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/kallsyms.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
131 #include "net-sysfs.h"
134 * The list of packet types we will receive (as opposed to discard)
135 * and the routines to invoke.
137 * Why 16. Because with 16 the only overlap we get on a hash of the
138 * low nibble of the protocol value is RARP/SNAP/X.25.
140 * NOTE: That is no longer true with the addition of VLAN tags. Not
141 * sure which should go first, but I bet it won't make much
142 * difference if we are running VLANs. The good news is that
143 * this protocol won't be in the list unless compiled in, so
144 * the average user (w/out VLANs) will not be adversely affected.
161 #define PTYPE_HASH_SIZE (16)
162 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
164 static DEFINE_SPINLOCK(ptype_lock);
165 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
166 static struct list_head ptype_all __read_mostly; /* Taps */
168 #ifdef CONFIG_NET_DMA
170 struct dma_client client;
172 cpumask_t channel_mask;
173 struct dma_chan **channels;
176 static enum dma_state_client
177 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
178 enum dma_state state);
180 static struct net_dma net_dma = {
182 .event_callback = netdev_dma_event,
188 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
191 * Pure readers hold dev_base_lock for reading.
193 * Writers must hold the rtnl semaphore while they loop through the
194 * dev_base_head list, and hold dev_base_lock for writing when they do the
195 * actual updates. This allows pure readers to access the list even
196 * while a writer is preparing to update it.
198 * To put it another way, dev_base_lock is held for writing only to
199 * protect against pure readers; the rtnl semaphore provides the
200 * protection against other writers.
202 * See, for example usages, register_netdevice() and
203 * unregister_netdevice(), which must be called with the rtnl
206 DEFINE_RWLOCK(dev_base_lock);
208 EXPORT_SYMBOL(dev_base_lock);
210 #define NETDEV_HASHBITS 8
211 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
213 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
215 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
216 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
219 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
221 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
224 /* Device list insertion */
225 static int list_netdevice(struct net_device *dev)
227 struct net *net = dev_net(dev);
231 write_lock_bh(&dev_base_lock);
232 list_add_tail(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
235 write_unlock_bh(&dev_base_lock);
239 /* Device list removal */
240 static void unlist_netdevice(struct net_device *dev)
244 /* Unlink dev from the device chain */
245 write_lock_bh(&dev_base_lock);
246 list_del(&dev->dev_list);
247 hlist_del(&dev->name_hlist);
248 hlist_del(&dev->index_hlist);
249 write_unlock_bh(&dev_base_lock);
256 static RAW_NOTIFIER_HEAD(netdev_chain);
259 * Device drivers call our routines to queue packets here. We empty the
260 * queue in the local softnet handler.
263 DEFINE_PER_CPU(struct softnet_data, softnet_data);
265 #ifdef CONFIG_LOCKDEP
267 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
268 * according to dev->type
270 static const unsigned short netdev_lock_type[] =
271 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
272 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
273 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
274 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
275 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
276 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
277 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
278 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
279 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
280 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
281 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
282 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
283 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
284 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
287 static const char *netdev_lock_name[] =
288 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
289 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
290 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
291 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
292 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
293 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
294 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
295 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
296 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
297 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
298 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
299 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
300 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
301 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
304 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
305 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
307 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
311 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
312 if (netdev_lock_type[i] == dev_type)
314 /* the last key is used by default */
315 return ARRAY_SIZE(netdev_lock_type) - 1;
318 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
319 unsigned short dev_type)
323 i = netdev_lock_pos(dev_type);
324 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
325 netdev_lock_name[i]);
328 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
332 i = netdev_lock_pos(dev->type);
333 lockdep_set_class_and_name(&dev->addr_list_lock,
334 &netdev_addr_lock_key[i],
335 netdev_lock_name[i]);
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
342 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
347 /*******************************************************************************
349 Protocol management and registration routines
351 *******************************************************************************/
354 * Add a protocol ID to the list. Now that the input handler is
355 * smarter we can dispense with all the messy stuff that used to be
358 * BEWARE!!! Protocol handlers, mangling input packets,
359 * MUST BE last in hash buckets and checking protocol handlers
360 * MUST start from promiscuous ptype_all chain in net_bh.
361 * It is true now, do not change it.
362 * Explanation follows: if protocol handler, mangling packet, will
363 * be the first on list, it is not able to sense, that packet
364 * is cloned and should be copied-on-write, so that it will
365 * change it and subsequent readers will get broken packet.
370 * dev_add_pack - add packet handler
371 * @pt: packet type declaration
373 * Add a protocol handler to the networking stack. The passed &packet_type
374 * is linked into kernel lists and may not be freed until it has been
375 * removed from the kernel lists.
377 * This call does not sleep therefore it can not
378 * guarantee all CPU's that are in middle of receiving packets
379 * will see the new packet type (until the next received packet).
382 void dev_add_pack(struct packet_type *pt)
386 spin_lock_bh(&ptype_lock);
387 if (pt->type == htons(ETH_P_ALL))
388 list_add_rcu(&pt->list, &ptype_all);
390 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
391 list_add_rcu(&pt->list, &ptype_base[hash]);
393 spin_unlock_bh(&ptype_lock);
397 * __dev_remove_pack - remove packet handler
398 * @pt: packet type declaration
400 * Remove a protocol handler that was previously added to the kernel
401 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
402 * from the kernel lists and can be freed or reused once this function
405 * The packet type might still be in use by receivers
406 * and must not be freed until after all the CPU's have gone
407 * through a quiescent state.
409 void __dev_remove_pack(struct packet_type *pt)
411 struct list_head *head;
412 struct packet_type *pt1;
414 spin_lock_bh(&ptype_lock);
416 if (pt->type == htons(ETH_P_ALL))
419 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
421 list_for_each_entry(pt1, head, list) {
423 list_del_rcu(&pt->list);
428 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
430 spin_unlock_bh(&ptype_lock);
433 * dev_remove_pack - remove packet handler
434 * @pt: packet type declaration
436 * Remove a protocol handler that was previously added to the kernel
437 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
438 * from the kernel lists and can be freed or reused once this function
441 * This call sleeps to guarantee that no CPU is looking at the packet
444 void dev_remove_pack(struct packet_type *pt)
446 __dev_remove_pack(pt);
451 /******************************************************************************
453 Device Boot-time Settings Routines
455 *******************************************************************************/
457 /* Boot time configuration table */
458 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
461 * netdev_boot_setup_add - add new setup entry
462 * @name: name of the device
463 * @map: configured settings for the device
465 * Adds new setup entry to the dev_boot_setup list. The function
466 * returns 0 on error and 1 on success. This is a generic routine to
469 static int netdev_boot_setup_add(char *name, struct ifmap *map)
471 struct netdev_boot_setup *s;
475 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
476 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
477 memset(s[i].name, 0, sizeof(s[i].name));
478 strlcpy(s[i].name, name, IFNAMSIZ);
479 memcpy(&s[i].map, map, sizeof(s[i].map));
484 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
488 * netdev_boot_setup_check - check boot time settings
489 * @dev: the netdevice
491 * Check boot time settings for the device.
492 * The found settings are set for the device to be used
493 * later in the device probing.
494 * Returns 0 if no settings found, 1 if they are.
496 int netdev_boot_setup_check(struct net_device *dev)
498 struct netdev_boot_setup *s = dev_boot_setup;
501 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
502 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
503 !strcmp(dev->name, s[i].name)) {
504 dev->irq = s[i].map.irq;
505 dev->base_addr = s[i].map.base_addr;
506 dev->mem_start = s[i].map.mem_start;
507 dev->mem_end = s[i].map.mem_end;
516 * netdev_boot_base - get address from boot time settings
517 * @prefix: prefix for network device
518 * @unit: id for network device
520 * Check boot time settings for the base address of device.
521 * The found settings are set for the device to be used
522 * later in the device probing.
523 * Returns 0 if no settings found.
525 unsigned long netdev_boot_base(const char *prefix, int unit)
527 const struct netdev_boot_setup *s = dev_boot_setup;
531 sprintf(name, "%s%d", prefix, unit);
534 * If device already registered then return base of 1
535 * to indicate not to probe for this interface
537 if (__dev_get_by_name(&init_net, name))
540 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
541 if (!strcmp(name, s[i].name))
542 return s[i].map.base_addr;
547 * Saves at boot time configured settings for any netdevice.
549 int __init netdev_boot_setup(char *str)
554 str = get_options(str, ARRAY_SIZE(ints), ints);
559 memset(&map, 0, sizeof(map));
563 map.base_addr = ints[2];
565 map.mem_start = ints[3];
567 map.mem_end = ints[4];
569 /* Add new entry to the list */
570 return netdev_boot_setup_add(str, &map);
573 __setup("netdev=", netdev_boot_setup);
575 /*******************************************************************************
577 Device Interface Subroutines
579 *******************************************************************************/
582 * __dev_get_by_name - find a device by its name
583 * @net: the applicable net namespace
584 * @name: name to find
586 * Find an interface by name. Must be called under RTNL semaphore
587 * or @dev_base_lock. If the name is found a pointer to the device
588 * is returned. If the name is not found then %NULL is returned. The
589 * reference counters are not incremented so the caller must be
590 * careful with locks.
593 struct net_device *__dev_get_by_name(struct net *net, const char *name)
595 struct hlist_node *p;
597 hlist_for_each(p, dev_name_hash(net, name)) {
598 struct net_device *dev
599 = hlist_entry(p, struct net_device, name_hlist);
600 if (!strncmp(dev->name, name, IFNAMSIZ))
607 * dev_get_by_name - find a device by its name
608 * @net: the applicable net namespace
609 * @name: name to find
611 * Find an interface by name. This can be called from any
612 * context and does its own locking. The returned handle has
613 * the usage count incremented and the caller must use dev_put() to
614 * release it when it is no longer needed. %NULL is returned if no
615 * matching device is found.
618 struct net_device *dev_get_by_name(struct net *net, const char *name)
620 struct net_device *dev;
622 read_lock(&dev_base_lock);
623 dev = __dev_get_by_name(net, name);
626 read_unlock(&dev_base_lock);
631 * __dev_get_by_index - find a device by its ifindex
632 * @net: the applicable net namespace
633 * @ifindex: index of device
635 * Search for an interface by index. Returns %NULL if the device
636 * is not found or a pointer to the device. The device has not
637 * had its reference counter increased so the caller must be careful
638 * about locking. The caller must hold either the RTNL semaphore
642 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
644 struct hlist_node *p;
646 hlist_for_each(p, dev_index_hash(net, ifindex)) {
647 struct net_device *dev
648 = hlist_entry(p, struct net_device, index_hlist);
649 if (dev->ifindex == ifindex)
657 * dev_get_by_index - find a device by its ifindex
658 * @net: the applicable net namespace
659 * @ifindex: index of device
661 * Search for an interface by index. Returns NULL if the device
662 * is not found or a pointer to the device. The device returned has
663 * had a reference added and the pointer is safe until the user calls
664 * dev_put to indicate they have finished with it.
667 struct net_device *dev_get_by_index(struct net *net, int ifindex)
669 struct net_device *dev;
671 read_lock(&dev_base_lock);
672 dev = __dev_get_by_index(net, ifindex);
675 read_unlock(&dev_base_lock);
680 * dev_getbyhwaddr - find a device by its hardware address
681 * @net: the applicable net namespace
682 * @type: media type of device
683 * @ha: hardware address
685 * Search for an interface by MAC address. Returns NULL if the device
686 * is not found or a pointer to the device. The caller must hold the
687 * rtnl semaphore. The returned device has not had its ref count increased
688 * and the caller must therefore be careful about locking
691 * If the API was consistent this would be __dev_get_by_hwaddr
694 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
696 struct net_device *dev;
700 for_each_netdev(net, dev)
701 if (dev->type == type &&
702 !memcmp(dev->dev_addr, ha, dev->addr_len))
708 EXPORT_SYMBOL(dev_getbyhwaddr);
710 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
712 struct net_device *dev;
715 for_each_netdev(net, dev)
716 if (dev->type == type)
722 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
724 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
726 struct net_device *dev;
729 dev = __dev_getfirstbyhwtype(net, type);
736 EXPORT_SYMBOL(dev_getfirstbyhwtype);
739 * dev_get_by_flags - find any device with given flags
740 * @net: the applicable net namespace
741 * @if_flags: IFF_* values
742 * @mask: bitmask of bits in if_flags to check
744 * Search for any interface with the given flags. Returns NULL if a device
745 * is not found or a pointer to the device. The device returned has
746 * had a reference added and the pointer is safe until the user calls
747 * dev_put to indicate they have finished with it.
