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>
122 #include <linux/if_vlan.h>
124 #include "net-sysfs.h"
127 * The list of packet types we will receive (as opposed to discard)
128 * and the routines to invoke.
130 * Why 16. Because with 16 the only overlap we get on a hash of the
131 * low nibble of the protocol value is RARP/SNAP/X.25.
133 * NOTE: That is no longer true with the addition of VLAN tags. Not
134 * sure which should go first, but I bet it won't make much
135 * difference if we are running VLANs. The good news is that
136 * this protocol won't be in the list unless compiled in, so
137 * the average user (w/out VLANs) will not be adversely affected.
154 #define PTYPE_HASH_SIZE (16)
155 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
157 static DEFINE_SPINLOCK(ptype_lock);
158 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
159 static struct list_head ptype_all __read_mostly; /* Taps */
161 #ifdef CONFIG_NET_DMA
163 struct dma_client client;
165 cpumask_t channel_mask;
166 struct dma_chan **channels;
169 static enum dma_state_client
170 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
171 enum dma_state state);
173 static struct net_dma net_dma = {
175 .event_callback = netdev_dma_event,
181 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
184 * Pure readers hold dev_base_lock for reading.
186 * Writers must hold the rtnl semaphore while they loop through the
187 * dev_base_head list, and hold dev_base_lock for writing when they do the
188 * actual updates. This allows pure readers to access the list even
189 * while a writer is preparing to update it.
191 * To put it another way, dev_base_lock is held for writing only to
192 * protect against pure readers; the rtnl semaphore provides the
193 * protection against other writers.
195 * See, for example usages, register_netdevice() and
196 * unregister_netdevice(), which must be called with the rtnl
199 DEFINE_RWLOCK(dev_base_lock);
201 EXPORT_SYMBOL(dev_base_lock);
203 #define NETDEV_HASHBITS 8
204 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
206 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
208 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
209 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
212 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
214 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
217 /* Device list insertion */
218 static int list_netdevice(struct net_device *dev)
220 struct net *net = dev_net(dev);
224 write_lock_bh(&dev_base_lock);
225 list_add_tail(&dev->dev_list, &net->dev_base_head);
226 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
227 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
228 write_unlock_bh(&dev_base_lock);
232 /* Device list removal */
233 static void unlist_netdevice(struct net_device *dev)
237 /* Unlink dev from the device chain */
238 write_lock_bh(&dev_base_lock);
239 list_del(&dev->dev_list);
240 hlist_del(&dev->name_hlist);
241 hlist_del(&dev->index_hlist);
242 write_unlock_bh(&dev_base_lock);
249 static RAW_NOTIFIER_HEAD(netdev_chain);
252 * Device drivers call our routines to queue packets here. We empty the
253 * queue in the local softnet handler.
256 DEFINE_PER_CPU(struct softnet_data, softnet_data);
258 #ifdef CONFIG_DEBUG_LOCK_ALLOC
260 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
261 * according to dev->type
263 static const unsigned short netdev_lock_type[] =
264 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
265 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
266 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
267 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
268 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
269 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
270 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
271 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
272 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
273 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
274 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
275 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
276 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
277 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
280 static const char *netdev_lock_name[] =
281 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
282 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
283 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
284 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
285 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
286 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
287 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
288 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
289 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
290 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
291 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
292 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
293 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
294 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
297 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
299 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
303 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
304 if (netdev_lock_type[i] == dev_type)
306 /* the last key is used by default */
307 return ARRAY_SIZE(netdev_lock_type) - 1;
310 static inline void netdev_set_lockdep_class(spinlock_t *lock,
311 unsigned short dev_type)
315 i = netdev_lock_pos(dev_type);
316 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
317 netdev_lock_name[i]);
320 static inline void netdev_set_lockdep_class(spinlock_t *lock,
321 unsigned short dev_type)
326 /*******************************************************************************
328 Protocol management and registration routines
330 *******************************************************************************/
333 * Add a protocol ID to the list. Now that the input handler is
334 * smarter we can dispense with all the messy stuff that used to be
337 * BEWARE!!! Protocol handlers, mangling input packets,
338 * MUST BE last in hash buckets and checking protocol handlers
339 * MUST start from promiscuous ptype_all chain in net_bh.
340 * It is true now, do not change it.
341 * Explanation follows: if protocol handler, mangling packet, will
342 * be the first on list, it is not able to sense, that packet
343 * is cloned and should be copied-on-write, so that it will
344 * change it and subsequent readers will get broken packet.
349 * dev_add_pack - add packet handler
350 * @pt: packet type declaration
352 * Add a protocol handler to the networking stack. The passed &packet_type
353 * is linked into kernel lists and may not be freed until it has been
354 * removed from the kernel lists.
356 * This call does not sleep therefore it can not
357 * guarantee all CPU's that are in middle of receiving packets
358 * will see the new packet type (until the next received packet).
361 void dev_add_pack(struct packet_type *pt)
365 spin_lock_bh(&ptype_lock);
366 if (pt->type == htons(ETH_P_ALL))
367 list_add_rcu(&pt->list, &ptype_all);
369 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
370 list_add_rcu(&pt->list, &ptype_base[hash]);
372 spin_unlock_bh(&ptype_lock);
376 * __dev_remove_pack - remove packet handler
377 * @pt: packet type declaration
379 * Remove a protocol handler that was previously added to the kernel
380 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
381 * from the kernel lists and can be freed or reused once this function
384 * The packet type might still be in use by receivers
385 * and must not be freed until after all the CPU's have gone
386 * through a quiescent state.
388 void __dev_remove_pack(struct packet_type *pt)
390 struct list_head *head;
391 struct packet_type *pt1;
393 spin_lock_bh(&ptype_lock);
395 if (pt->type == htons(ETH_P_ALL))
398 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
400 list_for_each_entry(pt1, head, list) {
402 list_del_rcu(&pt->list);
407 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
409 spin_unlock_bh(&ptype_lock);
412 * dev_remove_pack - remove packet handler
413 * @pt: packet type declaration
415 * Remove a protocol handler that was previously added to the kernel
416 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
417 * from the kernel lists and can be freed or reused once this function
420 * This call sleeps to guarantee that no CPU is looking at the packet
423 void dev_remove_pack(struct packet_type *pt)
425 __dev_remove_pack(pt);
430 /******************************************************************************
432 Device Boot-time Settings Routines
434 *******************************************************************************/
436 /* Boot time configuration table */
437 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
440 * netdev_boot_setup_add - add new setup entry
441 * @name: name of the device
442 * @map: configured settings for the device
444 * Adds new setup entry to the dev_boot_setup list. The function
445 * returns 0 on error and 1 on success. This is a generic routine to
448 static int netdev_boot_setup_add(char *name, struct ifmap *map)
450 struct netdev_boot_setup *s;
454 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
455 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
456 memset(s[i].name, 0, sizeof(s[i].name));
457 strcpy(s[i].name, name);
458 memcpy(&s[i].map, map, sizeof(s[i].map));
463 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
467 * netdev_boot_setup_check - check boot time settings
468 * @dev: the netdevice
470 * Check boot time settings for the device.
471 * The found settings are set for the device to be used
472 * later in the device probing.
473 * Returns 0 if no settings found, 1 if they are.
475 int netdev_boot_setup_check(struct net_device *dev)
477 struct netdev_boot_setup *s = dev_boot_setup;
480 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
481 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
482 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
483 dev->irq = s[i].map.irq;
484 dev->base_addr = s[i].map.base_addr;
485 dev->mem_start = s[i].map.mem_start;
486 dev->mem_end = s[i].map.mem_end;
495 * netdev_boot_base - get address from boot time settings
496 * @prefix: prefix for network device
497 * @unit: id for network device
499 * Check boot time settings for the base address of device.
500 * The found settings are set for the device to be used
501 * later in the device probing.
502 * Returns 0 if no settings found.
504 unsigned long netdev_boot_base(const char *prefix, int unit)
506 const struct netdev_boot_setup *s = dev_boot_setup;
510 sprintf(name, "%s%d", prefix, unit);
513 * If device already registered then return base of 1
514 * to indicate not to probe for this interface
516 if (__dev_get_by_name(&init_net, name))
519 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
520 if (!strcmp(name, s[i].name))
521 return s[i].map.base_addr;
526 * Saves at boot time configured settings for any netdevice.
528 int __init netdev_boot_setup(char *str)
533 str = get_options(str, ARRAY_SIZE(ints), ints);
538 memset(&map, 0, sizeof(map));
542 map.base_addr = ints[2];
544 map.mem_start = ints[3];
546 map.mem_end = ints[4];
548 /* Add new entry to the list */
549 return netdev_boot_setup_add(str, &map);
552 __setup("netdev=", netdev_boot_setup);
554 /*******************************************************************************
556 Device Interface Subroutines
558 *******************************************************************************/
561 * __dev_get_by_name - find a device by its name
562 * @net: the applicable net namespace
563 * @name: name to find
565 * Find an interface by name. Must be called under RTNL semaphore
566 * or @dev_base_lock. If the name is found a pointer to the device
567 * is returned. If the name is not found then %NULL is returned. The
568 * reference counters are not incremented so the caller must be
569 * careful with locks.
572 struct net_device *__dev_get_by_name(struct net *net, const char *name)
574 struct hlist_node *p;
576 hlist_for_each(p, dev_name_hash(net, name)) {
577 struct net_device *dev
578 = hlist_entry(p, struct net_device, name_hlist);
579 if (!strncmp(dev->name, name, IFNAMSIZ))
586 * dev_get_by_name - find a device by its name
587 * @net: the applicable net namespace
588 * @name: name to find
590 * Find an interface by name. This can be called from any
591 * context and does its own locking. The returned handle has
592 * the usage count incremented and the caller must use dev_put() to
593 * release it when it is no longer needed. %NULL is returned if no
594 * matching device is found.
597 struct net_device *dev_get_by_name(struct net *net, const char *name)
599 struct net_device *dev;
601 read_lock(&dev_base_lock);
602 dev = __dev_get_by_name(net, name);
605 read_unlock(&dev_base_lock);
610 * __dev_get_by_index - find a device by its ifindex
611 * @net: the applicable net namespace
612 * @ifindex: index of device
614 * Search for an interface by index. Returns %NULL if the device
615 * is not found or a pointer to the device. The device has not
616 * had its reference counter increased so the caller must be careful
617 * about locking. The caller must hold either the RTNL semaphore
621 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
623 struct hlist_node *p;
625 hlist_for_each(p, dev_index_hash(net, ifindex)) {
626 struct net_device *dev
627 = hlist_entry(p, struct net_device, index_hlist);
628 if (dev->ifindex == ifindex)
636 * dev_get_by_index - find a device by its ifindex
637 * @net: the applicable net namespace
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 * @net: the applicable net namespace
661 * @type: media type of device
662 * @ha: hardware address
664 * Search for an interface by MAC address. Returns NULL if the device
665 * is not found or a pointer to the device. The caller must hold the
666 * rtnl semaphore. The returned device has not had its ref count increased
667 * and the caller must therefore be careful about locking
670 * If the API was consistent this would be __dev_get_by_hwaddr
673 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
675 struct net_device *dev;
679 for_each_netdev(net, dev)
680 if (dev->type == type &&
681 !memcmp(dev->dev_addr, ha, dev->addr_len))
687 EXPORT_SYMBOL(dev_getbyhwaddr);
689 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
691 struct net_device *dev;
694 for_each_netdev(net, dev)
695 if (dev->type == type)
701 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
703 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
705 struct net_device *dev;
708 dev = __dev_getfirstbyhwtype(net, type);
715 EXPORT_SYMBOL(dev_getfirstbyhwtype);
718 * dev_get_by_flags - find any device with given flags
719 * @net: the applicable net namespace
720 * @if_flags: IFF_* values
721 * @mask: bitmask of bits in if_flags to check
723 * Search for any interface with the given flags. Returns NULL if a device
724 * is not found or a pointer to the device. The device returned has
725 * had a reference added and the pointer is safe until the user calls
726 * dev_put to indicate they have finished with it.
