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)))
2259 #ifdef CONFIG_NET_CLS_ACT
2260 if (skb->tc_verd & TC_NCLS) {
2261 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2266 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2267 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2268 ptype->dev == orig_dev) {
2270 ret = deliver_skb(skb, pt_prev, orig_dev);
2275 #ifdef CONFIG_NET_CLS_ACT
2276 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2282 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2285 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2289 type = skb->protocol;
2290 list_for_each_entry_rcu(ptype,
2291 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2292 if (ptype->type == type &&
2293 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2294 ptype->dev == orig_dev)) {
2296 ret = deliver_skb(skb, pt_prev, orig_dev);
2302 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2305 /* Jamal, now you will not able to escape explaining
2306 * me how you were going to use this. :-)
2316 /* Network device is going away, flush any packets still pending */
2317 static void flush_backlog(void *arg)
2319 struct net_device *dev = arg;
2320 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2321 struct sk_buff *skb, *tmp;
2323 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2324 if (skb->dev == dev) {
2325 __skb_unlink(skb, &queue->input_pkt_queue);
2330 static int process_backlog(struct napi_struct *napi, int quota)
2333 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2334 unsigned long start_time = jiffies;
2336 napi->weight = weight_p;
2338 struct sk_buff *skb;
2340 local_irq_disable();
2341 skb = __skb_dequeue(&queue->input_pkt_queue);
2343 __napi_complete(napi);
2349 netif_receive_skb(skb);
2350 } while (++work < quota && jiffies == start_time);
2356 * __napi_schedule - schedule for receive
2357 * @n: entry to schedule
2359 * The entry's receive function will be scheduled to run
2361 void __napi_schedule(struct napi_struct *n)
2363 unsigned long flags;
2365 local_irq_save(flags);
2366 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2367 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2368 local_irq_restore(flags);
2370 EXPORT_SYMBOL(__napi_schedule);
2373 static void net_rx_action(struct softirq_action *h)
2375 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2376 unsigned long time_limit = jiffies + 2;
2377 int budget = netdev_budget;
2380 local_irq_disable();
2382 while (!list_empty(list)) {
2383 struct napi_struct *n;
2386 /* If softirq window is exhuasted then punt.
2387 * Allow this to run for 2 jiffies since which will allow
2388 * an average latency of 1.5/HZ.
2390 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2395 /* Even though interrupts have been re-enabled, this
2396 * access is safe because interrupts can only add new
2397 * entries to the tail of this list, and only ->poll()
2398 * calls can remove this head entry from the list.
2400 n = list_entry(list->next, struct napi_struct, poll_list);
2402 have = netpoll_poll_lock(n);
2406 /* This NAPI_STATE_SCHED test is for avoiding a race
2407 * with netpoll's poll_napi(). Only the entity which
2408 * obtains the lock and sees NAPI_STATE_SCHED set will
2409 * actually make the ->poll() call. Therefore we avoid
2410 * accidently calling ->poll() when NAPI is not scheduled.
2413 if (test_bit(NAPI_STATE_SCHED, &n->state))
2414 work = n->poll(n, weight);
2416 WARN_ON_ONCE(work > weight);
2420 local_irq_disable();
2422 /* Drivers must not modify the NAPI state if they
2423 * consume the entire weight. In such cases this code
2424 * still "owns" the NAPI instance and therefore can
2425 * move the instance around on the list at-will.
2427 if (unlikely(work == weight)) {
2428 if (unlikely(napi_disable_pending(n)))
2431 list_move_tail(&n->poll_list, list);
2434 netpoll_poll_unlock(have);
2439 #ifdef CONFIG_NET_DMA
2441 * There may not be any more sk_buffs coming right now, so push
2442 * any pending DMA copies to hardware
2444 if (!cpus_empty(net_dma.channel_mask)) {
2446 for_each_cpu_mask_nr(chan_idx, net_dma.channel_mask) {
2447 struct dma_chan *chan = net_dma.channels[chan_idx];
2449 dma_async_memcpy_issue_pending(chan);
2457 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2458 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2462 static gifconf_func_t * gifconf_list [NPROTO];
2465 * register_gifconf - register a SIOCGIF handler
2466 * @family: Address family
2467 * @gifconf: Function handler
2469 * Register protocol dependent address dumping routines. The handler
2470 * that is passed must not be freed or reused until it has been replaced
2471 * by another handler.
2473 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2475 if (family >= NPROTO)
2477 gifconf_list[family] = gifconf;
2483 * Map an interface index to its name (SIOCGIFNAME)
2487 * We need this ioctl for efficient implementation of the
2488 * if_indextoname() function required by the IPv6 API. Without
2489 * it, we would have to search all the interfaces to find a
2493 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2495 struct net_device *dev;
2499 * Fetch the caller's info block.
2502 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2505 read_lock(&dev_base_lock);
2506 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2508 read_unlock(&dev_base_lock);
2512 strcpy(ifr.ifr_name, dev->name);
2513 read_unlock(&dev_base_lock);
2515 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2521 * Perform a SIOCGIFCONF call. This structure will change
2522 * size eventually, and there is nothing I can do about it.
2523 * Thus we will need a 'compatibility mode'.
2526 static int dev_ifconf(struct net *net, char __user *arg)
2529 struct net_device *dev;
2536 * Fetch the caller's info block.
2539 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2546 * Loop over the interfaces, and write an info block for each.
2550 for_each_netdev(net, dev) {
2551 for (i = 0; i < NPROTO; i++) {
2552 if (gifconf_list[i]) {
2555 done = gifconf_list[i](dev, NULL, 0);
2557 done = gifconf_list[i](dev, pos + total,
2567 * All done. Write the updated control block back to the caller.
2569 ifc.ifc_len = total;
2572 * Both BSD and Solaris return 0 here, so we do too.
2574 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2577 #ifdef CONFIG_PROC_FS
2579 * This is invoked by the /proc filesystem handler to display a device
2582 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2583 __acquires(dev_base_lock)
2585 struct net *net = seq_file_net(seq);
2587 struct net_device *dev;
2589 read_lock(&dev_base_lock);
2591 return SEQ_START_TOKEN;
2594 for_each_netdev(net, dev)
2601 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2603 struct net *net = seq_file_net(seq);
2605 return v == SEQ_START_TOKEN ?
