2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/sched.h>
83 #include <linux/mutex.h>
84 #include <linux/string.h>
86 #include <linux/socket.h>
87 #include <linux/sockios.h>
88 #include <linux/errno.h>
89 #include <linux/interrupt.h>
90 #include <linux/if_ether.h>
91 #include <linux/netdevice.h>
92 #include <linux/etherdevice.h>
93 #include <linux/notifier.h>
94 #include <linux/skbuff.h>
95 #include <net/net_namespace.h>
97 #include <linux/rtnetlink.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <linux/stat.h>
101 #include <linux/if_bridge.h>
102 #include <linux/if_macvlan.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/kmod.h>
109 #include <linux/module.h>
110 #include <linux/kallsyms.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
124 * The list of packet types we will receive (as opposed to discard)
125 * and the routines to invoke.
127 * Why 16. Because with 16 the only overlap we get on a hash of the
128 * low nibble of the protocol value is RARP/SNAP/X.25.
130 * NOTE: That is no longer true with the addition of VLAN tags. Not
131 * sure which should go first, but I bet it won't make much
132 * difference if we are running VLANs. The good news is that
133 * this protocol won't be in the list unless compiled in, so
134 * the average user (w/out VLANs) will not be adversely affected.
151 static DEFINE_SPINLOCK(ptype_lock);
152 static struct list_head ptype_base[16] __read_mostly; /* 16 way hashed list */
153 static struct list_head ptype_all __read_mostly; /* Taps */
155 #ifdef CONFIG_NET_DMA
157 struct dma_client client;
159 cpumask_t channel_mask;
160 struct dma_chan *channels[NR_CPUS];
163 static enum dma_state_client
164 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
165 enum dma_state state);
167 static struct net_dma net_dma = {
169 .event_callback = netdev_dma_event,
175 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
178 * Pure readers hold dev_base_lock for reading.
180 * Writers must hold the rtnl semaphore while they loop through the
181 * dev_base_head list, and hold dev_base_lock for writing when they do the
182 * actual updates. This allows pure readers to access the list even
183 * while a writer is preparing to update it.
185 * To put it another way, dev_base_lock is held for writing only to
186 * protect against pure readers; the rtnl semaphore provides the
187 * protection against other writers.
189 * See, for example usages, register_netdevice() and
190 * unregister_netdevice(), which must be called with the rtnl
193 DEFINE_RWLOCK(dev_base_lock);
195 EXPORT_SYMBOL(dev_base_lock);
197 #define NETDEV_HASHBITS 8
198 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
200 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
202 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
203 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
208 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
211 /* Device list insertion */
212 static int list_netdevice(struct net_device *dev)
214 struct net *net = dev->nd_net;
218 write_lock_bh(&dev_base_lock);
219 list_add_tail(&dev->dev_list, &net->dev_base_head);
220 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
221 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
222 write_unlock_bh(&dev_base_lock);
226 /* Device list removal */
227 static void unlist_netdevice(struct net_device *dev)
231 /* Unlink dev from the device chain */
232 write_lock_bh(&dev_base_lock);
233 list_del(&dev->dev_list);
234 hlist_del(&dev->name_hlist);
235 hlist_del(&dev->index_hlist);
236 write_unlock_bh(&dev_base_lock);
243 static RAW_NOTIFIER_HEAD(netdev_chain);
246 * Device drivers call our routines to queue packets here. We empty the
247 * queue in the local softnet handler.
250 DEFINE_PER_CPU(struct softnet_data, softnet_data);
253 extern int netdev_sysfs_init(void);
254 extern int netdev_register_sysfs(struct net_device *);
255 extern void netdev_unregister_sysfs(struct net_device *);
257 #define netdev_sysfs_init() (0)
258 #define netdev_register_sysfs(dev) (0)
259 #define netdev_unregister_sysfs(dev) do { } while(0)
262 #ifdef CONFIG_DEBUG_LOCK_ALLOC
264 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
265 * according to dev->type
267 static const unsigned short netdev_lock_type[] =
268 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
269 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
270 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
271 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
272 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
273 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
274 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
275 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
276 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
277 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
278 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
279 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
280 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
281 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
284 static const char *netdev_lock_name[] =
285 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
286 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
287 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
288 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
289 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
290 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
291 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
292 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
293 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
294 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
295 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
296 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
297 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
298 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
301 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
303 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
307 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
308 if (netdev_lock_type[i] == dev_type)
310 /* the last key is used by default */
311 return ARRAY_SIZE(netdev_lock_type) - 1;
314 static inline void netdev_set_lockdep_class(spinlock_t *lock,
315 unsigned short dev_type)
319 i = netdev_lock_pos(dev_type);
320 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
321 netdev_lock_name[i]);
324 static inline void netdev_set_lockdep_class(spinlock_t *lock,
325 unsigned short dev_type)
330 /*******************************************************************************
332 Protocol management and registration routines
334 *******************************************************************************/
337 * Add a protocol ID to the list. Now that the input handler is
338 * smarter we can dispense with all the messy stuff that used to be
341 * BEWARE!!! Protocol handlers, mangling input packets,
342 * MUST BE last in hash buckets and checking protocol handlers
343 * MUST start from promiscuous ptype_all chain in net_bh.
344 * It is true now, do not change it.
345 * Explanation follows: if protocol handler, mangling packet, will
346 * be the first on list, it is not able to sense, that packet
347 * is cloned and should be copied-on-write, so that it will
348 * change it and subsequent readers will get broken packet.
353 * dev_add_pack - add packet handler
354 * @pt: packet type declaration
356 * Add a protocol handler to the networking stack. The passed &packet_type
357 * is linked into kernel lists and may not be freed until it has been
358 * removed from the kernel lists.
360 * This call does not sleep therefore it can not
361 * guarantee all CPU's that are in middle of receiving packets
362 * will see the new packet type (until the next received packet).
365 void dev_add_pack(struct packet_type *pt)
369 spin_lock_bh(&ptype_lock);
370 if (pt->type == htons(ETH_P_ALL))
371 list_add_rcu(&pt->list, &ptype_all);
373 hash = ntohs(pt->type) & 15;
374 list_add_rcu(&pt->list, &ptype_base[hash]);
376 spin_unlock_bh(&ptype_lock);
380 * __dev_remove_pack - remove packet handler
381 * @pt: packet type declaration
383 * Remove a protocol handler that was previously added to the kernel
384 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
385 * from the kernel lists and can be freed or reused once this function
388 * The packet type might still be in use by receivers
389 * and must not be freed until after all the CPU's have gone
390 * through a quiescent state.
392 void __dev_remove_pack(struct packet_type *pt)
394 struct list_head *head;
395 struct packet_type *pt1;
397 spin_lock_bh(&ptype_lock);
399 if (pt->type == htons(ETH_P_ALL))
402 head = &ptype_base[ntohs(pt->type) & 15];
404 list_for_each_entry(pt1, head, list) {
406 list_del_rcu(&pt->list);
411 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
413 spin_unlock_bh(&ptype_lock);
416 * dev_remove_pack - remove packet handler
417 * @pt: packet type declaration
419 * Remove a protocol handler that was previously added to the kernel
420 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
421 * from the kernel lists and can be freed or reused once this function
424 * This call sleeps to guarantee that no CPU is looking at the packet
427 void dev_remove_pack(struct packet_type *pt)
429 __dev_remove_pack(pt);
434 /******************************************************************************
436 Device Boot-time Settings Routines
438 *******************************************************************************/
440 /* Boot time configuration table */
441 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
444 * netdev_boot_setup_add - add new setup entry
445 * @name: name of the device
446 * @map: configured settings for the device
448 * Adds new setup entry to the dev_boot_setup list. The function
449 * returns 0 on error and 1 on success. This is a generic routine to
452 static int netdev_boot_setup_add(char *name, struct ifmap *map)
454 struct netdev_boot_setup *s;
458 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
459 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
460 memset(s[i].name, 0, sizeof(s[i].name));
461 strcpy(s[i].name, name);
462 memcpy(&s[i].map, map, sizeof(s[i].map));
467 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
471 * netdev_boot_setup_check - check boot time settings
472 * @dev: the netdevice
474 * Check boot time settings for the device.
475 * The found settings are set for the device to be used
476 * later in the device probing.
477 * Returns 0 if no settings found, 1 if they are.
479 int netdev_boot_setup_check(struct net_device *dev)
481 struct netdev_boot_setup *s = dev_boot_setup;
484 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
485 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
486 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
487 dev->irq = s[i].map.irq;
488 dev->base_addr = s[i].map.base_addr;
489 dev->mem_start = s[i].map.mem_start;
490 dev->mem_end = s[i].map.mem_end;
499 * netdev_boot_base - get address from boot time settings
500 * @prefix: prefix for network device
501 * @unit: id for network device
503 * Check boot time settings for the base address of device.
504 * The found settings are set for the device to be used
505 * later in the device probing.
506 * Returns 0 if no settings found.
508 unsigned long netdev_boot_base(const char *prefix, int unit)
510 const struct netdev_boot_setup *s = dev_boot_setup;
514 sprintf(name, "%s%d", prefix, unit);
517 * If device already registered then return base of 1
518 * to indicate not to probe for this interface
520 if (__dev_get_by_name(&init_net, name))
523 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
524 if (!strcmp(name, s[i].name))
525 return s[i].map.base_addr;
530 * Saves at boot time configured settings for any netdevice.
532 int __init netdev_boot_setup(char *str)
537 str = get_options(str, ARRAY_SIZE(ints), ints);
542 memset(&map, 0, sizeof(map));
546 map.base_addr = ints[2];
548 map.mem_start = ints[3];
550 map.mem_end = ints[4];
552 /* Add new entry to the list */
553 return netdev_boot_setup_add(str, &map);
556 __setup("netdev=", netdev_boot_setup);
558 /*******************************************************************************
560 Device Interface Subroutines
562 *******************************************************************************/
565 * __dev_get_by_name - find a device by its name
566 * @name: name to find
568 * Find an interface by name. Must be called under RTNL semaphore
569 * or @dev_base_lock. If the name is found a pointer to the device
570 * is returned. If the name is not found then %NULL is returned. The
571 * reference counters are not incremented so the caller must be
572 * careful with locks.
575 struct net_device *__dev_get_by_name(struct net *net, const char *name)
577 struct hlist_node *p;
579 hlist_for_each(p, dev_name_hash(net, name)) {
580 struct net_device *dev
581 = hlist_entry(p, struct net_device, name_hlist);
582 if (!strncmp(dev->name, name, IFNAMSIZ))
589 * dev_get_by_name - find a device by its name
590 * @name: name to find
592 * Find an interface by name. This can be called from any
593 * context and does its own locking. The returned handle has
594 * the usage count incremented and the caller must use dev_put() to
595 * release it when it is no longer needed. %NULL is returned if no
596 * matching device is found.
599 struct net_device *dev_get_by_name(struct net *net, const char *name)
601 struct net_device *dev;
603 read_lock(&dev_base_lock);
604 dev = __dev_get_by_name(net, name);
607 read_unlock(&dev_base_lock);
612 * __dev_get_by_index - find a device by its ifindex
613 * @ifindex: index of device
615 * Search for an interface by index. Returns %NULL if the device
616 * is not found or a pointer to the device. The device has not
617 * had its reference counter increased so the caller must be careful
618 * about locking. The caller must hold either the RTNL semaphore
622 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
624 struct hlist_node *p;
626 hlist_for_each(p, dev_index_hash(net, ifindex)) {
627 struct net_device *dev
628 = hlist_entry(p, struct net_device, index_hlist);
629 if (dev->ifindex == ifindex)
637 * dev_get_by_index - find a device by its ifindex
638 * @ifindex: index of device
640 * Search for an interface by index. Returns NULL if the device
641 * is not found or a pointer to the device. The device returned has
642 * had a reference added and the pointer is safe until the user calls
643 * dev_put to indicate they have finished with it.