750 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
752 struct net_device *dev, *ret;
755 read_lock(&dev_base_lock);
756 for_each_netdev(net, dev) {
757 if (((dev->flags ^ if_flags) & mask) == 0) {
763 read_unlock(&dev_base_lock);
768 * dev_valid_name - check if name is okay for network device
771 * Network device names need to be valid file names to
772 * to allow sysfs to work. We also disallow any kind of
775 int dev_valid_name(const char *name)
779 if (strlen(name) >= IFNAMSIZ)
781 if (!strcmp(name, ".") || !strcmp(name, ".."))
785 if (*name == '/' || isspace(*name))
793 * __dev_alloc_name - allocate a name for a device
794 * @net: network namespace to allocate the device name in
795 * @name: name format string
796 * @buf: scratch buffer and result name string
798 * Passed a format string - eg "lt%d" it will try and find a suitable
799 * id. It scans list of devices to build up a free map, then chooses
800 * the first empty slot. The caller must hold the dev_base or rtnl lock
801 * while allocating the name and adding the device in order to avoid
803 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
804 * Returns the number of the unit assigned or a negative errno code.
807 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
811 const int max_netdevices = 8*PAGE_SIZE;
812 unsigned long *inuse;
813 struct net_device *d;
815 p = strnchr(name, IFNAMSIZ-1, '%');
818 * Verify the string as this thing may have come from
819 * the user. There must be either one "%d" and no other "%"
822 if (p[1] != 'd' || strchr(p + 2, '%'))
825 /* Use one page as a bit array of possible slots */
826 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
830 for_each_netdev(net, d) {
831 if (!sscanf(d->name, name, &i))
833 if (i < 0 || i >= max_netdevices)
836 /* avoid cases where sscanf is not exact inverse of printf */
837 snprintf(buf, IFNAMSIZ, name, i);
838 if (!strncmp(buf, d->name, IFNAMSIZ))
842 i = find_first_zero_bit(inuse, max_netdevices);
843 free_page((unsigned long) inuse);
846 snprintf(buf, IFNAMSIZ, name, i);
847 if (!__dev_get_by_name(net, buf))
850 /* It is possible to run out of possible slots
851 * when the name is long and there isn't enough space left
852 * for the digits, or if all bits are used.
858 * dev_alloc_name - allocate a name for a device
860 * @name: name format string
862 * Passed a format string - eg "lt%d" it will try and find a suitable
863 * id. It scans list of devices to build up a free map, then chooses
864 * the first empty slot. The caller must hold the dev_base or rtnl lock
865 * while allocating the name and adding the device in order to avoid
867 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
868 * Returns the number of the unit assigned or a negative errno code.
871 int dev_alloc_name(struct net_device *dev, const char *name)
877 BUG_ON(!dev_net(dev));
879 ret = __dev_alloc_name(net, name, buf);
881 strlcpy(dev->name, buf, IFNAMSIZ);
887 * dev_change_name - change name of a device
889 * @newname: name (or format string) must be at least IFNAMSIZ
891 * Change name of a device, can pass format strings "eth%d".
894 int dev_change_name(struct net_device *dev, const char *newname)
896 char oldname[IFNAMSIZ];
902 BUG_ON(!dev_net(dev));
905 if (dev->flags & IFF_UP)
908 if (!dev_valid_name(newname))
911 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
914 memcpy(oldname, dev->name, IFNAMSIZ);
916 if (strchr(newname, '%')) {
917 err = dev_alloc_name(dev, newname);
921 else if (__dev_get_by_name(net, newname))
924 strlcpy(dev->name, newname, IFNAMSIZ);
927 /* For now only devices in the initial network namespace
930 if (net == &init_net) {
931 ret = device_rename(&dev->dev, dev->name);
933 memcpy(dev->name, oldname, IFNAMSIZ);
938 write_lock_bh(&dev_base_lock);
939 hlist_del(&dev->name_hlist);
940 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
941 write_unlock_bh(&dev_base_lock);
943 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
944 ret = notifier_to_errno(ret);
949 "%s: name change rollback failed: %d.\n",
953 memcpy(dev->name, oldname, IFNAMSIZ);
962 * dev_set_alias - change ifalias of a device
964 * @alias: name up to IFALIASZ
965 * @len: limit of bytes to copy from info
967 * Set ifalias for a device,
969 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
984 dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL);
988 strlcpy(dev->ifalias, alias, len+1);
994 * netdev_features_change - device changes features
995 * @dev: device to cause notification
997 * Called to indicate a device has changed features.
999 void netdev_features_change(struct net_device *dev)
1001 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1003 EXPORT_SYMBOL(netdev_features_change);
1006 * netdev_state_change - device changes state
1007 * @dev: device to cause notification
1009 * Called to indicate a device has changed state. This function calls
1010 * the notifier chains for netdev_chain and sends a NEWLINK message
1011 * to the routing socket.
1013 void netdev_state_change(struct net_device *dev)
1015 if (dev->flags & IFF_UP) {
1016 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1017 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1021 void netdev_bonding_change(struct net_device *dev)
1023 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
1025 EXPORT_SYMBOL(netdev_bonding_change);
1028 * dev_load - load a network module
1029 * @net: the applicable net namespace
1030 * @name: name of interface
1032 * If a network interface is not present and the process has suitable
1033 * privileges this function loads the module. If module loading is not
1034 * available in this kernel then it becomes a nop.
1037 void dev_load(struct net *net, const char *name)
1039 struct net_device *dev;
1041 read_lock(&dev_base_lock);
1042 dev = __dev_get_by_name(net, name);
1043 read_unlock(&dev_base_lock);
1045 if (!dev && capable(CAP_SYS_MODULE))
1046 request_module("%s", name);
1050 * dev_open - prepare an interface for use.
1051 * @dev: device to open
1053 * Takes a device from down to up state. The device's private open
1054 * function is invoked and then the multicast lists are loaded. Finally
1055 * the device is moved into the up state and a %NETDEV_UP message is
1056 * sent to the netdev notifier chain.
1058 * Calling this function on an active interface is a nop. On a failure
1059 * a negative errno code is returned.
1061 int dev_open(struct net_device *dev)
1071 if (dev->flags & IFF_UP)
1075 * Is it even present?
1077 if (!netif_device_present(dev))
1081 * Call device private open method
1083 set_bit(__LINK_STATE_START, &dev->state);
1085 if (dev->validate_addr)
1086 ret = dev->validate_addr(dev);
1088 if (!ret && dev->open)
1089 ret = dev->open(dev);
1092 * If it went open OK then:
1096 clear_bit(__LINK_STATE_START, &dev->state);
1101 dev->flags |= IFF_UP;
1104 * Initialize multicasting status
1106 dev_set_rx_mode(dev);
1109 * Wakeup transmit queue engine
1114 * ... and announce new interface.
1116 call_netdevice_notifiers(NETDEV_UP, dev);
1123 * dev_close - shutdown an interface.
1124 * @dev: device to shutdown
1126 * This function moves an active device into down state. A
1127 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1128 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1131 int dev_close(struct net_device *dev)
1137 if (!(dev->flags & IFF_UP))
1141 * Tell people we are going down, so that they can
1142 * prepare to death, when device is still operating.
1144 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1146 clear_bit(__LINK_STATE_START, &dev->state);
1148 /* Synchronize to scheduled poll. We cannot touch poll list,
1149 * it can be even on different cpu. So just clear netif_running().
1151 * dev->stop() will invoke napi_disable() on all of it's
1152 * napi_struct instances on this device.
1154 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1156 dev_deactivate(dev);
1159 * Call the device specific close. This cannot fail.
1160 * Only if device is UP
1162 * We allow it to be called even after a DETACH hot-plug
1169 * Device is now down.
1172 dev->flags &= ~IFF_UP;
1175 * Tell people we are down
1177 call_netdevice_notifiers(NETDEV_DOWN, dev);
1184 * dev_disable_lro - disable Large Receive Offload on a device
1187 * Disable Large Receive Offload (LRO) on a net device. Must be
1188 * called under RTNL. This is needed if received packets may be
1189 * forwarded to another interface.
1191 void dev_disable_lro(struct net_device *dev)
1193 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1194 dev->ethtool_ops->set_flags) {
1195 u32 flags = dev->ethtool_ops->get_flags(dev);
1196 if (flags & ETH_FLAG_LRO) {
1197 flags &= ~ETH_FLAG_LRO;
1198 dev->ethtool_ops->set_flags(dev, flags);
1201 WARN_ON(dev->features & NETIF_F_LRO);
1203 EXPORT_SYMBOL(dev_disable_lro);
1206 static int dev_boot_phase = 1;
1209 * Device change register/unregister. These are not inline or static
1210 * as we export them to the world.
1214 * register_netdevice_notifier - register a network notifier block
1217 * Register a notifier to be called when network device events occur.
1218 * The notifier passed is linked into the kernel structures and must
1219 * not be reused until it has been unregistered. A negative errno code
1220 * is returned on a failure.
1222 * When registered all registration and up events are replayed
1223 * to the new notifier to allow device to have a race free
1224 * view of the network device list.
1227 int register_netdevice_notifier(struct notifier_block *nb)
1229 struct net_device *dev;
1230 struct net_device *last;
1235 err = raw_notifier_chain_register(&netdev_chain, nb);
1241 for_each_netdev(net, dev) {
1242 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1243 err = notifier_to_errno(err);
1247 if (!(dev->flags & IFF_UP))
1250 nb->notifier_call(nb, NETDEV_UP, dev);
1261 for_each_netdev(net, dev) {
1265 if (dev->flags & IFF_UP) {
1266 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1267 nb->notifier_call(nb, NETDEV_DOWN, dev);
1269 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1273 raw_notifier_chain_unregister(&netdev_chain, nb);
1278 * unregister_netdevice_notifier - unregister a network notifier block
1281 * Unregister a notifier previously registered by
1282 * register_netdevice_notifier(). The notifier is unlinked into the
1283 * kernel structures and may then be reused. A negative errno code
1284 * is returned on a failure.
1287 int unregister_netdevice_notifier(struct notifier_block *nb)
1292 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1298 * call_netdevice_notifiers - call all network notifier blocks
1299 * @val: value passed unmodified to notifier function
1300 * @dev: net_device pointer passed unmodified to notifier function
1302 * Call all network notifier blocks. Parameters and return value
1303 * are as for raw_notifier_call_chain().
1306 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1308 return raw_notifier_call_chain(&netdev_chain, val, dev);
1311 /* When > 0 there are consumers of rx skb time stamps */
1312 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1314 void net_enable_timestamp(void)
1316 atomic_inc(&netstamp_needed);
1319 void net_disable_timestamp(void)
1321 atomic_dec(&netstamp_needed);
1324 static inline void net_timestamp(struct sk_buff *skb)
1326 if (atomic_read(&netstamp_needed))
1327 __net_timestamp(skb);
1329 skb->tstamp.tv64 = 0;
1333 * Support routine. Sends outgoing frames to any network
1334 * taps currently in use.
1337 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1339 struct packet_type *ptype;
1344 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1345 /* Never send packets back to the socket
1346 * they originated from - MvS (miquels@drinkel.ow.org)
1348 if ((ptype->dev == dev || !ptype->dev) &&
1349 (ptype->af_packet_priv == NULL ||
1350 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1351 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1355 /* skb->nh should be correctly
1356 set by sender, so that the second statement is
1357 just protection against buggy protocols.
1359 skb_reset_mac_header(skb2);
1361 if (skb_network_header(skb2) < skb2->data ||
1362 skb2->network_header > skb2->tail) {
1363 if (net_ratelimit())
1364 printk(KERN_CRIT "protocol %04x is "
1366 skb2->protocol, dev->name);
1367 skb_reset_network_header(skb2);
1370 skb2->transport_header = skb2->network_header;
1371 skb2->pkt_type = PACKET_OUTGOING;
1372 ptype->func(skb2, skb->dev, ptype, skb->dev);
1379 static inline void __netif_reschedule(struct Qdisc *q)
1381 struct softnet_data *sd;
1382 unsigned long flags;
1384 local_irq_save(flags);
1385 sd = &__get_cpu_var(softnet_data);
1386 q->next_sched = sd->output_queue;
1387 sd->output_queue = q;
1388 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1389 local_irq_restore(flags);
1392 void __netif_schedule(struct Qdisc *q)
1394 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1395 __netif_reschedule(q);
1397 EXPORT_SYMBOL(__netif_schedule);
1399 void dev_kfree_skb_irq(struct sk_buff *skb)
1401 if (atomic_dec_and_test(&skb->users)) {
1402 struct softnet_data *sd;
1403 unsigned long flags;
1405 local_irq_save(flags);
1406 sd = &__get_cpu_var(softnet_data);
1407 skb->next = sd->completion_queue;
1408 sd->completion_queue = skb;
1409 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1410 local_irq_restore(flags);
1413 EXPORT_SYMBOL(dev_kfree_skb_irq);
1415 void dev_kfree_skb_any(struct sk_buff *skb)
1417 if (in_irq() || irqs_disabled())
1418 dev_kfree_skb_irq(skb);
1422 EXPORT_SYMBOL(dev_kfree_skb_any);
1426 * netif_device_detach - mark device as removed
1427 * @dev: network device
1429 * Mark device as removed from system and therefore no longer available.