2606 first_net_device(net) : next_net_device((struct net_device *)v);
2609 void dev_seq_stop(struct seq_file *seq, void *v)
2610 __releases(dev_base_lock)
2612 read_unlock(&dev_base_lock);
2615 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2617 struct net_device_stats *stats = dev->get_stats(dev);
2619 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2620 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2621 dev->name, stats->rx_bytes, stats->rx_packets,
2623 stats->rx_dropped + stats->rx_missed_errors,
2624 stats->rx_fifo_errors,
2625 stats->rx_length_errors + stats->rx_over_errors +
2626 stats->rx_crc_errors + stats->rx_frame_errors,
2627 stats->rx_compressed, stats->multicast,
2628 stats->tx_bytes, stats->tx_packets,
2629 stats->tx_errors, stats->tx_dropped,
2630 stats->tx_fifo_errors, stats->collisions,
2631 stats->tx_carrier_errors +
2632 stats->tx_aborted_errors +
2633 stats->tx_window_errors +
2634 stats->tx_heartbeat_errors,
2635 stats->tx_compressed);
2639 * Called from the PROCfs module. This now uses the new arbitrary sized
2640 * /proc/net interface to create /proc/net/dev
2642 static int dev_seq_show(struct seq_file *seq, void *v)
2644 if (v == SEQ_START_TOKEN)
2645 seq_puts(seq, "Inter-| Receive "
2647 " face |bytes packets errs drop fifo frame "
2648 "compressed multicast|bytes packets errs "
2649 "drop fifo colls carrier compressed\n");
2651 dev_seq_printf_stats(seq, v);
2655 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2657 struct netif_rx_stats *rc = NULL;
2659 while (*pos < nr_cpu_ids)
2660 if (cpu_online(*pos)) {
2661 rc = &per_cpu(netdev_rx_stat, *pos);
2668 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2670 return softnet_get_online(pos);
2673 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2676 return softnet_get_online(pos);
2679 static void softnet_seq_stop(struct seq_file *seq, void *v)
2683 static int softnet_seq_show(struct seq_file *seq, void *v)
2685 struct netif_rx_stats *s = v;
2687 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2688 s->total, s->dropped, s->time_squeeze, 0,
2689 0, 0, 0, 0, /* was fastroute */
2694 static const struct seq_operations dev_seq_ops = {
2695 .start = dev_seq_start,
2696 .next = dev_seq_next,
2697 .stop = dev_seq_stop,
2698 .show = dev_seq_show,
2701 static int dev_seq_open(struct inode *inode, struct file *file)
2703 return seq_open_net(inode, file, &dev_seq_ops,
2704 sizeof(struct seq_net_private));
2707 static const struct file_operations dev_seq_fops = {
2708 .owner = THIS_MODULE,
2709 .open = dev_seq_open,
2711 .llseek = seq_lseek,
2712 .release = seq_release_net,
2715 static const struct seq_operations softnet_seq_ops = {
2716 .start = softnet_seq_start,
2717 .next = softnet_seq_next,
2718 .stop = softnet_seq_stop,
2719 .show = softnet_seq_show,
2722 static int softnet_seq_open(struct inode *inode, struct file *file)
2724 return seq_open(file, &softnet_seq_ops);
2727 static const struct file_operations softnet_seq_fops = {
2728 .owner = THIS_MODULE,
2729 .open = softnet_seq_open,
2731 .llseek = seq_lseek,
2732 .release = seq_release,
2735 static void *ptype_get_idx(loff_t pos)
2737 struct packet_type *pt = NULL;
2741 list_for_each_entry_rcu(pt, &ptype_all, list) {
2747 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2748 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2757 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2761 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2764 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2766 struct packet_type *pt;
2767 struct list_head *nxt;
2771 if (v == SEQ_START_TOKEN)
2772 return ptype_get_idx(0);
2775 nxt = pt->list.next;
2776 if (pt->type == htons(ETH_P_ALL)) {
2777 if (nxt != &ptype_all)
2780 nxt = ptype_base[0].next;
2782 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2784 while (nxt == &ptype_base[hash]) {
2785 if (++hash >= PTYPE_HASH_SIZE)
2787 nxt = ptype_base[hash].next;
2790 return list_entry(nxt, struct packet_type, list);
2793 static void ptype_seq_stop(struct seq_file *seq, void *v)
2799 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2801 #ifdef CONFIG_KALLSYMS
2802 unsigned long offset = 0, symsize;
2803 const char *symname;
2807 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2814 modname = delim = "";
2815 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2821 seq_printf(seq, "[%p]", sym);
2824 static int ptype_seq_show(struct seq_file *seq, void *v)
2826 struct packet_type *pt = v;
2828 if (v == SEQ_START_TOKEN)
2829 seq_puts(seq, "Type Device Function\n");
2830 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2831 if (pt->type == htons(ETH_P_ALL))
2832 seq_puts(seq, "ALL ");
2834 seq_printf(seq, "%04x", ntohs(pt->type));
2836 seq_printf(seq, " %-8s ",
2837 pt->dev ? pt->dev->name : "");
2838 ptype_seq_decode(seq, pt->func);
2839 seq_putc(seq, '\n');
2845 static const struct seq_operations ptype_seq_ops = {
2846 .start = ptype_seq_start,
2847 .next = ptype_seq_next,
2848 .stop = ptype_seq_stop,
2849 .show = ptype_seq_show,
2852 static int ptype_seq_open(struct inode *inode, struct file *file)
2854 return seq_open_net(inode, file, &ptype_seq_ops,
2855 sizeof(struct seq_net_private));
2858 static const struct file_operations ptype_seq_fops = {
2859 .owner = THIS_MODULE,
2860 .open = ptype_seq_open,
2862 .llseek = seq_lseek,
2863 .release = seq_release_net,
2867 static int __net_init dev_proc_net_init(struct net *net)
2871 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2873 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2875 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2878 if (wext_proc_init(net))
2884 proc_net_remove(net, "ptype");
2886 proc_net_remove(net, "softnet_stat");
2888 proc_net_remove(net, "dev");
2892 static void __net_exit dev_proc_net_exit(struct net *net)
2894 wext_proc_exit(net);
2896 proc_net_remove(net, "ptype");
2897 proc_net_remove(net, "softnet_stat");
2898 proc_net_remove(net, "dev");
2901 static struct pernet_operations __net_initdata dev_proc_ops = {
2902 .init = dev_proc_net_init,
2903 .exit = dev_proc_net_exit,
2906 static int __init dev_proc_init(void)
2908 return register_pernet_subsys(&dev_proc_ops);
2911 #define dev_proc_init() 0
2912 #endif /* CONFIG_PROC_FS */
2916 * netdev_set_master - set up master/slave pair
2917 * @slave: slave device
2918 * @master: new master device
2920 * Changes the master device of the slave. Pass %NULL to break the
2921 * bonding. The caller must hold the RTNL semaphore. On a failure
2922 * a negative errno code is returned. On success the reference counts
2923 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2924 * function returns zero.
2926 int netdev_set_master(struct net_device *slave, struct net_device *master)
2928 struct net_device *old = slave->master;
2938 slave->master = master;
2946 slave->flags |= IFF_SLAVE;
2948 slave->flags &= ~IFF_SLAVE;
2950 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2954 static void dev_change_rx_flags(struct net_device *dev, int flags)
2956 if (dev->flags & IFF_UP && dev->change_rx_flags)
2957 dev->change_rx_flags(dev, flags);
2960 static int __dev_set_promiscuity(struct net_device *dev, int inc)
2962 unsigned short old_flags = dev->flags;
2966 dev->flags |= IFF_PROMISC;
2967 dev->promiscuity += inc;
2968 if (dev->promiscuity == 0) {
2971 * If inc causes overflow, untouch promisc and return error.
