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);
253 extern int netdev_sysfs_init(void);
254 extern int netdev_register_sysfs(struct net_device *);
255 extern void netdev_unregister_sysfs(struct net_device *);
257 #define netdev_sysfs_init() (0)
258 #define netdev_register_sysfs(dev) (0)
259 #define netdev_unregister_sysfs(dev) do { } while(0)
262 #ifdef CONFIG_DEBUG_LOCK_ALLOC
264 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
265 * according to dev->type
267 static const unsigned short netdev_lock_type[] =
268 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
269 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
270 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
271 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
272 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
273 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
274 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
275 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
276 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
277 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
278 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
279 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
280 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
281 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
284 static const char *netdev_lock_name[] =
285 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
286 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
287 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
288 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
289 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
290 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
291 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
292 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
293 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
294 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
295 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
296 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
297 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
298 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
301 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
303 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
307 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
308 if (netdev_lock_type[i] == dev_type)
310 /* the last key is used by default */
311 return ARRAY_SIZE(netdev_lock_type) - 1;
314 static inline void netdev_set_lockdep_class(spinlock_t *lock,
315 unsigned short dev_type)
319 i = netdev_lock_pos(dev_type);
320 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
321 netdev_lock_name[i]);
324 static inline void netdev_set_lockdep_class(spinlock_t *lock,
325 unsigned short dev_type)
330 /*******************************************************************************
332 Protocol management and registration routines
334 *******************************************************************************/
337 * Add a protocol ID to the list. Now that the input handler is
338 * smarter we can dispense with all the messy stuff that used to be
341 * BEWARE!!! Protocol handlers, mangling input packets,
342 * MUST BE last in hash buckets and checking protocol handlers
343 * MUST start from promiscuous ptype_all chain in net_bh.
344 * It is true now, do not change it.
345 * Explanation follows: if protocol handler, mangling packet, will
346 * be the first on list, it is not able to sense, that packet
347 * is cloned and should be copied-on-write, so that it will
348 * change it and subsequent readers will get broken packet.
353 * dev_add_pack - add packet handler
354 * @pt: packet type declaration
356 * Add a protocol handler to the networking stack. The passed &packet_type
357 * is linked into kernel lists and may not be freed until it has been
358 * removed from the kernel lists.
360 * This call does not sleep therefore it can not
361 * guarantee all CPU's that are in middle of receiving packets
362 * will see the new packet type (until the next received packet).
365 void dev_add_pack(struct packet_type *pt)
369 spin_lock_bh(&ptype_lock);
370 if (pt->type == htons(ETH_P_ALL))
371 list_add_rcu(&pt->list, &ptype_all);
373 hash = ntohs(pt->type) & 15;
374 list_add_rcu(&pt->list, &ptype_base[hash]);
376 spin_unlock_bh(&ptype_lock);
380 * __dev_remove_pack - remove packet handler
381 * @pt: packet type declaration
383 * Remove a protocol handler that was previously added to the kernel
384 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
385 * from the kernel lists and can be freed or reused once this function
388 * The packet type might still be in use by receivers
389 * and must not be freed until after all the CPU's have gone
390 * through a quiescent state.
392 void __dev_remove_pack(struct packet_type *pt)
394 struct list_head *head;
395 struct packet_type *pt1;
397 spin_lock_bh(&ptype_lock);
399 if (pt->type == htons(ETH_P_ALL))
402 head = &ptype_base[ntohs(pt->type) & 15];
404 list_for_each_entry(pt1, head, list) {
406 list_del_rcu(&pt->list);
411 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
413 spin_unlock_bh(&ptype_lock);
416 * dev_remove_pack - remove packet handler
417 * @pt: packet type declaration
419 * Remove a protocol handler that was previously added to the kernel
420 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
421 * from the kernel lists and can be freed or reused once this function
424 * This call sleeps to guarantee that no CPU is looking at the packet
427 void dev_remove_pack(struct packet_type *pt)
429 __dev_remove_pack(pt);
434 /******************************************************************************
436 Device Boot-time Settings Routines
438 *******************************************************************************/
440 /* Boot time configuration table */
441 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
444 * netdev_boot_setup_add - add new setup entry
445 * @name: name of the device
446 * @map: configured settings for the device
448 * Adds new setup entry to the dev_boot_setup list. The function
449 * returns 0 on error and 1 on success. This is a generic routine to
452 static int netdev_boot_setup_add(char *name, struct ifmap *map)
454 struct netdev_boot_setup *s;
458 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
459 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
460 memset(s[i].name, 0, sizeof(s[i].name));
461 strcpy(s[i].name, name);
462 memcpy(&s[i].map, map, sizeof(s[i].map));
467 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
471 * netdev_boot_setup_check - check boot time settings
472 * @dev: the netdevice
474 * Check boot time settings for the device.
475 * The found settings are set for the device to be used
476 * later in the device probing.
477 * Returns 0 if no settings found, 1 if they are.
479 int netdev_boot_setup_check(struct net_device *dev)
481 struct netdev_boot_setup *s = dev_boot_setup;
484 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
485 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
486 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
487 dev->irq = s[i].map.irq;
488 dev->base_addr = s[i].map.base_addr;
489 dev->mem_start = s[i].map.mem_start;
490 dev->mem_end = s[i].map.mem_end;
499 * netdev_boot_base - get address from boot time settings
500 * @prefix: prefix for network device
501 * @unit: id for network device
503 * Check boot time settings for the base address of device.
504 * The found settings are set for the device to be used
505 * later in the device probing.
506 * Returns 0 if no settings found.
508 unsigned long netdev_boot_base(const char *prefix, int unit)
510 const struct netdev_boot_setup *s = dev_boot_setup;
514 sprintf(name, "%s%d", prefix, unit);
517 * If device already registered then return base of 1
518 * to indicate not to probe for this interface
520 if (__dev_get_by_name(&init_net, name))
523 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
524 if (!strcmp(name, s[i].name))
525 return s[i].map.base_addr;
530 * Saves at boot time configured settings for any netdevice.
532 int __init netdev_boot_setup(char *str)
537 str = get_options(str, ARRAY_SIZE(ints), ints);
542 memset(&map, 0, sizeof(map));
546 map.base_addr = ints[2];
548 map.mem_start = ints[3];
550 map.mem_end = ints[4];
552 /* Add new entry to the list */
553 return netdev_boot_setup_add(str, &map);
556 __setup("netdev=", netdev_boot_setup);
558 /*******************************************************************************
560 Device Interface Subroutines
562 *******************************************************************************/
565 * __dev_get_by_name - find a device by its name
566 * @name: name to find
568 * Find an interface by name. Must be called under RTNL semaphore
569 * or @dev_base_lock. If the name is found a pointer to the device
570 * is returned. If the name is not found then %NULL is returned. The
571 * reference counters are not incremented so the caller must be
572 * careful with locks.
575 struct net_device *__dev_get_by_name(struct net *net, const char *name)
577 struct hlist_node *p;
579 hlist_for_each(p, dev_name_hash(net, name)) {
580 struct net_device *dev
581 = hlist_entry(p, struct net_device, name_hlist);
582 if (!strncmp(dev->name, name, IFNAMSIZ))
589 * dev_get_by_name - find a device by its name
590 * @name: name to find
592 * Find an interface by name. This can be called from any
593 * context and does its own locking. The returned handle has
594 * the usage count incremented and the caller must use dev_put() to
595 * release it when it is no longer needed. %NULL is returned if no
596 * matching device is found.
599 struct net_device *dev_get_by_name(struct net *net, const char *name)
601 struct net_device *dev;
603 read_lock(&dev_base_lock);
604 dev = __dev_get_by_name(net, name);
607 read_unlock(&dev_base_lock);
612 * __dev_get_by_index - find a device by its ifindex
613 * @ifindex: index of device
615 * Search for an interface by index. Returns %NULL if the device
616 * is not found or a pointer to the device. The device has not
617 * had its reference counter increased so the caller must be careful
618 * about locking. The caller must hold either the RTNL semaphore
622 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
624 struct hlist_node *p;
626 hlist_for_each(p, dev_index_hash(net, ifindex)) {
627 struct net_device *dev
628 = hlist_entry(p, struct net_device, index_hlist);
629 if (dev->ifindex == ifindex)
637 * dev_get_by_index - find a device by its ifindex
638 * @ifindex: index of device
640 * Search for an interface by index. Returns NULL if the device
641 * is not found or a pointer to the device. The device returned has
642 * had a reference added and the pointer is safe until the user calls
643 * dev_put to indicate they have finished with it.
646 struct net_device *dev_get_by_index(struct net *net, int ifindex)
648 struct net_device *dev;
650 read_lock(&dev_base_lock);
651 dev = __dev_get_by_index(net, ifindex);
654 read_unlock(&dev_base_lock);
659 * dev_getbyhwaddr - find a device by its hardware address
660 * @type: media type of device
661 * @ha: hardware address
663 * Search for an interface by MAC address. Returns NULL if the device
664 * is not found or a pointer to the device. The caller must hold the
665 * rtnl semaphore. The returned device has not had its ref count increased
666 * and the caller must therefore be careful about locking
669 * If the API was consistent this would be __dev_get_by_hwaddr
672 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
674 struct net_device *dev;
678 for_each_netdev(&init_net, dev)
679 if (dev->type == type &&
680 !memcmp(dev->dev_addr, ha, dev->addr_len))
686 EXPORT_SYMBOL(dev_getbyhwaddr);
688 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
690 struct net_device *dev;
693 for_each_netdev(net, dev)
694 if (dev->type == type)
700 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
702 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
704 struct net_device *dev;
707 dev = __dev_getfirstbyhwtype(net, type);
714 EXPORT_SYMBOL(dev_getfirstbyhwtype);
717 * dev_get_by_flags - find any device with given flags
718 * @if_flags: IFF_* values
719 * @mask: bitmask of bits in if_flags to check
721 * Search for any interface with the given flags. Returns NULL if a device
722 * is not found or a pointer to the device. The device returned has
723 * had a reference added and the pointer is safe until the user calls
724 * dev_put to indicate they have finished with it.
727 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
729 struct net_device *dev, *ret;
732 read_lock(&dev_base_lock);
733 for_each_netdev(net, dev) {
734 if (((dev->flags ^ if_flags) & mask) == 0) {
740 read_unlock(&dev_base_lock);
745 * dev_valid_name - check if name is okay for network device
748 * Network device names need to be valid file names to
749 * to allow sysfs to work. We also disallow any kind of
752 int dev_valid_name(const char *name)
756 if (strlen(name) >= IFNAMSIZ)
758 if (!strcmp(name, ".") || !strcmp(name, ".."))
762 if (*name == '/' || isspace(*name))
770 * __dev_alloc_name - allocate a name for a device
771 * @net: network namespace to allocate the device name in
772 * @name: name format string
773 * @buf: scratch buffer and result name string
775 * Passed a format string - eg "lt%d" it will try and find a suitable
776 * id. It scans list of devices to build up a free map, then chooses
777 * the first empty slot. The caller must hold the dev_base or rtnl lock
778 * while allocating the name and adding the device in order to avoid
780 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
781 * Returns the number of the unit assigned or a negative errno code.