729 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
731 struct net_device *dev, *ret;
734 read_lock(&dev_base_lock);
735 for_each_netdev(net, dev) {
736 if (((dev->flags ^ if_flags) & mask) == 0) {
742 read_unlock(&dev_base_lock);
747 * dev_valid_name - check if name is okay for network device
750 * Network device names need to be valid file names to
751 * to allow sysfs to work. We also disallow any kind of
754 int dev_valid_name(const char *name)
758 if (strlen(name) >= IFNAMSIZ)
760 if (!strcmp(name, ".") || !strcmp(name, ".."))
764 if (*name == '/' || isspace(*name))
772 * __dev_alloc_name - allocate a name for a device
773 * @net: network namespace to allocate the device name in
774 * @name: name format string
775 * @buf: scratch buffer and result name string
777 * Passed a format string - eg "lt%d" it will try and find a suitable
778 * id. It scans list of devices to build up a free map, then chooses
779 * the first empty slot. The caller must hold the dev_base or rtnl lock
780 * while allocating the name and adding the device in order to avoid
782 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
783 * Returns the number of the unit assigned or a negative errno code.
786 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
790 const int max_netdevices = 8*PAGE_SIZE;
791 unsigned long *inuse;
792 struct net_device *d;
794 p = strnchr(name, IFNAMSIZ-1, '%');
797 * Verify the string as this thing may have come from
798 * the user. There must be either one "%d" and no other "%"
801 if (p[1] != 'd' || strchr(p + 2, '%'))
804 /* Use one page as a bit array of possible slots */
805 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
809 for_each_netdev(net, d) {
810 if (!sscanf(d->name, name, &i))
812 if (i < 0 || i >= max_netdevices)
815 /* avoid cases where sscanf is not exact inverse of printf */
816 snprintf(buf, IFNAMSIZ, name, i);
817 if (!strncmp(buf, d->name, IFNAMSIZ))
821 i = find_first_zero_bit(inuse, max_netdevices);
822 free_page((unsigned long) inuse);
825 snprintf(buf, IFNAMSIZ, name, i);
826 if (!__dev_get_by_name(net, buf))
829 /* It is possible to run out of possible slots
830 * when the name is long and there isn't enough space left
831 * for the digits, or if all bits are used.
837 * dev_alloc_name - allocate a name for a device
839 * @name: name format string
841 * Passed a format string - eg "lt%d" it will try and find a suitable
842 * id. It scans list of devices to build up a free map, then chooses
843 * the first empty slot. The caller must hold the dev_base or rtnl lock
844 * while allocating the name and adding the device in order to avoid
846 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
847 * Returns the number of the unit assigned or a negative errno code.
850 int dev_alloc_name(struct net_device *dev, const char *name)
856 BUG_ON(!dev_net(dev));
858 ret = __dev_alloc_name(net, name, buf);
860 strlcpy(dev->name, buf, IFNAMSIZ);
866 * dev_change_name - change name of a device
868 * @newname: name (or format string) must be at least IFNAMSIZ
870 * Change name of a device, can pass format strings "eth%d".
873 int dev_change_name(struct net_device *dev, char *newname)
875 char oldname[IFNAMSIZ];
881 BUG_ON(!dev_net(dev));
884 if (dev->flags & IFF_UP)
887 if (!dev_valid_name(newname))
890 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
893 memcpy(oldname, dev->name, IFNAMSIZ);
895 if (strchr(newname, '%')) {
896 err = dev_alloc_name(dev, newname);
899 strcpy(newname, dev->name);
901 else if (__dev_get_by_name(net, newname))
904 strlcpy(dev->name, newname, IFNAMSIZ);
907 err = device_rename(&dev->dev, dev->name);
909 memcpy(dev->name, oldname, IFNAMSIZ);
913 write_lock_bh(&dev_base_lock);
914 hlist_del(&dev->name_hlist);
915 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
916 write_unlock_bh(&dev_base_lock);
918 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
919 ret = notifier_to_errno(ret);
924 "%s: name change rollback failed: %d.\n",
928 memcpy(dev->name, oldname, IFNAMSIZ);
937 * netdev_features_change - device changes features
938 * @dev: device to cause notification
940 * Called to indicate a device has changed features.
942 void netdev_features_change(struct net_device *dev)
944 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
946 EXPORT_SYMBOL(netdev_features_change);
949 * netdev_state_change - device changes state
950 * @dev: device to cause notification
952 * Called to indicate a device has changed state. This function calls
953 * the notifier chains for netdev_chain and sends a NEWLINK message
954 * to the routing socket.
956 void netdev_state_change(struct net_device *dev)
958 if (dev->flags & IFF_UP) {
959 call_netdevice_notifiers(NETDEV_CHANGE, dev);
960 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
964 void netdev_bonding_change(struct net_device *dev)
966 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
968 EXPORT_SYMBOL(netdev_bonding_change);
971 * dev_load - load a network module
972 * @net: the applicable net namespace
973 * @name: name of interface
975 * If a network interface is not present and the process has suitable
976 * privileges this function loads the module. If module loading is not
977 * available in this kernel then it becomes a nop.
980 void dev_load(struct net *net, const char *name)
982 struct net_device *dev;
984 read_lock(&dev_base_lock);
985 dev = __dev_get_by_name(net, name);
986 read_unlock(&dev_base_lock);
988 if (!dev && capable(CAP_SYS_MODULE))
989 request_module("%s", name);
993 * dev_open - prepare an interface for use.
994 * @dev: device to open
996 * Takes a device from down to up state. The device's private open
997 * function is invoked and then the multicast lists are loaded. Finally
998 * the device is moved into the up state and a %NETDEV_UP message is
999 * sent to the netdev notifier chain.
1001 * Calling this function on an active interface is a nop. On a failure
1002 * a negative errno code is returned.
1004 int dev_open(struct net_device *dev)
1014 if (dev->flags & IFF_UP)
1018 * Is it even present?
1020 if (!netif_device_present(dev))
1024 * Call device private open method
1026 set_bit(__LINK_STATE_START, &dev->state);
1028 if (dev->validate_addr)
1029 ret = dev->validate_addr(dev);
1031 if (!ret && dev->open)
1032 ret = dev->open(dev);
1035 * If it went open OK then:
1039 clear_bit(__LINK_STATE_START, &dev->state);
1044 dev->flags |= IFF_UP;
1047 * Initialize multicasting status
1049 dev_set_rx_mode(dev);
1052 * Wakeup transmit queue engine
1057 * ... and announce new interface.
1059 call_netdevice_notifiers(NETDEV_UP, dev);
1066 * dev_close - shutdown an interface.
1067 * @dev: device to shutdown
1069 * This function moves an active device into down state. A
1070 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1071 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1074 int dev_close(struct net_device *dev)
1080 if (!(dev->flags & IFF_UP))
1084 * Tell people we are going down, so that they can
1085 * prepare to death, when device is still operating.
1087 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1089 clear_bit(__LINK_STATE_START, &dev->state);
1091 /* Synchronize to scheduled poll. We cannot touch poll list,
1092 * it can be even on different cpu. So just clear netif_running().
1094 * dev->stop() will invoke napi_disable() on all of it's
1095 * napi_struct instances on this device.
1097 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1099 dev_deactivate(dev);
1102 * Call the device specific close. This cannot fail.
1103 * Only if device is UP
1105 * We allow it to be called even after a DETACH hot-plug
1112 * Device is now down.
1115 dev->flags &= ~IFF_UP;
1118 * Tell people we are down
1120 call_netdevice_notifiers(NETDEV_DOWN, dev);
1126 static int dev_boot_phase = 1;
1129 * Device change register/unregister. These are not inline or static
1130 * as we export them to the world.
1134 * register_netdevice_notifier - register a network notifier block
1137 * Register a notifier to be called when network device events occur.
1138 * The notifier passed is linked into the kernel structures and must
1139 * not be reused until it has been unregistered. A negative errno code
1140 * is returned on a failure.
1142 * When registered all registration and up events are replayed
1143 * to the new notifier to allow device to have a race free
1144 * view of the network device list.
1147 int register_netdevice_notifier(struct notifier_block *nb)
1149 struct net_device *dev;
1150 struct net_device *last;
1155 err = raw_notifier_chain_register(&netdev_chain, nb);
1161 for_each_netdev(net, dev) {
1162 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1163 err = notifier_to_errno(err);
1167 if (!(dev->flags & IFF_UP))
1170 nb->notifier_call(nb, NETDEV_UP, dev);
1181 for_each_netdev(net, dev) {
1185 if (dev->flags & IFF_UP) {
1186 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1187 nb->notifier_call(nb, NETDEV_DOWN, dev);
1189 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1193 raw_notifier_chain_unregister(&netdev_chain, nb);
1198 * unregister_netdevice_notifier - unregister a network notifier block
1201 * Unregister a notifier previously registered by
1202 * register_netdevice_notifier(). The notifier is unlinked into the
1203 * kernel structures and may then be reused. A negative errno code
1204 * is returned on a failure.
1207 int unregister_netdevice_notifier(struct notifier_block *nb)
1212 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1218 * call_netdevice_notifiers - call all network notifier blocks
1219 * @val: value passed unmodified to notifier function
1220 * @dev: net_device pointer passed unmodified to notifier function
1222 * Call all network notifier blocks. Parameters and return value
1223 * are as for raw_notifier_call_chain().
1226 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1228 return raw_notifier_call_chain(&netdev_chain, val, dev);
1231 /* When > 0 there are consumers of rx skb time stamps */
1232 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1234 void net_enable_timestamp(void)
1236 atomic_inc(&netstamp_needed);
1239 void net_disable_timestamp(void)
1241 atomic_dec(&netstamp_needed);
1244 static inline void net_timestamp(struct sk_buff *skb)
1246 if (atomic_read(&netstamp_needed))
1247 __net_timestamp(skb);
1249 skb->tstamp.tv64 = 0;
1253 * Support routine. Sends outgoing frames to any network
1254 * taps currently in use.
1257 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1259 struct packet_type *ptype;
1264 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1265 /* Never send packets back to the socket
1266 * they originated from - MvS (miquels@drinkel.ow.org)
1268 if ((ptype->dev == dev || !ptype->dev) &&
1269 (ptype->af_packet_priv == NULL ||
1270 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1271 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1275 /* skb->nh should be correctly
1276 set by sender, so that the second statement is
1277 just protection against buggy protocols.
1279 skb_reset_mac_header(skb2);
1281 if (skb_network_header(skb2) < skb2->data ||
1282 skb2->network_header > skb2->tail) {
1283 if (net_ratelimit())
1284 printk(KERN_CRIT "protocol %04x is "
1286 skb2->protocol, dev->name);
1287 skb_reset_network_header(skb2);
1290 skb2->transport_header = skb2->network_header;
1291 skb2->pkt_type = PACKET_OUTGOING;
1292 ptype->func(skb2, skb->dev, ptype, skb->dev);
1299 void __netif_schedule(struct net_device *dev)
1301 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1302 unsigned long flags;
1303 struct softnet_data *sd;
1305 local_irq_save(flags);
1306 sd = &__get_cpu_var(softnet_data);
1307 dev->next_sched = sd->output_queue;
1308 sd->output_queue = dev;
1309 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1310 local_irq_restore(flags);
1313 EXPORT_SYMBOL(__netif_schedule);
1315 void dev_kfree_skb_irq(struct sk_buff *skb)
1317 if (atomic_dec_and_test(&skb->users)) {
1318 struct softnet_data *sd;
1319 unsigned long flags;
1321 local_irq_save(flags);
1322 sd = &__get_cpu_var(softnet_data);
1323 skb->next = sd->completion_queue;
1324 sd->completion_queue = skb;
1325 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1326 local_irq_restore(flags);
1329 EXPORT_SYMBOL(dev_kfree_skb_irq);
1331 void dev_kfree_skb_any(struct sk_buff *skb)
1333 if (in_irq() || irqs_disabled())
1334 dev_kfree_skb_irq(skb);
1338 EXPORT_SYMBOL(dev_kfree_skb_any);
1342 * netif_device_detach - mark device as removed
1343 * @dev: network device
1345 * Mark device as removed from system and therefore no longer available.