646 struct net_device *dev_get_by_index(struct net *net, int ifindex)
648 struct net_device *dev;
650 read_lock(&dev_base_lock);
651 dev = __dev_get_by_index(net, ifindex);
654 read_unlock(&dev_base_lock);
659 * dev_getbyhwaddr - find a device by its hardware address
660 * @type: media type of device
661 * @ha: hardware address
663 * Search for an interface by MAC address. Returns NULL if the device
664 * is not found or a pointer to the device. The caller must hold the
665 * rtnl semaphore. The returned device has not had its ref count increased
666 * and the caller must therefore be careful about locking
669 * If the API was consistent this would be __dev_get_by_hwaddr
672 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
674 struct net_device *dev;
678 for_each_netdev(&init_net, dev)
679 if (dev->type == type &&
680 !memcmp(dev->dev_addr, ha, dev->addr_len))
686 EXPORT_SYMBOL(dev_getbyhwaddr);
688 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
690 struct net_device *dev;
693 for_each_netdev(net, dev)
694 if (dev->type == type)
700 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
702 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
704 struct net_device *dev;
707 dev = __dev_getfirstbyhwtype(net, type);
714 EXPORT_SYMBOL(dev_getfirstbyhwtype);
717 * dev_get_by_flags - find any device with given flags
718 * @if_flags: IFF_* values
719 * @mask: bitmask of bits in if_flags to check
721 * Search for any interface with the given flags. Returns NULL if a device
722 * is not found or a pointer to the device. The device returned has
723 * had a reference added and the pointer is safe until the user calls
724 * dev_put to indicate they have finished with it.
727 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
729 struct net_device *dev, *ret;
732 read_lock(&dev_base_lock);
733 for_each_netdev(net, dev) {
734 if (((dev->flags ^ if_flags) & mask) == 0) {
740 read_unlock(&dev_base_lock);
745 * dev_valid_name - check if name is okay for network device
748 * Network device names need to be valid file names to
749 * to allow sysfs to work. We also disallow any kind of
752 int dev_valid_name(const char *name)
756 if (strlen(name) >= IFNAMSIZ)
758 if (!strcmp(name, ".") || !strcmp(name, ".."))
762 if (*name == '/' || isspace(*name))
770 * __dev_alloc_name - allocate a name for a device
771 * @net: network namespace to allocate the device name in
772 * @name: name format string
773 * @buf: scratch buffer and result name string
775 * Passed a format string - eg "lt%d" it will try and find a suitable
776 * id. It scans list of devices to build up a free map, then chooses
777 * the first empty slot. The caller must hold the dev_base or rtnl lock
778 * while allocating the name and adding the device in order to avoid
780 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
781 * Returns the number of the unit assigned or a negative errno code.
784 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
788 const int max_netdevices = 8*PAGE_SIZE;
790 struct net_device *d;
792 p = strnchr(name, IFNAMSIZ-1, '%');
795 * Verify the string as this thing may have come from
796 * the user. There must be either one "%d" and no other "%"
799 if (p[1] != 'd' || strchr(p + 2, '%'))
802 /* Use one page as a bit array of possible slots */
803 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
807 for_each_netdev(net, d) {
808 if (!sscanf(d->name, name, &i))
810 if (i < 0 || i >= max_netdevices)
813 /* avoid cases where sscanf is not exact inverse of printf */
814 snprintf(buf, IFNAMSIZ, name, i);
815 if (!strncmp(buf, d->name, IFNAMSIZ))
819 i = find_first_zero_bit(inuse, max_netdevices);
820 free_page((unsigned long) inuse);
823 snprintf(buf, IFNAMSIZ, name, i);
824 if (!__dev_get_by_name(net, buf))
827 /* It is possible to run out of possible slots
828 * when the name is long and there isn't enough space left
829 * for the digits, or if all bits are used.
835 * dev_alloc_name - allocate a name for a device
837 * @name: name format string
839 * Passed a format string - eg "lt%d" it will try and find a suitable
840 * id. It scans list of devices to build up a free map, then chooses
841 * the first empty slot. The caller must hold the dev_base or rtnl lock
842 * while allocating the name and adding the device in order to avoid
844 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
845 * Returns the number of the unit assigned or a negative errno code.
848 int dev_alloc_name(struct net_device *dev, const char *name)
854 BUG_ON(!dev->nd_net);
856 ret = __dev_alloc_name(net, name, buf);
858 strlcpy(dev->name, buf, IFNAMSIZ);
864 * dev_change_name - change name of a device
866 * @newname: name (or format string) must be at least IFNAMSIZ
868 * Change name of a device, can pass format strings "eth%d".
871 int dev_change_name(struct net_device *dev, char *newname)
873 char oldname[IFNAMSIZ];
879 BUG_ON(!dev->nd_net);
882 if (dev->flags & IFF_UP)
885 if (!dev_valid_name(newname))
888 memcpy(oldname, dev->name, IFNAMSIZ);
890 if (strchr(newname, '%')) {
891 err = dev_alloc_name(dev, newname);
894 strcpy(newname, dev->name);
896 else if (__dev_get_by_name(net, newname))
899 strlcpy(dev->name, newname, IFNAMSIZ);
902 device_rename(&dev->dev, dev->name);
904 write_lock_bh(&dev_base_lock);
905 hlist_del(&dev->name_hlist);
906 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
907 write_unlock_bh(&dev_base_lock);
909 ret = raw_notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
910 ret = notifier_to_errno(ret);
915 "%s: name change rollback failed: %d.\n",
919 memcpy(dev->name, oldname, IFNAMSIZ);
928 * netdev_features_change - device changes features
929 * @dev: device to cause notification
931 * Called to indicate a device has changed features.
933 void netdev_features_change(struct net_device *dev)
935 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
937 EXPORT_SYMBOL(netdev_features_change);
940 * netdev_state_change - device changes state
941 * @dev: device to cause notification
943 * Called to indicate a device has changed state. This function calls
944 * the notifier chains for netdev_chain and sends a NEWLINK message
945 * to the routing socket.
947 void netdev_state_change(struct net_device *dev)
949 if (dev->flags & IFF_UP) {
950 raw_notifier_call_chain(&netdev_chain,
952 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
957 * dev_load - load a network module
958 * @name: name of interface
960 * If a network interface is not present and the process has suitable
961 * privileges this function loads the module. If module loading is not
962 * available in this kernel then it becomes a nop.
965 void dev_load(struct net *net, const char *name)
967 struct net_device *dev;
969 read_lock(&dev_base_lock);
970 dev = __dev_get_by_name(net, name);
971 read_unlock(&dev_base_lock);
973 if (!dev && capable(CAP_SYS_MODULE))
974 request_module("%s", name);
977 static int default_rebuild_header(struct sk_buff *skb)
979 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
980 skb->dev ? skb->dev->name : "NULL!!!");
986 * dev_open - prepare an interface for use.
987 * @dev: device to open
989 * Takes a device from down to up state. The device's private open
990 * function is invoked and then the multicast lists are loaded. Finally
991 * the device is moved into the up state and a %NETDEV_UP message is
992 * sent to the netdev notifier chain.
994 * Calling this function on an active interface is a nop. On a failure
995 * a negative errno code is returned.
997 int dev_open(struct net_device *dev)
1005 if (dev->flags & IFF_UP)
1009 * Is it even present?
1011 if (!netif_device_present(dev))
1015 * Call device private open method
1017 set_bit(__LINK_STATE_START, &dev->state);
1019 ret = dev->open(dev);
1021 clear_bit(__LINK_STATE_START, &dev->state);
1025 * If it went open OK then:
1032 dev->flags |= IFF_UP;
1035 * Initialize multicasting status
1037 dev_set_rx_mode(dev);
1040 * Wakeup transmit queue engine
1045 * ... and announce new interface.
1047 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
1053 * dev_close - shutdown an interface.
1054 * @dev: device to shutdown
1056 * This function moves an active device into down state. A
1057 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1058 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1061 int dev_close(struct net_device *dev)
1065 if (!(dev->flags & IFF_UP))
1069 * Tell people we are going down, so that they can
1070 * prepare to death, when device is still operating.
1072 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
1074 dev_deactivate(dev);
1076 clear_bit(__LINK_STATE_START, &dev->state);
1078 /* Synchronize to scheduled poll. We cannot touch poll list,
1079 * it can be even on different cpu. So just clear netif_running().
1081 * dev->stop() will invoke napi_disable() on all of it's
1082 * napi_struct instances on this device.
1084 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1087 * Call the device specific close. This cannot fail.
1088 * Only if device is UP
1090 * We allow it to be called even after a DETACH hot-plug
1097 * Device is now down.
1100 dev->flags &= ~IFF_UP;
1103 * Tell people we are down
1105 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1111 static int dev_boot_phase = 1;
1114 * Device change register/unregister. These are not inline or static
1115 * as we export them to the world.
1119 * register_netdevice_notifier - register a network notifier block
1122 * Register a notifier to be called when network device events occur.
1123 * The notifier passed is linked into the kernel structures and must
1124 * not be reused until it has been unregistered. A negative errno code
1125 * is returned on a failure.
1127 * When registered all registration and up events are replayed
1128 * to the new notifier to allow device to have a race free
1129 * view of the network device list.
1132 int register_netdevice_notifier(struct notifier_block *nb)
1134 struct net_device *dev;
1135 struct net_device *last;
1140 err = raw_notifier_chain_register(&netdev_chain, nb);
1146 for_each_netdev(net, dev) {
1147 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1148 err = notifier_to_errno(err);
1152 if (!(dev->flags & IFF_UP))
1155 nb->notifier_call(nb, NETDEV_UP, dev);
1166 for_each_netdev(net, dev) {
1170 if (dev->flags & IFF_UP) {
1171 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1172 nb->notifier_call(nb, NETDEV_DOWN, dev);
1174 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1181 * unregister_netdevice_notifier - unregister a network notifier block
1184 * Unregister a notifier previously registered by
1185 * register_netdevice_notifier(). The notifier is unlinked into the
1186 * kernel structures and may then be reused. A negative errno code
1187 * is returned on a failure.
1190 int unregister_netdevice_notifier(struct notifier_block *nb)
1195 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1201 * call_netdevice_notifiers - call all network notifier blocks
1202 * @val: value passed unmodified to notifier function
1203 * @v: pointer passed unmodified to notifier function
1205 * Call all network notifier blocks. Parameters and return value
1206 * are as for raw_notifier_call_chain().
1209 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1211 return raw_notifier_call_chain(&netdev_chain, val, dev);
1214 /* When > 0 there are consumers of rx skb time stamps */
1215 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1217 void net_enable_timestamp(void)
1219 atomic_inc(&netstamp_needed);
1222 void net_disable_timestamp(void)
1224 atomic_dec(&netstamp_needed);
1227 static inline void net_timestamp(struct sk_buff *skb)
1229 if (atomic_read(&netstamp_needed))
1230 __net_timestamp(skb);
1232 skb->tstamp.tv64 = 0;
1236 * Support routine. Sends outgoing frames to any network
1237 * taps currently in use.
1240 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1242 struct packet_type *ptype;
1247 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1248 /* Never send packets back to the socket
1249 * they originated from - MvS (miquels@drinkel.ow.org)
1251 if ((ptype->dev == dev || !ptype->dev) &&
1252 (ptype->af_packet_priv == NULL ||
1253 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1254 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1258 /* skb->nh should be correctly
1259 set by sender, so that the second statement is
1260 just protection against buggy protocols.
