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/notifier.h>
94 #include <linux/skbuff.h>
95 #include <net/net_namespace.h>
97 #include <linux/rtnetlink.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <linux/stat.h>
101 #include <linux/if_bridge.h>
102 #include <linux/if_macvlan.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/kmod.h>
109 #include <linux/module.h>
110 #include <linux/kallsyms.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
124 * The list of packet types we will receive (as opposed to discard)
125 * and the routines to invoke.
127 * Why 16. Because with 16 the only overlap we get on a hash of the
128 * low nibble of the protocol value is RARP/SNAP/X.25.
130 * NOTE: That is no longer true with the addition of VLAN tags. Not
131 * sure which should go first, but I bet it won't make much
132 * difference if we are running VLANs. The good news is that
133 * this protocol won't be in the list unless compiled in, so
134 * the average user (w/out VLANs) will not be adversely affected.
151 static DEFINE_SPINLOCK(ptype_lock);
152 static struct list_head ptype_base[16] __read_mostly; /* 16 way hashed list */
153 static struct list_head ptype_all __read_mostly; /* Taps */
155 #ifdef CONFIG_NET_DMA
157 struct dma_client client;
159 cpumask_t channel_mask;
160 struct dma_chan *channels[NR_CPUS];
163 static enum dma_state_client
164 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
165 enum dma_state state);
167 static struct net_dma net_dma = {
169 .event_callback = netdev_dma_event,
175 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
178 * Pure readers hold dev_base_lock for reading.
180 * Writers must hold the rtnl semaphore while they loop through the
181 * dev_base_head list, and hold dev_base_lock for writing when they do the
182 * actual updates. This allows pure readers to access the list even
183 * while a writer is preparing to update it.
185 * To put it another way, dev_base_lock is held for writing only to
186 * protect against pure readers; the rtnl semaphore provides the
187 * protection against other writers.
189 * See, for example usages, register_netdevice() and
190 * unregister_netdevice(), which must be called with the rtnl
193 DEFINE_RWLOCK(dev_base_lock);
195 EXPORT_SYMBOL(dev_base_lock);
197 #define NETDEV_HASHBITS 8
198 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
200 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
202 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
203 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
208 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
211 /* Device list insertion */
212 static int list_netdevice(struct net_device *dev)
214 struct net *net = dev->nd_net;
218 write_lock_bh(&dev_base_lock);
219 list_add_tail(&dev->dev_list, &net->dev_base_head);
220 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
221 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
222 write_unlock_bh(&dev_base_lock);
226 /* Device list removal */
227 static void unlist_netdevice(struct net_device *dev)
231 /* Unlink dev from the device chain */
232 write_lock_bh(&dev_base_lock);
233 list_del(&dev->dev_list);
234 hlist_del(&dev->name_hlist);
235 hlist_del(&dev->index_hlist);
236 write_unlock_bh(&dev_base_lock);
243 static RAW_NOTIFIER_HEAD(netdev_chain);
246 * Device drivers call our routines to queue packets here. We empty the
247 * queue in the local softnet handler.
250 DEFINE_PER_CPU(struct softnet_data, softnet_data);
252 extern int netdev_kobject_init(void);
253 extern int netdev_register_kobject(struct net_device *);
254 extern void netdev_unregister_kobject(struct net_device *);
256 #ifdef CONFIG_DEBUG_LOCK_ALLOC
258 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
259 * according to dev->type
261 static const unsigned short netdev_lock_type[] =
262 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
263 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
264 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
265 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
266 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
267 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
268 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
269 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
270 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
271 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
272 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
273 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
274 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
275 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
278 static const char *netdev_lock_name[] =
279 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
280 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
281 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
282 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
283 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
284 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
285 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
286 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
287 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
288 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
289 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
290 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
291 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
292 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
295 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
297 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
301 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
302 if (netdev_lock_type[i] == dev_type)
304 /* the last key is used by default */
305 return ARRAY_SIZE(netdev_lock_type) - 1;
308 static inline void netdev_set_lockdep_class(spinlock_t *lock,
309 unsigned short dev_type)
313 i = netdev_lock_pos(dev_type);
314 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
315 netdev_lock_name[i]);
318 static inline void netdev_set_lockdep_class(spinlock_t *lock,
319 unsigned short dev_type)
324 /*******************************************************************************
326 Protocol management and registration routines
328 *******************************************************************************/
331 * Add a protocol ID to the list. Now that the input handler is
332 * smarter we can dispense with all the messy stuff that used to be
335 * BEWARE!!! Protocol handlers, mangling input packets,
336 * MUST BE last in hash buckets and checking protocol handlers
337 * MUST start from promiscuous ptype_all chain in net_bh.
338 * It is true now, do not change it.
339 * Explanation follows: if protocol handler, mangling packet, will
340 * be the first on list, it is not able to sense, that packet
341 * is cloned and should be copied-on-write, so that it will
342 * change it and subsequent readers will get broken packet.
347 * dev_add_pack - add packet handler
348 * @pt: packet type declaration
350 * Add a protocol handler to the networking stack. The passed &packet_type
351 * is linked into kernel lists and may not be freed until it has been
352 * removed from the kernel lists.
354 * This call does not sleep therefore it can not
355 * guarantee all CPU's that are in middle of receiving packets
356 * will see the new packet type (until the next received packet).
359 void dev_add_pack(struct packet_type *pt)
363 spin_lock_bh(&ptype_lock);
364 if (pt->type == htons(ETH_P_ALL))
365 list_add_rcu(&pt->list, &ptype_all);
367 hash = ntohs(pt->type) & 15;
368 list_add_rcu(&pt->list, &ptype_base[hash]);
370 spin_unlock_bh(&ptype_lock);
374 * __dev_remove_pack - remove packet handler
375 * @pt: packet type declaration
377 * Remove a protocol handler that was previously added to the kernel
378 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
379 * from the kernel lists and can be freed or reused once this function
382 * The packet type might still be in use by receivers
383 * and must not be freed until after all the CPU's have gone
384 * through a quiescent state.
386 void __dev_remove_pack(struct packet_type *pt)
388 struct list_head *head;
389 struct packet_type *pt1;
391 spin_lock_bh(&ptype_lock);
393 if (pt->type == htons(ETH_P_ALL))
396 head = &ptype_base[ntohs(pt->type) & 15];
398 list_for_each_entry(pt1, head, list) {
400 list_del_rcu(&pt->list);
405 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
407 spin_unlock_bh(&ptype_lock);
410 * dev_remove_pack - remove packet handler
411 * @pt: packet type declaration
413 * Remove a protocol handler that was previously added to the kernel
414 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
415 * from the kernel lists and can be freed or reused once this function
418 * This call sleeps to guarantee that no CPU is looking at the packet
421 void dev_remove_pack(struct packet_type *pt)
423 __dev_remove_pack(pt);
428 /******************************************************************************
430 Device Boot-time Settings Routines
432 *******************************************************************************/
434 /* Boot time configuration table */
435 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
438 * netdev_boot_setup_add - add new setup entry
439 * @name: name of the device
440 * @map: configured settings for the device
442 * Adds new setup entry to the dev_boot_setup list. The function
443 * returns 0 on error and 1 on success. This is a generic routine to
446 static int netdev_boot_setup_add(char *name, struct ifmap *map)
448 struct netdev_boot_setup *s;
452 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
453 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
454 memset(s[i].name, 0, sizeof(s[i].name));
455 strcpy(s[i].name, name);
456 memcpy(&s[i].map, map, sizeof(s[i].map));
461 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
465 * netdev_boot_setup_check - check boot time settings
466 * @dev: the netdevice
468 * Check boot time settings for the device.
469 * The found settings are set for the device to be used
470 * later in the device probing.
471 * Returns 0 if no settings found, 1 if they are.
473 int netdev_boot_setup_check(struct net_device *dev)
475 struct netdev_boot_setup *s = dev_boot_setup;
478 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
479 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
480 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
481 dev->irq = s[i].map.irq;
482 dev->base_addr = s[i].map.base_addr;
483 dev->mem_start = s[i].map.mem_start;
484 dev->mem_end = s[i].map.mem_end;
493 * netdev_boot_base - get address from boot time settings
494 * @prefix: prefix for network device
495 * @unit: id for network device
497 * Check boot time settings for the base address of device.
498 * The found settings are set for the device to be used
499 * later in the device probing.
500 * Returns 0 if no settings found.
502 unsigned long netdev_boot_base(const char *prefix, int unit)
504 const struct netdev_boot_setup *s = dev_boot_setup;
508 sprintf(name, "%s%d", prefix, unit);
511 * If device already registered then return base of 1
512 * to indicate not to probe for this interface
514 if (__dev_get_by_name(&init_net, name))
517 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
518 if (!strcmp(name, s[i].name))
519 return s[i].map.base_addr;
524 * Saves at boot time configured settings for any netdevice.
526 int __init netdev_boot_setup(char *str)
531 str = get_options(str, ARRAY_SIZE(ints), ints);
536 memset(&map, 0, sizeof(map));
540 map.base_addr = ints[2];
542 map.mem_start = ints[3];
544 map.mem_end = ints[4];
546 /* Add new entry to the list */
547 return netdev_boot_setup_add(str, &map);
550 __setup("netdev=", netdev_boot_setup);
552 /*******************************************************************************
554 Device Interface Subroutines
556 *******************************************************************************/
559 * __dev_get_by_name - find a device by its name
560 * @name: name to find
562 * Find an interface by name. Must be called under RTNL semaphore
563 * or @dev_base_lock. If the name is found a pointer to the device
564 * is returned. If the name is not found then %NULL is returned. The
565 * reference counters are not incremented so the caller must be
566 * careful with locks.
569 struct net_device *__dev_get_by_name(struct net *net, const char *name)
571 struct hlist_node *p;
573 hlist_for_each(p, dev_name_hash(net, name)) {
574 struct net_device *dev
575 = hlist_entry(p, struct net_device, name_hlist);
576 if (!strncmp(dev->name, name, IFNAMSIZ))
583 * dev_get_by_name - find a device by its name
584 * @name: name to find
586 * Find an interface by name. This can be called from any
587 * context and does its own locking. The returned handle has
588 * the usage count incremented and the caller must use dev_put() to
589 * release it when it is no longer needed. %NULL is returned if no
590 * matching device is found.
593 struct net_device *dev_get_by_name(struct net *net, const char *name)
595 struct net_device *dev;
597 read_lock(&dev_base_lock);
598 dev = __dev_get_by_name(net, name);
601 read_unlock(&dev_base_lock);
606 * __dev_get_by_index - find a device by its ifindex
607 * @ifindex: index of device
609 * Search for an interface by index. Returns %NULL if the device
610 * is not found or a pointer to the device. The device has not
611 * had its reference counter increased so the caller must be careful
612 * about locking. The caller must hold either the RTNL semaphore
616 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
618 struct hlist_node *p;
620 hlist_for_each(p, dev_index_hash(net, ifindex)) {
621 struct net_device *dev
622 = hlist_entry(p, struct net_device, index_hlist);
623 if (dev->ifindex == ifindex)
631 * dev_get_by_index - find a device by its ifindex
632 * @ifindex: index of device
634 * Search for an interface by index. Returns NULL if the device
635 * is not found or a pointer to the device. The device returned has
636 * had a reference added and the pointer is safe until the user calls
637 * dev_put to indicate they have finished with it.