1347 void netif_device_detach(struct net_device *dev)
1349 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1350 netif_running(dev)) {
1351 netif_stop_queue(dev);
1354 EXPORT_SYMBOL(netif_device_detach);
1357 * netif_device_attach - mark device as attached
1358 * @dev: network device
1360 * Mark device as attached from system and restart if needed.
1362 void netif_device_attach(struct net_device *dev)
1364 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1365 netif_running(dev)) {
1366 netif_wake_queue(dev);
1367 __netdev_watchdog_up(dev);
1370 EXPORT_SYMBOL(netif_device_attach);
1372 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1374 return ((features & NETIF_F_GEN_CSUM) ||
1375 ((features & NETIF_F_IP_CSUM) &&
1376 protocol == htons(ETH_P_IP)) ||
1377 ((features & NETIF_F_IPV6_CSUM) &&
1378 protocol == htons(ETH_P_IPV6)));
1381 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1383 if (can_checksum_protocol(dev->features, skb->protocol))
1386 if (skb->protocol == htons(ETH_P_8021Q)) {
1387 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1388 if (can_checksum_protocol(dev->features & dev->vlan_features,
1389 veh->h_vlan_encapsulated_proto))
1397 * Invalidate hardware checksum when packet is to be mangled, and
1398 * complete checksum manually on outgoing path.
1400 int skb_checksum_help(struct sk_buff *skb)
1403 int ret = 0, offset;
1405 if (skb->ip_summed == CHECKSUM_COMPLETE)
1406 goto out_set_summed;
1408 if (unlikely(skb_shinfo(skb)->gso_size)) {
1409 /* Let GSO fix up the checksum. */
1410 goto out_set_summed;
1413 offset = skb->csum_start - skb_headroom(skb);
1414 BUG_ON(offset >= skb_headlen(skb));
1415 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1417 offset += skb->csum_offset;
1418 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1420 if (skb_cloned(skb) &&
1421 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1422 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1427 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1429 skb->ip_summed = CHECKSUM_NONE;
1435 * skb_gso_segment - Perform segmentation on skb.
1436 * @skb: buffer to segment
1437 * @features: features for the output path (see dev->features)
1439 * This function segments the given skb and returns a list of segments.
1441 * It may return NULL if the skb requires no segmentation. This is
1442 * only possible when GSO is used for verifying header integrity.
1444 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1446 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1447 struct packet_type *ptype;
1448 __be16 type = skb->protocol;
1451 BUG_ON(skb_shinfo(skb)->frag_list);
1453 skb_reset_mac_header(skb);
1454 skb->mac_len = skb->network_header - skb->mac_header;
1455 __skb_pull(skb, skb->mac_len);
1457 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1458 if (skb_header_cloned(skb) &&
1459 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1460 return ERR_PTR(err);
1464 list_for_each_entry_rcu(ptype,
1465 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1466 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1467 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1468 err = ptype->gso_send_check(skb);
1469 segs = ERR_PTR(err);
1470 if (err || skb_gso_ok(skb, features))
1472 __skb_push(skb, (skb->data -
1473 skb_network_header(skb)));
1475 segs = ptype->gso_segment(skb, features);
1481 __skb_push(skb, skb->data - skb_mac_header(skb));
1486 EXPORT_SYMBOL(skb_gso_segment);
1488 /* Take action when hardware reception checksum errors are detected. */
1490 void netdev_rx_csum_fault(struct net_device *dev)
1492 if (net_ratelimit()) {
1493 printk(KERN_ERR "%s: hw csum failure.\n",
1494 dev ? dev->name : "<unknown>");
1498 EXPORT_SYMBOL(netdev_rx_csum_fault);
1501 /* Actually, we should eliminate this check as soon as we know, that:
1502 * 1. IOMMU is present and allows to map all the memory.
1503 * 2. No high memory really exists on this machine.
1506 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1508 #ifdef CONFIG_HIGHMEM
1511 if (dev->features & NETIF_F_HIGHDMA)
1514 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1515 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1523 void (*destructor)(struct sk_buff *skb);
1526 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1528 static void dev_gso_skb_destructor(struct sk_buff *skb)
1530 struct dev_gso_cb *cb;
1533 struct sk_buff *nskb = skb->next;
1535 skb->next = nskb->next;
1538 } while (skb->next);
1540 cb = DEV_GSO_CB(skb);
1542 cb->destructor(skb);
1546 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1547 * @skb: buffer to segment
1549 * This function segments the given skb and stores the list of segments
1552 static int dev_gso_segment(struct sk_buff *skb)
1554 struct net_device *dev = skb->dev;
1555 struct sk_buff *segs;
1556 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1559 segs = skb_gso_segment(skb, features);
1561 /* Verifying header integrity only. */
1566 return PTR_ERR(segs);
1569 DEV_GSO_CB(skb)->destructor = skb->destructor;
1570 skb->destructor = dev_gso_skb_destructor;
1575 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1577 if (likely(!skb->next)) {
1578 if (!list_empty(&ptype_all))
1579 dev_queue_xmit_nit(skb, dev);
1581 if (netif_needs_gso(dev, skb)) {
1582 if (unlikely(dev_gso_segment(skb)))
1588 return dev->hard_start_xmit(skb, dev);
1593 struct sk_buff *nskb = skb->next;
1596 skb->next = nskb->next;
1598 rc = dev->hard_start_xmit(nskb, dev);
1600 nskb->next = skb->next;
1604 if (unlikely((netif_queue_stopped(dev) ||
1605 netif_subqueue_stopped(dev, skb)) &&
1607 return NETDEV_TX_BUSY;
1608 } while (skb->next);
1610 skb->destructor = DEV_GSO_CB(skb)->destructor;
1618 * dev_queue_xmit - transmit a buffer
1619 * @skb: buffer to transmit
1621 * Queue a buffer for transmission to a network device. The caller must
1622 * have set the device and priority and built the buffer before calling
1623 * this function. The function can be called from an interrupt.
1625 * A negative errno code is returned on a failure. A success does not
1626 * guarantee the frame will be transmitted as it may be dropped due
1627 * to congestion or traffic shaping.
1629 * -----------------------------------------------------------------------------------
1630 * I notice this method can also return errors from the queue disciplines,
1631 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1634 * Regardless of the return value, the skb is consumed, so it is currently
1635 * difficult to retry a send to this method. (You can bump the ref count
1636 * before sending to hold a reference for retry if you are careful.)
1638 * When calling this method, interrupts MUST be enabled. This is because
1639 * the BH enable code must have IRQs enabled so that it will not deadlock.
1643 int dev_queue_xmit(struct sk_buff *skb)
1645 struct net_device *dev = skb->dev;
1649 /* GSO will handle the following emulations directly. */
1650 if (netif_needs_gso(dev, skb))
1653 if (skb_shinfo(skb)->frag_list &&
1654 !(dev->features & NETIF_F_FRAGLIST) &&
1655 __skb_linearize(skb))
1658 /* Fragmented skb is linearized if device does not support SG,
1659 * or if at least one of fragments is in highmem and device
1660 * does not support DMA from it.
1662 if (skb_shinfo(skb)->nr_frags &&
1663 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1664 __skb_linearize(skb))
1667 /* If packet is not checksummed and device does not support
1668 * checksumming for this protocol, complete checksumming here.
1670 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1671 skb_set_transport_header(skb, skb->csum_start -
1673 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1678 spin_lock_prefetch(&dev->queue_lock);
1680 /* Disable soft irqs for various locks below. Also
1681 * stops preemption for RCU.
1685 /* Updates of qdisc are serialized by queue_lock.
1686 * The struct Qdisc which is pointed to by qdisc is now a
1687 * rcu structure - it may be accessed without acquiring
1688 * a lock (but the structure may be stale.) The freeing of the
1689 * qdisc will be deferred until it's known that there are no
1690 * more references to it.
1692 * If the qdisc has an enqueue function, we still need to
1693 * hold the queue_lock before calling it, since queue_lock
1694 * also serializes access to the device queue.
1697 q = rcu_dereference(dev->qdisc);
1698 #ifdef CONFIG_NET_CLS_ACT
1699 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1702 /* Grab device queue */
1703 spin_lock(&dev->queue_lock);
1706 /* reset queue_mapping to zero */
1707 skb_set_queue_mapping(skb, 0);
1708 rc = q->enqueue(skb, q);
1710 spin_unlock(&dev->queue_lock);
1712 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1715 spin_unlock(&dev->queue_lock);
1718 /* The device has no queue. Common case for software devices:
1719 loopback, all the sorts of tunnels...
1721 Really, it is unlikely that netif_tx_lock protection is necessary
1722 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1724 However, it is possible, that they rely on protection
1727 Check this and shot the lock. It is not prone from deadlocks.
1728 Either shot noqueue qdisc, it is even simpler 8)
1730 if (dev->flags & IFF_UP) {
1731 int cpu = smp_processor_id(); /* ok because BHs are off */
1733 if (dev->xmit_lock_owner != cpu) {
1735 HARD_TX_LOCK(dev, cpu);
1737 if (!netif_queue_stopped(dev) &&
1738 !netif_subqueue_stopped(dev, skb)) {
1740 if (!dev_hard_start_xmit(skb, dev)) {
1741 HARD_TX_UNLOCK(dev);
1745 HARD_TX_UNLOCK(dev);
1746 if (net_ratelimit())
1747 printk(KERN_CRIT "Virtual device %s asks to "
1748 "queue packet!\n", dev->name);
1750 /* Recursion is detected! It is possible,
1752 if (net_ratelimit())
1753 printk(KERN_CRIT "Dead loop on virtual device "
1754 "%s, fix it urgently!\n", dev->name);
1759 rcu_read_unlock_bh();
1765 rcu_read_unlock_bh();
1770 /*=======================================================================
1772 =======================================================================*/
1774 int netdev_max_backlog __read_mostly = 1000;
1775 int netdev_budget __read_mostly = 300;
1776 int weight_p __read_mostly = 64; /* old backlog weight */
1778 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1782 * netif_rx - post buffer to the network code
1783 * @skb: buffer to post
1785 * This function receives a packet from a device driver and queues it for
1786 * the upper (protocol) levels to process. It always succeeds. The buffer
1787 * may be dropped during processing for congestion control or by the
1791 * NET_RX_SUCCESS (no congestion)
1792 * NET_RX_DROP (packet was dropped)
1796 int netif_rx(struct sk_buff *skb)
1798 struct softnet_data *queue;
1799 unsigned long flags;
1801 /* if netpoll wants it, pretend we never saw it */
1802 if (netpoll_rx(skb))
1805 if (!skb->tstamp.tv64)
1809 * The code is rearranged so that the path is the most
1810 * short when CPU is congested, but is still operating.