2974 dev->flags &= ~IFF_PROMISC;
2976 dev->promiscuity -= inc;
2977 printk(KERN_WARNING "%s: promiscuity touches roof, "
2978 "set promiscuity failed, promiscuity feature "
2979 "of device might be broken.\n", dev->name);
2983 if (dev->flags != old_flags) {
2984 printk(KERN_INFO "device %s %s promiscuous mode\n",
2985 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2988 audit_log(current->audit_context, GFP_ATOMIC,
2989 AUDIT_ANOM_PROMISCUOUS,
2990 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2991 dev->name, (dev->flags & IFF_PROMISC),
2992 (old_flags & IFF_PROMISC),
2993 audit_get_loginuid(current),
2994 current->uid, current->gid,
2995 audit_get_sessionid(current));
2997 dev_change_rx_flags(dev, IFF_PROMISC);
3003 * dev_set_promiscuity - update promiscuity count on a device
3007 * Add or remove promiscuity from a device. While the count in the device
3008 * remains above zero the interface remains promiscuous. Once it hits zero
3009 * the device reverts back to normal filtering operation. A negative inc
3010 * value is used to drop promiscuity on the device.
3011 * Return 0 if successful or a negative errno code on error.
3013 int dev_set_promiscuity(struct net_device *dev, int inc)
3015 unsigned short old_flags = dev->flags;
3018 err = __dev_set_promiscuity(dev, inc);
3021 if (dev->flags != old_flags)
3022 dev_set_rx_mode(dev);
3027 * dev_set_allmulti - update allmulti count on a device
3031 * Add or remove reception of all multicast frames to a device. While the
3032 * count in the device remains above zero the interface remains listening
3033 * to all interfaces. Once it hits zero the device reverts back to normal
3034 * filtering operation. A negative @inc value is used to drop the counter
3035 * when releasing a resource needing all multicasts.
3036 * Return 0 if successful or a negative errno code on error.
3039 int dev_set_allmulti(struct net_device *dev, int inc)
3041 unsigned short old_flags = dev->flags;
3045 dev->flags |= IFF_ALLMULTI;
3046 dev->allmulti += inc;
3047 if (dev->allmulti == 0) {
3050 * If inc causes overflow, untouch allmulti and return error.
3053 dev->flags &= ~IFF_ALLMULTI;
3055 dev->allmulti -= inc;
3056 printk(KERN_WARNING "%s: allmulti touches roof, "
3057 "set allmulti failed, allmulti feature of "
3058 "device might be broken.\n", dev->name);
3062 if (dev->flags ^ old_flags) {
3063 dev_change_rx_flags(dev, IFF_ALLMULTI);
3064 dev_set_rx_mode(dev);
3070 * Upload unicast and multicast address lists to device and
3071 * configure RX filtering. When the device doesn't support unicast
3072 * filtering it is put in promiscuous mode while unicast addresses
3075 void __dev_set_rx_mode(struct net_device *dev)
3077 /* dev_open will call this function so the list will stay sane. */
3078 if (!(dev->flags&IFF_UP))
3081 if (!netif_device_present(dev))
3084 if (dev->set_rx_mode)
3085 dev->set_rx_mode(dev);
3087 /* Unicast addresses changes may only happen under the rtnl,
3088 * therefore calling __dev_set_promiscuity here is safe.
3090 if (dev->uc_count > 0 && !dev->uc_promisc) {
3091 __dev_set_promiscuity(dev, 1);
3092 dev->uc_promisc = 1;
3093 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3094 __dev_set_promiscuity(dev, -1);
3095 dev->uc_promisc = 0;
3098 if (dev->set_multicast_list)
3099 dev->set_multicast_list(dev);
3103 void dev_set_rx_mode(struct net_device *dev)
3105 netif_addr_lock_bh(dev);
3106 __dev_set_rx_mode(dev);
3107 netif_addr_unlock_bh(dev);
3110 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3111 void *addr, int alen, int glbl)
3113 struct dev_addr_list *da;
3115 for (; (da = *list) != NULL; list = &da->next) {
3116 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3117 alen == da->da_addrlen) {
3119 int old_glbl = da->da_gusers;
3136 int __dev_addr_add(struct dev_addr_list **list, int *count,
3137 void *addr, int alen, int glbl)
3139 struct dev_addr_list *da;
3141 for (da = *list; da != NULL; da = da->next) {
3142 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3143 da->da_addrlen == alen) {
3145 int old_glbl = da->da_gusers;
3155 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3158 memcpy(da->da_addr, addr, alen);
3159 da->da_addrlen = alen;
3161 da->da_gusers = glbl ? 1 : 0;
3169 * dev_unicast_delete - Release secondary unicast address.
3171 * @addr: address to delete
3172 * @alen: length of @addr
3174 * Release reference to a secondary unicast address and remove it
3175 * from the device if the reference count drops to zero.
3177 * The caller must hold the rtnl_mutex.
3179 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3185 netif_addr_lock_bh(dev);
3186 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3188 __dev_set_rx_mode(dev);
3189 netif_addr_unlock_bh(dev);
3192 EXPORT_SYMBOL(dev_unicast_delete);
3195 * dev_unicast_add - add a secondary unicast address
3197 * @addr: address to add
3198 * @alen: length of @addr
3200 * Add a secondary unicast address to the device or increase
3201 * the reference count if it already exists.
3203 * The caller must hold the rtnl_mutex.
3205 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3211 netif_addr_lock_bh(dev);
3212 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3214 __dev_set_rx_mode(dev);
3215 netif_addr_unlock_bh(dev);
3218 EXPORT_SYMBOL(dev_unicast_add);
3220 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3221 struct dev_addr_list **from, int *from_count)
3223 struct dev_addr_list *da, *next;
3227 while (da != NULL) {
3229 if (!da->da_synced) {
3230 err = __dev_addr_add(to, to_count,
3231 da->da_addr, da->da_addrlen, 0);
3236 } else if (da->da_users == 1) {
3237 __dev_addr_delete(to, to_count,
3238 da->da_addr, da->da_addrlen, 0);
3239 __dev_addr_delete(from, from_count,
3240 da->da_addr, da->da_addrlen, 0);
3247 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3248 struct dev_addr_list **from, int *from_count)
3250 struct dev_addr_list *da, *next;
3253 while (da != NULL) {
3255 if (da->da_synced) {
3256 __dev_addr_delete(to, to_count,
3257 da->da_addr, da->da_addrlen, 0);
3259 __dev_addr_delete(from, from_count,
3260 da->da_addr, da->da_addrlen, 0);
3267 * dev_unicast_sync - Synchronize device's unicast list to another device
3268 * @to: destination device
3269 * @from: source device
3271 * Add newly added addresses to the destination device and release
3272 * addresses that have no users left. The source device must be
3273 * locked by netif_tx_lock_bh.
3275 * This function is intended to be called from the dev->set_rx_mode
3276 * function of layered software devices.
3278 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3282 netif_addr_lock_bh(to);
3283 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3284 &from->uc_list, &from->uc_count);
3286 __dev_set_rx_mode(to);
3287 netif_addr_unlock_bh(to);
3290 EXPORT_SYMBOL(dev_unicast_sync);
3293 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3294 * @to: destination device
3295 * @from: source device
3297 * Remove all addresses that were added to the destination device by
3298 * dev_unicast_sync(). This function is intended to be called from the
3299 * dev->stop function of layered software devices.