640 struct net_device *dev_get_by_index(struct net *net, int ifindex)
642 struct net_device *dev;
644 read_lock(&dev_base_lock);
645 dev = __dev_get_by_index(net, ifindex);
648 read_unlock(&dev_base_lock);
653 * dev_getbyhwaddr - find a device by its hardware address
654 * @type: media type of device
655 * @ha: hardware address
657 * Search for an interface by MAC address. Returns NULL if the device
658 * is not found or a pointer to the device. The caller must hold the
659 * rtnl semaphore. The returned device has not had its ref count increased
660 * and the caller must therefore be careful about locking
663 * If the API was consistent this would be __dev_get_by_hwaddr
666 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
668 struct net_device *dev;
672 for_each_netdev(&init_net, dev)
673 if (dev->type == type &&
674 !memcmp(dev->dev_addr, ha, dev->addr_len))
680 EXPORT_SYMBOL(dev_getbyhwaddr);
682 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
684 struct net_device *dev;
687 for_each_netdev(net, dev)
688 if (dev->type == type)
694 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
696 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
698 struct net_device *dev;
701 dev = __dev_getfirstbyhwtype(net, type);
708 EXPORT_SYMBOL(dev_getfirstbyhwtype);
711 * dev_get_by_flags - find any device with given flags
712 * @if_flags: IFF_* values
713 * @mask: bitmask of bits in if_flags to check
715 * Search for any interface with the given flags. Returns NULL if a device
716 * is not found or a pointer to the device. The device returned has
717 * had a reference added and the pointer is safe until the user calls
718 * dev_put to indicate they have finished with it.
721 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
723 struct net_device *dev, *ret;
726 read_lock(&dev_base_lock);
727 for_each_netdev(net, dev) {
728 if (((dev->flags ^ if_flags) & mask) == 0) {
734 read_unlock(&dev_base_lock);
739 * dev_valid_name - check if name is okay for network device
742 * Network device names need to be valid file names to
743 * to allow sysfs to work. We also disallow any kind of
746 int dev_valid_name(const char *name)
750 if (strlen(name) >= IFNAMSIZ)
752 if (!strcmp(name, ".") || !strcmp(name, ".."))
756 if (*name == '/' || isspace(*name))
764 * __dev_alloc_name - allocate a name for a device
765 * @net: network namespace to allocate the device name in
766 * @name: name format string
767 * @buf: scratch buffer and result name string
769 * Passed a format string - eg "lt%d" it will try and find a suitable
770 * id. It scans list of devices to build up a free map, then chooses
771 * the first empty slot. The caller must hold the dev_base or rtnl lock
772 * while allocating the name and adding the device in order to avoid
774 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
775 * Returns the number of the unit assigned or a negative errno code.
778 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
782 const int max_netdevices = 8*PAGE_SIZE;
784 struct net_device *d;
786 p = strnchr(name, IFNAMSIZ-1, '%');
789 * Verify the string as this thing may have come from
790 * the user. There must be either one "%d" and no other "%"
793 if (p[1] != 'd' || strchr(p + 2, '%'))
796 /* Use one page as a bit array of possible slots */
797 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
801 for_each_netdev(net, d) {
802 if (!sscanf(d->name, name, &i))
804 if (i < 0 || i >= max_netdevices)
807 /* avoid cases where sscanf is not exact inverse of printf */
808 snprintf(buf, IFNAMSIZ, name, i);
809 if (!strncmp(buf, d->name, IFNAMSIZ))
813 i = find_first_zero_bit(inuse, max_netdevices);
814 free_page((unsigned long) inuse);
817 snprintf(buf, IFNAMSIZ, name, i);
818 if (!__dev_get_by_name(net, buf))
821 /* It is possible to run out of possible slots
822 * when the name is long and there isn't enough space left
823 * for the digits, or if all bits are used.
829 * dev_alloc_name - allocate a name for a device
831 * @name: name format string
833 * Passed a format string - eg "lt%d" it will try and find a suitable
834 * id. It scans list of devices to build up a free map, then chooses
835 * the first empty slot. The caller must hold the dev_base or rtnl lock
836 * while allocating the name and adding the device in order to avoid
838 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
839 * Returns the number of the unit assigned or a negative errno code.
842 int dev_alloc_name(struct net_device *dev, const char *name)
848 BUG_ON(!dev->nd_net);
850 ret = __dev_alloc_name(net, name, buf);
852 strlcpy(dev->name, buf, IFNAMSIZ);
858 * dev_change_name - change name of a device
860 * @newname: name (or format string) must be at least IFNAMSIZ
862 * Change name of a device, can pass format strings "eth%d".
865 int dev_change_name(struct net_device *dev, char *newname)
867 char oldname[IFNAMSIZ];
873 BUG_ON(!dev->nd_net);
876 if (dev->flags & IFF_UP)
879 if (!dev_valid_name(newname))
882 memcpy(oldname, dev->name, IFNAMSIZ);
884 if (strchr(newname, '%')) {
885 err = dev_alloc_name(dev, newname);
888 strcpy(newname, dev->name);
890 else if (__dev_get_by_name(net, newname))
893 strlcpy(dev->name, newname, IFNAMSIZ);
896 device_rename(&dev->dev, dev->name);
898 write_lock_bh(&dev_base_lock);
899 hlist_del(&dev->name_hlist);
900 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
901 write_unlock_bh(&dev_base_lock);
903 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
904 ret = notifier_to_errno(ret);
909 "%s: name change rollback failed: %d.\n",
913 memcpy(dev->name, oldname, IFNAMSIZ);
922 * netdev_features_change - device changes features
923 * @dev: device to cause notification
925 * Called to indicate a device has changed features.
927 void netdev_features_change(struct net_device *dev)
929 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
931 EXPORT_SYMBOL(netdev_features_change);
934 * netdev_state_change - device changes state
935 * @dev: device to cause notification
937 * Called to indicate a device has changed state. This function calls
938 * the notifier chains for netdev_chain and sends a NEWLINK message
939 * to the routing socket.
941 void netdev_state_change(struct net_device *dev)
943 if (dev->flags & IFF_UP) {
944 call_netdevice_notifiers(NETDEV_CHANGE, dev);
945 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
950 * dev_load - load a network module
951 * @name: name of interface
953 * If a network interface is not present and the process has suitable
954 * privileges this function loads the module. If module loading is not
955 * available in this kernel then it becomes a nop.
958 void dev_load(struct net *net, const char *name)
960 struct net_device *dev;
962 read_lock(&dev_base_lock);
963 dev = __dev_get_by_name(net, name);
964 read_unlock(&dev_base_lock);
966 if (!dev && capable(CAP_SYS_MODULE))
967 request_module("%s", name);
970 static int default_rebuild_header(struct sk_buff *skb)
972 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
973 skb->dev ? skb->dev->name : "NULL!!!");
979 * dev_open - prepare an interface for use.
980 * @dev: device to open
982 * Takes a device from down to up state. The device's private open
983 * function is invoked and then the multicast lists are loaded. Finally
984 * the device is moved into the up state and a %NETDEV_UP message is
985 * sent to the netdev notifier chain.
987 * Calling this function on an active interface is a nop. On a failure
988 * a negative errno code is returned.
990 int dev_open(struct net_device *dev)
998 if (dev->flags & IFF_UP)
1002 * Is it even present?
1004 if (!netif_device_present(dev))
1008 * Call device private open method
1010 set_bit(__LINK_STATE_START, &dev->state);
1012 ret = dev->open(dev);
1014 clear_bit(__LINK_STATE_START, &dev->state);
1018 * If it went open OK then:
1025 dev->flags |= IFF_UP;
1028 * Initialize multicasting status
1030 dev_set_rx_mode(dev);
1033 * Wakeup transmit queue engine
1038 * ... and announce new interface.
1040 call_netdevice_notifiers(NETDEV_UP, dev);
1046 * dev_close - shutdown an interface.
1047 * @dev: device to shutdown
1049 * This function moves an active device into down state. A
1050 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1051 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1054 int dev_close(struct net_device *dev)
1058 if (!(dev->flags & IFF_UP))
1062 * Tell people we are going down, so that they can
1063 * prepare to death, when device is still operating.
1065 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1067 dev_deactivate(dev);
1069 clear_bit(__LINK_STATE_START, &dev->state);
1071 /* Synchronize to scheduled poll. We cannot touch poll list,
1072 * it can be even on different cpu. So just clear netif_running().
1074 * dev->stop() will invoke napi_disable() on all of it's
1075 * napi_struct instances on this device.
1077 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1080 * Call the device specific close. This cannot fail.
1081 * Only if device is UP
1083 * We allow it to be called even after a DETACH hot-plug
1090 * Device is now down.
1093 dev->flags &= ~IFF_UP;
1096 * Tell people we are down
1098 call_netdevice_notifiers(NETDEV_DOWN, dev);
1104 static int dev_boot_phase = 1;
1107 * Device change register/unregister. These are not inline or static
1108 * as we export them to the world.
1112 * register_netdevice_notifier - register a network notifier block
1115 * Register a notifier to be called when network device events occur.
1116 * The notifier passed is linked into the kernel structures and must
1117 * not be reused until it has been unregistered. A negative errno code
1118 * is returned on a failure.
1120 * When registered all registration and up events are replayed
1121 * to the new notifier to allow device to have a race free
1122 * view of the network device list.
1125 int register_netdevice_notifier(struct notifier_block *nb)
1127 struct net_device *dev;
1128 struct net_device *last;
1133 err = raw_notifier_chain_register(&netdev_chain, nb);
1139 for_each_netdev(net, dev) {
1140 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1141 err = notifier_to_errno(err);
1145 if (!(dev->flags & IFF_UP))
1148 nb->notifier_call(nb, NETDEV_UP, dev);
1159 for_each_netdev(net, dev) {
1163 if (dev->flags & IFF_UP) {
1164 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1165 nb->notifier_call(nb, NETDEV_DOWN, dev);
1167 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1174 * unregister_netdevice_notifier - unregister a network notifier block
1177 * Unregister a notifier previously registered by
1178 * register_netdevice_notifier(). The notifier is unlinked into the
1179 * kernel structures and may then be reused. A negative errno code
1180 * is returned on a failure.
1183 int unregister_netdevice_notifier(struct notifier_block *nb)
1188 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1194 * call_netdevice_notifiers - call all network notifier blocks
1195 * @val: value passed unmodified to notifier function
1196 * @v: pointer passed unmodified to notifier function
1198 * Call all network notifier blocks. Parameters and return value
1199 * are as for raw_notifier_call_chain().
1202 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1204 return raw_notifier_call_chain(&netdev_chain, val, dev);
1207 /* When > 0 there are consumers of rx skb time stamps */
1208 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1210 void net_enable_timestamp(void)
1212 atomic_inc(&netstamp_needed);
1215 void net_disable_timestamp(void)
1217 atomic_dec(&netstamp_needed);
1220 static inline void net_timestamp(struct sk_buff *skb)
1222 if (atomic_read(&netstamp_needed))
1223 __net_timestamp(skb);
1225 skb->tstamp.tv64 = 0;
1229 * Support routine. Sends outgoing frames to any network
1230 * taps currently in use.
1233 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1235 struct packet_type *ptype;
1240 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1241 /* Never send packets back to the socket
1242 * they originated from - MvS (miquels@drinkel.ow.org)
1244 if ((ptype->dev == dev || !ptype->dev) &&
1245 (ptype->af_packet_priv == NULL ||
1246 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1247 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1251 /* skb->nh should be correctly
1252 set by sender, so that the second statement is
1253 just protection against buggy protocols.