784 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
788 const int max_netdevices = 8*PAGE_SIZE;
790 struct net_device *d;
792 p = strnchr(name, IFNAMSIZ-1, '%');
795 * Verify the string as this thing may have come from
796 * the user. There must be either one "%d" and no other "%"
799 if (p[1] != 'd' || strchr(p + 2, '%'))
802 /* Use one page as a bit array of possible slots */
803 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
807 for_each_netdev(net, d) {
808 if (!sscanf(d->name, name, &i))
810 if (i < 0 || i >= max_netdevices)
813 /* avoid cases where sscanf is not exact inverse of printf */
814 snprintf(buf, IFNAMSIZ, name, i);
815 if (!strncmp(buf, d->name, IFNAMSIZ))
819 i = find_first_zero_bit(inuse, max_netdevices);
820 free_page((unsigned long) inuse);
823 snprintf(buf, IFNAMSIZ, name, i);
824 if (!__dev_get_by_name(net, buf))
827 /* It is possible to run out of possible slots
828 * when the name is long and there isn't enough space left
829 * for the digits, or if all bits are used.
835 * dev_alloc_name - allocate a name for a device
837 * @name: name format string
839 * Passed a format string - eg "lt%d" it will try and find a suitable
840 * id. It scans list of devices to build up a free map, then chooses
841 * the first empty slot. The caller must hold the dev_base or rtnl lock
842 * while allocating the name and adding the device in order to avoid
844 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
845 * Returns the number of the unit assigned or a negative errno code.
848 int dev_alloc_name(struct net_device *dev, const char *name)
854 BUG_ON(!dev->nd_net);
856 ret = __dev_alloc_name(net, name, buf);
858 strlcpy(dev->name, buf, IFNAMSIZ);
864 * dev_change_name - change name of a device
866 * @newname: name (or format string) must be at least IFNAMSIZ
868 * Change name of a device, can pass format strings "eth%d".
871 int dev_change_name(struct net_device *dev, char *newname)
873 char oldname[IFNAMSIZ];
879 BUG_ON(!dev->nd_net);
882 if (dev->flags & IFF_UP)
885 if (!dev_valid_name(newname))
888 memcpy(oldname, dev->name, IFNAMSIZ);
890 if (strchr(newname, '%')) {
891 err = dev_alloc_name(dev, newname);
894 strcpy(newname, dev->name);
896 else if (__dev_get_by_name(net, newname))
899 strlcpy(dev->name, newname, IFNAMSIZ);
902 device_rename(&dev->dev, dev->name);
904 write_lock_bh(&dev_base_lock);
905 hlist_del(&dev->name_hlist);
906 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
907 write_unlock_bh(&dev_base_lock);
909 ret = raw_notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
910 ret = notifier_to_errno(ret);
915 "%s: name change rollback failed: %d.\n",
919 memcpy(dev->name, oldname, IFNAMSIZ);
928 * netdev_features_change - device changes features
929 * @dev: device to cause notification
931 * Called to indicate a device has changed features.
933 void netdev_features_change(struct net_device *dev)
935 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
937 EXPORT_SYMBOL(netdev_features_change);
940 * netdev_state_change - device changes state
941 * @dev: device to cause notification
943 * Called to indicate a device has changed state. This function calls
944 * the notifier chains for netdev_chain and sends a NEWLINK message
945 * to the routing socket.
947 void netdev_state_change(struct net_device *dev)
949 if (dev->flags & IFF_UP) {
950 raw_notifier_call_chain(&netdev_chain,
952 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
957 * dev_load - load a network module
958 * @name: name of interface
960 * If a network interface is not present and the process has suitable
961 * privileges this function loads the module. If module loading is not
962 * available in this kernel then it becomes a nop.
965 void dev_load(struct net *net, const char *name)
967 struct net_device *dev;
969 read_lock(&dev_base_lock);
970 dev = __dev_get_by_name(net, name);
971 read_unlock(&dev_base_lock);
973 if (!dev && capable(CAP_SYS_MODULE))
974 request_module("%s", name);
977 static int default_rebuild_header(struct sk_buff *skb)
979 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
980 skb->dev ? skb->dev->name : "NULL!!!");
986 * dev_open - prepare an interface for use.
987 * @dev: device to open
989 * Takes a device from down to up state. The device's private open
990 * function is invoked and then the multicast lists are loaded. Finally
991 * the device is moved into the up state and a %NETDEV_UP message is
992 * sent to the netdev notifier chain.
994 * Calling this function on an active interface is a nop. On a failure
995 * a negative errno code is returned.
997 int dev_open(struct net_device *dev)
1005 if (dev->flags & IFF_UP)
1009 * Is it even present?
1011 if (!netif_device_present(dev))
1015 * Call device private open method
1017 set_bit(__LINK_STATE_START, &dev->state);
1019 ret = dev->open(dev);
1021 clear_bit(__LINK_STATE_START, &dev->state);
1025 * If it went open OK then:
1032 dev->flags |= IFF_UP;
1035 * Initialize multicasting status
1037 dev_set_rx_mode(dev);
1040 * Wakeup transmit queue engine
1045 * ... and announce new interface.
1047 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
1053 * dev_close - shutdown an interface.
1054 * @dev: device to shutdown
1056 * This function moves an active device into down state. A
1057 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1058 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1061 int dev_close(struct net_device *dev)
1065 if (!(dev->flags & IFF_UP))
1069 * Tell people we are going down, so that they can
1070 * prepare to death, when device is still operating.
1072 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
1074 dev_deactivate(dev);
1076 clear_bit(__LINK_STATE_START, &dev->state);
1078 /* Synchronize to scheduled poll. We cannot touch poll list,
1079 * it can be even on different cpu. So just clear netif_running().
1081 * dev->stop() will invoke napi_disable() on all of it's
1082 * napi_struct instances on this device.
1084 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1087 * Call the device specific close. This cannot fail.
1088 * Only if device is UP
1090 * We allow it to be called even after a DETACH hot-plug
1097 * Device is now down.
1100 dev->flags &= ~IFF_UP;
1103 * Tell people we are down
1105 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1111 static int dev_boot_phase = 1;
1114 * Device change register/unregister. These are not inline or static
1115 * as we export them to the world.
1119 * register_netdevice_notifier - register a network notifier block
1122 * Register a notifier to be called when network device events occur.
1123 * The notifier passed is linked into the kernel structures and must
1124 * not be reused until it has been unregistered. A negative errno code
1125 * is returned on a failure.
1127 * When registered all registration and up events are replayed
1128 * to the new notifier to allow device to have a race free
1129 * view of the network device list.
1132 int register_netdevice_notifier(struct notifier_block *nb)
1134 struct net_device *dev;
1135 struct net_device *last;
1140 err = raw_notifier_chain_register(&netdev_chain, nb);
1146 for_each_netdev(net, dev) {
1147 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1148 err = notifier_to_errno(err);
1152 if (!(dev->flags & IFF_UP))
1155 nb->notifier_call(nb, NETDEV_UP, dev);
1166 for_each_netdev(net, dev) {
1170 if (dev->flags & IFF_UP) {
1171 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1172 nb->notifier_call(nb, NETDEV_DOWN, dev);
1174 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1181 * unregister_netdevice_notifier - unregister a network notifier block
1184 * Unregister a notifier previously registered by
1185 * register_netdevice_notifier(). The notifier is unlinked into the
1186 * kernel structures and may then be reused. A negative errno code
1187 * is returned on a failure.
1190 int unregister_netdevice_notifier(struct notifier_block *nb)
1195 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1201 * call_netdevice_notifiers - call all network notifier blocks
1202 * @val: value passed unmodified to notifier function
1203 * @v: pointer passed unmodified to notifier function
1205 * Call all network notifier blocks. Parameters and return value
1206 * are as for raw_notifier_call_chain().
1209 int call_netdevice_notifiers(unsigned long val, void *v)
1211 return raw_notifier_call_chain(&netdev_chain, val, v);
1214 /* When > 0 there are consumers of rx skb time stamps */
1215 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1217 void net_enable_timestamp(void)
1219 atomic_inc(&netstamp_needed);
1222 void net_disable_timestamp(void)
1224 atomic_dec(&netstamp_needed);
1227 static inline void net_timestamp(struct sk_buff *skb)
1229 if (atomic_read(&netstamp_needed))
1230 __net_timestamp(skb);
1232 skb->tstamp.tv64 = 0;
1236 * Support routine. Sends outgoing frames to any network
1237 * taps currently in use.
1240 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1242 struct packet_type *ptype;
1247 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1248 /* Never send packets back to the socket
1249 * they originated from - MvS (miquels@drinkel.ow.org)
1251 if ((ptype->dev == dev || !ptype->dev) &&
1252 (ptype->af_packet_priv == NULL ||
1253 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1254 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1258 /* skb->nh should be correctly
1259 set by sender, so that the second statement is
1260 just protection against buggy protocols.
1262 skb_reset_mac_header(skb2);
1264 if (skb_network_header(skb2) < skb2->data ||
1265 skb2->network_header > skb2->tail) {
1266 if (net_ratelimit())
1267 printk(KERN_CRIT "protocol %04x is "
1269 skb2->protocol, dev->name);
1270 skb_reset_network_header(skb2);
1273 skb2->transport_header = skb2->network_header;
1274 skb2->pkt_type = PACKET_OUTGOING;
1275 ptype->func(skb2, skb->dev, ptype, skb->dev);
1282 void __netif_schedule(struct net_device *dev)
1284 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1285 unsigned long flags;
1286 struct softnet_data *sd;
1288 local_irq_save(flags);
1289 sd = &__get_cpu_var(softnet_data);
1290 dev->next_sched = sd->output_queue;
1291 sd->output_queue = dev;
1292 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1293 local_irq_restore(flags);
1296 EXPORT_SYMBOL(__netif_schedule);
1298 void dev_kfree_skb_irq(struct sk_buff *skb)
1300 if (atomic_dec_and_test(&skb->users)) {
1301 struct softnet_data *sd;
1302 unsigned long flags;
1304 local_irq_save(flags);
1305 sd = &__get_cpu_var(softnet_data);
1306 skb->next = sd->completion_queue;
1307 sd->completion_queue = skb;
1308 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1309 local_irq_restore(flags);
1312 EXPORT_SYMBOL(dev_kfree_skb_irq);
1314 void dev_kfree_skb_any(struct sk_buff *skb)
1316 if (in_irq() || irqs_disabled())
1317 dev_kfree_skb_irq(skb);
1321 EXPORT_SYMBOL(dev_kfree_skb_any);
1325 * netif_device_detach - mark device as removed
1326 * @dev: network device
1328 * Mark device as removed from system and therefore no longer available.
1330 void netif_device_detach(struct net_device *dev)
1332 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1333 netif_running(dev)) {
1334 netif_stop_queue(dev);
1337 EXPORT_SYMBOL(netif_device_detach);
1340 * netif_device_attach - mark device as attached
1341 * @dev: network device
1343 * Mark device as attached from system and restart if needed.
1345 void netif_device_attach(struct net_device *dev)
1347 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1348 netif_running(dev)) {
1349 netif_wake_queue(dev);
1350 __netdev_watchdog_up(dev);
1353 EXPORT_SYMBOL(netif_device_attach);
1357 * Invalidate hardware checksum when packet is to be mangled, and
1358 * complete checksum manually on outgoing path.