3301 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3303 netif_addr_lock_bh(from);
3304 netif_addr_lock(to);
3306 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3307 &from->uc_list, &from->uc_count);
3308 __dev_set_rx_mode(to);
3310 netif_addr_unlock(to);
3311 netif_addr_unlock_bh(from);
3313 EXPORT_SYMBOL(dev_unicast_unsync);
3315 static void __dev_addr_discard(struct dev_addr_list **list)
3317 struct dev_addr_list *tmp;
3319 while (*list != NULL) {
3322 if (tmp->da_users > tmp->da_gusers)
3323 printk("__dev_addr_discard: address leakage! "
3324 "da_users=%d\n", tmp->da_users);
3329 static void dev_addr_discard(struct net_device *dev)
3331 netif_addr_lock_bh(dev);
3333 __dev_addr_discard(&dev->uc_list);
3336 __dev_addr_discard(&dev->mc_list);
3339 netif_addr_unlock_bh(dev);
3343 * dev_get_flags - get flags reported to userspace
3346 * Get the combination of flag bits exported through APIs to userspace.
3348 unsigned dev_get_flags(const struct net_device *dev)
3352 flags = (dev->flags & ~(IFF_PROMISC |
3357 (dev->gflags & (IFF_PROMISC |
3360 if (netif_running(dev)) {
3361 if (netif_oper_up(dev))
3362 flags |= IFF_RUNNING;
3363 if (netif_carrier_ok(dev))
3364 flags |= IFF_LOWER_UP;
3365 if (netif_dormant(dev))
3366 flags |= IFF_DORMANT;
3373 * dev_change_flags - change device settings
3375 * @flags: device state flags
3377 * Change settings on device based state flags. The flags are
3378 * in the userspace exported format.
3380 int dev_change_flags(struct net_device *dev, unsigned flags)
3383 int old_flags = dev->flags;
3388 * Set the flags on our device.
3391 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3392 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3394 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3398 * Load in the correct multicast list now the flags have changed.
3401 if ((old_flags ^ flags) & IFF_MULTICAST)
3402 dev_change_rx_flags(dev, IFF_MULTICAST);
3404 dev_set_rx_mode(dev);
3407 * Have we downed the interface. We handle IFF_UP ourselves
3408 * according to user attempts to set it, rather than blindly
3413 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3414 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3417 dev_set_rx_mode(dev);
3420 if (dev->flags & IFF_UP &&
3421 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3423 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3425 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3426 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3427 dev->gflags ^= IFF_PROMISC;
3428 dev_set_promiscuity(dev, inc);
3431 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3432 is important. Some (broken) drivers set IFF_PROMISC, when
3433 IFF_ALLMULTI is requested not asking us and not reporting.
3435 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3436 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3437 dev->gflags ^= IFF_ALLMULTI;
3438 dev_set_allmulti(dev, inc);
3441 /* Exclude state transition flags, already notified */
3442 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3444 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3450 * dev_set_mtu - Change maximum transfer unit
3452 * @new_mtu: new transfer unit
3454 * Change the maximum transfer size of the network device.
3456 int dev_set_mtu(struct net_device *dev, int new_mtu)
3460 if (new_mtu == dev->mtu)
3463 /* MTU must be positive. */
3467 if (!netif_device_present(dev))
3471 if (dev->change_mtu)
3472 err = dev->change_mtu(dev, new_mtu);
3475 if (!err && dev->flags & IFF_UP)
3476 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3481 * dev_set_mac_address - Change Media Access Control Address
3485 * Change the hardware (MAC) address of the device
3487 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3491 if (!dev->set_mac_address)
3493 if (sa->sa_family != dev->type)
3495 if (!netif_device_present(dev))
3497 err = dev->set_mac_address(dev, sa);
3499 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3504 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3506 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3509 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3515 case SIOCGIFFLAGS: /* Get interface flags */
3516 ifr->ifr_flags = dev_get_flags(dev);
3519 case SIOCGIFMETRIC: /* Get the metric on the interface
3520 (currently unused) */
3521 ifr->ifr_metric = 0;
3524 case SIOCGIFMTU: /* Get the MTU of a device */
3525 ifr->ifr_mtu = dev->mtu;
3530 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3532 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3533 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3534 ifr->ifr_hwaddr.sa_family = dev->type;
3542 ifr->ifr_map.mem_start = dev->mem_start;
3543 ifr->ifr_map.mem_end = dev->mem_end;
3544 ifr->ifr_map.base_addr = dev->base_addr;
3545 ifr->ifr_map.irq = dev->irq;
3546 ifr->ifr_map.dma = dev->dma;
3547 ifr->ifr_map.port = dev->if_port;
3551 ifr->ifr_ifindex = dev->ifindex;
3555 ifr->ifr_qlen = dev->tx_queue_len;
3559 /* dev_ioctl() should ensure this case
3571 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3573 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3576 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3582 case SIOCSIFFLAGS: /* Set interface flags */
3583 return dev_change_flags(dev, ifr->ifr_flags);
3585 case SIOCSIFMETRIC: /* Set the metric on the interface
3586 (currently unused) */
3589 case SIOCSIFMTU: /* Set the MTU of a device */
3590 return dev_set_mtu(dev, ifr->ifr_mtu);
3593 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3595 case SIOCSIFHWBROADCAST:
3596 if (ifr->ifr_hwaddr.sa_family != dev->type)
3598 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3599 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3600 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3604 if (dev->set_config) {
3605 if (!netif_device_present(dev))
3607 return dev->set_config(dev, &ifr->ifr_map);
3612 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3613 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3615 if (!netif_device_present(dev))
3617 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3621 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3622 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3624 if (!netif_device_present(dev))
3626 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3630 if (ifr->ifr_qlen < 0)
3632 dev->tx_queue_len = ifr->ifr_qlen;
3636 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3637 return dev_change_name(dev, ifr->ifr_newname);
3640 * Unknown or private ioctl
3644 if ((cmd >= SIOCDEVPRIVATE &&
3645 cmd <= SIOCDEVPRIVATE + 15) ||
3646 cmd == SIOCBONDENSLAVE ||
3647 cmd == SIOCBONDRELEASE ||
3648 cmd == SIOCBONDSETHWADDR ||
3649 cmd == SIOCBONDSLAVEINFOQUERY ||
3650 cmd == SIOCBONDINFOQUERY ||
3651 cmd == SIOCBONDCHANGEACTIVE ||
3652 cmd == SIOCGMIIPHY ||
3653 cmd == SIOCGMIIREG ||
3654 cmd == SIOCSMIIREG ||
3655 cmd == SIOCBRADDIF ||
3656 cmd == SIOCBRDELIF ||
3657 cmd == SIOCWANDEV) {
3659 if (dev->do_ioctl) {
3660 if (netif_device_present(dev))
3661 err = dev->do_ioctl(dev, ifr,
3674 * This function handles all "interface"-type I/O control requests. The actual
3675 * 'doing' part of this is dev_ifsioc above.
3679 * dev_ioctl - network device ioctl
3680 * @net: the applicable net namespace
3681 * @cmd: command to issue
3682 * @arg: pointer to a struct ifreq in user space
3684 * Issue ioctl functions to devices. This is normally called by the
3685 * user space syscall interfaces but can sometimes be useful for
3686 * other purposes. The return value is the return from the syscall if
3687 * positive or a negative errno code on error.
