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>
123 #include "net-sysfs.h"
126 * The list of packet types we will receive (as opposed to discard)
127 * and the routines to invoke.
129 * Why 16. Because with 16 the only overlap we get on a hash of the
130 * low nibble of the protocol value is RARP/SNAP/X.25.
132 * NOTE: That is no longer true with the addition of VLAN tags. Not
133 * sure which should go first, but I bet it won't make much
134 * difference if we are running VLANs. The good news is that
135 * this protocol won't be in the list unless compiled in, so
136 * the average user (w/out VLANs) will not be adversely affected.
153 #define PTYPE_HASH_SIZE (16)
154 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
156 static DEFINE_SPINLOCK(ptype_lock);
157 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
158 static struct list_head ptype_all __read_mostly; /* Taps */
160 #ifdef CONFIG_NET_DMA
162 struct dma_client client;
164 cpumask_t channel_mask;
165 struct dma_chan *channels[NR_CPUS];
168 static enum dma_state_client
169 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
170 enum dma_state state);
172 static struct net_dma net_dma = {
174 .event_callback = netdev_dma_event,
180 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
183 * Pure readers hold dev_base_lock for reading.
185 * Writers must hold the rtnl semaphore while they loop through the
186 * dev_base_head list, and hold dev_base_lock for writing when they do the
187 * actual updates. This allows pure readers to access the list even
188 * while a writer is preparing to update it.
190 * To put it another way, dev_base_lock is held for writing only to
191 * protect against pure readers; the rtnl semaphore provides the
192 * protection against other writers.
194 * See, for example usages, register_netdevice() and
195 * unregister_netdevice(), which must be called with the rtnl
198 DEFINE_RWLOCK(dev_base_lock);
200 EXPORT_SYMBOL(dev_base_lock);
202 #define NETDEV_HASHBITS 8
203 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
205 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
207 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
208 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
211 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
213 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
216 /* Device list insertion */
217 static int list_netdevice(struct net_device *dev)
219 struct net *net = dev->nd_net;
223 write_lock_bh(&dev_base_lock);
224 list_add_tail(&dev->dev_list, &net->dev_base_head);
225 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
226 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
227 write_unlock_bh(&dev_base_lock);
231 /* Device list removal */
232 static void unlist_netdevice(struct net_device *dev)
236 /* Unlink dev from the device chain */
237 write_lock_bh(&dev_base_lock);
238 list_del(&dev->dev_list);
239 hlist_del(&dev->name_hlist);
240 hlist_del(&dev->index_hlist);
241 write_unlock_bh(&dev_base_lock);
248 static RAW_NOTIFIER_HEAD(netdev_chain);
251 * Device drivers call our routines to queue packets here. We empty the
252 * queue in the local softnet handler.
255 DEFINE_PER_CPU(struct softnet_data, softnet_data);
257 #ifdef CONFIG_DEBUG_LOCK_ALLOC
259 * register_netdevice() inits dev->_xmit_lock and sets lockdep class
260 * according to dev->type
262 static const unsigned short netdev_lock_type[] =
263 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
264 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
265 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
266 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
267 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
268 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
269 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
270 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
271 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
272 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
273 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
274 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
275 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
276 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_VOID,
279 static const char *netdev_lock_name[] =
280 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
281 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
282 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
283 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
284 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
285 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
286 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
287 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
288 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
289 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
290 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
291 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
292 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
293 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_VOID",
296 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
298 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
302 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
303 if (netdev_lock_type[i] == dev_type)
305 /* the last key is used by default */
306 return ARRAY_SIZE(netdev_lock_type) - 1;
309 static inline void netdev_set_lockdep_class(spinlock_t *lock,
310 unsigned short dev_type)
314 i = netdev_lock_pos(dev_type);
315 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
316 netdev_lock_name[i]);
319 static inline void netdev_set_lockdep_class(spinlock_t *lock,
320 unsigned short dev_type)
325 /*******************************************************************************
327 Protocol management and registration routines
329 *******************************************************************************/
332 * Add a protocol ID to the list. Now that the input handler is
333 * smarter we can dispense with all the messy stuff that used to be
336 * BEWARE!!! Protocol handlers, mangling input packets,
337 * MUST BE last in hash buckets and checking protocol handlers
338 * MUST start from promiscuous ptype_all chain in net_bh.
339 * It is true now, do not change it.
340 * Explanation follows: if protocol handler, mangling packet, will
341 * be the first on list, it is not able to sense, that packet
342 * is cloned and should be copied-on-write, so that it will
343 * change it and subsequent readers will get broken packet.
348 * dev_add_pack - add packet handler
349 * @pt: packet type declaration
351 * Add a protocol handler to the networking stack. The passed &packet_type
352 * is linked into kernel lists and may not be freed until it has been
353 * removed from the kernel lists.
355 * This call does not sleep therefore it can not
356 * guarantee all CPU's that are in middle of receiving packets
357 * will see the new packet type (until the next received packet).
360 void dev_add_pack(struct packet_type *pt)
364 spin_lock_bh(&ptype_lock);
365 if (pt->type == htons(ETH_P_ALL))
366 list_add_rcu(&pt->list, &ptype_all);
368 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
369 list_add_rcu(&pt->list, &ptype_base[hash]);
371 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 * The packet type might still be in use by receivers
384 * and must not be freed until after all the CPU's have gone
385 * through a quiescent state.
387 void __dev_remove_pack(struct packet_type *pt)
389 struct list_head *head;
390 struct packet_type *pt1;
392 spin_lock_bh(&ptype_lock);
394 if (pt->type == htons(ETH_P_ALL))
397 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
399 list_for_each_entry(pt1, head, list) {
401 list_del_rcu(&pt->list);
406 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
408 spin_unlock_bh(&ptype_lock);
411 * dev_remove_pack - remove packet handler
412 * @pt: packet type declaration
414 * Remove a protocol handler that was previously added to the kernel
415 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
416 * from the kernel lists and can be freed or reused once this function
419 * This call sleeps to guarantee that no CPU is looking at the packet
422 void dev_remove_pack(struct packet_type *pt)
424 __dev_remove_pack(pt);
429 /******************************************************************************
431 Device Boot-time Settings Routines
433 *******************************************************************************/
435 /* Boot time configuration table */
436 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
439 * netdev_boot_setup_add - add new setup entry
440 * @name: name of the device
441 * @map: configured settings for the device
443 * Adds new setup entry to the dev_boot_setup list. The function
444 * returns 0 on error and 1 on success. This is a generic routine to
447 static int netdev_boot_setup_add(char *name, struct ifmap *map)
449 struct netdev_boot_setup *s;
453 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
454 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
455 memset(s[i].name, 0, sizeof(s[i].name));
456 strcpy(s[i].name, name);
457 memcpy(&s[i].map, map, sizeof(s[i].map));
462 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
466 * netdev_boot_setup_check - check boot time settings
467 * @dev: the netdevice
469 * Check boot time settings for the device.
470 * The found settings are set for the device to be used
471 * later in the device probing.
472 * Returns 0 if no settings found, 1 if they are.
474 int netdev_boot_setup_check(struct net_device *dev)
476 struct netdev_boot_setup *s = dev_boot_setup;
479 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
480 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
481 !strncmp(dev->name, s[i].name, strlen(s[i].name))) {
482 dev->irq = s[i].map.irq;
483 dev->base_addr = s[i].map.base_addr;
484 dev->mem_start = s[i].map.mem_start;
485 dev->mem_end = s[i].map.mem_end;
494 * netdev_boot_base - get address from boot time settings
495 * @prefix: prefix for network device
496 * @unit: id for network device
498 * Check boot time settings for the base address of device.
499 * The found settings are set for the device to be used
500 * later in the device probing.
501 * Returns 0 if no settings found.
503 unsigned long netdev_boot_base(const char *prefix, int unit)
505 const struct netdev_boot_setup *s = dev_boot_setup;
509 sprintf(name, "%s%d", prefix, unit);
512 * If device already registered then return base of 1
513 * to indicate not to probe for this interface
515 if (__dev_get_by_name(&init_net, name))
518 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
519 if (!strcmp(name, s[i].name))
520 return s[i].map.base_addr;
525 * Saves at boot time configured settings for any netdevice.