1360 int skb_checksum_help(struct sk_buff *skb)
1363 int ret = 0, offset;
1365 if (skb->ip_summed == CHECKSUM_COMPLETE)
1366 goto out_set_summed;
1368 if (unlikely(skb_shinfo(skb)->gso_size)) {
1369 /* Let GSO fix up the checksum. */
1370 goto out_set_summed;
1373 if (skb_cloned(skb)) {
1374 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1379 offset = skb->csum_start - skb_headroom(skb);
1380 BUG_ON(offset > (int)skb->len);
1381 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1383 offset = skb_headlen(skb) - offset;
1384 BUG_ON(offset <= 0);
1385 BUG_ON(skb->csum_offset + 2 > offset);
1387 *(__sum16 *)(skb->head + skb->csum_start + skb->csum_offset) =
1390 skb->ip_summed = CHECKSUM_NONE;
1396 * skb_gso_segment - Perform segmentation on skb.
1397 * @skb: buffer to segment
1398 * @features: features for the output path (see dev->features)
1400 * This function segments the given skb and returns a list of segments.
1402 * It may return NULL if the skb requires no segmentation. This is
1403 * only possible when GSO is used for verifying header integrity.
1405 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1407 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1408 struct packet_type *ptype;
1409 __be16 type = skb->protocol;
1412 BUG_ON(skb_shinfo(skb)->frag_list);
1414 skb_reset_mac_header(skb);
1415 skb->mac_len = skb->network_header - skb->mac_header;
1416 __skb_pull(skb, skb->mac_len);
1418 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1419 if (skb_header_cloned(skb) &&
1420 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1421 return ERR_PTR(err);
1425 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1426 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1427 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1428 err = ptype->gso_send_check(skb);
1429 segs = ERR_PTR(err);
1430 if (err || skb_gso_ok(skb, features))
1432 __skb_push(skb, (skb->data -
1433 skb_network_header(skb)));
1435 segs = ptype->gso_segment(skb, features);
1441 __skb_push(skb, skb->data - skb_mac_header(skb));
1446 EXPORT_SYMBOL(skb_gso_segment);
1448 /* Take action when hardware reception checksum errors are detected. */
1450 void netdev_rx_csum_fault(struct net_device *dev)
1452 if (net_ratelimit()) {
1453 printk(KERN_ERR "%s: hw csum failure.\n",
1454 dev ? dev->name : "<unknown>");
1458 EXPORT_SYMBOL(netdev_rx_csum_fault);
1461 /* Actually, we should eliminate this check as soon as we know, that:
1462 * 1. IOMMU is present and allows to map all the memory.
1463 * 2. No high memory really exists on this machine.
1466 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1468 #ifdef CONFIG_HIGHMEM
1471 if (dev->features & NETIF_F_HIGHDMA)
1474 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1475 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1483 void (*destructor)(struct sk_buff *skb);
1486 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1488 static void dev_gso_skb_destructor(struct sk_buff *skb)
1490 struct dev_gso_cb *cb;
1493 struct sk_buff *nskb = skb->next;
1495 skb->next = nskb->next;
1498 } while (skb->next);
1500 cb = DEV_GSO_CB(skb);
1502 cb->destructor(skb);
1506 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1507 * @skb: buffer to segment
1509 * This function segments the given skb and stores the list of segments
1512 static int dev_gso_segment(struct sk_buff *skb)
1514 struct net_device *dev = skb->dev;
1515 struct sk_buff *segs;
1516 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1519 segs = skb_gso_segment(skb, features);
1521 /* Verifying header integrity only. */
1525 if (unlikely(IS_ERR(segs)))
1526 return PTR_ERR(segs);
1529 DEV_GSO_CB(skb)->destructor = skb->destructor;
1530 skb->destructor = dev_gso_skb_destructor;
1535 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1537 if (likely(!skb->next)) {
1538 if (!list_empty(&ptype_all))
1539 dev_queue_xmit_nit(skb, dev);
1541 if (netif_needs_gso(dev, skb)) {
1542 if (unlikely(dev_gso_segment(skb)))
1548 return dev->hard_start_xmit(skb, dev);
1553 struct sk_buff *nskb = skb->next;
1556 skb->next = nskb->next;
1558 rc = dev->hard_start_xmit(nskb, dev);
1560 nskb->next = skb->next;
1564 if (unlikely((netif_queue_stopped(dev) ||
1565 netif_subqueue_stopped(dev, skb->queue_mapping)) &&
1567 return NETDEV_TX_BUSY;
1568 } while (skb->next);
1570 skb->destructor = DEV_GSO_CB(skb)->destructor;
1577 #define HARD_TX_LOCK(dev, cpu) { \
1578 if ((dev->features & NETIF_F_LLTX) == 0) { \
1579 netif_tx_lock(dev); \
1583 #define HARD_TX_UNLOCK(dev) { \
1584 if ((dev->features & NETIF_F_LLTX) == 0) { \
1585 netif_tx_unlock(dev); \
1590 * dev_queue_xmit - transmit a buffer
1591 * @skb: buffer to transmit
1593 * Queue a buffer for transmission to a network device. The caller must
1594 * have set the device and priority and built the buffer before calling
1595 * this function. The function can be called from an interrupt.
1597 * A negative errno code is returned on a failure. A success does not
1598 * guarantee the frame will be transmitted as it may be dropped due
1599 * to congestion or traffic shaping.
1601 * -----------------------------------------------------------------------------------
1602 * I notice this method can also return errors from the queue disciplines,
1603 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1606 * Regardless of the return value, the skb is consumed, so it is currently
1607 * difficult to retry a send to this method. (You can bump the ref count
1608 * before sending to hold a reference for retry if you are careful.)
1610 * When calling this method, interrupts MUST be enabled. This is because
1611 * the BH enable code must have IRQs enabled so that it will not deadlock.
1615 int dev_queue_xmit(struct sk_buff *skb)
1617 struct net_device *dev = skb->dev;
1621 /* GSO will handle the following emulations directly. */
1622 if (netif_needs_gso(dev, skb))
1625 if (skb_shinfo(skb)->frag_list &&
1626 !(dev->features & NETIF_F_FRAGLIST) &&
1627 __skb_linearize(skb))
1630 /* Fragmented skb is linearized if device does not support SG,
1631 * or if at least one of fragments is in highmem and device
1632 * does not support DMA from it.
1634 if (skb_shinfo(skb)->nr_frags &&
1635 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1636 __skb_linearize(skb))
1639 /* If packet is not checksummed and device does not support
1640 * checksumming for this protocol, complete checksumming here.
1642 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1643 skb_set_transport_header(skb, skb->csum_start -
1646 if (!(dev->features & NETIF_F_GEN_CSUM) &&
1647 !((dev->features & NETIF_F_IP_CSUM) &&
1648 skb->protocol == htons(ETH_P_IP)) &&
1649 !((dev->features & NETIF_F_IPV6_CSUM) &&
1650 skb->protocol == htons(ETH_P_IPV6)))
1651 if (skb_checksum_help(skb))
1656 spin_lock_prefetch(&dev->queue_lock);
1658 /* Disable soft irqs for various locks below. Also
1659 * stops preemption for RCU.
1663 /* Updates of qdisc are serialized by queue_lock.
1664 * The struct Qdisc which is pointed to by qdisc is now a
1665 * rcu structure - it may be accessed without acquiring
1666 * a lock (but the structure may be stale.) The freeing of the
1667 * qdisc will be deferred until it's known that there are no
1668 * more references to it.
1670 * If the qdisc has an enqueue function, we still need to
1671 * hold the queue_lock before calling it, since queue_lock
1672 * also serializes access to the device queue.
1675 q = rcu_dereference(dev->qdisc);
1676 #ifdef CONFIG_NET_CLS_ACT
1677 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1680 /* Grab device queue */
1681 spin_lock(&dev->queue_lock);
1684 /* reset queue_mapping to zero */
1685 skb->queue_mapping = 0;
1686 rc = q->enqueue(skb, q);
1688 spin_unlock(&dev->queue_lock);
1690 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1693 spin_unlock(&dev->queue_lock);
1696 /* The device has no queue. Common case for software devices:
1697 loopback, all the sorts of tunnels...
1699 Really, it is unlikely that netif_tx_lock protection is necessary
1700 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1702 However, it is possible, that they rely on protection
1705 Check this and shot the lock. It is not prone from deadlocks.
1706 Either shot noqueue qdisc, it is even simpler 8)
1708 if (dev->flags & IFF_UP) {
1709 int cpu = smp_processor_id(); /* ok because BHs are off */
1711 if (dev->xmit_lock_owner != cpu) {
1713 HARD_TX_LOCK(dev, cpu);
1715 if (!netif_queue_stopped(dev) &&
1716 !netif_subqueue_stopped(dev, skb->queue_mapping)) {
1718 if (!dev_hard_start_xmit(skb, dev)) {
1719 HARD_TX_UNLOCK(dev);
1723 HARD_TX_UNLOCK(dev);
1724 if (net_ratelimit())
1725 printk(KERN_CRIT "Virtual device %s asks to "
1726 "queue packet!\n", dev->name);
1728 /* Recursion is detected! It is possible,
1730 if (net_ratelimit())
1731 printk(KERN_CRIT "Dead loop on virtual device "
1732 "%s, fix it urgently!\n", dev->name);
1737 rcu_read_unlock_bh();
1743 rcu_read_unlock_bh();
1748 /*=======================================================================
1750 =======================================================================*/
1752 int netdev_max_backlog __read_mostly = 1000;
1753 int netdev_budget __read_mostly = 300;
1754 int weight_p __read_mostly = 64; /* old backlog weight */
1756 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1760 * netif_rx - post buffer to the network code
1761 * @skb: buffer to post
1763 * This function receives a packet from a device driver and queues it for
1764 * the upper (protocol) levels to process. It always succeeds. The buffer
1765 * may be dropped during processing for congestion control or by the
1769 * NET_RX_SUCCESS (no congestion)
1770 * NET_RX_CN_LOW (low congestion)
1771 * NET_RX_CN_MOD (moderate congestion)
1772 * NET_RX_CN_HIGH (high congestion)
1773 * NET_RX_DROP (packet was dropped)
1777 int netif_rx(struct sk_buff *skb)
1779 struct softnet_data *queue;
1780 unsigned long flags;
1782 /* if netpoll wants it, pretend we never saw it */
1783 if (netpoll_rx(skb))
1786 if (!skb->tstamp.tv64)
1790 * The code is rearranged so that the path is the most
1791 * short when CPU is congested, but is still operating.
