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>
96 #include <linux/rtnetlink.h>
97 #include <linux/proc_fs.h>
98 #include <linux/seq_file.h>
99 #include <linux/stat.h>
100 #include <linux/if_bridge.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <linux/highmem.h>
105 #include <linux/init.h>
106 #include <linux/kmod.h>
107 #include <linux/module.h>
108 #include <linux/kallsyms.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <net/wext.h>
113 #include <net/iw_handler.h>
114 #include <asm/current.h>
115 #include <linux/audit.h>
116 #include <linux/dmaengine.h>
117 #include <linux/err.h>
118 #include <linux/ctype.h>
119 #include <linux/if_arp.h>
122 * The list of packet types we will receive (as opposed to discard)
123 * and the routines to invoke.
125 * Why 16. Because with 16 the only overlap we get on a hash of the
126 * low nibble of the protocol value is RARP/SNAP/X.25.
128 * NOTE: That is no longer true with the addition of VLAN tags. Not
129 * sure which should go first, but I bet it won't make much
130 * difference if we are running VLANs. The good news is that
131 * this protocol won't be in the list unless compiled in, so
132 * the average user (w/out VLANs) will not be adversely affected.
149 static DEFINE_SPINLOCK(ptype_lock);
150 static struct list_head ptype_base[16] __read_mostly; /* 16 way hashed list */
151 static struct list_head ptype_all __read_mostly; /* Taps */
153 #ifdef CONFIG_NET_DMA
154 static struct dma_client *net_dma_client;
155 static unsigned int net_dma_count;
156 static spinlock_t net_dma_event_lock;
160 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
163 * Pure readers hold dev_base_lock for reading.
165 * Writers must hold the rtnl semaphore while they loop through the
166 * dev_base_head list, and hold dev_base_lock for writing when they do the
167 * actual updates. This allows pure readers to access the list even
168 * while a writer is preparing to update it.
170 * To put it another way, dev_base_lock is held for writing only to
171 * protect against pure readers; the rtnl semaphore provides the
172 * protection against other writers.
174 * See, for example usages, register_netdevice() and
175 * unregister_netdevice(), which must be called with the rtnl
178 LIST_HEAD(dev_base_head);
179 DEFINE_RWLOCK(dev_base_lock);
181 EXPORT_SYMBOL(dev_base_head);
182 EXPORT_SYMBOL(dev_base_lock);
184 #define NETDEV_HASHBITS 8
185 static struct hlist_head dev_name_head[1<<NETDEV_HASHBITS];
186 static struct hlist_head dev_index_head[1<<NETDEV_HASHBITS];
188 static inline struct hlist_head *dev_name_hash(const char *name)
190 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
191 return &dev_name_head[hash & ((1<<NETDEV_HASHBITS)-1)];
194 static inline struct hlist_head *dev_index_hash(int ifindex)
196 return &dev_index_head[ifindex & ((1<<NETDEV_HASHBITS)-1)];
203 static RAW_NOTIFIER_HEAD(netdev_chain);
206 * Device drivers call our routines to queue packets here. We empty the
207 * queue in the local softnet handler.
209 DEFINE_PER_CPU(struct softnet_data, softnet_data) = { NULL };
212 extern int netdev_sysfs_init(void);
213 extern int netdev_register_sysfs(struct net_device *);
214 extern void netdev_unregister_sysfs(struct net_device *);
216 #define netdev_sysfs_init() (0)
217 #define netdev_register_sysfs(dev) (0)
218 #define netdev_unregister_sysfs(dev) do { } while(0)
221 #ifdef CONFIG_DEBUG_LOCK_ALLOC
223 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
224 * according to dev->type
226 static const unsigned short netdev_lock_type[] =
227 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
228 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
229 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
230 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
231 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
232 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
233 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
234 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
235 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
236 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
237 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
238 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
239 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
240 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
243 static const char *netdev_lock_name[] =
244 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
245 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
246 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
247 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
248 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
249 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
250 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
251 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
252 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
253 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
254 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
255 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
256 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
257 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
260 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
262 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
266 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
267 if (netdev_lock_type[i] == dev_type)
269 /* the last key is used by default */
270 return ARRAY_SIZE(netdev_lock_type) - 1;
273 static inline void netdev_set_lockdep_class(spinlock_t *lock,
274 unsigned short dev_type)
278 i = netdev_lock_pos(dev_type);
279 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
280 netdev_lock_name[i]);
283 static inline void netdev_set_lockdep_class(spinlock_t *lock,
284 unsigned short dev_type)
289 /*******************************************************************************
291 Protocol management and registration routines
293 *******************************************************************************/
296 * Add a protocol ID to the list. Now that the input handler is
297 * smarter we can dispense with all the messy stuff that used to be
300 * BEWARE!!! Protocol handlers, mangling input packets,
301 * MUST BE last in hash buckets and checking protocol handlers
302 * MUST start from promiscuous ptype_all chain in net_bh.
303 * It is true now, do not change it.
304 * Explanation follows: if protocol handler, mangling packet, will
305 * be the first on list, it is not able to sense, that packet
306 * is cloned and should be copied-on-write, so that it will
307 * change it and subsequent readers will get broken packet.
312 * dev_add_pack - add packet handler
313 * @pt: packet type declaration
315 * Add a protocol handler to the networking stack. The passed &packet_type
316 * is linked into kernel lists and may not be freed until it has been
317 * removed from the kernel lists.
319 * This call does not sleep therefore it can not
320 * guarantee all CPU's that are in middle of receiving packets
321 * will see the new packet type (until the next received packet).
324 void dev_add_pack(struct packet_type *pt)
328 spin_lock_bh(&ptype_lock);
329 if (pt->type == htons(ETH_P_ALL))
330 list_add_rcu(&pt->list, &ptype_all);
332 hash = ntohs(pt->type) & 15;
333 list_add_rcu(&pt->list, &ptype_base[hash]);
335 spin_unlock_bh(&ptype_lock);
339 * __dev_remove_pack - remove packet handler
340 * @pt: packet type declaration
342 * Remove a protocol handler that was previously added to the kernel
343 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
344 * from the kernel lists and can be freed or reused once this function
347 * The packet type might still be in use by receivers
348 * and must not be freed until after all the CPU's have gone
349 * through a quiescent state.
351 void __dev_remove_pack(struct packet_type *pt)
353 struct list_head *head;
354 struct packet_type *pt1;
356 spin_lock_bh(&ptype_lock);
358 if (pt->type == htons(ETH_P_ALL))
361 head = &ptype_base[ntohs(pt->type) & 15];
363 list_for_each_entry(pt1, head, list) {
365 list_del_rcu(&pt->list);
370 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
372 spin_unlock_bh(&ptype_lock);
375 * dev_remove_pack - remove packet handler
376 * @pt: packet type declaration
378 * Remove a protocol handler that was previously added to the kernel
379 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
380 * from the kernel lists and can be freed or reused once this function
383 * This call sleeps to guarantee that no CPU is looking at the packet
386 void dev_remove_pack(struct packet_type *pt)
388 __dev_remove_pack(pt);
393 /******************************************************************************
395 Device Boot-time Settings Routines
397 *******************************************************************************/
399 /* Boot time configuration table */
400 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
403 * netdev_boot_setup_add - add new setup entry
404 * @name: name of the device
405 * @map: configured settings for the device
407 * Adds new setup entry to the dev_boot_setup list. The function
408 * returns 0 on error and 1 on success. This is a generic routine to
411 static int netdev_boot_setup_add(char *name, struct ifmap *map)
413 struct netdev_boot_setup *s;
417 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
418 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
419 memset(s[i].name, 0, sizeof(s[i].name));
420 strcpy(s[i].name, name);
421 memcpy(&s[i].map, map, sizeof(s[i].map));
426 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
430 * netdev_boot_setup_check - check boot time settings
431 * @dev: the netdevice
433 * Check boot time settings for the device.
434 * The found settings are set for the device to be used
435 * later in the device probing.
436 * Returns 0 if no settings found, 1 if they are.
438 int netdev_boot_setup_check(struct net_device *dev)
440 struct netdev_boot_setup *s = dev_boot_setup;
443 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
444 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
445 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
446 dev->irq = s[i].map.irq;
447 dev->base_addr = s[i].map.base_addr;
448 dev->mem_start = s[i].map.mem_start;
449 dev->mem_end = s[i].map.mem_end;
458 * netdev_boot_base - get address from boot time settings
459 * @prefix: prefix for network device
460 * @unit: id for network device
462 * Check boot time settings for the base address of device.
463 * The found settings are set for the device to be used
464 * later in the device probing.
465 * Returns 0 if no settings found.
467 unsigned long netdev_boot_base(const char *prefix, int unit)
469 const struct netdev_boot_setup *s = dev_boot_setup;
473 sprintf(name, "%s%d", prefix, unit);
476 * If device already registered then return base of 1
477 * to indicate not to probe for this interface
479 if (__dev_get_by_name(name))
482 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
483 if (!strcmp(name, s[i].name))
484 return s[i].map.base_addr;
489 * Saves at boot time configured settings for any netdevice.
491 int __init netdev_boot_setup(char *str)
496 str = get_options(str, ARRAY_SIZE(ints), ints);
501 memset(&map, 0, sizeof(map));
505 map.base_addr = ints[2];
507 map.mem_start = ints[3];
509 map.mem_end = ints[4];
511 /* Add new entry to the list */
512 return netdev_boot_setup_add(str, &map);
515 __setup("netdev=", netdev_boot_setup);
517 /*******************************************************************************
519 Device Interface Subroutines
521 *******************************************************************************/
524 * __dev_get_by_name - find a device by its name
525 * @name: name to find
527 * Find an interface by name. Must be called under RTNL semaphore
528 * or @dev_base_lock. If the name is found a pointer to the device
529 * is returned. If the name is not found then %NULL is returned. The
530 * reference counters are not incremented so the caller must be
531 * careful with locks.
534 struct net_device *__dev_get_by_name(const char *name)
536 struct hlist_node *p;
538 hlist_for_each(p, dev_name_hash(name)) {
539 struct net_device *dev
540 = hlist_entry(p, struct net_device, name_hlist);
541 if (!strncmp(dev->name, name, IFNAMSIZ))
548 * dev_get_by_name - find a device by its name
549 * @name: name to find
551 * Find an interface by name. This can be called from any
552 * context and does its own locking. The returned handle has
553 * the usage count incremented and the caller must use dev_put() to
554 * release it when it is no longer needed. %NULL is returned if no
555 * matching device is found.