1255 skb_reset_mac_header(skb2);
1257 if (skb_network_header(skb2) < skb2->data ||
1258 skb2->network_header > skb2->tail) {
1259 if (net_ratelimit())
1260 printk(KERN_CRIT "protocol %04x is "
1262 skb2->protocol, dev->name);
1263 skb_reset_network_header(skb2);
1266 skb2->transport_header = skb2->network_header;
1267 skb2->pkt_type = PACKET_OUTGOING;
1268 ptype->func(skb2, skb->dev, ptype, skb->dev);
1275 void __netif_schedule(struct net_device *dev)
1277 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1278 unsigned long flags;
1279 struct softnet_data *sd;
1281 local_irq_save(flags);
1282 sd = &__get_cpu_var(softnet_data);
1283 dev->next_sched = sd->output_queue;
1284 sd->output_queue = dev;
1285 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1286 local_irq_restore(flags);
1289 EXPORT_SYMBOL(__netif_schedule);
1291 void dev_kfree_skb_irq(struct sk_buff *skb)
1293 if (atomic_dec_and_test(&skb->users)) {
1294 struct softnet_data *sd;
1295 unsigned long flags;
1297 local_irq_save(flags);
1298 sd = &__get_cpu_var(softnet_data);
1299 skb->next = sd->completion_queue;
1300 sd->completion_queue = skb;
1301 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1302 local_irq_restore(flags);
1305 EXPORT_SYMBOL(dev_kfree_skb_irq);
1307 void dev_kfree_skb_any(struct sk_buff *skb)
1309 if (in_irq() || irqs_disabled())
1310 dev_kfree_skb_irq(skb);
1314 EXPORT_SYMBOL(dev_kfree_skb_any);
1318 * netif_device_detach - mark device as removed
1319 * @dev: network device
1321 * Mark device as removed from system and therefore no longer available.
1323 void netif_device_detach(struct net_device *dev)
1325 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1326 netif_running(dev)) {
1327 netif_stop_queue(dev);
1330 EXPORT_SYMBOL(netif_device_detach);
1333 * netif_device_attach - mark device as attached
1334 * @dev: network device
1336 * Mark device as attached from system and restart if needed.
1338 void netif_device_attach(struct net_device *dev)
1340 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1341 netif_running(dev)) {
1342 netif_wake_queue(dev);
1343 __netdev_watchdog_up(dev);
1346 EXPORT_SYMBOL(netif_device_attach);
1350 * Invalidate hardware checksum when packet is to be mangled, and
1351 * complete checksum manually on outgoing path.
1353 int skb_checksum_help(struct sk_buff *skb)
1356 int ret = 0, offset;
1358 if (skb->ip_summed == CHECKSUM_COMPLETE)
1359 goto out_set_summed;
1361 if (unlikely(skb_shinfo(skb)->gso_size)) {
1362 /* Let GSO fix up the checksum. */
1363 goto out_set_summed;
1366 if (skb_cloned(skb)) {
1367 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1372 offset = skb->csum_start - skb_headroom(skb);
1373 BUG_ON(offset > (int)skb->len);
1374 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1376 offset = skb_headlen(skb) - offset;
1377 BUG_ON(offset <= 0);
1378 BUG_ON(skb->csum_offset + 2 > offset);
1380 *(__sum16 *)(skb->head + skb->csum_start + skb->csum_offset) =
1383 skb->ip_summed = CHECKSUM_NONE;
1389 * skb_gso_segment - Perform segmentation on skb.
1390 * @skb: buffer to segment
1391 * @features: features for the output path (see dev->features)
1393 * This function segments the given skb and returns a list of segments.
1395 * It may return NULL if the skb requires no segmentation. This is
1396 * only possible when GSO is used for verifying header integrity.
1398 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1400 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1401 struct packet_type *ptype;
1402 __be16 type = skb->protocol;
1405 BUG_ON(skb_shinfo(skb)->frag_list);
1407 skb_reset_mac_header(skb);
1408 skb->mac_len = skb->network_header - skb->mac_header;
1409 __skb_pull(skb, skb->mac_len);
1411 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1412 if (skb_header_cloned(skb) &&
1413 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1414 return ERR_PTR(err);
1418 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1419 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1420 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1421 err = ptype->gso_send_check(skb);
1422 segs = ERR_PTR(err);
1423 if (err || skb_gso_ok(skb, features))
1425 __skb_push(skb, (skb->data -
1426 skb_network_header(skb)));
1428 segs = ptype->gso_segment(skb, features);
1434 __skb_push(skb, skb->data - skb_mac_header(skb));
1439 EXPORT_SYMBOL(skb_gso_segment);
1441 /* Take action when hardware reception checksum errors are detected. */
1443 void netdev_rx_csum_fault(struct net_device *dev)
1445 if (net_ratelimit()) {
1446 printk(KERN_ERR "%s: hw csum failure.\n",
1447 dev ? dev->name : "<unknown>");
1451 EXPORT_SYMBOL(netdev_rx_csum_fault);
1454 /* Actually, we should eliminate this check as soon as we know, that:
1455 * 1. IOMMU is present and allows to map all the memory.
1456 * 2. No high memory really exists on this machine.
1459 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1461 #ifdef CONFIG_HIGHMEM
1464 if (dev->features & NETIF_F_HIGHDMA)
1467 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1468 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1476 void (*destructor)(struct sk_buff *skb);
1479 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1481 static void dev_gso_skb_destructor(struct sk_buff *skb)
1483 struct dev_gso_cb *cb;
1486 struct sk_buff *nskb = skb->next;
1488 skb->next = nskb->next;
1491 } while (skb->next);
1493 cb = DEV_GSO_CB(skb);
1495 cb->destructor(skb);
1499 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1500 * @skb: buffer to segment
1502 * This function segments the given skb and stores the list of segments
1505 static int dev_gso_segment(struct sk_buff *skb)
1507 struct net_device *dev = skb->dev;
1508 struct sk_buff *segs;
1509 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1512 segs = skb_gso_segment(skb, features);
1514 /* Verifying header integrity only. */
1518 if (unlikely(IS_ERR(segs)))
1519 return PTR_ERR(segs);
1522 DEV_GSO_CB(skb)->destructor = skb->destructor;
1523 skb->destructor = dev_gso_skb_destructor;
1528 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1530 if (likely(!skb->next)) {
1531 if (!list_empty(&ptype_all))
1532 dev_queue_xmit_nit(skb, dev);
1534 if (netif_needs_gso(dev, skb)) {
1535 if (unlikely(dev_gso_segment(skb)))
1541 return dev->hard_start_xmit(skb, dev);
1546 struct sk_buff *nskb = skb->next;
1549 skb->next = nskb->next;
1551 rc = dev->hard_start_xmit(nskb, dev);
1553 nskb->next = skb->next;
1557 if (unlikely((netif_queue_stopped(dev) ||
1558 netif_subqueue_stopped(dev, skb->queue_mapping)) &&
1560 return NETDEV_TX_BUSY;
1561 } while (skb->next);
1563 skb->destructor = DEV_GSO_CB(skb)->destructor;
1571 * dev_queue_xmit - transmit a buffer
1572 * @skb: buffer to transmit
1574 * Queue a buffer for transmission to a network device. The caller must
1575 * have set the device and priority and built the buffer before calling
1576 * this function. The function can be called from an interrupt.
1578 * A negative errno code is returned on a failure. A success does not
1579 * guarantee the frame will be transmitted as it may be dropped due
1580 * to congestion or traffic shaping.
1582 * -----------------------------------------------------------------------------------
1583 * I notice this method can also return errors from the queue disciplines,
1584 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1587 * Regardless of the return value, the skb is consumed, so it is currently
1588 * difficult to retry a send to this method. (You can bump the ref count
1589 * before sending to hold a reference for retry if you are careful.)
1591 * When calling this method, interrupts MUST be enabled. This is because
1592 * the BH enable code must have IRQs enabled so that it will not deadlock.
1596 int dev_queue_xmit(struct sk_buff *skb)
1598 struct net_device *dev = skb->dev;
1602 /* GSO will handle the following emulations directly. */
1603 if (netif_needs_gso(dev, skb))
1606 if (skb_shinfo(skb)->frag_list &&
1607 !(dev->features & NETIF_F_FRAGLIST) &&
1608 __skb_linearize(skb))
1611 /* Fragmented skb is linearized if device does not support SG,
1612 * or if at least one of fragments is in highmem and device
1613 * does not support DMA from it.
1615 if (skb_shinfo(skb)->nr_frags &&
1616 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1617 __skb_linearize(skb))
1620 /* If packet is not checksummed and device does not support
1621 * checksumming for this protocol, complete checksumming here.
1623 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1624 skb_set_transport_header(skb, skb->csum_start -
1627 if (!(dev->features & NETIF_F_GEN_CSUM) &&
1628 !((dev->features & NETIF_F_IP_CSUM) &&
1629 skb->protocol == htons(ETH_P_IP)) &&
1630 !((dev->features & NETIF_F_IPV6_CSUM) &&
1631 skb->protocol == htons(ETH_P_IPV6)))
1632 if (skb_checksum_help(skb))
1637 spin_lock_prefetch(&dev->queue_lock);
1639 /* Disable soft irqs for various locks below. Also
1640 * stops preemption for RCU.
1644 /* Updates of qdisc are serialized by queue_lock.
1645 * The struct Qdisc which is pointed to by qdisc is now a
1646 * rcu structure - it may be accessed without acquiring
1647 * a lock (but the structure may be stale.) The freeing of the
1648 * qdisc will be deferred until it's known that there are no
1649 * more references to it.
1651 * If the qdisc has an enqueue function, we still need to
1652 * hold the queue_lock before calling it, since queue_lock
1653 * also serializes access to the device queue.
1656 q = rcu_dereference(dev->qdisc);
1657 #ifdef CONFIG_NET_CLS_ACT
1658 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1661 /* Grab device queue */
1662 spin_lock(&dev->queue_lock);
1665 /* reset queue_mapping to zero */
1666 skb->queue_mapping = 0;
1667 rc = q->enqueue(skb, q);
1669 spin_unlock(&dev->queue_lock);
1671 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1674 spin_unlock(&dev->queue_lock);
1677 /* The device has no queue. Common case for software devices:
1678 loopback, all the sorts of tunnels...
1680 Really, it is unlikely that netif_tx_lock protection is necessary
1681 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1683 However, it is possible, that they rely on protection
1686 Check this and shot the lock. It is not prone from deadlocks.
1687 Either shot noqueue qdisc, it is even simpler 8)
1689 if (dev->flags & IFF_UP) {
1690 int cpu = smp_processor_id(); /* ok because BHs are off */
1692 if (dev->xmit_lock_owner != cpu) {
1694 HARD_TX_LOCK(dev, cpu);
1696 if (!netif_queue_stopped(dev) &&
1697 !netif_subqueue_stopped(dev, skb->queue_mapping)) {
1699 if (!dev_hard_start_xmit(skb, dev)) {
1700 HARD_TX_UNLOCK(dev);
1704 HARD_TX_UNLOCK(dev);
1705 if (net_ratelimit())
1706 printk(KERN_CRIT "Virtual device %s asks to "
1707 "queue packet!\n", dev->name);
1709 /* Recursion is detected! It is possible,
1711 if (net_ratelimit())
1712 printk(KERN_CRIT "Dead loop on virtual device "
1713 "%s, fix it urgently!\n", dev->name);
1718 rcu_read_unlock_bh();
1724 rcu_read_unlock_bh();
1729 /*=======================================================================
1731 =======================================================================*/
1733 int netdev_max_backlog __read_mostly = 1000;
1734 int netdev_budget __read_mostly = 300;
1735 int weight_p __read_mostly = 64; /* old backlog weight */
1737 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1741 * netif_rx - post buffer to the network code
1742 * @skb: buffer to post
1744 * This function receives a packet from a device driver and queues it for
1745 * the upper (protocol) levels to process. It always succeeds. The buffer
1746 * may be dropped during processing for congestion control or by the
1750 * NET_RX_SUCCESS (no congestion)
1751 * NET_RX_CN_LOW (low congestion)
1752 * NET_RX_CN_MOD (moderate congestion)
1753 * NET_RX_CN_HIGH (high congestion)
1754 * NET_RX_DROP (packet was dropped)
1758 int netif_rx(struct sk_buff *skb)
1760 struct softnet_data *queue;
1761 unsigned long flags;
1763 /* if netpoll wants it, pretend we never saw it */
1764 if (netpoll_rx(skb))
1767 if (!skb->tstamp.tv64)
1771 * The code is rearranged so that the path is the most
1772 * short when CPU is congested, but is still operating.