1262 skb_reset_mac_header(skb2);
1264 if (skb_network_header(skb2) < skb2->data ||
1265 skb2->network_header > skb2->tail) {
1266 if (net_ratelimit())
1267 printk(KERN_CRIT "protocol %04x is "
1269 skb2->protocol, dev->name);
1270 skb_reset_network_header(skb2);
1273 skb2->transport_header = skb2->network_header;
1274 skb2->pkt_type = PACKET_OUTGOING;
1275 ptype->func(skb2, skb->dev, ptype, skb->dev);
1282 void __netif_schedule(struct net_device *dev)
1284 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1285 unsigned long flags;
1286 struct softnet_data *sd;
1288 local_irq_save(flags);
1289 sd = &__get_cpu_var(softnet_data);
1290 dev->next_sched = sd->output_queue;
1291 sd->output_queue = dev;
1292 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1293 local_irq_restore(flags);
1296 EXPORT_SYMBOL(__netif_schedule);
1298 void dev_kfree_skb_irq(struct sk_buff *skb)
1300 if (atomic_dec_and_test(&skb->users)) {
1301 struct softnet_data *sd;
1302 unsigned long flags;
1304 local_irq_save(flags);
1305 sd = &__get_cpu_var(softnet_data);
1306 skb->next = sd->completion_queue;
1307 sd->completion_queue = skb;
1308 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1309 local_irq_restore(flags);
1312 EXPORT_SYMBOL(dev_kfree_skb_irq);
1314 void dev_kfree_skb_any(struct sk_buff *skb)
1316 if (in_irq() || irqs_disabled())
1317 dev_kfree_skb_irq(skb);
1321 EXPORT_SYMBOL(dev_kfree_skb_any);
1325 * netif_device_detach - mark device as removed
1326 * @dev: network device
1328 * Mark device as removed from system and therefore no longer available.
1330 void netif_device_detach(struct net_device *dev)
1332 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1333 netif_running(dev)) {
1334 netif_stop_queue(dev);
1337 EXPORT_SYMBOL(netif_device_detach);
1340 * netif_device_attach - mark device as attached
1341 * @dev: network device
1343 * Mark device as attached from system and restart if needed.
1345 void netif_device_attach(struct net_device *dev)
1347 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1348 netif_running(dev)) {
1349 netif_wake_queue(dev);
1350 __netdev_watchdog_up(dev);
1353 EXPORT_SYMBOL(netif_device_attach);
1357 * Invalidate hardware checksum when packet is to be mangled, and
1358 * complete checksum manually on outgoing path.
1360 int skb_checksum_help(struct sk_buff *skb)
1363 int ret = 0, offset;
1365 if (skb->ip_summed == CHECKSUM_COMPLETE)
1366 goto out_set_summed;
1368 if (unlikely(skb_shinfo(skb)->gso_size)) {
1369 /* Let GSO fix up the checksum. */
1370 goto out_set_summed;
1373 if (skb_cloned(skb)) {
1374 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1379 offset = skb->csum_start - skb_headroom(skb);
1380 BUG_ON(offset > (int)skb->len);
1381 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1383 offset = skb_headlen(skb) - offset;
1384 BUG_ON(offset <= 0);
1385 BUG_ON(skb->csum_offset + 2 > offset);
1387 *(__sum16 *)(skb->head + skb->csum_start + skb->csum_offset) =
1390 skb->ip_summed = CHECKSUM_NONE;
1396 * skb_gso_segment - Perform segmentation on skb.
1397 * @skb: buffer to segment
1398 * @features: features for the output path (see dev->features)
1400 * This function segments the given skb and returns a list of segments.
1402 * It may return NULL if the skb requires no segmentation. This is
1403 * only possible when GSO is used for verifying header integrity.
1405 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1407 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1408 struct packet_type *ptype;
1409 __be16 type = skb->protocol;
1412 BUG_ON(skb_shinfo(skb)->frag_list);
1414 skb_reset_mac_header(skb);
1415 skb->mac_len = skb->network_header - skb->mac_header;
1416 __skb_pull(skb, skb->mac_len);
1418 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1419 if (skb_header_cloned(skb) &&
1420 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1421 return ERR_PTR(err);
1425 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1426 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1427 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1428 err = ptype->gso_send_check(skb);
1429 segs = ERR_PTR(err);
1430 if (err || skb_gso_ok(skb, features))
1432 __skb_push(skb, (skb->data -
1433 skb_network_header(skb)));
1435 segs = ptype->gso_segment(skb, features);
1441 __skb_push(skb, skb->data - skb_mac_header(skb));
1446 EXPORT_SYMBOL(skb_gso_segment);
1448 /* Take action when hardware reception checksum errors are detected. */
1450 void netdev_rx_csum_fault(struct net_device *dev)
1452 if (net_ratelimit()) {
1453 printk(KERN_ERR "%s: hw csum failure.\n",
1454 dev ? dev->name : "<unknown>");
1458 EXPORT_SYMBOL(netdev_rx_csum_fault);
1461 /* Actually, we should eliminate this check as soon as we know, that:
1462 * 1. IOMMU is present and allows to map all the memory.
1463 * 2. No high memory really exists on this machine.
1466 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1468 #ifdef CONFIG_HIGHMEM
1471 if (dev->features & NETIF_F_HIGHDMA)
1474 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1475 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1483 void (*destructor)(struct sk_buff *skb);
1486 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1488 static void dev_gso_skb_destructor(struct sk_buff *skb)
1490 struct dev_gso_cb *cb;
1493 struct sk_buff *nskb = skb->next;
1495 skb->next = nskb->next;
1498 } while (skb->next);
1500 cb = DEV_GSO_CB(skb);
1502 cb->destructor(skb);
1506 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1507 * @skb: buffer to segment
1509 * This function segments the given skb and stores the list of segments
1512 static int dev_gso_segment(struct sk_buff *skb)
1514 struct net_device *dev = skb->dev;
1515 struct sk_buff *segs;
1516 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1519 segs = skb_gso_segment(skb, features);
1521 /* Verifying header integrity only. */
1525 if (unlikely(IS_ERR(segs)))
1526 return PTR_ERR(segs);
1529 DEV_GSO_CB(skb)->destructor = skb->destructor;
1530 skb->destructor = dev_gso_skb_destructor;
1535 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1537 if (likely(!skb->next)) {
1538 if (!list_empty(&ptype_all))
1539 dev_queue_xmit_nit(skb, dev);
1541 if (netif_needs_gso(dev, skb)) {
1542 if (unlikely(dev_gso_segment(skb)))
1548 return dev->hard_start_xmit(skb, dev);
1553 struct sk_buff *nskb = skb->next;
1556 skb->next = nskb->next;
1558 rc = dev->hard_start_xmit(nskb, dev);
1560 nskb->next = skb->next;
1564 if (unlikely((netif_queue_stopped(dev) ||
1565 netif_subqueue_stopped(dev, skb->queue_mapping)) &&
1567 return NETDEV_TX_BUSY;
1568 } while (skb->next);
1570 skb->destructor = DEV_GSO_CB(skb)->destructor;
1578 * dev_queue_xmit - transmit a buffer
1579 * @skb: buffer to transmit
1581 * Queue a buffer for transmission to a network device. The caller must
1582 * have set the device and priority and built the buffer before calling
1583 * this function. The function can be called from an interrupt.
1585 * A negative errno code is returned on a failure. A success does not
1586 * guarantee the frame will be transmitted as it may be dropped due
1587 * to congestion or traffic shaping.
1589 * -----------------------------------------------------------------------------------
1590 * I notice this method can also return errors from the queue disciplines,
1591 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1594 * Regardless of the return value, the skb is consumed, so it is currently
1595 * difficult to retry a send to this method. (You can bump the ref count
1596 * before sending to hold a reference for retry if you are careful.)
1598 * When calling this method, interrupts MUST be enabled. This is because
1599 * the BH enable code must have IRQs enabled so that it will not deadlock.
1603 int dev_queue_xmit(struct sk_buff *skb)
1605 struct net_device *dev = skb->dev;
1609 /* GSO will handle the following emulations directly. */
1610 if (netif_needs_gso(dev, skb))
1613 if (skb_shinfo(skb)->frag_list &&
1614 !(dev->features & NETIF_F_FRAGLIST) &&
1615 __skb_linearize(skb))
1618 /* Fragmented skb is linearized if device does not support SG,
1619 * or if at least one of fragments is in highmem and device
1620 * does not support DMA from it.
1622 if (skb_shinfo(skb)->nr_frags &&
1623 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1624 __skb_linearize(skb))
1627 /* If packet is not checksummed and device does not support
1628 * checksumming for this protocol, complete checksumming here.
1630 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1631 skb_set_transport_header(skb, skb->csum_start -
1634 if (!(dev->features & NETIF_F_GEN_CSUM) &&
1635 !((dev->features & NETIF_F_IP_CSUM) &&
1636 skb->protocol == htons(ETH_P_IP)) &&
1637 !((dev->features & NETIF_F_IPV6_CSUM) &&
1638 skb->protocol == htons(ETH_P_IPV6)))
1639 if (skb_checksum_help(skb))
1644 spin_lock_prefetch(&dev->queue_lock);
1646 /* Disable soft irqs for various locks below. Also
1647 * stops preemption for RCU.
1651 /* Updates of qdisc are serialized by queue_lock.
1652 * The struct Qdisc which is pointed to by qdisc is now a
1653 * rcu structure - it may be accessed without acquiring
1654 * a lock (but the structure may be stale.) The freeing of the
1655 * qdisc will be deferred until it's known that there are no
1656 * more references to it.
1658 * If the qdisc has an enqueue function, we still need to
1659 * hold the queue_lock before calling it, since queue_lock
1660 * also serializes access to the device queue.
1663 q = rcu_dereference(dev->qdisc);
1664 #ifdef CONFIG_NET_CLS_ACT
1665 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1668 /* Grab device queue */
1669 spin_lock(&dev->queue_lock);
1672 /* reset queue_mapping to zero */
1673 skb->queue_mapping = 0;
1674 rc = q->enqueue(skb, q);
1676 spin_unlock(&dev->queue_lock);
1678 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1681 spin_unlock(&dev->queue_lock);
1684 /* The device has no queue. Common case for software devices:
1685 loopback, all the sorts of tunnels...
1687 Really, it is unlikely that netif_tx_lock protection is necessary
1688 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1690 However, it is possible, that they rely on protection
1693 Check this and shot the lock. It is not prone from deadlocks.
1694 Either shot noqueue qdisc, it is even simpler 8)
1696 if (dev->flags & IFF_UP) {
1697 int cpu = smp_processor_id(); /* ok because BHs are off */
1699 if (dev->xmit_lock_owner != cpu) {
1701 HARD_TX_LOCK(dev, cpu);
1703 if (!netif_queue_stopped(dev) &&
1704 !netif_subqueue_stopped(dev, skb->queue_mapping)) {
1706 if (!dev_hard_start_xmit(skb, dev)) {
1707 HARD_TX_UNLOCK(dev);
1711 HARD_TX_UNLOCK(dev);
1712 if (net_ratelimit())
1713 printk(KERN_CRIT "Virtual device %s asks to "
1714 "queue packet!\n", dev->name);
1716 /* Recursion is detected! It is possible,
1718 if (net_ratelimit())
1719 printk(KERN_CRIT "Dead loop on virtual device "
1720 "%s, fix it urgently!\n", dev->name);
1725 rcu_read_unlock_bh();
1731 rcu_read_unlock_bh();
1736 /*=======================================================================
1738 =======================================================================*/
1740 int netdev_max_backlog __read_mostly = 1000;
1741 int netdev_budget __read_mostly = 300;
1742 int weight_p __read_mostly = 64; /* old backlog weight */
1744 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1748 * netif_rx - post buffer to the network code
1749 * @skb: buffer to post
1751 * This function receives a packet from a device driver and queues it for
1752 * the upper (protocol) levels to process. It always succeeds. The buffer
1753 * may be dropped during processing for congestion control or by the
1757 * NET_RX_SUCCESS (no congestion)
1758 * NET_RX_CN_LOW (low congestion)
1759 * NET_RX_CN_MOD (moderate congestion)
1760 * NET_RX_CN_HIGH (high congestion)
1761 * NET_RX_DROP (packet was dropped)
1765 int netif_rx(struct sk_buff *skb)
1767 struct softnet_data *queue;
1768 unsigned long flags;
1770 /* if netpoll wants it, pretend we never saw it */
1771 if (netpoll_rx(skb))
1774 if (!skb->tstamp.tv64)
1778 * The code is rearranged so that the path is the most
1779 * short when CPU is congested, but is still operating.