558 struct net_device *dev_get_by_name(const char *name)
560 struct net_device *dev;
562 read_lock(&dev_base_lock);
563 dev = __dev_get_by_name(name);
566 read_unlock(&dev_base_lock);
571 * __dev_get_by_index - find a device by its ifindex
572 * @ifindex: index of device
574 * Search for an interface by index. Returns %NULL if the device
575 * is not found or a pointer to the device. The device has not
576 * had its reference counter increased so the caller must be careful
577 * about locking. The caller must hold either the RTNL semaphore
581 struct net_device *__dev_get_by_index(int ifindex)
583 struct hlist_node *p;
585 hlist_for_each(p, dev_index_hash(ifindex)) {
586 struct net_device *dev
587 = hlist_entry(p, struct net_device, index_hlist);
588 if (dev->ifindex == ifindex)
596 * dev_get_by_index - find a device by its ifindex
597 * @ifindex: index of device
599 * Search for an interface by index. Returns NULL if the device
600 * is not found or a pointer to the device. The device returned has
601 * had a reference added and the pointer is safe until the user calls
602 * dev_put to indicate they have finished with it.
605 struct net_device *dev_get_by_index(int ifindex)
607 struct net_device *dev;
609 read_lock(&dev_base_lock);
610 dev = __dev_get_by_index(ifindex);
613 read_unlock(&dev_base_lock);
618 * dev_getbyhwaddr - find a device by its hardware address
619 * @type: media type of device
620 * @ha: hardware address
622 * Search for an interface by MAC address. Returns NULL if the device
623 * is not found or a pointer to the device. The caller must hold the
624 * rtnl semaphore. The returned device has not had its ref count increased
625 * and the caller must therefore be careful about locking
628 * If the API was consistent this would be __dev_get_by_hwaddr
631 struct net_device *dev_getbyhwaddr(unsigned short type, char *ha)
633 struct net_device *dev;
638 if (dev->type == type &&
639 !memcmp(dev->dev_addr, ha, dev->addr_len))
645 EXPORT_SYMBOL(dev_getbyhwaddr);
647 struct net_device *__dev_getfirstbyhwtype(unsigned short type)
649 struct net_device *dev;
653 if (dev->type == type)
659 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
661 struct net_device *dev_getfirstbyhwtype(unsigned short type)
663 struct net_device *dev;
666 dev = __dev_getfirstbyhwtype(type);
673 EXPORT_SYMBOL(dev_getfirstbyhwtype);
676 * dev_get_by_flags - find any device with given flags
677 * @if_flags: IFF_* values
678 * @mask: bitmask of bits in if_flags to check
680 * Search for any interface with the given flags. Returns NULL if a device
681 * is not found or a pointer to the device. The device returned has
682 * had a reference added and the pointer is safe until the user calls
683 * dev_put to indicate they have finished with it.
686 struct net_device * dev_get_by_flags(unsigned short if_flags, unsigned short mask)
688 struct net_device *dev, *ret;
691 read_lock(&dev_base_lock);
692 for_each_netdev(dev) {
693 if (((dev->flags ^ if_flags) & mask) == 0) {
699 read_unlock(&dev_base_lock);
704 * dev_valid_name - check if name is okay for network device
707 * Network device names need to be valid file names to
708 * to allow sysfs to work. We also disallow any kind of
711 int dev_valid_name(const char *name)
715 if (strlen(name) >= IFNAMSIZ)
717 if (!strcmp(name, ".") || !strcmp(name, ".."))
721 if (*name == '/' || isspace(*name))
729 * dev_alloc_name - allocate a name for a device
731 * @name: name format string
733 * Passed a format string - eg "lt%d" it will try and find a suitable
734 * id. It scans list of devices to build up a free map, then chooses
735 * the first empty slot. The caller must hold the dev_base or rtnl lock
736 * while allocating the name and adding the device in order to avoid
738 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
739 * Returns the number of the unit assigned or a negative errno code.
742 int dev_alloc_name(struct net_device *dev, const char *name)
747 const int max_netdevices = 8*PAGE_SIZE;
749 struct net_device *d;
751 p = strnchr(name, IFNAMSIZ-1, '%');
754 * Verify the string as this thing may have come from
755 * the user. There must be either one "%d" and no other "%"
758 if (p[1] != 'd' || strchr(p + 2, '%'))
761 /* Use one page as a bit array of possible slots */
762 inuse = (long *) get_zeroed_page(GFP_ATOMIC);
767 if (!sscanf(d->name, name, &i))
769 if (i < 0 || i >= max_netdevices)
772 /* avoid cases where sscanf is not exact inverse of printf */
773 snprintf(buf, sizeof(buf), name, i);
774 if (!strncmp(buf, d->name, IFNAMSIZ))
778 i = find_first_zero_bit(inuse, max_netdevices);
779 free_page((unsigned long) inuse);
782 snprintf(buf, sizeof(buf), name, i);
783 if (!__dev_get_by_name(buf)) {
784 strlcpy(dev->name, buf, IFNAMSIZ);
788 /* It is possible to run out of possible slots
789 * when the name is long and there isn't enough space left
790 * for the digits, or if all bits are used.
797 * dev_change_name - change name of a device
799 * @newname: name (or format string) must be at least IFNAMSIZ
801 * Change name of a device, can pass format strings "eth%d".
804 int dev_change_name(struct net_device *dev, char *newname)
810 if (dev->flags & IFF_UP)
813 if (!dev_valid_name(newname))
816 if (strchr(newname, '%')) {
817 err = dev_alloc_name(dev, newname);
820 strcpy(newname, dev->name);
822 else if (__dev_get_by_name(newname))
825 strlcpy(dev->name, newname, IFNAMSIZ);
827 device_rename(&dev->dev, dev->name);
828 hlist_del(&dev->name_hlist);
829 hlist_add_head(&dev->name_hlist, dev_name_hash(dev->name));
830 raw_notifier_call_chain(&netdev_chain, NETDEV_CHANGENAME, dev);
836 * netdev_features_change - device changes features
837 * @dev: device to cause notification
839 * Called to indicate a device has changed features.
841 void netdev_features_change(struct net_device *dev)
843 raw_notifier_call_chain(&netdev_chain, NETDEV_FEAT_CHANGE, dev);
845 EXPORT_SYMBOL(netdev_features_change);
848 * netdev_state_change - device changes state
849 * @dev: device to cause notification
851 * Called to indicate a device has changed state. This function calls
852 * the notifier chains for netdev_chain and sends a NEWLINK message
853 * to the routing socket.
855 void netdev_state_change(struct net_device *dev)
857 if (dev->flags & IFF_UP) {
858 raw_notifier_call_chain(&netdev_chain,
860 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
865 * dev_load - load a network module
866 * @name: name of interface
868 * If a network interface is not present and the process has suitable
869 * privileges this function loads the module. If module loading is not
870 * available in this kernel then it becomes a nop.
873 void dev_load(const char *name)
875 struct net_device *dev;
877 read_lock(&dev_base_lock);
878 dev = __dev_get_by_name(name);
879 read_unlock(&dev_base_lock);
881 if (!dev && capable(CAP_SYS_MODULE))
882 request_module("%s", name);
885 static int default_rebuild_header(struct sk_buff *skb)
887 printk(KERN_DEBUG "%s: default_rebuild_header called -- BUG!\n",
888 skb->dev ? skb->dev->name : "NULL!!!");
894 * dev_open - prepare an interface for use.
895 * @dev: device to open
897 * Takes a device from down to up state. The device's private open
898 * function is invoked and then the multicast lists are loaded. Finally
899 * the device is moved into the up state and a %NETDEV_UP message is
900 * sent to the netdev notifier chain.
902 * Calling this function on an active interface is a nop. On a failure
903 * a negative errno code is returned.
905 int dev_open(struct net_device *dev)
913 if (dev->flags & IFF_UP)
917 * Is it even present?
919 if (!netif_device_present(dev))
923 * Call device private open method
925 set_bit(__LINK_STATE_START, &dev->state);
927 ret = dev->open(dev);
929 clear_bit(__LINK_STATE_START, &dev->state);
933 * If it went open OK then:
940 dev->flags |= IFF_UP;
943 * Initialize multicasting status
945 dev_set_rx_mode(dev);
948 * Wakeup transmit queue engine
953 * ... and announce new interface.
955 raw_notifier_call_chain(&netdev_chain, NETDEV_UP, dev);
961 * dev_close - shutdown an interface.
962 * @dev: device to shutdown
964 * This function moves an active device into down state. A
965 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
966 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
969 int dev_close(struct net_device *dev)
971 if (!(dev->flags & IFF_UP))
975 * Tell people we are going down, so that they can
976 * prepare to death, when device is still operating.
978 raw_notifier_call_chain(&netdev_chain, NETDEV_GOING_DOWN, dev);
982 clear_bit(__LINK_STATE_START, &dev->state);
984 /* Synchronize to scheduled poll. We cannot touch poll list,
985 * it can be even on different cpu. So just clear netif_running(),
986 * and wait when poll really will happen. Actually, the best place
987 * for this is inside dev->stop() after device stopped its irq
988 * engine, but this requires more changes in devices. */
990 smp_mb__after_clear_bit(); /* Commit netif_running(). */
991 while (test_bit(__LINK_STATE_RX_SCHED, &dev->state)) {
997 * Call the device specific close. This cannot fail.
998 * Only if device is UP
1000 * We allow it to be called even after a DETACH hot-plug
1007 * Device is now down.
1010 dev->flags &= ~IFF_UP;
1013 * Tell people we are down
1015 raw_notifier_call_chain(&netdev_chain, NETDEV_DOWN, dev);
1022 * Device change register/unregister. These are not inline or static
1023 * as we export them to the world.
1027 * register_netdevice_notifier - register a network notifier block
1030 * Register a notifier to be called when network device events occur.
1031 * The notifier passed is linked into the kernel structures and must
1032 * not be reused until it has been unregistered. A negative errno code
1033 * is returned on a failure.
1035 * When registered all registration and up events are replayed
1036 * to the new notifier to allow device to have a race free
1037 * view of the network device list.
1040 int register_netdevice_notifier(struct notifier_block *nb)
1042 struct net_device *dev;
1046 err = raw_notifier_chain_register(&netdev_chain, nb);
1048 for_each_netdev(dev) {
1049 nb->notifier_call(nb, NETDEV_REGISTER, dev);
1051 if (dev->flags & IFF_UP)
1052 nb->notifier_call(nb, NETDEV_UP, dev);
1060 * unregister_netdevice_notifier - unregister a network notifier block
1063 * Unregister a notifier previously registered by
1064 * register_netdevice_notifier(). The notifier is unlinked into the
1065 * kernel structures and may then be reused. A negative errno code
1066 * is returned on a failure.
1069 int unregister_netdevice_notifier(struct notifier_block *nb)
1074 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1080 * call_netdevice_notifiers - call all network notifier blocks
1081 * @val: value passed unmodified to notifier function
1082 * @v: pointer passed unmodified to notifier function
1084 * Call all network notifier blocks. Parameters and return value
1085 * are as for raw_notifier_call_chain().