1774 local_irq_save(flags);
1775 queue = &__get_cpu_var(softnet_data);
1777 __get_cpu_var(netdev_rx_stat).total++;
1778 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1779 if (queue->input_pkt_queue.qlen) {
1782 __skb_queue_tail(&queue->input_pkt_queue, skb);
1783 local_irq_restore(flags);
1784 return NET_RX_SUCCESS;
1787 napi_schedule(&queue->backlog);
1791 __get_cpu_var(netdev_rx_stat).dropped++;
1792 local_irq_restore(flags);
1798 int netif_rx_ni(struct sk_buff *skb)
1803 err = netif_rx(skb);
1804 if (local_softirq_pending())
1811 EXPORT_SYMBOL(netif_rx_ni);
1813 static inline struct net_device *skb_bond(struct sk_buff *skb)
1815 struct net_device *dev = skb->dev;
1818 if (skb_bond_should_drop(skb)) {
1822 skb->dev = dev->master;
1829 static void net_tx_action(struct softirq_action *h)
1831 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1833 if (sd->completion_queue) {
1834 struct sk_buff *clist;
1836 local_irq_disable();
1837 clist = sd->completion_queue;
1838 sd->completion_queue = NULL;
1842 struct sk_buff *skb = clist;
1843 clist = clist->next;
1845 BUG_TRAP(!atomic_read(&skb->users));
1850 if (sd->output_queue) {
1851 struct net_device *head;
1853 local_irq_disable();
1854 head = sd->output_queue;
1855 sd->output_queue = NULL;
1859 struct net_device *dev = head;
1860 head = head->next_sched;
1862 smp_mb__before_clear_bit();
1863 clear_bit(__LINK_STATE_SCHED, &dev->state);
1865 if (spin_trylock(&dev->queue_lock)) {
1867 spin_unlock(&dev->queue_lock);
1869 netif_schedule(dev);
1875 static inline int deliver_skb(struct sk_buff *skb,
1876 struct packet_type *pt_prev,
1877 struct net_device *orig_dev)
1879 atomic_inc(&skb->users);
1880 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1883 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1884 /* These hooks defined here for ATM */
1886 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1887 unsigned char *addr);
1888 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1891 * If bridge module is loaded call bridging hook.
1892 * returns NULL if packet was consumed.
1894 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
1895 struct sk_buff *skb) __read_mostly;
1896 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
1897 struct packet_type **pt_prev, int *ret,
1898 struct net_device *orig_dev)
1900 struct net_bridge_port *port;
1902 if (skb->pkt_type == PACKET_LOOPBACK ||
1903 (port = rcu_dereference(skb->dev->br_port)) == NULL)
1907 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1911 return br_handle_frame_hook(port, skb);
1914 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1917 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1918 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
1919 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
1921 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
1922 struct packet_type **pt_prev,
1924 struct net_device *orig_dev)
1926 if (skb->dev->macvlan_port == NULL)
1930 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1933 return macvlan_handle_frame_hook(skb);
1936 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1939 #ifdef CONFIG_NET_CLS_ACT
1940 /* TODO: Maybe we should just force sch_ingress to be compiled in
1941 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1942 * a compare and 2 stores extra right now if we dont have it on
1943 * but have CONFIG_NET_CLS_ACT
1944 * NOTE: This doesnt stop any functionality; if you dont have
1945 * the ingress scheduler, you just cant add policies on ingress.
1948 static int ing_filter(struct sk_buff *skb)
1951 struct net_device *dev = skb->dev;
1952 int result = TC_ACT_OK;
1954 if (dev->qdisc_ingress) {
1955 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1956 if (MAX_RED_LOOP < ttl++) {
1957 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1958 skb->iif, skb->dev->ifindex);
1962 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1964 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1966 spin_lock(&dev->ingress_lock);
1967 if ((q = dev->qdisc_ingress) != NULL)
1968 result = q->enqueue(skb, q);
1969 spin_unlock(&dev->ingress_lock);
1977 int netif_receive_skb(struct sk_buff *skb)
1979 struct packet_type *ptype, *pt_prev;
1980 struct net_device *orig_dev;
1981 int ret = NET_RX_DROP;
1984 /* if we've gotten here through NAPI, check netpoll */
1985 if (netpoll_receive_skb(skb))
1988 if (!skb->tstamp.tv64)
1992 skb->iif = skb->dev->ifindex;
1994 orig_dev = skb_bond(skb);
1999 __get_cpu_var(netdev_rx_stat).total++;
2001 skb_reset_network_header(skb);
2002 skb_reset_transport_header(skb);
2003 skb->mac_len = skb->network_header - skb->mac_header;
2009 #ifdef CONFIG_NET_CLS_ACT
2010 if (skb->tc_verd & TC_NCLS) {
2011 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2016 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2017 if (!ptype->dev || ptype->dev == skb->dev) {
2019 ret = deliver_skb(skb, pt_prev, orig_dev);
2024 #ifdef CONFIG_NET_CLS_ACT
2026 ret = deliver_skb(skb, pt_prev, orig_dev);
2027 pt_prev = NULL; /* noone else should process this after*/
2029 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2032 ret = ing_filter(skb);
2034 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
2043 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2046 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2050 type = skb->protocol;
2051 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
2052 if (ptype->type == type &&
2053 (!ptype->dev || ptype->dev == skb->dev)) {
2055 ret = deliver_skb(skb, pt_prev, orig_dev);
2061 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2064 /* Jamal, now you will not able to escape explaining
2065 * me how you were going to use this. :-)
2075 static int process_backlog(struct napi_struct *napi, int quota)
2078 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2079 unsigned long start_time = jiffies;
2081 napi->weight = weight_p;
2083 struct sk_buff *skb;
2084 struct net_device *dev;
2086 local_irq_disable();
2087 skb = __skb_dequeue(&queue->input_pkt_queue);
2089 __napi_complete(napi);
2098 netif_receive_skb(skb);
2101 } while (++work < quota && jiffies == start_time);
2107 * __napi_schedule - schedule for receive
2108 * @napi: entry to schedule
2110 * The entry's receive function will be scheduled to run
2112 void fastcall __napi_schedule(struct napi_struct *n)
2114 unsigned long flags;
2116 local_irq_save(flags);
2117 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2118 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2119 local_irq_restore(flags);
2121 EXPORT_SYMBOL(__napi_schedule);
2124 static void net_rx_action(struct softirq_action *h)
2126 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2127 unsigned long start_time = jiffies;
2128 int budget = netdev_budget;
2131 local_irq_disable();
2133 while (!list_empty(list)) {
2134 struct napi_struct *n;
2137 /* If softirq window is exhuasted then punt.
2139 * Note that this is a slight policy change from the
2140 * previous NAPI code, which would allow up to 2
2141 * jiffies to pass before breaking out. The test
2142 * used to be "jiffies - start_time > 1".
2144 if (unlikely(budget <= 0 || jiffies != start_time))
2149 /* Even though interrupts have been re-enabled, this
2150 * access is safe because interrupts can only add new
2151 * entries to the tail of this list, and only ->poll()
2152 * calls can remove this head entry from the list.
2154 n = list_entry(list->next, struct napi_struct, poll_list);
2156 have = netpoll_poll_lock(n);
2160 work = n->poll(n, weight);
2162 WARN_ON_ONCE(work > weight);
2166 local_irq_disable();
2168 /* Drivers must not modify the NAPI state if they
2169 * consume the entire weight. In such cases this code
2170 * still "owns" the NAPI instance and therefore can
2171 * move the instance around on the list at-will.
2173 if (unlikely(work == weight))
2174 list_move_tail(&n->poll_list, list);
2176 netpoll_poll_unlock(have);
2181 #ifdef CONFIG_NET_DMA
2183 * There may not be any more sk_buffs coming right now, so push
2184 * any pending DMA copies to hardware
2186 if (!cpus_empty(net_dma.channel_mask)) {
2188 for_each_cpu_mask(chan_idx, net_dma.channel_mask) {
2189 struct dma_chan *chan = net_dma.channels[chan_idx];
2191 dma_async_memcpy_issue_pending(chan);
2199 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2200 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2204 static gifconf_func_t * gifconf_list [NPROTO];
2207 * register_gifconf - register a SIOCGIF handler
2208 * @family: Address family
2209 * @gifconf: Function handler
2211 * Register protocol dependent address dumping routines. The handler
2212 * that is passed must not be freed or reused until it has been replaced
2213 * by another handler.
2215 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2217 if (family >= NPROTO)
2219 gifconf_list[family] = gifconf;
2225 * Map an interface index to its name (SIOCGIFNAME)
2229 * We need this ioctl for efficient implementation of the
2230 * if_indextoname() function required by the IPv6 API. Without
2231 * it, we would have to search all the interfaces to find a
2235 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2237 struct net_device *dev;
2241 * Fetch the caller's info block.
2244 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2247 read_lock(&dev_base_lock);
2248 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2250 read_unlock(&dev_base_lock);
2254 strcpy(ifr.ifr_name, dev->name);
2255 read_unlock(&dev_base_lock);
2257 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2263 * Perform a SIOCGIFCONF call. This structure will change
2264 * size eventually, and there is nothing I can do about it.
2265 * Thus we will need a 'compatibility mode'.
2268 static int dev_ifconf(struct net *net, char __user *arg)
2271 struct net_device *dev;
2278 * Fetch the caller's info block.
2281 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2288 * Loop over the interfaces, and write an info block for each.
2292 for_each_netdev(net, dev) {
2293 for (i = 0; i < NPROTO; i++) {
2294 if (gifconf_list[i]) {
2297 done = gifconf_list[i](dev, NULL, 0);
2299 done = gifconf_list[i](dev, pos + total,
2309 * All done. Write the updated control block back to the caller.
2311 ifc.ifc_len = total;
2314 * Both BSD and Solaris return 0 here, so we do too.
2316 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2319 #ifdef CONFIG_PROC_FS
2321 * This is invoked by the /proc filesystem handler to display a device
2324 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2326 struct net *net = seq->private;
2328 struct net_device *dev;
2330 read_lock(&dev_base_lock);
2332 return SEQ_START_TOKEN;
2335 for_each_netdev(net, dev)
2342 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2344 struct net *net = seq->private;
2346 return v == SEQ_START_TOKEN ?