1793 local_irq_save(flags);
1794 queue = &__get_cpu_var(softnet_data);
1796 __get_cpu_var(netdev_rx_stat).total++;
1797 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1798 if (queue->input_pkt_queue.qlen) {
1801 __skb_queue_tail(&queue->input_pkt_queue, skb);
1802 local_irq_restore(flags);
1803 return NET_RX_SUCCESS;
1806 napi_schedule(&queue->backlog);
1810 __get_cpu_var(netdev_rx_stat).dropped++;
1811 local_irq_restore(flags);
1817 int netif_rx_ni(struct sk_buff *skb)
1822 err = netif_rx(skb);
1823 if (local_softirq_pending())
1830 EXPORT_SYMBOL(netif_rx_ni);
1832 static inline struct net_device *skb_bond(struct sk_buff *skb)
1834 struct net_device *dev = skb->dev;
1837 if (skb_bond_should_drop(skb)) {
1841 skb->dev = dev->master;
1848 static void net_tx_action(struct softirq_action *h)
1850 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1852 if (sd->completion_queue) {
1853 struct sk_buff *clist;
1855 local_irq_disable();
1856 clist = sd->completion_queue;
1857 sd->completion_queue = NULL;
1861 struct sk_buff *skb = clist;
1862 clist = clist->next;
1864 BUG_TRAP(!atomic_read(&skb->users));
1869 if (sd->output_queue) {
1870 struct net_device *head;
1872 local_irq_disable();
1873 head = sd->output_queue;
1874 sd->output_queue = NULL;
1878 struct net_device *dev = head;
1879 head = head->next_sched;
1881 smp_mb__before_clear_bit();
1882 clear_bit(__LINK_STATE_SCHED, &dev->state);
1884 if (spin_trylock(&dev->queue_lock)) {
1886 spin_unlock(&dev->queue_lock);
1888 netif_schedule(dev);
1894 static inline int deliver_skb(struct sk_buff *skb,
1895 struct packet_type *pt_prev,
1896 struct net_device *orig_dev)
1898 atomic_inc(&skb->users);
1899 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1902 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1903 /* These hooks defined here for ATM */
1905 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1906 unsigned char *addr);
1907 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1910 * If bridge module is loaded call bridging hook.
1911 * returns NULL if packet was consumed.
1913 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
1914 struct sk_buff *skb) __read_mostly;
1915 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
1916 struct packet_type **pt_prev, int *ret,
1917 struct net_device *orig_dev)
1919 struct net_bridge_port *port;
1921 if (skb->pkt_type == PACKET_LOOPBACK ||
1922 (port = rcu_dereference(skb->dev->br_port)) == NULL)
1926 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1930 return br_handle_frame_hook(port, skb);
1933 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1936 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1937 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
1938 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
1940 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
1941 struct packet_type **pt_prev,
1943 struct net_device *orig_dev)
1945 if (skb->dev->macvlan_port == NULL)
1949 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1952 return macvlan_handle_frame_hook(skb);
1955 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1958 #ifdef CONFIG_NET_CLS_ACT
1959 /* TODO: Maybe we should just force sch_ingress to be compiled in
1960 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1961 * a compare and 2 stores extra right now if we dont have it on
1962 * but have CONFIG_NET_CLS_ACT
1963 * NOTE: This doesnt stop any functionality; if you dont have
1964 * the ingress scheduler, you just cant add policies on ingress.
1967 static int ing_filter(struct sk_buff *skb)
1970 struct net_device *dev = skb->dev;
1971 int result = TC_ACT_OK;
1973 if (dev->qdisc_ingress) {
1974 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1975 if (MAX_RED_LOOP < ttl++) {
1976 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1977 skb->iif, skb->dev->ifindex);
1981 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1983 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1985 spin_lock(&dev->ingress_lock);
1986 if ((q = dev->qdisc_ingress) != NULL)
1987 result = q->enqueue(skb, q);
1988 spin_unlock(&dev->ingress_lock);
1996 int netif_receive_skb(struct sk_buff *skb)
1998 struct packet_type *ptype, *pt_prev;
1999 struct net_device *orig_dev;
2000 int ret = NET_RX_DROP;
2003 /* if we've gotten here through NAPI, check netpoll */
2004 if (netpoll_receive_skb(skb))
2007 if (!skb->tstamp.tv64)
2011 skb->iif = skb->dev->ifindex;
2013 orig_dev = skb_bond(skb);
2018 __get_cpu_var(netdev_rx_stat).total++;
2020 skb_reset_network_header(skb);
2021 skb_reset_transport_header(skb);
2022 skb->mac_len = skb->network_header - skb->mac_header;
2028 #ifdef CONFIG_NET_CLS_ACT
2029 if (skb->tc_verd & TC_NCLS) {
2030 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2035 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2036 if (!ptype->dev || ptype->dev == skb->dev) {
2038 ret = deliver_skb(skb, pt_prev, orig_dev);
2043 #ifdef CONFIG_NET_CLS_ACT
2045 ret = deliver_skb(skb, pt_prev, orig_dev);
2046 pt_prev = NULL; /* noone else should process this after*/
2048 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2051 ret = ing_filter(skb);
2053 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
2062 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2065 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2069 type = skb->protocol;
2070 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
2071 if (ptype->type == type &&
2072 (!ptype->dev || ptype->dev == skb->dev)) {
2074 ret = deliver_skb(skb, pt_prev, orig_dev);
2080 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2083 /* Jamal, now you will not able to escape explaining
2084 * me how you were going to use this. :-)
2094 static int process_backlog(struct napi_struct *napi, int quota)
2097 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2098 unsigned long start_time = jiffies;
2100 napi->weight = weight_p;
2102 struct sk_buff *skb;
2103 struct net_device *dev;
2105 local_irq_disable();
2106 skb = __skb_dequeue(&queue->input_pkt_queue);
2108 __napi_complete(napi);
2117 netif_receive_skb(skb);
2120 } while (++work < quota && jiffies == start_time);
2126 * __napi_schedule - schedule for receive
2127 * @napi: entry to schedule
2129 * The entry's receive function will be scheduled to run
2131 void fastcall __napi_schedule(struct napi_struct *n)
2133 unsigned long flags;
2135 local_irq_save(flags);
2136 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2137 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2138 local_irq_restore(flags);
2140 EXPORT_SYMBOL(__napi_schedule);
2143 static void net_rx_action(struct softirq_action *h)
2145 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2146 unsigned long start_time = jiffies;
2147 int budget = netdev_budget;
2150 local_irq_disable();
2152 while (!list_empty(list)) {
2153 struct napi_struct *n;
2156 /* If softirq window is exhuasted then punt.
2158 * Note that this is a slight policy change from the
2159 * previous NAPI code, which would allow up to 2
2160 * jiffies to pass before breaking out. The test
2161 * used to be "jiffies - start_time > 1".
2163 if (unlikely(budget <= 0 || jiffies != start_time))
2168 /* Even though interrupts have been re-enabled, this
2169 * access is safe because interrupts can only add new
2170 * entries to the tail of this list, and only ->poll()
2171 * calls can remove this head entry from the list.
2173 n = list_entry(list->next, struct napi_struct, poll_list);
2175 have = netpoll_poll_lock(n);
2179 work = n->poll(n, weight);
2181 WARN_ON_ONCE(work > weight);
2185 local_irq_disable();
2187 /* Drivers must not modify the NAPI state if they
2188 * consume the entire weight. In such cases this code
2189 * still "owns" the NAPI instance and therefore can
2190 * move the instance around on the list at-will.
2192 if (unlikely(work == weight))
2193 list_move_tail(&n->poll_list, list);
2195 netpoll_poll_unlock(have);
2200 #ifdef CONFIG_NET_DMA
2202 * There may not be any more sk_buffs coming right now, so push
2203 * any pending DMA copies to hardware
2205 if (!cpus_empty(net_dma.channel_mask)) {
2207 for_each_cpu_mask(chan_idx, net_dma.channel_mask) {
2208 struct dma_chan *chan = net_dma.channels[chan_idx];
2210 dma_async_memcpy_issue_pending(chan);
2218 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2219 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2223 static gifconf_func_t * gifconf_list [NPROTO];
2226 * register_gifconf - register a SIOCGIF handler
2227 * @family: Address family
2228 * @gifconf: Function handler
2230 * Register protocol dependent address dumping routines. The handler
2231 * that is passed must not be freed or reused until it has been replaced
2232 * by another handler.
2234 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2236 if (family >= NPROTO)
2238 gifconf_list[family] = gifconf;
2244 * Map an interface index to its name (SIOCGIFNAME)
2248 * We need this ioctl for efficient implementation of the
2249 * if_indextoname() function required by the IPv6 API. Without
2250 * it, we would have to search all the interfaces to find a
2254 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2256 struct net_device *dev;
2260 * Fetch the caller's info block.
2263 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2266 read_lock(&dev_base_lock);
2267 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2269 read_unlock(&dev_base_lock);
2273 strcpy(ifr.ifr_name, dev->name);
2274 read_unlock(&dev_base_lock);
2276 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2282 * Perform a SIOCGIFCONF call. This structure will change
2283 * size eventually, and there is nothing I can do about it.
2284 * Thus we will need a 'compatibility mode'.
2287 static int dev_ifconf(struct net *net, char __user *arg)
2290 struct net_device *dev;
2297 * Fetch the caller's info block.
2300 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2307 * Loop over the interfaces, and write an info block for each.
2311 for_each_netdev(net, dev) {
2312 for (i = 0; i < NPROTO; i++) {
2313 if (gifconf_list[i]) {
2316 done = gifconf_list[i](dev, NULL, 0);
2318 done = gifconf_list[i](dev, pos + total,
2328 * All done. Write the updated control block back to the caller.
2330 ifc.ifc_len = total;
2333 * Both BSD and Solaris return 0 here, so we do too.
2335 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2338 #ifdef CONFIG_PROC_FS
2340 * This is invoked by the /proc filesystem handler to display a device
2343 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2345 struct net *net = seq->private;
2347 struct net_device *dev;
2349 read_lock(&dev_base_lock);
2351 return SEQ_START_TOKEN;
2354 for_each_netdev(net, dev)
2361 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2363 struct net *net = seq->private;
2365 return v == SEQ_START_TOKEN ?
