevice(dev);
5435 /* Shutdown queueing discipline. */
5438 /* Notify protocols, that we are about to destroy
5439 this device. They should clean all the things.
5441 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5444 * Flush the unicast and multicast chains
5446 dev_unicast_flush(dev);
5447 dev_addr_discard(dev);
5449 netdev_unregister_kobject(dev);
5451 /* Actually switch the network namespace */
5452 dev_net_set(dev, net);
5454 /* Assign the new device name */
5455 if (destname != dev->name)
5456 strcpy(dev->name, destname);
5458 /* If there is an ifindex conflict assign a new one */
5459 if (__dev_get_by_index(net, dev->ifindex)) {
5460 int iflink = (dev->iflink == dev->ifindex);
5461 dev->ifindex = dev_new_index(net);
5463 dev->iflink = dev->ifindex;
5466 /* Fixup kobjects */
5467 err = netdev_register_kobject(dev);
5470 /* Add the device back in the hashes */
5471 list_netdevice(dev);
5473 /* Notify protocols, that a new device appeared. */
5474 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5481 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5483 static int dev_cpu_callback(struct notifier_block *nfb,
5484 unsigned long action,
5487 struct sk_buff **list_skb;
5488 struct Qdisc **list_net;
5489 struct sk_buff *skb;
5490 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5491 struct softnet_data *sd, *oldsd;
5493 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5496 local_irq_disable();
5497 cpu = smp_processor_id();
5498 sd = &per_cpu(softnet_data, cpu);
5499 oldsd = &per_cpu(softnet_data, oldcpu);
5501 /* Find end of our completion_queue. */
5502 list_skb = &sd->completion_queue;
5504 list_skb = &(*list_skb)->next;
5505 /* Append completion queue from offline CPU. */
5506 *list_skb = oldsd->completion_queue;
5507 oldsd->completion_queue = NULL;
5509 /* Find end of our output_queue. */
5510 list_net = &sd->output_queue;
5512 list_net = &(*list_net)->next_sched;
5513 /* Append output queue from offline CPU. */
5514 *list_net = oldsd->output_queue;
5515 oldsd->output_queue = NULL;
5517 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5520 /* Process offline CPU's input_pkt_queue */
5521 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5529 * netdev_increment_features - increment feature set by one
5530 * @all: current feature set
5531 * @one: new feature set
5532 * @mask: mask feature set
5534 * Computes a new feature set after adding a device with feature set
5535 * @one to the master device with current feature set @all. Will not
5536 * enable anything that is off in @mask. Returns the new feature set.
5538 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5541 /* If device needs checksumming, downgrade to it. */
5542 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5543 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5544 else if (mask & NETIF_F_ALL_CSUM) {
5545 /* If one device supports v4/v6 checksumming, set for all. */
5546 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5547 !(all & NETIF_F_GEN_CSUM)) {
5548 all &= ~NETIF_F_ALL_CSUM;
5549 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5552 /* If one device supports hw checksumming, set for all. */
5553 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5554 all &= ~NETIF_F_ALL_CSUM;
5555 all |= NETIF_F_HW_CSUM;
5559 one |= NETIF_F_ALL_CSUM;
5561 one |= all & NETIF_F_ONE_FOR_ALL;
5562 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5563 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5567 EXPORT_SYMBOL(netdev_increment_features);
5569 static struct hlist_head *netdev_create_hash(void)
5572 struct hlist_head *hash;
5574 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5576 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5577 INIT_HLIST_HEAD(&hash[i]);
5582 /* Initialize per network namespace state */
5583 static int __net_init netdev_init(struct net *net)
5585 INIT_LIST_HEAD(&net->dev_base_head);
5587 net->dev_name_head = netdev_create_hash();
5588 if (net->dev_name_head == NULL)
5591 net->dev_index_head = netdev_create_hash();
5592 if (net->dev_index_head == NULL)
5598 kfree(net->dev_name_head);
5604 * netdev_drivername - network driver for the device
5605 * @dev: network device
5606 * @buffer: buffer for resulting name
5607 * @len: size of buffer
5609 * Determine network driver for device.
5611 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5613 const struct device_driver *driver;
5614 const struct device *parent;
5616 if (len <= 0 || !buffer)
5620 parent = dev->dev.parent;
5625 driver = parent->driver;
5626 if (driver && driver->name)
5627 strlcpy(buffer, driver->name, len);
5631 static void __net_exit netdev_exit(struct net *net)
5633 kfree(net->dev_name_head);
5634 kfree(net->dev_index_head);
5637 static struct pernet_operations __net_initdata netdev_net_ops = {
5638 .init = netdev_init,
5639 .exit = netdev_exit,
5642 static void __net_exit default_device_exit(struct net *net)
5644 struct net_device *dev;
5646 * Push all migratable of the network devices back to the
5647 * initial network namespace
5651 for_each_netdev(net, dev) {
5653 char fb_name[IFNAMSIZ];
5655 /* Ignore unmoveable devices (i.e. loopback) */
5656 if (dev->features & NETIF_F_NETNS_LOCAL)
5659 /* Delete virtual devices */
5660 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5661 dev->rtnl_link_ops->dellink(dev, NULL);
5665 /* Push remaing network devices to init_net */
5666 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5667 err = dev_change_net_namespace(dev, &init_net, fb_name);
5669 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5670 __func__, dev->name, err);
5678 static struct pernet_operations __net_initdata default_device_ops = {
5679 .exit = default_device_exit,
5683 * Initialize the DEV module. At boot time this walks the device list and
5684 * unhooks any devices that fail to initialise (normally hardware not
5685 * present) and leaves us with a valid list of present and active devices.
5690 * This is called single threaded during boot, so no need
5691 * to take the rtnl semaphore.
5693 static int __init net_dev_init(void)
5695 int i, rc = -ENOMEM;
5697 BUG_ON(!dev_boot_phase);
5699 if (dev_proc_init())
5702 if (netdev_kobject_init())
5705 INIT_LIST_HEAD(&ptype_all);
5706 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5707 INIT_LIST_HEAD(&ptype_base[i]);
5709 if (register_pernet_subsys(&netdev_net_ops))
5713 * Initialise the packet receive queues.
5716 for_each_possible_cpu(i) {
5717 struct softnet_data *queue;
5719 queue = &per_cpu(softnet_data, i);
5720 skb_queue_head_init(&queue->input_pkt_queue);
5721 queue->completion_queue = NULL;
5722 INIT_LIST_HEAD(&queue->poll_list);
5724 queue->backlog.poll = process_backlog;
5725 queue->backlog.weight = weight_p;
5726 queue->backlog.gro_list = NULL;
5727 queue->backlog.gro_count = 0;
5732 /* The loopback device is special if any other network devices
5733 * is present in a network namespace the loopback device must
5734 * be present. Since we now dynamically allocate and free the
5735 * loopback device ensure this invariant is maintained by
5736 * keeping the loopback device as the first device on the
5737 * list of network devices. Ensuring the loopback devices
5738 * is the first device that appears and the last network device
5741 if (register_pernet_device(&loopback_net_ops))
5744 if (register_pernet_device(&default_device_ops))
5747 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5748 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5750 hotcpu_notifier(dev_cpu_callback, 0);
5758 subsys_initcall(net_dev_init);
5760 static int __init initialize_hashrnd(void)
5762 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5766 late_initcall_sync(initialize_hashrnd);