527 int __init netdev_boot_setup(char *str)
532 str = get_options(str, ARRAY_SIZE(ints), ints);
537 memset(&map, 0, sizeof(map));
541 map.base_addr = ints[2];
543 map.mem_start = ints[3];
545 map.mem_end = ints[4];
547 /* Add new entry to the list */
548 return netdev_boot_setup_add(str, &map);
551 __setup("netdev=", netdev_boot_setup);
553 /*******************************************************************************
555 Device Interface Subroutines
557 *******************************************************************************/
560 * __dev_get_by_name - find a device by its name
561 * @net: the applicable net namespace
562 * @name: name to find
564 * Find an interface by name. Must be called under RTNL semaphore
565 * or @dev_base_lock. If the name is found a pointer to the device
566 * is returned. If the name is not found then %NULL is returned. The
567 * reference counters are not incremented so the caller must be
568 * careful with locks.
571 struct net_device *__dev_get_by_name(struct net *net, const char *name)
573 struct hlist_node *p;
575 hlist_for_each(p, dev_name_hash(net, name)) {
576 struct net_device *dev
577 = hlist_entry(p, struct net_device, name_hlist);
578 if (!strncmp(dev->name, name, IFNAMSIZ))
585 * dev_get_by_name - find a device by its name
586 * @net: the applicable net namespace
587 * @name: name to find
589 * Find an interface by name. This can be called from any
590 * context and does its own locking. The returned handle has
591 * the usage count incremented and the caller must use dev_put() to
592 * release it when it is no longer needed. %NULL is returned if no
593 * matching device is found.
596 struct net_device *dev_get_by_name(struct net *net, const char *name)
598 struct net_device *dev;
600 read_lock(&dev_base_lock);
601 dev = __dev_get_by_name(net, name);
604 read_unlock(&dev_base_lock);
609 * __dev_get_by_index - find a device by its ifindex
610 * @net: the applicable net namespace
611 * @ifindex: index of device
613 * Search for an interface by index. Returns %NULL if the device
614 * is not found or a pointer to the device. The device has not
615 * had its reference counter increased so the caller must be careful
616 * about locking. The caller must hold either the RTNL semaphore
620 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
622 struct hlist_node *p;
624 hlist_for_each(p, dev_index_hash(net, ifindex)) {
625 struct net_device *dev
626 = hlist_entry(p, struct net_device, index_hlist);
627 if (dev->ifindex == ifindex)
635 * dev_get_by_index - find a device by its ifindex
636 * @net: the applicable net namespace
637 * @ifindex: index of device
639 * Search for an interface by index. Returns NULL if the device
640 * is not found or a pointer to the device. The device returned has
641 * had a reference added and the pointer is safe until the user calls
642 * dev_put to indicate they have finished with it.
645 struct net_device *dev_get_by_index(struct net *net, int ifindex)
647 struct net_device *dev;
649 read_lock(&dev_base_lock);
650 dev = __dev_get_by_index(net, ifindex);
653 read_unlock(&dev_base_lock);
658 * dev_getbyhwaddr - find a device by its hardware address
659 * @net: the applicable net namespace
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(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 * @net: the applicable net namespace
719 * @if_flags: IFF_* values
720 * @mask: bitmask of bits in if_flags to check
722 * Search for any interface with the given flags. Returns NULL if a device
723 * is not found or a pointer to the device. The device returned has
724 * had a reference added and the pointer is safe until the user calls
725 * dev_put to indicate they have finished with it.
728 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
730 struct net_device *dev, *ret;
733 read_lock(&dev_base_lock);
734 for_each_netdev(net, dev) {
735 if (((dev->flags ^ if_flags) & mask) == 0) {
741 read_unlock(&dev_base_lock);
746 * dev_valid_name - check if name is okay for network device
749 * Network device names need to be valid file names to
750 * to allow sysfs to work. We also disallow any kind of
753 int dev_valid_name(const char *name)
757 if (strlen(name) >= IFNAMSIZ)
759 if (!strcmp(name, ".") || !strcmp(name, ".."))
763 if (*name == '/' || isspace(*name))
771 * __dev_alloc_name - allocate a name for a device
772 * @net: network namespace to allocate the device name in
773 * @name: name format string
774 * @buf: scratch buffer and result name string
776 * Passed a format string - eg "lt%d" it will try and find a suitable
777 * id. It scans list of devices to build up a free map, then chooses
778 * the first empty slot. The caller must hold the dev_base or rtnl lock
779 * while allocating the name and adding the device in order to avoid
781 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
782 * Returns the number of the unit assigned or a negative errno code.
785 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
789 const int max_netdevices = 8*PAGE_SIZE;
790 unsigned long *inuse;
791 struct net_device *d;
793 p = strnchr(name, IFNAMSIZ-1, '%');
796 * Verify the string as this thing may have come from
797 * the user. There must be either one "%d" and no other "%"
800 if (p[1] != 'd' || strchr(p + 2, '%'))
803 /* Use one page as a bit array of possible slots */
804 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
808 for_each_netdev(net, d) {
809 if (!sscanf(d->name, name, &i))
811 if (i < 0 || i >= max_netdevices)
814 /* avoid cases where sscanf is not exact inverse of printf */
815 snprintf(buf, IFNAMSIZ, name, i);
816 if (!strncmp(buf, d->name, IFNAMSIZ))
820 i = find_first_zero_bit(inuse, max_netdevices);
821 free_page((unsigned long) inuse);
824 snprintf(buf, IFNAMSIZ, name, i);
825 if (!__dev_get_by_name(net, buf))
828 /* It is possible to run out of possible slots
829 * when the name is long and there isn't enough space left
830 * for the digits, or if all bits are used.
836 * dev_alloc_name - allocate a name for a device
838 * @name: name format string
840 * Passed a format string - eg "lt%d" it will try and find a suitable
841 * id. It scans list of devices to build up a free map, then chooses
842 * the first empty slot. The caller must hold the dev_base or rtnl lock
843 * while allocating the name and adding the device in order to avoid
845 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
846 * Returns the number of the unit assigned or a negative errno code.
849 int dev_alloc_name(struct net_device *dev, const char *name)
855 BUG_ON(!dev->nd_net);
857 ret = __dev_alloc_name(net, name, buf);
859 strlcpy(dev->name, buf, IFNAMSIZ);
865 * dev_change_name - change name of a device
867 * @newname: name (or format string) must be at least IFNAMSIZ
869 * Change name of a device, can pass format strings "eth%d".
872 int dev_change_name(struct net_device *dev, char *newname)
874 char oldname[IFNAMSIZ];
880 BUG_ON(!dev->nd_net);
883 if (dev->flags & IFF_UP)
886 if (!dev_valid_name(newname))
889 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
892 memcpy(oldname, dev->name, IFNAMSIZ);
894 if (strchr(newname, '%')) {
895 err = dev_alloc_name(dev, newname);
898 strcpy(newname, dev->name);
900 else if (__dev_get_by_name(net, newname))
903 strlcpy(dev->name, newname, IFNAMSIZ);
906 device_rename(&dev->dev, dev->name);
908 write_lock_bh(&dev_base_lock);
909 hlist_del(&dev->name_hlist);
910 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
911 write_unlock_bh(&dev_base_lock);
913 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
914 ret = notifier_to_errno(ret);
919 "%s: name change rollback failed: %d.\n",
923 memcpy(dev->name, oldname, IFNAMSIZ);
932 * netdev_features_change - device changes features
933 * @dev: device to cause notification
935 * Called to indicate a device has changed features.
937 void netdev_features_change(struct net_device *dev)
939 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
941 EXPORT_SYMBOL(netdev_features_change);
944 * netdev_state_change - device changes state
945 * @dev: device to cause notification
947 * Called to indicate a device has changed state. This function calls
948 * the notifier chains for netdev_chain and sends a NEWLINK message
949 * to the routing socket.
951 void netdev_state_change(struct net_device *dev)
953 if (dev->flags & IFF_UP) {
954 call_netdevice_notifiers(NETDEV_CHANGE, dev);
955 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
960 * dev_load - load a network module
961 * @net: the applicable net namespace
962 * @name: name of interface
964 * If a network interface is not present and the process has suitable
965 * privileges this function loads the module. If module loading is not
966 * available in this kernel then it becomes a nop.
969 void dev_load(struct net *net, const char *name)
971 struct net_device *dev;
973 read_lock(&dev_base_lock);
974 dev = __dev_get_by_name(net, name);
975 read_unlock(&dev_base_lock);
977 if (!dev && capable(CAP_SYS_MODULE))
978 request_module("%s", name);
982 * dev_open - prepare an interface for use.
983 * @dev: device to open
985 * Takes a device from down to up state. The device's private open
986 * function is invoked and then the multicast lists are loaded. Finally
987 * the device is moved into the up state and a %NETDEV_UP message is
988 * sent to the netdev notifier chain.