1431 void netif_device_detach(struct net_device *dev)
1433 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1434 netif_running(dev)) {
1435 netif_stop_queue(dev);
1438 EXPORT_SYMBOL(netif_device_detach);
1441 * netif_device_attach - mark device as attached
1442 * @dev: network device
1444 * Mark device as attached from system and restart if needed.
1446 void netif_device_attach(struct net_device *dev)
1448 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1449 netif_running(dev)) {
1450 netif_wake_queue(dev);
1451 __netdev_watchdog_up(dev);
1454 EXPORT_SYMBOL(netif_device_attach);
1456 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1458 return ((features & NETIF_F_GEN_CSUM) ||
1459 ((features & NETIF_F_IP_CSUM) &&
1460 protocol == htons(ETH_P_IP)) ||
1461 ((features & NETIF_F_IPV6_CSUM) &&
1462 protocol == htons(ETH_P_IPV6)));
1465 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1467 if (can_checksum_protocol(dev->features, skb->protocol))
1470 if (skb->protocol == htons(ETH_P_8021Q)) {
1471 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1472 if (can_checksum_protocol(dev->features & dev->vlan_features,
1473 veh->h_vlan_encapsulated_proto))
1481 * Invalidate hardware checksum when packet is to be mangled, and
1482 * complete checksum manually on outgoing path.
1484 int skb_checksum_help(struct sk_buff *skb)
1487 int ret = 0, offset;
1489 if (skb->ip_summed == CHECKSUM_COMPLETE)
1490 goto out_set_summed;
1492 if (unlikely(skb_shinfo(skb)->gso_size)) {
1493 /* Let GSO fix up the checksum. */
1494 goto out_set_summed;
1497 offset = skb->csum_start - skb_headroom(skb);
1498 BUG_ON(offset >= skb_headlen(skb));
1499 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1501 offset += skb->csum_offset;
1502 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1504 if (skb_cloned(skb) &&
1505 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1506 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1511 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1513 skb->ip_summed = CHECKSUM_NONE;
1519 * skb_gso_segment - Perform segmentation on skb.
1520 * @skb: buffer to segment
1521 * @features: features for the output path (see dev->features)
1523 * This function segments the given skb and returns a list of segments.
1525 * It may return NULL if the skb requires no segmentation. This is
1526 * only possible when GSO is used for verifying header integrity.
1528 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1530 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1531 struct packet_type *ptype;
1532 __be16 type = skb->protocol;
1535 BUG_ON(skb_shinfo(skb)->frag_list);
1537 skb_reset_mac_header(skb);
1538 skb->mac_len = skb->network_header - skb->mac_header;
1539 __skb_pull(skb, skb->mac_len);
1541 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1542 if (skb_header_cloned(skb) &&
1543 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1544 return ERR_PTR(err);
1548 list_for_each_entry_rcu(ptype,
1549 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1550 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1551 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1552 err = ptype->gso_send_check(skb);
1553 segs = ERR_PTR(err);
1554 if (err || skb_gso_ok(skb, features))
1556 __skb_push(skb, (skb->data -
1557 skb_network_header(skb)));
1559 segs = ptype->gso_segment(skb, features);
1565 __skb_push(skb, skb->data - skb_mac_header(skb));
1570 EXPORT_SYMBOL(skb_gso_segment);
1572 /* Take action when hardware reception checksum errors are detected. */
1574 void netdev_rx_csum_fault(struct net_device *dev)
1576 if (net_ratelimit()) {
1577 printk(KERN_ERR "%s: hw csum failure.\n",
1578 dev ? dev->name : "<unknown>");
1582 EXPORT_SYMBOL(netdev_rx_csum_fault);
1585 /* Actually, we should eliminate this check as soon as we know, that:
1586 * 1. IOMMU is present and allows to map all the memory.
1587 * 2. No high memory really exists on this machine.
1590 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1592 #ifdef CONFIG_HIGHMEM
1595 if (dev->features & NETIF_F_HIGHDMA)
1598 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1599 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1607 void (*destructor)(struct sk_buff *skb);
1610 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1612 static void dev_gso_skb_destructor(struct sk_buff *skb)
1614 struct dev_gso_cb *cb;
1617 struct sk_buff *nskb = skb->next;
1619 skb->next = nskb->next;
1622 } while (skb->next);
1624 cb = DEV_GSO_CB(skb);
1626 cb->destructor(skb);
1630 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1631 * @skb: buffer to segment
1633 * This function segments the given skb and stores the list of segments
1636 static int dev_gso_segment(struct sk_buff *skb)
1638 struct net_device *dev = skb->dev;
1639 struct sk_buff *segs;
1640 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1643 segs = skb_gso_segment(skb, features);
1645 /* Verifying header integrity only. */
1650 return PTR_ERR(segs);
1653 DEV_GSO_CB(skb)->destructor = skb->destructor;
1654 skb->destructor = dev_gso_skb_destructor;
1659 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1660 struct netdev_queue *txq)
1662 if (likely(!skb->next)) {
1663 if (!list_empty(&ptype_all))
1664 dev_queue_xmit_nit(skb, dev);
1666 if (netif_needs_gso(dev, skb)) {
1667 if (unlikely(dev_gso_segment(skb)))
1673 return dev->hard_start_xmit(skb, dev);
1678 struct sk_buff *nskb = skb->next;
1681 skb->next = nskb->next;
1683 rc = dev->hard_start_xmit(nskb, dev);
1685 nskb->next = skb->next;
1689 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1690 return NETDEV_TX_BUSY;
1691 } while (skb->next);
1693 skb->destructor = DEV_GSO_CB(skb)->destructor;
1700 static u32 simple_tx_hashrnd;
1701 static int simple_tx_hashrnd_initialized = 0;
1703 static u16 simple_tx_hash(struct net_device *dev, struct sk_buff *skb)
1705 u32 addr1, addr2, ports;
1709 if (unlikely(!simple_tx_hashrnd_initialized)) {
1710 get_random_bytes(&simple_tx_hashrnd, 4);
1711 simple_tx_hashrnd_initialized = 1;
1714 switch (skb->protocol) {
1715 case htons(ETH_P_IP):
1716 if (!(ip_hdr(skb)->frag_off & htons(IP_MF | IP_OFFSET)))
1717 ip_proto = ip_hdr(skb)->protocol;
1718 addr1 = ip_hdr(skb)->saddr;
1719 addr2 = ip_hdr(skb)->daddr;
1720 ihl = ip_hdr(skb)->ihl;
1722 case htons(ETH_P_IPV6):
1723 ip_proto = ipv6_hdr(skb)->nexthdr;
1724 addr1 = ipv6_hdr(skb)->saddr.s6_addr32[3];
1725 addr2 = ipv6_hdr(skb)->daddr.s6_addr32[3];
1740 case IPPROTO_UDPLITE:
1741 ports = *((u32 *) (skb_network_header(skb) + (ihl * 4)));
1749 hash = jhash_3words(addr1, addr2, ports, simple_tx_hashrnd);
1751 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1754 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1755 struct sk_buff *skb)
1757 u16 queue_index = 0;
1759 if (dev->select_queue)
1760 queue_index = dev->select_queue(dev, skb);
1761 else if (dev->real_num_tx_queues > 1)
1762 queue_index = simple_tx_hash(dev, skb);
1764 skb_set_queue_mapping(skb, queue_index);
1765 return netdev_get_tx_queue(dev, queue_index);
1769 * dev_queue_xmit - transmit a buffer
1770 * @skb: buffer to transmit
1772 * Queue a buffer for transmission to a network device. The caller must
1773 * have set the device and priority and built the buffer before calling
1774 * this function. The function can be called from an interrupt.
1776 * A negative errno code is returned on a failure. A success does not
1777 * guarantee the frame will be transmitted as it may be dropped due
1778 * to congestion or traffic shaping.
1780 * -----------------------------------------------------------------------------------
1781 * I notice this method can also return errors from the queue disciplines,
1782 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1785 * Regardless of the return value, the skb is consumed, so it is currently
1786 * difficult to retry a send to this method. (You can bump the ref count
1787 * before sending to hold a reference for retry if you are careful.)
1789 * When calling this method, interrupts MUST be enabled. This is because
1790 * the BH enable code must have IRQs enabled so that it will not deadlock.
1793 int dev_queue_xmit(struct sk_buff *skb)
1795 struct net_device *dev = skb->dev;
1796 struct netdev_queue *txq;
1800 /* GSO will handle the following emulations directly. */
1801 if (netif_needs_gso(dev, skb))
1804 if (skb_shinfo(skb)->frag_list &&
1805 !(dev->features & NETIF_F_FRAGLIST) &&
1806 __skb_linearize(skb))
1809 /* Fragmented skb is linearized if device does not support SG,
1810 * or if at least one of fragments is in highmem and device
1811 * does not support DMA from it.
1813 if (skb_shinfo(skb)->nr_frags &&
1814 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1815 __skb_linearize(skb))
1818 /* If packet is not checksummed and device does not support
1819 * checksumming for this protocol, complete checksumming here.
1821 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1822 skb_set_transport_header(skb, skb->csum_start -
1824 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1829 /* Disable soft irqs for various locks below. Also
1830 * stops preemption for RCU.
1834 txq = dev_pick_tx(dev, skb);
1835 q = rcu_dereference(txq->qdisc);
1837 #ifdef CONFIG_NET_CLS_ACT
1838 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1841 spinlock_t *root_lock = qdisc_lock(q);
1843 spin_lock(root_lock);
1845 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1849 rc = qdisc_enqueue_root(skb, q);
1852 spin_unlock(root_lock);
1857 /* The device has no queue. Common case for software devices:
1858 loopback, all the sorts of tunnels...
1860 Really, it is unlikely that netif_tx_lock protection is necessary
1861 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1863 However, it is possible, that they rely on protection
1866 Check this and shot the lock. It is not prone from deadlocks.
1867 Either shot noqueue qdisc, it is even simpler 8)
1869 if (dev->flags & IFF_UP) {
1870 int cpu = smp_processor_id(); /* ok because BHs are off */
1872 if (txq->xmit_lock_owner != cpu) {
1874 HARD_TX_LOCK(dev, txq, cpu);
1876 if (!netif_tx_queue_stopped(txq)) {
1878 if (!dev_hard_start_xmit(skb, dev, txq)) {
1879 HARD_TX_UNLOCK(dev, txq);
1883 HARD_TX_UNLOCK(dev, txq);
1884 if (net_ratelimit())
1885 printk(KERN_CRIT "Virtual device %s asks to "
1886 "queue packet!\n", dev->name);
1888 /* Recursion is detected! It is possible,
1890 if (net_ratelimit())
1891 printk(KERN_CRIT "Dead loop on virtual device "
1892 "%s, fix it urgently!\n", dev->name);
1897 rcu_read_unlock_bh();
1903 rcu_read_unlock_bh();
1908 /*=======================================================================
1910 =======================================================================*/
1912 int netdev_max_backlog __read_mostly = 1000;
1913 int netdev_budget __read_mostly = 300;
1914 int weight_p __read_mostly = 64; /* old backlog weight */
1916 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1920 * netif_rx - post buffer to the network code
1921 * @skb: buffer to post
1923 * This function receives a packet from a device driver and queues it for
1924 * the upper (protocol) levels to process. It always succeeds. The buffer
1925 * may be dropped during processing for congestion control or by the
1929 * NET_RX_SUCCESS (no congestion)
1930 * NET_RX_DROP (packet was dropped)
1934 int netif_rx(struct sk_buff *skb)
1936 struct softnet_data *queue;
1937 unsigned long flags;
1939 /* if netpoll wants it, pretend we never saw it */
1940 if (netpoll_rx(skb))
1943 if (!skb->tstamp.tv64)
1947 * The code is rearranged so that the path is the most
1948 * short when CPU is congested, but is still operating.