1812 local_irq_save(flags);
1813 queue = &__get_cpu_var(softnet_data);
1815 __get_cpu_var(netdev_rx_stat).total++;
1816 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1817 if (queue->input_pkt_queue.qlen) {
1820 __skb_queue_tail(&queue->input_pkt_queue, skb);
1821 local_irq_restore(flags);
1822 return NET_RX_SUCCESS;
1825 napi_schedule(&queue->backlog);
1829 __get_cpu_var(netdev_rx_stat).dropped++;
1830 local_irq_restore(flags);
1836 int netif_rx_ni(struct sk_buff *skb)
1841 err = netif_rx(skb);
1842 if (local_softirq_pending())
1849 EXPORT_SYMBOL(netif_rx_ni);
1851 static inline struct net_device *skb_bond(struct sk_buff *skb)
1853 struct net_device *dev = skb->dev;
1856 if (skb_bond_should_drop(skb)) {
1860 skb->dev = dev->master;
1867 static void net_tx_action(struct softirq_action *h)
1869 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1871 if (sd->completion_queue) {
1872 struct sk_buff *clist;
1874 local_irq_disable();
1875 clist = sd->completion_queue;
1876 sd->completion_queue = NULL;
1880 struct sk_buff *skb = clist;
1881 clist = clist->next;
1883 BUG_TRAP(!atomic_read(&skb->users));
1888 if (sd->output_queue) {
1889 struct net_device *head;
1891 local_irq_disable();
1892 head = sd->output_queue;
1893 sd->output_queue = NULL;
1897 struct net_device *dev = head;
1898 head = head->next_sched;
1900 smp_mb__before_clear_bit();
1901 clear_bit(__LINK_STATE_SCHED, &dev->state);
1903 if (spin_trylock(&dev->queue_lock)) {
1905 spin_unlock(&dev->queue_lock);
1907 netif_schedule(dev);
1913 static inline int deliver_skb(struct sk_buff *skb,
1914 struct packet_type *pt_prev,
1915 struct net_device *orig_dev)
1917 atomic_inc(&skb->users);
1918 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1921 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1922 /* These hooks defined here for ATM */
1924 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1925 unsigned char *addr);
1926 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1929 * If bridge module is loaded call bridging hook.
1930 * returns NULL if packet was consumed.
1932 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
1933 struct sk_buff *skb) __read_mostly;
1934 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
1935 struct packet_type **pt_prev, int *ret,
1936 struct net_device *orig_dev)
1938 struct net_bridge_port *port;
1940 if (skb->pkt_type == PACKET_LOOPBACK ||
1941 (port = rcu_dereference(skb->dev->br_port)) == NULL)
1945 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1949 return br_handle_frame_hook(port, skb);
1952 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1955 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1956 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
1957 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
1959 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
1960 struct packet_type **pt_prev,
1962 struct net_device *orig_dev)
1964 if (skb->dev->macvlan_port == NULL)
1968 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1971 return macvlan_handle_frame_hook(skb);
1974 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1977 #ifdef CONFIG_NET_CLS_ACT
1978 /* TODO: Maybe we should just force sch_ingress to be compiled in
1979 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1980 * a compare and 2 stores extra right now if we dont have it on
1981 * but have CONFIG_NET_CLS_ACT
1982 * NOTE: This doesnt stop any functionality; if you dont have
1983 * the ingress scheduler, you just cant add policies on ingress.
1986 static int ing_filter(struct sk_buff *skb)
1989 struct net_device *dev = skb->dev;
1990 int result = TC_ACT_OK;
1991 u32 ttl = G_TC_RTTL(skb->tc_verd);
1993 if (MAX_RED_LOOP < ttl++) {
1995 "Redir loop detected Dropping packet (%d->%d)\n",
1996 skb->iif, dev->ifindex);
2000 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2001 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2003 spin_lock(&dev->ingress_lock);
2004 if ((q = dev->qdisc_ingress) != NULL)
2005 result = q->enqueue(skb, q);
2006 spin_unlock(&dev->ingress_lock);
2011 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2012 struct packet_type **pt_prev,
2013 int *ret, struct net_device *orig_dev)
2015 if (!skb->dev->qdisc_ingress)
2019 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2022 /* Huh? Why does turning on AF_PACKET affect this? */
2023 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2026 switch (ing_filter(skb)) {
2040 * netif_receive_skb - process receive buffer from network
2041 * @skb: buffer to process
2043 * netif_receive_skb() is the main receive data processing function.
2044 * It always succeeds. The buffer may be dropped during processing
2045 * for congestion control or by the protocol layers.
2047 * This function may only be called from softirq context and interrupts
2048 * should be enabled.
2050 * Return values (usually ignored):
2051 * NET_RX_SUCCESS: no congestion
2052 * NET_RX_DROP: packet was dropped
2054 int netif_receive_skb(struct sk_buff *skb)
2056 struct packet_type *ptype, *pt_prev;
2057 struct net_device *orig_dev;
2058 int ret = NET_RX_DROP;
2061 /* if we've gotten here through NAPI, check netpoll */
2062 if (netpoll_receive_skb(skb))
2065 if (!skb->tstamp.tv64)
2069 skb->iif = skb->dev->ifindex;
2071 orig_dev = skb_bond(skb);
2076 __get_cpu_var(netdev_rx_stat).total++;
2078 skb_reset_network_header(skb);
2079 skb_reset_transport_header(skb);
2080 skb->mac_len = skb->network_header - skb->mac_header;
2086 #ifdef CONFIG_NET_CLS_ACT
2087 if (skb->tc_verd & TC_NCLS) {
2088 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2093 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2094 if (!ptype->dev || ptype->dev == skb->dev) {
2096 ret = deliver_skb(skb, pt_prev, orig_dev);
2101 #ifdef CONFIG_NET_CLS_ACT
2102 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2108 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2111 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2115 type = skb->protocol;
2116 list_for_each_entry_rcu(ptype,
2117 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2118 if (ptype->type == type &&
2119 (!ptype->dev || ptype->dev == skb->dev)) {
2121 ret = deliver_skb(skb, pt_prev, orig_dev);
2127 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2130 /* Jamal, now you will not able to escape explaining
2131 * me how you were going to use this. :-)
2141 static int process_backlog(struct napi_struct *napi, int quota)
2144 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2145 unsigned long start_time = jiffies;
2147 napi->weight = weight_p;
2149 struct sk_buff *skb;
2150 struct net_device *dev;
2152 local_irq_disable();
2153 skb = __skb_dequeue(&queue->input_pkt_queue);
2155 __napi_complete(napi);
2164 netif_receive_skb(skb);
2167 } while (++work < quota && jiffies == start_time);
2173 * __napi_schedule - schedule for receive
2174 * @n: entry to schedule
2176 * The entry's receive function will be scheduled to run
2178 void __napi_schedule(struct napi_struct *n)
2180 unsigned long flags;
2182 local_irq_save(flags);
2183 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2184 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2185 local_irq_restore(flags);
2187 EXPORT_SYMBOL(__napi_schedule);
2190 static void net_rx_action(struct softirq_action *h)
2192 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2193 unsigned long start_time = jiffies;
2194 int budget = netdev_budget;
2197 local_irq_disable();
2199 while (!list_empty(list)) {
2200 struct napi_struct *n;
2203 /* If softirq window is exhuasted then punt.
2205 * Note that this is a slight policy change from the
2206 * previous NAPI code, which would allow up to 2
2207 * jiffies to pass before breaking out. The test
2208 * used to be "jiffies - start_time > 1".
2210 if (unlikely(budget <= 0 || jiffies != start_time))
2215 /* Even though interrupts have been re-enabled, this
2216 * access is safe because interrupts can only add new
2217 * entries to the tail of this list, and only ->poll()
2218 * calls can remove this head entry from the list.
2220 n = list_entry(list->next, struct napi_struct, poll_list);
2222 have = netpoll_poll_lock(n);
2226 /* This NAPI_STATE_SCHED test is for avoiding a race
2227 * with netpoll's poll_napi(). Only the entity which
2228 * obtains the lock and sees NAPI_STATE_SCHED set will
2229 * actually make the ->poll() call. Therefore we avoid
2230 * accidently calling ->poll() when NAPI is not scheduled.
2233 if (test_bit(NAPI_STATE_SCHED, &n->state))
2234 work = n->poll(n, weight);
2236 WARN_ON_ONCE(work > weight);
2240 local_irq_disable();
2242 /* Drivers must not modify the NAPI state if they
2243 * consume the entire weight. In such cases this code
2244 * still "owns" the NAPI instance and therefore can
2245 * move the instance around on the list at-will.
2247 if (unlikely(work == weight)) {
2248 if (unlikely(napi_disable_pending(n)))
2251 list_move_tail(&n->poll_list, list);
2254 netpoll_poll_unlock(have);
2259 #ifdef CONFIG_NET_DMA
2261 * There may not be any more sk_buffs coming right now, so push
2262 * any pending DMA copies to hardware
2264 if (!cpus_empty(net_dma.channel_mask)) {
2266 for_each_cpu_mask(chan_idx, net_dma.channel_mask) {
2267 struct dma_chan *chan = net_dma.channels[chan_idx];
2269 dma_async_memcpy_issue_pending(chan);
2277 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2278 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2282 static gifconf_func_t * gifconf_list [NPROTO];
2285 * register_gifconf - register a SIOCGIF handler
2286 * @family: Address family
2287 * @gifconf: Function handler
2289 * Register protocol dependent address dumping routines. The handler
2290 * that is passed must not be freed or reused until it has been replaced
2291 * by another handler.
2293 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2295 if (family >= NPROTO)
2297 gifconf_list[family] = gifconf;
2303 * Map an interface index to its name (SIOCGIFNAME)
2307 * We need this ioctl for efficient implementation of the
2308 * if_indextoname() function required by the IPv6 API. Without
2309 * it, we would have to search all the interfaces to find a
2313 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2315 struct net_device *dev;
2319 * Fetch the caller's info block.
2322 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2325 read_lock(&dev_base_lock);
2326 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2328 read_unlock(&dev_base_lock);
2332 strcpy(ifr.ifr_name, dev->name);
2333 read_unlock(&dev_base_lock);
2335 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2341 * Perform a SIOCGIFCONF call. This structure will change
2342 * size eventually, and there is nothing I can do about it.
2343 * Thus we will need a 'compatibility mode'.
2346 static int dev_ifconf(struct net *net, char __user *arg)
2349 struct net_device *dev;
2356 * Fetch the caller's info block.
2359 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2366 * Loop over the interfaces, and write an info block for each.
2370 for_each_netdev(net, dev) {
2371 for (i = 0; i < NPROTO; i++) {
2372 if (gifconf_list[i]) {
2375 done = gifconf_list[i](dev, NULL, 0);
2377 done = gifconf_list[i](dev, pos + total,
2387 * All done. Write the updated control block back to the caller.
2389 ifc.ifc_len = total;
2392 * Both BSD and Solaris return 0 here, so we do too.
2394 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2397 #ifdef CONFIG_PROC_FS
2399 * This is invoked by the /proc filesystem handler to display a device
2402 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2403 __acquires(dev_base_lock)
2405 struct net *net = seq_file_net(seq);
2407 struct net_device *dev;
2409 read_lock(&dev_base_lock);
2411 return SEQ_START_TOKEN;
2414 for_each_netdev(net, dev)
2421 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2423 struct net *net = seq_file_net(seq);
2425 return v == SEQ_START_TOKEN ?