1088 int call_netdevice_notifiers(unsigned long val, void *v)
1090 return raw_notifier_call_chain(&netdev_chain, val, v);
1093 /* When > 0 there are consumers of rx skb time stamps */
1094 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1096 void net_enable_timestamp(void)
1098 atomic_inc(&netstamp_needed);
1101 void net_disable_timestamp(void)
1103 atomic_dec(&netstamp_needed);
1106 static inline void net_timestamp(struct sk_buff *skb)
1108 if (atomic_read(&netstamp_needed))
1109 __net_timestamp(skb);
1111 skb->tstamp.tv64 = 0;
1115 * Support routine. Sends outgoing frames to any network
1116 * taps currently in use.
1119 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1121 struct packet_type *ptype;
1126 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1127 /* Never send packets back to the socket
1128 * they originated from - MvS (miquels@drinkel.ow.org)
1130 if ((ptype->dev == dev || !ptype->dev) &&
1131 (ptype->af_packet_priv == NULL ||
1132 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1133 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1137 /* skb->nh should be correctly
1138 set by sender, so that the second statement is
1139 just protection against buggy protocols.
1141 skb_reset_mac_header(skb2);
1143 if (skb_network_header(skb2) < skb2->data ||
1144 skb2->network_header > skb2->tail) {
1145 if (net_ratelimit())
1146 printk(KERN_CRIT "protocol %04x is "
1148 skb2->protocol, dev->name);
1149 skb_reset_network_header(skb2);
1152 skb2->transport_header = skb2->network_header;
1153 skb2->pkt_type = PACKET_OUTGOING;
1154 ptype->func(skb2, skb->dev, ptype, skb->dev);
1161 void __netif_schedule(struct net_device *dev)
1163 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1164 unsigned long flags;
1165 struct softnet_data *sd;
1167 local_irq_save(flags);
1168 sd = &__get_cpu_var(softnet_data);
1169 dev->next_sched = sd->output_queue;
1170 sd->output_queue = dev;
1171 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1172 local_irq_restore(flags);
1175 EXPORT_SYMBOL(__netif_schedule);
1177 void __netif_rx_schedule(struct net_device *dev)
1179 unsigned long flags;
1181 local_irq_save(flags);
1183 list_add_tail(&dev->poll_list, &__get_cpu_var(softnet_data).poll_list);
1185 dev->quota += dev->weight;
1187 dev->quota = dev->weight;
1188 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
1189 local_irq_restore(flags);
1191 EXPORT_SYMBOL(__netif_rx_schedule);
1193 void dev_kfree_skb_any(struct sk_buff *skb)
1195 if (in_irq() || irqs_disabled())
1196 dev_kfree_skb_irq(skb);
1200 EXPORT_SYMBOL(dev_kfree_skb_any);
1204 void netif_device_detach(struct net_device *dev)
1206 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1207 netif_running(dev)) {
1208 netif_stop_queue(dev);
1211 EXPORT_SYMBOL(netif_device_detach);
1213 void netif_device_attach(struct net_device *dev)
1215 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1216 netif_running(dev)) {
1217 netif_wake_queue(dev);
1218 __netdev_watchdog_up(dev);
1221 EXPORT_SYMBOL(netif_device_attach);
1225 * Invalidate hardware checksum when packet is to be mangled, and
1226 * complete checksum manually on outgoing path.
1228 int skb_checksum_help(struct sk_buff *skb)
1231 int ret = 0, offset;
1233 if (skb->ip_summed == CHECKSUM_COMPLETE)
1234 goto out_set_summed;
1236 if (unlikely(skb_shinfo(skb)->gso_size)) {
1237 /* Let GSO fix up the checksum. */
1238 goto out_set_summed;
1241 if (skb_cloned(skb)) {
1242 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1247 offset = skb->csum_start - skb_headroom(skb);
1248 BUG_ON(offset > (int)skb->len);
1249 csum = skb_checksum(skb, offset, skb->len-offset, 0);
1251 offset = skb_headlen(skb) - offset;
1252 BUG_ON(offset <= 0);
1253 BUG_ON(skb->csum_offset + 2 > offset);
1255 *(__sum16 *)(skb->head + skb->csum_start + skb->csum_offset) =
1258 skb->ip_summed = CHECKSUM_NONE;
1264 * skb_gso_segment - Perform segmentation on skb.
1265 * @skb: buffer to segment
1266 * @features: features for the output path (see dev->features)
1268 * This function segments the given skb and returns a list of segments.
1270 * It may return NULL if the skb requires no segmentation. This is
1271 * only possible when GSO is used for verifying header integrity.
1273 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1275 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1276 struct packet_type *ptype;
1277 __be16 type = skb->protocol;
1280 BUG_ON(skb_shinfo(skb)->frag_list);
1282 skb_reset_mac_header(skb);
1283 skb->mac_len = skb->network_header - skb->mac_header;
1284 __skb_pull(skb, skb->mac_len);
1286 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1287 if (skb_header_cloned(skb) &&
1288 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1289 return ERR_PTR(err);
1293 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type) & 15], list) {
1294 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1295 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1296 err = ptype->gso_send_check(skb);
1297 segs = ERR_PTR(err);
1298 if (err || skb_gso_ok(skb, features))
1300 __skb_push(skb, (skb->data -
1301 skb_network_header(skb)));
1303 segs = ptype->gso_segment(skb, features);
1309 __skb_push(skb, skb->data - skb_mac_header(skb));
1314 EXPORT_SYMBOL(skb_gso_segment);
1316 /* Take action when hardware reception checksum errors are detected. */
1318 void netdev_rx_csum_fault(struct net_device *dev)
1320 if (net_ratelimit()) {
1321 printk(KERN_ERR "%s: hw csum failure.\n",
1322 dev ? dev->name : "<unknown>");
1326 EXPORT_SYMBOL(netdev_rx_csum_fault);
1329 /* Actually, we should eliminate this check as soon as we know, that:
1330 * 1. IOMMU is present and allows to map all the memory.
1331 * 2. No high memory really exists on this machine.
1334 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1336 #ifdef CONFIG_HIGHMEM
1339 if (dev->features & NETIF_F_HIGHDMA)
1342 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1343 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1351 void (*destructor)(struct sk_buff *skb);
1354 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1356 static void dev_gso_skb_destructor(struct sk_buff *skb)
1358 struct dev_gso_cb *cb;
1361 struct sk_buff *nskb = skb->next;
1363 skb->next = nskb->next;
1366 } while (skb->next);
1368 cb = DEV_GSO_CB(skb);
1370 cb->destructor(skb);
1374 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1375 * @skb: buffer to segment
1377 * This function segments the given skb and stores the list of segments
1380 static int dev_gso_segment(struct sk_buff *skb)
1382 struct net_device *dev = skb->dev;
1383 struct sk_buff *segs;
1384 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1387 segs = skb_gso_segment(skb, features);
1389 /* Verifying header integrity only. */
1393 if (unlikely(IS_ERR(segs)))
1394 return PTR_ERR(segs);
1397 DEV_GSO_CB(skb)->destructor = skb->destructor;
1398 skb->destructor = dev_gso_skb_destructor;
1403 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1405 if (likely(!skb->next)) {
1406 if (!list_empty(&ptype_all))
1407 dev_queue_xmit_nit(skb, dev);
1409 if (netif_needs_gso(dev, skb)) {
1410 if (unlikely(dev_gso_segment(skb)))
1416 return dev->hard_start_xmit(skb, dev);
1421 struct sk_buff *nskb = skb->next;
1424 skb->next = nskb->next;
1426 rc = dev->hard_start_xmit(nskb, dev);
1428 nskb->next = skb->next;
1432 if (unlikely(netif_queue_stopped(dev) && skb->next))
1433 return NETDEV_TX_BUSY;
1434 } while (skb->next);
1436 skb->destructor = DEV_GSO_CB(skb)->destructor;
1443 #define HARD_TX_LOCK(dev, cpu) { \
1444 if ((dev->features & NETIF_F_LLTX) == 0) { \
1445 netif_tx_lock(dev); \
1449 #define HARD_TX_UNLOCK(dev) { \
1450 if ((dev->features & NETIF_F_LLTX) == 0) { \
1451 netif_tx_unlock(dev); \
1456 * dev_queue_xmit - transmit a buffer
1457 * @skb: buffer to transmit
1459 * Queue a buffer for transmission to a network device. The caller must
1460 * have set the device and priority and built the buffer before calling
1461 * this function. The function can be called from an interrupt.
1463 * A negative errno code is returned on a failure. A success does not
1464 * guarantee the frame will be transmitted as it may be dropped due
1465 * to congestion or traffic shaping.
1467 * -----------------------------------------------------------------------------------
1468 * I notice this method can also return errors from the queue disciplines,
1469 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1472 * Regardless of the return value, the skb is consumed, so it is currently
1473 * difficult to retry a send to this method. (You can bump the ref count
1474 * before sending to hold a reference for retry if you are careful.)
1476 * When calling this method, interrupts MUST be enabled. This is because
1477 * the BH enable code must have IRQs enabled so that it will not deadlock.
1481 int dev_queue_xmit(struct sk_buff *skb)
1483 struct net_device *dev = skb->dev;
1487 /* GSO will handle the following emulations directly. */
1488 if (netif_needs_gso(dev, skb))
1491 if (skb_shinfo(skb)->frag_list &&
1492 !(dev->features & NETIF_F_FRAGLIST) &&
1493 __skb_linearize(skb))
1496 /* Fragmented skb is linearized if device does not support SG,
1497 * or if at least one of fragments is in highmem and device
1498 * does not support DMA from it.
1500 if (skb_shinfo(skb)->nr_frags &&
1501 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1502 __skb_linearize(skb))
1505 /* If packet is not checksummed and device does not support
1506 * checksumming for this protocol, complete checksumming here.
1508 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1509 skb_set_transport_header(skb, skb->csum_start -
1512 if (!(dev->features & NETIF_F_GEN_CSUM) &&
1513 !((dev->features & NETIF_F_IP_CSUM) &&
1514 skb->protocol == htons(ETH_P_IP)) &&
1515 !((dev->features & NETIF_F_IPV6_CSUM) &&
1516 skb->protocol == htons(ETH_P_IPV6)))
1517 if (skb_checksum_help(skb))
1522 spin_lock_prefetch(&dev->queue_lock);
1524 /* Disable soft irqs for various locks below. Also
1525 * stops preemption for RCU.
1529 /* Updates of qdisc are serialized by queue_lock.
1530 * The struct Qdisc which is pointed to by qdisc is now a
1531 * rcu structure - it may be accessed without acquiring
1532 * a lock (but the structure may be stale.) The freeing of the
1533 * qdisc will be deferred until it's known that there are no
1534 * more references to it.