990 * Calling this function on an active interface is a nop. On a failure
991 * a negative errno code is returned.
993 int dev_open(struct net_device *dev)
1001 if (dev->flags & IFF_UP)
1005 * Is it even present?
1007 if (!netif_device_present(dev))
1011 * Call device private open method
1013 set_bit(__LINK_STATE_START, &dev->state);
1015 if (dev->validate_addr)
1016 ret = dev->validate_addr(dev);
1018 if (!ret && dev->open)
1019 ret = dev->open(dev);
1022 * If it went open OK then:
1026 clear_bit(__LINK_STATE_START, &dev->state);
1031 dev->flags |= IFF_UP;
1034 * Initialize multicasting status
1036 dev_set_rx_mode(dev);
1039 * Wakeup transmit queue engine
1044 * ... and announce new interface.
1046 call_netdevice_notifiers(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 call_netdevice_notifiers(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 call_netdevice_notifiers(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);
1178 raw_notifier_chain_unregister(&netdev_chain, nb);
1183 * unregister_netdevice_notifier - unregister a network notifier block
1186 * Unregister a notifier previously registered by
1187 * register_netdevice_notifier(). The notifier is unlinked into the
1188 * kernel structures and may then be reused. A negative errno code
1189 * is returned on a failure.
1192 int unregister_netdevice_notifier(struct notifier_block *nb)
1197 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1203 * call_netdevice_notifiers - call all network notifier blocks
1204 * @val: value passed unmodified to notifier function
1205 * @dev: net_device pointer passed unmodified to notifier function
1207 * Call all network notifier blocks. Parameters and return value
1208 * are as for raw_notifier_call_chain().
1211 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1213 return raw_notifier_call_chain(&netdev_chain, val, dev);
1216 /* When > 0 there are consumers of rx skb time stamps */
1217 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1219 void net_enable_timestamp(void)
1221 atomic_inc(&netstamp_needed);
1224 void net_disable_timestamp(void)
1226 atomic_dec(&netstamp_needed);
1229 static inline void net_timestamp(struct sk_buff *skb)
1231 if (atomic_read(&netstamp_needed))
1232 __net_timestamp(skb);
1234 skb->tstamp.tv64 = 0;
1238 * Support routine. Sends outgoing frames to any network
1239 * taps currently in use.
1242 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1244 struct packet_type *ptype;
1249 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1250 /* Never send packets back to the socket
1251 * they originated from - MvS (miquels@drinkel.ow.org)
1253 if ((ptype->dev == dev || !ptype->dev) &&
1254 (ptype->af_packet_priv == NULL ||
1255 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1256 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1260 /* skb->nh should be correctly
1261 set by sender, so that the second statement is
1262 just protection against buggy protocols.
1264 skb_reset_mac_header(skb2);
1266 if (skb_network_header(skb2) < skb2->data ||
1267 skb2->network_header > skb2->tail) {
1268 if (net_ratelimit())
1269 printk(KERN_CRIT "protocol %04x is "
1271 skb2->protocol, dev->name);
1272 skb_reset_network_header(skb2);
1275 skb2->transport_header = skb2->network_header;
1276 skb2->pkt_type = PACKET_OUTGOING;
1277 ptype->func(skb2, skb->dev, ptype, skb->dev);
1284 void __netif_schedule(struct net_device *dev)
1286 if (!test_and_set_bit(__LINK_STATE_SCHED, &dev->state)) {
1287 unsigned long flags;
1288 struct softnet_data *sd;
1290 local_irq_save(flags);
1291 sd = &__get_cpu_var(softnet_data);
1292 dev->next_sched = sd->output_queue;
1293 sd->output_queue = dev;
1294 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1295 local_irq_restore(flags);
1298 EXPORT_SYMBOL(__netif_schedule);
1300 void dev_kfree_skb_irq(struct sk_buff *skb)
1302 if (atomic_dec_and_test(&skb->users)) {
1303 struct softnet_data *sd;
1304 unsigned long flags;
1306 local_irq_save(flags);
1307 sd = &__get_cpu_var(softnet_data);
1308 skb->next = sd->completion_queue;
1309 sd->completion_queue = skb;
1310 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1311 local_irq_restore(flags);
1314 EXPORT_SYMBOL(dev_kfree_skb_irq);
1316 void dev_kfree_skb_any(struct sk_buff *skb)
1318 if (in_irq() || irqs_disabled())
1319 dev_kfree_skb_irq(skb);
1323 EXPORT_SYMBOL(dev_kfree_skb_any);
1327 * netif_device_detach - mark device as removed
1328 * @dev: network device
1330 * Mark device as removed from system and therefore no longer available.
1332 void netif_device_detach(struct net_device *dev)
1334 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1335 netif_running(dev)) {
1336 netif_stop_queue(dev);
1339 EXPORT_SYMBOL(netif_device_detach);
1342 * netif_device_attach - mark device as attached
1343 * @dev: network device
1345 * Mark device as attached from system and restart if needed.
1347 void netif_device_attach(struct net_device *dev)
1349 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1350 netif_running(dev)) {
1351 netif_wake_queue(dev);
1352 __netdev_watchdog_up(dev);
1355 EXPORT_SYMBOL(netif_device_attach);
1359 * Invalidate hardware checksum when packet is to be mangled, and
1360 * complete checksum manually on outgoing path.
1362 int skb_checksum_help(struct sk_buff *skb)
1365 int ret = 0, offset;
1367 if (skb->ip_summed == CHECKSUM_COMPLETE)
1368 goto out_set_summed;
1370 if (unlikely(skb_shinfo(skb)->gso_size)) {
1371 /* Let GSO fix up the checksum. */
1372 goto out_set_summed;
1375 offset = skb->csum_start - skb_headroom(skb);
1376 BUG_ON(offset >= skb_headlen(skb));
1377 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1379 offset += skb->csum_offset;
1380 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1382 if (skb_cloned(skb) &&
1383 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1384 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1389 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1391 skb->ip_summed = CHECKSUM_NONE;
1397 * skb_gso_segment - Perform segmentation on skb.
1398 * @skb: buffer to segment
1399 * @features: features for the output path (see dev->features)
1401 * This function segments the given skb and returns a list of segments.
1403 * It may return NULL if the skb requires no segmentation. This is
1404 * only possible when GSO is used for verifying header integrity.
1406 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1408 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1409 struct packet_type *ptype;
1410 __be16 type = skb->protocol;
1413 BUG_ON(skb_shinfo(skb)->frag_list);
1415 skb_reset_mac_header(skb);
1416 skb->mac_len = skb->network_header - skb->mac_header;
1417 __skb_pull(skb, skb->mac_len);
1419 if (WARN_ON(skb->ip_summed != CHECKSUM_PARTIAL)) {
1420 if (skb_header_cloned(skb) &&
1421 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1422 return ERR_PTR(err);
1426 list_for_each_entry_rcu(ptype,
1427 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1428 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1429 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1430 err = ptype->gso_send_check(skb);
1431 segs = ERR_PTR(err);
1432 if (err || skb_gso_ok(skb, features))
1434 __skb_push(skb, (skb->data -
1435 skb_network_header(skb)));
1437 segs = ptype->gso_segment(skb, features);
1443 __skb_push(skb, skb->data - skb_mac_header(skb));
1448 EXPORT_SYMBOL(skb_gso_segment);
1450 /* Take action when hardware reception checksum errors are detected. */
1452 void netdev_rx_csum_fault(struct net_device *dev)
1454 if (net_ratelimit()) {
1455 printk(KERN_ERR "%s: hw csum failure.\n",
1456 dev ? dev->name : "<unknown>");
1460 EXPORT_SYMBOL(netdev_rx_csum_fault);
1463 /* Actually, we should eliminate this check as soon as we know, that:
1464 * 1. IOMMU is present and allows to map all the memory.
1465 * 2. No high memory really exists on this machine.