2347 first_net_device(net) : next_net_device((struct net_device *)v);
2350 void dev_seq_stop(struct seq_file *seq, void *v)
2352 read_unlock(&dev_base_lock);
2355 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2357 struct net_device_stats *stats = dev->get_stats(dev);
2359 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2360 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2361 dev->name, stats->rx_bytes, stats->rx_packets,
2363 stats->rx_dropped + stats->rx_missed_errors,
2364 stats->rx_fifo_errors,
2365 stats->rx_length_errors + stats->rx_over_errors +
2366 stats->rx_crc_errors + stats->rx_frame_errors,
2367 stats->rx_compressed, stats->multicast,
2368 stats->tx_bytes, stats->tx_packets,
2369 stats->tx_errors, stats->tx_dropped,
2370 stats->tx_fifo_errors, stats->collisions,
2371 stats->tx_carrier_errors +
2372 stats->tx_aborted_errors +
2373 stats->tx_window_errors +
2374 stats->tx_heartbeat_errors,
2375 stats->tx_compressed);
2379 * Called from the PROCfs module. This now uses the new arbitrary sized
2380 * /proc/net interface to create /proc/net/dev
2382 static int dev_seq_show(struct seq_file *seq, void *v)
2384 if (v == SEQ_START_TOKEN)
2385 seq_puts(seq, "Inter-| Receive "
2387 " face |bytes packets errs drop fifo frame "
2388 "compressed multicast|bytes packets errs "
2389 "drop fifo colls carrier compressed\n");
2391 dev_seq_printf_stats(seq, v);
2395 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2397 struct netif_rx_stats *rc = NULL;
2399 while (*pos < NR_CPUS)
2400 if (cpu_online(*pos)) {
2401 rc = &per_cpu(netdev_rx_stat, *pos);
2408 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2410 return softnet_get_online(pos);
2413 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2416 return softnet_get_online(pos);
2419 static void softnet_seq_stop(struct seq_file *seq, void *v)
2423 static int softnet_seq_show(struct seq_file *seq, void *v)
2425 struct netif_rx_stats *s = v;
2427 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2428 s->total, s->dropped, s->time_squeeze, 0,
2429 0, 0, 0, 0, /* was fastroute */
2434 static const struct seq_operations dev_seq_ops = {
2435 .start = dev_seq_start,
2436 .next = dev_seq_next,
2437 .stop = dev_seq_stop,
2438 .show = dev_seq_show,
2441 static int dev_seq_open(struct inode *inode, struct file *file)
2443 struct seq_file *seq;
2445 res = seq_open(file, &dev_seq_ops);
2447 seq = file->private_data;
2448 seq->private = get_proc_net(inode);
2449 if (!seq->private) {
2450 seq_release(inode, file);
2457 static int dev_seq_release(struct inode *inode, struct file *file)
2459 struct seq_file *seq = file->private_data;
2460 struct net *net = seq->private;
2462 return seq_release(inode, file);
2465 static const struct file_operations dev_seq_fops = {
2466 .owner = THIS_MODULE,
2467 .open = dev_seq_open,
2469 .llseek = seq_lseek,
2470 .release = dev_seq_release,
2473 static const struct seq_operations softnet_seq_ops = {
2474 .start = softnet_seq_start,
2475 .next = softnet_seq_next,
2476 .stop = softnet_seq_stop,
2477 .show = softnet_seq_show,
2480 static int softnet_seq_open(struct inode *inode, struct file *file)
2482 return seq_open(file, &softnet_seq_ops);
2485 static const struct file_operations softnet_seq_fops = {
2486 .owner = THIS_MODULE,
2487 .open = softnet_seq_open,
2489 .llseek = seq_lseek,
2490 .release = seq_release,
2493 static void *ptype_get_idx(loff_t pos)
2495 struct packet_type *pt = NULL;
2499 list_for_each_entry_rcu(pt, &ptype_all, list) {
2505 for (t = 0; t < 16; t++) {
2506 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2515 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2518 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2521 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2523 struct packet_type *pt;
2524 struct list_head *nxt;
2528 if (v == SEQ_START_TOKEN)
2529 return ptype_get_idx(0);
2532 nxt = pt->list.next;
2533 if (pt->type == htons(ETH_P_ALL)) {
2534 if (nxt != &ptype_all)
2537 nxt = ptype_base[0].next;
2539 hash = ntohs(pt->type) & 15;
2541 while (nxt == &ptype_base[hash]) {
2544 nxt = ptype_base[hash].next;
2547 return list_entry(nxt, struct packet_type, list);
2550 static void ptype_seq_stop(struct seq_file *seq, void *v)
2555 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2557 #ifdef CONFIG_KALLSYMS
2558 unsigned long offset = 0, symsize;
2559 const char *symname;
2563 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2570 modname = delim = "";
2571 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2577 seq_printf(seq, "[%p]", sym);
2580 static int ptype_seq_show(struct seq_file *seq, void *v)
2582 struct packet_type *pt = v;
2584 if (v == SEQ_START_TOKEN)
2585 seq_puts(seq, "Type Device Function\n");
2587 if (pt->type == htons(ETH_P_ALL))
2588 seq_puts(seq, "ALL ");
2590 seq_printf(seq, "%04x", ntohs(pt->type));
2592 seq_printf(seq, " %-8s ",
2593 pt->dev ? pt->dev->name : "");
2594 ptype_seq_decode(seq, pt->func);
2595 seq_putc(seq, '\n');
2601 static const struct seq_operations ptype_seq_ops = {
2602 .start = ptype_seq_start,
2603 .next = ptype_seq_next,
2604 .stop = ptype_seq_stop,
2605 .show = ptype_seq_show,
2608 static int ptype_seq_open(struct inode *inode, struct file *file)
2610 return seq_open(file, &ptype_seq_ops);
2613 static const struct file_operations ptype_seq_fops = {
2614 .owner = THIS_MODULE,
2615 .open = ptype_seq_open,
2617 .llseek = seq_lseek,
2618 .release = seq_release,
2622 static int dev_proc_net_init(struct net *net)
2626 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2628 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2630 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2633 if (wext_proc_init(net))
2639 proc_net_remove(net, "ptype");
2641 proc_net_remove(net, "softnet_stat");
2643 proc_net_remove(net, "dev");
2647 static void dev_proc_net_exit(struct net *net)
2649 wext_proc_exit(net);
2651 proc_net_remove(net, "ptype");
2652 proc_net_remove(net, "softnet_stat");
2653 proc_net_remove(net, "dev");
2656 static struct pernet_operations dev_proc_ops = {
2657 .init = dev_proc_net_init,
2658 .exit = dev_proc_net_exit,
2661 static int __init dev_proc_init(void)
2663 return register_pernet_subsys(&dev_proc_ops);
2666 #define dev_proc_init() 0
2667 #endif /* CONFIG_PROC_FS */
2671 * netdev_set_master - set up master/slave pair
2672 * @slave: slave device
2673 * @master: new master device
2675 * Changes the master device of the slave. Pass %NULL to break the
2676 * bonding. The caller must hold the RTNL semaphore. On a failure
2677 * a negative errno code is returned. On success the reference counts
2678 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2679 * function returns zero.
2681 int netdev_set_master(struct net_device *slave, struct net_device *master)
2683 struct net_device *old = slave->master;
2693 slave->master = master;
2701 slave->flags |= IFF_SLAVE;
2703 slave->flags &= ~IFF_SLAVE;
2705 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2709 static void __dev_set_promiscuity(struct net_device *dev, int inc)
2711 unsigned short old_flags = dev->flags;
2715 if ((dev->promiscuity += inc) == 0)
2716 dev->flags &= ~IFF_PROMISC;
2718 dev->flags |= IFF_PROMISC;
2719 if (dev->flags != old_flags) {
2720 printk(KERN_INFO "device %s %s promiscuous mode\n",
2721 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2723 audit_log(current->audit_context, GFP_ATOMIC,
2724 AUDIT_ANOM_PROMISCUOUS,
2725 "dev=%s prom=%d old_prom=%d auid=%u",
2726 dev->name, (dev->flags & IFF_PROMISC),
2727 (old_flags & IFF_PROMISC),
2728 audit_get_loginuid(current->audit_context));
2730 if (dev->change_rx_flags)
2731 dev->change_rx_flags(dev, IFF_PROMISC);
2736 * dev_set_promiscuity - update promiscuity count on a device
2740 * Add or remove promiscuity from a device. While the count in the device
2741 * remains above zero the interface remains promiscuous. Once it hits zero
2742 * the device reverts back to normal filtering operation. A negative inc
2743 * value is used to drop promiscuity on the device.
2745 void dev_set_promiscuity(struct net_device *dev, int inc)
2747 unsigned short old_flags = dev->flags;
2749 __dev_set_promiscuity(dev, inc);
2750 if (dev->flags != old_flags)
2751 dev_set_rx_mode(dev);
2755 * dev_set_allmulti - update allmulti count on a device
2759 * Add or remove reception of all multicast frames to a device. While the
2760 * count in the device remains above zero the interface remains listening
2761 * to all interfaces. Once it hits zero the device reverts back to normal
2762 * filtering operation. A negative @inc value is used to drop the counter
2763 * when releasing a resource needing all multicasts.
2766 void dev_set_allmulti(struct net_device *dev, int inc)
2768 unsigned short old_flags = dev->flags;
2772 dev->flags |= IFF_ALLMULTI;
2773 if ((dev->allmulti += inc) == 0)
2774 dev->flags &= ~IFF_ALLMULTI;
2775 if (dev->flags ^ old_flags) {
2776 if (dev->change_rx_flags)
2777 dev->change_rx_flags(dev, IFF_ALLMULTI);
2778 dev_set_rx_mode(dev);
2783 * Upload unicast and multicast address lists to device and
2784 * configure RX filtering. When the device doesn't support unicast
2785 * filtering it is put in promiscous mode while unicast addresses
2788 void __dev_set_rx_mode(struct net_device *dev)
2790 /* dev_open will call this function so the list will stay sane. */
2791 if (!(dev->flags&IFF_UP))
2794 if (!netif_device_present(dev))
2797 if (dev->set_rx_mode)
2798 dev->set_rx_mode(dev);
2800 /* Unicast addresses changes may only happen under the rtnl,
2801 * therefore calling __dev_set_promiscuity here is safe.
2803 if (dev->uc_count > 0 && !dev->uc_promisc) {
2804 __dev_set_promiscuity(dev, 1);
2805 dev->uc_promisc = 1;
2806 } else if (dev->uc_count == 0 && dev->uc_promisc) {
2807 __dev_set_promiscuity(dev, -1);
2808 dev->uc_promisc = 0;
2811 if (dev->set_multicast_list)
2812 dev->set_multicast_list(dev);
2816 void dev_set_rx_mode(struct net_device *dev)
2818 netif_tx_lock_bh(dev);
2819 __dev_set_rx_mode(dev);
2820 netif_tx_unlock_bh(dev);
2823 int __dev_addr_delete(struct dev_addr_list **list, int *count,
2824 void *addr, int alen, int glbl)
2826 struct dev_addr_list *da;
2828 for (; (da = *list) != NULL; list = &da->next) {
2829 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2830 alen == da->da_addrlen) {
2832 int old_glbl = da->da_gusers;
2849 int __dev_addr_add(struct dev_addr_list **list, int *count,
2850 void *addr, int alen, int glbl)
2852 struct dev_addr_list *da;
2854 for (da = *list; da != NULL; da = da->next) {
2855 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2856 da->da_addrlen == alen) {
2858 int old_glbl = da->da_gusers;
2868 da = kmalloc(sizeof(*da), GFP_ATOMIC);
2871 memcpy(da->da_addr, addr, alen);
2872 da->da_addrlen = alen;
2874 da->da_gusers = glbl ? 1 : 0;
2882 * dev_unicast_delete - Release secondary unicast address.
2884 * @addr: address to delete
2885 * @alen: length of @addr
2887 * Release reference to a secondary unicast address and remove it
2888 * from the device if the reference count drops to zero.