1536 * If the qdisc has an enqueue function, we still need to
1537 * hold the queue_lock before calling it, since queue_lock
1538 * also serializes access to the device queue.
1541 q = rcu_dereference(dev->qdisc);
1542 #ifdef CONFIG_NET_CLS_ACT
1543 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1546 /* Grab device queue */
1547 spin_lock(&dev->queue_lock);
1550 rc = q->enqueue(skb, q);
1552 spin_unlock(&dev->queue_lock);
1554 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1557 spin_unlock(&dev->queue_lock);
1560 /* The device has no queue. Common case for software devices:
1561 loopback, all the sorts of tunnels...
1563 Really, it is unlikely that netif_tx_lock protection is necessary
1564 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1566 However, it is possible, that they rely on protection
1569 Check this and shot the lock. It is not prone from deadlocks.
1570 Either shot noqueue qdisc, it is even simpler 8)
1572 if (dev->flags & IFF_UP) {
1573 int cpu = smp_processor_id(); /* ok because BHs are off */
1575 if (dev->xmit_lock_owner != cpu) {
1577 HARD_TX_LOCK(dev, cpu);
1579 if (!netif_queue_stopped(dev)) {
1581 if (!dev_hard_start_xmit(skb, dev)) {
1582 HARD_TX_UNLOCK(dev);
1586 HARD_TX_UNLOCK(dev);
1587 if (net_ratelimit())
1588 printk(KERN_CRIT "Virtual device %s asks to "
1589 "queue packet!\n", dev->name);
1591 /* Recursion is detected! It is possible,
1593 if (net_ratelimit())
1594 printk(KERN_CRIT "Dead loop on virtual device "
1595 "%s, fix it urgently!\n", dev->name);
1600 rcu_read_unlock_bh();
1606 rcu_read_unlock_bh();
1611 /*=======================================================================
1613 =======================================================================*/
1615 int netdev_max_backlog __read_mostly = 1000;
1616 int netdev_budget __read_mostly = 300;
1617 int weight_p __read_mostly = 64; /* old backlog weight */
1619 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1623 * netif_rx - post buffer to the network code
1624 * @skb: buffer to post
1626 * This function receives a packet from a device driver and queues it for
1627 * the upper (protocol) levels to process. It always succeeds. The buffer
1628 * may be dropped during processing for congestion control or by the
1632 * NET_RX_SUCCESS (no congestion)
1633 * NET_RX_CN_LOW (low congestion)
1634 * NET_RX_CN_MOD (moderate congestion)
1635 * NET_RX_CN_HIGH (high congestion)
1636 * NET_RX_DROP (packet was dropped)
1640 int netif_rx(struct sk_buff *skb)
1642 struct softnet_data *queue;
1643 unsigned long flags;
1645 /* if netpoll wants it, pretend we never saw it */
1646 if (netpoll_rx(skb))
1649 if (!skb->tstamp.tv64)
1653 * The code is rearranged so that the path is the most
1654 * short when CPU is congested, but is still operating.
1656 local_irq_save(flags);
1657 queue = &__get_cpu_var(softnet_data);
1659 __get_cpu_var(netdev_rx_stat).total++;
1660 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1661 if (queue->input_pkt_queue.qlen) {
1664 __skb_queue_tail(&queue->input_pkt_queue, skb);
1665 local_irq_restore(flags);
1666 return NET_RX_SUCCESS;
1669 netif_rx_schedule(&queue->backlog_dev);
1673 __get_cpu_var(netdev_rx_stat).dropped++;
1674 local_irq_restore(flags);
1680 int netif_rx_ni(struct sk_buff *skb)
1685 err = netif_rx(skb);
1686 if (local_softirq_pending())
1693 EXPORT_SYMBOL(netif_rx_ni);
1695 static inline struct net_device *skb_bond(struct sk_buff *skb)
1697 struct net_device *dev = skb->dev;
1700 if (skb_bond_should_drop(skb)) {
1704 skb->dev = dev->master;
1710 static void net_tx_action(struct softirq_action *h)
1712 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1714 if (sd->completion_queue) {
1715 struct sk_buff *clist;
1717 local_irq_disable();
1718 clist = sd->completion_queue;
1719 sd->completion_queue = NULL;
1723 struct sk_buff *skb = clist;
1724 clist = clist->next;
1726 BUG_TRAP(!atomic_read(&skb->users));
1731 if (sd->output_queue) {
1732 struct net_device *head;
1734 local_irq_disable();
1735 head = sd->output_queue;
1736 sd->output_queue = NULL;
1740 struct net_device *dev = head;
1741 head = head->next_sched;
1743 smp_mb__before_clear_bit();
1744 clear_bit(__LINK_STATE_SCHED, &dev->state);
1746 if (spin_trylock(&dev->queue_lock)) {
1748 spin_unlock(&dev->queue_lock);
1750 netif_schedule(dev);
1756 static inline int deliver_skb(struct sk_buff *skb,
1757 struct packet_type *pt_prev,
1758 struct net_device *orig_dev)
1760 atomic_inc(&skb->users);
1761 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1764 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1765 /* These hooks defined here for ATM */
1767 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1768 unsigned char *addr);
1769 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1772 * If bridge module is loaded call bridging hook.
1773 * returns NULL if packet was consumed.
1775 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
1776 struct sk_buff *skb) __read_mostly;
1777 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
1778 struct packet_type **pt_prev, int *ret,
1779 struct net_device *orig_dev)
1781 struct net_bridge_port *port;
1783 if (skb->pkt_type == PACKET_LOOPBACK ||
1784 (port = rcu_dereference(skb->dev->br_port)) == NULL)
1788 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1792 return br_handle_frame_hook(port, skb);
1795 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1798 #ifdef CONFIG_NET_CLS_ACT
1799 /* TODO: Maybe we should just force sch_ingress to be compiled in
1800 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1801 * a compare and 2 stores extra right now if we dont have it on
1802 * but have CONFIG_NET_CLS_ACT
1803 * NOTE: This doesnt stop any functionality; if you dont have
1804 * the ingress scheduler, you just cant add policies on ingress.
1807 static int ing_filter(struct sk_buff *skb)
1810 struct net_device *dev = skb->dev;
1811 int result = TC_ACT_OK;
1813 if (dev->qdisc_ingress) {
1814 __u32 ttl = (__u32) G_TC_RTTL(skb->tc_verd);
1815 if (MAX_RED_LOOP < ttl++) {
1816 printk(KERN_WARNING "Redir loop detected Dropping packet (%d->%d)\n",
1817 skb->iif, skb->dev->ifindex);
1821 skb->tc_verd = SET_TC_RTTL(skb->tc_verd,ttl);
1823 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_INGRESS);
1825 spin_lock(&dev->ingress_lock);
1826 if ((q = dev->qdisc_ingress) != NULL)
1827 result = q->enqueue(skb, q);
1828 spin_unlock(&dev->ingress_lock);
1836 int netif_receive_skb(struct sk_buff *skb)
1838 struct packet_type *ptype, *pt_prev;
1839 struct net_device *orig_dev;
1840 int ret = NET_RX_DROP;
1843 /* if we've gotten here through NAPI, check netpoll */
1844 if (skb->dev->poll && netpoll_rx(skb))
1847 if (!skb->tstamp.tv64)
1851 skb->iif = skb->dev->ifindex;
1853 orig_dev = skb_bond(skb);
1858 __get_cpu_var(netdev_rx_stat).total++;
1860 skb_reset_network_header(skb);
1861 skb_reset_transport_header(skb);
1862 skb->mac_len = skb->network_header - skb->mac_header;
1868 #ifdef CONFIG_NET_CLS_ACT
1869 if (skb->tc_verd & TC_NCLS) {
1870 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
1875 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1876 if (!ptype->dev || ptype->dev == skb->dev) {
1878 ret = deliver_skb(skb, pt_prev, orig_dev);
1883 #ifdef CONFIG_NET_CLS_ACT
1885 ret = deliver_skb(skb, pt_prev, orig_dev);
1886 pt_prev = NULL; /* noone else should process this after*/
1888 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1891 ret = ing_filter(skb);
1893 if (ret == TC_ACT_SHOT || (ret == TC_ACT_STOLEN)) {
1902 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
1906 type = skb->protocol;
1907 list_for_each_entry_rcu(ptype, &ptype_base[ntohs(type)&15], list) {
1908 if (ptype->type == type &&
1909 (!ptype->dev || ptype->dev == skb->dev)) {
1911 ret = deliver_skb(skb, pt_prev, orig_dev);
1917 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1920 /* Jamal, now you will not able to escape explaining
1921 * me how you were going to use this. :-)
1931 static int process_backlog(struct net_device *backlog_dev, int *budget)
1934 int quota = min(backlog_dev->quota, *budget);
1935 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1936 unsigned long start_time = jiffies;
1938 backlog_dev->weight = weight_p;
1940 struct sk_buff *skb;
1941 struct net_device *dev;
1943 local_irq_disable();
1944 skb = __skb_dequeue(&queue->input_pkt_queue);
1951 netif_receive_skb(skb);
1957 if (work >= quota || jiffies - start_time > 1)
1962 backlog_dev->quota -= work;
1967 backlog_dev->quota -= work;
1970 list_del(&backlog_dev->poll_list);
1971 smp_mb__before_clear_bit();
1972 netif_poll_enable(backlog_dev);
1978 static void net_rx_action(struct softirq_action *h)
1980 struct softnet_data *queue = &__get_cpu_var(softnet_data);
1981 unsigned long start_time = jiffies;
1982 int budget = netdev_budget;
1985 local_irq_disable();
1987 while (!list_empty(&queue->poll_list)) {
1988 struct net_device *dev;
1990 if (budget <= 0 || jiffies - start_time > 1)
1995 dev = list_entry(queue->poll_list.next,
1996 struct net_device, poll_list);
1997 have = netpoll_poll_lock(dev);
1999 if (dev->quota <= 0 || dev->poll(dev, &budget)) {
2000 netpoll_poll_unlock(have);
2001 local_irq_disable();
2002 list_move_tail(&dev->poll_list, &queue->poll_list);
2004 dev->quota += dev->weight;
2006 dev->quota = dev->weight;
2008 netpoll_poll_unlock(have);
2010 local_irq_disable();
2015 #ifdef CONFIG_NET_DMA
2017 * There may not be any more sk_buffs coming right now, so push
2018 * any pending DMA copies to hardware
2020 if (net_dma_client) {
2021 struct dma_chan *chan;
2023 list_for_each_entry_rcu(chan, &net_dma_client->channels, client_node)
2024 dma_async_memcpy_issue_pending(chan);
2031 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2032 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2036 static gifconf_func_t * gifconf_list [NPROTO];
2039 * register_gifconf - register a SIOCGIF handler
2040 * @family: Address family
2041 * @gifconf: Function handler
2043 * Register protocol dependent address dumping routines. The handler
2044 * that is passed must not be freed or reused until it has been replaced
2045 * by another handler.