2366 first_net_device(net) : next_net_device((struct net_device *)v);
2369 void dev_seq_stop(struct seq_file *seq, void *v)
2371 read_unlock(&dev_base_lock);
2374 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2376 struct net_device_stats *stats = dev->get_stats(dev);
2378 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2379 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2380 dev->name, stats->rx_bytes, stats->rx_packets,
2382 stats->rx_dropped + stats->rx_missed_errors,
2383 stats->rx_fifo_errors,
2384 stats->rx_length_errors + stats->rx_over_errors +
2385 stats->rx_crc_errors + stats->rx_frame_errors,
2386 stats->rx_compressed, stats->multicast,
2387 stats->tx_bytes, stats->tx_packets,
2388 stats->tx_errors, stats->tx_dropped,
2389 stats->tx_fifo_errors, stats->collisions,
2390 stats->tx_carrier_errors +
2391 stats->tx_aborted_errors +
2392 stats->tx_window_errors +
2393 stats->tx_heartbeat_errors,
2394 stats->tx_compressed);
2398 * Called from the PROCfs module. This now uses the new arbitrary sized
2399 * /proc/net interface to create /proc/net/dev
2401 static int dev_seq_show(struct seq_file *seq, void *v)
2403 if (v == SEQ_START_TOKEN)
2404 seq_puts(seq, "Inter-| Receive "
2406 " face |bytes packets errs drop fifo frame "
2407 "compressed multicast|bytes packets errs "
2408 "drop fifo colls carrier compressed\n");
2410 dev_seq_printf_stats(seq, v);
2414 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2416 struct netif_rx_stats *rc = NULL;
2418 while (*pos < NR_CPUS)
2419 if (cpu_online(*pos)) {
2420 rc = &per_cpu(netdev_rx_stat, *pos);
2427 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2429 return softnet_get_online(pos);
2432 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2435 return softnet_get_online(pos);
2438 static void softnet_seq_stop(struct seq_file *seq, void *v)
2442 static int softnet_seq_show(struct seq_file *seq, void *v)
2444 struct netif_rx_stats *s = v;
2446 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2447 s->total, s->dropped, s->time_squeeze, 0,
2448 0, 0, 0, 0, /* was fastroute */
2453 static const struct seq_operations dev_seq_ops = {
2454 .start = dev_seq_start,
2455 .next = dev_seq_next,
2456 .stop = dev_seq_stop,
2457 .show = dev_seq_show,
2460 static int dev_seq_open(struct inode *inode, struct file *file)
2462 struct seq_file *seq;
2464 res = seq_open(file, &dev_seq_ops);
2466 seq = file->private_data;
2467 seq->private = get_net(PROC_NET(inode));
2472 static int dev_seq_release(struct inode *inode, struct file *file)
2474 struct seq_file *seq = file->private_data;
2475 struct net *net = seq->private;
2477 return seq_release(inode, file);
2480 static const struct file_operations dev_seq_fops = {
2481 .owner = THIS_MODULE,
2482 .open = dev_seq_open,
2484 .llseek = seq_lseek,
2485 .release = dev_seq_release,
2488 static const struct seq_operations softnet_seq_ops = {
2489 .start = softnet_seq_start,
2490 .next = softnet_seq_next,
2491 .stop = softnet_seq_stop,
2492 .show = softnet_seq_show,
2495 static int softnet_seq_open(struct inode *inode, struct file *file)
2497 return seq_open(file, &softnet_seq_ops);
2500 static const struct file_operations softnet_seq_fops = {
2501 .owner = THIS_MODULE,
2502 .open = softnet_seq_open,
2504 .llseek = seq_lseek,
2505 .release = seq_release,
2508 static void *ptype_get_idx(loff_t pos)
2510 struct packet_type *pt = NULL;
2514 list_for_each_entry_rcu(pt, &ptype_all, list) {
2520 for (t = 0; t < 16; t++) {
2521 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2530 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2533 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2536 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2538 struct packet_type *pt;
2539 struct list_head *nxt;
2543 if (v == SEQ_START_TOKEN)
2544 return ptype_get_idx(0);
2547 nxt = pt->list.next;
2548 if (pt->type == htons(ETH_P_ALL)) {
2549 if (nxt != &ptype_all)
2552 nxt = ptype_base[0].next;
2554 hash = ntohs(pt->type) & 15;
2556 while (nxt == &ptype_base[hash]) {
2559 nxt = ptype_base[hash].next;
2562 return list_entry(nxt, struct packet_type, list);
2565 static void ptype_seq_stop(struct seq_file *seq, void *v)
2570 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2572 #ifdef CONFIG_KALLSYMS
2573 unsigned long offset = 0, symsize;
2574 const char *symname;
2578 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2585 modname = delim = "";
2586 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2592 seq_printf(seq, "[%p]", sym);
2595 static int ptype_seq_show(struct seq_file *seq, void *v)
2597 struct packet_type *pt = v;
2599 if (v == SEQ_START_TOKEN)
2600 seq_puts(seq, "Type Device Function\n");
2602 if (pt->type == htons(ETH_P_ALL))
2603 seq_puts(seq, "ALL ");
2605 seq_printf(seq, "%04x", ntohs(pt->type));
2607 seq_printf(seq, " %-8s ",
2608 pt->dev ? pt->dev->name : "");
2609 ptype_seq_decode(seq, pt->func);
2610 seq_putc(seq, '\n');
2616 static const struct seq_operations ptype_seq_ops = {
2617 .start = ptype_seq_start,
2618 .next = ptype_seq_next,
2619 .stop = ptype_seq_stop,
2620 .show = ptype_seq_show,
2623 static int ptype_seq_open(struct inode *inode, struct file *file)
2625 return seq_open(file, &ptype_seq_ops);
2628 static const struct file_operations ptype_seq_fops = {
2629 .owner = THIS_MODULE,
2630 .open = ptype_seq_open,
2632 .llseek = seq_lseek,
2633 .release = seq_release,
2637 static int dev_proc_net_init(struct net *net)
2641 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2643 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2645 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2648 if (wext_proc_init(net))
2654 proc_net_remove(net, "ptype");
2656 proc_net_remove(net, "softnet_stat");
2658 proc_net_remove(net, "dev");
2662 static void dev_proc_net_exit(struct net *net)
2664 wext_proc_exit(net);
2666 proc_net_remove(net, "ptype");
2667 proc_net_remove(net, "softnet_stat");
2668 proc_net_remove(net, "dev");
2671 static struct pernet_operations dev_proc_ops = {
2672 .init = dev_proc_net_init,
2673 .exit = dev_proc_net_exit,
2676 static int __init dev_proc_init(void)
2678 return register_pernet_subsys(&dev_proc_ops);
2681 #define dev_proc_init() 0
2682 #endif /* CONFIG_PROC_FS */
2686 * netdev_set_master - set up master/slave pair
2687 * @slave: slave device
2688 * @master: new master device
2690 * Changes the master device of the slave. Pass %NULL to break the
2691 * bonding. The caller must hold the RTNL semaphore. On a failure
2692 * a negative errno code is returned. On success the reference counts
2693 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2694 * function returns zero.
2696 int netdev_set_master(struct net_device *slave, struct net_device *master)
2698 struct net_device *old = slave->master;
2708 slave->master = master;
2716 slave->flags |= IFF_SLAVE;
2718 slave->flags &= ~IFF_SLAVE;
2720 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2724 static void __dev_set_promiscuity(struct net_device *dev, int inc)
2726 unsigned short old_flags = dev->flags;
2730 if ((dev->promiscuity += inc) == 0)
2731 dev->flags &= ~IFF_PROMISC;
2733 dev->flags |= IFF_PROMISC;
2734 if (dev->flags != old_flags) {
2735 printk(KERN_INFO "device %s %s promiscuous mode\n",
2736 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2738 audit_log(current->audit_context, GFP_ATOMIC,
2739 AUDIT_ANOM_PROMISCUOUS,
2740 "dev=%s prom=%d old_prom=%d auid=%u",
2741 dev->name, (dev->flags & IFF_PROMISC),
2742 (old_flags & IFF_PROMISC),
2743 audit_get_loginuid(current->audit_context));
2745 if (dev->change_rx_flags)
2746 dev->change_rx_flags(dev, IFF_PROMISC);
2751 * dev_set_promiscuity - update promiscuity count on a device
2755 * Add or remove promiscuity from a device. While the count in the device
2756 * remains above zero the interface remains promiscuous. Once it hits zero
2757 * the device reverts back to normal filtering operation. A negative inc
2758 * value is used to drop promiscuity on the device.
2760 void dev_set_promiscuity(struct net_device *dev, int inc)
2762 unsigned short old_flags = dev->flags;
2764 __dev_set_promiscuity(dev, inc);
2765 if (dev->flags != old_flags)
2766 dev_set_rx_mode(dev);
2770 * dev_set_allmulti - update allmulti count on a device
2774 * Add or remove reception of all multicast frames to a device. While the
2775 * count in the device remains above zero the interface remains listening
2776 * to all interfaces. Once it hits zero the device reverts back to normal
2777 * filtering operation. A negative @inc value is used to drop the counter
2778 * when releasing a resource needing all multicasts.
2781 void dev_set_allmulti(struct net_device *dev, int inc)
2783 unsigned short old_flags = dev->flags;
2787 dev->flags |= IFF_ALLMULTI;
2788 if ((dev->allmulti += inc) == 0)
2789 dev->flags &= ~IFF_ALLMULTI;
2790 if (dev->flags ^ old_flags) {
2791 if (dev->change_rx_flags)
2792 dev->change_rx_flags(dev, IFF_ALLMULTI);
2793 dev_set_rx_mode(dev);
2798 * Upload unicast and multicast address lists to device and
2799 * configure RX filtering. When the device doesn't support unicast
2800 * filtering it is put in promiscous mode while unicast addresses
2803 void __dev_set_rx_mode(struct net_device *dev)
2805 /* dev_open will call this function so the list will stay sane. */
2806 if (!(dev->flags&IFF_UP))
2809 if (!netif_device_present(dev))
2812 if (dev->set_rx_mode)
2813 dev->set_rx_mode(dev);
2815 /* Unicast addresses changes may only happen under the rtnl,
2816 * therefore calling __dev_set_promiscuity here is safe.
2818 if (dev->uc_count > 0 && !dev->uc_promisc) {
2819 __dev_set_promiscuity(dev, 1);
2820 dev->uc_promisc = 1;
2821 } else if (dev->uc_count == 0 && dev->uc_promisc) {
2822 __dev_set_promiscuity(dev, -1);
2823 dev->uc_promisc = 0;
2826 if (dev->set_multicast_list)
2827 dev->set_multicast_list(dev);
2831 void dev_set_rx_mode(struct net_device *dev)
2833 netif_tx_lock_bh(dev);
2834 __dev_set_rx_mode(dev);
2835 netif_tx_unlock_bh(dev);
2838 int __dev_addr_delete(struct dev_addr_list **list, int *count,
2839 void *addr, int alen, int glbl)
2841 struct dev_addr_list *da;
2843 for (; (da = *list) != NULL; list = &da->next) {
2844 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2845 alen == da->da_addrlen) {
2847 int old_glbl = da->da_gusers;
2864 int __dev_addr_add(struct dev_addr_list **list, int *count,
2865 void *addr, int alen, int glbl)
2867 struct dev_addr_list *da;
2869 for (da = *list; da != NULL; da = da->next) {
2870 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2871 da->da_addrlen == alen) {
2873 int old_glbl = da->da_gusers;
2883 da = kmalloc(sizeof(*da), GFP_ATOMIC);
2886 memcpy(da->da_addr, addr, alen);
2887 da->da_addrlen = alen;
2889 da->da_gusers = glbl ? 1 : 0;
2897 * dev_unicast_delete - Release secondary unicast address.
2899 * @addr: address to delete
2900 * @alen: length of @addr
2902 * Release reference to a secondary unicast address and remove it
2903 * from the device if the reference count drops to zero.
2905 * The caller must hold the rtnl_mutex.
2907 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
2913 netif_tx_lock_bh(dev);
2914 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2916 __dev_set_rx_mode(dev);
2917 netif_tx_unlock_bh(dev);
2920 EXPORT_SYMBOL(dev_unicast_delete);
2923 * dev_unicast_add - add a secondary unicast address
2925 * @addr: address to delete
2926 * @alen: length of @addr
2928 * Add a secondary unicast address to the device or increase
2929 * the reference count if it already exists.
2931 * The caller must hold the rtnl_mutex.
2933 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
2939 netif_tx_lock_bh(dev);
2940 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2942 __dev_set_rx_mode(dev);
2943 netif_tx_unlock_bh(dev);
2946 EXPORT_SYMBOL(dev_unicast_add);
2948 static void __dev_addr_discard(struct dev_addr_list **list)
2950 struct dev_addr_list *tmp;
2952 while (*list != NULL) {
2955 if (tmp->da_users > tmp->da_gusers)
2956 printk("__dev_addr_discard: address leakage! "
2957 "da_users=%d\n", tmp->da_users);
2962 static void dev_addr_discard(struct net_device *dev)
2964 netif_tx_lock_bh(dev);
2966 __dev_addr_discard(&dev->uc_list);
2969 __dev_addr_discard(&dev->mc_list);
2972 netif_tx_unlock_bh(dev);
2975 unsigned dev_get_flags(const struct net_device *dev)
2979 flags = (dev->flags & ~(IFF_PROMISC |
2984 (dev->gflags & (IFF_PROMISC |
2987 if (netif_running(dev)) {
2988 if (netif_oper_up(dev))
2989 flags |= IFF_RUNNING;
2990 if (netif_carrier_ok(dev))
2991 flags |= IFF_LOWER_UP;
2992 if (netif_dormant(dev))
2993 flags |= IFF_DORMANT;
2999 int dev_change_flags(struct net_device *dev, unsigned flags)
3002 int old_flags = dev->flags;
3007 * Set the flags on our device.