1468 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1470 #ifdef CONFIG_HIGHMEM
1473 if (dev->features & NETIF_F_HIGHDMA)
1476 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1477 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1485 void (*destructor)(struct sk_buff *skb);
1488 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1490 static void dev_gso_skb_destructor(struct sk_buff *skb)
1492 struct dev_gso_cb *cb;
1495 struct sk_buff *nskb = skb->next;
1497 skb->next = nskb->next;
1500 } while (skb->next);
1502 cb = DEV_GSO_CB(skb);
1504 cb->destructor(skb);
1508 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1509 * @skb: buffer to segment
1511 * This function segments the given skb and stores the list of segments
1514 static int dev_gso_segment(struct sk_buff *skb)
1516 struct net_device *dev = skb->dev;
1517 struct sk_buff *segs;
1518 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1521 segs = skb_gso_segment(skb, features);
1523 /* Verifying header integrity only. */
1527 if (unlikely(IS_ERR(segs)))
1528 return PTR_ERR(segs);
1531 DEV_GSO_CB(skb)->destructor = skb->destructor;
1532 skb->destructor = dev_gso_skb_destructor;
1537 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev)
1539 if (likely(!skb->next)) {
1540 if (!list_empty(&ptype_all))
1541 dev_queue_xmit_nit(skb, dev);
1543 if (netif_needs_gso(dev, skb)) {
1544 if (unlikely(dev_gso_segment(skb)))
1550 return dev->hard_start_xmit(skb, dev);
1555 struct sk_buff *nskb = skb->next;
1558 skb->next = nskb->next;
1560 rc = dev->hard_start_xmit(nskb, dev);
1562 nskb->next = skb->next;
1566 if (unlikely((netif_queue_stopped(dev) ||
1567 netif_subqueue_stopped(dev, skb)) &&
1569 return NETDEV_TX_BUSY;
1570 } while (skb->next);
1572 skb->destructor = DEV_GSO_CB(skb)->destructor;
1580 * dev_queue_xmit - transmit a buffer
1581 * @skb: buffer to transmit
1583 * Queue a buffer for transmission to a network device. The caller must
1584 * have set the device and priority and built the buffer before calling
1585 * this function. The function can be called from an interrupt.
1587 * A negative errno code is returned on a failure. A success does not
1588 * guarantee the frame will be transmitted as it may be dropped due
1589 * to congestion or traffic shaping.
1591 * -----------------------------------------------------------------------------------
1592 * I notice this method can also return errors from the queue disciplines,
1593 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1596 * Regardless of the return value, the skb is consumed, so it is currently
1597 * difficult to retry a send to this method. (You can bump the ref count
1598 * before sending to hold a reference for retry if you are careful.)
1600 * When calling this method, interrupts MUST be enabled. This is because
1601 * the BH enable code must have IRQs enabled so that it will not deadlock.
1605 int dev_queue_xmit(struct sk_buff *skb)
1607 struct net_device *dev = skb->dev;
1611 /* GSO will handle the following emulations directly. */
1612 if (netif_needs_gso(dev, skb))
1615 if (skb_shinfo(skb)->frag_list &&
1616 !(dev->features & NETIF_F_FRAGLIST) &&
1617 __skb_linearize(skb))
1620 /* Fragmented skb is linearized if device does not support SG,
1621 * or if at least one of fragments is in highmem and device
1622 * does not support DMA from it.
1624 if (skb_shinfo(skb)->nr_frags &&
1625 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1626 __skb_linearize(skb))
1629 /* If packet is not checksummed and device does not support
1630 * checksumming for this protocol, complete checksumming here.
1632 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1633 skb_set_transport_header(skb, skb->csum_start -
1636 if (!(dev->features & NETIF_F_GEN_CSUM) &&
1637 !((dev->features & NETIF_F_IP_CSUM) &&
1638 skb->protocol == htons(ETH_P_IP)) &&
1639 !((dev->features & NETIF_F_IPV6_CSUM) &&
1640 skb->protocol == htons(ETH_P_IPV6)))
1641 if (skb_checksum_help(skb))
1646 spin_lock_prefetch(&dev->queue_lock);
1648 /* Disable soft irqs for various locks below. Also
1649 * stops preemption for RCU.
1653 /* Updates of qdisc are serialized by queue_lock.
1654 * The struct Qdisc which is pointed to by qdisc is now a
1655 * rcu structure - it may be accessed without acquiring
1656 * a lock (but the structure may be stale.) The freeing of the
1657 * qdisc will be deferred until it's known that there are no
1658 * more references to it.
1660 * If the qdisc has an enqueue function, we still need to
1661 * hold the queue_lock before calling it, since queue_lock
1662 * also serializes access to the device queue.
1665 q = rcu_dereference(dev->qdisc);
1666 #ifdef CONFIG_NET_CLS_ACT
1667 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1670 /* Grab device queue */
1671 spin_lock(&dev->queue_lock);
1674 /* reset queue_mapping to zero */
1675 skb_set_queue_mapping(skb, 0);
1676 rc = q->enqueue(skb, q);
1678 spin_unlock(&dev->queue_lock);
1680 rc = rc == NET_XMIT_BYPASS ? NET_XMIT_SUCCESS : rc;
1683 spin_unlock(&dev->queue_lock);
1686 /* The device has no queue. Common case for software devices:
1687 loopback, all the sorts of tunnels...
1689 Really, it is unlikely that netif_tx_lock protection is necessary
1690 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1692 However, it is possible, that they rely on protection
1695 Check this and shot the lock. It is not prone from deadlocks.
1696 Either shot noqueue qdisc, it is even simpler 8)
1698 if (dev->flags & IFF_UP) {
1699 int cpu = smp_processor_id(); /* ok because BHs are off */
1701 if (dev->xmit_lock_owner != cpu) {
1703 HARD_TX_LOCK(dev, cpu);
1705 if (!netif_queue_stopped(dev) &&
1706 !netif_subqueue_stopped(dev, skb)) {
1708 if (!dev_hard_start_xmit(skb, dev)) {
1709 HARD_TX_UNLOCK(dev);
1713 HARD_TX_UNLOCK(dev);
1714 if (net_ratelimit())
1715 printk(KERN_CRIT "Virtual device %s asks to "
1716 "queue packet!\n", dev->name);
1718 /* Recursion is detected! It is possible,
1720 if (net_ratelimit())
1721 printk(KERN_CRIT "Dead loop on virtual device "
1722 "%s, fix it urgently!\n", dev->name);
1727 rcu_read_unlock_bh();
1733 rcu_read_unlock_bh();
1738 /*=======================================================================
1740 =======================================================================*/
1742 int netdev_max_backlog __read_mostly = 1000;
1743 int netdev_budget __read_mostly = 300;
1744 int weight_p __read_mostly = 64; /* old backlog weight */
1746 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1750 * netif_rx - post buffer to the network code
1751 * @skb: buffer to post
1753 * This function receives a packet from a device driver and queues it for
1754 * the upper (protocol) levels to process. It always succeeds. The buffer
1755 * may be dropped during processing for congestion control or by the
1759 * NET_RX_SUCCESS (no congestion)
1760 * NET_RX_DROP (packet was dropped)
1764 int netif_rx(struct sk_buff *skb)
1766 struct softnet_data *queue;
1767 unsigned long flags;
1769 /* if netpoll wants it, pretend we never saw it */
1770 if (netpoll_rx(skb))
1773 if (!skb->tstamp.tv64)
1777 * The code is rearranged so that the path is the most
1778 * short when CPU is congested, but is still operating.
1780 local_irq_save(flags);
1781 queue = &__get_cpu_var(softnet_data);
1783 __get_cpu_var(netdev_rx_stat).total++;
1784 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1785 if (queue->input_pkt_queue.qlen) {
1788 __skb_queue_tail(&queue->input_pkt_queue, skb);
1789 local_irq_restore(flags);
1790 return NET_RX_SUCCESS;
1793 napi_schedule(&queue->backlog);
1797 __get_cpu_var(netdev_rx_stat).dropped++;
1798 local_irq_restore(flags);
1804 int netif_rx_ni(struct sk_buff *skb)
1809 err = netif_rx(skb);
1810 if (local_softirq_pending())
1817 EXPORT_SYMBOL(netif_rx_ni);
1819 static inline struct net_device *skb_bond(struct sk_buff *skb)
1821 struct net_device *dev = skb->dev;
1824 if (skb_bond_should_drop(skb)) {
1828 skb->dev = dev->master;
1835 static void net_tx_action(struct softirq_action *h)
1837 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1839 if (sd->completion_queue) {
1840 struct sk_buff *clist;
1842 local_irq_disable();
1843 clist = sd->completion_queue;
1844 sd->completion_queue = NULL;
1848 struct sk_buff *skb = clist;
1849 clist = clist->next;
1851 BUG_TRAP(!atomic_read(&skb->users));
1856 if (sd->output_queue) {
1857 struct net_device *head;
1859 local_irq_disable();
1860 head = sd->output_queue;
1861 sd->output_queue = NULL;
1865 struct net_device *dev = head;
1866 head = head->next_sched;
1868 smp_mb__before_clear_bit();
1869 clear_bit(__LINK_STATE_SCHED, &dev->state);
1871 if (spin_trylock(&dev->queue_lock)) {
1873 spin_unlock(&dev->queue_lock);
1875 netif_schedule(dev);
1881 static inline int deliver_skb(struct sk_buff *skb,
1882 struct packet_type *pt_prev,
1883 struct net_device *orig_dev)
1885 atomic_inc(&skb->users);
1886 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1889 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
1890 /* These hooks defined here for ATM */
1892 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
1893 unsigned char *addr);
1894 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
1897 * If bridge module is loaded call bridging hook.