2890 * The caller must hold the rtnl_mutex.
2892 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
2898 netif_tx_lock_bh(dev);
2899 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2901 __dev_set_rx_mode(dev);
2902 netif_tx_unlock_bh(dev);
2905 EXPORT_SYMBOL(dev_unicast_delete);
2908 * dev_unicast_add - add a secondary unicast address
2910 * @addr: address to delete
2911 * @alen: length of @addr
2913 * Add a secondary unicast address to the device or increase
2914 * the reference count if it already exists.
2916 * The caller must hold the rtnl_mutex.
2918 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
2924 netif_tx_lock_bh(dev);
2925 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2927 __dev_set_rx_mode(dev);
2928 netif_tx_unlock_bh(dev);
2931 EXPORT_SYMBOL(dev_unicast_add);
2933 static void __dev_addr_discard(struct dev_addr_list **list)
2935 struct dev_addr_list *tmp;
2937 while (*list != NULL) {
2940 if (tmp->da_users > tmp->da_gusers)
2941 printk("__dev_addr_discard: address leakage! "
2942 "da_users=%d\n", tmp->da_users);
2947 static void dev_addr_discard(struct net_device *dev)
2949 netif_tx_lock_bh(dev);
2951 __dev_addr_discard(&dev->uc_list);
2954 __dev_addr_discard(&dev->mc_list);
2957 netif_tx_unlock_bh(dev);
2960 unsigned dev_get_flags(const struct net_device *dev)
2964 flags = (dev->flags & ~(IFF_PROMISC |
2969 (dev->gflags & (IFF_PROMISC |
2972 if (netif_running(dev)) {
2973 if (netif_oper_up(dev))
2974 flags |= IFF_RUNNING;
2975 if (netif_carrier_ok(dev))
2976 flags |= IFF_LOWER_UP;
2977 if (netif_dormant(dev))
2978 flags |= IFF_DORMANT;
2984 int dev_change_flags(struct net_device *dev, unsigned flags)
2987 int old_flags = dev->flags;
2992 * Set the flags on our device.
2995 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2996 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2998 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3002 * Load in the correct multicast list now the flags have changed.
3005 if (dev->change_rx_flags && (dev->flags ^ flags) & IFF_MULTICAST)
3006 dev->change_rx_flags(dev, IFF_MULTICAST);
3008 dev_set_rx_mode(dev);
3011 * Have we downed the interface. We handle IFF_UP ourselves
3012 * according to user attempts to set it, rather than blindly
3017 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3018 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3021 dev_set_rx_mode(dev);
3024 if (dev->flags & IFF_UP &&
3025 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3027 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3029 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3030 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3031 dev->gflags ^= IFF_PROMISC;
3032 dev_set_promiscuity(dev, inc);
3035 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3036 is important. Some (broken) drivers set IFF_PROMISC, when
3037 IFF_ALLMULTI is requested not asking us and not reporting.
3039 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3040 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3041 dev->gflags ^= IFF_ALLMULTI;
3042 dev_set_allmulti(dev, inc);
3045 /* Exclude state transition flags, already notified */
3046 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3048 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3053 int dev_set_mtu(struct net_device *dev, int new_mtu)
3057 if (new_mtu == dev->mtu)
3060 /* MTU must be positive. */
3064 if (!netif_device_present(dev))
3068 if (dev->change_mtu)
3069 err = dev->change_mtu(dev, new_mtu);
3072 if (!err && dev->flags & IFF_UP)
3073 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3077 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3081 if (!dev->set_mac_address)
3083 if (sa->sa_family != dev->type)
3085 if (!netif_device_present(dev))
3087 err = dev->set_mac_address(dev, sa);
3089 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3094 * Perform the SIOCxIFxxx calls.
3096 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3099 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3105 case SIOCGIFFLAGS: /* Get interface flags */
3106 ifr->ifr_flags = dev_get_flags(dev);
3109 case SIOCSIFFLAGS: /* Set interface flags */
3110 return dev_change_flags(dev, ifr->ifr_flags);
3112 case SIOCGIFMETRIC: /* Get the metric on the interface
3113 (currently unused) */
3114 ifr->ifr_metric = 0;
3117 case SIOCSIFMETRIC: /* Set the metric on the interface
3118 (currently unused) */
3121 case SIOCGIFMTU: /* Get the MTU of a device */
3122 ifr->ifr_mtu = dev->mtu;
3125 case SIOCSIFMTU: /* Set the MTU of a device */
3126 return dev_set_mtu(dev, ifr->ifr_mtu);
3130 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3132 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3133 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3134 ifr->ifr_hwaddr.sa_family = dev->type;
3138 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3140 case SIOCSIFHWBROADCAST:
3141 if (ifr->ifr_hwaddr.sa_family != dev->type)
3143 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3144 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3145 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3149 ifr->ifr_map.mem_start = dev->mem_start;
3150 ifr->ifr_map.mem_end = dev->mem_end;
3151 ifr->ifr_map.base_addr = dev->base_addr;
3152 ifr->ifr_map.irq = dev->irq;
3153 ifr->ifr_map.dma = dev->dma;
3154 ifr->ifr_map.port = dev->if_port;
3158 if (dev->set_config) {
3159 if (!netif_device_present(dev))
3161 return dev->set_config(dev, &ifr->ifr_map);
3166 if (!dev->set_multicast_list ||
3167 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3169 if (!netif_device_present(dev))
3171 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3175 if (!dev->set_multicast_list ||
3176 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3178 if (!netif_device_present(dev))
3180 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3184 ifr->ifr_ifindex = dev->ifindex;
3188 ifr->ifr_qlen = dev->tx_queue_len;
3192 if (ifr->ifr_qlen < 0)
3194 dev->tx_queue_len = ifr->ifr_qlen;
3198 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3199 return dev_change_name(dev, ifr->ifr_newname);
3202 * Unknown or private ioctl
3206 if ((cmd >= SIOCDEVPRIVATE &&
3207 cmd <= SIOCDEVPRIVATE + 15) ||
3208 cmd == SIOCBONDENSLAVE ||
3209 cmd == SIOCBONDRELEASE ||
3210 cmd == SIOCBONDSETHWADDR ||
3211 cmd == SIOCBONDSLAVEINFOQUERY ||
3212 cmd == SIOCBONDINFOQUERY ||
3213 cmd == SIOCBONDCHANGEACTIVE ||
3214 cmd == SIOCGMIIPHY ||
3215 cmd == SIOCGMIIREG ||
3216 cmd == SIOCSMIIREG ||
3217 cmd == SIOCBRADDIF ||
3218 cmd == SIOCBRDELIF ||
3219 cmd == SIOCWANDEV) {
3221 if (dev->do_ioctl) {
3222 if (netif_device_present(dev))
3223 err = dev->do_ioctl(dev, ifr,
3236 * This function handles all "interface"-type I/O control requests. The actual
3237 * 'doing' part of this is dev_ifsioc above.
3241 * dev_ioctl - network device ioctl
3242 * @cmd: command to issue
3243 * @arg: pointer to a struct ifreq in user space
3245 * Issue ioctl functions to devices. This is normally called by the
3246 * user space syscall interfaces but can sometimes be useful for
3247 * other purposes. The return value is the return from the syscall if
3248 * positive or a negative errno code on error.
3251 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3257 /* One special case: SIOCGIFCONF takes ifconf argument
3258 and requires shared lock, because it sleeps writing
3262 if (cmd == SIOCGIFCONF) {
3264 ret = dev_ifconf(net, (char __user *) arg);
3268 if (cmd == SIOCGIFNAME)
3269 return dev_ifname(net, (struct ifreq __user *)arg);
3271 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3274 ifr.ifr_name[IFNAMSIZ-1] = 0;
3276 colon = strchr(ifr.ifr_name, ':');
3281 * See which interface the caller is talking about.
3286 * These ioctl calls:
3287 * - can be done by all.
3288 * - atomic and do not require locking.
3299 dev_load(net, ifr.ifr_name);
3300 read_lock(&dev_base_lock);
3301 ret = dev_ifsioc(net, &ifr, cmd);
3302 read_unlock(&dev_base_lock);
3306 if (copy_to_user(arg, &ifr,
3307 sizeof(struct ifreq)))
3313 dev_load(net, ifr.ifr_name);
3315 ret = dev_ethtool(net, &ifr);
3320 if (copy_to_user(arg, &ifr,
3321 sizeof(struct ifreq)))
3327 * These ioctl calls:
3328 * - require superuser power.
3329 * - require strict serialization.
3335 if (!capable(CAP_NET_ADMIN))
3337 dev_load(net, ifr.ifr_name);
3339 ret = dev_ifsioc(net, &ifr, cmd);
3344 if (copy_to_user(arg, &ifr,
3345 sizeof(struct ifreq)))
3351 * These ioctl calls:
3352 * - require superuser power.
3353 * - require strict serialization.
3354 * - do not return a value
3364 case SIOCSIFHWBROADCAST:
3367 case SIOCBONDENSLAVE:
3368 case SIOCBONDRELEASE:
3369 case SIOCBONDSETHWADDR:
3370 case SIOCBONDCHANGEACTIVE:
3373 if (!capable(CAP_NET_ADMIN))
3376 case SIOCBONDSLAVEINFOQUERY:
3377 case SIOCBONDINFOQUERY:
3378 dev_load(net, ifr.ifr_name);
3380 ret = dev_ifsioc(net, &ifr, cmd);
3385 /* Get the per device memory space. We can add this but
3386 * currently do not support it */
3388 /* Set the per device memory buffer space.
3389 * Not applicable in our case */
3394 * Unknown or private ioctl.
3397 if (cmd == SIOCWANDEV ||
3398 (cmd >= SIOCDEVPRIVATE &&
3399 cmd <= SIOCDEVPRIVATE + 15)) {
3400 dev_load(net, ifr.ifr_name);
3402 ret = dev_ifsioc(net, &ifr, cmd);
3404 if (!ret && copy_to_user(arg, &ifr,
3405 sizeof(struct ifreq)))
3409 /* Take care of Wireless Extensions */
3410 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3411 return wext_handle_ioctl(net, &ifr, cmd, arg);
3418 * dev_new_index - allocate an ifindex
3420 * Returns a suitable unique value for a new device interface
3421 * number. The caller must hold the rtnl semaphore or the
3422 * dev_base_lock to be sure it remains unique.
3424 static int dev_new_index(struct net *net)
3430 if (!__dev_get_by_index(net, ifindex))
3435 /* Delayed registration/unregisteration */
3436 static DEFINE_SPINLOCK(net_todo_list_lock);
3437 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
3439 static void net_set_todo(struct net_device *dev)
3441 spin_lock(&net_todo_list_lock);
3442 list_add_tail(&dev->todo_list, &net_todo_list);
3443 spin_unlock(&net_todo_list_lock);
3447 * register_netdevice - register a network device
3448 * @dev: device to register
3450 * Take a completed network device structure and add it to the kernel
3451 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3452 * chain. 0 is returned on success. A negative errno code is returned
3453 * on a failure to set up the device, or if the name is a duplicate.
3455 * Callers must hold the rtnl semaphore. You may want
3456 * register_netdev() instead of this.
3459 * The locking appears insufficient to guarantee two parallel registers
3460 * will not get the same name.