2047 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2049 if (family >= NPROTO)
2051 gifconf_list[family] = gifconf;
2057 * Map an interface index to its name (SIOCGIFNAME)
2061 * We need this ioctl for efficient implementation of the
2062 * if_indextoname() function required by the IPv6 API. Without
2063 * it, we would have to search all the interfaces to find a
2067 static int dev_ifname(struct ifreq __user *arg)
2069 struct net_device *dev;
2073 * Fetch the caller's info block.
2076 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2079 read_lock(&dev_base_lock);
2080 dev = __dev_get_by_index(ifr.ifr_ifindex);
2082 read_unlock(&dev_base_lock);
2086 strcpy(ifr.ifr_name, dev->name);
2087 read_unlock(&dev_base_lock);
2089 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2095 * Perform a SIOCGIFCONF call. This structure will change
2096 * size eventually, and there is nothing I can do about it.
2097 * Thus we will need a 'compatibility mode'.
2100 static int dev_ifconf(char __user *arg)
2103 struct net_device *dev;
2110 * Fetch the caller's info block.
2113 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2120 * Loop over the interfaces, and write an info block for each.
2124 for_each_netdev(dev) {
2125 for (i = 0; i < NPROTO; i++) {
2126 if (gifconf_list[i]) {
2129 done = gifconf_list[i](dev, NULL, 0);
2131 done = gifconf_list[i](dev, pos + total,
2141 * All done. Write the updated control block back to the caller.
2143 ifc.ifc_len = total;
2146 * Both BSD and Solaris return 0 here, so we do too.
2148 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2151 #ifdef CONFIG_PROC_FS
2153 * This is invoked by the /proc filesystem handler to display a device
2156 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2159 struct net_device *dev;
2161 read_lock(&dev_base_lock);
2163 return SEQ_START_TOKEN;
2166 for_each_netdev(dev)
2173 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2176 return v == SEQ_START_TOKEN ?
2177 first_net_device() : next_net_device((struct net_device *)v);
2180 void dev_seq_stop(struct seq_file *seq, void *v)
2182 read_unlock(&dev_base_lock);
2185 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2187 struct net_device_stats *stats = dev->get_stats(dev);
2189 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2190 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2191 dev->name, stats->rx_bytes, stats->rx_packets,
2193 stats->rx_dropped + stats->rx_missed_errors,
2194 stats->rx_fifo_errors,
2195 stats->rx_length_errors + stats->rx_over_errors +
2196 stats->rx_crc_errors + stats->rx_frame_errors,
2197 stats->rx_compressed, stats->multicast,
2198 stats->tx_bytes, stats->tx_packets,
2199 stats->tx_errors, stats->tx_dropped,
2200 stats->tx_fifo_errors, stats->collisions,
2201 stats->tx_carrier_errors +
2202 stats->tx_aborted_errors +
2203 stats->tx_window_errors +
2204 stats->tx_heartbeat_errors,
2205 stats->tx_compressed);
2209 * Called from the PROCfs module. This now uses the new arbitrary sized
2210 * /proc/net interface to create /proc/net/dev
2212 static int dev_seq_show(struct seq_file *seq, void *v)
2214 if (v == SEQ_START_TOKEN)
2215 seq_puts(seq, "Inter-| Receive "
2217 " face |bytes packets errs drop fifo frame "
2218 "compressed multicast|bytes packets errs "
2219 "drop fifo colls carrier compressed\n");
2221 dev_seq_printf_stats(seq, v);
2225 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2227 struct netif_rx_stats *rc = NULL;
2229 while (*pos < NR_CPUS)
2230 if (cpu_online(*pos)) {
2231 rc = &per_cpu(netdev_rx_stat, *pos);
2238 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2240 return softnet_get_online(pos);
2243 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2246 return softnet_get_online(pos);
2249 static void softnet_seq_stop(struct seq_file *seq, void *v)
2253 static int softnet_seq_show(struct seq_file *seq, void *v)
2255 struct netif_rx_stats *s = v;
2257 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2258 s->total, s->dropped, s->time_squeeze, 0,
2259 0, 0, 0, 0, /* was fastroute */
2264 static const struct seq_operations dev_seq_ops = {
2265 .start = dev_seq_start,
2266 .next = dev_seq_next,
2267 .stop = dev_seq_stop,
2268 .show = dev_seq_show,
2271 static int dev_seq_open(struct inode *inode, struct file *file)
2273 return seq_open(file, &dev_seq_ops);
2276 static const struct file_operations dev_seq_fops = {
2277 .owner = THIS_MODULE,
2278 .open = dev_seq_open,
2280 .llseek = seq_lseek,
2281 .release = seq_release,
2284 static const struct seq_operations softnet_seq_ops = {
2285 .start = softnet_seq_start,
2286 .next = softnet_seq_next,
2287 .stop = softnet_seq_stop,
2288 .show = softnet_seq_show,
2291 static int softnet_seq_open(struct inode *inode, struct file *file)
2293 return seq_open(file, &softnet_seq_ops);
2296 static const struct file_operations softnet_seq_fops = {
2297 .owner = THIS_MODULE,
2298 .open = softnet_seq_open,
2300 .llseek = seq_lseek,
2301 .release = seq_release,
2304 static void *ptype_get_idx(loff_t pos)
2306 struct packet_type *pt = NULL;
2310 list_for_each_entry_rcu(pt, &ptype_all, list) {
2316 for (t = 0; t < 16; t++) {
2317 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2326 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2329 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2332 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2334 struct packet_type *pt;
2335 struct list_head *nxt;
2339 if (v == SEQ_START_TOKEN)
2340 return ptype_get_idx(0);
2343 nxt = pt->list.next;
2344 if (pt->type == htons(ETH_P_ALL)) {
2345 if (nxt != &ptype_all)
2348 nxt = ptype_base[0].next;
2350 hash = ntohs(pt->type) & 15;
2352 while (nxt == &ptype_base[hash]) {
2355 nxt = ptype_base[hash].next;
2358 return list_entry(nxt, struct packet_type, list);
2361 static void ptype_seq_stop(struct seq_file *seq, void *v)
2366 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2368 #ifdef CONFIG_KALLSYMS
2369 unsigned long offset = 0, symsize;
2370 const char *symname;
2374 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2381 modname = delim = "";
2382 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2388 seq_printf(seq, "[%p]", sym);
2391 static int ptype_seq_show(struct seq_file *seq, void *v)
2393 struct packet_type *pt = v;
2395 if (v == SEQ_START_TOKEN)
2396 seq_puts(seq, "Type Device Function\n");
2398 if (pt->type == htons(ETH_P_ALL))
2399 seq_puts(seq, "ALL ");
2401 seq_printf(seq, "%04x", ntohs(pt->type));
2403 seq_printf(seq, " %-8s ",
2404 pt->dev ? pt->dev->name : "");
2405 ptype_seq_decode(seq, pt->func);
2406 seq_putc(seq, '\n');
2412 static const struct seq_operations ptype_seq_ops = {
2413 .start = ptype_seq_start,
2414 .next = ptype_seq_next,
2415 .stop = ptype_seq_stop,
2416 .show = ptype_seq_show,
2419 static int ptype_seq_open(struct inode *inode, struct file *file)
2421 return seq_open(file, &ptype_seq_ops);
2424 static const struct file_operations ptype_seq_fops = {
2425 .owner = THIS_MODULE,
2426 .open = ptype_seq_open,
2428 .llseek = seq_lseek,
2429 .release = seq_release,
2433 static int __init dev_proc_init(void)
2437 if (!proc_net_fops_create("dev", S_IRUGO, &dev_seq_fops))
2439 if (!proc_net_fops_create("softnet_stat", S_IRUGO, &softnet_seq_fops))
2441 if (!proc_net_fops_create("ptype", S_IRUGO, &ptype_seq_fops))
2444 if (wext_proc_init())
2450 proc_net_remove("ptype");
2452 proc_net_remove("softnet_stat");
2454 proc_net_remove("dev");
2458 #define dev_proc_init() 0
2459 #endif /* CONFIG_PROC_FS */
2463 * netdev_set_master - set up master/slave pair
2464 * @slave: slave device
2465 * @master: new master device
2467 * Changes the master device of the slave. Pass %NULL to break the
2468 * bonding. The caller must hold the RTNL semaphore. On a failure
2469 * a negative errno code is returned. On success the reference counts
2470 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2471 * function returns zero.
2473 int netdev_set_master(struct net_device *slave, struct net_device *master)
2475 struct net_device *old = slave->master;
2485 slave->master = master;
2493 slave->flags |= IFF_SLAVE;
2495 slave->flags &= ~IFF_SLAVE;
2497 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2501 static void __dev_set_promiscuity(struct net_device *dev, int inc)
2503 unsigned short old_flags = dev->flags;
2505 if ((dev->promiscuity += inc) == 0)
2506 dev->flags &= ~IFF_PROMISC;
2508 dev->flags |= IFF_PROMISC;
2509 if (dev->flags != old_flags) {
2510 printk(KERN_INFO "device %s %s promiscuous mode\n",
2511 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2513 audit_log(current->audit_context, GFP_ATOMIC,
2514 AUDIT_ANOM_PROMISCUOUS,
2515 "dev=%s prom=%d old_prom=%d auid=%u",
2516 dev->name, (dev->flags & IFF_PROMISC),
2517 (old_flags & IFF_PROMISC),
2518 audit_get_loginuid(current->audit_context));
2523 * dev_set_promiscuity - update promiscuity count on a device
2527 * Add or remove promiscuity from a device. While the count in the device
2528 * remains above zero the interface remains promiscuous. Once it hits zero
2529 * the device reverts back to normal filtering operation. A negative inc
2530 * value is used to drop promiscuity on the device.
2532 void dev_set_promiscuity(struct net_device *dev, int inc)
2534 unsigned short old_flags = dev->flags;
2536 __dev_set_promiscuity(dev, inc);
2537 if (dev->flags != old_flags)
2538 dev_set_rx_mode(dev);
2542 * dev_set_allmulti - update allmulti count on a device
2546 * Add or remove reception of all multicast frames to a device. While the
2547 * count in the device remains above zero the interface remains listening
2548 * to all interfaces. Once it hits zero the device reverts back to normal
2549 * filtering operation. A negative @inc value is used to drop the counter
2550 * when releasing a resource needing all multicasts.