1898 * returns NULL if packet was consumed.
1900 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
1901 struct sk_buff *skb) __read_mostly;
1902 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
1903 struct packet_type **pt_prev, int *ret,
1904 struct net_device *orig_dev)
1906 struct net_bridge_port *port;
1908 if (skb->pkt_type == PACKET_LOOPBACK ||
1909 (port = rcu_dereference(skb->dev->br_port)) == NULL)
1913 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1917 return br_handle_frame_hook(port, skb);
1920 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
1923 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
1924 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
1925 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
1927 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
1928 struct packet_type **pt_prev,
1930 struct net_device *orig_dev)
1932 if (skb->dev->macvlan_port == NULL)
1936 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1939 return macvlan_handle_frame_hook(skb);
1942 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
1945 #ifdef CONFIG_NET_CLS_ACT
1946 /* TODO: Maybe we should just force sch_ingress to be compiled in
1947 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
1948 * a compare and 2 stores extra right now if we dont have it on
1949 * but have CONFIG_NET_CLS_ACT
1950 * NOTE: This doesnt stop any functionality; if you dont have
1951 * the ingress scheduler, you just cant add policies on ingress.
1954 static int ing_filter(struct sk_buff *skb)
1957 struct net_device *dev = skb->dev;
1958 int result = TC_ACT_OK;
1959 u32 ttl = G_TC_RTTL(skb->tc_verd);
1961 if (MAX_RED_LOOP < ttl++) {
1963 "Redir loop detected Dropping packet (%d->%d)\n",
1964 skb->iif, dev->ifindex);
1968 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
1969 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
1971 spin_lock(&dev->ingress_lock);
1972 if ((q = dev->qdisc_ingress) != NULL)
1973 result = q->enqueue(skb, q);
1974 spin_unlock(&dev->ingress_lock);
1979 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
1980 struct packet_type **pt_prev,
1981 int *ret, struct net_device *orig_dev)
1983 if (!skb->dev->qdisc_ingress)
1987 *ret = deliver_skb(skb, *pt_prev, orig_dev);
1990 /* Huh? Why does turning on AF_PACKET affect this? */
1991 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
1994 switch (ing_filter(skb)) {
2008 * netif_receive_skb - process receive buffer from network
2009 * @skb: buffer to process
2011 * netif_receive_skb() is the main receive data processing function.
2012 * It always succeeds. The buffer may be dropped during processing
2013 * for congestion control or by the protocol layers.
2015 * This function may only be called from softirq context and interrupts
2016 * should be enabled.
2018 * Return values (usually ignored):
2019 * NET_RX_SUCCESS: no congestion
2020 * NET_RX_DROP: packet was dropped
2022 int netif_receive_skb(struct sk_buff *skb)
2024 struct packet_type *ptype, *pt_prev;
2025 struct net_device *orig_dev;
2026 int ret = NET_RX_DROP;
2029 /* if we've gotten here through NAPI, check netpoll */
2030 if (netpoll_receive_skb(skb))
2033 if (!skb->tstamp.tv64)
2037 skb->iif = skb->dev->ifindex;
2039 orig_dev = skb_bond(skb);
2044 __get_cpu_var(netdev_rx_stat).total++;
2046 skb_reset_network_header(skb);
2047 skb_reset_transport_header(skb);
2048 skb->mac_len = skb->network_header - skb->mac_header;
2054 #ifdef CONFIG_NET_CLS_ACT
2055 if (skb->tc_verd & TC_NCLS) {
2056 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2061 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2062 if (!ptype->dev || ptype->dev == skb->dev) {
2064 ret = deliver_skb(skb, pt_prev, orig_dev);
2069 #ifdef CONFIG_NET_CLS_ACT
2070 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2076 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2079 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2083 type = skb->protocol;
2084 list_for_each_entry_rcu(ptype,
2085 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2086 if (ptype->type == type &&
2087 (!ptype->dev || ptype->dev == skb->dev)) {
2089 ret = deliver_skb(skb, pt_prev, orig_dev);
2095 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2098 /* Jamal, now you will not able to escape explaining
2099 * me how you were going to use this. :-)
2109 static int process_backlog(struct napi_struct *napi, int quota)
2112 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2113 unsigned long start_time = jiffies;
2115 napi->weight = weight_p;
2117 struct sk_buff *skb;
2118 struct net_device *dev;
2120 local_irq_disable();
2121 skb = __skb_dequeue(&queue->input_pkt_queue);
2123 __napi_complete(napi);
2132 netif_receive_skb(skb);
2135 } while (++work < quota && jiffies == start_time);
2141 * __napi_schedule - schedule for receive
2142 * @n: entry to schedule
2144 * The entry's receive function will be scheduled to run
2146 void fastcall __napi_schedule(struct napi_struct *n)
2148 unsigned long flags;
2150 local_irq_save(flags);
2151 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2152 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2153 local_irq_restore(flags);
2155 EXPORT_SYMBOL(__napi_schedule);
2158 static void net_rx_action(struct softirq_action *h)
2160 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2161 unsigned long start_time = jiffies;
2162 int budget = netdev_budget;
2165 local_irq_disable();
2167 while (!list_empty(list)) {
2168 struct napi_struct *n;
2171 /* If softirq window is exhuasted then punt.
2173 * Note that this is a slight policy change from the
2174 * previous NAPI code, which would allow up to 2
2175 * jiffies to pass before breaking out. The test
2176 * used to be "jiffies - start_time > 1".
2178 if (unlikely(budget <= 0 || jiffies != start_time))
2183 /* Even though interrupts have been re-enabled, this
2184 * access is safe because interrupts can only add new
2185 * entries to the tail of this list, and only ->poll()
2186 * calls can remove this head entry from the list.
2188 n = list_entry(list->next, struct napi_struct, poll_list);
2190 have = netpoll_poll_lock(n);
2194 /* This NAPI_STATE_SCHED test is for avoiding a race
2195 * with netpoll's poll_napi(). Only the entity which
2196 * obtains the lock and sees NAPI_STATE_SCHED set will
2197 * actually make the ->poll() call. Therefore we avoid
2198 * accidently calling ->poll() when NAPI is not scheduled.
2201 if (test_bit(NAPI_STATE_SCHED, &n->state))
2202 work = n->poll(n, weight);
2204 WARN_ON_ONCE(work > weight);
2208 local_irq_disable();
2210 /* Drivers must not modify the NAPI state if they
2211 * consume the entire weight. In such cases this code
2212 * still "owns" the NAPI instance and therefore can
2213 * move the instance around on the list at-will.
2215 if (unlikely(work == weight)) {
2216 if (unlikely(napi_disable_pending(n)))
2219 list_move_tail(&n->poll_list, list);
2222 netpoll_poll_unlock(have);
2227 #ifdef CONFIG_NET_DMA
2229 * There may not be any more sk_buffs coming right now, so push
2230 * any pending DMA copies to hardware
2232 if (!cpus_empty(net_dma.channel_mask)) {
2234 for_each_cpu_mask(chan_idx, net_dma.channel_mask) {
2235 struct dma_chan *chan = net_dma.channels[chan_idx];
2237 dma_async_memcpy_issue_pending(chan);
2245 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2246 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2250 static gifconf_func_t * gifconf_list [NPROTO];
2253 * register_gifconf - register a SIOCGIF handler
2254 * @family: Address family
2255 * @gifconf: Function handler
2257 * Register protocol dependent address dumping routines. The handler
2258 * that is passed must not be freed or reused until it has been replaced
2259 * by another handler.
2261 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2263 if (family >= NPROTO)
2265 gifconf_list[family] = gifconf;
2271 * Map an interface index to its name (SIOCGIFNAME)
2275 * We need this ioctl for efficient implementation of the
2276 * if_indextoname() function required by the IPv6 API. Without
2277 * it, we would have to search all the interfaces to find a
2281 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2283 struct net_device *dev;
2287 * Fetch the caller's info block.
2290 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2293 read_lock(&dev_base_lock);
2294 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2296 read_unlock(&dev_base_lock);
2300 strcpy(ifr.ifr_name, dev->name);
2301 read_unlock(&dev_base_lock);
2303 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2309 * Perform a SIOCGIFCONF call. This structure will change
2310 * size eventually, and there is nothing I can do about it.