3463 int register_netdevice(struct net_device *dev)
3465 struct hlist_head *head;
3466 struct hlist_node *p;
3470 BUG_ON(dev_boot_phase);
3475 /* When net_device's are persistent, this will be fatal. */
3476 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3477 BUG_ON(!dev->nd_net);
3480 spin_lock_init(&dev->queue_lock);
3481 spin_lock_init(&dev->_xmit_lock);
3482 netdev_set_lockdep_class(&dev->_xmit_lock, dev->type);
3483 dev->xmit_lock_owner = -1;
3484 spin_lock_init(&dev->ingress_lock);
3488 /* Init, if this function is available */
3490 ret = dev->init(dev);
3498 if (!dev_valid_name(dev->name)) {
3503 dev->ifindex = dev_new_index(net);
3504 if (dev->iflink == -1)
3505 dev->iflink = dev->ifindex;
3507 /* Check for existence of name */
3508 head = dev_name_hash(net, dev->name);
3509 hlist_for_each(p, head) {
3510 struct net_device *d
3511 = hlist_entry(p, struct net_device, name_hlist);
3512 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3518 /* Fix illegal checksum combinations */
3519 if ((dev->features & NETIF_F_HW_CSUM) &&
3520 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3521 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3523 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3526 if ((dev->features & NETIF_F_NO_CSUM) &&
3527 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3528 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3530 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3534 /* Fix illegal SG+CSUM combinations. */
3535 if ((dev->features & NETIF_F_SG) &&
3536 !(dev->features & NETIF_F_ALL_CSUM)) {
3537 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3539 dev->features &= ~NETIF_F_SG;
3542 /* TSO requires that SG is present as well. */
3543 if ((dev->features & NETIF_F_TSO) &&
3544 !(dev->features & NETIF_F_SG)) {
3545 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3547 dev->features &= ~NETIF_F_TSO;
3549 if (dev->features & NETIF_F_UFO) {
3550 if (!(dev->features & NETIF_F_HW_CSUM)) {
3551 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3552 "NETIF_F_HW_CSUM feature.\n",
3554 dev->features &= ~NETIF_F_UFO;
3556 if (!(dev->features & NETIF_F_SG)) {
3557 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3558 "NETIF_F_SG feature.\n",
3560 dev->features &= ~NETIF_F_UFO;
3565 * nil rebuild_header routine,
3566 * that should be never called and used as just bug trap.
3569 if (!dev->rebuild_header)
3570 dev->rebuild_header = default_rebuild_header;
3572 ret = netdev_register_kobject(dev);
3575 dev->reg_state = NETREG_REGISTERED;
3578 * Default initial state at registry is that the
3579 * device is present.
3582 set_bit(__LINK_STATE_PRESENT, &dev->state);
3584 dev_init_scheduler(dev);
3586 list_netdevice(dev);
3588 /* Notify protocols, that a new device appeared. */
3589 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
3590 ret = notifier_to_errno(ret);
3592 unregister_netdevice(dev);
3604 * register_netdev - register a network device
3605 * @dev: device to register
3607 * Take a completed network device structure and add it to the kernel
3608 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3609 * chain. 0 is returned on success. A negative errno code is returned
3610 * on a failure to set up the device, or if the name is a duplicate.
3612 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3613 * and expands the device name if you passed a format string to
3616 int register_netdev(struct net_device *dev)
3623 * If the name is a format string the caller wants us to do a
3626 if (strchr(dev->name, '%')) {
3627 err = dev_alloc_name(dev, dev->name);
3632 err = register_netdevice(dev);
3637 EXPORT_SYMBOL(register_netdev);
3640 * netdev_wait_allrefs - wait until all references are gone.
3642 * This is called when unregistering network devices.
3644 * Any protocol or device that holds a reference should register
3645 * for netdevice notification, and cleanup and put back the
3646 * reference if they receive an UNREGISTER event.
3647 * We can get stuck here if buggy protocols don't correctly
3650 static void netdev_wait_allrefs(struct net_device *dev)
3652 unsigned long rebroadcast_time, warning_time;
3654 rebroadcast_time = warning_time = jiffies;
3655 while (atomic_read(&dev->refcnt) != 0) {
3656 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3659 /* Rebroadcast unregister notification */
3660 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3662 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3664 /* We must not have linkwatch events
3665 * pending on unregister. If this
3666 * happens, we simply run the queue
3667 * unscheduled, resulting in a noop
3670 linkwatch_run_queue();
3675 rebroadcast_time = jiffies;
3680 if (time_after(jiffies, warning_time + 10 * HZ)) {
3681 printk(KERN_EMERG "unregister_netdevice: "
3682 "waiting for %s to become free. Usage "
3684 dev->name, atomic_read(&dev->refcnt));
3685 warning_time = jiffies;
3694 * register_netdevice(x1);
3695 * register_netdevice(x2);
3697 * unregister_netdevice(y1);
3698 * unregister_netdevice(y2);
3704 * We are invoked by rtnl_unlock() after it drops the semaphore.
3705 * This allows us to deal with problems:
3706 * 1) We can delete sysfs objects which invoke hotplug
3707 * without deadlocking with linkwatch via keventd.
3708 * 2) Since we run with the RTNL semaphore not held, we can sleep
3709 * safely in order to wait for the netdev refcnt to drop to zero.
3711 static DEFINE_MUTEX(net_todo_run_mutex);
3712 void netdev_run_todo(void)
3714 struct list_head list;
3716 /* Need to guard against multiple cpu's getting out of order. */
3717 mutex_lock(&net_todo_run_mutex);
3719 /* Not safe to do outside the semaphore. We must not return
3720 * until all unregister events invoked by the local processor
3721 * have been completed (either by this todo run, or one on
3724 if (list_empty(&net_todo_list))
3727 /* Snapshot list, allow later requests */
3728 spin_lock(&net_todo_list_lock);
3729 list_replace_init(&net_todo_list, &list);
3730 spin_unlock(&net_todo_list_lock);
3732 while (!list_empty(&list)) {
3733 struct net_device *dev
3734 = list_entry(list.next, struct net_device, todo_list);
3735 list_del(&dev->todo_list);
3737 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3738 printk(KERN_ERR "network todo '%s' but state %d\n",
3739 dev->name, dev->reg_state);
3744 dev->reg_state = NETREG_UNREGISTERED;
3746 netdev_wait_allrefs(dev);
3749 BUG_ON(atomic_read(&dev->refcnt));
3750 BUG_TRAP(!dev->ip_ptr);
3751 BUG_TRAP(!dev->ip6_ptr);
3752 BUG_TRAP(!dev->dn_ptr);
3754 if (dev->destructor)
3755 dev->destructor(dev);
3757 /* Free network device */
3758 kobject_put(&dev->dev.kobj);
3762 mutex_unlock(&net_todo_run_mutex);
3765 static struct net_device_stats *internal_stats(struct net_device *dev)
3771 * alloc_netdev_mq - allocate network device
3772 * @sizeof_priv: size of private data to allocate space for
3773 * @name: device name format string
3774 * @setup: callback to initialize device
3775 * @queue_count: the number of subqueues to allocate
3777 * Allocates a struct net_device with private data area for driver use
3778 * and performs basic initialization. Also allocates subquue structs
3779 * for each queue on the device at the end of the netdevice.
3781 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
3782 void (*setup)(struct net_device *), unsigned int queue_count)
3785 struct net_device *dev;
3788 BUG_ON(strlen(name) >= sizeof(dev->name));
3790 /* ensure 32-byte alignment of both the device and private area */
3791 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST +
3792 (sizeof(struct net_device_subqueue) * (queue_count - 1))) &
3793 ~NETDEV_ALIGN_CONST;
3794 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3796 p = kzalloc(alloc_size, GFP_KERNEL);
3798 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
3802 dev = (struct net_device *)
3803 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3804 dev->padded = (char *)dev - (char *)p;
3805 dev->nd_net = &init_net;
3808 dev->priv = ((char *)dev +
3809 ((sizeof(struct net_device) +
3810 (sizeof(struct net_device_subqueue) *
3811 (queue_count - 1)) + NETDEV_ALIGN_CONST)
3812 & ~NETDEV_ALIGN_CONST));
3815 dev->egress_subqueue_count = queue_count;
3817 dev->get_stats = internal_stats;
3818 netpoll_netdev_init(dev);
3820 strcpy(dev->name, name);
3823 EXPORT_SYMBOL(alloc_netdev_mq);
3826 * free_netdev - free network device
3829 * This function does the last stage of destroying an allocated device
3830 * interface. The reference to the device object is released.
3831 * If this is the last reference then it will be freed.
3833 void free_netdev(struct net_device *dev)
3835 /* Compatibility with error handling in drivers */
3836 if (dev->reg_state == NETREG_UNINITIALIZED) {
3837 kfree((char *)dev - dev->padded);
3841 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3842 dev->reg_state = NETREG_RELEASED;
3844 /* will free via device release */
3845 put_device(&dev->dev);
3848 /* Synchronize with packet receive processing. */
3849 void synchronize_net(void)
3856 * unregister_netdevice - remove device from the kernel
3859 * This function shuts down a device interface and removes it
3860 * from the kernel tables. On success 0 is returned, on a failure
3861 * a negative errno code is returned.
3863 * Callers must hold the rtnl semaphore. You may want
3864 * unregister_netdev() instead of this.
3867 void unregister_netdevice(struct net_device *dev)
3869 BUG_ON(dev_boot_phase);
3872 /* Some devices call without registering for initialization unwind. */
3873 if (dev->reg_state == NETREG_UNINITIALIZED) {
3874 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3875 "was registered\n", dev->name, dev);
3881 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3883 /* If device is running, close it first. */
3884 if (dev->flags & IFF_UP)
3887 /* And unlink it from device chain. */
3888 unlist_netdevice(dev);
3890 dev->reg_state = NETREG_UNREGISTERING;
3894 /* Shutdown queueing discipline. */
3898 /* Notify protocols, that we are about to destroy
3899 this device. They should clean all the things.
3901 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3904 * Flush the unicast and multicast chains
3906 dev_addr_discard(dev);
3911 /* Notifier chain MUST detach us from master device. */
3912 BUG_TRAP(!dev->master);
3914 /* Remove entries from kobject tree */
3915 netdev_unregister_kobject(dev);
3917 /* Finish processing unregister after unlock */
3926 * unregister_netdev - remove device from the kernel
3929 * This function shuts down a device interface and removes it
3930 * from the kernel tables. On success 0 is returned, on a failure
3931 * a negative errno code is returned.
3933 * This is just a wrapper for unregister_netdevice that takes
3934 * the rtnl semaphore. In general you want to use this and not
3935 * unregister_netdevice.
3937 void unregister_netdev(struct net_device *dev)
3940 unregister_netdevice(dev);
3944 EXPORT_SYMBOL(unregister_netdev);
3947 * dev_change_net_namespace - move device to different nethost namespace
3949 * @net: network namespace
3950 * @pat: If not NULL name pattern to try if the current device name
3951 * is already taken in the destination network namespace.
3953 * This function shuts down a device interface and moves it
3954 * to a new network namespace. On success 0 is returned, on
3955 * a failure a netagive errno code is returned.
3957 * Callers must hold the rtnl semaphore.
3960 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
3963 const char *destname;
3968 /* Don't allow namespace local devices to be moved. */
3970 if (dev->features & NETIF_F_NETNS_LOCAL)
3973 /* Ensure the device has been registrered */
3975 if (dev->reg_state != NETREG_REGISTERED)
3978 /* Get out if there is nothing todo */
3980 if (dev->nd_net == net)
3983 /* Pick the destination device name, and ensure
3984 * we can use it in the destination network namespace.
3987 destname = dev->name;
3988 if (__dev_get_by_name(net, destname)) {
3989 /* We get here if we can't use the current device name */
3992 if (!dev_valid_name(pat))
3994 if (strchr(pat, '%')) {
3995 if (__dev_alloc_name(net, pat, buf) < 0)
4000 if (__dev_get_by_name(net, destname))
4005 * And now a mini version of register_netdevice unregister_netdevice.