3010 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3011 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3013 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3017 * Load in the correct multicast list now the flags have changed.
3020 if (dev->change_rx_flags && (dev->flags ^ flags) & IFF_MULTICAST)
3021 dev->change_rx_flags(dev, IFF_MULTICAST);
3023 dev_set_rx_mode(dev);
3026 * Have we downed the interface. We handle IFF_UP ourselves
3027 * according to user attempts to set it, rather than blindly
3032 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3033 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3036 dev_set_rx_mode(dev);
3039 if (dev->flags & IFF_UP &&
3040 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3042 raw_notifier_call_chain(&netdev_chain,
3043 NETDEV_CHANGE, dev);
3045 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3046 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3047 dev->gflags ^= IFF_PROMISC;
3048 dev_set_promiscuity(dev, inc);
3051 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3052 is important. Some (broken) drivers set IFF_PROMISC, when
3053 IFF_ALLMULTI is requested not asking us and not reporting.
3055 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3056 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3057 dev->gflags ^= IFF_ALLMULTI;
3058 dev_set_allmulti(dev, inc);
3061 /* Exclude state transition flags, already notified */
3062 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3064 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3069 int dev_set_mtu(struct net_device *dev, int new_mtu)
3073 if (new_mtu == dev->mtu)
3076 /* MTU must be positive. */
3080 if (!netif_device_present(dev))
3084 if (dev->change_mtu)
3085 err = dev->change_mtu(dev, new_mtu);
3088 if (!err && dev->flags & IFF_UP)
3089 raw_notifier_call_chain(&netdev_chain,
3090 NETDEV_CHANGEMTU, dev);
3094 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3098 if (!dev->set_mac_address)
3100 if (sa->sa_family != dev->type)
3102 if (!netif_device_present(dev))
3104 err = dev->set_mac_address(dev, sa);
3106 raw_notifier_call_chain(&netdev_chain,
3107 NETDEV_CHANGEADDR, dev);
3112 * Perform the SIOCxIFxxx calls.
3114 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3117 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3123 case SIOCGIFFLAGS: /* Get interface flags */
3124 ifr->ifr_flags = dev_get_flags(dev);
3127 case SIOCSIFFLAGS: /* Set interface flags */
3128 return dev_change_flags(dev, ifr->ifr_flags);
3130 case SIOCGIFMETRIC: /* Get the metric on the interface
3131 (currently unused) */
3132 ifr->ifr_metric = 0;
3135 case SIOCSIFMETRIC: /* Set the metric on the interface
3136 (currently unused) */
3139 case SIOCGIFMTU: /* Get the MTU of a device */
3140 ifr->ifr_mtu = dev->mtu;
3143 case SIOCSIFMTU: /* Set the MTU of a device */
3144 return dev_set_mtu(dev, ifr->ifr_mtu);
3148 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3150 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3151 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3152 ifr->ifr_hwaddr.sa_family = dev->type;
3156 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3158 case SIOCSIFHWBROADCAST:
3159 if (ifr->ifr_hwaddr.sa_family != dev->type)
3161 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3162 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3163 raw_notifier_call_chain(&netdev_chain,
3164 NETDEV_CHANGEADDR, dev);
3168 ifr->ifr_map.mem_start = dev->mem_start;
3169 ifr->ifr_map.mem_end = dev->mem_end;
3170 ifr->ifr_map.base_addr = dev->base_addr;
3171 ifr->ifr_map.irq = dev->irq;
3172 ifr->ifr_map.dma = dev->dma;
3173 ifr->ifr_map.port = dev->if_port;
3177 if (dev->set_config) {
3178 if (!netif_device_present(dev))
3180 return dev->set_config(dev, &ifr->ifr_map);
3185 if (!dev->set_multicast_list ||
3186 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3188 if (!netif_device_present(dev))
3190 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3194 if (!dev->set_multicast_list ||
3195 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3197 if (!netif_device_present(dev))
3199 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3203 ifr->ifr_ifindex = dev->ifindex;
3207 ifr->ifr_qlen = dev->tx_queue_len;
3211 if (ifr->ifr_qlen < 0)
3213 dev->tx_queue_len = ifr->ifr_qlen;
3217 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3218 return dev_change_name(dev, ifr->ifr_newname);
3221 * Unknown or private ioctl
3225 if ((cmd >= SIOCDEVPRIVATE &&
3226 cmd <= SIOCDEVPRIVATE + 15) ||
3227 cmd == SIOCBONDENSLAVE ||
3228 cmd == SIOCBONDRELEASE ||
3229 cmd == SIOCBONDSETHWADDR ||
3230 cmd == SIOCBONDSLAVEINFOQUERY ||
3231 cmd == SIOCBONDINFOQUERY ||
3232 cmd == SIOCBONDCHANGEACTIVE ||
3233 cmd == SIOCGMIIPHY ||
3234 cmd == SIOCGMIIREG ||
3235 cmd == SIOCSMIIREG ||
3236 cmd == SIOCBRADDIF ||
3237 cmd == SIOCBRDELIF ||
3238 cmd == SIOCWANDEV) {
3240 if (dev->do_ioctl) {
3241 if (netif_device_present(dev))
3242 err = dev->do_ioctl(dev, ifr,
3255 * This function handles all "interface"-type I/O control requests. The actual
3256 * 'doing' part of this is dev_ifsioc above.
3260 * dev_ioctl - network device ioctl
3261 * @cmd: command to issue
3262 * @arg: pointer to a struct ifreq in user space
3264 * Issue ioctl functions to devices. This is normally called by the
3265 * user space syscall interfaces but can sometimes be useful for
3266 * other purposes. The return value is the return from the syscall if
3267 * positive or a negative errno code on error.
3270 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3276 /* One special case: SIOCGIFCONF takes ifconf argument
3277 and requires shared lock, because it sleeps writing
3281 if (cmd == SIOCGIFCONF) {
3283 ret = dev_ifconf(net, (char __user *) arg);
3287 if (cmd == SIOCGIFNAME)
3288 return dev_ifname(net, (struct ifreq __user *)arg);
3290 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3293 ifr.ifr_name[IFNAMSIZ-1] = 0;
3295 colon = strchr(ifr.ifr_name, ':');
3300 * See which interface the caller is talking about.
3305 * These ioctl calls:
3306 * - can be done by all.
3307 * - atomic and do not require locking.
3318 dev_load(net, ifr.ifr_name);
3319 read_lock(&dev_base_lock);
3320 ret = dev_ifsioc(net, &ifr, cmd);
3321 read_unlock(&dev_base_lock);
3325 if (copy_to_user(arg, &ifr,
3326 sizeof(struct ifreq)))
3332 dev_load(net, ifr.ifr_name);
3334 ret = dev_ethtool(net, &ifr);
3339 if (copy_to_user(arg, &ifr,
3340 sizeof(struct ifreq)))
3346 * These ioctl calls:
3347 * - require superuser power.
3348 * - require strict serialization.
3354 if (!capable(CAP_NET_ADMIN))
3356 dev_load(net, ifr.ifr_name);
3358 ret = dev_ifsioc(net, &ifr, cmd);
3363 if (copy_to_user(arg, &ifr,
3364 sizeof(struct ifreq)))
3370 * These ioctl calls:
3371 * - require superuser power.
3372 * - require strict serialization.
3373 * - do not return a value
3383 case SIOCSIFHWBROADCAST:
3386 case SIOCBONDENSLAVE:
3387 case SIOCBONDRELEASE:
3388 case SIOCBONDSETHWADDR:
3389 case SIOCBONDCHANGEACTIVE:
3392 if (!capable(CAP_NET_ADMIN))
3395 case SIOCBONDSLAVEINFOQUERY:
3396 case SIOCBONDINFOQUERY:
3397 dev_load(net, ifr.ifr_name);
3399 ret = dev_ifsioc(net, &ifr, cmd);
3404 /* Get the per device memory space. We can add this but
3405 * currently do not support it */
3407 /* Set the per device memory buffer space.
3408 * Not applicable in our case */
3413 * Unknown or private ioctl.
3416 if (cmd == SIOCWANDEV ||
3417 (cmd >= SIOCDEVPRIVATE &&
3418 cmd <= SIOCDEVPRIVATE + 15)) {
3419 dev_load(net, ifr.ifr_name);
3421 ret = dev_ifsioc(net, &ifr, cmd);
3423 if (!ret && copy_to_user(arg, &ifr,
3424 sizeof(struct ifreq)))
3428 /* Take care of Wireless Extensions */
3429 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3430 return wext_handle_ioctl(net, &ifr, cmd, arg);
3437 * dev_new_index - allocate an ifindex
3439 * Returns a suitable unique value for a new device interface
3440 * number. The caller must hold the rtnl semaphore or the
3441 * dev_base_lock to be sure it remains unique.
3443 static int dev_new_index(struct net *net)
3449 if (!__dev_get_by_index(net, ifindex))
3454 /* Delayed registration/unregisteration */
3455 static DEFINE_SPINLOCK(net_todo_list_lock);
3456 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
3458 static void net_set_todo(struct net_device *dev)
3460 spin_lock(&net_todo_list_lock);
3461 list_add_tail(&dev->todo_list, &net_todo_list);
3462 spin_unlock(&net_todo_list_lock);
3466 * register_netdevice - register a network device
3467 * @dev: device to register
3469 * Take a completed network device structure and add it to the kernel
3470 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3471 * chain. 0 is returned on success. A negative errno code is returned
3472 * on a failure to set up the device, or if the name is a duplicate.
3474 * Callers must hold the rtnl semaphore. You may want
3475 * register_netdev() instead of this.
3478 * The locking appears insufficient to guarantee two parallel registers
3479 * will not get the same name.