2426 first_net_device(net) : next_net_device((struct net_device *)v);
2429 void dev_seq_stop(struct seq_file *seq, void *v)
2430 __releases(dev_base_lock)
2432 read_unlock(&dev_base_lock);
2435 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2437 struct net_device_stats *stats = dev->get_stats(dev);
2439 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2440 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2441 dev->name, stats->rx_bytes, stats->rx_packets,
2443 stats->rx_dropped + stats->rx_missed_errors,
2444 stats->rx_fifo_errors,
2445 stats->rx_length_errors + stats->rx_over_errors +
2446 stats->rx_crc_errors + stats->rx_frame_errors,
2447 stats->rx_compressed, stats->multicast,
2448 stats->tx_bytes, stats->tx_packets,
2449 stats->tx_errors, stats->tx_dropped,
2450 stats->tx_fifo_errors, stats->collisions,
2451 stats->tx_carrier_errors +
2452 stats->tx_aborted_errors +
2453 stats->tx_window_errors +
2454 stats->tx_heartbeat_errors,
2455 stats->tx_compressed);
2459 * Called from the PROCfs module. This now uses the new arbitrary sized
2460 * /proc/net interface to create /proc/net/dev
2462 static int dev_seq_show(struct seq_file *seq, void *v)
2464 if (v == SEQ_START_TOKEN)
2465 seq_puts(seq, "Inter-| Receive "
2467 " face |bytes packets errs drop fifo frame "
2468 "compressed multicast|bytes packets errs "
2469 "drop fifo colls carrier compressed\n");
2471 dev_seq_printf_stats(seq, v);
2475 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2477 struct netif_rx_stats *rc = NULL;
2479 while (*pos < nr_cpu_ids)
2480 if (cpu_online(*pos)) {
2481 rc = &per_cpu(netdev_rx_stat, *pos);
2488 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2490 return softnet_get_online(pos);
2493 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2496 return softnet_get_online(pos);
2499 static void softnet_seq_stop(struct seq_file *seq, void *v)
2503 static int softnet_seq_show(struct seq_file *seq, void *v)
2505 struct netif_rx_stats *s = v;
2507 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2508 s->total, s->dropped, s->time_squeeze, 0,
2509 0, 0, 0, 0, /* was fastroute */
2514 static const struct seq_operations dev_seq_ops = {
2515 .start = dev_seq_start,
2516 .next = dev_seq_next,
2517 .stop = dev_seq_stop,
2518 .show = dev_seq_show,
2521 static int dev_seq_open(struct inode *inode, struct file *file)
2523 return seq_open_net(inode, file, &dev_seq_ops,
2524 sizeof(struct seq_net_private));
2527 static const struct file_operations dev_seq_fops = {
2528 .owner = THIS_MODULE,
2529 .open = dev_seq_open,
2531 .llseek = seq_lseek,
2532 .release = seq_release_net,
2535 static const struct seq_operations softnet_seq_ops = {
2536 .start = softnet_seq_start,
2537 .next = softnet_seq_next,
2538 .stop = softnet_seq_stop,
2539 .show = softnet_seq_show,
2542 static int softnet_seq_open(struct inode *inode, struct file *file)
2544 return seq_open(file, &softnet_seq_ops);
2547 static const struct file_operations softnet_seq_fops = {
2548 .owner = THIS_MODULE,
2549 .open = softnet_seq_open,
2551 .llseek = seq_lseek,
2552 .release = seq_release,
2555 static void *ptype_get_idx(loff_t pos)
2557 struct packet_type *pt = NULL;
2561 list_for_each_entry_rcu(pt, &ptype_all, list) {
2567 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2568 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2577 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2581 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2584 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2586 struct packet_type *pt;
2587 struct list_head *nxt;
2591 if (v == SEQ_START_TOKEN)
2592 return ptype_get_idx(0);
2595 nxt = pt->list.next;
2596 if (pt->type == htons(ETH_P_ALL)) {
2597 if (nxt != &ptype_all)
2600 nxt = ptype_base[0].next;
2602 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2604 while (nxt == &ptype_base[hash]) {
2605 if (++hash >= PTYPE_HASH_SIZE)
2607 nxt = ptype_base[hash].next;
2610 return list_entry(nxt, struct packet_type, list);
2613 static void ptype_seq_stop(struct seq_file *seq, void *v)
2619 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2621 #ifdef CONFIG_KALLSYMS
2622 unsigned long offset = 0, symsize;
2623 const char *symname;
2627 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2634 modname = delim = "";
2635 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2641 seq_printf(seq, "[%p]", sym);
2644 static int ptype_seq_show(struct seq_file *seq, void *v)
2646 struct packet_type *pt = v;
2648 if (v == SEQ_START_TOKEN)
2649 seq_puts(seq, "Type Device Function\n");
2650 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
2651 if (pt->type == htons(ETH_P_ALL))
2652 seq_puts(seq, "ALL ");
2654 seq_printf(seq, "%04x", ntohs(pt->type));
2656 seq_printf(seq, " %-8s ",
2657 pt->dev ? pt->dev->name : "");
2658 ptype_seq_decode(seq, pt->func);
2659 seq_putc(seq, '\n');
2665 static const struct seq_operations ptype_seq_ops = {
2666 .start = ptype_seq_start,
2667 .next = ptype_seq_next,
2668 .stop = ptype_seq_stop,
2669 .show = ptype_seq_show,
2672 static int ptype_seq_open(struct inode *inode, struct file *file)
2674 return seq_open_net(inode, file, &ptype_seq_ops,
2675 sizeof(struct seq_net_private));
2678 static const struct file_operations ptype_seq_fops = {
2679 .owner = THIS_MODULE,
2680 .open = ptype_seq_open,
2682 .llseek = seq_lseek,
2683 .release = seq_release_net,
2687 static int __net_init dev_proc_net_init(struct net *net)
2691 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2693 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2695 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2698 if (wext_proc_init(net))
2704 proc_net_remove(net, "ptype");
2706 proc_net_remove(net, "softnet_stat");
2708 proc_net_remove(net, "dev");
2712 static void __net_exit dev_proc_net_exit(struct net *net)
2714 wext_proc_exit(net);
2716 proc_net_remove(net, "ptype");
2717 proc_net_remove(net, "softnet_stat");
2718 proc_net_remove(net, "dev");
2721 static struct pernet_operations __net_initdata dev_proc_ops = {
2722 .init = dev_proc_net_init,
2723 .exit = dev_proc_net_exit,
2726 static int __init dev_proc_init(void)
2728 return register_pernet_subsys(&dev_proc_ops);
2731 #define dev_proc_init() 0
2732 #endif /* CONFIG_PROC_FS */
2736 * netdev_set_master - set up master/slave pair
2737 * @slave: slave device
2738 * @master: new master device
2740 * Changes the master device of the slave. Pass %NULL to break the
2741 * bonding. The caller must hold the RTNL semaphore. On a failure
2742 * a negative errno code is returned. On success the reference counts
2743 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2744 * function returns zero.
2746 int netdev_set_master(struct net_device *slave, struct net_device *master)
2748 struct net_device *old = slave->master;
2758 slave->master = master;
2766 slave->flags |= IFF_SLAVE;
2768 slave->flags &= ~IFF_SLAVE;
2770 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2774 static void __dev_set_promiscuity(struct net_device *dev, int inc)
2776 unsigned short old_flags = dev->flags;
2780 if ((dev->promiscuity += inc) == 0)
2781 dev->flags &= ~IFF_PROMISC;
2783 dev->flags |= IFF_PROMISC;
2784 if (dev->flags != old_flags) {
2785 printk(KERN_INFO "device %s %s promiscuous mode\n",
2786 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2789 audit_log(current->audit_context, GFP_ATOMIC,
2790 AUDIT_ANOM_PROMISCUOUS,
2791 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
2792 dev->name, (dev->flags & IFF_PROMISC),
2793 (old_flags & IFF_PROMISC),
2794 audit_get_loginuid(current),
2795 current->uid, current->gid,
2796 audit_get_sessionid(current));
2798 if (dev->change_rx_flags)
2799 dev->change_rx_flags(dev, IFF_PROMISC);
2804 * dev_set_promiscuity - update promiscuity count on a device
2808 * Add or remove promiscuity from a device. While the count in the device
2809 * remains above zero the interface remains promiscuous. Once it hits zero
2810 * the device reverts back to normal filtering operation. A negative inc
2811 * value is used to drop promiscuity on the device.
2813 void dev_set_promiscuity(struct net_device *dev, int inc)
2815 unsigned short old_flags = dev->flags;
2817 __dev_set_promiscuity(dev, inc);
2818 if (dev->flags != old_flags)
2819 dev_set_rx_mode(dev);
2823 * dev_set_allmulti - update allmulti count on a device
2827 * Add or remove reception of all multicast frames to a device. While the
2828 * count in the device remains above zero the interface remains listening
2829 * to all interfaces. Once it hits zero the device reverts back to normal
2830 * filtering operation. A negative @inc value is used to drop the counter
2831 * when releasing a resource needing all multicasts.
2834 void dev_set_allmulti(struct net_device *dev, int inc)
2836 unsigned short old_flags = dev->flags;
2840 dev->flags |= IFF_ALLMULTI;
2841 if ((dev->allmulti += inc) == 0)
2842 dev->flags &= ~IFF_ALLMULTI;
2843 if (dev->flags ^ old_flags) {
2844 if (dev->change_rx_flags)
2845 dev->change_rx_flags(dev, IFF_ALLMULTI);
2846 dev_set_rx_mode(dev);
2851 * Upload unicast and multicast address lists to device and
2852 * configure RX filtering. When the device doesn't support unicast
2853 * filtering it is put in promiscuous mode while unicast addresses
2856 void __dev_set_rx_mode(struct net_device *dev)
2858 /* dev_open will call this function so the list will stay sane. */
2859 if (!(dev->flags&IFF_UP))
2862 if (!netif_device_present(dev))
2865 if (dev->set_rx_mode)
2866 dev->set_rx_mode(dev);
2868 /* Unicast addresses changes may only happen under the rtnl,
2869 * therefore calling __dev_set_promiscuity here is safe.
2871 if (dev->uc_count > 0 && !dev->uc_promisc) {
2872 __dev_set_promiscuity(dev, 1);
2873 dev->uc_promisc = 1;
2874 } else if (dev->uc_count == 0 && dev->uc_promisc) {
2875 __dev_set_promiscuity(dev, -1);
2876 dev->uc_promisc = 0;
2879 if (dev->set_multicast_list)
2880 dev->set_multicast_list(dev);
2884 void dev_set_rx_mode(struct net_device *dev)
2886 netif_tx_lock_bh(dev);
2887 __dev_set_rx_mode(dev);
2888 netif_tx_unlock_bh(dev);
2891 int __dev_addr_delete(struct dev_addr_list **list, int *count,
2892 void *addr, int alen, int glbl)
2894 struct dev_addr_list *da;
2896 for (; (da = *list) != NULL; list = &da->next) {
2897 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2898 alen == da->da_addrlen) {
2900 int old_glbl = da->da_gusers;
2917 int __dev_addr_add(struct dev_addr_list **list, int *count,
2918 void *addr, int alen, int glbl)
2920 struct dev_addr_list *da;
2922 for (da = *list; da != NULL; da = da->next) {
2923 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2924 da->da_addrlen == alen) {
2926 int old_glbl = da->da_gusers;
2936 da = kzalloc(sizeof(*da), GFP_ATOMIC);
2939 memcpy(da->da_addr, addr, alen);
2940 da->da_addrlen = alen;
2942 da->da_gusers = glbl ? 1 : 0;
2950 * dev_unicast_delete - Release secondary unicast address.
2952 * @addr: address to delete
2953 * @alen: length of @addr
2955 * Release reference to a secondary unicast address and remove it
2956 * from the device if the reference count drops to zero.
2958 * The caller must hold the rtnl_mutex.
2960 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
2966 netif_tx_lock_bh(dev);
2967 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2969 __dev_set_rx_mode(dev);
2970 netif_tx_unlock_bh(dev);
2973 EXPORT_SYMBOL(dev_unicast_delete);
2976 * dev_unicast_add - add a secondary unicast address
2978 * @addr: address to delete
2979 * @alen: length of @addr
2981 * Add a secondary unicast address to the device or increase
2982 * the reference count if it already exists.
2984 * The caller must hold the rtnl_mutex.
2986 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
2992 netif_tx_lock_bh(dev);
2993 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2995 __dev_set_rx_mode(dev);
2996 netif_tx_unlock_bh(dev);
2999 EXPORT_SYMBOL(dev_unicast_add);
3001 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3002 struct dev_addr_list **from, int *from_count)
3004 struct dev_addr_list *da, *next;
3008 while (da != NULL) {
3010 if (!da->da_synced) {
3011 err = __dev_addr_add(to, to_count,
3012 da->da_addr, da->da_addrlen, 0);
3017 } else if (da->da_users == 1) {
3018 __dev_addr_delete(to, to_count,
3019 da->da_addr, da->da_addrlen, 0);
3020 __dev_addr_delete(from, from_count,
3021 da->da_addr, da->da_addrlen, 0);
3028 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3029 struct dev_addr_list **from, int *from_count)
3031 struct dev_addr_list *da, *next;
3034 while (da != NULL) {
3036 if (da->da_synced) {
3037 __dev_addr_delete(to, to_count,
3038 da->da_addr, da->da_addrlen, 0);
3040 __dev_addr_delete(from, from_count,
3041 da->da_addr, da->da_addrlen, 0);
3048 * dev_unicast_sync - Synchronize device's unicast list to another device
3049 * @to: destination device
3050 * @from: source device
3052 * Add newly added addresses to the destination device and release
3053 * addresses that have no users left. The source device must be
3054 * locked by netif_tx_lock_bh.