2311 * Thus we will need a 'compatibility mode'.
2314 static int dev_ifconf(struct net *net, char __user *arg)
2317 struct net_device *dev;
2324 * Fetch the caller's info block.
2327 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2334 * Loop over the interfaces, and write an info block for each.
2338 for_each_netdev(net, dev) {
2339 for (i = 0; i < NPROTO; i++) {
2340 if (gifconf_list[i]) {
2343 done = gifconf_list[i](dev, NULL, 0);
2345 done = gifconf_list[i](dev, pos + total,
2355 * All done. Write the updated control block back to the caller.
2357 ifc.ifc_len = total;
2360 * Both BSD and Solaris return 0 here, so we do too.
2362 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2365 #ifdef CONFIG_PROC_FS
2367 * This is invoked by the /proc filesystem handler to display a device
2370 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2372 struct net *net = seq_file_net(seq);
2374 struct net_device *dev;
2376 read_lock(&dev_base_lock);
2378 return SEQ_START_TOKEN;
2381 for_each_netdev(net, dev)
2388 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2390 struct net *net = seq_file_net(seq);
2392 return v == SEQ_START_TOKEN ?
2393 first_net_device(net) : next_net_device((struct net_device *)v);
2396 void dev_seq_stop(struct seq_file *seq, void *v)
2398 read_unlock(&dev_base_lock);
2401 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2403 struct net_device_stats *stats = dev->get_stats(dev);
2405 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2406 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2407 dev->name, stats->rx_bytes, stats->rx_packets,
2409 stats->rx_dropped + stats->rx_missed_errors,
2410 stats->rx_fifo_errors,
2411 stats->rx_length_errors + stats->rx_over_errors +
2412 stats->rx_crc_errors + stats->rx_frame_errors,
2413 stats->rx_compressed, stats->multicast,
2414 stats->tx_bytes, stats->tx_packets,
2415 stats->tx_errors, stats->tx_dropped,
2416 stats->tx_fifo_errors, stats->collisions,
2417 stats->tx_carrier_errors +
2418 stats->tx_aborted_errors +
2419 stats->tx_window_errors +
2420 stats->tx_heartbeat_errors,
2421 stats->tx_compressed);
2425 * Called from the PROCfs module. This now uses the new arbitrary sized
2426 * /proc/net interface to create /proc/net/dev
2428 static int dev_seq_show(struct seq_file *seq, void *v)
2430 if (v == SEQ_START_TOKEN)
2431 seq_puts(seq, "Inter-| Receive "
2433 " face |bytes packets errs drop fifo frame "
2434 "compressed multicast|bytes packets errs "
2435 "drop fifo colls carrier compressed\n");
2437 dev_seq_printf_stats(seq, v);
2441 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2443 struct netif_rx_stats *rc = NULL;
2445 while (*pos < NR_CPUS)
2446 if (cpu_online(*pos)) {
2447 rc = &per_cpu(netdev_rx_stat, *pos);
2454 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
2456 return softnet_get_online(pos);
2459 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2462 return softnet_get_online(pos);
2465 static void softnet_seq_stop(struct seq_file *seq, void *v)
2469 static int softnet_seq_show(struct seq_file *seq, void *v)
2471 struct netif_rx_stats *s = v;
2473 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
2474 s->total, s->dropped, s->time_squeeze, 0,
2475 0, 0, 0, 0, /* was fastroute */
2480 static const struct seq_operations dev_seq_ops = {
2481 .start = dev_seq_start,
2482 .next = dev_seq_next,
2483 .stop = dev_seq_stop,
2484 .show = dev_seq_show,
2487 static int dev_seq_open(struct inode *inode, struct file *file)
2489 return seq_open_net(inode, file, &dev_seq_ops,
2490 sizeof(struct seq_net_private));
2493 static const struct file_operations dev_seq_fops = {
2494 .owner = THIS_MODULE,
2495 .open = dev_seq_open,
2497 .llseek = seq_lseek,
2498 .release = seq_release_net,
2501 static const struct seq_operations softnet_seq_ops = {
2502 .start = softnet_seq_start,
2503 .next = softnet_seq_next,
2504 .stop = softnet_seq_stop,
2505 .show = softnet_seq_show,
2508 static int softnet_seq_open(struct inode *inode, struct file *file)
2510 return seq_open(file, &softnet_seq_ops);
2513 static const struct file_operations softnet_seq_fops = {
2514 .owner = THIS_MODULE,
2515 .open = softnet_seq_open,
2517 .llseek = seq_lseek,
2518 .release = seq_release,
2521 static void *ptype_get_idx(loff_t pos)
2523 struct packet_type *pt = NULL;
2527 list_for_each_entry_rcu(pt, &ptype_all, list) {
2533 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
2534 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
2543 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
2546 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
2549 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2551 struct packet_type *pt;
2552 struct list_head *nxt;
2556 if (v == SEQ_START_TOKEN)
2557 return ptype_get_idx(0);
2560 nxt = pt->list.next;
2561 if (pt->type == htons(ETH_P_ALL)) {
2562 if (nxt != &ptype_all)
2565 nxt = ptype_base[0].next;
2567 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
2569 while (nxt == &ptype_base[hash]) {
2570 if (++hash >= PTYPE_HASH_SIZE)
2572 nxt = ptype_base[hash].next;
2575 return list_entry(nxt, struct packet_type, list);
2578 static void ptype_seq_stop(struct seq_file *seq, void *v)
2583 static void ptype_seq_decode(struct seq_file *seq, void *sym)
2585 #ifdef CONFIG_KALLSYMS
2586 unsigned long offset = 0, symsize;
2587 const char *symname;
2591 symname = kallsyms_lookup((unsigned long)sym, &symsize, &offset,
2598 modname = delim = "";
2599 seq_printf(seq, "%s%s%s%s+0x%lx", delim, modname, delim,
2605 seq_printf(seq, "[%p]", sym);
2608 static int ptype_seq_show(struct seq_file *seq, void *v)
2610 struct packet_type *pt = v;
2612 if (v == SEQ_START_TOKEN)
2613 seq_puts(seq, "Type Device Function\n");
2615 if (pt->type == htons(ETH_P_ALL))
2616 seq_puts(seq, "ALL ");
2618 seq_printf(seq, "%04x", ntohs(pt->type));
2620 seq_printf(seq, " %-8s ",
2621 pt->dev ? pt->dev->name : "");
2622 ptype_seq_decode(seq, pt->func);
2623 seq_putc(seq, '\n');
2629 static const struct seq_operations ptype_seq_ops = {
2630 .start = ptype_seq_start,
2631 .next = ptype_seq_next,
2632 .stop = ptype_seq_stop,
2633 .show = ptype_seq_show,
2636 static int ptype_seq_open(struct inode *inode, struct file *file)
2638 return seq_open(file, &ptype_seq_ops);
2641 static const struct file_operations ptype_seq_fops = {
2642 .owner = THIS_MODULE,
2643 .open = ptype_seq_open,
2645 .llseek = seq_lseek,
2646 .release = seq_release,
2650 static int __net_init dev_proc_net_init(struct net *net)
2654 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
2656 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
2658 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
2661 if (wext_proc_init(net))
2667 proc_net_remove(net, "ptype");
2669 proc_net_remove(net, "softnet_stat");
2671 proc_net_remove(net, "dev");
2675 static void __net_exit dev_proc_net_exit(struct net *net)
2677 wext_proc_exit(net);
2679 proc_net_remove(net, "ptype");
2680 proc_net_remove(net, "softnet_stat");
2681 proc_net_remove(net, "dev");
2684 static struct pernet_operations __net_initdata dev_proc_ops = {
2685 .init = dev_proc_net_init,
2686 .exit = dev_proc_net_exit,
2689 static int __init dev_proc_init(void)
2691 return register_pernet_subsys(&dev_proc_ops);
2694 #define dev_proc_init() 0
2695 #endif /* CONFIG_PROC_FS */
2699 * netdev_set_master - set up master/slave pair
2700 * @slave: slave device
2701 * @master: new master device
2703 * Changes the master device of the slave. Pass %NULL to break the
2704 * bonding. The caller must hold the RTNL semaphore. On a failure
2705 * a negative errno code is returned. On success the reference counts
2706 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
2707 * function returns zero.