1781 local_irq_save(flags);
1782 queue = &__get_cpu_var(softnet_data);
1784 __get_cpu_var(netdev_rx_stat).total++;
1785 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1786 if (queue->input_pkt_queue.qlen) {
1789 __skb_queue_tail(&queue->input_pkt_queue, skb);
1790 local_irq_restore(flags);
1791 return NET_RX_SUCCESS;
1794 napi_schedule(&queue->backlog);
1798 __get_cpu_var(netdev_rx_stat).dropped++;
1799 local_irq_restore(flags);
1805 int netif_rx_ni(struct sk_buff *skb)
1810 err = netif_rx(skb);
1811 if (local_softirq_pending())
1818 EXPORT_SYMBOL(netif_rx_ni);
1820 static inline struct net_device *skb_bond(struct sk_buff *skb)
1822 struct net_device *dev = skb->dev;
1825 if (skb_bond_should_drop(skb)) {
1829 skb->dev = dev->master;
1836 static void net_tx_action(struct softirq_action *h)
1838 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1840 if (sd->completion_queue) {
1841 struct sk_buff *clist;
1843 local_irq_disable();
1844 clist = sd->completion_queue;
1845 sd->completion_queue = NULL;
1849 struct sk_buff *skb = clist;
1850 clist = clist->next;
1852 BUG_TRAP(!atomic_read(&skb->users));
1857 if (sd->output_queue) {
1858 struct net_device *head;
1860 local_irq_disable();
1861 head = sd->output_queue;
1862 sd->output_queue = NULL;
1866 struct net_device *dev = head;
1867 head = head->next_sched;
1869 smp_mb__before_clear_bit();
1870 clear_bit(__LINK_STATE_SCHED, &dev->state);
1872 if (spin_trylock(&dev->queue_lock)) {
1874 spin_unlock(&dev->queue_lock);
1876 netif_schedule(dev);
1882 static inline int deliver_skb(struct sk_buff *skb,
1883 struct packet_type *pt_prev,
1884 struct net_device *orig_dev)
1886 atomic_inc(&skb->users);
1887 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1890 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1891 /* These hooks defined here for ATM */
1893 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1894 unsigned char *addr);
1895 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1898 * If bridge module is loaded call bridging hook.
1899 * returns NULL if packet was consumed.
1901 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
1902 struct sk_buff *skb) __read_mostly;
1903 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
1904 struct packet_type **pt_prev, int *ret,
1905 struct net_device *orig_dev)
1907 struct net_bridge_port *port;
1909 if (skb->pkt_type == PACKET_LOOPBACK ||
1910 (port = rcu_dereference(skb->dev->br_port)) == NULL)
1914 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1918 return br_handle_frame_hook(port, skb);
1921 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1924 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1925 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
1926 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
1928 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
1929 struct packet_type **pt_prev,
1931 struct net_device *orig_dev)
1933 if (skb->dev->macvlan_port == NULL)
1937 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1940 return macvlan_handle_frame_hook(skb);
1943 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1946 #ifdef CONFIG_NET_CLS_ACT
1947 /* TODO: Maybe we should just force sch_ingress to be compiled in
1948 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1949 * a compare and 2 stores extra right now if we dont have it on
1950 * but have CONFIG_NET_CLS_ACT
1951 * NOTE: This doesnt stop any functionality; if you dont have
1952 * the ingress scheduler, you just cant add policies on ingress.
1955 static int ing_filter(struct sk_buff *skb)
1958 struct net_device *dev = skb->dev;
1959 int result = TC_ACT_OK;
1961 if (dev->qdisc_ingress) {
1962 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1963 if (MAX_RED_LOOP < ttl++) {
1964 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1965 skb->iif, skb->dev->ifindex);
1969 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1971 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1973 spin_lock(&dev->ingress_lock);
1974 if ((q = dev->qdisc_ingress) != NULL)
1975 result = q->enqueue(skb, q);
1976 spin_unlock(&dev->ingress_lock);
1984 int netif_receive_skb(struct sk_buff *skb)
1986 struct packet_type *ptype, *pt_prev;
1987 struct net_device *orig_dev;
1988 int ret = NET_RX_DROP;
1991 /* if we've gotten here through NAPI, check netpoll */
1992 if (netpoll_receive_skb(skb))
1995 if (!skb->tstamp.tv64)
1999 skb->iif = skb->dev->ifindex;
2001 orig_dev = skb_bond(skb);
2006 __get_cpu_var(netdev_rx_stat).total++;
2008 skb_reset_network_header(skb);
2009 skb_reset_transport_header(skb);
2010 skb->mac_len = skb->network_header - skb->mac_header;
2016 #ifdef CONFIG_NET_CLS_ACT
2017 if (skb->tc_verd & TC_NCLS) {
2018 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2023 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2024 if (!ptype->dev || ptype->dev == skb->dev) {
2026 ret = deliver_skb(skb, pt_prev, orig_dev);
2031 #ifdef CONFIG_NET_CLS_ACT
2033 ret = deliver_skb(skb, pt_prev, orig_dev);
2034 pt_prev = NULL; /* noone else should process this after*/
2036 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2039 ret = ing_filter(skb);
2041 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
2050 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2053 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2057 type = skb->protocol;
2058 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
2059 if (ptype->type == type &&
2060 (!ptype->dev || ptype->dev == skb->dev)) {
2062 ret = deliver_skb(skb, pt_prev, orig_dev);
2068 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2071 /* Jamal, now you will not able to escape explaining
2072 * me how you were going to use this. :-)
2082 static int process_backlog(struct napi_struct *napi, int quota)
2085 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2086 unsigned long start_time = jiffies;
2088 napi->weight = weight_p;
2090 struct sk_buff *skb;
2091 struct net_device *dev;
2093 local_irq_disable();
2094 skb = __skb_dequeue(&queue->input_pkt_queue);
2096 __napi_complete(napi);
2105 netif_receive_skb(skb);
2108 } while (++work < quota && jiffies == start_time);
2114 * __napi_schedule - schedule for receive
2115 * @napi: entry to schedule
2117 * The entry's receive function will be scheduled to run
2119 void fastcall __napi_schedule(struct napi_struct *n)
2121 unsigned long flags;
2123 local_irq_save(flags);
2124 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2125 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2126 local_irq_restore(flags);
2128 EXPORT_SYMBOL(__napi_schedule);
2131 static void net_rx_action(struct softirq_action *h)
2133 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2134 unsigned long start_time = jiffies;
2135 int budget = netdev_budget;
2138 local_irq_disable();
2140 while (!list_empty(list)) {
2141 struct napi_struct *n;
2144 /* If softirq window is exhuasted then punt.
2146 * Note that this is a slight policy change from the
2147 * previous NAPI code, which would allow up to 2
2148 * jiffies to pass before breaking out. The test
2149 * used to be "jiffies - start_time > 1".
2151 if (unlikely(budget <= 0 || jiffies != start_time))
2156 /* Even though interrupts have been re-enabled, this
2157 * access is safe because interrupts can only add new
2158 * entries to the tail of this list, and only ->poll()
2159 * calls can remove this head entry from the list.
2161 n = list_entry(list->next, struct napi_struct, poll_list);
2163 have = netpoll_poll_lock(n);
2167 work = n->poll(n, weight);
2169 WARN_ON_ONCE(work > weight);
2173 local_irq_disable();
2175 /* Drivers must not modify the NAPI state if they
2176 * consume the entire weight. In such cases this code
2177 * still "owns" the NAPI instance and therefore can
2178 * move the instance around on the list at-will.
2180 if (unlikely(work == weight))
2181 list_move_tail(&n->poll_list, list);
2183 netpoll_poll_unlock(have);
2188 #ifdef CONFIG_NET_DMA
2190 * There may not be any more sk_buffs coming right now, so push
2191 * any pending DMA copies to hardware
2193 if (!cpus_empty(net_dma.channel_mask)) {
2195 for_each_cpu_mask(chan_idx, net_dma.channel_mask) {
2196 struct dma_chan *chan = net_dma.channels[chan_idx];
2198 dma_async_memcpy_issue_pending(chan);
2206 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2207 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2211 static gifconf_func_t * gifconf_list [NPROTO];
2214 * register_gifconf - register a SIOCGIF handler
2215 * @family: Address family
2216 * @gifconf: Function handler
2218 * Register protocol dependent address dumping routines. The handler
2219 * that is passed must not be freed or reused until it has been replaced
2220 * by another handler.
2222 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2224 if (family >= NPROTO)
2226 gifconf_list[family] = gifconf;
2232 * Map an interface index to its name (SIOCGIFNAME)
2236 * We need this ioctl for efficient implementation of the
2237 * if_indextoname() function required by the IPv6 API. Without
2238 * it, we would have to search all the interfaces to find a
2242 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2244 struct net_device *dev;
2248 * Fetch the caller's info block.
2251 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2254 read_lock(&dev_base_lock);
2255 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2257 read_unlock(&dev_base_lock);
2261 strcpy(ifr.ifr_name, dev->name);
2262 read_unlock(&dev_base_lock);
2264 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2270 * Perform a SIOCGIFCONF call. This structure will change
2271 * size eventually, and there is nothing I can do about it.
2272 * Thus we will need a 'compatibility mode'.
2275 static int dev_ifconf(struct net *net, char __user *arg)
2278 struct net_device *dev;
2285 * Fetch the caller's info block.
2288 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2295 * Loop over the interfaces, and write an info block for each.
2299 for_each_netdev(net, dev) {
2300 for (i = 0; i < NPROTO; i++) {
2301 if (gifconf_list[i]) {
2304 done = gifconf_list[i](dev, NULL, 0);
2306 done = gifconf_list[i](dev, pos + total,
2316 * All done. Write the updated control block back to the caller.
2318 ifc.ifc_len = total;
2321 * Both BSD and Solaris return 0 here, so we do too.
2323 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2326 #ifdef CONFIG_PROC_FS
2328 * This is invoked by the /proc filesystem handler to display a device
2331 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2333 struct net *net = seq->private;
2335 struct net_device *dev;
2337 read_lock(&dev_base_lock);
2339 return SEQ_START_TOKEN;
2342 for_each_netdev(net, dev)
2349 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2351 struct net *net = seq->private;
2353 return v == SEQ_START_TOKEN ?