3056 * This function is intended to be called from the dev->set_rx_mode
3057 * function of layered software devices.
3059 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3063 netif_tx_lock_bh(to);
3064 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3065 &from->uc_list, &from->uc_count);
3067 __dev_set_rx_mode(to);
3068 netif_tx_unlock_bh(to);
3071 EXPORT_SYMBOL(dev_unicast_sync);
3074 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3075 * @to: destination device
3076 * @from: source device
3078 * Remove all addresses that were added to the destination device by
3079 * dev_unicast_sync(). This function is intended to be called from the
3080 * dev->stop function of layered software devices.
3082 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3084 netif_tx_lock_bh(from);
3085 netif_tx_lock_bh(to);
3087 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3088 &from->uc_list, &from->uc_count);
3089 __dev_set_rx_mode(to);
3091 netif_tx_unlock_bh(to);
3092 netif_tx_unlock_bh(from);
3094 EXPORT_SYMBOL(dev_unicast_unsync);
3096 static void __dev_addr_discard(struct dev_addr_list **list)
3098 struct dev_addr_list *tmp;
3100 while (*list != NULL) {
3103 if (tmp->da_users > tmp->da_gusers)
3104 printk("__dev_addr_discard: address leakage! "
3105 "da_users=%d\n", tmp->da_users);
3110 static void dev_addr_discard(struct net_device *dev)
3112 netif_tx_lock_bh(dev);
3114 __dev_addr_discard(&dev->uc_list);
3117 __dev_addr_discard(&dev->mc_list);
3120 netif_tx_unlock_bh(dev);
3123 unsigned dev_get_flags(const struct net_device *dev)
3127 flags = (dev->flags & ~(IFF_PROMISC |
3132 (dev->gflags & (IFF_PROMISC |
3135 if (netif_running(dev)) {
3136 if (netif_oper_up(dev))
3137 flags |= IFF_RUNNING;
3138 if (netif_carrier_ok(dev))
3139 flags |= IFF_LOWER_UP;
3140 if (netif_dormant(dev))
3141 flags |= IFF_DORMANT;
3147 int dev_change_flags(struct net_device *dev, unsigned flags)
3150 int old_flags = dev->flags;
3155 * Set the flags on our device.
3158 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3159 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3161 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3165 * Load in the correct multicast list now the flags have changed.
3168 if (dev->change_rx_flags && (old_flags ^ flags) & IFF_MULTICAST)
3169 dev->change_rx_flags(dev, IFF_MULTICAST);
3171 dev_set_rx_mode(dev);
3174 * Have we downed the interface. We handle IFF_UP ourselves
3175 * according to user attempts to set it, rather than blindly
3180 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3181 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3184 dev_set_rx_mode(dev);
3187 if (dev->flags & IFF_UP &&
3188 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3190 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3192 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3193 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3194 dev->gflags ^= IFF_PROMISC;
3195 dev_set_promiscuity(dev, inc);
3198 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3199 is important. Some (broken) drivers set IFF_PROMISC, when
3200 IFF_ALLMULTI is requested not asking us and not reporting.
3202 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3203 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3204 dev->gflags ^= IFF_ALLMULTI;
3205 dev_set_allmulti(dev, inc);
3208 /* Exclude state transition flags, already notified */
3209 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3211 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3216 int dev_set_mtu(struct net_device *dev, int new_mtu)
3220 if (new_mtu == dev->mtu)
3223 /* MTU must be positive. */
3227 if (!netif_device_present(dev))
3231 if (dev->change_mtu)
3232 err = dev->change_mtu(dev, new_mtu);
3235 if (!err && dev->flags & IFF_UP)
3236 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3240 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3244 if (!dev->set_mac_address)
3246 if (sa->sa_family != dev->type)
3248 if (!netif_device_present(dev))
3250 err = dev->set_mac_address(dev, sa);
3252 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3257 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3259 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3262 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3268 case SIOCGIFFLAGS: /* Get interface flags */
3269 ifr->ifr_flags = dev_get_flags(dev);
3272 case SIOCGIFMETRIC: /* Get the metric on the interface
3273 (currently unused) */
3274 ifr->ifr_metric = 0;
3277 case SIOCGIFMTU: /* Get the MTU of a device */
3278 ifr->ifr_mtu = dev->mtu;
3283 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3285 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3286 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3287 ifr->ifr_hwaddr.sa_family = dev->type;
3295 ifr->ifr_map.mem_start = dev->mem_start;
3296 ifr->ifr_map.mem_end = dev->mem_end;
3297 ifr->ifr_map.base_addr = dev->base_addr;
3298 ifr->ifr_map.irq = dev->irq;
3299 ifr->ifr_map.dma = dev->dma;
3300 ifr->ifr_map.port = dev->if_port;
3304 ifr->ifr_ifindex = dev->ifindex;
3308 ifr->ifr_qlen = dev->tx_queue_len;
3312 /* dev_ioctl() should ensure this case
3324 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3326 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3329 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3335 case SIOCSIFFLAGS: /* Set interface flags */
3336 return dev_change_flags(dev, ifr->ifr_flags);
3338 case SIOCSIFMETRIC: /* Set the metric on the interface
3339 (currently unused) */
3342 case SIOCSIFMTU: /* Set the MTU of a device */
3343 return dev_set_mtu(dev, ifr->ifr_mtu);
3346 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3348 case SIOCSIFHWBROADCAST:
3349 if (ifr->ifr_hwaddr.sa_family != dev->type)
3351 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3352 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3353 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3357 if (dev->set_config) {
3358 if (!netif_device_present(dev))
3360 return dev->set_config(dev, &ifr->ifr_map);
3365 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3366 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3368 if (!netif_device_present(dev))
3370 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3374 if ((!dev->set_multicast_list && !dev->set_rx_mode) ||
3375 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3377 if (!netif_device_present(dev))
3379 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3383 if (ifr->ifr_qlen < 0)
3385 dev->tx_queue_len = ifr->ifr_qlen;
3389 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3390 return dev_change_name(dev, ifr->ifr_newname);
3393 * Unknown or private ioctl
3397 if ((cmd >= SIOCDEVPRIVATE &&
3398 cmd <= SIOCDEVPRIVATE + 15) ||
3399 cmd == SIOCBONDENSLAVE ||
3400 cmd == SIOCBONDRELEASE ||
3401 cmd == SIOCBONDSETHWADDR ||
3402 cmd == SIOCBONDSLAVEINFOQUERY ||
3403 cmd == SIOCBONDINFOQUERY ||
3404 cmd == SIOCBONDCHANGEACTIVE ||
3405 cmd == SIOCGMIIPHY ||
3406 cmd == SIOCGMIIREG ||
3407 cmd == SIOCSMIIREG ||
3408 cmd == SIOCBRADDIF ||
3409 cmd == SIOCBRDELIF ||
3410 cmd == SIOCWANDEV) {
3412 if (dev->do_ioctl) {
3413 if (netif_device_present(dev))
3414 err = dev->do_ioctl(dev, ifr,
3427 * This function handles all "interface"-type I/O control requests. The actual
3428 * 'doing' part of this is dev_ifsioc above.
3432 * dev_ioctl - network device ioctl
3433 * @net: the applicable net namespace
3434 * @cmd: command to issue
3435 * @arg: pointer to a struct ifreq in user space
3437 * Issue ioctl functions to devices. This is normally called by the
3438 * user space syscall interfaces but can sometimes be useful for
3439 * other purposes. The return value is the return from the syscall if
3440 * positive or a negative errno code on error.
3443 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3449 /* One special case: SIOCGIFCONF takes ifconf argument
3450 and requires shared lock, because it sleeps writing
3454 if (cmd == SIOCGIFCONF) {
3456 ret = dev_ifconf(net, (char __user *) arg);
3460 if (cmd == SIOCGIFNAME)
3461 return dev_ifname(net, (struct ifreq __user *)arg);
3463 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3466 ifr.ifr_name[IFNAMSIZ-1] = 0;
3468 colon = strchr(ifr.ifr_name, ':');
3473 * See which interface the caller is talking about.
3478 * These ioctl calls:
3479 * - can be done by all.
3480 * - atomic and do not require locking.
3491 dev_load(net, ifr.ifr_name);
3492 read_lock(&dev_base_lock);
3493 ret = dev_ifsioc_locked(net, &ifr, cmd);
3494 read_unlock(&dev_base_lock);
3498 if (copy_to_user(arg, &ifr,
3499 sizeof(struct ifreq)))
3505 dev_load(net, ifr.ifr_name);
3507 ret = dev_ethtool(net, &ifr);
3512 if (copy_to_user(arg, &ifr,
3513 sizeof(struct ifreq)))
3519 * These ioctl calls:
3520 * - require superuser power.
3521 * - require strict serialization.
3527 if (!capable(CAP_NET_ADMIN))
3529 dev_load(net, ifr.ifr_name);
3531 ret = dev_ifsioc(net, &ifr, cmd);
3536 if (copy_to_user(arg, &ifr,
3537 sizeof(struct ifreq)))
3543 * These ioctl calls:
3544 * - require superuser power.
3545 * - require strict serialization.
3546 * - do not return a value
3556 case SIOCSIFHWBROADCAST:
3559 case SIOCBONDENSLAVE:
3560 case SIOCBONDRELEASE:
3561 case SIOCBONDSETHWADDR:
3562 case SIOCBONDCHANGEACTIVE:
3565 if (!capable(CAP_NET_ADMIN))
3568 case SIOCBONDSLAVEINFOQUERY:
3569 case SIOCBONDINFOQUERY:
3570 dev_load(net, ifr.ifr_name);
3572 ret = dev_ifsioc(net, &ifr, cmd);
3577 /* Get the per device memory space. We can add this but
3578 * currently do not support it */
3580 /* Set the per device memory buffer space.
3581 * Not applicable in our case */
3586 * Unknown or private ioctl.
3589 if (cmd == SIOCWANDEV ||
3590 (cmd >= SIOCDEVPRIVATE &&
3591 cmd <= SIOCDEVPRIVATE + 15)) {
3592 dev_load(net, ifr.ifr_name);
3594 ret = dev_ifsioc(net, &ifr, cmd);
3596 if (!ret && copy_to_user(arg, &ifr,
3597 sizeof(struct ifreq)))
3601 /* Take care of Wireless Extensions */
3602 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3603 return wext_handle_ioctl(net, &ifr, cmd, arg);
3610 * dev_new_index - allocate an ifindex
3611 * @net: the applicable net namespace
3613 * Returns a suitable unique value for a new device interface
3614 * number. The caller must hold the rtnl semaphore or the
3615 * dev_base_lock to be sure it remains unique.
3617 static int dev_new_index(struct net *net)
3623 if (!__dev_get_by_index(net, ifindex))
3628 /* Delayed registration/unregisteration */
3629 static DEFINE_SPINLOCK(net_todo_list_lock);
3630 static LIST_HEAD(net_todo_list);
3632 static void net_set_todo(struct net_device *dev)
3634 spin_lock(&net_todo_list_lock);
3635 list_add_tail(&dev->todo_list, &net_todo_list);
3636 spin_unlock(&net_todo_list_lock);
3639 static void rollback_registered(struct net_device *dev)
3641 BUG_ON(dev_boot_phase);
3644 /* Some devices call without registering for initialization unwind. */
3645 if (dev->reg_state == NETREG_UNINITIALIZED) {
3646 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3647 "was registered\n", dev->name, dev);
3653 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3655 /* If device is running, close it first. */
3658 /* And unlink it from device chain. */
3659 unlist_netdevice(dev);
3661 dev->reg_state = NETREG_UNREGISTERING;
3665 /* Shutdown queueing discipline. */
3669 /* Notify protocols, that we are about to destroy
3670 this device. They should clean all the things.