2709 int netdev_set_master(struct net_device *slave, struct net_device *master)
2711 struct net_device *old = slave->master;
2721 slave->master = master;
2729 slave->flags |= IFF_SLAVE;
2731 slave->flags &= ~IFF_SLAVE;
2733 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
2737 static void __dev_set_promiscuity(struct net_device *dev, int inc)
2739 unsigned short old_flags = dev->flags;
2743 if ((dev->promiscuity += inc) == 0)
2744 dev->flags &= ~IFF_PROMISC;
2746 dev->flags |= IFF_PROMISC;
2747 if (dev->flags != old_flags) {
2748 printk(KERN_INFO "device %s %s promiscuous mode\n",
2749 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
2751 audit_log(current->audit_context, GFP_ATOMIC,
2752 AUDIT_ANOM_PROMISCUOUS,
2753 "dev=%s prom=%d old_prom=%d auid=%u",
2754 dev->name, (dev->flags & IFF_PROMISC),
2755 (old_flags & IFF_PROMISC),
2756 audit_get_loginuid(current->audit_context));
2758 if (dev->change_rx_flags)
2759 dev->change_rx_flags(dev, IFF_PROMISC);
2764 * dev_set_promiscuity - update promiscuity count on a device
2768 * Add or remove promiscuity from a device. While the count in the device
2769 * remains above zero the interface remains promiscuous. Once it hits zero
2770 * the device reverts back to normal filtering operation. A negative inc
2771 * value is used to drop promiscuity on the device.
2773 void dev_set_promiscuity(struct net_device *dev, int inc)
2775 unsigned short old_flags = dev->flags;
2777 __dev_set_promiscuity(dev, inc);
2778 if (dev->flags != old_flags)
2779 dev_set_rx_mode(dev);
2783 * dev_set_allmulti - update allmulti count on a device
2787 * Add or remove reception of all multicast frames to a device. While the
2788 * count in the device remains above zero the interface remains listening
2789 * to all interfaces. Once it hits zero the device reverts back to normal
2790 * filtering operation. A negative @inc value is used to drop the counter
2791 * when releasing a resource needing all multicasts.
2794 void dev_set_allmulti(struct net_device *dev, int inc)
2796 unsigned short old_flags = dev->flags;
2800 dev->flags |= IFF_ALLMULTI;
2801 if ((dev->allmulti += inc) == 0)
2802 dev->flags &= ~IFF_ALLMULTI;
2803 if (dev->flags ^ old_flags) {
2804 if (dev->change_rx_flags)
2805 dev->change_rx_flags(dev, IFF_ALLMULTI);
2806 dev_set_rx_mode(dev);
2811 * Upload unicast and multicast address lists to device and
2812 * configure RX filtering. When the device doesn't support unicast
2813 * filtering it is put in promiscuous mode while unicast addresses
2816 void __dev_set_rx_mode(struct net_device *dev)
2818 /* dev_open will call this function so the list will stay sane. */
2819 if (!(dev->flags&IFF_UP))
2822 if (!netif_device_present(dev))
2825 if (dev->set_rx_mode)
2826 dev->set_rx_mode(dev);
2828 /* Unicast addresses changes may only happen under the rtnl,
2829 * therefore calling __dev_set_promiscuity here is safe.
2831 if (dev->uc_count > 0 && !dev->uc_promisc) {
2832 __dev_set_promiscuity(dev, 1);
2833 dev->uc_promisc = 1;
2834 } else if (dev->uc_count == 0 && dev->uc_promisc) {
2835 __dev_set_promiscuity(dev, -1);
2836 dev->uc_promisc = 0;
2839 if (dev->set_multicast_list)
2840 dev->set_multicast_list(dev);
2844 void dev_set_rx_mode(struct net_device *dev)
2846 netif_tx_lock_bh(dev);
2847 __dev_set_rx_mode(dev);
2848 netif_tx_unlock_bh(dev);
2851 int __dev_addr_delete(struct dev_addr_list **list, int *count,
2852 void *addr, int alen, int glbl)
2854 struct dev_addr_list *da;
2856 for (; (da = *list) != NULL; list = &da->next) {
2857 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2858 alen == da->da_addrlen) {
2860 int old_glbl = da->da_gusers;
2877 int __dev_addr_add(struct dev_addr_list **list, int *count,
2878 void *addr, int alen, int glbl)
2880 struct dev_addr_list *da;
2882 for (da = *list; da != NULL; da = da->next) {
2883 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
2884 da->da_addrlen == alen) {
2886 int old_glbl = da->da_gusers;
2896 da = kmalloc(sizeof(*da), GFP_ATOMIC);
2899 memcpy(da->da_addr, addr, alen);
2900 da->da_addrlen = alen;
2902 da->da_gusers = glbl ? 1 : 0;
2910 * dev_unicast_delete - Release secondary unicast address.
2912 * @addr: address to delete
2913 * @alen: length of @addr
2915 * Release reference to a secondary unicast address and remove it
2916 * from the device if the reference count drops to zero.
2918 * The caller must hold the rtnl_mutex.
2920 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
2926 netif_tx_lock_bh(dev);
2927 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2929 __dev_set_rx_mode(dev);
2930 netif_tx_unlock_bh(dev);
2933 EXPORT_SYMBOL(dev_unicast_delete);
2936 * dev_unicast_add - add a secondary unicast address
2938 * @addr: address to delete
2939 * @alen: length of @addr
2941 * Add a secondary unicast address to the device or increase
2942 * the reference count if it already exists.
2944 * The caller must hold the rtnl_mutex.
2946 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
2952 netif_tx_lock_bh(dev);
2953 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
2955 __dev_set_rx_mode(dev);
2956 netif_tx_unlock_bh(dev);
2959 EXPORT_SYMBOL(dev_unicast_add);
2961 static void __dev_addr_discard(struct dev_addr_list **list)
2963 struct dev_addr_list *tmp;
2965 while (*list != NULL) {
2968 if (tmp->da_users > tmp->da_gusers)
2969 printk("__dev_addr_discard: address leakage! "
2970 "da_users=%d\n", tmp->da_users);
2975 static void dev_addr_discard(struct net_device *dev)
2977 netif_tx_lock_bh(dev);
2979 __dev_addr_discard(&dev->uc_list);
2982 __dev_addr_discard(&dev->mc_list);
2985 netif_tx_unlock_bh(dev);
2988 unsigned dev_get_flags(const struct net_device *dev)
2992 flags = (dev->flags & ~(IFF_PROMISC |
2997 (dev->gflags & (IFF_PROMISC |
3000 if (netif_running(dev)) {
3001 if (netif_oper_up(dev))
3002 flags |= IFF_RUNNING;
3003 if (netif_carrier_ok(dev))
3004 flags |= IFF_LOWER_UP;
3005 if (netif_dormant(dev))
3006 flags |= IFF_DORMANT;
3012 int dev_change_flags(struct net_device *dev, unsigned flags)
3015 int old_flags = dev->flags;
3020 * Set the flags on our device.
3023 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3024 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3026 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3030 * Load in the correct multicast list now the flags have changed.
3033 if (dev->change_rx_flags && (dev->flags ^ flags) & IFF_MULTICAST)
3034 dev->change_rx_flags(dev, IFF_MULTICAST);
3036 dev_set_rx_mode(dev);
3039 * Have we downed the interface. We handle IFF_UP ourselves
3040 * according to user attempts to set it, rather than blindly
3045 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3046 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3049 dev_set_rx_mode(dev);
3052 if (dev->flags & IFF_UP &&
3053 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3055 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3057 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3058 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3059 dev->gflags ^= IFF_PROMISC;
3060 dev_set_promiscuity(dev, inc);
3063 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3064 is important. Some (broken) drivers set IFF_PROMISC, when