3690 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3696 /* One special case: SIOCGIFCONF takes ifconf argument
3697 and requires shared lock, because it sleeps writing
3701 if (cmd == SIOCGIFCONF) {
3703 ret = dev_ifconf(net, (char __user *) arg);
3707 if (cmd == SIOCGIFNAME)
3708 return dev_ifname(net, (struct ifreq __user *)arg);
3710 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3713 ifr.ifr_name[IFNAMSIZ-1] = 0;
3715 colon = strchr(ifr.ifr_name, ':');
3720 * See which interface the caller is talking about.
3725 * These ioctl calls:
3726 * - can be done by all.
3727 * - atomic and do not require locking.
3738 dev_load(net, ifr.ifr_name);
3739 read_lock(&dev_base_lock);
3740 ret = dev_ifsioc_locked(net, &ifr, cmd);
3741 read_unlock(&dev_base_lock);
3745 if (copy_to_user(arg, &ifr,
3746 sizeof(struct ifreq)))
3752 dev_load(net, ifr.ifr_name);
3754 ret = dev_ethtool(net, &ifr);
3759 if (copy_to_user(arg, &ifr,
3760 sizeof(struct ifreq)))
3766 * These ioctl calls:
3767 * - require superuser power.
3768 * - require strict serialization.
3774 if (!capable(CAP_NET_ADMIN))
3776 dev_load(net, ifr.ifr_name);
3778 ret = dev_ifsioc(net, &ifr, cmd);
3783 if (copy_to_user(arg, &ifr,
3784 sizeof(struct ifreq)))
3790 * These ioctl calls:
3791 * - require superuser power.
3792 * - require strict serialization.
3793 * - do not return a value
3803 case SIOCSIFHWBROADCAST:
3806 case SIOCBONDENSLAVE:
3807 case SIOCBONDRELEASE:
3808 case SIOCBONDSETHWADDR:
3809 case SIOCBONDCHANGEACTIVE:
3812 if (!capable(CAP_NET_ADMIN))
3815 case SIOCBONDSLAVEINFOQUERY:
3816 case SIOCBONDINFOQUERY:
3817 dev_load(net, ifr.ifr_name);
3819 ret = dev_ifsioc(net, &ifr, cmd);
3824 /* Get the per device memory space. We can add this but
3825 * currently do not support it */
3827 /* Set the per device memory buffer space.
3828 * Not applicable in our case */
3833 * Unknown or private ioctl.
3836 if (cmd == SIOCWANDEV ||
3837 (cmd >= SIOCDEVPRIVATE &&
3838 cmd <= SIOCDEVPRIVATE + 15)) {
3839 dev_load(net, ifr.ifr_name);
3841 ret = dev_ifsioc(net, &ifr, cmd);
3843 if (!ret && copy_to_user(arg, &ifr,
3844 sizeof(struct ifreq)))
3848 /* Take care of Wireless Extensions */
3849 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3850 return wext_handle_ioctl(net, &ifr, cmd, arg);
3857 * dev_new_index - allocate an ifindex
3858 * @net: the applicable net namespace
3860 * Returns a suitable unique value for a new device interface
3861 * number. The caller must hold the rtnl semaphore or the
3862 * dev_base_lock to be sure it remains unique.
3864 static int dev_new_index(struct net *net)
3870 if (!__dev_get_by_index(net, ifindex))
3875 /* Delayed registration/unregisteration */
3876 static LIST_HEAD(net_todo_list);
3878 static void net_set_todo(struct net_device *dev)
3880 list_add_tail(&dev->todo_list, &net_todo_list);
3883 static void rollback_registered(struct net_device *dev)
3885 BUG_ON(dev_boot_phase);
3888 /* Some devices call without registering for initialization unwind. */
3889 if (dev->reg_state == NETREG_UNINITIALIZED) {
3890 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3891 "was registered\n", dev->name, dev);
3897 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3899 /* If device is running, close it first. */
3902 /* And unlink it from device chain. */
3903 unlist_netdevice(dev);
3905 dev->reg_state = NETREG_UNREGISTERING;
3909 /* Shutdown queueing discipline. */
3913 /* Notify protocols, that we are about to destroy
3914 this device. They should clean all the things.
3916 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3919 * Flush the unicast and multicast chains
3921 dev_addr_discard(dev);
3926 /* Notifier chain MUST detach us from master device. */
3927 WARN_ON(dev->master);
3929 /* Remove entries from kobject tree */
3930 netdev_unregister_kobject(dev);
3937 static void __netdev_init_queue_locks_one(struct net_device *dev,
3938 struct netdev_queue *dev_queue,
3941 spin_lock_init(&dev_queue->_xmit_lock);
3942 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
3943 dev_queue->xmit_lock_owner = -1;
3946 static void netdev_init_queue_locks(struct net_device *dev)
3948 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
3949 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
3952 unsigned long netdev_fix_features(unsigned long features, const char *name)
3954 /* Fix illegal SG+CSUM combinations. */
3955 if ((features & NETIF_F_SG) &&
3956 !(features & NETIF_F_ALL_CSUM)) {
3958 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
3959 "checksum feature.\n", name);
3960 features &= ~NETIF_F_SG;
3963 /* TSO requires that SG is present as well. */
3964 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
3966 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
3967 "SG feature.\n", name);
3968 features &= ~NETIF_F_TSO;
3971 if (features & NETIF_F_UFO) {
3972 if (!(features & NETIF_F_GEN_CSUM)) {
3974 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3975 "since no NETIF_F_HW_CSUM feature.\n",
3977 features &= ~NETIF_F_UFO;
3980 if (!(features & NETIF_F_SG)) {
3982 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
3983 "since no NETIF_F_SG feature.\n", name);
3984 features &= ~NETIF_F_UFO;
3990 EXPORT_SYMBOL(netdev_fix_features);
3993 * register_netdevice - register a network device
3994 * @dev: device to register
3996 * Take a completed network device structure and add it to the kernel
3997 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3998 * chain. 0 is returned on success. A negative errno code is returned
3999 * on a failure to set up the device, or if the name is a duplicate.
4001 * Callers must hold the rtnl semaphore. You may want
4002 * register_netdev() instead of this.
4005 * The locking appears insufficient to guarantee two parallel registers
4006 * will not get the same name.
4009 int register_netdevice(struct net_device *dev)
4011 struct hlist_head *head;
4012 struct hlist_node *p;
4016 BUG_ON(dev_boot_phase);
4021 /* When net_device's are persistent, this will be fatal. */
4022 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4023 BUG_ON(!dev_net(dev));
4026 spin_lock_init(&dev->addr_list_lock);
4027 netdev_set_addr_lockdep_class(dev);
4028 netdev_init_queue_locks(dev);
4032 /* Init, if this function is available */
4034 ret = dev->init(dev);
4042 if (!dev_valid_name(dev->name)) {
4047 dev->ifindex = dev_new_index(net);
4048 if (dev->iflink == -1)
4049 dev->iflink = dev->ifindex;
4051 /* Check for existence of name */
4052 head = dev_name_hash(net, dev->name);
4053 hlist_for_each(p, head) {
4054 struct net_device *d
4055 = hlist_entry(p, struct net_device, name_hlist);
4056 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4062 /* Fix illegal checksum combinations */
4063 if ((dev->features & NETIF_F_HW_CSUM) &&
4064 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4065 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4067 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4070 if ((dev->features & NETIF_F_NO_CSUM) &&
4071 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4072 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4074 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4077 dev->features = netdev_fix_features(dev->features, dev->name);
4079 /* Enable software GSO if SG is supported. */
4080 if (dev->features & NETIF_F_SG)
4081 dev->features |= NETIF_F_GSO;
4083 netdev_initialize_kobject(dev);
4084 ret = netdev_register_kobject(dev);
4087 dev->reg_state = NETREG_REGISTERED;
4090 * Default initial state at registry is that the
4091 * device is present.