1950 local_irq_save(flags);
1951 queue = &__get_cpu_var(softnet_data);
1953 __get_cpu_var(netdev_rx_stat).total++;
1954 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1955 if (queue->input_pkt_queue.qlen) {
1957 __skb_queue_tail(&queue->input_pkt_queue, skb);
1958 local_irq_restore(flags);
1959 return NET_RX_SUCCESS;
1962 napi_schedule(&queue->backlog);
1966 __get_cpu_var(netdev_rx_stat).dropped++;
1967 local_irq_restore(flags);
1973 int netif_rx_ni(struct sk_buff *skb)
1978 err = netif_rx(skb);
1979 if (local_softirq_pending())
1986 EXPORT_SYMBOL(netif_rx_ni);
1988 static void net_tx_action(struct softirq_action *h)
1990 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1992 if (sd->completion_queue) {
1993 struct sk_buff *clist;
1995 local_irq_disable();
1996 clist = sd->completion_queue;
1997 sd->completion_queue = NULL;
2001 struct sk_buff *skb = clist;
2002 clist = clist->next;
2004 WARN_ON(atomic_read(&skb->users));
2009 if (sd->output_queue) {
2012 local_irq_disable();
2013 head = sd->output_queue;
2014 sd->output_queue = NULL;
2018 struct Qdisc *q = head;
2019 spinlock_t *root_lock;
2021 head = head->next_sched;
2023 root_lock = qdisc_lock(q);
2024 if (spin_trylock(root_lock)) {
2025 smp_mb__before_clear_bit();
2026 clear_bit(__QDISC_STATE_SCHED,
2029 spin_unlock(root_lock);
2031 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2033 __netif_reschedule(q);
2035 smp_mb__before_clear_bit();
2036 clear_bit(__QDISC_STATE_SCHED,
2044 static inline int deliver_skb(struct sk_buff *skb,
2045 struct packet_type *pt_prev,
2046 struct net_device *orig_dev)
2048 atomic_inc(&skb->users);
2049 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2052 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2053 /* These hooks defined here for ATM */
2055 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2056 unsigned char *addr);
2057 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2060 * If bridge module is loaded call bridging hook.
2061 * returns NULL if packet was consumed.
2063 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2064 struct sk_buff *skb) __read_mostly;
2065 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2066 struct packet_type **pt_prev, int *ret,
2067 struct net_device *orig_dev)
2069 struct net_bridge_port *port;
2071 if (skb->pkt_type == PACKET_LOOPBACK ||
2072 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2076 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2080 return br_handle_frame_hook(port, skb);
2083 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2086 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2087 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2088 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2090 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2091 struct packet_type **pt_prev,
2093 struct net_device *orig_dev)
2095 if (skb->dev->macvlan_port == NULL)
2099 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2102 return macvlan_handle_frame_hook(skb);
2105 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2108 #ifdef CONFIG_NET_CLS_ACT
2109 /* TODO: Maybe we should just force sch_ingress to be compiled in
2110 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2111 * a compare and 2 stores extra right now if we dont have it on
2112 * but have CONFIG_NET_CLS_ACT
2113 * NOTE: This doesnt stop any functionality; if you dont have
2114 * the ingress scheduler, you just cant add policies on ingress.
2117 static int ing_filter(struct sk_buff *skb)
2119 struct net_device *dev = skb->dev;
2120 u32 ttl = G_TC_RTTL(skb->tc_verd);
2121 struct netdev_queue *rxq;
2122 int result = TC_ACT_OK;
2125 if (MAX_RED_LOOP < ttl++) {
2127 "Redir loop detected Dropping packet (%d->%d)\n",
2128 skb->iif, dev->ifindex);
2132 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2133 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2135 rxq = &dev->rx_queue;
2138 if (q != &noop_qdisc) {
2139 spin_lock(qdisc_lock(q));
2140 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2141 result = qdisc_enqueue_root(skb, q);
2142 spin_unlock(qdisc_lock(q));
2148 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2149 struct packet_type **pt_prev,
2150 int *ret, struct net_device *orig_dev)
2152 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2156 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2159 /* Huh? Why does turning on AF_PACKET affect this? */
2160 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2163 switch (ing_filter(skb)) {
2177 * netif_nit_deliver - deliver received packets to network taps
2180 * This function is used to deliver incoming packets to network
2181 * taps. It should be used when the normal netif_receive_skb path
2182 * is bypassed, for example because of VLAN acceleration.
2184 void netif_nit_deliver(struct sk_buff *skb)
2186 struct packet_type *ptype;
2188 if (list_empty(&ptype_all))
2191 skb_reset_network_header(skb);
2192 skb_reset_transport_header(skb);
2193 skb->mac_len = skb->network_header - skb->mac_header;
2196 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2197 if (!ptype->dev || ptype->dev == skb->dev)
2198 deliver_skb(skb, ptype, skb->dev);
2204 * netif_receive_skb - process receive buffer from network
2205 * @skb: buffer to process
2207 * netif_receive_skb() is the main receive data processing function.
2208 * It always succeeds. The buffer may be dropped during processing
2209 * for congestion control or by the protocol layers.
2211 * This function may only be called from softirq context and interrupts
2212 * should be enabled.
2214 * Return values (usually ignored):
2215 * NET_RX_SUCCESS: no congestion
2216 * NET_RX_DROP: packet was dropped
2218 int netif_receive_skb(struct sk_buff *skb)
2220 struct packet_type *ptype, *pt_prev;
2221 struct net_device *orig_dev;
2222 struct net_device *null_or_orig;
2223 int ret = NET_RX_DROP;
2226 /* if we've gotten here through NAPI, check netpoll */
2227 if (netpoll_receive_skb(skb))
2230 if (!skb->tstamp.tv64)
2234 skb->iif = skb->dev->ifindex;
2236 null_or_orig = NULL;
2237 orig_dev = skb->dev;
2238 if (orig_dev->master) {
2239 if (skb_bond_should_drop(skb))
2240 null_or_orig = orig_dev; /* deliver only exact match */
2242 skb->dev = orig_dev->master;
2245 __get_cpu_var(netdev_rx_stat).total++;
2247 skb_reset_network_header(skb);
2248 skb_reset_transport_header(skb);
2249 skb->mac_len = skb->network_header - skb->mac_header;
2255 /* Don't receive packets in an exiting network namespace */
2256 if (!net_alive(dev_net(skb->dev))) {
2261 #ifdef CONFIG_NET_CLS_ACT
2262 if (skb->tc_verd & TC_NCLS) {
2263 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2268 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2269 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2270 ptype->dev == orig_dev) {
2272 ret = deliver_skb(skb, pt_prev, orig_dev);
2277 #ifdef CONFIG_NET_CLS_ACT
2278 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2284 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2287 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2291 type = skb->protocol;
2292 list_for_each_entry_rcu(ptype,
2293 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2294 if (ptype->type == type &&
2295 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2296 ptype->dev == orig_dev)) {
2298 ret = deliver_skb(skb, pt_prev, orig_dev);
2304 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2307 /* Jamal, now you will not able to escape explaining
2308 * me how you were going to use this. :-)
2318 /* Network device is going away, flush any packets still pending */
2319 static void flush_backlog(void *arg)
2321 struct net_device *dev = arg;
2322 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2323 struct sk_buff *skb, *tmp;
2325 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2326 if (skb->dev == dev) {
2327 __skb_unlink(skb, &queue->input_pkt_queue);
2332 static int process_backlog(struct napi_struct *napi, int quota)
2335 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2336 unsigned long start_time = jiffies;
2338 napi->weight = weight_p;
2340 struct sk_buff *skb;
2342 local_irq_disable();
2343 skb = __skb_dequeue(&queue->input_pkt_queue);
2345 __napi_complete(napi);
2351 netif_receive_skb(skb);
2352 } while (++work < quota && jiffies == start_time);
2358 * __napi_schedule - schedule for receive
2359 * @n: entry to schedule
2361 * The entry's receive function will be scheduled to run
2363 void __napi_schedule(struct napi_struct *n)
2365 unsigned long flags;
2367 local_irq_save(flags);
2368 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2369 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2370 local_irq_restore(flags);
2372 EXPORT_SYMBOL(__napi_schedule);
2375 static void net_rx_action(struct softirq_action *h)
2377 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2378 unsigned long time_limit = jiffies + 2;
2379 int budget = netdev_budget;
2382 local_irq_disable();
2384 while (!list_empty(list)) {
2385 struct napi_struct *n;
2388 /* If softirq window is exhuasted then punt.
2389 * Allow this to run for 2 jiffies since which will allow
2390 * an average latency of 1.5/HZ.
2392 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2397 /* Even though interrupts have been re-enabled, this
2398 * access is safe because interrupts can only add new
2399 * entries to the tail of this list, and only ->poll()
2400 * calls can remove this head entry from the list.
2402 n = list_entry(list->next, struct napi_struct, poll_list);
2404 have = netpoll_poll_lock(n);
2408 /* This NAPI_STATE_SCHED test is for avoiding a race
2409 * with netpoll's poll_napi(). Only the entity which
2410 * obtains the lock and sees NAPI_STATE_SCHED set will
2411 * actually make the ->poll() call. Therefore we avoid
2412 * accidently calling ->poll() when NAPI is not scheduled.
2415 if (test_bit(NAPI_STATE_SCHED, &n->state))
2416 work = n->poll(n, weight);
2418 WARN_ON_ONCE(work > weight);
2422 local_irq_disable();
2424 /* Drivers must not modify the NAPI state if they
2425 * consume the entire weight. In such cases this code
2426 * still "owns" the NAPI instance and therefore can
2427 * move the instance around on the list at-will.
2429 if (unlikely(work == weight)) {
2430 if (unlikely(napi_disable_pending(n)))
2433 list_move_tail(&n->poll_list, list);
2436 netpoll_poll_unlock(have);
2441 #ifdef CONFIG_NET_DMA
2443 * There may not be any more sk_buffs coming right now, so push
2444 * any pending DMA copies to hardware
2446 if (!cpus_empty(net_dma.channel_mask)) {
2448 for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) {
2449 struct dma_chan *chan = net_dma.channels[chan_idx];
2451 dma_async_memcpy_issue_pending(chan);
2459 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2460 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2464 static gifconf_func_t * gifconf_list [NPROTO];
2467 * register_gifconf - register a SIOCGIF handler
2468 * @family: Address family
2469 * @gifconf: Function handler
2471 * Register protocol dependent address dumping routines. The handler
2472 * that is passed must not be freed or reused until it has been replaced
2473 * by another handler.
2475 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2477 if (family >= NPROTO)
2479 gifconf_list[family] = gifconf;
2485 * Map an interface index to its name (SIOCGIFNAME)
2489 * We need this ioctl for efficient implementation of the
2490 * if_indextoname() function required by the IPv6 API. Without
2491 * it, we would have to search all the interfaces to find a
2495 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2497 struct net_device *dev;
2501 * Fetch the caller's info block.
2504 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2507 read_lock(&dev_base_lock);
2508 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2510 read_unlock(&dev_base_lock);
2514 strcpy(ifr.ifr_name, dev->name);
2515 read_unlock(&dev_base_lock);
2517 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2523 * Perform a SIOCGIFCONF call. This structure will change
2524 * size eventually, and there is nothing I can do about it.
2525 * Thus we will need a 'compatibility mode'.
2528 static int dev_ifconf(struct net *net, char __user *arg)
2531 struct net_device *dev;
2538 * Fetch the caller's info block.
2541 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2548 * Loop over the interfaces, and write an info block for each.
2552 for_each_netdev(net, dev) {
2553 for (i = 0; i < NPROTO; i++) {
2554 if (gifconf_list[i]) {
2557 done = gifconf_list[i](dev, NULL, 0);
2559 done = gifconf_list[i](dev, pos + total,
2569 * All done. Write the updated control block back to the caller.
2571 ifc.ifc_len = total;
2574 * Both BSD and Solaris return 0 here, so we do too.
2576 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2579 #ifdef CONFIG_PROC_FS
2581 * This is invoked by the /proc filesystem handler to display a device
2584 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2585 __acquires(dev_base_lock)
2587 struct net *net = seq_file_net(seq);
2589 struct net_device *dev;
2591 read_lock(&dev_base_lock);
2593 return SEQ_START_TOKEN;
2596 for_each_netdev(net, dev)
2603 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2605 struct net *net = seq_file_net(seq);
2607 return v == SEQ_START_TOKEN ?