2354 first_net_device(net) : next_net_device((struct net_device *)v);
2357 void dev_seq_stop(struct seq_file *seq, void *v)
2359 read_unlock(&dev_base_lock);
2362 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2364 struct net_device_stats *stats = dev->get_stats(dev);
2366 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2367 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2368 dev->name, stats->rx_bytes, stats->rx_packets,
2370 stats->rx_dropped + stats->rx_missed_errors,
2371 stats->rx_fifo_errors,
2372 stats->rx_length_errors + stats->rx_over_errors +
2373 stats->rx_crc_errors + stats->rx_frame_errors,
2374 stats->rx_compressed, stats->multicast,
2375 stats->tx_bytes, stats->tx_packets,
2376 stats->tx_errors, stats->tx_dropped,
2377 stats->tx_fifo_errors, stats->collisions,
2378 stats->tx_carrier_errors +
2379 stats->tx_aborted_errors +
2380 stats->tx_window_errors +
2381 stats->tx_heartbeat_errors,
2382 stats->tx_compressed);
2386 * Called from the PROCfs module. This now uses the new arbitrary sized
2387 * /proc/net interface to create /proc/net/dev
2389 static int dev_seq_show(struct seq_file *seq, void *v)
2391 if (v == SEQ_START_TOKEN)
2392 seq_puts(seq, "Inter-| Receive "
2394 " face |bytes packets errs drop fifo frame "
2395 "compressed multicast|bytes packets errs "
2396 "drop fifo colls carrier compressed\n");
2398 dev_seq_printf_stats(seq, v);
2402 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2404 struct netif_rx_stats *rc = NULL;
2406 while (*pos < NR_CPUS)
2407 if (cpu_online(*pos)) {
2408 rc = &per_cpu(netdev_rx_stat, *pos);
2415 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2417 return softnet_get_online(pos);
2420 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2423 return softnet_get_online(pos);
2426 static void softnet_seq_stop(struct seq_file *seq, void *v)
2430 static int softnet_seq_show(struct seq_file *seq, void *v)
2432 struct netif_rx_stats *s = v;
2434 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2435 s->total, s->dropped, s->time_squeeze, 0,
2436 0, 0, 0, 0, /* was fastroute */
2441 static const struct seq_operations dev_seq_ops = {
2442 .start = dev_seq_start,
2443 .next = dev_seq_next,
2444 .stop = dev_seq_stop,
2445 .show = dev_seq_show,
2448 static int dev_seq_open(struct inode *inode, struct file *file)
2450 struct seq_file *seq;
2452 res = seq_open(file, &dev_seq_ops);
2454 seq = file->private_data;
2455 seq->private = get_proc_net(inode);
2456 if (!seq->private) {
2457 seq_release(inode, file);
2464 static int dev_seq_release(struct inode *inode, struct file *file)
2466 struct seq_file *seq = file->private_data;
2467 struct net *net = seq->private;
2469 return seq_release(inode, file);
2472 static const struct file_operations dev_seq_fops = {
2473 .owner = THIS_MODULE,
2474 .open = dev_seq_open,
2476 .llseek = seq_lseek,
2477 .release = dev_seq_release,
2480 static const struct seq_operations softnet_seq_ops = {
2481 .start = softnet_seq_start,
2482 .next = softnet_seq_next,
2483 .stop = softnet_seq_stop,
2484 .show = softnet_seq_show,
2487 static int softnet_seq_open(struct inode *inode, struct file *file)
2489 return seq_open(file, &softnet_seq_ops);
2492 static const struct file_operations softnet_seq_fops = {
2493 .owner = THIS_MODULE,
2494 .open = softnet_seq_open,
2496 .llseek = seq_lseek,
2497 .release = seq_release,
2500 static void *ptype_get_idx(loff_t pos)
2502 struct packet_type *pt = NULL;
2506 list_for_each_entry_rcu(pt, &ptype_all, list) {
2512 for (t = 0; t < 16; t++) {
2513 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2522 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2525 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2528 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2530 struct packet_type *pt;
2531 struct list_head *nxt;
2535 if (v == SEQ_START_TOKEN)
2536 return ptype_get_idx(0);
2539 nxt = pt->list.next;
2540 if (pt->type == htons(ETH_P_ALL)) {
2541 if (nxt != &ptype_all)
2544 nxt = ptype_base[0].next;
2546 hash = ntohs(pt->type) & 15;
2548 while (nxt == &ptype_base[hash]) {
2551 nxt = ptype_base[hash].next;
2554 return list_entry(nxt, struct packet_type, list);
2557 static void ptype_seq_stop(struct seq_file *seq, void *v)
2562 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2564 #ifdef CONFIG_KALLSYMS
2565 unsigned long offset = 0, symsize;
2566 const char *symname;
2570 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2577 modname = delim = "";
2578 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2584 seq_printf(seq, "[%p]", sym);
2587 static int ptype_seq_show(struct seq_file *seq, void *v)
2589 struct packet_type *pt = v;
2591 if (v == SEQ_START_TOKEN)
2592 seq_puts(seq, "Type Device Function\n");
2594 if (pt->type == htons(ETH_P_ALL))
2595 seq_puts(seq, "ALL ");
2597 seq_printf(seq, "%04x", ntohs(pt->type));
2599 seq_printf(seq, " %-8s ",
2600 pt->dev ? pt->dev->name : "");
2601 ptype_seq_decode(seq, pt->func);
2602 seq_putc(seq, '\n');
2608 static const struct seq_operations ptype_seq_ops = {
2609 .start = ptype_seq_start,
2610 .next = ptype_seq_next,
2611 .stop = ptype_seq_stop,
2612 .show = ptype_seq_show,
2615 static int ptype_seq_open(struct inode *inode, struct file *file)
2617 return seq_open(file, &ptype_seq_ops);
2620 static const struct file_operations ptype_seq_fops = {
2621 .owner = THIS_MODULE,
2622 .open = ptype_seq_open,
2624 .llseek = seq_lseek,
2625 .release = seq_release,
2629 static int dev_proc_net_init(struct net *net)
2633 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2635 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2637 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2640 if (wext_proc_init(net))
2646 proc_net_remove(net, "ptype");
2648 proc_net_remove(net, "softnet_stat");
2650 proc_net_remove(net, "dev");
2654 static void dev_proc_net_exit(struct net *net)
2656 wext_proc_exit(net);
2658 proc_net_remove(net, "ptype");
2659 proc_net_remove(net, "softnet_stat");
2660 proc_net_remove(net, "dev");
2663 static struct pernet_operations dev_proc_ops = {
2664 .init = dev_proc_net_init,
2665 .exit = dev_proc_net_exit,
2668 static int __init dev_proc_init(void)
2670 return register_pernet_subsys(&dev_proc_ops);
2673 #define dev_proc_init() 0
2674 #endif /* CONFIG_PROC_FS */
2678 * netdev_set_master - set up master/slave pair
2679 * @slave: slave device
2680 * @master: new master device
2682 * Changes the master device of the slave. Pass %NULL to break the
2683 * bonding. The caller must hold the RTNL semaphore. On a failure
2684 * a negative errno code is returned. On success the reference counts
2685 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2686 * function returns zero.
2688 int netdev_set_master(struct net_device *slave, struct net_device *master)
2690 struct net_device *old = slave->master;
2700 slave->master = master;
2708 slave->flags |= IFF_SLAVE;
2710 slave->flags &= ~IFF_SLAVE;
2712 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2716 static void __dev_set_promiscuity(struct net_device *dev, int inc)
2718 unsigned short old_flags = dev->flags;
2722 if ((dev->promiscuity += inc) == 0)
2723 dev->flags &= ~IFF_PROMISC;
2725 dev->flags |= IFF_PROMISC;
2726 if (dev->flags != old_flags) {
2727 printk(KERN_INFO "device %s %s promiscuous mode\n",
2728 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2730 audit_log(current->audit_context, GFP_ATOMIC,
2731 AUDIT_ANOM_PROMISCUOUS,
2732 "dev=%s prom=%d old_prom=%d auid=%u",
2733 dev->name, (dev->flags & IFF_PROMISC),
2734 (old_flags & IFF_PROMISC),
2735 audit_get_loginuid(current->audit_context));
2737 if (dev->change_rx_flags)
2738 dev->change_rx_flags(dev, IFF_PROMISC);
2743 * dev_set_promiscuity - update promiscuity count on a device
2747 * Add or remove promiscuity from a device. While the count in the device
2748 * remains above zero the interface remains promiscuous. Once it hits zero
2749 * the device reverts back to normal filtering operation. A negative inc
2750 * value is used to drop promiscuity on the device.
2752 void dev_set_promiscuity(struct net_device *dev, int inc)
2754 unsigned short old_flags = dev->flags;
2756 __dev_set_promiscuity(dev, inc);
2757 if (dev->flags != old_flags)
2758 dev_set_rx_mode(dev);
2762 * dev_set_allmulti - update allmulti count on a device
2766 * Add or remove reception of all multicast frames to a device. While the
2767 * count in the device remains above zero the interface remains listening
2768 * to all interfaces. Once it hits zero the device reverts back to normal
2769 * filtering operation. A negative @inc value is used to drop the counter
2770 * when releasing a resource needing all multicasts.
2773 void dev_set_allmulti(struct net_device *dev, int inc)
2775 unsigned short old_flags = dev->flags;
2779 dev->flags |= IFF_ALLMULTI;
2780 if ((dev->allmulti += inc) == 0)
2781 dev->flags &= ~IFF_ALLMULTI;
2782 if (dev->flags ^ old_flags) {
2783 if (dev->change_rx_flags)
2784 dev->change_rx_flags(dev, IFF_ALLMULTI);
2785 dev_set_rx_mode(dev);
2790 * Upload unicast and multicast address lists to device and
2791 * configure RX filtering. When the device doesn't support unicast
2792 * filtering it is put in promiscous mode while unicast addresses
2795 void __dev_set_rx_mode(struct net_device *dev)
2797 /* dev_open will call this function so the list will stay sane. */
2798 if (!(dev->flags&IFF_UP))
2801 if (!netif_device_present(dev))
2804 if (dev->set_rx_mode)
2805 dev->set_rx_mode(dev);
2807 /* Unicast addresses changes may only happen under the rtnl,
2808 * therefore calling __dev_set_promiscuity here is safe.
2810 if (dev->uc_count > 0 && !dev->uc_promisc) {
2811 __dev_set_promiscuity(dev, 1);
2812 dev->uc_promisc = 1;
2813 } else if (dev->uc_count == 0 && dev->uc_promisc) {
2814 __dev_set_promiscuity(dev, -1);
2815 dev->uc_promisc = 0;
2818 if (dev->set_multicast_list)
2819 dev->set_multicast_list(dev);
2823 void dev_set_rx_mode(struct net_device *dev)
2825 netif_tx_lock_bh(dev);
2826 __dev_set_rx_mode(dev);
2827 netif_tx_unlock_bh(dev);
2830 int __dev_addr_delete(struct dev_addr_list **list, int *count,
2831 void *addr, int alen, int glbl)
2833 struct dev_addr_list *da;
2835 for (; (da = *list) != NULL; list = &da->next) {
2836 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2837 alen == da->da_addrlen) {
2839 int old_glbl = da->da_gusers;
2856 int __dev_addr_add(struct dev_addr_list **list, int *count,
2857 void *addr, int alen, int glbl)
2859 struct dev_addr_list *da;
2861 for (da = *list; da != NULL; da = da->next) {
2862 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2863 da->da_addrlen == alen) {
2865 int old_glbl = da->da_gusers;
2875 da = kmalloc(sizeof(*da), GFP_ATOMIC);
2878 memcpy(da->da_addr, addr, alen);
2879 da->da_addrlen = alen;
2881 da->da_gusers = glbl ? 1 : 0;
2889 * dev_unicast_delete - Release secondary unicast address.
2891 * @addr: address to delete
2892 * @alen: length of @addr
2894 * Release reference to a secondary unicast address and remove it
2895 * from the device if the reference count drops to zero.
2897 * The caller must hold the rtnl_mutex.