3672 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3675 * Flush the unicast and multicast chains
3677 dev_addr_discard(dev);
3682 /* Notifier chain MUST detach us from master device. */
3683 BUG_TRAP(!dev->master);
3685 /* Remove entries from kobject tree */
3686 netdev_unregister_kobject(dev);
3694 * register_netdevice - register a network device
3695 * @dev: device to register
3697 * Take a completed network device structure and add it to the kernel
3698 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3699 * chain. 0 is returned on success. A negative errno code is returned
3700 * on a failure to set up the device, or if the name is a duplicate.
3702 * Callers must hold the rtnl semaphore. You may want
3703 * register_netdev() instead of this.
3706 * The locking appears insufficient to guarantee two parallel registers
3707 * will not get the same name.
3710 int register_netdevice(struct net_device *dev)
3712 struct hlist_head *head;
3713 struct hlist_node *p;
3717 BUG_ON(dev_boot_phase);
3722 /* When net_device's are persistent, this will be fatal. */
3723 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3724 BUG_ON(!dev_net(dev));
3727 spin_lock_init(&dev->queue_lock);
3728 spin_lock_init(&dev->_xmit_lock);
3729 netdev_set_lockdep_class(&dev->_xmit_lock, dev->type);
3730 dev->xmit_lock_owner = -1;
3731 spin_lock_init(&dev->ingress_lock);
3735 /* Init, if this function is available */
3737 ret = dev->init(dev);
3745 if (!dev_valid_name(dev->name)) {
3750 dev->ifindex = dev_new_index(net);
3751 if (dev->iflink == -1)
3752 dev->iflink = dev->ifindex;
3754 /* Check for existence of name */
3755 head = dev_name_hash(net, dev->name);
3756 hlist_for_each(p, head) {
3757 struct net_device *d
3758 = hlist_entry(p, struct net_device, name_hlist);
3759 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3765 /* Fix illegal checksum combinations */
3766 if ((dev->features & NETIF_F_HW_CSUM) &&
3767 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3768 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3770 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3773 if ((dev->features & NETIF_F_NO_CSUM) &&
3774 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3775 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3777 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3781 /* Fix illegal SG+CSUM combinations. */
3782 if ((dev->features & NETIF_F_SG) &&
3783 !(dev->features & NETIF_F_ALL_CSUM)) {
3784 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3786 dev->features &= ~NETIF_F_SG;
3789 /* TSO requires that SG is present as well. */
3790 if ((dev->features & NETIF_F_TSO) &&
3791 !(dev->features & NETIF_F_SG)) {
3792 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3794 dev->features &= ~NETIF_F_TSO;
3796 if (dev->features & NETIF_F_UFO) {
3797 if (!(dev->features & NETIF_F_HW_CSUM)) {
3798 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3799 "NETIF_F_HW_CSUM feature.\n",
3801 dev->features &= ~NETIF_F_UFO;
3803 if (!(dev->features & NETIF_F_SG)) {
3804 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3805 "NETIF_F_SG feature.\n",
3807 dev->features &= ~NETIF_F_UFO;
3811 netdev_initialize_kobject(dev);
3812 ret = netdev_register_kobject(dev);
3815 dev->reg_state = NETREG_REGISTERED;
3818 * Default initial state at registry is that the
3819 * device is present.
3822 set_bit(__LINK_STATE_PRESENT, &dev->state);
3824 dev_init_scheduler(dev);
3826 list_netdevice(dev);
3828 /* Notify protocols, that a new device appeared. */
3829 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
3830 ret = notifier_to_errno(ret);
3832 rollback_registered(dev);
3833 dev->reg_state = NETREG_UNREGISTERED;
3846 * register_netdev - register a network device
3847 * @dev: device to register
3849 * Take a completed network device structure and add it to the kernel
3850 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3851 * chain. 0 is returned on success. A negative errno code is returned
3852 * on a failure to set up the device, or if the name is a duplicate.
3854 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3855 * and expands the device name if you passed a format string to
3858 int register_netdev(struct net_device *dev)
3865 * If the name is a format string the caller wants us to do a
3868 if (strchr(dev->name, '%')) {
3869 err = dev_alloc_name(dev, dev->name);
3874 err = register_netdevice(dev);
3879 EXPORT_SYMBOL(register_netdev);
3882 * netdev_wait_allrefs - wait until all references are gone.
3884 * This is called when unregistering network devices.
3886 * Any protocol or device that holds a reference should register
3887 * for netdevice notification, and cleanup and put back the
3888 * reference if they receive an UNREGISTER event.
3889 * We can get stuck here if buggy protocols don't correctly
3892 static void netdev_wait_allrefs(struct net_device *dev)
3894 unsigned long rebroadcast_time, warning_time;
3896 rebroadcast_time = warning_time = jiffies;
3897 while (atomic_read(&dev->refcnt) != 0) {
3898 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3901 /* Rebroadcast unregister notification */
3902 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3904 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3906 /* We must not have linkwatch events
3907 * pending on unregister. If this
3908 * happens, we simply run the queue
3909 * unscheduled, resulting in a noop
3912 linkwatch_run_queue();
3917 rebroadcast_time = jiffies;
3922 if (time_after(jiffies, warning_time + 10 * HZ)) {
3923 printk(KERN_EMERG "unregister_netdevice: "
3924 "waiting for %s to become free. Usage "
3926 dev->name, atomic_read(&dev->refcnt));
3927 warning_time = jiffies;
3936 * register_netdevice(x1);
3937 * register_netdevice(x2);
3939 * unregister_netdevice(y1);
3940 * unregister_netdevice(y2);
3946 * We are invoked by rtnl_unlock() after it drops the semaphore.
3947 * This allows us to deal with problems:
3948 * 1) We can delete sysfs objects which invoke hotplug
3949 * without deadlocking with linkwatch via keventd.
3950 * 2) Since we run with the RTNL semaphore not held, we can sleep
3951 * safely in order to wait for the netdev refcnt to drop to zero.
3953 static DEFINE_MUTEX(net_todo_run_mutex);
3954 void netdev_run_todo(void)
3956 struct list_head list;
3958 /* Need to guard against multiple cpu's getting out of order. */
3959 mutex_lock(&net_todo_run_mutex);
3961 /* Not safe to do outside the semaphore. We must not return
3962 * until all unregister events invoked by the local processor
3963 * have been completed (either by this todo run, or one on
3966 if (list_empty(&net_todo_list))
3969 /* Snapshot list, allow later requests */
3970 spin_lock(&net_todo_list_lock);
3971 list_replace_init(&net_todo_list, &list);
3972 spin_unlock(&net_todo_list_lock);
3974 while (!list_empty(&list)) {
3975 struct net_device *dev
3976 = list_entry(list.next, struct net_device, todo_list);
3977 list_del(&dev->todo_list);
3979 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3980 printk(KERN_ERR "network todo '%s' but state %d\n",
3981 dev->name, dev->reg_state);
3986 dev->reg_state = NETREG_UNREGISTERED;
3988 netdev_wait_allrefs(dev);
3991 BUG_ON(atomic_read(&dev->refcnt));
3992 BUG_TRAP(!dev->ip_ptr);
3993 BUG_TRAP(!dev->ip6_ptr);
3994 BUG_TRAP(!dev->dn_ptr);
3996 if (dev->destructor)
3997 dev->destructor(dev);
3999 /* Free network device */
4000 kobject_put(&dev->dev.kobj);
4004 mutex_unlock(&net_todo_run_mutex);
4007 static struct net_device_stats *internal_stats(struct net_device *dev)
4013 * alloc_netdev_mq - allocate network device
4014 * @sizeof_priv: size of private data to allocate space for
4015 * @name: device name format string
4016 * @setup: callback to initialize device
4017 * @queue_count: the number of subqueues to allocate
4019 * Allocates a struct net_device with private data area for driver use
4020 * and performs basic initialization. Also allocates subquue structs
4021 * for each queue on the device at the end of the netdevice.
4023 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4024 void (*setup)(struct net_device *), unsigned int queue_count)
4027 struct net_device *dev;
4030 BUG_ON(strlen(name) >= sizeof(dev->name));
4032 alloc_size = sizeof(struct net_device) +
4033 sizeof(struct net_device_subqueue) * (queue_count - 1);
4035 /* ensure 32-byte alignment of private area */
4036 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4037 alloc_size += sizeof_priv;
4039 /* ensure 32-byte alignment of whole construct */
4040 alloc_size += NETDEV_ALIGN_CONST;
4042 p = kzalloc(alloc_size, GFP_KERNEL);
4044 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4048 dev = (struct net_device *)
4049 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4050 dev->padded = (char *)dev - (char *)p;
4051 dev_net_set(dev, &init_net);
4054 dev->priv = ((char *)dev +
4055 ((sizeof(struct net_device) +
4056 (sizeof(struct net_device_subqueue) *
4057 (queue_count - 1)) + NETDEV_ALIGN_CONST)
4058 & ~NETDEV_ALIGN_CONST));
4061 dev->egress_subqueue_count = queue_count;
4062 dev->gso_max_size = GSO_MAX_SIZE;
4064 dev->get_stats = internal_stats;
4065 netpoll_netdev_init(dev);
4067 strcpy(dev->name, name);
4070 EXPORT_SYMBOL(alloc_netdev_mq);
4073 * free_netdev - free network device
4076 * This function does the last stage of destroying an allocated device
4077 * interface. The reference to the device object is released.
4078 * If this is the last reference then it will be freed.
4080 void free_netdev(struct net_device *dev)
4082 release_net(dev_net(dev));
4084 /* Compatibility with error handling in drivers */
4085 if (dev->reg_state == NETREG_UNINITIALIZED) {
4086 kfree((char *)dev - dev->padded);
4090 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4091 dev->reg_state = NETREG_RELEASED;
4093 /* will free via device release */
4094 put_device(&dev->dev);
4097 /* Synchronize with packet receive processing. */
4098 void synchronize_net(void)
4105 * unregister_netdevice - remove device from the kernel
4108 * This function shuts down a device interface and removes it
4109 * from the kernel tables.
4111 * Callers must hold the rtnl semaphore. You may want
4112 * unregister_netdev() instead of this.
4115 void unregister_netdevice(struct net_device *dev)
4119 rollback_registered(dev);
4120 /* Finish processing unregister after unlock */
4125 * unregister_netdev - remove device from the kernel
4128 * This function shuts down a device interface and removes it
4129 * from the kernel tables.
4131 * This is just a wrapper for unregister_netdevice that takes
4132 * the rtnl semaphore. In general you want to use this and not
4133 * unregister_netdevice.
4135 void unregister_netdev(struct net_device *dev)
4138 unregister_netdevice(dev);
4142 EXPORT_SYMBOL(unregister_netdev);
4145 * dev_change_net_namespace - move device to different nethost namespace
4147 * @net: network namespace
4148 * @pat: If not NULL name pattern to try if the current device name
4149 * is already taken in the destination network namespace.
4151 * This function shuts down a device interface and moves it
4152 * to a new network namespace. On success 0 is returned, on
4153 * a failure a netagive errno code is returned.
4155 * Callers must hold the rtnl semaphore.
4158 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4161 const char *destname;
4166 /* Don't allow namespace local devices to be moved. */
4168 if (dev->features & NETIF_F_NETNS_LOCAL)
4171 /* Ensure the device has been registrered */
4173 if (dev->reg_state != NETREG_REGISTERED)
4176 /* Get out if there is nothing todo */
4178 if (net_eq(dev_net(dev), net))
4181 /* Pick the destination device name, and ensure
4182 * we can use it in the destination network namespace.