2553 void dev_set_allmulti(struct net_device *dev, int inc)
2555 unsigned short old_flags = dev->flags;
2557 dev->flags |= IFF_ALLMULTI;
2558 if ((dev->allmulti += inc) == 0)
2559 dev->flags &= ~IFF_ALLMULTI;
2560 if (dev->flags ^ old_flags)
2561 dev_set_rx_mode(dev);
2565 * Upload unicast and multicast address lists to device and
2566 * configure RX filtering. When the device doesn't support unicast
2567 * filtering it is put in promiscous mode while unicast addresses
2570 void __dev_set_rx_mode(struct net_device *dev)
2572 /* dev_open will call this function so the list will stay sane. */
2573 if (!(dev->flags&IFF_UP))
2576 if (!netif_device_present(dev))
2579 if (dev->set_rx_mode)
2580 dev->set_rx_mode(dev);
2582 /* Unicast addresses changes may only happen under the rtnl,
2583 * therefore calling __dev_set_promiscuity here is safe.
2585 if (dev->uc_count > 0 && !dev->uc_promisc) {
2586 __dev_set_promiscuity(dev, 1);
2587 dev->uc_promisc = 1;
2588 } else if (dev->uc_count == 0 && dev->uc_promisc) {
2589 __dev_set_promiscuity(dev, -1);
2590 dev->uc_promisc = 0;
2593 if (dev->set_multicast_list)
2594 dev->set_multicast_list(dev);
2598 void dev_set_rx_mode(struct net_device *dev)
2600 netif_tx_lock_bh(dev);
2601 __dev_set_rx_mode(dev);
2602 netif_tx_unlock_bh(dev);
2605 int __dev_addr_delete(struct dev_addr_list **list, void *addr, int alen,
2608 struct dev_addr_list *da;
2610 for (; (da = *list) != NULL; list = &da->next) {
2611 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2612 alen == da->da_addrlen) {
2614 int old_glbl = da->da_gusers;
2630 int __dev_addr_add(struct dev_addr_list **list, void *addr, int alen, int glbl)
2632 struct dev_addr_list *da;
2634 for (da = *list; da != NULL; da = da->next) {
2635 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2636 da->da_addrlen == alen) {
2638 int old_glbl = da->da_gusers;
2648 da = kmalloc(sizeof(*da), GFP_ATOMIC);
2651 memcpy(da->da_addr, addr, alen);
2652 da->da_addrlen = alen;
2654 da->da_gusers = glbl ? 1 : 0;
2660 void __dev_addr_discard(struct dev_addr_list **list)
2662 struct dev_addr_list *tmp;
2664 while (*list != NULL) {
2667 if (tmp->da_users > tmp->da_gusers)
2668 printk("__dev_addr_discard: address leakage! "
2669 "da_users=%d\n", tmp->da_users);
2675 * dev_unicast_delete - Release secondary unicast address.
2678 * Release reference to a secondary unicast address and remove it
2679 * from the device if the reference count drop to zero.
2681 * The caller must hold the rtnl_mutex.
2683 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
2689 netif_tx_lock_bh(dev);
2690 err = __dev_addr_delete(&dev->uc_list, addr, alen, 0);
2693 __dev_set_rx_mode(dev);
2695 netif_tx_unlock_bh(dev);
2698 EXPORT_SYMBOL(dev_unicast_delete);
2701 * dev_unicast_add - add a secondary unicast address
2704 * Add a secondary unicast address to the device or increase
2705 * the reference count if it already exists.
2707 * The caller must hold the rtnl_mutex.
2709 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
2715 netif_tx_lock_bh(dev);
2716 err = __dev_addr_add(&dev->uc_list, addr, alen, 0);
2719 __dev_set_rx_mode(dev);
2721 netif_tx_unlock_bh(dev);
2724 EXPORT_SYMBOL(dev_unicast_add);
2726 static void dev_unicast_discard(struct net_device *dev)
2728 netif_tx_lock_bh(dev);
2729 __dev_addr_discard(&dev->uc_list);
2731 netif_tx_unlock_bh(dev);
2734 unsigned dev_get_flags(const struct net_device *dev)
2738 flags = (dev->flags & ~(IFF_PROMISC |
2743 (dev->gflags & (IFF_PROMISC |
2746 if (netif_running(dev)) {
2747 if (netif_oper_up(dev))
2748 flags |= IFF_RUNNING;
2749 if (netif_carrier_ok(dev))
2750 flags |= IFF_LOWER_UP;
2751 if (netif_dormant(dev))
2752 flags |= IFF_DORMANT;
2758 int dev_change_flags(struct net_device *dev, unsigned flags)
2761 int old_flags = dev->flags;
2764 * Set the flags on our device.
2767 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
2768 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
2770 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
2774 * Load in the correct multicast list now the flags have changed.
2777 dev_set_rx_mode(dev);
2780 * Have we downed the interface. We handle IFF_UP ourselves
2781 * according to user attempts to set it, rather than blindly
2786 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
2787 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
2790 dev_set_rx_mode(dev);
2793 if (dev->flags & IFF_UP &&
2794 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
2796 raw_notifier_call_chain(&netdev_chain,
2797 NETDEV_CHANGE, dev);
2799 if ((flags ^ dev->gflags) & IFF_PROMISC) {
2800 int inc = (flags & IFF_PROMISC) ? +1 : -1;
2801 dev->gflags ^= IFF_PROMISC;
2802 dev_set_promiscuity(dev, inc);
2805 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
2806 is important. Some (broken) drivers set IFF_PROMISC, when
2807 IFF_ALLMULTI is requested not asking us and not reporting.
2809 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
2810 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
2811 dev->gflags ^= IFF_ALLMULTI;
2812 dev_set_allmulti(dev, inc);
2815 /* Exclude state transition flags, already notified */
2816 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
2818 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
2823 int dev_set_mtu(struct net_device *dev, int new_mtu)
2827 if (new_mtu == dev->mtu)
2830 /* MTU must be positive. */
2834 if (!netif_device_present(dev))
2838 if (dev->change_mtu)
2839 err = dev->change_mtu(dev, new_mtu);
2842 if (!err && dev->flags & IFF_UP)
2843 raw_notifier_call_chain(&netdev_chain,
2844 NETDEV_CHANGEMTU, dev);
2848 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
2852 if (!dev->set_mac_address)
2854 if (sa->sa_family != dev->type)
2856 if (!netif_device_present(dev))
2858 err = dev->set_mac_address(dev, sa);
2860 raw_notifier_call_chain(&netdev_chain,
2861 NETDEV_CHANGEADDR, dev);
2866 * Perform the SIOCxIFxxx calls.
2868 static int dev_ifsioc(struct ifreq *ifr, unsigned int cmd)
2871 struct net_device *dev = __dev_get_by_name(ifr->ifr_name);
2877 case SIOCGIFFLAGS: /* Get interface flags */
2878 ifr->ifr_flags = dev_get_flags(dev);
2881 case SIOCSIFFLAGS: /* Set interface flags */
2882 return dev_change_flags(dev, ifr->ifr_flags);
2884 case SIOCGIFMETRIC: /* Get the metric on the interface
2885 (currently unused) */
2886 ifr->ifr_metric = 0;
2889 case SIOCSIFMETRIC: /* Set the metric on the interface
2890 (currently unused) */
2893 case SIOCGIFMTU: /* Get the MTU of a device */
2894 ifr->ifr_mtu = dev->mtu;
2897 case SIOCSIFMTU: /* Set the MTU of a device */
2898 return dev_set_mtu(dev, ifr->ifr_mtu);
2902 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
2904 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
2905 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2906 ifr->ifr_hwaddr.sa_family = dev->type;
2910 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
2912 case SIOCSIFHWBROADCAST:
2913 if (ifr->ifr_hwaddr.sa_family != dev->type)
2915 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
2916 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
2917 raw_notifier_call_chain(&netdev_chain,
2918 NETDEV_CHANGEADDR, dev);
2922 ifr->ifr_map.mem_start = dev->mem_start;
2923 ifr->ifr_map.mem_end = dev->mem_end;
2924 ifr->ifr_map.base_addr = dev->base_addr;
2925 ifr->ifr_map.irq = dev->irq;
2926 ifr->ifr_map.dma = dev->dma;
2927 ifr->ifr_map.port = dev->if_port;
2931 if (dev->set_config) {
2932 if (!netif_device_present(dev))
2934 return dev->set_config(dev, &ifr->ifr_map);
2939 if (!dev->set_multicast_list ||
2940 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2942 if (!netif_device_present(dev))
2944 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
2948 if (!dev->set_multicast_list ||
2949 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
2951 if (!netif_device_present(dev))
2953 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
2957 ifr->ifr_ifindex = dev->ifindex;
2961 ifr->ifr_qlen = dev->tx_queue_len;
2965 if (ifr->ifr_qlen < 0)
2967 dev->tx_queue_len = ifr->ifr_qlen;
2971 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
2972 return dev_change_name(dev, ifr->ifr_newname);
2975 * Unknown or private ioctl
2979 if ((cmd >= SIOCDEVPRIVATE &&
2980 cmd <= SIOCDEVPRIVATE + 15) ||
2981 cmd == SIOCBONDENSLAVE ||
2982 cmd == SIOCBONDRELEASE ||
2983 cmd == SIOCBONDSETHWADDR ||
2984 cmd == SIOCBONDSLAVEINFOQUERY ||
2985 cmd == SIOCBONDINFOQUERY ||
2986 cmd == SIOCBONDCHANGEACTIVE ||
2987 cmd == SIOCGMIIPHY ||
2988 cmd == SIOCGMIIREG ||
2989 cmd == SIOCSMIIREG ||
2990 cmd == SIOCBRADDIF ||
2991 cmd == SIOCBRDELIF ||
2992 cmd == SIOCWANDEV) {
2994 if (dev->do_ioctl) {
2995 if (netif_device_present(dev))
2996 err = dev->do_ioctl(dev, ifr,
3009 * This function handles all "interface"-type I/O control requests. The actual
3010 * 'doing' part of this is dev_ifsioc above.
3014 * dev_ioctl - network device ioctl
3015 * @cmd: command to issue
3016 * @arg: pointer to a struct ifreq in user space
3018 * Issue ioctl functions to devices. This is normally called by the
3019 * user space syscall interfaces but can sometimes be useful for
3020 * other purposes. The return value is the return from the syscall if
3021 * positive or a negative errno code on error.
3024 int dev_ioctl(unsigned int cmd, void __user *arg)
3030 /* One special case: SIOCGIFCONF takes ifconf argument
3031 and requires shared lock, because it sleeps writing
3035 if (cmd == SIOCGIFCONF) {
3037 ret = dev_ifconf((char __user *) arg);
3041 if (cmd == SIOCGIFNAME)
3042 return dev_ifname((struct ifreq __user *)arg);
3044 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3047 ifr.ifr_name[IFNAMSIZ-1] = 0;
3049 colon = strchr(ifr.ifr_name, ':');
3054 * See which interface the caller is talking about.
3059 * These ioctl calls:
3060 * - can be done by all.
3061 * - atomic and do not require locking.