3482 int register_netdevice(struct net_device *dev)
3484 struct hlist_head *head;
3485 struct hlist_node *p;
3489 BUG_ON(dev_boot_phase);
3494 /* When net_device's are persistent, this will be fatal. */
3495 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3496 BUG_ON(!dev->nd_net);
3499 spin_lock_init(&dev->queue_lock);
3500 spin_lock_init(&dev->_xmit_lock);
3501 netdev_set_lockdep_class(&dev->_xmit_lock, dev->type);
3502 dev->xmit_lock_owner = -1;
3503 spin_lock_init(&dev->ingress_lock);
3507 /* Init, if this function is available */
3509 ret = dev->init(dev);
3517 if (!dev_valid_name(dev->name)) {
3522 dev->ifindex = dev_new_index(net);
3523 if (dev->iflink == -1)
3524 dev->iflink = dev->ifindex;
3526 /* Check for existence of name */
3527 head = dev_name_hash(net, dev->name);
3528 hlist_for_each(p, head) {
3529 struct net_device *d
3530 = hlist_entry(p, struct net_device, name_hlist);
3531 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3537 /* Fix illegal checksum combinations */
3538 if ((dev->features & NETIF_F_HW_CSUM) &&
3539 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3540 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3542 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3545 if ((dev->features & NETIF_F_NO_CSUM) &&
3546 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3547 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3549 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3553 /* Fix illegal SG+CSUM combinations. */
3554 if ((dev->features & NETIF_F_SG) &&
3555 !(dev->features & NETIF_F_ALL_CSUM)) {
3556 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3558 dev->features &= ~NETIF_F_SG;
3561 /* TSO requires that SG is present as well. */
3562 if ((dev->features & NETIF_F_TSO) &&
3563 !(dev->features & NETIF_F_SG)) {
3564 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3566 dev->features &= ~NETIF_F_TSO;
3568 if (dev->features & NETIF_F_UFO) {
3569 if (!(dev->features & NETIF_F_HW_CSUM)) {
3570 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3571 "NETIF_F_HW_CSUM feature.\n",
3573 dev->features &= ~NETIF_F_UFO;
3575 if (!(dev->features & NETIF_F_SG)) {
3576 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3577 "NETIF_F_SG feature.\n",
3579 dev->features &= ~NETIF_F_UFO;
3584 * nil rebuild_header routine,
3585 * that should be never called and used as just bug trap.
3588 if (!dev->rebuild_header)
3589 dev->rebuild_header = default_rebuild_header;
3591 ret = netdev_register_sysfs(dev);
3594 dev->reg_state = NETREG_REGISTERED;
3597 * Default initial state at registry is that the
3598 * device is present.
3601 set_bit(__LINK_STATE_PRESENT, &dev->state);
3603 dev_init_scheduler(dev);
3605 list_netdevice(dev);
3607 /* Notify protocols, that a new device appeared. */
3608 ret = raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3609 ret = notifier_to_errno(ret);
3611 unregister_netdevice(dev);
3623 * register_netdev - register a network device
3624 * @dev: device to register
3626 * Take a completed network device structure and add it to the kernel
3627 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3628 * chain. 0 is returned on success. A negative errno code is returned
3629 * on a failure to set up the device, or if the name is a duplicate.
3631 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3632 * and expands the device name if you passed a format string to
3635 int register_netdev(struct net_device *dev)
3642 * If the name is a format string the caller wants us to do a
3645 if (strchr(dev->name, '%')) {
3646 err = dev_alloc_name(dev, dev->name);
3651 err = register_netdevice(dev);
3656 EXPORT_SYMBOL(register_netdev);
3659 * netdev_wait_allrefs - wait until all references are gone.
3661 * This is called when unregistering network devices.
3663 * Any protocol or device that holds a reference should register
3664 * for netdevice notification, and cleanup and put back the
3665 * reference if they receive an UNREGISTER event.
3666 * We can get stuck here if buggy protocols don't correctly
3669 static void netdev_wait_allrefs(struct net_device *dev)
3671 unsigned long rebroadcast_time, warning_time;
3673 rebroadcast_time = warning_time = jiffies;
3674 while (atomic_read(&dev->refcnt) != 0) {
3675 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3678 /* Rebroadcast unregister notification */
3679 raw_notifier_call_chain(&netdev_chain,
3680 NETDEV_UNREGISTER, dev);
3682 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3684 /* We must not have linkwatch events
3685 * pending on unregister. If this
3686 * happens, we simply run the queue
3687 * unscheduled, resulting in a noop
3690 linkwatch_run_queue();
3695 rebroadcast_time = jiffies;
3700 if (time_after(jiffies, warning_time + 10 * HZ)) {
3701 printk(KERN_EMERG "unregister_netdevice: "
3702 "waiting for %s to become free. Usage "
3704 dev->name, atomic_read(&dev->refcnt));
3705 warning_time = jiffies;
3714 * register_netdevice(x1);
3715 * register_netdevice(x2);
3717 * unregister_netdevice(y1);
3718 * unregister_netdevice(y2);
3724 * We are invoked by rtnl_unlock() after it drops the semaphore.
3725 * This allows us to deal with problems:
3726 * 1) We can delete sysfs objects which invoke hotplug
3727 * without deadlocking with linkwatch via keventd.
3728 * 2) Since we run with the RTNL semaphore not held, we can sleep
3729 * safely in order to wait for the netdev refcnt to drop to zero.
3731 static DEFINE_MUTEX(net_todo_run_mutex);
3732 void netdev_run_todo(void)
3734 struct list_head list;
3736 /* Need to guard against multiple cpu's getting out of order. */
3737 mutex_lock(&net_todo_run_mutex);
3739 /* Not safe to do outside the semaphore. We must not return
3740 * until all unregister events invoked by the local processor
3741 * have been completed (either by this todo run, or one on
3744 if (list_empty(&net_todo_list))
3747 /* Snapshot list, allow later requests */
3748 spin_lock(&net_todo_list_lock);
3749 list_replace_init(&net_todo_list, &list);
3750 spin_unlock(&net_todo_list_lock);
3752 while (!list_empty(&list)) {
3753 struct net_device *dev
3754 = list_entry(list.next, struct net_device, todo_list);
3755 list_del(&dev->todo_list);
3757 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3758 printk(KERN_ERR "network todo '%s' but state %d\n",
3759 dev->name, dev->reg_state);
3764 dev->reg_state = NETREG_UNREGISTERED;
3766 netdev_wait_allrefs(dev);
3769 BUG_ON(atomic_read(&dev->refcnt));
3770 BUG_TRAP(!dev->ip_ptr);
3771 BUG_TRAP(!dev->ip6_ptr);
3772 BUG_TRAP(!dev->dn_ptr);
3774 if (dev->destructor)
3775 dev->destructor(dev);
3777 /* Free network device */
3778 kobject_put(&dev->dev.kobj);
3782 mutex_unlock(&net_todo_run_mutex);
3785 static struct net_device_stats *internal_stats(struct net_device *dev)
3791 * alloc_netdev_mq - allocate network device
3792 * @sizeof_priv: size of private data to allocate space for
3793 * @name: device name format string
3794 * @setup: callback to initialize device
3795 * @queue_count: the number of subqueues to allocate
3797 * Allocates a struct net_device with private data area for driver use
3798 * and performs basic initialization. Also allocates subquue structs
3799 * for each queue on the device at the end of the netdevice.
3801 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
3802 void (*setup)(struct net_device *), unsigned int queue_count)
3805 struct net_device *dev;
3808 BUG_ON(strlen(name) >= sizeof(dev->name));
3810 /* ensure 32-byte alignment of both the device and private area */
3811 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST +
3812 (sizeof(struct net_device_subqueue) * (queue_count - 1))) &
3813 ~NETDEV_ALIGN_CONST;
3814 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3816 p = kzalloc(alloc_size, GFP_KERNEL);
3818 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
3822 dev = (struct net_device *)
3823 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3824 dev->padded = (char *)dev - (char *)p;
3825 dev->nd_net = &init_net;
3828 dev->priv = ((char *)dev +
3829 ((sizeof(struct net_device) +
3830 (sizeof(struct net_device_subqueue) *
3831 (queue_count - 1)) + NETDEV_ALIGN_CONST)
3832 & ~NETDEV_ALIGN_CONST));
3835 dev->egress_subqueue_count = queue_count;
3837 dev->get_stats = internal_stats;
3838 netpoll_netdev_init(dev);
3840 strcpy(dev->name, name);
3843 EXPORT_SYMBOL(alloc_netdev_mq);
3846 * free_netdev - free network device
3849 * This function does the last stage of destroying an allocated device
3850 * interface. The reference to the device object is released.
3851 * If this is the last reference then it will be freed.
3853 void free_netdev(struct net_device *dev)
3856 /* Compatibility with error handling in drivers */
3857 if (dev->reg_state == NETREG_UNINITIALIZED) {
3858 kfree((char *)dev - dev->padded);
3862 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3863 dev->reg_state = NETREG_RELEASED;
3865 /* will free via device release */
3866 put_device(&dev->dev);
3868 kfree((char *)dev - dev->padded);
3872 /* Synchronize with packet receive processing. */
3873 void synchronize_net(void)
3880 * unregister_netdevice - remove device from the kernel
3883 * This function shuts down a device interface and removes it
3884 * from the kernel tables. On success 0 is returned, on a failure
3885 * a negative errno code is returned.
3887 * Callers must hold the rtnl semaphore. You may want
3888 * unregister_netdev() instead of this.
3891 void unregister_netdevice(struct net_device *dev)
3893 BUG_ON(dev_boot_phase);
3896 /* Some devices call without registering for initialization unwind. */
3897 if (dev->reg_state == NETREG_UNINITIALIZED) {
3898 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3899 "was registered\n", dev->name, dev);
3905 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3907 /* If device is running, close it first. */
3908 if (dev->flags & IFF_UP)
3911 /* And unlink it from device chain. */
3912 unlist_netdevice(dev);
3914 dev->reg_state = NETREG_UNREGISTERING;
3918 /* Shutdown queueing discipline. */
3922 /* Notify protocols, that we are about to destroy
3923 this device. They should clean all the things.
3925 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3928 * Flush the unicast and multicast chains
3930 dev_addr_discard(dev);
3935 /* Notifier chain MUST detach us from master device. */
3936 BUG_TRAP(!dev->master);
3938 /* Remove entries from sysfs */
3939 netdev_unregister_sysfs(dev);
3941 /* Finish processing unregister after unlock */
3950 * unregister_netdev - remove device from the kernel
3953 * This function shuts down a device interface and removes it
3954 * from the kernel tables. On success 0 is returned, on a failure
3955 * a negative errno code is returned.
3957 * This is just a wrapper for unregister_netdevice that takes
3958 * the rtnl semaphore. In general you want to use this and not
3959 * unregister_netdevice.
3961 void unregister_netdev(struct net_device *dev)
3964 unregister_netdevice(dev);
3968 EXPORT_SYMBOL(unregister_netdev);
3971 * dev_change_net_namespace - move device to different nethost namespace
3973 * @net: network namespace
3974 * @pat: If not NULL name pattern to try if the current device name
3975 * is already taken in the destination network namespace.
3977 * This function shuts down a device interface and moves it
3978 * to a new network namespace. On success 0 is returned, on
3979 * a failure a netagive errno code is returned.
3981 * Callers must hold the rtnl semaphore.
3984 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
3987 const char *destname;
3992 /* Don't allow namespace local devices to be moved. */
3994 if (dev->features & NETIF_F_NETNS_LOCAL)
3997 /* Ensure the device has been registrered */
3999 if (dev->reg_state != NETREG_REGISTERED)
4002 /* Get out if there is nothing todo */
4004 if (dev->nd_net == net)
4007 /* Pick the destination device name, and ensure
4008 * we can use it in the destination network namespace.
4011 destname = dev->name;
4012 if (__dev_get_by_name(net, destname)) {
4013 /* We get here if we can't use the current device name */
4016 if (!dev_valid_name(pat))
4018 if (strchr(pat, '%')) {
4019 if (__dev_alloc_name(net, pat, buf) < 0)
4024 if (__dev_get_by_name(net, destname))
4029 * And now a mini version of register_netdevice unregister_netdevice.
4032 /* If device is running close it first. */
4033 if (dev->flags & IFF_UP)
4036 /* And unlink it from device chain */
4038 unlist_netdevice(dev);
4042 /* Shutdown queueing discipline. */
4045 /* Notify protocols, that we are about to destroy
4046 this device. They should clean all the things.