2899 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
2905 netif_tx_lock_bh(dev);
2906 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2908 __dev_set_rx_mode(dev);
2909 netif_tx_unlock_bh(dev);
2912 EXPORT_SYMBOL(dev_unicast_delete);
2915 * dev_unicast_add - add a secondary unicast address
2917 * @addr: address to delete
2918 * @alen: length of @addr
2920 * Add a secondary unicast address to the device or increase
2921 * the reference count if it already exists.
2923 * The caller must hold the rtnl_mutex.
2925 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
2931 netif_tx_lock_bh(dev);
2932 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2934 __dev_set_rx_mode(dev);
2935 netif_tx_unlock_bh(dev);
2938 EXPORT_SYMBOL(dev_unicast_add);
2940 static void __dev_addr_discard(struct dev_addr_list **list)
2942 struct dev_addr_list *tmp;
2944 while (*list != NULL) {
2947 if (tmp->da_users > tmp->da_gusers)
2948 printk("__dev_addr_discard: address leakage! "
2949 "da_users=%d\n", tmp->da_users);
2954 static void dev_addr_discard(struct net_device *dev)
2956 netif_tx_lock_bh(dev);
2958 __dev_addr_discard(&dev->uc_list);
2961 __dev_addr_discard(&dev->mc_list);
2964 netif_tx_unlock_bh(dev);
2967 unsigned dev_get_flags(const struct net_device *dev)
2971 flags = (dev->flags & ~(IFF_PROMISC |
2976 (dev->gflags & (IFF_PROMISC |
2979 if (netif_running(dev)) {
2980 if (netif_oper_up(dev))
2981 flags |= IFF_RUNNING;
2982 if (netif_carrier_ok(dev))
2983 flags |= IFF_LOWER_UP;
2984 if (netif_dormant(dev))
2985 flags |= IFF_DORMANT;
2991 int dev_change_flags(struct net_device *dev, unsigned flags)
2994 int old_flags = dev->flags;
2999 * Set the flags on our device.
3002 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3003 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3005 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3009 * Load in the correct multicast list now the flags have changed.
3012 if (dev->change_rx_flags && (dev->flags ^ flags) & IFF_MULTICAST)
3013 dev->change_rx_flags(dev, IFF_MULTICAST);
3015 dev_set_rx_mode(dev);
3018 * Have we downed the interface. We handle IFF_UP ourselves
3019 * according to user attempts to set it, rather than blindly
3024 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3025 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3028 dev_set_rx_mode(dev);
3031 if (dev->flags & IFF_UP &&
3032 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3034 raw_notifier_call_chain(&netdev_chain,
3035 NETDEV_CHANGE, dev);
3037 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3038 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3039 dev->gflags ^= IFF_PROMISC;
3040 dev_set_promiscuity(dev, inc);
3043 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3044 is important. Some (broken) drivers set IFF_PROMISC, when
3045 IFF_ALLMULTI is requested not asking us and not reporting.
3047 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3048 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3049 dev->gflags ^= IFF_ALLMULTI;
3050 dev_set_allmulti(dev, inc);
3053 /* Exclude state transition flags, already notified */
3054 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3056 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3061 int dev_set_mtu(struct net_device *dev, int new_mtu)
3065 if (new_mtu == dev->mtu)
3068 /* MTU must be positive. */
3072 if (!netif_device_present(dev))
3076 if (dev->change_mtu)
3077 err = dev->change_mtu(dev, new_mtu);
3080 if (!err && dev->flags & IFF_UP)
3081 raw_notifier_call_chain(&netdev_chain,
3082 NETDEV_CHANGEMTU, dev);
3086 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3090 if (!dev->set_mac_address)
3092 if (sa->sa_family != dev->type)
3094 if (!netif_device_present(dev))
3096 err = dev->set_mac_address(dev, sa);
3098 raw_notifier_call_chain(&netdev_chain,
3099 NETDEV_CHANGEADDR, dev);
3104 * Perform the SIOCxIFxxx calls.
3106 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3109 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3115 case SIOCGIFFLAGS: /* Get interface flags */
3116 ifr->ifr_flags = dev_get_flags(dev);
3119 case SIOCSIFFLAGS: /* Set interface flags */
3120 return dev_change_flags(dev, ifr->ifr_flags);
3122 case SIOCGIFMETRIC: /* Get the metric on the interface
3123 (currently unused) */
3124 ifr->ifr_metric = 0;
3127 case SIOCSIFMETRIC: /* Set the metric on the interface
3128 (currently unused) */
3131 case SIOCGIFMTU: /* Get the MTU of a device */
3132 ifr->ifr_mtu = dev->mtu;
3135 case SIOCSIFMTU: /* Set the MTU of a device */
3136 return dev_set_mtu(dev, ifr->ifr_mtu);
3140 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3142 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3143 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3144 ifr->ifr_hwaddr.sa_family = dev->type;
3148 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3150 case SIOCSIFHWBROADCAST:
3151 if (ifr->ifr_hwaddr.sa_family != dev->type)
3153 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3154 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3155 raw_notifier_call_chain(&netdev_chain,
3156 NETDEV_CHANGEADDR, dev);
3160 ifr->ifr_map.mem_start = dev->mem_start;
3161 ifr->ifr_map.mem_end = dev->mem_end;
3162 ifr->ifr_map.base_addr = dev->base_addr;
3163 ifr->ifr_map.irq = dev->irq;
3164 ifr->ifr_map.dma = dev->dma;
3165 ifr->ifr_map.port = dev->if_port;
3169 if (dev->set_config) {
3170 if (!netif_device_present(dev))
3172 return dev->set_config(dev, &ifr->ifr_map);
3177 if (!dev->set_multicast_list ||
3178 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3180 if (!netif_device_present(dev))
3182 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3186 if (!dev->set_multicast_list ||
3187 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3189 if (!netif_device_present(dev))
3191 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3195 ifr->ifr_ifindex = dev->ifindex;
3199 ifr->ifr_qlen = dev->tx_queue_len;
3203 if (ifr->ifr_qlen < 0)
3205 dev->tx_queue_len = ifr->ifr_qlen;
3209 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3210 return dev_change_name(dev, ifr->ifr_newname);
3213 * Unknown or private ioctl
3217 if ((cmd >= SIOCDEVPRIVATE &&
3218 cmd <= SIOCDEVPRIVATE + 15) ||
3219 cmd == SIOCBONDENSLAVE ||
3220 cmd == SIOCBONDRELEASE ||
3221 cmd == SIOCBONDSETHWADDR ||
3222 cmd == SIOCBONDSLAVEINFOQUERY ||
3223 cmd == SIOCBONDINFOQUERY ||
3224 cmd == SIOCBONDCHANGEACTIVE ||
3225 cmd == SIOCGMIIPHY ||
3226 cmd == SIOCGMIIREG ||
3227 cmd == SIOCSMIIREG ||
3228 cmd == SIOCBRADDIF ||
3229 cmd == SIOCBRDELIF ||
3230 cmd == SIOCWANDEV) {
3232 if (dev->do_ioctl) {
3233 if (netif_device_present(dev))
3234 err = dev->do_ioctl(dev, ifr,
3247 * This function handles all "interface"-type I/O control requests. The actual
3248 * 'doing' part of this is dev_ifsioc above.
3252 * dev_ioctl - network device ioctl
3253 * @cmd: command to issue
3254 * @arg: pointer to a struct ifreq in user space
3256 * Issue ioctl functions to devices. This is normally called by the
3257 * user space syscall interfaces but can sometimes be useful for
3258 * other purposes. The return value is the return from the syscall if
3259 * positive or a negative errno code on error.
3262 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3268 /* One special case: SIOCGIFCONF takes ifconf argument
3269 and requires shared lock, because it sleeps writing
3273 if (cmd == SIOCGIFCONF) {
3275 ret = dev_ifconf(net, (char __user *) arg);
3279 if (cmd == SIOCGIFNAME)
3280 return dev_ifname(net, (struct ifreq __user *)arg);
3282 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3285 ifr.ifr_name[IFNAMSIZ-1] = 0;
3287 colon = strchr(ifr.ifr_name, ':');
3292 * See which interface the caller is talking about.
3297 * These ioctl calls:
3298 * - can be done by all.
3299 * - atomic and do not require locking.
3310 dev_load(net, ifr.ifr_name);
3311 read_lock(&dev_base_lock);
3312 ret = dev_ifsioc(net, &ifr, cmd);
3313 read_unlock(&dev_base_lock);
3317 if (copy_to_user(arg, &ifr,
3318 sizeof(struct ifreq)))
3324 dev_load(net, ifr.ifr_name);
3326 ret = dev_ethtool(net, &ifr);
3331 if (copy_to_user(arg, &ifr,
3332 sizeof(struct ifreq)))
3338 * These ioctl calls:
3339 * - require superuser power.
3340 * - require strict serialization.
3346 if (!capable(CAP_NET_ADMIN))
3348 dev_load(net, ifr.ifr_name);
3350 ret = dev_ifsioc(net, &ifr, cmd);
3355 if (copy_to_user(arg, &ifr,
3356 sizeof(struct ifreq)))
3362 * These ioctl calls:
3363 * - require superuser power.
3364 * - require strict serialization.
3365 * - do not return a value
3375 case SIOCSIFHWBROADCAST:
3378 case SIOCBONDENSLAVE:
3379 case SIOCBONDRELEASE:
3380 case SIOCBONDSETHWADDR:
3381 case SIOCBONDCHANGEACTIVE:
3384 if (!capable(CAP_NET_ADMIN))
3387 case SIOCBONDSLAVEINFOQUERY:
3388 case SIOCBONDINFOQUERY:
3389 dev_load(net, ifr.ifr_name);
3391 ret = dev_ifsioc(net, &ifr, cmd);
3396 /* Get the per device memory space. We can add this but
3397 * currently do not support it */
3399 /* Set the per device memory buffer space.
3400 * Not applicable in our case */
3405 * Unknown or private ioctl.
3408 if (cmd == SIOCWANDEV ||
3409 (cmd >= SIOCDEVPRIVATE &&
3410 cmd <= SIOCDEVPRIVATE + 15)) {
3411 dev_load(net, ifr.ifr_name);
3413 ret = dev_ifsioc(net, &ifr, cmd);
3415 if (!ret && copy_to_user(arg, &ifr,
3416 sizeof(struct ifreq)))
3420 /* Take care of Wireless Extensions */
3421 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3422 return wext_handle_ioctl(net, &ifr, cmd, arg);
3429 * dev_new_index - allocate an ifindex
3431 * Returns a suitable unique value for a new device interface
3432 * number. The caller must hold the rtnl semaphore or the
3433 * dev_base_lock to be sure it remains unique.
3435 static int dev_new_index(struct net *net)
3441 if (!__dev_get_by_index(net, ifindex))
3446 /* Delayed registration/unregisteration */
3447 static DEFINE_SPINLOCK(net_todo_list_lock);
3448 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
3450 static void net_set_todo(struct net_device *dev)
3452 spin_lock(&net_todo_list_lock);
3453 list_add_tail(&dev->todo_list, &net_todo_list);
3454 spin_unlock(&net_todo_list_lock);
3458 * register_netdevice - register a network device
3459 * @dev: device to register
3461 * Take a completed network device structure and add it to the kernel
3462 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3463 * chain. 0 is returned on success. A negative errno code is returned
3464 * on a failure to set up the device, or if the name is a duplicate.
3466 * Callers must hold the rtnl semaphore. You may want
3467 * register_netdev() instead of this.
3470 * The locking appears insufficient to guarantee two parallel registers
3471 * will not get the same name.