3072 dev_load(ifr.ifr_name);
3073 read_lock(&dev_base_lock);
3074 ret = dev_ifsioc(&ifr, cmd);
3075 read_unlock(&dev_base_lock);
3079 if (copy_to_user(arg, &ifr,
3080 sizeof(struct ifreq)))
3086 dev_load(ifr.ifr_name);
3088 ret = dev_ethtool(&ifr);
3093 if (copy_to_user(arg, &ifr,
3094 sizeof(struct ifreq)))
3100 * These ioctl calls:
3101 * - require superuser power.
3102 * - require strict serialization.
3108 if (!capable(CAP_NET_ADMIN))
3110 dev_load(ifr.ifr_name);
3112 ret = dev_ifsioc(&ifr, cmd);
3117 if (copy_to_user(arg, &ifr,
3118 sizeof(struct ifreq)))
3124 * These ioctl calls:
3125 * - require superuser power.
3126 * - require strict serialization.
3127 * - do not return a value
3137 case SIOCSIFHWBROADCAST:
3140 case SIOCBONDENSLAVE:
3141 case SIOCBONDRELEASE:
3142 case SIOCBONDSETHWADDR:
3143 case SIOCBONDCHANGEACTIVE:
3146 if (!capable(CAP_NET_ADMIN))
3149 case SIOCBONDSLAVEINFOQUERY:
3150 case SIOCBONDINFOQUERY:
3151 dev_load(ifr.ifr_name);
3153 ret = dev_ifsioc(&ifr, cmd);
3158 /* Get the per device memory space. We can add this but
3159 * currently do not support it */
3161 /* Set the per device memory buffer space.
3162 * Not applicable in our case */
3167 * Unknown or private ioctl.
3170 if (cmd == SIOCWANDEV ||
3171 (cmd >= SIOCDEVPRIVATE &&
3172 cmd <= SIOCDEVPRIVATE + 15)) {
3173 dev_load(ifr.ifr_name);
3175 ret = dev_ifsioc(&ifr, cmd);
3177 if (!ret && copy_to_user(arg, &ifr,
3178 sizeof(struct ifreq)))
3182 /* Take care of Wireless Extensions */
3183 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3184 return wext_handle_ioctl(&ifr, cmd, arg);
3191 * dev_new_index - allocate an ifindex
3193 * Returns a suitable unique value for a new device interface
3194 * number. The caller must hold the rtnl semaphore or the
3195 * dev_base_lock to be sure it remains unique.
3197 static int dev_new_index(void)
3203 if (!__dev_get_by_index(ifindex))
3208 static int dev_boot_phase = 1;
3210 /* Delayed registration/unregisteration */
3211 static DEFINE_SPINLOCK(net_todo_list_lock);
3212 static struct list_head net_todo_list = LIST_HEAD_INIT(net_todo_list);
3214 static void net_set_todo(struct net_device *dev)
3216 spin_lock(&net_todo_list_lock);
3217 list_add_tail(&dev->todo_list, &net_todo_list);
3218 spin_unlock(&net_todo_list_lock);
3222 * register_netdevice - register a network device
3223 * @dev: device to register
3225 * Take a completed network device structure and add it to the kernel
3226 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3227 * chain. 0 is returned on success. A negative errno code is returned
3228 * on a failure to set up the device, or if the name is a duplicate.
3230 * Callers must hold the rtnl semaphore. You may want
3231 * register_netdev() instead of this.
3234 * The locking appears insufficient to guarantee two parallel registers
3235 * will not get the same name.
3238 int register_netdevice(struct net_device *dev)
3240 struct hlist_head *head;
3241 struct hlist_node *p;
3244 BUG_ON(dev_boot_phase);
3249 /* When net_device's are persistent, this will be fatal. */
3250 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3252 spin_lock_init(&dev->queue_lock);
3253 spin_lock_init(&dev->_xmit_lock);
3254 netdev_set_lockdep_class(&dev->_xmit_lock, dev->type);
3255 dev->xmit_lock_owner = -1;
3256 spin_lock_init(&dev->ingress_lock);
3260 /* Init, if this function is available */
3262 ret = dev->init(dev);
3270 if (!dev_valid_name(dev->name)) {
3275 dev->ifindex = dev_new_index();
3276 if (dev->iflink == -1)
3277 dev->iflink = dev->ifindex;
3279 /* Check for existence of name */
3280 head = dev_name_hash(dev->name);
3281 hlist_for_each(p, head) {
3282 struct net_device *d
3283 = hlist_entry(p, struct net_device, name_hlist);
3284 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3290 /* Fix illegal checksum combinations */
3291 if ((dev->features & NETIF_F_HW_CSUM) &&
3292 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3293 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3295 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3298 if ((dev->features & NETIF_F_NO_CSUM) &&
3299 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3300 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3302 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3306 /* Fix illegal SG+CSUM combinations. */
3307 if ((dev->features & NETIF_F_SG) &&
3308 !(dev->features & NETIF_F_ALL_CSUM)) {
3309 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3311 dev->features &= ~NETIF_F_SG;
3314 /* TSO requires that SG is present as well. */
3315 if ((dev->features & NETIF_F_TSO) &&
3316 !(dev->features & NETIF_F_SG)) {
3317 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3319 dev->features &= ~NETIF_F_TSO;
3321 if (dev->features & NETIF_F_UFO) {
3322 if (!(dev->features & NETIF_F_HW_CSUM)) {
3323 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3324 "NETIF_F_HW_CSUM feature.\n",
3326 dev->features &= ~NETIF_F_UFO;
3328 if (!(dev->features & NETIF_F_SG)) {
3329 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3330 "NETIF_F_SG feature.\n",
3332 dev->features &= ~NETIF_F_UFO;
3337 * nil rebuild_header routine,
3338 * that should be never called and used as just bug trap.
3341 if (!dev->rebuild_header)
3342 dev->rebuild_header = default_rebuild_header;
3344 ret = netdev_register_sysfs(dev);
3347 dev->reg_state = NETREG_REGISTERED;
3350 * Default initial state at registry is that the
3351 * device is present.
3354 set_bit(__LINK_STATE_PRESENT, &dev->state);
3356 dev_init_scheduler(dev);
3357 write_lock_bh(&dev_base_lock);
3358 list_add_tail(&dev->dev_list, &dev_base_head);
3359 hlist_add_head(&dev->name_hlist, head);
3360 hlist_add_head(&dev->index_hlist, dev_index_hash(dev->ifindex));
3362 write_unlock_bh(&dev_base_lock);
3364 /* Notify protocols, that a new device appeared. */
3365 raw_notifier_call_chain(&netdev_chain, NETDEV_REGISTER, dev);
3374 * register_netdev - register a network device
3375 * @dev: device to register
3377 * Take a completed network device structure and add it to the kernel
3378 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3379 * chain. 0 is returned on success. A negative errno code is returned
3380 * on a failure to set up the device, or if the name is a duplicate.
3382 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3383 * and expands the device name if you passed a format string to
3386 int register_netdev(struct net_device *dev)
3393 * If the name is a format string the caller wants us to do a
3396 if (strchr(dev->name, '%')) {
3397 err = dev_alloc_name(dev, dev->name);
3402 err = register_netdevice(dev);
3407 EXPORT_SYMBOL(register_netdev);
3410 * netdev_wait_allrefs - wait until all references are gone.
3412 * This is called when unregistering network devices.
3414 * Any protocol or device that holds a reference should register
3415 * for netdevice notification, and cleanup and put back the
3416 * reference if they receive an UNREGISTER event.
3417 * We can get stuck here if buggy protocols don't correctly
3420 static void netdev_wait_allrefs(struct net_device *dev)
3422 unsigned long rebroadcast_time, warning_time;
3424 rebroadcast_time = warning_time = jiffies;
3425 while (atomic_read(&dev->refcnt) != 0) {
3426 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3429 /* Rebroadcast unregister notification */
3430 raw_notifier_call_chain(&netdev_chain,
3431 NETDEV_UNREGISTER, dev);
3433 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3435 /* We must not have linkwatch events
3436 * pending on unregister. If this
3437 * happens, we simply run the queue
3438 * unscheduled, resulting in a noop
3441 linkwatch_run_queue();
3446 rebroadcast_time = jiffies;
3451 if (time_after(jiffies, warning_time + 10 * HZ)) {
3452 printk(KERN_EMERG "unregister_netdevice: "
3453 "waiting for %s to become free. Usage "
3455 dev->name, atomic_read(&dev->refcnt));
3456 warning_time = jiffies;
3465 * register_netdevice(x1);
3466 * register_netdevice(x2);
3468 * unregister_netdevice(y1);
3469 * unregister_netdevice(y2);
3475 * We are invoked by rtnl_unlock() after it drops the semaphore.
3476 * This allows us to deal with problems:
3477 * 1) We can delete sysfs objects which invoke hotplug
3478 * without deadlocking with linkwatch via keventd.
3479 * 2) Since we run with the RTNL semaphore not held, we can sleep
3480 * safely in order to wait for the netdev refcnt to drop to zero.
3482 static DEFINE_MUTEX(net_todo_run_mutex);
3483 void netdev_run_todo(void)
3485 struct list_head list;
3487 /* Need to guard against multiple cpu's getting out of order. */
3488 mutex_lock(&net_todo_run_mutex);
3490 /* Not safe to do outside the semaphore. We must not return
3491 * until all unregister events invoked by the local processor
3492 * have been completed (either by this todo run, or one on
3495 if (list_empty(&net_todo_list))
3498 /* Snapshot list, allow later requests */
3499 spin_lock(&net_todo_list_lock);
3500 list_replace_init(&net_todo_list, &list);
3501 spin_unlock(&net_todo_list_lock);
3503 while (!list_empty(&list)) {
3504 struct net_device *dev
3505 = list_entry(list.next, struct net_device, todo_list);
3506 list_del(&dev->todo_list);
3508 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3509 printk(KERN_ERR "network todo '%s' but state %d\n",
3510 dev->name, dev->reg_state);
3515 dev->reg_state = NETREG_UNREGISTERED;
3517 netdev_wait_allrefs(dev);
3520 BUG_ON(atomic_read(&dev->refcnt));
3521 BUG_TRAP(!dev->ip_ptr);
3522 BUG_TRAP(!dev->ip6_ptr);
3523 BUG_TRAP(!dev->dn_ptr);
3525 if (dev->destructor)
3526 dev->destructor(dev);
3528 /* Free network device */
3529 kobject_put(&dev->dev.kobj);
3533 mutex_unlock(&net_todo_run_mutex);
3536 static struct net_device_stats *internal_stats(struct net_device *dev)
3542 * alloc_netdev - allocate network device
3543 * @sizeof_priv: size of private data to allocate space for
3544 * @name: device name format string
3545 * @setup: callback to initialize device
3547 * Allocates a struct net_device with private data area for driver use
3548 * and performs basic initialization.