2608 first_net_device(net) : next_net_device((struct net_device *)v);
2611 void dev_seq_stop(struct seq_file *seq, void *v)
2612 __releases(dev_base_lock)
2614 read_unlock(&dev_base_lock);
2617 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2619 struct net_device_stats *stats = dev->get_stats(dev);
2621 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2622 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2623 dev->name, stats->rx_bytes, stats->rx_packets,
2625 stats->rx_dropped + stats->rx_missed_errors,
2626 stats->rx_fifo_errors,
2627 stats->rx_length_errors + stats->rx_over_errors +
2628 stats->rx_crc_errors + stats->rx_frame_errors,
2629 stats->rx_compressed, stats->multicast,
2630 stats->tx_bytes, stats->tx_packets,
2631 stats->tx_errors, stats->tx_dropped,
2632 stats->tx_fifo_errors, stats->collisions,
2633 stats->tx_carrier_errors +
2634 stats->tx_aborted_errors +
2635 stats->tx_window_errors +
2636 stats->tx_heartbeat_errors,
2637 stats->tx_compressed);
2641 * Called from the PROCfs module. This now uses the new arbitrary sized
2642 * /proc/net interface to create /proc/net/dev
2644 static int dev_seq_show(struct seq_file *seq, void *v)
2646 if (v == SEQ_START_TOKEN)
2647 seq_puts(seq, "Inter-| Receive "
2649 " face |bytes packets errs drop fifo frame "
2650 "compressed multicast|bytes packets errs "
2651 "drop fifo colls carrier compressed\n");
2653 dev_seq_printf_stats(seq, v);
2657 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2659 struct netif_rx_stats *rc = NULL;
2661 while (*pos < nr_cpu_ids)
2662 if (cpu_online(*pos)) {
2663 rc = &per_cpu(netdev_rx_stat, *pos);
2670 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2672 return softnet_get_online(pos);
2675 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2678 return softnet_get_online(pos);
2681 static void softnet_seq_stop(struct seq_file *seq, void *v)
2685 static int softnet_seq_show(struct seq_file *seq, void *v)
2687 struct netif_rx_stats *s = v;
2689 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2690 s->total, s->dropped, s->time_squeeze, 0,
2691 0, 0, 0, 0, /* was fastroute */
2696 static const struct seq_operations dev_seq_ops = {
2697 .start = dev_seq_start,
2698 .next = dev_seq_next,
2699 .stop = dev_seq_stop,
2700 .show = dev_seq_show,
2703 static int dev_seq_open(struct inode *inode, struct file *file)
2705 return seq_open_net(inode, file, &dev_seq_ops,
2706 sizeof(struct seq_net_private));
2709 static const struct file_operations dev_seq_fops = {
2710 .owner = THIS_MODULE,
2711 .open = dev_seq_open,
2713 .llseek = seq_lseek,
2714 .release = seq_release_net,
2717 static const struct seq_operations softnet_seq_ops = {
2718 .start = softnet_seq_start,
2719 .next = softnet_seq_next,
2720 .stop = softnet_seq_stop,
2721 .show = softnet_seq_show,
2724 static int softnet_seq_open(struct inode *inode, struct file *file)
2726 return seq_open(file, &softnet_seq_ops);
2729 static const struct file_operations softnet_seq_fops = {
2730 .owner = THIS_MODULE,
2731 .open = softnet_seq_open,
2733 .llseek = seq_lseek,
2734 .release = seq_release,
2737 static void *ptype_get_idx(loff_t pos)
2739 struct packet_type *pt = NULL;
2743 list_for_each_entry_rcu(pt, &ptype_all, list) {
2749 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2750 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2759 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2763 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2766 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2768 struct packet_type *pt;
2769 struct list_head *nxt;
2773 if (v == SEQ_START_TOKEN)
2774 return ptype_get_idx(0);
2777 nxt = pt->list.next;
2778 if (pt->type == htons(ETH_P_ALL)) {
2779 if (nxt != &ptype_all)
2782 nxt = ptype_base[0].next;
2784 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2786 while (nxt == &ptype_base[hash]) {
2787 if (++hash >= PTYPE_HASH_SIZE)
2789 nxt = ptype_base[hash].next;
2792 return list_entry(nxt, struct packet_type, list);
2795 static void ptype_seq_stop(struct seq_file *seq, void *v)
2801 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2803 #ifdef CONFIG_KALLSYMS
2804 unsigned long offset = 0, symsize;
2805 const char *symname;
2809 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2816 modname = delim = "";
2817 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2823 seq_printf(seq, "[%p]", sym);
2826 static int ptype_seq_show(struct seq_file *seq, void *v)
2828 struct packet_type *pt = v;
2830 if (v == SEQ_START_TOKEN)
2831 seq_puts(seq, "Type Device Function\n");
2832 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2833 if (pt->type == htons(ETH_P_ALL))
2834 seq_puts(seq, "ALL ");
2836 seq_printf(seq, "%04x", ntohs(pt->type));
2838 seq_printf(seq, " %-8s ",
2839 pt->dev ? pt->dev->name : "");
2840 ptype_seq_decode(seq, pt->func);
2841 seq_putc(seq, '\n');
2847 static const struct seq_operations ptype_seq_ops = {
2848 .start = ptype_seq_start,
2849 .next = ptype_seq_next,
2850 .stop = ptype_seq_stop,
2851 .show = ptype_seq_show,
2854 static int ptype_seq_open(struct inode *inode, struct file *file)
2856 return seq_open_net(inode, file, &ptype_seq_ops,
2857 sizeof(struct seq_net_private));
2860 static const struct file_operations ptype_seq_fops = {
2861 .owner = THIS_MODULE,
2862 .open = ptype_seq_open,
2864 .llseek = seq_lseek,
2865 .release = seq_release_net,
2869 static int __net_init dev_proc_net_init(struct net *net)
2873 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2875 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2877 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2880 if (wext_proc_init(net))
2886 proc_net_remove(net, "ptype");
2888 proc_net_remove(net, "softnet_stat");
2890 proc_net_remove(net, "dev");
2894 static void __net_exit dev_proc_net_exit(struct net *net)
2896 wext_proc_exit(net);
2898 proc_net_remove(net, "ptype");
2899 proc_net_remove(net, "softnet_stat");
2900 proc_net_remove(net, "dev");
2903 static struct pernet_operations __net_initdata dev_proc_ops = {
2904 .init = dev_proc_net_init,
2905 .exit = dev_proc_net_exit,
2908 static int __init dev_proc_init(void)
2910 return register_pernet_subsys(&dev_proc_ops);
2913 #define dev_proc_init() 0
2914 #endif /* CONFIG_PROC_FS */
2918 * netdev_set_master - set up master/slave pair
2919 * @slave: slave device
2920 * @master: new master device
2922 * Changes the master device of the slave. Pass %NULL to break the
2923 * bonding. The caller must hold the RTNL semaphore. On a failure
2924 * a negative errno code is returned. On success the reference counts
2925 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2926 * function returns zero.
2928 int netdev_set_master(struct net_device *slave, struct net_device *master)
2930 struct net_device *old = slave->master;
2940 slave->master = master;
2948 slave->flags |= IFF_SLAVE;
2950 slave->flags &= ~IFF_SLAVE;
2952 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2956 static void dev_change_rx_flags(struct net_device *dev, int flags)
2958 if (dev->flags & IFF_UP && dev->change_rx_flags)
2959 dev->change_rx_flags(dev, flags);
2962 static int __dev_set_promiscuity(struct net_device *dev, int inc)
2964 unsigned short old_flags = dev->flags;
2968 dev->flags |= IFF_PROMISC;
2969 dev->promiscuity += inc;
2970 if (dev->promiscuity == 0) {
2973 * If inc causes overflow, untouch promisc and return error.
2976 dev->flags &= ~IFF_PROMISC;
2978 dev->promiscuity -= inc;
2979 printk(KERN_WARNING "%s: promiscuity touches roof, "
2980 "set promiscuity failed, promiscuity feature "
2981 "of device might be broken.\n", dev->name);
2985 if (dev->flags != old_flags) {
2986 printk(KERN_INFO "device %s %s promiscuous mode\n",
2987 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2990 audit_log(current->audit_context, GFP_ATOMIC,
2991 AUDIT_ANOM_PROMISCUOUS,
2992 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2993 dev->name, (dev->flags & IFF_PROMISC),
2994 (old_flags & IFF_PROMISC),
2995 audit_get_loginuid(current),
2996 current->uid, current->gid,
2997 audit_get_sessionid(current));
2999 dev_change_rx_flags(dev, IFF_PROMISC);
3005 * dev_set_promiscuity - update promiscuity count on a device
3009 * Add or remove promiscuity from a device. While the count in the device
3010 * remains above zero the interface remains promiscuous. Once it hits zero
3011 * the device reverts back to normal filtering operation. A negative inc
3012 * value is used to drop promiscuity on the device.
3013 * Return 0 if successful or a negative errno code on error.
3015 int dev_set_promiscuity(struct net_device *dev, int inc)
3017 unsigned short old_flags = dev->flags;
3020 err = __dev_set_promiscuity(dev, inc);
3023 if (dev->flags != old_flags)
3024 dev_set_rx_mode(dev);
3029 * dev_set_allmulti - update allmulti count on a device
3033 * Add or remove reception of all multicast frames to a device. While the
3034 * count in the device remains above zero the interface remains listening
3035 * to all interfaces. Once it hits zero the device reverts back to normal
3036 * filtering operation. A negative @inc value is used to drop the counter
3037 * when releasing a resource needing all multicasts.
3038 * Return 0 if successful or a negative errno code on error.
3041 int dev_set_allmulti(struct net_device *dev, int inc)
3043 unsigned short old_flags = dev->flags;
3047 dev->flags |= IFF_ALLMULTI;
3048 dev->allmulti += inc;
3049 if (dev->allmulti == 0) {
3052 * If inc causes overflow, untouch allmulti and return error.
3055 dev->flags &= ~IFF_ALLMULTI;
3057 dev->allmulti -= inc;
3058 printk(KERN_WARNING "%s: allmulti touches roof, "
3059 "set allmulti failed, allmulti feature of "
3060 "device might be broken.\n", dev->name);
3064 if (dev->flags ^ old_flags) {
3065 dev_change_rx_flags(dev, IFF_ALLMULTI);
3066 dev_set_rx_mode(dev);
3072 * Upload unicast and multicast address lists to device and
3073 * configure RX filtering. When the device doesn't support unicast
3074 * filtering it is put in promiscuous mode while unicast addresses
3077 void __dev_set_rx_mode(struct net_device *dev)
3079 /* dev_open will call this function so the list will stay sane. */
3080 if (!(dev->flags&IFF_UP))
3083 if (!netif_device_present(dev))
3086 if (dev->set_rx_mode)
3087 dev->set_rx_mode(dev);
3089 /* Unicast addresses changes may only happen under the rtnl,
3090 * therefore calling __dev_set_promiscuity here is safe.
3092 if (dev->uc_count > 0 && !dev->uc_promisc) {
3093 __dev_set_promiscuity(dev, 1);
3094 dev->uc_promisc = 1;
3095 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3096 __dev_set_promiscuity(dev, -1);
3097 dev->uc_promisc = 0;
3100 if (dev->set_multicast_list)
3101 dev->set_multicast_list(dev);
3105 void dev_set_rx_mode(struct net_device *dev)
3107 netif_addr_lock_bh(dev);
3108 __dev_set_rx_mode(dev);
3109 netif_addr_unlock_bh(dev);
3112 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3113 void *addr, int alen, int glbl)
3115 struct dev_addr_list *da;
3117 for (; (da = *list) != NULL; list = &da->next) {
3118 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3119 alen == da->da_addrlen) {
3121 int old_glbl = da->da_gusers;
3138 int __dev_addr_add(struct dev_addr_list **list, int *count,
3139 void *addr, int alen, int glbl)
3141 struct dev_addr_list *da;
3143 for (da = *list; da != NULL; da = da->next) {
3144 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3145 da->da_addrlen == alen) {
3147 int old_glbl = da->da_gusers;
3157 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3160 memcpy(da->da_addr, addr, alen);
3161 da->da_addrlen = alen;
3163 da->da_gusers = glbl ? 1 : 0;
3171 * dev_unicast_delete - Release secondary unicast address.
3173 * @addr: address to delete
3174 * @alen: length of @addr
3176 * Release reference to a secondary unicast address and remove it
3177 * from the device if the reference count drops to zero.
3179 * The caller must hold the rtnl_mutex.
3181 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3187 netif_addr_lock_bh(dev);
3188 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3190 __dev_set_rx_mode(dev);
3191 netif_addr_unlock_bh(dev);
3194 EXPORT_SYMBOL(dev_unicast_delete);
3197 * dev_unicast_add - add a secondary unicast address
3199 * @addr: address to add
3200 * @alen: length of @addr
3202 * Add a secondary unicast address to the device or increase
3203 * the reference count if it already exists.
3205 * The caller must hold the rtnl_mutex.