4008 /* If device is running close it first. */
4009 if (dev->flags & IFF_UP)
4012 /* And unlink it from device chain */
4014 unlist_netdevice(dev);
4018 /* Shutdown queueing discipline. */
4021 /* Notify protocols, that we are about to destroy
4022 this device. They should clean all the things.
4024 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4027 * Flush the unicast and multicast chains
4029 dev_addr_discard(dev);
4031 /* Actually switch the network namespace */
4034 /* Assign the new device name */
4035 if (destname != dev->name)
4036 strcpy(dev->name, destname);
4038 /* If there is an ifindex conflict assign a new one */
4039 if (__dev_get_by_index(net, dev->ifindex)) {
4040 int iflink = (dev->iflink == dev->ifindex);
4041 dev->ifindex = dev_new_index(net);
4043 dev->iflink = dev->ifindex;
4046 /* Fixup kobjects */
4047 err = device_rename(&dev->dev, dev->name);
4050 /* Add the device back in the hashes */
4051 list_netdevice(dev);
4053 /* Notify protocols, that a new device appeared. */
4054 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4062 static int dev_cpu_callback(struct notifier_block *nfb,
4063 unsigned long action,
4066 struct sk_buff **list_skb;
4067 struct net_device **list_net;
4068 struct sk_buff *skb;
4069 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4070 struct softnet_data *sd, *oldsd;
4072 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4075 local_irq_disable();
4076 cpu = smp_processor_id();
4077 sd = &per_cpu(softnet_data, cpu);
4078 oldsd = &per_cpu(softnet_data, oldcpu);
4080 /* Find end of our completion_queue. */
4081 list_skb = &sd->completion_queue;
4083 list_skb = &(*list_skb)->next;
4084 /* Append completion queue from offline CPU. */
4085 *list_skb = oldsd->completion_queue;
4086 oldsd->completion_queue = NULL;
4088 /* Find end of our output_queue. */
4089 list_net = &sd->output_queue;
4091 list_net = &(*list_net)->next_sched;
4092 /* Append output queue from offline CPU. */
4093 *list_net = oldsd->output_queue;
4094 oldsd->output_queue = NULL;
4096 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4099 /* Process offline CPU's input_pkt_queue */
4100 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4106 #ifdef CONFIG_NET_DMA
4108 * net_dma_rebalance - try to maintain one DMA channel per CPU
4109 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4111 * This is called when the number of channels allocated to the net_dma client
4112 * changes. The net_dma client tries to have one DMA channel per CPU.
4115 static void net_dma_rebalance(struct net_dma *net_dma)
4117 unsigned int cpu, i, n, chan_idx;
4118 struct dma_chan *chan;
4120 if (cpus_empty(net_dma->channel_mask)) {
4121 for_each_online_cpu(cpu)
4122 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4127 cpu = first_cpu(cpu_online_map);
4129 for_each_cpu_mask(chan_idx, net_dma->channel_mask) {
4130 chan = net_dma->channels[chan_idx];
4132 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4133 + (i < (num_online_cpus() %
4134 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4137 per_cpu(softnet_data, cpu).net_dma = chan;
4138 cpu = next_cpu(cpu, cpu_online_map);
4146 * netdev_dma_event - event callback for the net_dma_client
4147 * @client: should always be net_dma_client
4148 * @chan: DMA channel for the event
4149 * @state: DMA state to be handled
4151 static enum dma_state_client
4152 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4153 enum dma_state state)
4155 int i, found = 0, pos = -1;
4156 struct net_dma *net_dma =
4157 container_of(client, struct net_dma, client);
4158 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4160 spin_lock(&net_dma->lock);
4162 case DMA_RESOURCE_AVAILABLE:
4163 for (i = 0; i < NR_CPUS; i++)
4164 if (net_dma->channels[i] == chan) {
4167 } else if (net_dma->channels[i] == NULL && pos < 0)
4170 if (!found && pos >= 0) {
4172 net_dma->channels[pos] = chan;
4173 cpu_set(pos, net_dma->channel_mask);
4174 net_dma_rebalance(net_dma);
4177 case DMA_RESOURCE_REMOVED:
4178 for (i = 0; i < NR_CPUS; i++)
4179 if (net_dma->channels[i] == chan) {
4187 cpu_clear(pos, net_dma->channel_mask);
4188 net_dma->channels[i] = NULL;
4189 net_dma_rebalance(net_dma);
4195 spin_unlock(&net_dma->lock);
4201 * netdev_dma_regiser - register the networking subsystem as a DMA client
4203 static int __init netdev_dma_register(void)
4205 spin_lock_init(&net_dma.lock);
4206 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4207 dma_async_client_register(&net_dma.client);
4208 dma_async_client_chan_request(&net_dma.client);
4213 static int __init netdev_dma_register(void) { return -ENODEV; }
4214 #endif /* CONFIG_NET_DMA */
4217 * netdev_compute_feature - compute conjunction of two feature sets
4218 * @all: first feature set
4219 * @one: second feature set
4221 * Computes a new feature set after adding a device with feature set
4222 * @one to the master device with current feature set @all. Returns
4223 * the new feature set.
4225 int netdev_compute_features(unsigned long all, unsigned long one)
4227 /* if device needs checksumming, downgrade to hw checksumming */
4228 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4229 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4231 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4232 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4233 all ^= NETIF_F_HW_CSUM
4234 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4236 if (one & NETIF_F_GSO)
4237 one |= NETIF_F_GSO_SOFTWARE;
4240 /* If even one device supports robust GSO, enable it for all. */
4241 if (one & NETIF_F_GSO_ROBUST)
4242 all |= NETIF_F_GSO_ROBUST;
4244 all &= one | NETIF_F_LLTX;
4246 if (!(all & NETIF_F_ALL_CSUM))
4248 if (!(all & NETIF_F_SG))
4249 all &= ~NETIF_F_GSO_MASK;
4253 EXPORT_SYMBOL(netdev_compute_features);
4255 static struct hlist_head *netdev_create_hash(void)
4258 struct hlist_head *hash;
4260 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4262 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4263 INIT_HLIST_HEAD(&hash[i]);
4268 /* Initialize per network namespace state */
4269 static int netdev_init(struct net *net)
4271 INIT_LIST_HEAD(&net->dev_base_head);
4272 rwlock_init(&dev_base_lock);
4274 net->dev_name_head = netdev_create_hash();
4275 if (net->dev_name_head == NULL)
4278 net->dev_index_head = netdev_create_hash();
4279 if (net->dev_index_head == NULL)
4285 kfree(net->dev_name_head);
4290 static void netdev_exit(struct net *net)
4292 kfree(net->dev_name_head);
4293 kfree(net->dev_index_head);
4296 static struct pernet_operations netdev_net_ops = {
4297 .init = netdev_init,
4298 .exit = netdev_exit,
4301 static void default_device_exit(struct net *net)
4303 struct net_device *dev, *next;
4305 * Push all migratable of the network devices back to the
4306 * initial network namespace
4309 for_each_netdev_safe(net, dev, next) {
4312 /* Ignore unmoveable devices (i.e. loopback) */
4313 if (dev->features & NETIF_F_NETNS_LOCAL)
4316 /* Push remaing network devices to init_net */
4317 err = dev_change_net_namespace(dev, &init_net, "dev%d");
4319 printk(KERN_WARNING "%s: failed to move %s to init_net: %d\n",
4320 __func__, dev->name, err);
4321 unregister_netdevice(dev);
4327 static struct pernet_operations default_device_ops = {
4328 .exit = default_device_exit,
4332 * Initialize the DEV module. At boot time this walks the device list and
4333 * unhooks any devices that fail to initialise (normally hardware not
4334 * present) and leaves us with a valid list of present and active devices.
4339 * This is called single threaded during boot, so no need
4340 * to take the rtnl semaphore.
4342 static int __init net_dev_init(void)
4344 int i, rc = -ENOMEM;
4346 BUG_ON(!dev_boot_phase);
4348 if (dev_proc_init())
4351 if (netdev_kobject_init())
4354 INIT_LIST_HEAD(&ptype_all);
4355 for (i = 0; i < 16; i++)
4356 INIT_LIST_HEAD(&ptype_base[i]);
4358 if (register_pernet_subsys(&netdev_net_ops))
4361 if (register_pernet_device(&default_device_ops))
4365 * Initialise the packet receive queues.
4368 for_each_possible_cpu(i) {
4369 struct softnet_data *queue;
4371 queue = &per_cpu(softnet_data, i);
4372 skb_queue_head_init(&queue->input_pkt_queue);
4373 queue->completion_queue = NULL;
4374 INIT_LIST_HEAD(&queue->poll_list);
4376 queue->backlog.poll = process_backlog;
4377 queue->backlog.weight = weight_p;
4380 netdev_dma_register();
4384 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
4385 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
4387 hotcpu_notifier(dev_cpu_callback, 0);
4395 subsys_initcall(net_dev_init);
4397 EXPORT_SYMBOL(__dev_get_by_index);
4398 EXPORT_SYMBOL(__dev_get_by_name);
4399 EXPORT_SYMBOL(__dev_remove_pack);
4400 EXPORT_SYMBOL(dev_valid_name);
4401 EXPORT_SYMBOL(dev_add_pack);
4402 EXPORT_SYMBOL(dev_alloc_name);
4403 EXPORT_SYMBOL(dev_close);
4404 EXPORT_SYMBOL(dev_get_by_flags);
4405 EXPORT_SYMBOL(dev_get_by_index);
4406 EXPORT_SYMBOL(dev_get_by_name);
4407 EXPORT_SYMBOL(dev_open);
4408 EXPORT_SYMBOL(dev_queue_xmit);
4409 EXPORT_SYMBOL(dev_remove_pack);
4410 EXPORT_SYMBOL(dev_set_allmulti);
4411 EXPORT_SYMBOL(dev_set_promiscuity);
4412 EXPORT_SYMBOL(dev_change_flags);
4413 EXPORT_SYMBOL(dev_set_mtu);
4414 EXPORT_SYMBOL(dev_set_mac_address);
4415 EXPORT_SYMBOL(free_netdev);
4416 EXPORT_SYMBOL(netdev_boot_setup_check);
4417 EXPORT_SYMBOL(netdev_set_master);
4418 EXPORT_SYMBOL(netdev_state_change);
4419 EXPORT_SYMBOL(netif_receive_skb);
4420 EXPORT_SYMBOL(netif_rx);
4421 EXPORT_SYMBOL(register_gifconf);
4422 EXPORT_SYMBOL(register_netdevice);
4423 EXPORT_SYMBOL(register_netdevice_notifier);
4424 EXPORT_SYMBOL(skb_checksum_help);
4425 EXPORT_SYMBOL(synchronize_net);
4426 EXPORT_SYMBOL(unregister_netdevice);
4427 EXPORT_SYMBOL(unregister_netdevice_notifier);
4428 EXPORT_SYMBOL(net_enable_timestamp);
4429 EXPORT_SYMBOL(net_disable_timestamp);
4430 EXPORT_SYMBOL(dev_get_flags);
4432 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4433 EXPORT_SYMBOL(br_handle_frame_hook);
4434 EXPORT_SYMBOL(br_fdb_get_hook);
4435 EXPORT_SYMBOL(br_fdb_put_hook);
4439 EXPORT_SYMBOL(dev_load);
4442 EXPORT_PER_CPU_SYMBOL(softnet_data);