4185 destname = dev->name;
4186 if (__dev_get_by_name(net, destname)) {
4187 /* We get here if we can't use the current device name */
4190 if (!dev_valid_name(pat))
4192 if (strchr(pat, '%')) {
4193 if (__dev_alloc_name(net, pat, buf) < 0)
4198 if (__dev_get_by_name(net, destname))
4203 * And now a mini version of register_netdevice unregister_netdevice.
4206 /* If device is running close it first. */
4209 /* And unlink it from device chain */
4211 unlist_netdevice(dev);
4215 /* Shutdown queueing discipline. */
4218 /* Notify protocols, that we are about to destroy
4219 this device. They should clean all the things.
4221 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4224 * Flush the unicast and multicast chains
4226 dev_addr_discard(dev);
4228 /* Actually switch the network namespace */
4229 dev_net_set(dev, net);
4231 /* Assign the new device name */
4232 if (destname != dev->name)
4233 strcpy(dev->name, destname);
4235 /* If there is an ifindex conflict assign a new one */
4236 if (__dev_get_by_index(net, dev->ifindex)) {
4237 int iflink = (dev->iflink == dev->ifindex);
4238 dev->ifindex = dev_new_index(net);
4240 dev->iflink = dev->ifindex;
4243 /* Fixup kobjects */
4244 netdev_unregister_kobject(dev);
4245 err = netdev_register_kobject(dev);
4248 /* Add the device back in the hashes */
4249 list_netdevice(dev);
4251 /* Notify protocols, that a new device appeared. */
4252 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4260 static int dev_cpu_callback(struct notifier_block *nfb,
4261 unsigned long action,
4264 struct sk_buff **list_skb;
4265 struct net_device **list_net;
4266 struct sk_buff *skb;
4267 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4268 struct softnet_data *sd, *oldsd;
4270 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4273 local_irq_disable();
4274 cpu = smp_processor_id();
4275 sd = &per_cpu(softnet_data, cpu);
4276 oldsd = &per_cpu(softnet_data, oldcpu);
4278 /* Find end of our completion_queue. */
4279 list_skb = &sd->completion_queue;
4281 list_skb = &(*list_skb)->next;
4282 /* Append completion queue from offline CPU. */
4283 *list_skb = oldsd->completion_queue;
4284 oldsd->completion_queue = NULL;
4286 /* Find end of our output_queue. */
4287 list_net = &sd->output_queue;
4289 list_net = &(*list_net)->next_sched;
4290 /* Append output queue from offline CPU. */
4291 *list_net = oldsd->output_queue;
4292 oldsd->output_queue = NULL;
4294 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4297 /* Process offline CPU's input_pkt_queue */
4298 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4304 #ifdef CONFIG_NET_DMA
4306 * net_dma_rebalance - try to maintain one DMA channel per CPU
4307 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4309 * This is called when the number of channels allocated to the net_dma client
4310 * changes. The net_dma client tries to have one DMA channel per CPU.
4313 static void net_dma_rebalance(struct net_dma *net_dma)
4315 unsigned int cpu, i, n, chan_idx;
4316 struct dma_chan *chan;
4318 if (cpus_empty(net_dma->channel_mask)) {
4319 for_each_online_cpu(cpu)
4320 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4325 cpu = first_cpu(cpu_online_map);
4327 for_each_cpu_mask(chan_idx, net_dma->channel_mask) {
4328 chan = net_dma->channels[chan_idx];
4330 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4331 + (i < (num_online_cpus() %
4332 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4335 per_cpu(softnet_data, cpu).net_dma = chan;
4336 cpu = next_cpu(cpu, cpu_online_map);
4344 * netdev_dma_event - event callback for the net_dma_client
4345 * @client: should always be net_dma_client
4346 * @chan: DMA channel for the event
4347 * @state: DMA state to be handled
4349 static enum dma_state_client
4350 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4351 enum dma_state state)
4353 int i, found = 0, pos = -1;
4354 struct net_dma *net_dma =
4355 container_of(client, struct net_dma, client);
4356 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4358 spin_lock(&net_dma->lock);
4360 case DMA_RESOURCE_AVAILABLE:
4361 for (i = 0; i < nr_cpu_ids; i++)
4362 if (net_dma->channels[i] == chan) {
4365 } else if (net_dma->channels[i] == NULL && pos < 0)
4368 if (!found && pos >= 0) {
4370 net_dma->channels[pos] = chan;
4371 cpu_set(pos, net_dma->channel_mask);
4372 net_dma_rebalance(net_dma);
4375 case DMA_RESOURCE_REMOVED:
4376 for (i = 0; i < nr_cpu_ids; i++)
4377 if (net_dma->channels[i] == chan) {
4385 cpu_clear(pos, net_dma->channel_mask);
4386 net_dma->channels[i] = NULL;
4387 net_dma_rebalance(net_dma);
4393 spin_unlock(&net_dma->lock);
4399 * netdev_dma_regiser - register the networking subsystem as a DMA client
4401 static int __init netdev_dma_register(void)
4403 net_dma.channels = kzalloc(nr_cpu_ids * sizeof(struct net_dma),
4405 if (unlikely(!net_dma.channels)) {
4407 "netdev_dma: no memory for net_dma.channels\n");
4410 spin_lock_init(&net_dma.lock);
4411 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4412 dma_async_client_register(&net_dma.client);
4413 dma_async_client_chan_request(&net_dma.client);
4418 static int __init netdev_dma_register(void) { return -ENODEV; }
4419 #endif /* CONFIG_NET_DMA */
4422 * netdev_compute_feature - compute conjunction of two feature sets
4423 * @all: first feature set
4424 * @one: second feature set
4426 * Computes a new feature set after adding a device with feature set
4427 * @one to the master device with current feature set @all. Returns
4428 * the new feature set.
4430 int netdev_compute_features(unsigned long all, unsigned long one)
4432 /* if device needs checksumming, downgrade to hw checksumming */
4433 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4434 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4436 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4437 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4438 all ^= NETIF_F_HW_CSUM
4439 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4441 if (one & NETIF_F_GSO)
4442 one |= NETIF_F_GSO_SOFTWARE;
4445 /* If even one device supports robust GSO, enable it for all. */
4446 if (one & NETIF_F_GSO_ROBUST)
4447 all |= NETIF_F_GSO_ROBUST;
4449 all &= one | NETIF_F_LLTX;
4451 if (!(all & NETIF_F_ALL_CSUM))
4453 if (!(all & NETIF_F_SG))
4454 all &= ~NETIF_F_GSO_MASK;
4458 EXPORT_SYMBOL(netdev_compute_features);
4460 static struct hlist_head *netdev_create_hash(void)
4463 struct hlist_head *hash;
4465 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4467 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4468 INIT_HLIST_HEAD(&hash[i]);
4473 /* Initialize per network namespace state */
4474 static int __net_init netdev_init(struct net *net)
4476 INIT_LIST_HEAD(&net->dev_base_head);
4478 net->dev_name_head = netdev_create_hash();
4479 if (net->dev_name_head == NULL)
4482 net->dev_index_head = netdev_create_hash();
4483 if (net->dev_index_head == NULL)
4489 kfree(net->dev_name_head);
4494 static void __net_exit netdev_exit(struct net *net)
4496 kfree(net->dev_name_head);
4497 kfree(net->dev_index_head);
4500 static struct pernet_operations __net_initdata netdev_net_ops = {
4501 .init = netdev_init,
4502 .exit = netdev_exit,
4505 static void __net_exit default_device_exit(struct net *net)
4507 struct net_device *dev, *next;
4509 * Push all migratable of the network devices back to the
4510 * initial network namespace
4513 for_each_netdev_safe(net, dev, next) {
4515 char fb_name[IFNAMSIZ];
4517 /* Ignore unmoveable devices (i.e. loopback) */
4518 if (dev->features & NETIF_F_NETNS_LOCAL)
4521 /* Push remaing network devices to init_net */
4522 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
4523 err = dev_change_net_namespace(dev, &init_net, fb_name);
4525 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
4526 __func__, dev->name, err);
4533 static struct pernet_operations __net_initdata default_device_ops = {
4534 .exit = default_device_exit,
4538 * Initialize the DEV module. At boot time this walks the device list and
4539 * unhooks any devices that fail to initialise (normally hardware not
4540 * present) and leaves us with a valid list of present and active devices.
4545 * This is called single threaded during boot, so no need
4546 * to take the rtnl semaphore.
4548 static int __init net_dev_init(void)
4550 int i, rc = -ENOMEM;
4552 BUG_ON(!dev_boot_phase);
4554 if (dev_proc_init())
4557 if (netdev_kobject_init())
4560 INIT_LIST_HEAD(&ptype_all);
4561 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4562 INIT_LIST_HEAD(&ptype_base[i]);
4564 if (register_pernet_subsys(&netdev_net_ops))
4567 if (register_pernet_device(&default_device_ops))
4571 * Initialise the packet receive queues.
4574 for_each_possible_cpu(i) {
4575 struct softnet_data *queue;
4577 queue = &per_cpu(softnet_data, i);
4578 skb_queue_head_init(&queue->input_pkt_queue);
4579 queue->completion_queue = NULL;
4580 INIT_LIST_HEAD(&queue->poll_list);
4582 queue->backlog.poll = process_backlog;
4583 queue->backlog.weight = weight_p;
4586 netdev_dma_register();
4590 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
4591 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
4593 hotcpu_notifier(dev_cpu_callback, 0);
4601 subsys_initcall(net_dev_init);
4603 EXPORT_SYMBOL(__dev_get_by_index);
4604 EXPORT_SYMBOL(__dev_get_by_name);
4605 EXPORT_SYMBOL(__dev_remove_pack);
4606 EXPORT_SYMBOL(dev_valid_name);
4607 EXPORT_SYMBOL(dev_add_pack);
4608 EXPORT_SYMBOL(dev_alloc_name);
4609 EXPORT_SYMBOL(dev_close);
4610 EXPORT_SYMBOL(dev_get_by_flags);
4611 EXPORT_SYMBOL(dev_get_by_index);
4612 EXPORT_SYMBOL(dev_get_by_name);
4613 EXPORT_SYMBOL(dev_open);
4614 EXPORT_SYMBOL(dev_queue_xmit);
4615 EXPORT_SYMBOL(dev_remove_pack);
4616 EXPORT_SYMBOL(dev_set_allmulti);
4617 EXPORT_SYMBOL(dev_set_promiscuity);
4618 EXPORT_SYMBOL(dev_change_flags);
4619 EXPORT_SYMBOL(dev_set_mtu);
4620 EXPORT_SYMBOL(dev_set_mac_address);
4621 EXPORT_SYMBOL(free_netdev);
4622 EXPORT_SYMBOL(netdev_boot_setup_check);
4623 EXPORT_SYMBOL(netdev_set_master);
4624 EXPORT_SYMBOL(netdev_state_change);
4625 EXPORT_SYMBOL(netif_receive_skb);
4626 EXPORT_SYMBOL(netif_rx);
4627 EXPORT_SYMBOL(register_gifconf);
4628 EXPORT_SYMBOL(register_netdevice);
4629 EXPORT_SYMBOL(register_netdevice_notifier);
4630 EXPORT_SYMBOL(skb_checksum_help);
4631 EXPORT_SYMBOL(synchronize_net);
4632 EXPORT_SYMBOL(unregister_netdevice);
4633 EXPORT_SYMBOL(unregister_netdevice_notifier);
4634 EXPORT_SYMBOL(net_enable_timestamp);
4635 EXPORT_SYMBOL(net_disable_timestamp);
4636 EXPORT_SYMBOL(dev_get_flags);
4638 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4639 EXPORT_SYMBOL(br_handle_frame_hook);
4640 EXPORT_SYMBOL(br_fdb_get_hook);
4641 EXPORT_SYMBOL(br_fdb_put_hook);
4645 EXPORT_SYMBOL(dev_load);
4648 EXPORT_PER_CPU_SYMBOL(softnet_data);