3474 int register_netdevice(struct net_device *dev)
3476 struct hlist_head *head;
3477 struct hlist_node *p;
3481 BUG_ON(dev_boot_phase);
3486 /* When net_device's are persistent, this will be fatal. */
3487 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3488 BUG_ON(!dev->nd_net);
3491 spin_lock_init(&dev->queue_lock);
3492 spin_lock_init(&dev->_xmit_lock);
3493 netdev_set_lockdep_class(&dev->_xmit_lock, dev->type);
3494 dev->xmit_lock_owner = -1;
3495 spin_lock_init(&dev->ingress_lock);
3499 /* Init, if this function is available */
3501 ret = dev->init(dev);
3509 if (!dev_valid_name(dev->name)) {
3514 dev->ifindex = dev_new_index(net);
3515 if (dev->iflink == -1)
3516 dev->iflink = dev->ifindex;
3518 /* Check for existence of name */
3519 head = dev_name_hash(net, dev->name);
3520 hlist_for_each(p, head) {
3521 struct net_device *d
3522 = hlist_entry(p, struct net_device, name_hlist);
3523 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3529 /* Fix illegal checksum combinations */
3530 if ((dev->features & NETIF_F_HW_CSUM) &&
3531 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3532 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3534 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3537 if ((dev->features & NETIF_F_NO_CSUM) &&
3538 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3539 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3541 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3545 /* Fix illegal SG+CSUM combinations. */
3546 if ((dev->features & NETIF_F_SG) &&
3547 !(dev->features & NETIF_F_ALL_CSUM)) {
3548 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3550 dev->features &= ~NETIF_F_SG;
3553 /* TSO requires that SG is present as well. */
3554 if ((dev->features & NETIF_F_TSO) &&
3555 !(dev->features & NETIF_F_SG)) {
3556 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3558 dev->features &= ~NETIF_F_TSO;
3560 if (dev->features & NETIF_F_UFO) {
3561 if (!(dev->features & NETIF_F_HW_CSUM)) {
3562 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3563 "NETIF_F_HW_CSUM feature.\n",
3565 dev->features &= ~NETIF_F_UFO;
3567 if (!(dev->features & NETIF_F_SG)) {
3568 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3569 "NETIF_F_SG feature.\n",
3571 dev->features &= ~NETIF_F_UFO;
3576 * nil rebuild_header routine,
3577 * that should be never called and used as just bug trap.
3580 if (!dev->rebuild_header)
3581 dev->rebuild_header = default_rebuild_header;
3583 ret = netdev_register_sysfs(dev);
3586 dev->reg_state = NETREG_REGISTERED;
3589 * Default initial state at registry is that the
3590 * device is present.
3593 set_bit(__LINK_STATE_PRESENT, &dev->state);
3595 dev_init_scheduler(dev);
3597 list_netdevice(dev);
3599 /* Notify protocols, that a new device appeared. */
3600 ret = raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3601 ret = notifier_to_errno(ret);
3603 unregister_netdevice(dev);
3615 * register_netdev - register a network device
3616 * @dev: device to register
3618 * Take a completed network device structure and add it to the kernel
3619 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3620 * chain. 0 is returned on success. A negative errno code is returned
3621 * on a failure to set up the device, or if the name is a duplicate.
3623 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3624 * and expands the device name if you passed a format string to
3627 int register_netdev(struct net_device *dev)
3634 * If the name is a format string the caller wants us to do a
3637 if (strchr(dev->name, '%')) {
3638 err = dev_alloc_name(dev, dev->name);
3643 err = register_netdevice(dev);
3648 EXPORT_SYMBOL(register_netdev);
3651 * netdev_wait_allrefs - wait until all references are gone.
3653 * This is called when unregistering network devices.
3655 * Any protocol or device that holds a reference should register
3656 * for netdevice notification, and cleanup and put back the
3657 * reference if they receive an UNREGISTER event.
3658 * We can get stuck here if buggy protocols don't correctly
3661 static void netdev_wait_allrefs(struct net_device *dev)
3663 unsigned long rebroadcast_time, warning_time;
3665 rebroadcast_time = warning_time = jiffies;
3666 while (atomic_read(&dev->refcnt) != 0) {
3667 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3670 /* Rebroadcast unregister notification */
3671 raw_notifier_call_chain(&netdev_chain,
3672 NETDEV_UNREGISTER, dev);
3674 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3676 /* We must not have linkwatch events
3677 * pending on unregister. If this
3678 * happens, we simply run the queue
3679 * unscheduled, resulting in a noop
3682 linkwatch_run_queue();
3687 rebroadcast_time = jiffies;
3692 if (time_after(jiffies, warning_time + 10 * HZ)) {
3693 printk(KERN_EMERG "unregister_netdevice: "
3694 "waiting for %s to become free. Usage "
3696 dev->name, atomic_read(&dev->refcnt));
3697 warning_time = jiffies;
3706 * register_netdevice(x1);
3707 * register_netdevice(x2);
3709 * unregister_netdevice(y1);
3710 * unregister_netdevice(y2);
3716 * We are invoked by rtnl_unlock() after it drops the semaphore.
3717 * This allows us to deal with problems:
3718 * 1) We can delete sysfs objects which invoke hotplug
3719 * without deadlocking with linkwatch via keventd.
3720 * 2) Since we run with the RTNL semaphore not held, we can sleep
3721 * safely in order to wait for the netdev refcnt to drop to zero.
3723 static DEFINE_MUTEX(net_todo_run_mutex);
3724 void netdev_run_todo(void)
3726 struct list_head list;
3728 /* Need to guard against multiple cpu's getting out of order. */
3729 mutex_lock(&net_todo_run_mutex);
3731 /* Not safe to do outside the semaphore. We must not return
3732 * until all unregister events invoked by the local processor
3733 * have been completed (either by this todo run, or one on
3736 if (list_empty(&net_todo_list))
3739 /* Snapshot list, allow later requests */
3740 spin_lock(&net_todo_list_lock);
3741 list_replace_init(&net_todo_list, &list);
3742 spin_unlock(&net_todo_list_lock);
3744 while (!list_empty(&list)) {
3745 struct net_device *dev
3746 = list_entry(list.next, struct net_device, todo_list);
3747 list_del(&dev->todo_list);
3749 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3750 printk(KERN_ERR "network todo '%s' but state %d\n",
3751 dev->name, dev->reg_state);
3756 dev->reg_state = NETREG_UNREGISTERED;
3758 netdev_wait_allrefs(dev);
3761 BUG_ON(atomic_read(&dev->refcnt));
3762 BUG_TRAP(!dev->ip_ptr);
3763 BUG_TRAP(!dev->ip6_ptr);
3764 BUG_TRAP(!dev->dn_ptr);
3766 if (dev->destructor)
3767 dev->destructor(dev);
3769 /* Free network device */
3770 kobject_put(&dev->dev.kobj);
3774 mutex_unlock(&net_todo_run_mutex);
3777 static struct net_device_stats *internal_stats(struct net_device *dev)
3783 * alloc_netdev_mq - allocate network device
3784 * @sizeof_priv: size of private data to allocate space for
3785 * @name: device name format string
3786 * @setup: callback to initialize device
3787 * @queue_count: the number of subqueues to allocate
3789 * Allocates a struct net_device with private data area for driver use
3790 * and performs basic initialization. Also allocates subquue structs
3791 * for each queue on the device at the end of the netdevice.
3793 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
3794 void (*setup)(struct net_device *), unsigned int queue_count)
3797 struct net_device *dev;
3800 BUG_ON(strlen(name) >= sizeof(dev->name));
3802 /* ensure 32-byte alignment of both the device and private area */
3803 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST +
3804 (sizeof(struct net_device_subqueue) * (queue_count - 1))) &
3805 ~NETDEV_ALIGN_CONST;
3806 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3808 p = kzalloc(alloc_size, GFP_KERNEL);
3810 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
3814 dev = (struct net_device *)
3815 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3816 dev->padded = (char *)dev - (char *)p;
3817 dev->nd_net = &init_net;
3820 dev->priv = ((char *)dev +
3821 ((sizeof(struct net_device) +
3822 (sizeof(struct net_device_subqueue) *
3823 (queue_count - 1)) + NETDEV_ALIGN_CONST)
3824 & ~NETDEV_ALIGN_CONST));
3827 dev->egress_subqueue_count = queue_count;
3829 dev->get_stats = internal_stats;
3830 netpoll_netdev_init(dev);
3832 strcpy(dev->name, name);
3835 EXPORT_SYMBOL(alloc_netdev_mq);
3838 * free_netdev - free network device
3841 * This function does the last stage of destroying an allocated device
3842 * interface. The reference to the device object is released.
3843 * If this is the last reference then it will be freed.
3845 void free_netdev(struct net_device *dev)
3848 /* Compatibility with error handling in drivers */
3849 if (dev->reg_state == NETREG_UNINITIALIZED) {
3850 kfree((char *)dev - dev->padded);
3854 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3855 dev->reg_state = NETREG_RELEASED;
3857 /* will free via device release */
3858 put_device(&dev->dev);
3860 kfree((char *)dev - dev->padded);
3864 /* Synchronize with packet receive processing. */
3865 void synchronize_net(void)
3872 * unregister_netdevice - remove device from the kernel
3875 * This function shuts down a device interface and removes it
3876 * from the kernel tables. On success 0 is returned, on a failure
3877 * a negative errno code is returned.
3879 * Callers must hold the rtnl semaphore. You may want
3880 * unregister_netdev() instead of this.
3883 void unregister_netdevice(struct net_device *dev)
3885 BUG_ON(dev_boot_phase);
3888 /* Some devices call without registering for initialization unwind. */
3889 if (dev->reg_state == NETREG_UNINITIALIZED) {
3890 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3891 "was registered\n", dev->name, dev);
3897 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3899 /* If device is running, close it first. */
3900 if (dev->flags & IFF_UP)
3903 /* And unlink it from device chain. */
3904 unlist_netdevice(dev);
3906 dev->reg_state = NETREG_UNREGISTERING;
3910 /* Shutdown queueing discipline. */
3914 /* Notify protocols, that we are about to destroy
3915 this device. They should clean all the things.
3917 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3920 * Flush the unicast and multicast chains
3922 dev_addr_discard(dev);
3927 /* Notifier chain MUST detach us from master device. */
3928 BUG_TRAP(!dev->master);
3930 /* Remove entries from sysfs */
3931 netdev_unregister_sysfs(dev);
3933 /* Finish processing unregister after unlock */
3942 * unregister_netdev - remove device from the kernel
3945 * This function shuts down a device interface and removes it
3946 * from the kernel tables. On success 0 is returned, on a failure
3947 * a negative errno code is returned.
3949 * This is just a wrapper for unregister_netdevice that takes
3950 * the rtnl semaphore. In general you want to use this and not
3951 * unregister_netdevice.
3953 void unregister_netdev(struct net_device *dev)
3956 unregister_netdevice(dev);
3960 EXPORT_SYMBOL(unregister_netdev);
3963 * dev_change_net_namespace - move device to different nethost namespace
3965 * @net: network namespace
3966 * @pat: If not NULL name pattern to try if the current device name
3967 * is already taken in the destination network namespace.
3969 * This function shuts down a device interface and moves it
3970 * to a new network namespace. On success 0 is returned, on
3971 * a failure a netagive errno code is returned.
3973 * Callers must hold the rtnl semaphore.
3976 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
3979 const char *destname;
3984 /* Don't allow namespace local devices to be moved. */
3986 if (dev->features & NETIF_F_NETNS_LOCAL)
3989 /* Ensure the device has been registrered */
3991 if (dev->reg_state != NETREG_REGISTERED)
3994 /* Get out if there is nothing todo */
3996 if (dev->nd_net == net)
3999 /* Pick the destination device name, and ensure
4000 * we can use it in the destination network namespace.