4048 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4051 * Flush the unicast and multicast chains
4053 dev_addr_discard(dev);
4055 /* Actually switch the network namespace */
4058 /* Assign the new device name */
4059 if (destname != dev->name)
4060 strcpy(dev->name, destname);
4062 /* If there is an ifindex conflict assign a new one */
4063 if (__dev_get_by_index(net, dev->ifindex)) {
4064 int iflink = (dev->iflink == dev->ifindex);
4065 dev->ifindex = dev_new_index(net);
4067 dev->iflink = dev->ifindex;
4071 err = device_rename(&dev->dev, dev->name);
4074 /* Add the device back in the hashes */
4075 list_netdevice(dev);
4077 /* Notify protocols, that a new device appeared. */
4078 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4086 static int dev_cpu_callback(struct notifier_block *nfb,
4087 unsigned long action,
4090 struct sk_buff **list_skb;
4091 struct net_device **list_net;
4092 struct sk_buff *skb;
4093 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4094 struct softnet_data *sd, *oldsd;
4096 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4099 local_irq_disable();
4100 cpu = smp_processor_id();
4101 sd = &per_cpu(softnet_data, cpu);
4102 oldsd = &per_cpu(softnet_data, oldcpu);
4104 /* Find end of our completion_queue. */
4105 list_skb = &sd->completion_queue;
4107 list_skb = &(*list_skb)->next;
4108 /* Append completion queue from offline CPU. */
4109 *list_skb = oldsd->completion_queue;
4110 oldsd->completion_queue = NULL;
4112 /* Find end of our output_queue. */
4113 list_net = &sd->output_queue;
4115 list_net = &(*list_net)->next_sched;
4116 /* Append output queue from offline CPU. */
4117 *list_net = oldsd->output_queue;
4118 oldsd->output_queue = NULL;
4120 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4123 /* Process offline CPU's input_pkt_queue */
4124 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4130 #ifdef CONFIG_NET_DMA
4132 * net_dma_rebalance - try to maintain one DMA channel per CPU
4133 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4135 * This is called when the number of channels allocated to the net_dma client
4136 * changes. The net_dma client tries to have one DMA channel per CPU.
4139 static void net_dma_rebalance(struct net_dma *net_dma)
4141 unsigned int cpu, i, n, chan_idx;
4142 struct dma_chan *chan;
4144 if (cpus_empty(net_dma->channel_mask)) {
4145 for_each_online_cpu(cpu)
4146 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4151 cpu = first_cpu(cpu_online_map);
4153 for_each_cpu_mask(chan_idx, net_dma->channel_mask) {
4154 chan = net_dma->channels[chan_idx];
4156 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4157 + (i < (num_online_cpus() %
4158 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4161 per_cpu(softnet_data, cpu).net_dma = chan;
4162 cpu = next_cpu(cpu, cpu_online_map);
4170 * netdev_dma_event - event callback for the net_dma_client
4171 * @client: should always be net_dma_client
4172 * @chan: DMA channel for the event
4173 * @state: DMA state to be handled
4175 static enum dma_state_client
4176 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4177 enum dma_state state)
4179 int i, found = 0, pos = -1;
4180 struct net_dma *net_dma =
4181 container_of(client, struct net_dma, client);
4182 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4184 spin_lock(&net_dma->lock);
4186 case DMA_RESOURCE_AVAILABLE:
4187 for (i = 0; i < NR_CPUS; i++)
4188 if (net_dma->channels[i] == chan) {
4191 } else if (net_dma->channels[i] == NULL && pos < 0)
4194 if (!found && pos >= 0) {
4196 net_dma->channels[pos] = chan;
4197 cpu_set(pos, net_dma->channel_mask);
4198 net_dma_rebalance(net_dma);
4201 case DMA_RESOURCE_REMOVED:
4202 for (i = 0; i < NR_CPUS; i++)
4203 if (net_dma->channels[i] == chan) {
4211 cpu_clear(pos, net_dma->channel_mask);
4212 net_dma->channels[i] = NULL;
4213 net_dma_rebalance(net_dma);
4219 spin_unlock(&net_dma->lock);
4225 * netdev_dma_regiser - register the networking subsystem as a DMA client
4227 static int __init netdev_dma_register(void)
4229 spin_lock_init(&net_dma.lock);
4230 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4231 dma_async_client_register(&net_dma.client);
4232 dma_async_client_chan_request(&net_dma.client);
4237 static int __init netdev_dma_register(void) { return -ENODEV; }
4238 #endif /* CONFIG_NET_DMA */
4241 * netdev_compute_feature - compute conjunction of two feature sets
4242 * @all: first feature set
4243 * @one: second feature set
4245 * Computes a new feature set after adding a device with feature set
4246 * @one to the master device with current feature set @all. Returns
4247 * the new feature set.
4249 int netdev_compute_features(unsigned long all, unsigned long one)
4251 /* if device needs checksumming, downgrade to hw checksumming */
4252 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4253 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4255 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4256 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4257 all ^= NETIF_F_HW_CSUM
4258 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4260 if (one & NETIF_F_GSO)
4261 one |= NETIF_F_GSO_SOFTWARE;
4264 /* If even one device supports robust GSO, enable it for all. */
4265 if (one & NETIF_F_GSO_ROBUST)
4266 all |= NETIF_F_GSO_ROBUST;
4268 all &= one | NETIF_F_LLTX;
4270 if (!(all & NETIF_F_ALL_CSUM))
4272 if (!(all & NETIF_F_SG))
4273 all &= ~NETIF_F_GSO_MASK;
4277 EXPORT_SYMBOL(netdev_compute_features);
4279 /* Initialize per network namespace state */
4280 static int netdev_init(struct net *net)
4283 INIT_LIST_HEAD(&net->dev_base_head);
4284 rwlock_init(&dev_base_lock);
4286 net->dev_name_head = kmalloc(
4287 sizeof(*net->dev_name_head)*NETDEV_HASHENTRIES, GFP_KERNEL);
4288 if (!net->dev_name_head)
4291 net->dev_index_head = kmalloc(
4292 sizeof(*net->dev_index_head)*NETDEV_HASHENTRIES, GFP_KERNEL);
4293 if (!net->dev_index_head) {
4294 kfree(net->dev_name_head);
4298 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4299 INIT_HLIST_HEAD(&net->dev_name_head[i]);
4301 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4302 INIT_HLIST_HEAD(&net->dev_index_head[i]);
4307 static void netdev_exit(struct net *net)
4309 kfree(net->dev_name_head);
4310 kfree(net->dev_index_head);
4313 static struct pernet_operations netdev_net_ops = {
4314 .init = netdev_init,
4315 .exit = netdev_exit,
4318 static void default_device_exit(struct net *net)
4320 struct net_device *dev, *next;
4322 * Push all migratable of the network devices back to the
4323 * initial network namespace
4326 for_each_netdev_safe(net, dev, next) {
4329 /* Ignore unmoveable devices (i.e. loopback) */
4330 if (dev->features & NETIF_F_NETNS_LOCAL)
4333 /* Push remaing network devices to init_net */
4334 err = dev_change_net_namespace(dev, &init_net, "dev%d");
4336 printk(KERN_WARNING "%s: failed to move %s to init_net: %d\n",
4337 __func__, dev->name, err);
4338 unregister_netdevice(dev);
4344 static struct pernet_operations default_device_ops = {
4345 .exit = default_device_exit,
4349 * Initialize the DEV module. At boot time this walks the device list and
4350 * unhooks any devices that fail to initialise (normally hardware not
4351 * present) and leaves us with a valid list of present and active devices.
4356 * This is called single threaded during boot, so no need
4357 * to take the rtnl semaphore.
4359 static int __init net_dev_init(void)
4361 int i, rc = -ENOMEM;
4363 BUG_ON(!dev_boot_phase);
4365 if (dev_proc_init())
4368 if (netdev_sysfs_init())
4371 INIT_LIST_HEAD(&ptype_all);
4372 for (i = 0; i < 16; i++)
4373 INIT_LIST_HEAD(&ptype_base[i]);
4375 if (register_pernet_subsys(&netdev_net_ops))
4378 if (register_pernet_device(&default_device_ops))
4382 * Initialise the packet receive queues.
4385 for_each_possible_cpu(i) {
4386 struct softnet_data *queue;
4388 queue = &per_cpu(softnet_data, i);
4389 skb_queue_head_init(&queue->input_pkt_queue);
4390 queue->completion_queue = NULL;
4391 INIT_LIST_HEAD(&queue->poll_list);
4393 queue->backlog.poll = process_backlog;
4394 queue->backlog.weight = weight_p;
4397 netdev_dma_register();
4401 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
4402 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
4404 hotcpu_notifier(dev_cpu_callback, 0);
4412 subsys_initcall(net_dev_init);
4414 EXPORT_SYMBOL(__dev_get_by_index);
4415 EXPORT_SYMBOL(__dev_get_by_name);
4416 EXPORT_SYMBOL(__dev_remove_pack);
4417 EXPORT_SYMBOL(dev_valid_name);
4418 EXPORT_SYMBOL(dev_add_pack);
4419 EXPORT_SYMBOL(dev_alloc_name);
4420 EXPORT_SYMBOL(dev_close);
4421 EXPORT_SYMBOL(dev_get_by_flags);
4422 EXPORT_SYMBOL(dev_get_by_index);
4423 EXPORT_SYMBOL(dev_get_by_name);
4424 EXPORT_SYMBOL(dev_open);
4425 EXPORT_SYMBOL(dev_queue_xmit);
4426 EXPORT_SYMBOL(dev_remove_pack);
4427 EXPORT_SYMBOL(dev_set_allmulti);
4428 EXPORT_SYMBOL(dev_set_promiscuity);
4429 EXPORT_SYMBOL(dev_change_flags);
4430 EXPORT_SYMBOL(dev_set_mtu);
4431 EXPORT_SYMBOL(dev_set_mac_address);
4432 EXPORT_SYMBOL(free_netdev);
4433 EXPORT_SYMBOL(netdev_boot_setup_check);
4434 EXPORT_SYMBOL(netdev_set_master);
4435 EXPORT_SYMBOL(netdev_state_change);
4436 EXPORT_SYMBOL(netif_receive_skb);
4437 EXPORT_SYMBOL(netif_rx);
4438 EXPORT_SYMBOL(register_gifconf);
4439 EXPORT_SYMBOL(register_netdevice);
4440 EXPORT_SYMBOL(register_netdevice_notifier);
4441 EXPORT_SYMBOL(skb_checksum_help);
4442 EXPORT_SYMBOL(synchronize_net);
4443 EXPORT_SYMBOL(unregister_netdevice);
4444 EXPORT_SYMBOL(unregister_netdevice_notifier);
4445 EXPORT_SYMBOL(net_enable_timestamp);
4446 EXPORT_SYMBOL(net_disable_timestamp);
4447 EXPORT_SYMBOL(dev_get_flags);
4449 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4450 EXPORT_SYMBOL(br_handle_frame_hook);
4451 EXPORT_SYMBOL(br_fdb_get_hook);
4452 EXPORT_SYMBOL(br_fdb_put_hook);
4456 EXPORT_SYMBOL(dev_load);
4459 EXPORT_PER_CPU_SYMBOL(softnet_data);