3207 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3213 netif_addr_lock_bh(dev);
3214 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3216 __dev_set_rx_mode(dev);
3217 netif_addr_unlock_bh(dev);
3220 EXPORT_SYMBOL(dev_unicast_add);
3222 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3223 struct dev_addr_list **from, int *from_count)
3225 struct dev_addr_list *da, *next;
3229 while (da != NULL) {
3231 if (!da->da_synced) {
3232 err = __dev_addr_add(to, to_count,
3233 da->da_addr, da->da_addrlen, 0);
3238 } else if (da->da_users == 1) {
3239 __dev_addr_delete(to, to_count,
3240 da->da_addr, da->da_addrlen, 0);
3241 __dev_addr_delete(from, from_count,
3242 da->da_addr, da->da_addrlen, 0);
3249 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3250 struct dev_addr_list **from, int *from_count)
3252 struct dev_addr_list *da, *next;
3255 while (da != NULL) {
3257 if (da->da_synced) {
3258 __dev_addr_delete(to, to_count,
3259 da->da_addr, da->da_addrlen, 0);
3261 __dev_addr_delete(from, from_count,
3262 da->da_addr, da->da_addrlen, 0);
3269 * dev_unicast_sync - Synchronize device's unicast list to another device
3270 * @to: destination device
3271 * @from: source device
3273 * Add newly added addresses to the destination device and release
3274 * addresses that have no users left. The source device must be
3275 * locked by netif_tx_lock_bh.
3277 * This function is intended to be called from the dev->set_rx_mode
3278 * function of layered software devices.
3280 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3284 netif_addr_lock_bh(to);
3285 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3286 &from->uc_list, &from->uc_count);
3288 __dev_set_rx_mode(to);
3289 netif_addr_unlock_bh(to);
3292 EXPORT_SYMBOL(dev_unicast_sync);
3295 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3296 * @to: destination device
3297 * @from: source device
3299 * Remove all addresses that were added to the destination device by
3300 * dev_unicast_sync(). This function is intended to be called from the
3301 * dev->stop function of layered software devices.
3303 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3305 netif_addr_lock_bh(from);
3306 netif_addr_lock(to);
3308 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3309 &from->uc_list, &from->uc_count);
3310 __dev_set_rx_mode(to);
3312 netif_addr_unlock(to);
3313 netif_addr_unlock_bh(from);
3315 EXPORT_SYMBOL(dev_unicast_unsync);
3317 static void __dev_addr_discard(struct dev_addr_list **list)
3319 struct dev_addr_list *tmp;
3321 while (*list != NULL) {
3324 if (tmp->da_users > tmp->da_gusers)
3325 printk("__dev_addr_discard: address leakage! "
3326 "da_users=%d\n", tmp->da_users);
3331 static void dev_addr_discard(struct net_device *dev)
3333 netif_addr_lock_bh(dev);
3335 __dev_addr_discard(&dev->uc_list);
3338 __dev_addr_discard(&dev->mc_list);
3341 netif_addr_unlock_bh(dev);
3345 * dev_get_flags - get flags reported to userspace
3348 * Get the combination of flag bits exported through APIs to userspace.
3350 unsigned dev_get_flags(const struct net_device *dev)
3354 flags = (dev->flags & ~(IFF_PROMISC |
3359 (dev->gflags & (IFF_PROMISC |
3362 if (netif_running(dev)) {
3363 if (netif_oper_up(dev))
3364 flags |= IFF_RUNNING;
3365 if (netif_carrier_ok(dev))
3366 flags |= IFF_LOWER_UP;
3367 if (netif_dormant(dev))
3368 flags |= IFF_DORMANT;
3375 * dev_change_flags - change device settings
3377 * @flags: device state flags
3379 * Change settings on device based state flags. The flags are
3380 * in the userspace exported format.
3382 int dev_change_flags(struct net_device *dev, unsigned flags)
3385 int old_flags = dev->flags;
3390 * Set the flags on our device.
3393 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3394 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3396 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3400 * Load in the correct multicast list now the flags have changed.
3403 if ((old_flags ^ flags) & IFF_MULTICAST)
3404 dev_change_rx_flags(dev, IFF_MULTICAST);
3406 dev_set_rx_mode(dev);
3409 * Have we downed the interface. We handle IFF_UP ourselves
3410 * according to user attempts to set it, rather than blindly
3415 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3416 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3419 dev_set_rx_mode(dev);
3422 if (dev->flags & IFF_UP &&
3423 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3425 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3427 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3428 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3429 dev->gflags ^= IFF_PROMISC;
3430 dev_set_promiscuity(dev, inc);
3433 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3434 is important. Some (broken) drivers set IFF_PROMISC, when
3435 IFF_ALLMULTI is requested not asking us and not reporting.
3437 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3438 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3439 dev->gflags ^= IFF_ALLMULTI;
3440 dev_set_allmulti(dev, inc);
3443 /* Exclude state transition flags, already notified */
3444 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3446 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3452 * dev_set_mtu - Change maximum transfer unit
3454 * @new_mtu: new transfer unit
3456 * Change the maximum transfer size of the network device.
3458 int dev_set_mtu(struct net_device *dev, int new_mtu)
3462 if (new_mtu == dev->mtu)
3465 /* MTU must be positive. */
3469 if (!netif_device_present(dev))
3473 if (dev->change_mtu)
3474 err = dev->change_mtu(dev, new_mtu);
3477 if (!err && dev->flags & IFF_UP)
3478 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3483 * dev_set_mac_address - Change Media Access Control Address
3487 * Change the hardware (MAC) address of the device
3489 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3493 if (!dev->set_mac_address)
3495 if (sa->sa_family != dev->type)
3497 if (!netif_device_present(dev))
3499 err = dev->set_mac_address(dev, sa);
3501 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3506 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3508 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3511 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3517 case SIOCGIFFLAGS: /* Get interface flags */
3518 ifr->ifr_flags = dev_get_flags(dev);
3521 case SIOCGIFMETRIC: /* Get the metric on the interface
3522 (currently unused) */
3523 ifr->ifr_metric = 0;
3526 case SIOCGIFMTU: /* Get the MTU of a device */
3527 ifr->ifr_mtu = dev->mtu;
3532 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3534 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3535 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3536 ifr->ifr_hwaddr.sa_family = dev->type;
3544 ifr->ifr_map.mem_start = dev->mem_start;
3545 ifr->ifr_map.mem_end = dev->mem_end;
3546 ifr->ifr_map.base_addr = dev->base_addr;
3547 ifr->ifr_map.irq = dev->irq;
3548 ifr->ifr_map.dma = dev->dma;
3549 ifr->ifr_map.port = dev->if_port;
3553 ifr->ifr_ifindex = dev->ifindex;
3557 ifr->ifr_qlen = dev->tx_queue_len;
3561 /* dev_ioctl() should ensure this case
3573 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3575 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3578 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3584 case SIOCSIFFLAGS: /* Set interface flags */
3585 return dev_change_flags(dev, ifr->ifr_flags);
3587 case SIOCSIFMETRIC: /* Set the metric on the interface
3588 (currently unused) */
3591 case SIOCSIFMTU: /* Set the MTU of a device */
3592 return dev_set_mtu(dev, ifr->ifr_mtu);
3595 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3597 case SIOCSIFHWBROADCAST:
3598 if (ifr->ifr_hwaddr.sa_family != dev->type)
3600 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3601 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3602 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3606 if (dev->set_config) {
3607 if (!netif_device_present(dev))
3609 return dev->set_config(dev, &ifr->ifr_map);
3614 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3615 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3617 if (!netif_device_present(dev))
3619 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3623 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3624 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3626 if (!netif_device_present(dev))
3628 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3632 if (ifr->ifr_qlen < 0)
3634 dev->tx_queue_len = ifr->ifr_qlen;
3638 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3639 return dev_change_name(dev, ifr->ifr_newname);
3642 * Unknown or private ioctl
3646 if ((cmd >= SIOCDEVPRIVATE &&
3647 cmd <= SIOCDEVPRIVATE + 15) ||
3648 cmd == SIOCBONDENSLAVE ||
3649 cmd == SIOCBONDRELEASE ||
3650 cmd == SIOCBONDSETHWADDR ||
3651 cmd == SIOCBONDSLAVEINFOQUERY ||
3652 cmd == SIOCBONDINFOQUERY ||
3653 cmd == SIOCBONDCHANGEACTIVE ||
3654 cmd == SIOCGMIIPHY ||
3655 cmd == SIOCGMIIREG ||
3656 cmd == SIOCSMIIREG ||
3657 cmd == SIOCBRADDIF ||
3658 cmd == SIOCBRDELIF ||
3659 cmd == SIOCWANDEV) {
3661 if (dev->do_ioctl) {
3662 if (netif_device_present(dev))
3663 err = dev->do_ioctl(dev, ifr,
3676 * This function handles all "interface"-type I/O control requests. The actual
3677 * 'doing' part of this is dev_ifsioc above.
3681 * dev_ioctl - network device ioctl
3682 * @net: the applicable net namespace
3683 * @cmd: command to issue
3684 * @arg: pointer to a struct ifreq in user space
3686 * Issue ioctl functions to devices. This is normally called by the
3687 * user space syscall interfaces but can sometimes be useful for
3688 * other purposes. The return value is the return from the syscall if
3689 * positive or a negative errno code on error.
3692 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3698 /* One special case: SIOCGIFCONF takes ifconf argument
3699 and requires shared lock, because it sleeps writing
3703 if (cmd == SIOCGIFCONF) {
3705 ret = dev_ifconf(net, (char __user *) arg);
3709 if (cmd == SIOCGIFNAME)
3710 return dev_ifname(net, (struct ifreq __user *)arg);
3712 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3715 ifr.ifr_name[IFNAMSIZ-1] = 0;
3717 colon = strchr(ifr.ifr_name, ':');
3722 * See which interface the caller is talking about.
3727 * These ioctl calls:
3728 * - can be done by all.
3729 * - atomic and do not require locking.
3740 dev_load(net, ifr.ifr_name);
3741 read_lock(&dev_base_lock);
3742 ret = dev_ifsioc_locked(net, &ifr, cmd);
3743 read_unlock(&dev_base_lock);
3747 if (copy_to_user(arg, &ifr,
3748 sizeof(struct ifreq)))
3754 dev_load(net, ifr.ifr_name);
3756 ret = dev_ethtool(net, &ifr);
3761 if (copy_to_user(arg, &ifr,
3762 sizeof(struct ifreq)))
3768 * These ioctl calls:
3769 * - require superuser power.
3770 * - require strict serialization.
3776 if (!capable(CAP_NET_ADMIN))
3778 dev_load(net, ifr.ifr_name);
3780 ret = dev_ifsioc(net, &ifr, cmd);
3785 if (copy_to_user(arg, &ifr,
3786 sizeof(struct ifreq)))
3792 * These ioctl calls:
3793 * - require superuser power.
3794 * - require strict serialization.
3795 * - do not return a value
3805 case SIOCSIFHWBROADCAST:
3808 case SIOCBONDENSLAVE:
3809 case SIOCBONDRELEASE:
3810 case SIOCBONDSETHWADDR:
3811 case SIOCBONDCHANGEACTIVE:
3814 if (!capable(CAP_NET_ADMIN))
3817 case SIOCBONDSLAVEINFOQUERY:
3818 case SIOCBONDINFOQUERY:
3819 dev_load(net, ifr.ifr_name);
3821 ret = dev_ifsioc(net, &ifr, cmd);
3826 /* Get the per device memory space. We can add this but
3827 * currently do not support it */
3829 /* Set the per device memory buffer space.
3830 * Not applicable in our case */
3835 * Unknown or private ioctl.
3838 if (cmd == SIOCWANDEV ||
3839 (cmd >= SIOCDEVPRIVATE &&
3840 cmd <= SIOCDEVPRIVATE + 15)) {
3841 dev_load(net, ifr.ifr_name);
3843 ret = dev_ifsioc(net, &ifr, cmd);
3845 if (!ret && copy_to_user(arg, &ifr,
3846 sizeof(struct ifreq)))
3850 /* Take care of Wireless Extensions */
3851 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3852 return wext_handle_ioctl(net, &ifr, cmd, arg);
3859 * dev_new_index - allocate an ifindex
3860 * @net: the applicable net namespace
3862 * Returns a suitable unique value for a new device interface
3863 * number. The caller must hold the rtnl semaphore or the
3864 * dev_base_lock to be sure it remains unique.