4094 set_bit(__LINK_STATE_PRESENT, &dev->state);
4096 dev_init_scheduler(dev);
4098 list_netdevice(dev);
4100 /* Notify protocols, that a new device appeared. */
4101 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4102 ret = notifier_to_errno(ret);
4104 rollback_registered(dev);
4105 dev->reg_state = NETREG_UNREGISTERED;
4118 * register_netdev - register a network device
4119 * @dev: device to register
4121 * Take a completed network device structure and add it to the kernel
4122 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4123 * chain. 0 is returned on success. A negative errno code is returned
4124 * on a failure to set up the device, or if the name is a duplicate.
4126 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4127 * and expands the device name if you passed a format string to
4130 int register_netdev(struct net_device *dev)
4137 * If the name is a format string the caller wants us to do a
4140 if (strchr(dev->name, '%')) {
4141 err = dev_alloc_name(dev, dev->name);
4146 err = register_netdevice(dev);
4151 EXPORT_SYMBOL(register_netdev);
4154 * netdev_wait_allrefs - wait until all references are gone.
4156 * This is called when unregistering network devices.
4158 * Any protocol or device that holds a reference should register
4159 * for netdevice notification, and cleanup and put back the
4160 * reference if they receive an UNREGISTER event.
4161 * We can get stuck here if buggy protocols don't correctly
4164 static void netdev_wait_allrefs(struct net_device *dev)
4166 unsigned long rebroadcast_time, warning_time;
4168 rebroadcast_time = warning_time = jiffies;
4169 while (atomic_read(&dev->refcnt) != 0) {
4170 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4173 /* Rebroadcast unregister notification */
4174 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4176 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4178 /* We must not have linkwatch events
4179 * pending on unregister. If this
4180 * happens, we simply run the queue
4181 * unscheduled, resulting in a noop
4184 linkwatch_run_queue();
4189 rebroadcast_time = jiffies;
4194 if (time_after(jiffies, warning_time + 10 * HZ)) {
4195 printk(KERN_EMERG "unregister_netdevice: "
4196 "waiting for %s to become free. Usage "
4198 dev->name, atomic_read(&dev->refcnt));
4199 warning_time = jiffies;
4208 * register_netdevice(x1);
4209 * register_netdevice(x2);
4211 * unregister_netdevice(y1);
4212 * unregister_netdevice(y2);
4218 * We are invoked by rtnl_unlock().
4219 * This allows us to deal with problems:
4220 * 1) We can delete sysfs objects which invoke hotplug
4221 * without deadlocking with linkwatch via keventd.
4222 * 2) Since we run with the RTNL semaphore not held, we can sleep
4223 * safely in order to wait for the netdev refcnt to drop to zero.
4225 * We must not return until all unregister events added during
4226 * the interval the lock was held have been completed.
4228 void netdev_run_todo(void)
4230 struct list_head list;
4232 /* Snapshot list, allow later requests */
4233 list_replace_init(&net_todo_list, &list);
4237 while (!list_empty(&list)) {
4238 struct net_device *dev
4239 = list_entry(list.next, struct net_device, todo_list);
4240 list_del(&dev->todo_list);
4242 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4243 printk(KERN_ERR "network todo '%s' but state %d\n",
4244 dev->name, dev->reg_state);
4249 dev->reg_state = NETREG_UNREGISTERED;
4251 on_each_cpu(flush_backlog, dev, 1);
4253 netdev_wait_allrefs(dev);
4256 BUG_ON(atomic_read(&dev->refcnt));
4257 WARN_ON(dev->ip_ptr);
4258 WARN_ON(dev->ip6_ptr);
4259 WARN_ON(dev->dn_ptr);
4261 if (dev->destructor)
4262 dev->destructor(dev);
4264 /* Free network device */
4265 kobject_put(&dev->dev.kobj);
4269 static struct net_device_stats *internal_stats(struct net_device *dev)
4274 static void netdev_init_one_queue(struct net_device *dev,
4275 struct netdev_queue *queue,
4281 static void netdev_init_queues(struct net_device *dev)
4283 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4284 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4285 spin_lock_init(&dev->tx_global_lock);
4289 * alloc_netdev_mq - allocate network device
4290 * @sizeof_priv: size of private data to allocate space for
4291 * @name: device name format string
4292 * @setup: callback to initialize device
4293 * @queue_count: the number of subqueues to allocate
4295 * Allocates a struct net_device with private data area for driver use
4296 * and performs basic initialization. Also allocates subquue structs
4297 * for each queue on the device at the end of the netdevice.
4299 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4300 void (*setup)(struct net_device *), unsigned int queue_count)
4302 struct netdev_queue *tx;
4303 struct net_device *dev;
4307 BUG_ON(strlen(name) >= sizeof(dev->name));
4309 alloc_size = sizeof(struct net_device);
4311 /* ensure 32-byte alignment of private area */
4312 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4313 alloc_size += sizeof_priv;
4315 /* ensure 32-byte alignment of whole construct */
4316 alloc_size += NETDEV_ALIGN_CONST;
4318 p = kzalloc(alloc_size, GFP_KERNEL);
4320 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4324 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4326 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4332 dev = (struct net_device *)
4333 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4334 dev->padded = (char *)dev - (char *)p;
4335 dev_net_set(dev, &init_net);
4338 dev->num_tx_queues = queue_count;
4339 dev->real_num_tx_queues = queue_count;
4342 dev->priv = ((char *)dev +
4343 ((sizeof(struct net_device) + NETDEV_ALIGN_CONST)
4344 & ~NETDEV_ALIGN_CONST));
4347 dev->gso_max_size = GSO_MAX_SIZE;
4349 netdev_init_queues(dev);
4351 dev->get_stats = internal_stats;
4352 netpoll_netdev_init(dev);
4354 strcpy(dev->name, name);
4357 EXPORT_SYMBOL(alloc_netdev_mq);
4360 * free_netdev - free network device
4363 * This function does the last stage of destroying an allocated device
4364 * interface. The reference to the device object is released.
4365 * If this is the last reference then it will be freed.
4367 void free_netdev(struct net_device *dev)
4369 release_net(dev_net(dev));
4373 /* Compatibility with error handling in drivers */
4374 if (dev->reg_state == NETREG_UNINITIALIZED) {
4375 kfree((char *)dev - dev->padded);
4379 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4380 dev->reg_state = NETREG_RELEASED;
4382 /* will free via device release */
4383 put_device(&dev->dev);
4387 * synchronize_net - Synchronize with packet receive processing
4389 * Wait for packets currently being received to be done.
4390 * Does not block later packets from starting.
4392 void synchronize_net(void)
4399 * unregister_netdevice - remove device from the kernel
4402 * This function shuts down a device interface and removes it
4403 * from the kernel tables.
4405 * Callers must hold the rtnl semaphore. You may want
4406 * unregister_netdev() instead of this.