4003 destname = dev->name;
4004 if (__dev_get_by_name(net, destname)) {
4005 /* We get here if we can't use the current device name */
4008 if (!dev_valid_name(pat))
4010 if (strchr(pat, '%')) {
4011 if (__dev_alloc_name(net, pat, buf) < 0)
4016 if (__dev_get_by_name(net, destname))
4021 * And now a mini version of register_netdevice unregister_netdevice.
4024 /* If device is running close it first. */
4025 if (dev->flags & IFF_UP)
4028 /* And unlink it from device chain */
4030 unlist_netdevice(dev);
4034 /* Shutdown queueing discipline. */
4037 /* Notify protocols, that we are about to destroy
4038 this device. They should clean all the things.
4040 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4043 * Flush the unicast and multicast chains
4045 dev_addr_discard(dev);
4047 /* Actually switch the network namespace */
4050 /* Assign the new device name */
4051 if (destname != dev->name)
4052 strcpy(dev->name, destname);
4054 /* If there is an ifindex conflict assign a new one */
4055 if (__dev_get_by_index(net, dev->ifindex)) {
4056 int iflink = (dev->iflink == dev->ifindex);
4057 dev->ifindex = dev_new_index(net);
4059 dev->iflink = dev->ifindex;
4063 err = device_rename(&dev->dev, dev->name);
4066 /* Add the device back in the hashes */
4067 list_netdevice(dev);
4069 /* Notify protocols, that a new device appeared. */
4070 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4078 static int dev_cpu_callback(struct notifier_block *nfb,
4079 unsigned long action,
4082 struct sk_buff **list_skb;
4083 struct net_device **list_net;
4084 struct sk_buff *skb;
4085 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4086 struct softnet_data *sd, *oldsd;
4088 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4091 local_irq_disable();
4092 cpu = smp_processor_id();
4093 sd = &per_cpu(softnet_data, cpu);
4094 oldsd = &per_cpu(softnet_data, oldcpu);
4096 /* Find end of our completion_queue. */
4097 list_skb = &sd->completion_queue;
4099 list_skb = &(*list_skb)->next;
4100 /* Append completion queue from offline CPU. */
4101 *list_skb = oldsd->completion_queue;
4102 oldsd->completion_queue = NULL;
4104 /* Find end of our output_queue. */
4105 list_net = &sd->output_queue;
4107 list_net = &(*list_net)->next_sched;
4108 /* Append output queue from offline CPU. */
4109 *list_net = oldsd->output_queue;
4110 oldsd->output_queue = NULL;
4112 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4115 /* Process offline CPU's input_pkt_queue */
4116 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4122 #ifdef CONFIG_NET_DMA
4124 * net_dma_rebalance - try to maintain one DMA channel per CPU
4125 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4127 * This is called when the number of channels allocated to the net_dma client
4128 * changes. The net_dma client tries to have one DMA channel per CPU.
4131 static void net_dma_rebalance(struct net_dma *net_dma)
4133 unsigned int cpu, i, n, chan_idx;
4134 struct dma_chan *chan;
4136 if (cpus_empty(net_dma->channel_mask)) {
4137 for_each_online_cpu(cpu)
4138 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4143 cpu = first_cpu(cpu_online_map);
4145 for_each_cpu_mask(chan_idx, net_dma->channel_mask) {
4146 chan = net_dma->channels[chan_idx];
4148 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4149 + (i < (num_online_cpus() %
4150 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4153 per_cpu(softnet_data, cpu).net_dma = chan;
4154 cpu = next_cpu(cpu, cpu_online_map);
4162 * netdev_dma_event - event callback for the net_dma_client
4163 * @client: should always be net_dma_client
4164 * @chan: DMA channel for the event
4165 * @state: DMA state to be handled
4167 static enum dma_state_client
4168 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4169 enum dma_state state)
4171 int i, found = 0, pos = -1;
4172 struct net_dma *net_dma =
4173 container_of(client, struct net_dma, client);
4174 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4176 spin_lock(&net_dma->lock);
4178 case DMA_RESOURCE_AVAILABLE:
4179 for (i = 0; i < NR_CPUS; i++)
4180 if (net_dma->channels[i] == chan) {
4183 } else if (net_dma->channels[i] == NULL && pos < 0)
4186 if (!found && pos >= 0) {
4188 net_dma->channels[pos] = chan;
4189 cpu_set(pos, net_dma->channel_mask);
4190 net_dma_rebalance(net_dma);
4193 case DMA_RESOURCE_REMOVED:
4194 for (i = 0; i < NR_CPUS; i++)
4195 if (net_dma->channels[i] == chan) {
4203 cpu_clear(pos, net_dma->channel_mask);
4204 net_dma->channels[i] = NULL;
4205 net_dma_rebalance(net_dma);
4211 spin_unlock(&net_dma->lock);
4217 * netdev_dma_regiser - register the networking subsystem as a DMA client
4219 static int __init netdev_dma_register(void)
4221 spin_lock_init(&net_dma.lock);
4222 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4223 dma_async_client_register(&net_dma.client);
4224 dma_async_client_chan_request(&net_dma.client);
4229 static int __init netdev_dma_register(void) { return -ENODEV; }
4230 #endif /* CONFIG_NET_DMA */
4233 * netdev_compute_feature - compute conjunction of two feature sets
4234 * @all: first feature set
4235 * @one: second feature set
4237 * Computes a new feature set after adding a device with feature set
4238 * @one to the master device with current feature set @all. Returns
4239 * the new feature set.
4241 int netdev_compute_features(unsigned long all, unsigned long one)
4243 /* if device needs checksumming, downgrade to hw checksumming */
4244 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4245 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4247 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4248 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4249 all ^= NETIF_F_HW_CSUM
4250 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4252 if (one & NETIF_F_GSO)
4253 one |= NETIF_F_GSO_SOFTWARE;
4256 /* If even one device supports robust GSO, enable it for all. */
4257 if (one & NETIF_F_GSO_ROBUST)
4258 all |= NETIF_F_GSO_ROBUST;
4260 all &= one | NETIF_F_LLTX;
4262 if (!(all & NETIF_F_ALL_CSUM))
4264 if (!(all & NETIF_F_SG))
4265 all &= ~NETIF_F_GSO_MASK;
4269 EXPORT_SYMBOL(netdev_compute_features);
4271 static struct hlist_head *netdev_create_hash(void)
4274 struct hlist_head *hash;
4276 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4278 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4279 INIT_HLIST_HEAD(&hash[i]);
4284 /* Initialize per network namespace state */
4285 static int netdev_init(struct net *net)
4287 INIT_LIST_HEAD(&net->dev_base_head);
4288 rwlock_init(&dev_base_lock);
4290 net->dev_name_head = netdev_create_hash();
4291 if (net->dev_name_head == NULL)
4294 net->dev_index_head = netdev_create_hash();
4295 if (net->dev_index_head == NULL)
4301 kfree(net->dev_name_head);
4306 static void netdev_exit(struct net *net)
4308 kfree(net->dev_name_head);
4309 kfree(net->dev_index_head);
4312 static struct pernet_operations netdev_net_ops = {
4313 .init = netdev_init,
4314 .exit = netdev_exit,
4317 static void default_device_exit(struct net *net)
4319 struct net_device *dev, *next;
4321 * Push all migratable of the network devices back to the
4322 * initial network namespace
4325 for_each_netdev_safe(net, dev, next) {
4328 /* Ignore unmoveable devices (i.e. loopback) */
4329 if (dev->features & NETIF_F_NETNS_LOCAL)
4332 /* Push remaing network devices to init_net */
4333 err = dev_change_net_namespace(dev, &init_net, "dev%d");
4335 printk(KERN_WARNING "%s: failed to move %s to init_net: %d\n",
4336 __func__, dev->name, err);
4337 unregister_netdevice(dev);
4343 static struct pernet_operations default_device_ops = {
4344 .exit = default_device_exit,
4348 * Initialize the DEV module. At boot time this walks the device list and
4349 * unhooks any devices that fail to initialise (normally hardware not
4350 * present) and leaves us with a valid list of present and active devices.
4355 * This is called single threaded during boot, so no need
4356 * to take the rtnl semaphore.
4358 static int __init net_dev_init(void)
4360 int i, rc = -ENOMEM;
4362 BUG_ON(!dev_boot_phase);
4364 if (dev_proc_init())
4367 if (netdev_sysfs_init())
4370 INIT_LIST_HEAD(&ptype_all);
4371 for (i = 0; i < 16; i++)
4372 INIT_LIST_HEAD(&ptype_base[i]);
4374 if (register_pernet_subsys(&netdev_net_ops))
4377 if (register_pernet_device(&default_device_ops))
4381 * Initialise the packet receive queues.
4384 for_each_possible_cpu(i) {
4385 struct softnet_data *queue;
4387 queue = &per_cpu(softnet_data, i);
4388 skb_queue_head_init(&queue->input_pkt_queue);
4389 queue->completion_queue = NULL;
4390 INIT_LIST_HEAD(&queue->poll_list);
4392 queue->backlog.poll = process_backlog;
4393 queue->backlog.weight = weight_p;
4396 netdev_dma_register();
4400 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
4401 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
4403 hotcpu_notifier(dev_cpu_callback, 0);
4411 subsys_initcall(net_dev_init);
4413 EXPORT_SYMBOL(__dev_get_by_index);
4414 EXPORT_SYMBOL(__dev_get_by_name);
4415 EXPORT_SYMBOL(__dev_remove_pack);
4416 EXPORT_SYMBOL(dev_valid_name);
4417 EXPORT_SYMBOL(dev_add_pack);
4418 EXPORT_SYMBOL(dev_alloc_name);
4419 EXPORT_SYMBOL(dev_close);
4420 EXPORT_SYMBOL(dev_get_by_flags);
4421 EXPORT_SYMBOL(dev_get_by_index);
4422 EXPORT_SYMBOL(dev_get_by_name);
4423 EXPORT_SYMBOL(dev_open);
4424 EXPORT_SYMBOL(dev_queue_xmit);
4425 EXPORT_SYMBOL(dev_remove_pack);
4426 EXPORT_SYMBOL(dev_set_allmulti);
4427 EXPORT_SYMBOL(dev_set_promiscuity);
4428 EXPORT_SYMBOL(dev_change_flags);
4429 EXPORT_SYMBOL(dev_set_mtu);
4430 EXPORT_SYMBOL(dev_set_mac_address);
4431 EXPORT_SYMBOL(free_netdev);
4432 EXPORT_SYMBOL(netdev_boot_setup_check);
4433 EXPORT_SYMBOL(netdev_set_master);
4434 EXPORT_SYMBOL(netdev_state_change);
4435 EXPORT_SYMBOL(netif_receive_skb);
4436 EXPORT_SYMBOL(netif_rx);
4437 EXPORT_SYMBOL(register_gifconf);
4438 EXPORT_SYMBOL(register_netdevice);
4439 EXPORT_SYMBOL(register_netdevice_notifier);
4440 EXPORT_SYMBOL(skb_checksum_help);
4441 EXPORT_SYMBOL(synchronize_net);
4442 EXPORT_SYMBOL(unregister_netdevice);
4443 EXPORT_SYMBOL(unregister_netdevice_notifier);
4444 EXPORT_SYMBOL(net_enable_timestamp);
4445 EXPORT_SYMBOL(net_disable_timestamp);
4446 EXPORT_SYMBOL(dev_get_flags);
4448 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4449 EXPORT_SYMBOL(br_handle_frame_hook);
4450 EXPORT_SYMBOL(br_fdb_get_hook);
4451 EXPORT_SYMBOL(br_fdb_put_hook);
4455 EXPORT_SYMBOL(dev_load);
4458 EXPORT_PER_CPU_SYMBOL(softnet_data);