3866 static int dev_new_index(struct net *net)
3872 if (!__dev_get_by_index(net, ifindex))
3877 /* Delayed registration/unregisteration */
3878 static LIST_HEAD(net_todo_list);
3880 static void net_set_todo(struct net_device *dev)
3882 list_add_tail(&dev->todo_list, &net_todo_list);
3885 static void rollback_registered(struct net_device *dev)
3887 BUG_ON(dev_boot_phase);
3890 /* Some devices call without registering for initialization unwind. */
3891 if (dev->reg_state == NETREG_UNINITIALIZED) {
3892 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3893 "was registered\n", dev->name, dev);
3899 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3901 /* If device is running, close it first. */
3904 /* And unlink it from device chain. */
3905 unlist_netdevice(dev);
3907 dev->reg_state = NETREG_UNREGISTERING;
3911 /* Shutdown queueing discipline. */
3915 /* Notify protocols, that we are about to destroy
3916 this device. They should clean all the things.
3918 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3921 * Flush the unicast and multicast chains
3923 dev_addr_discard(dev);
3928 /* Notifier chain MUST detach us from master device. */
3929 WARN_ON(dev->master);
3931 /* Remove entries from kobject tree */
3932 netdev_unregister_kobject(dev);
3939 static void __netdev_init_queue_locks_one(struct net_device *dev,
3940 struct netdev_queue *dev_queue,
3943 spin_lock_init(&dev_queue->_xmit_lock);
3944 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
3945 dev_queue->xmit_lock_owner = -1;
3948 static void netdev_init_queue_locks(struct net_device *dev)
3950 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3951 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
3954 unsigned long netdev_fix_features(unsigned long features, const char *name)
3956 /* Fix illegal SG+CSUM combinations. */
3957 if ((features & NETIF_F_SG) &&
3958 !(features & NETIF_F_ALL_CSUM)) {
3960 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
3961 "checksum feature.\n", name);
3962 features &= ~NETIF_F_SG;
3965 /* TSO requires that SG is present as well. */
3966 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
3968 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
3969 "SG feature.\n", name);
3970 features &= ~NETIF_F_TSO;
3973 if (features & NETIF_F_UFO) {
3974 if (!(features & NETIF_F_GEN_CSUM)) {
3976 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3977 "since no NETIF_F_HW_CSUM feature.\n",
3979 features &= ~NETIF_F_UFO;
3982 if (!(features & NETIF_F_SG)) {
3984 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3985 "since no NETIF_F_SG feature.\n", name);
3986 features &= ~NETIF_F_UFO;
3992 EXPORT_SYMBOL(netdev_fix_features);
3995 * register_netdevice - register a network device
3996 * @dev: device to register
3998 * Take a completed network device structure and add it to the kernel
3999 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4000 * chain. 0 is returned on success. A negative errno code is returned
4001 * on a failure to set up the device, or if the name is a duplicate.
4003 * Callers must hold the rtnl semaphore. You may want
4004 * register_netdev() instead of this.
4007 * The locking appears insufficient to guarantee two parallel registers
4008 * will not get the same name.
4011 int register_netdevice(struct net_device *dev)
4013 struct hlist_head *head;
4014 struct hlist_node *p;
4018 BUG_ON(dev_boot_phase);
4023 /* When net_device's are persistent, this will be fatal. */
4024 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4025 BUG_ON(!dev_net(dev));
4028 spin_lock_init(&dev->addr_list_lock);
4029 netdev_set_addr_lockdep_class(dev);
4030 netdev_init_queue_locks(dev);
4034 /* Init, if this function is available */
4036 ret = dev->init(dev);
4044 if (!dev_valid_name(dev->name)) {
4049 dev->ifindex = dev_new_index(net);
4050 if (dev->iflink == -1)
4051 dev->iflink = dev->ifindex;
4053 /* Check for existence of name */
4054 head = dev_name_hash(net, dev->name);
4055 hlist_for_each(p, head) {
4056 struct net_device *d
4057 = hlist_entry(p, struct net_device, name_hlist);
4058 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4064 /* Fix illegal checksum combinations */
4065 if ((dev->features & NETIF_F_HW_CSUM) &&
4066 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4067 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4069 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4072 if ((dev->features & NETIF_F_NO_CSUM) &&
4073 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4074 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4076 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4079 dev->features = netdev_fix_features(dev->features, dev->name);
4081 /* Enable software GSO if SG is supported. */
4082 if (dev->features & NETIF_F_SG)
4083 dev->features |= NETIF_F_GSO;
4085 netdev_initialize_kobject(dev);
4086 ret = netdev_register_kobject(dev);
4089 dev->reg_state = NETREG_REGISTERED;
4092 * Default initial state at registry is that the
4093 * device is present.
4096 set_bit(__LINK_STATE_PRESENT, &dev->state);
4098 dev_init_scheduler(dev);
4100 list_netdevice(dev);
4102 /* Notify protocols, that a new device appeared. */
4103 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4104 ret = notifier_to_errno(ret);
4106 rollback_registered(dev);
4107 dev->reg_state = NETREG_UNREGISTERED;
4120 * register_netdev - register a network device
4121 * @dev: device to register
4123 * Take a completed network device structure and add it to the kernel
4124 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4125 * chain. 0 is returned on success. A negative errno code is returned
4126 * on a failure to set up the device, or if the name is a duplicate.
4128 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4129 * and expands the device name if you passed a format string to
4132 int register_netdev(struct net_device *dev)
4139 * If the name is a format string the caller wants us to do a
4142 if (strchr(dev->name, '%')) {
4143 err = dev_alloc_name(dev, dev->name);
4148 err = register_netdevice(dev);
4153 EXPORT_SYMBOL(register_netdev);
4156 * netdev_wait_allrefs - wait until all references are gone.
4158 * This is called when unregistering network devices.
4160 * Any protocol or device that holds a reference should register
4161 * for netdevice notification, and cleanup and put back the
4162 * reference if they receive an UNREGISTER event.
4163 * We can get stuck here if buggy protocols don't correctly
4166 static void netdev_wait_allrefs(struct net_device *dev)
4168 unsigned long rebroadcast_time, warning_time;
4170 rebroadcast_time = warning_time = jiffies;
4171 while (atomic_read(&dev->refcnt) != 0) {
4172 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4175 /* Rebroadcast unregister notification */
4176 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4178 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4180 /* We must not have linkwatch events
4181 * pending on unregister. If this
4182 * happens, we simply run the queue
4183 * unscheduled, resulting in a noop
4186 linkwatch_run_queue();
4191 rebroadcast_time = jiffies;
4196 if (time_after(jiffies, warning_time + 10 * HZ)) {
4197 printk(KERN_EMERG "unregister_netdevice: "
4198 "waiting for %s to become free. Usage "
4200 dev->name, atomic_read(&dev->refcnt));
4201 warning_time = jiffies;
4210 * register_netdevice(x1);
4211 * register_netdevice(x2);
4213 * unregister_netdevice(y1);
4214 * unregister_netdevice(y2);
4220 * We are invoked by rtnl_unlock().
4221 * This allows us to deal with problems:
4222 * 1) We can delete sysfs objects which invoke hotplug
4223 * without deadlocking with linkwatch via keventd.
4224 * 2) Since we run with the RTNL semaphore not held, we can sleep
4225 * safely in order to wait for the netdev refcnt to drop to zero.
4227 * We must not return until all unregister events added during
4228 * the interval the lock was held have been completed.
4230 void netdev_run_todo(void)
4232 struct list_head list;
4234 /* Snapshot list, allow later requests */
4235 list_replace_init(&net_todo_list, &list);
4239 while (!list_empty(&list)) {
4240 struct net_device *dev
4241 = list_entry(list.next, struct net_device, todo_list);
4242 list_del(&dev->todo_list);
4244 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4245 printk(KERN_ERR "network todo '%s' but state %d\n",
4246 dev->name, dev->reg_state);
4251 dev->reg_state = NETREG_UNREGISTERED;
4253 on_each_cpu(flush_backlog, dev, 1);
4255 netdev_wait_allrefs(dev);
4258 BUG_ON(atomic_read(&dev->refcnt));
4259 WARN_ON(dev->ip_ptr);
4260 WARN_ON(dev->ip6_ptr);
4261 WARN_ON(dev->dn_ptr);
4263 if (dev->destructor)
4264 dev->destructor(dev);
4266 /* Free network device */
4267 kobject_put(&dev->dev.kobj);
4271 static struct net_device_stats *internal_stats(struct net_device *dev)
4276 static void netdev_init_one_queue(struct net_device *dev,
4277 struct netdev_queue *queue,
4283 static void netdev_init_queues(struct net_device *dev)
4285 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4286 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4287 spin_lock_init(&dev->tx_global_lock);
4291 * alloc_netdev_mq - allocate network device
4292 * @sizeof_priv: size of private data to allocate space for
4293 * @name: device name format string
4294 * @setup: callback to initialize device
4295 * @queue_count: the number of subqueues to allocate
4297 * Allocates a struct net_device with private data area for driver use
4298 * and performs basic initialization. Also allocates subquue structs
4299 * for each queue on the device at the end of the netdevice.
4301 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4302 void (*setup)(struct net_device *), unsigned int queue_count)
4304 struct netdev_queue *tx;
4305 struct net_device *dev;
4309 BUG_ON(strlen(name) >= sizeof(dev->name));
4311 alloc_size = sizeof(struct net_device);
4313 /* ensure 32-byte alignment of private area */
4314 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4315 alloc_size += sizeof_priv;
4317 /* ensure 32-byte alignment of whole construct */
4318 alloc_size += NETDEV_ALIGN_CONST;
4320 p = kzalloc(alloc_size, GFP_KERNEL);
4322 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4326 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4328 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4334 dev = (struct net_device *)
4335 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4336 dev->padded = (char *)dev - (char *)p;
4337 dev_net_set(dev, &init_net);
4340 dev->num_tx_queues = queue_count;
4341 dev->real_num_tx_queues = queue_count;
4344 dev->priv = ((char *)dev +
4345 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
4346 & ~NETDEV_ALIGN_CONST));
4349 dev->gso_max_size = GSO_MAX_SIZE;
4351 netdev_init_queues(dev);
4353 dev->get_stats = internal_stats;
4354 netpoll_netdev_init(dev);
4356 strcpy(dev->name, name);
4359 EXPORT_SYMBOL(alloc_netdev_mq);
4362 * free_netdev - free network device
4365 * This function does the last stage of destroying an allocated device
4366 * interface. The reference to the device object is released.
4367 * If this is the last reference then it will be freed.
4369 void free_netdev(struct net_device *dev)
4371 release_net(dev_net(dev));
4375 /* Compatibility with error handling in drivers */
4376 if (dev->reg_state == NETREG_UNINITIALIZED) {
4377 kfree((char *)dev - dev->padded);
4381 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4382 dev->reg_state = NETREG_RELEASED;
4384 /* will free via device release */
4385 put_device(&dev->dev);
4389 * synchronize_net - Synchronize with packet receive processing
4391 * Wait for packets currently being received to be done.
4392 * Does not block later packets from starting.
4394 void synchronize_net(void)
4401 * unregister_netdevice - remove device from the kernel
4404 * This function shuts down a device interface and removes it
4405 * from the kernel tables.
4407 * Callers must hold the rtnl semaphore. You may want
4408 * unregister_netdev() instead of this.
4411 void unregister_netdevice(struct net_device *dev)
4415 rollback_registered(dev);
4416 /* Finish processing unregister after unlock */
4421 * unregister_netdev - remove device from the kernel
4424 * This function shuts down a device interface and removes it
4425 * from the kernel tables.
4427 * This is just a wrapper for unregister_netdevice that takes
4428 * the rtnl semaphore. In general you want to use this and not
4429 * unregister_netdevice.
4431 void unregister_netdev(struct net_device *dev)
4434 unregister_netdevice(dev);
4438 EXPORT_SYMBOL(unregister_netdev);
4441 * dev_change_net_namespace - move device to different nethost namespace
4443 * @net: network namespace
4444 * @pat: If not NULL name pattern to try if the current device name
4445 * is already taken in the destination network namespace.
4447 * This function shuts down a device interface and moves it
4448 * to a new network namespace. On success 0 is returned, on
4449 * a failure a netagive errno code is returned.
4451 * Callers must hold the rtnl semaphore.
4454 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4457 const char *destname;
4462 /* Don't allow namespace local devices to be moved. */
4464 if (dev->features & NETIF_F_NETNS_LOCAL)
4468 /* Don't allow real devices to be moved when sysfs
4472 if (dev->dev.parent)
4476 /* Ensure the device has been registrered */
4478 if (dev->reg_state != NETREG_REGISTERED)
4481 /* Get out if there is nothing todo */
4483 if (net_eq(dev_net(dev), net))
4486 /* Pick the destination device name, and ensure
4487 * we can use it in the destination network namespace.