4409 void unregister_netdevice(struct net_device *dev)
4413 rollback_registered(dev);
4414 /* Finish processing unregister after unlock */
4419 * unregister_netdev - remove device from the kernel
4422 * This function shuts down a device interface and removes it
4423 * from the kernel tables.
4425 * This is just a wrapper for unregister_netdevice that takes
4426 * the rtnl semaphore. In general you want to use this and not
4427 * unregister_netdevice.
4429 void unregister_netdev(struct net_device *dev)
4432 unregister_netdevice(dev);
4436 EXPORT_SYMBOL(unregister_netdev);
4439 * dev_change_net_namespace - move device to different nethost namespace
4441 * @net: network namespace
4442 * @pat: If not NULL name pattern to try if the current device name
4443 * is already taken in the destination network namespace.
4445 * This function shuts down a device interface and moves it
4446 * to a new network namespace. On success 0 is returned, on
4447 * a failure a netagive errno code is returned.
4449 * Callers must hold the rtnl semaphore.
4452 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4455 const char *destname;
4460 /* Don't allow namespace local devices to be moved. */
4462 if (dev->features & NETIF_F_NETNS_LOCAL)
4466 /* Don't allow real devices to be moved when sysfs
4470 if (dev->dev.parent)
4474 /* Ensure the device has been registrered */
4476 if (dev->reg_state != NETREG_REGISTERED)
4479 /* Get out if there is nothing todo */
4481 if (net_eq(dev_net(dev), net))
4484 /* Pick the destination device name, and ensure
4485 * we can use it in the destination network namespace.
4488 destname = dev->name;
4489 if (__dev_get_by_name(net, destname)) {
4490 /* We get here if we can't use the current device name */
4493 if (!dev_valid_name(pat))
4495 if (strchr(pat, '%')) {
4496 if (__dev_alloc_name(net, pat, buf) < 0)
4501 if (__dev_get_by_name(net, destname))
4506 * And now a mini version of register_netdevice unregister_netdevice.
4509 /* If device is running close it first. */
4512 /* And unlink it from device chain */
4514 unlist_netdevice(dev);
4518 /* Shutdown queueing discipline. */
4521 /* Notify protocols, that we are about to destroy
4522 this device. They should clean all the things.
4524 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4527 * Flush the unicast and multicast chains
4529 dev_addr_discard(dev);
4531 netdev_unregister_kobject(dev);
4533 /* Actually switch the network namespace */
4534 dev_net_set(dev, net);
4536 /* Assign the new device name */
4537 if (destname != dev->name)
4538 strcpy(dev->name, destname);
4540 /* If there is an ifindex conflict assign a new one */
4541 if (__dev_get_by_index(net, dev->ifindex)) {
4542 int iflink = (dev->iflink == dev->ifindex);
4543 dev->ifindex = dev_new_index(net);
4545 dev->iflink = dev->ifindex;
4548 /* Fixup kobjects */
4549 err = netdev_register_kobject(dev);
4552 /* Add the device back in the hashes */
4553 list_netdevice(dev);
4555 /* Notify protocols, that a new device appeared. */
4556 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4564 static int dev_cpu_callback(struct notifier_block *nfb,
4565 unsigned long action,
4568 struct sk_buff **list_skb;
4569 struct Qdisc **list_net;
4570 struct sk_buff *skb;
4571 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4572 struct softnet_data *sd, *oldsd;
4574 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4577 local_irq_disable();
4578 cpu = smp_processor_id();
4579 sd = &per_cpu(softnet_data, cpu);
4580 oldsd = &per_cpu(softnet_data, oldcpu);
4582 /* Find end of our completion_queue. */
4583 list_skb = &sd->completion_queue;
4585 list_skb = &(*list_skb)->next;
4586 /* Append completion queue from offline CPU. */
4587 *list_skb = oldsd->completion_queue;
4588 oldsd->completion_queue = NULL;
4590 /* Find end of our output_queue. */
4591 list_net = &sd->output_queue;
4593 list_net = &(*list_net)->next_sched;
4594 /* Append output queue from offline CPU. */
4595 *list_net = oldsd->output_queue;
4596 oldsd->output_queue = NULL;
4598 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4601 /* Process offline CPU's input_pkt_queue */
4602 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4608 #ifdef CONFIG_NET_DMA
4610 * net_dma_rebalance - try to maintain one DMA channel per CPU
4611 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4613 * This is called when the number of channels allocated to the net_dma client
4614 * changes. The net_dma client tries to have one DMA channel per CPU.
4617 static void net_dma_rebalance(struct net_dma *net_dma)
4619 unsigned int cpu, i, n, chan_idx;
4620 struct dma_chan *chan;
4622 if (cpus_empty(net_dma->channel_mask)) {
4623 for_each_online_cpu(cpu)
4624 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4629 cpu = first_cpu(cpu_online_map);
4631 for_each_cpu_mask_nr(chan_idx, net_dma->channel_mask) {
4632 chan = net_dma->channels[chan_idx];
4634 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4635 + (i < (num_online_cpus() %
4636 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4639 per_cpu(softnet_data, cpu).net_dma = chan;
4640 cpu = next_cpu(cpu, cpu_online_map);
4648 * netdev_dma_event - event callback for the net_dma_client
4649 * @client: should always be net_dma_client
4650 * @chan: DMA channel for the event
4651 * @state: DMA state to be handled
4653 static enum dma_state_client
4654 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4655 enum dma_state state)
4657 int i, found = 0, pos = -1;
4658 struct net_dma *net_dma =
4659 container_of(client, struct net_dma, client);
4660 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4662 spin_lock(&net_dma->lock);
4664 case DMA_RESOURCE_AVAILABLE:
4665 for (i = 0; i < nr_cpu_ids; i++)
4666 if (net_dma->channels[i] == chan) {
4669 } else if (net_dma->channels[i] == NULL && pos < 0)
4672 if (!found && pos >= 0) {
4674 net_dma->channels[pos] = chan;
4675 cpu_set(pos, net_dma->channel_mask);
4676 net_dma_rebalance(net_dma);
4679 case DMA_RESOURCE_REMOVED:
4680 for (i = 0; i < nr_cpu_ids; i++)
4681 if (net_dma->channels[i] == chan) {
4689 cpu_clear(pos, net_dma->channel_mask);
4690 net_dma->channels[i] = NULL;
4691 net_dma_rebalance(net_dma);
4697 spin_unlock(&net_dma->lock);
4703 * netdev_dma_register - register the networking subsystem as a DMA client
4705 static int __init netdev_dma_register(void)
4707 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4709 if (unlikely(!net_dma.channels)) {
4711 "netdev_dma: no memory for net_dma.channels\n");
4714 spin_lock_init(&net_dma.lock);
4715 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4716 dma_async_client_register(&net_dma.client);
4717 dma_async_client_chan_request(&net_dma.client);
4722 static int __init netdev_dma_register(void) { return -ENODEV; }
4723 #endif /* CONFIG_NET_DMA */
4726 * netdev_increment_features - increment feature set by one
4727 * @all: current feature set
4728 * @one: new feature set
4729 * @mask: mask feature set
4731 * Computes a new feature set after adding a device with feature set
4732 * @one to the master device with current feature set @all. Will not
4733 * enable anything that is off in @mask. Returns the new feature set.