3550 struct net_device *alloc_netdev(int sizeof_priv, const char *name,
3551 void (*setup)(struct net_device *))
3554 struct net_device *dev;
3557 BUG_ON(strlen(name) >= sizeof(dev->name));
3559 /* ensure 32-byte alignment of both the device and private area */
3560 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
3561 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3563 p = kzalloc(alloc_size, GFP_KERNEL);
3565 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
3569 dev = (struct net_device *)
3570 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3571 dev->padded = (char *)dev - (char *)p;
3574 dev->priv = netdev_priv(dev);
3576 dev->get_stats = internal_stats;
3578 strcpy(dev->name, name);
3581 EXPORT_SYMBOL(alloc_netdev);
3584 * free_netdev - free network device
3587 * This function does the last stage of destroying an allocated device
3588 * interface. The reference to the device object is released.
3589 * If this is the last reference then it will be freed.
3591 void free_netdev(struct net_device *dev)
3594 /* Compatibility with error handling in drivers */
3595 if (dev->reg_state == NETREG_UNINITIALIZED) {
3596 kfree((char *)dev - dev->padded);
3600 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3601 dev->reg_state = NETREG_RELEASED;
3603 /* will free via device release */
3604 put_device(&dev->dev);
3606 kfree((char *)dev - dev->padded);
3610 /* Synchronize with packet receive processing. */
3611 void synchronize_net(void)
3618 * unregister_netdevice - remove device from the kernel
3621 * This function shuts down a device interface and removes it
3622 * from the kernel tables. On success 0 is returned, on a failure
3623 * a negative errno code is returned.
3625 * Callers must hold the rtnl semaphore. You may want
3626 * unregister_netdev() instead of this.
3629 void unregister_netdevice(struct net_device *dev)
3631 BUG_ON(dev_boot_phase);
3634 /* Some devices call without registering for initialization unwind. */
3635 if (dev->reg_state == NETREG_UNINITIALIZED) {
3636 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3637 "was registered\n", dev->name, dev);
3643 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3645 /* If device is running, close it first. */
3646 if (dev->flags & IFF_UP)
3649 /* And unlink it from device chain. */
3650 write_lock_bh(&dev_base_lock);
3651 list_del(&dev->dev_list);
3652 hlist_del(&dev->name_hlist);
3653 hlist_del(&dev->index_hlist);
3654 write_unlock_bh(&dev_base_lock);
3656 dev->reg_state = NETREG_UNREGISTERING;
3660 /* Shutdown queueing discipline. */
3664 /* Notify protocols, that we are about to destroy
3665 this device. They should clean all the things.
3667 raw_notifier_call_chain(&netdev_chain, NETDEV_UNREGISTER, dev);
3670 * Flush the unicast and multicast chains
3672 dev_unicast_discard(dev);
3673 dev_mc_discard(dev);
3678 /* Notifier chain MUST detach us from master device. */
3679 BUG_TRAP(!dev->master);
3681 /* Remove entries from sysfs */
3682 netdev_unregister_sysfs(dev);
3684 /* Finish processing unregister after unlock */
3693 * unregister_netdev - remove device from the kernel
3696 * This function shuts down a device interface and removes it
3697 * from the kernel tables. On success 0 is returned, on a failure
3698 * a negative errno code is returned.
3700 * This is just a wrapper for unregister_netdevice that takes
3701 * the rtnl semaphore. In general you want to use this and not
3702 * unregister_netdevice.
3704 void unregister_netdev(struct net_device *dev)
3707 unregister_netdevice(dev);
3711 EXPORT_SYMBOL(unregister_netdev);
3713 static int dev_cpu_callback(struct notifier_block *nfb,
3714 unsigned long action,
3717 struct sk_buff **list_skb;
3718 struct net_device **list_net;
3719 struct sk_buff *skb;
3720 unsigned int cpu, oldcpu = (unsigned long)ocpu;
3721 struct softnet_data *sd, *oldsd;
3723 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
3726 local_irq_disable();
3727 cpu = smp_processor_id();
3728 sd = &per_cpu(softnet_data, cpu);
3729 oldsd = &per_cpu(softnet_data, oldcpu);
3731 /* Find end of our completion_queue. */
3732 list_skb = &sd->completion_queue;
3734 list_skb = &(*list_skb)->next;
3735 /* Append completion queue from offline CPU. */
3736 *list_skb = oldsd->completion_queue;
3737 oldsd->completion_queue = NULL;
3739 /* Find end of our output_queue. */
3740 list_net = &sd->output_queue;
3742 list_net = &(*list_net)->next_sched;
3743 /* Append output queue from offline CPU. */
3744 *list_net = oldsd->output_queue;
3745 oldsd->output_queue = NULL;
3747 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3750 /* Process offline CPU's input_pkt_queue */
3751 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
3757 #ifdef CONFIG_NET_DMA
3759 * net_dma_rebalance -
3760 * This is called when the number of channels allocated to the net_dma_client
3761 * changes. The net_dma_client tries to have one DMA channel per CPU.
3763 static void net_dma_rebalance(void)
3765 unsigned int cpu, i, n;
3766 struct dma_chan *chan;
3768 if (net_dma_count == 0) {
3769 for_each_online_cpu(cpu)
3770 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
3775 cpu = first_cpu(cpu_online_map);
3778 list_for_each_entry(chan, &net_dma_client->channels, client_node) {
3779 n = ((num_online_cpus() / net_dma_count)
3780 + (i < (num_online_cpus() % net_dma_count) ? 1 : 0));
3783 per_cpu(softnet_data, cpu).net_dma = chan;
3784 cpu = next_cpu(cpu, cpu_online_map);
3793 * netdev_dma_event - event callback for the net_dma_client
3794 * @client: should always be net_dma_client
3795 * @chan: DMA channel for the event
3796 * @event: event type
3798 static void netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
3799 enum dma_event event)
3801 spin_lock(&net_dma_event_lock);
3803 case DMA_RESOURCE_ADDED:
3805 net_dma_rebalance();
3807 case DMA_RESOURCE_REMOVED:
3809 net_dma_rebalance();
3814 spin_unlock(&net_dma_event_lock);
3818 * netdev_dma_regiser - register the networking subsystem as a DMA client
3820 static int __init netdev_dma_register(void)
3822 spin_lock_init(&net_dma_event_lock);
3823 net_dma_client = dma_async_client_register(netdev_dma_event);
3824 if (net_dma_client == NULL)
3827 dma_async_client_chan_request(net_dma_client, num_online_cpus());
3832 static int __init netdev_dma_register(void) { return -ENODEV; }
3833 #endif /* CONFIG_NET_DMA */
3836 * Initialize the DEV module. At boot time this walks the device list and
3837 * unhooks any devices that fail to initialise (normally hardware not
3838 * present) and leaves us with a valid list of present and active devices.
3843 * This is called single threaded during boot, so no need
3844 * to take the rtnl semaphore.
3846 static int __init net_dev_init(void)
3848 int i, rc = -ENOMEM;
3850 BUG_ON(!dev_boot_phase);
3852 if (dev_proc_init())
3855 if (netdev_sysfs_init())
3858 INIT_LIST_HEAD(&ptype_all);
3859 for (i = 0; i < 16; i++)
3860 INIT_LIST_HEAD(&ptype_base[i]);
3862 for (i = 0; i < ARRAY_SIZE(dev_name_head); i++)
3863 INIT_HLIST_HEAD(&dev_name_head[i]);
3865 for (i = 0; i < ARRAY_SIZE(dev_index_head); i++)
3866 INIT_HLIST_HEAD(&dev_index_head[i]);
3869 * Initialise the packet receive queues.
3872 for_each_possible_cpu(i) {
3873 struct softnet_data *queue;
3875 queue = &per_cpu(softnet_data, i);
3876 skb_queue_head_init(&queue->input_pkt_queue);
3877 queue->completion_queue = NULL;
3878 INIT_LIST_HEAD(&queue->poll_list);
3879 set_bit(__LINK_STATE_START, &queue->backlog_dev.state);
3880 queue->backlog_dev.weight = weight_p;
3881 queue->backlog_dev.poll = process_backlog;
3882 atomic_set(&queue->backlog_dev.refcnt, 1);
3885 netdev_dma_register();
3889 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
3890 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
3892 hotcpu_notifier(dev_cpu_callback, 0);
3900 subsys_initcall(net_dev_init);
3902 EXPORT_SYMBOL(__dev_get_by_index);
3903 EXPORT_SYMBOL(__dev_get_by_name);
3904 EXPORT_SYMBOL(__dev_remove_pack);
3905 EXPORT_SYMBOL(dev_valid_name);
3906 EXPORT_SYMBOL(dev_add_pack);
3907 EXPORT_SYMBOL(dev_alloc_name);
3908 EXPORT_SYMBOL(dev_close);
3909 EXPORT_SYMBOL(dev_get_by_flags);
3910 EXPORT_SYMBOL(dev_get_by_index);
3911 EXPORT_SYMBOL(dev_get_by_name);
3912 EXPORT_SYMBOL(dev_open);
3913 EXPORT_SYMBOL(dev_queue_xmit);
3914 EXPORT_SYMBOL(dev_remove_pack);
3915 EXPORT_SYMBOL(dev_set_allmulti);
3916 EXPORT_SYMBOL(dev_set_promiscuity);
3917 EXPORT_SYMBOL(dev_change_flags);
3918 EXPORT_SYMBOL(dev_set_mtu);
3919 EXPORT_SYMBOL(dev_set_mac_address);
3920 EXPORT_SYMBOL(free_netdev);
3921 EXPORT_SYMBOL(netdev_boot_setup_check);
3922 EXPORT_SYMBOL(netdev_set_master);
3923 EXPORT_SYMBOL(netdev_state_change);
3924 EXPORT_SYMBOL(netif_receive_skb);
3925 EXPORT_SYMBOL(netif_rx);
3926 EXPORT_SYMBOL(register_gifconf);
3927 EXPORT_SYMBOL(register_netdevice);
3928 EXPORT_SYMBOL(register_netdevice_notifier);
3929 EXPORT_SYMBOL(skb_checksum_help);
3930 EXPORT_SYMBOL(synchronize_net);
3931 EXPORT_SYMBOL(unregister_netdevice);
3932 EXPORT_SYMBOL(unregister_netdevice_notifier);
3933 EXPORT_SYMBOL(net_enable_timestamp);
3934 EXPORT_SYMBOL(net_disable_timestamp);
3935 EXPORT_SYMBOL(dev_get_flags);
3937 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
3938 EXPORT_SYMBOL(br_handle_frame_hook);
3939 EXPORT_SYMBOL(br_fdb_get_hook);
3940 EXPORT_SYMBOL(br_fdb_put_hook);
3944 EXPORT_SYMBOL(dev_load);
3947 EXPORT_PER_CPU_SYMBOL(softnet_data);