4490 destname = dev->name;
4491 if (__dev_get_by_name(net, destname)) {
4492 /* We get here if we can't use the current device name */
4495 if (!dev_valid_name(pat))
4497 if (strchr(pat, '%')) {
4498 if (__dev_alloc_name(net, pat, buf) < 0)
4503 if (__dev_get_by_name(net, destname))
4508 * And now a mini version of register_netdevice unregister_netdevice.
4511 /* If device is running close it first. */
4514 /* And unlink it from device chain */
4516 unlist_netdevice(dev);
4520 /* Shutdown queueing discipline. */
4523 /* Notify protocols, that we are about to destroy
4524 this device. They should clean all the things.
4526 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4529 * Flush the unicast and multicast chains
4531 dev_addr_discard(dev);
4533 netdev_unregister_kobject(dev);
4535 /* Actually switch the network namespace */
4536 dev_net_set(dev, net);
4538 /* Assign the new device name */
4539 if (destname != dev->name)
4540 strcpy(dev->name, destname);
4542 /* If there is an ifindex conflict assign a new one */
4543 if (__dev_get_by_index(net, dev->ifindex)) {
4544 int iflink = (dev->iflink == dev->ifindex);
4545 dev->ifindex = dev_new_index(net);
4547 dev->iflink = dev->ifindex;
4550 /* Fixup kobjects */
4551 err = netdev_register_kobject(dev);
4554 /* Add the device back in the hashes */
4555 list_netdevice(dev);
4557 /* Notify protocols, that a new device appeared. */
4558 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4566 static int dev_cpu_callback(struct notifier_block *nfb,
4567 unsigned long action,
4570 struct sk_buff **list_skb;
4571 struct Qdisc **list_net;
4572 struct sk_buff *skb;
4573 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4574 struct softnet_data *sd, *oldsd;
4576 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4579 local_irq_disable();
4580 cpu = smp_processor_id();
4581 sd = &per_cpu(softnet_data, cpu);
4582 oldsd = &per_cpu(softnet_data, oldcpu);
4584 /* Find end of our completion_queue. */
4585 list_skb = &sd->completion_queue;
4587 list_skb = &(*list_skb)->next;
4588 /* Append completion queue from offline CPU. */
4589 *list_skb = oldsd->completion_queue;
4590 oldsd->completion_queue = NULL;
4592 /* Find end of our output_queue. */
4593 list_net = &sd->output_queue;
4595 list_net = &(*list_net)->next_sched;
4596 /* Append output queue from offline CPU. */
4597 *list_net = oldsd->output_queue;
4598 oldsd->output_queue = NULL;
4600 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4603 /* Process offline CPU's input_pkt_queue */
4604 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4610 #ifdef CONFIG_NET_DMA
4612 * net_dma_rebalance - try to maintain one DMA channel per CPU
4613 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4615 * This is called when the number of channels allocated to the net_dma client
4616 * changes. The net_dma client tries to have one DMA channel per CPU.
4619 static void net_dma_rebalance(struct net_dma *net_dma)
4621 unsigned int cpu, i, n, chan_idx;
4622 struct dma_chan *chan;
4624 if (cpus_empty(net_dma->channel_mask)) {
4625 for_each_online_cpu(cpu)
4626 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4631 cpu = first_cpu(cpu_online_map);
4633 for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) {
4634 chan = net_dma->channels[chan_idx];
4636 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4637 + (i < (num_online_cpus() %
4638 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4641 per_cpu(softnet_data, cpu).net_dma = chan;
4642 cpu = next_cpu(cpu, cpu_online_map);
4650 * netdev_dma_event - event callback for the net_dma_client
4651 * @client: should always be net_dma_client
4652 * @chan: DMA channel for the event
4653 * @state: DMA state to be handled
4655 static enum dma_state_client
4656 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4657 enum dma_state state)
4659 int i, found = 0, pos = -1;
4660 struct net_dma *net_dma =
4661 container_of(client, struct net_dma, client);
4662 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4664 spin_lock(&net_dma->lock);
4666 case DMA_RESOURCE_AVAILABLE:
4667 for (i = 0; i < nr_cpu_ids; i++)
4668 if (net_dma->channels[i] == chan) {
4671 } else if (net_dma->channels[i] == NULL && pos < 0)
4674 if (!found && pos >= 0) {
4676 net_dma->channels[pos] = chan;
4677 cpu_set(pos, net_dma->channel_mask);
4678 net_dma_rebalance(net_dma);
4681 case DMA_RESOURCE_REMOVED:
4682 for (i = 0; i < nr_cpu_ids; i++)
4683 if (net_dma->channels[i] == chan) {
4691 cpu_clear(pos, net_dma->channel_mask);
4692 net_dma->channels[i] = NULL;
4693 net_dma_rebalance(net_dma);
4699 spin_unlock(&net_dma->lock);
4705 * netdev_dma_register - register the networking subsystem as a DMA client
4707 static int __init netdev_dma_register(void)
4709 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4711 if (unlikely(!net_dma.channels)) {
4713 "netdev_dma: no memory for net_dma.channels\n");
4716 spin_lock_init(&net_dma.lock);
4717 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4718 dma_async_client_register(&net_dma.client);
4719 dma_async_client_chan_request(&net_dma.client);
4724 static int __init netdev_dma_register(void) { return -ENODEV; }
4725 #endif /* CONFIG_NET_DMA */
4728 * netdev_increment_features - increment feature set by one
4729 * @all: current feature set
4730 * @one: new feature set
4731 * @mask: mask feature set
4733 * Computes a new feature set after adding a device with feature set
4734 * @one to the master device with current feature set @all. Will not
4735 * enable anything that is off in @mask. Returns the new feature set.
4737 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
4740 /* If device needs checksumming, downgrade to it. */
4741 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4742 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
4743 else if (mask & NETIF_F_ALL_CSUM) {
4744 /* If one device supports v4/v6 checksumming, set for all. */
4745 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
4746 !(all & NETIF_F_GEN_CSUM)) {
4747 all &= ~NETIF_F_ALL_CSUM;
4748 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
4751 /* If one device supports hw checksumming, set for all. */
4752 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
4753 all &= ~NETIF_F_ALL_CSUM;
4754 all |= NETIF_F_HW_CSUM;
4758 one |= NETIF_F_ALL_CSUM;
4760 one |= all & NETIF_F_ONE_FOR_ALL;
4761 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
4762 all |= one & mask & NETIF_F_ONE_FOR_ALL;
4766 EXPORT_SYMBOL(netdev_increment_features);
4768 static struct hlist_head *netdev_create_hash(void)
4771 struct hlist_head *hash;
4773 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4775 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4776 INIT_HLIST_HEAD(&hash[i]);
4781 /* Initialize per network namespace state */
4782 static int __net_init netdev_init(struct net *net)
4784 INIT_LIST_HEAD(&net->dev_base_head);
4786 net->dev_name_head = netdev_create_hash();
4787 if (net->dev_name_head == NULL)
4790 net->dev_index_head = netdev_create_hash();
4791 if (net->dev_index_head == NULL)
4797 kfree(net->dev_name_head);
4803 * netdev_drivername - network driver for the device
4804 * @dev: network device
4805 * @buffer: buffer for resulting name
4806 * @len: size of buffer
4808 * Determine network driver for device.
4810 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
4812 const struct device_driver *driver;
4813 const struct device *parent;
4815 if (len <= 0 || !buffer)
4819 parent = dev->dev.parent;
4824 driver = parent->driver;
4825 if (driver && driver->name)
4826 strlcpy(buffer, driver->name, len);
4830 static void __net_exit netdev_exit(struct net *net)
4832 kfree(net->dev_name_head);
4833 kfree(net->dev_index_head);
4836 static struct pernet_operations __net_initdata netdev_net_ops = {
4837 .init = netdev_init,
4838 .exit = netdev_exit,
4841 static void __net_exit default_device_exit(struct net *net)
4843 struct net_device *dev, *next;
4845 * Push all migratable of the network devices back to the
4846 * initial network namespace
4849 for_each_netdev_safe(net, dev, next) {
4851 char fb_name[IFNAMSIZ];
4853 /* Ignore unmoveable devices (i.e. loopback) */
4854 if (dev->features & NETIF_F_NETNS_LOCAL)
4857 /* Delete virtual devices */
4858 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
4859 dev->rtnl_link_ops->dellink(dev);
4863 /* Push remaing network devices to init_net */
4864 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4865 err = dev_change_net_namespace(dev, &init_net, fb_name);
4867 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4868 __func__, dev->name, err);
4875 static struct pernet_operations __net_initdata default_device_ops = {
4876 .exit = default_device_exit,
4880 * Initialize the DEV module. At boot time this walks the device list and
4881 * unhooks any devices that fail to initialise (normally hardware not
4882 * present) and leaves us with a valid list of present and active devices.
4887 * This is called single threaded during boot, so no need
4888 * to take the rtnl semaphore.
4890 static int __init net_dev_init(void)
4892 int i, rc = -ENOMEM;
4894 BUG_ON(!dev_boot_phase);
4896 if (dev_proc_init())
4899 if (netdev_kobject_init())
4902 INIT_LIST_HEAD(&ptype_all);
4903 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4904 INIT_LIST_HEAD(&ptype_base[i]);
4906 if (register_pernet_subsys(&netdev_net_ops))
4909 /* The loopback device is special if any other network devices
4910 * is present in a network namespace the loopback device must
4911 * be present. Since we now dynamically allocate and free the
4912 * loopback device ensure this invariant is maintained by
4913 * keeping the loopback device as the first device on the
4914 * list of network devices. Ensuring the loopback devices
4915 * is the first device that appears and the last network device
4918 if (register_pernet_device(&loopback_net_ops))
4921 if (register_pernet_device(&default_device_ops))
4925 * Initialise the packet receive queues.
4928 for_each_possible_cpu(i) {
4929 struct softnet_data *queue;
4931 queue = &per_cpu(softnet_data, i);
4932 skb_queue_head_init(&queue->input_pkt_queue);
4933 queue->completion_queue = NULL;
4934 INIT_LIST_HEAD(&queue->poll_list);
4936 queue->backlog.poll = process_backlog;
4937 queue->backlog.weight = weight_p;
4940 netdev_dma_register();
4944 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
4945 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
4947 hotcpu_notifier(dev_cpu_callback, 0);
4955 subsys_initcall(net_dev_init);
4957 EXPORT_SYMBOL(__dev_get_by_index);
4958 EXPORT_SYMBOL(__dev_get_by_name);
4959 EXPORT_SYMBOL(__dev_remove_pack);
4960 EXPORT_SYMBOL(dev_valid_name);
4961 EXPORT_SYMBOL(dev_add_pack);
4962 EXPORT_SYMBOL(dev_alloc_name);
4963 EXPORT_SYMBOL(dev_close);
4964 EXPORT_SYMBOL(dev_get_by_flags);
4965 EXPORT_SYMBOL(dev_get_by_index);
4966 EXPORT_SYMBOL(dev_get_by_name);
4967 EXPORT_SYMBOL(dev_open);
4968 EXPORT_SYMBOL(dev_queue_xmit);
4969 EXPORT_SYMBOL(dev_remove_pack);
4970 EXPORT_SYMBOL(dev_set_allmulti);
4971 EXPORT_SYMBOL(dev_set_promiscuity);
4972 EXPORT_SYMBOL(dev_change_flags);
4973 EXPORT_SYMBOL(dev_set_mtu);
4974 EXPORT_SYMBOL(dev_set_mac_address);
4975 EXPORT_SYMBOL(free_netdev);
4976 EXPORT_SYMBOL(netdev_boot_setup_check);
4977 EXPORT_SYMBOL(netdev_set_master);
4978 EXPORT_SYMBOL(netdev_state_change);
4979 EXPORT_SYMBOL(netif_receive_skb);
4980 EXPORT_SYMBOL(netif_rx);
4981 EXPORT_SYMBOL(register_gifconf);
4982 EXPORT_SYMBOL(register_netdevice);
4983 EXPORT_SYMBOL(register_netdevice_notifier);
4984 EXPORT_SYMBOL(skb_checksum_help);
4985 EXPORT_SYMBOL(synchronize_net);
4986 EXPORT_SYMBOL(unregister_netdevice);
4987 EXPORT_SYMBOL(unregister_netdevice_notifier);
4988 EXPORT_SYMBOL(net_enable_timestamp);
4989 EXPORT_SYMBOL(net_disable_timestamp);
4990 EXPORT_SYMBOL(dev_get_flags);
4992 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4993 EXPORT_SYMBOL(br_handle_frame_hook);
4994 EXPORT_SYMBOL(br_fdb_get_hook);
4995 EXPORT_SYMBOL(br_fdb_put_hook);
4998 EXPORT_SYMBOL(dev_load);
5000 EXPORT_PER_CPU_SYMBOL(softnet_data);