4735 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
4738 /* If device needs checksumming, downgrade to it. */
4739 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4740 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
4741 else if (mask & NETIF_F_ALL_CSUM) {
4742 /* If one device supports v4/v6 checksumming, set for all. */
4743 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
4744 !(all & NETIF_F_GEN_CSUM)) {
4745 all &= ~NETIF_F_ALL_CSUM;
4746 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
4749 /* If one device supports hw checksumming, set for all. */
4750 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
4751 all &= ~NETIF_F_ALL_CSUM;
4752 all |= NETIF_F_HW_CSUM;
4756 one |= NETIF_F_ALL_CSUM;
4758 one |= all & NETIF_F_ONE_FOR_ALL;
4759 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
4760 all |= one & mask & NETIF_F_ONE_FOR_ALL;
4764 EXPORT_SYMBOL(netdev_increment_features);
4766 static struct hlist_head *netdev_create_hash(void)
4769 struct hlist_head *hash;
4771 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4773 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4774 INIT_HLIST_HEAD(&hash[i]);
4779 /* Initialize per network namespace state */
4780 static int __net_init netdev_init(struct net *net)
4782 INIT_LIST_HEAD(&net->dev_base_head);
4784 net->dev_name_head = netdev_create_hash();
4785 if (net->dev_name_head == NULL)
4788 net->dev_index_head = netdev_create_hash();
4789 if (net->dev_index_head == NULL)
4795 kfree(net->dev_name_head);
4801 * netdev_drivername - network driver for the device
4802 * @dev: network device
4803 * @buffer: buffer for resulting name
4804 * @len: size of buffer
4806 * Determine network driver for device.
4808 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
4810 const struct device_driver *driver;
4811 const struct device *parent;
4813 if (len <= 0 || !buffer)
4817 parent = dev->dev.parent;
4822 driver = parent->driver;
4823 if (driver && driver->name)
4824 strlcpy(buffer, driver->name, len);
4828 static void __net_exit netdev_exit(struct net *net)
4830 kfree(net->dev_name_head);
4831 kfree(net->dev_index_head);
4834 static struct pernet_operations __net_initdata netdev_net_ops = {
4835 .init = netdev_init,
4836 .exit = netdev_exit,
4839 static void __net_exit default_device_exit(struct net *net)
4841 struct net_device *dev, *next;
4843 * Push all migratable of the network devices back to the
4844 * initial network namespace
4847 for_each_netdev_safe(net, dev, next) {
4849 char fb_name[IFNAMSIZ];
4851 /* Ignore unmoveable devices (i.e. loopback) */
4852 if (dev->features & NETIF_F_NETNS_LOCAL)
4855 /* Delete virtual devices */
4856 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
4857 dev->rtnl_link_ops->dellink(dev);
4861 /* Push remaing network devices to init_net */
4862 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4863 err = dev_change_net_namespace(dev, &init_net, fb_name);
4865 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4866 __func__, dev->name, err);
4873 static struct pernet_operations __net_initdata default_device_ops = {
4874 .exit = default_device_exit,
4878 * Initialize the DEV module. At boot time this walks the device list and
4879 * unhooks any devices that fail to initialise (normally hardware not
4880 * present) and leaves us with a valid list of present and active devices.
4885 * This is called single threaded during boot, so no need
4886 * to take the rtnl semaphore.
4888 static int __init net_dev_init(void)
4890 int i, rc = -ENOMEM;
4892 BUG_ON(!dev_boot_phase);
4894 if (dev_proc_init())
4897 if (netdev_kobject_init())
4900 INIT_LIST_HEAD(&ptype_all);
4901 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4902 INIT_LIST_HEAD(&ptype_base[i]);
4904 if (register_pernet_subsys(&netdev_net_ops))
4907 /* The loopback device is special if any other network devices
4908 * is present in a network namespace the loopback device must
4909 * be present. Since we now dynamically allocate and free the
4910 * loopback device ensure this invariant is maintained by
4911 * keeping the loopback device as the first device on the
4912 * list of network devices. Ensuring the loopback devices
4913 * is the first device that appears and the last network device
4916 if (register_pernet_device(&loopback_net_ops))
4919 if (register_pernet_device(&default_device_ops))
4923 * Initialise the packet receive queues.
4926 for_each_possible_cpu(i) {
4927 struct softnet_data *queue;
4929 queue = &per_cpu(softnet_data, i);
4930 skb_queue_head_init(&queue->input_pkt_queue);
4931 queue->completion_queue = NULL;
4932 INIT_LIST_HEAD(&queue->poll_list);
4934 queue->backlog.poll = process_backlog;
4935 queue->backlog.weight = weight_p;
4938 netdev_dma_register();
4942 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
4943 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
4945 hotcpu_notifier(dev_cpu_callback, 0);
4953 subsys_initcall(net_dev_init);
4955 EXPORT_SYMBOL(__dev_get_by_index);
4956 EXPORT_SYMBOL(__dev_get_by_name);
4957 EXPORT_SYMBOL(__dev_remove_pack);
4958 EXPORT_SYMBOL(dev_valid_name);
4959 EXPORT_SYMBOL(dev_add_pack);
4960 EXPORT_SYMBOL(dev_alloc_name);
4961 EXPORT_SYMBOL(dev_close);
4962 EXPORT_SYMBOL(dev_get_by_flags);
4963 EXPORT_SYMBOL(dev_get_by_index);
4964 EXPORT_SYMBOL(dev_get_by_name);
4965 EXPORT_SYMBOL(dev_open);
4966 EXPORT_SYMBOL(dev_queue_xmit);
4967 EXPORT_SYMBOL(dev_remove_pack);
4968 EXPORT_SYMBOL(dev_set_allmulti);
4969 EXPORT_SYMBOL(dev_set_promiscuity);
4970 EXPORT_SYMBOL(dev_change_flags);
4971 EXPORT_SYMBOL(dev_set_mtu);
4972 EXPORT_SYMBOL(dev_set_mac_address);
4973 EXPORT_SYMBOL(free_netdev);
4974 EXPORT_SYMBOL(netdev_boot_setup_check);
4975 EXPORT_SYMBOL(netdev_set_master);
4976 EXPORT_SYMBOL(netdev_state_change);
4977 EXPORT_SYMBOL(netif_receive_skb);
4978 EXPORT_SYMBOL(netif_rx);
4979 EXPORT_SYMBOL(register_gifconf);
4980 EXPORT_SYMBOL(register_netdevice);
4981 EXPORT_SYMBOL(register_netdevice_notifier);
4982 EXPORT_SYMBOL(skb_checksum_help);
4983 EXPORT_SYMBOL(synchronize_net);
4984 EXPORT_SYMBOL(unregister_netdevice);
4985 EXPORT_SYMBOL(unregister_netdevice_notifier);
4986 EXPORT_SYMBOL(net_enable_timestamp);
4987 EXPORT_SYMBOL(net_disable_timestamp);
4988 EXPORT_SYMBOL(dev_get_flags);
4990 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4991 EXPORT_SYMBOL(br_handle_frame_hook);
4992 EXPORT_SYMBOL(br_fdb_get_hook);
4993 EXPORT_SYMBOL(br_fdb_put_hook);
4996 EXPORT_SYMBOL(dev_load);
4998 EXPORT_PER_CPU_SYMBOL(softnet_data);