3065 IFF_ALLMULTI is requested not asking us and not reporting.
3067 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3068 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3069 dev->gflags ^= IFF_ALLMULTI;
3070 dev_set_allmulti(dev, inc);
3073 /* Exclude state transition flags, already notified */
3074 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3076 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3081 int dev_set_mtu(struct net_device *dev, int new_mtu)
3085 if (new_mtu == dev->mtu)
3088 /* MTU must be positive. */
3092 if (!netif_device_present(dev))
3096 if (dev->change_mtu)
3097 err = dev->change_mtu(dev, new_mtu);
3100 if (!err && dev->flags & IFF_UP)
3101 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3105 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3109 if (!dev->set_mac_address)
3111 if (sa->sa_family != dev->type)
3113 if (!netif_device_present(dev))
3115 err = dev->set_mac_address(dev, sa);
3117 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3122 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3124 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3127 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3133 case SIOCGIFFLAGS: /* Get interface flags */
3134 ifr->ifr_flags = dev_get_flags(dev);
3137 case SIOCGIFMETRIC: /* Get the metric on the interface
3138 (currently unused) */
3139 ifr->ifr_metric = 0;
3142 case SIOCGIFMTU: /* Get the MTU of a device */
3143 ifr->ifr_mtu = dev->mtu;
3148 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3150 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3151 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3152 ifr->ifr_hwaddr.sa_family = dev->type;
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 ifr->ifr_ifindex = dev->ifindex;
3173 ifr->ifr_qlen = dev->tx_queue_len;
3177 /* dev_ioctl() should ensure this case
3189 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3191 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3194 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3200 case SIOCSIFFLAGS: /* Set interface flags */
3201 return dev_change_flags(dev, ifr->ifr_flags);
3203 case SIOCSIFMETRIC: /* Set the metric on the interface
3204 (currently unused) */
3207 case SIOCSIFMTU: /* Set the MTU of a device */
3208 return dev_set_mtu(dev, ifr->ifr_mtu);
3211 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3213 case SIOCSIFHWBROADCAST:
3214 if (ifr->ifr_hwaddr.sa_family != dev->type)
3216 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3217 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3218 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3222 if (dev->set_config) {
3223 if (!netif_device_present(dev))
3225 return dev->set_config(dev, &ifr->ifr_map);
3230 if (!dev->set_multicast_list ||
3231 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3233 if (!netif_device_present(dev))
3235 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3239 if (!dev->set_multicast_list ||
3240 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3242 if (!netif_device_present(dev))
3244 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3248 if (ifr->ifr_qlen < 0)
3250 dev->tx_queue_len = ifr->ifr_qlen;
3254 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3255 return dev_change_name(dev, ifr->ifr_newname);
3258 * Unknown or private ioctl
3262 if ((cmd >= SIOCDEVPRIVATE &&
3263 cmd <= SIOCDEVPRIVATE + 15) ||
3264 cmd == SIOCBONDENSLAVE ||
3265 cmd == SIOCBONDRELEASE ||
3266 cmd == SIOCBONDSETHWADDR ||
3267 cmd == SIOCBONDSLAVEINFOQUERY ||
3268 cmd == SIOCBONDINFOQUERY ||
3269 cmd == SIOCBONDCHANGEACTIVE ||
3270 cmd == SIOCGMIIPHY ||
3271 cmd == SIOCGMIIREG ||
3272 cmd == SIOCSMIIREG ||
3273 cmd == SIOCBRADDIF ||
3274 cmd == SIOCBRDELIF ||
3275 cmd == SIOCWANDEV) {
3277 if (dev->do_ioctl) {
3278 if (netif_device_present(dev))
3279 err = dev->do_ioctl(dev, ifr,
3292 * This function handles all "interface"-type I/O control requests. The actual
3293 * 'doing' part of this is dev_ifsioc above.
3297 * dev_ioctl - network device ioctl
3298 * @net: the applicable net namespace
3299 * @cmd: command to issue
3300 * @arg: pointer to a struct ifreq in user space
3302 * Issue ioctl functions to devices. This is normally called by the
3303 * user space syscall interfaces but can sometimes be useful for
3304 * other purposes. The return value is the return from the syscall if
3305 * positive or a negative errno code on error.
3308 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
3314 /* One special case: SIOCGIFCONF takes ifconf argument
3315 and requires shared lock, because it sleeps writing
3319 if (cmd == SIOCGIFCONF) {
3321 ret = dev_ifconf(net, (char __user *) arg);
3325 if (cmd == SIOCGIFNAME)
3326 return dev_ifname(net, (struct ifreq __user *)arg);
3328 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3331 ifr.ifr_name[IFNAMSIZ-1] = 0;
3333 colon = strchr(ifr.ifr_name, ':');
3338 * See which interface the caller is talking about.
3343 * These ioctl calls:
3344 * - can be done by all.
3345 * - atomic and do not require locking.
3356 dev_load(net, ifr.ifr_name);
3357 read_lock(&dev_base_lock);
3358 ret = dev_ifsioc_locked(net, &ifr, cmd);
3359 read_unlock(&dev_base_lock);
3363 if (copy_to_user(arg, &ifr,
3364 sizeof(struct ifreq)))
3370 dev_load(net, ifr.ifr_name);
3372 ret = dev_ethtool(net, &ifr);
3377 if (copy_to_user(arg, &ifr,
3378 sizeof(struct ifreq)))
3384 * These ioctl calls:
3385 * - require superuser power.
3386 * - require strict serialization.
3392 if (!capable(CAP_NET_ADMIN))
3394 dev_load(net, ifr.ifr_name);
3396 ret = dev_ifsioc(net, &ifr, cmd);
3401 if (copy_to_user(arg, &ifr,
3402 sizeof(struct ifreq)))
3408 * These ioctl calls:
3409 * - require superuser power.
3410 * - require strict serialization.
3411 * - do not return a value
3421 case SIOCSIFHWBROADCAST:
3424 case SIOCBONDENSLAVE:
3425 case SIOCBONDRELEASE:
3426 case SIOCBONDSETHWADDR:
3427 case SIOCBONDCHANGEACTIVE:
3430 if (!capable(CAP_NET_ADMIN))
3433 case SIOCBONDSLAVEINFOQUERY:
3434 case SIOCBONDINFOQUERY:
3435 dev_load(net, ifr.ifr_name);
3437 ret = dev_ifsioc(net, &ifr, cmd);
3442 /* Get the per device memory space. We can add this but
3443 * currently do not support it */
3445 /* Set the per device memory buffer space.
3446 * Not applicable in our case */
3451 * Unknown or private ioctl.
3454 if (cmd == SIOCWANDEV ||
3455 (cmd >= SIOCDEVPRIVATE &&
3456 cmd <= SIOCDEVPRIVATE + 15)) {
3457 dev_load(net, ifr.ifr_name);
3459 ret = dev_ifsioc(net, &ifr, cmd);
3461 if (!ret && copy_to_user(arg, &ifr,
3462 sizeof(struct ifreq)))
3466 /* Take care of Wireless Extensions */
3467 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
3468 return wext_handle_ioctl(net, &ifr, cmd, arg);
3475 * dev_new_index - allocate an ifindex
3476 * @net: the applicable net namespace
3478 * Returns a suitable unique value for a new device interface
3479 * number. The caller must hold the rtnl semaphore or the
3480 * dev_base_lock to be sure it remains unique.
3482 static int dev_new_index(struct net *net)
3488 if (!__dev_get_by_index(net, ifindex))
3493 /* Delayed registration/unregisteration */
3494 static DEFINE_SPINLOCK(net_todo_list_lock);
3495 static LIST_HEAD(net_todo_list);
3497 static void net_set_todo(struct net_device *dev)
3499 spin_lock(&net_todo_list_lock);
3500 list_add_tail(&dev->todo_list, &net_todo_list);
3501 spin_unlock(&net_todo_list_lock);
3504 static void rollback_registered(struct net_device *dev)
3506 BUG_ON(dev_boot_phase);
3509 /* Some devices call without registering for initialization unwind. */
3510 if (dev->reg_state == NETREG_UNINITIALIZED) {
3511 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
3512 "was registered\n", dev->name, dev);
3518 BUG_ON(dev->reg_state != NETREG_REGISTERED);
3520 /* If device is running, close it first. */
3523 /* And unlink it from device chain. */
3524 unlist_netdevice(dev);
3526 dev->reg_state = NETREG_UNREGISTERING;
3530 /* Shutdown queueing discipline. */
3534 /* Notify protocols, that we are about to destroy
3535 this device. They should clean all the things.
3537 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3540 * Flush the unicast and multicast chains
3542 dev_addr_discard(dev);
3547 /* Notifier chain MUST detach us from master device. */
3548 BUG_TRAP(!dev->master);
3550 /* Remove entries from kobject tree */
3551 netdev_unregister_kobject(dev);
3559 * register_netdevice - register a network device
3560 * @dev: device to register
3562 * Take a completed network device structure and add it to the kernel
3563 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3564 * chain. 0 is returned on success. A negative errno code is returned
3565 * on a failure to set up the device, or if the name is a duplicate.
3567 * Callers must hold the rtnl semaphore. You may want
3568 * register_netdev() instead of this.
3571 * The locking appears insufficient to guarantee two parallel registers
3572 * will not get the same name.
3575 int register_netdevice(struct net_device *dev)
3577 struct hlist_head *head;
3578 struct hlist_node *p;
3582 BUG_ON(dev_boot_phase);
3587 /* When net_device's are persistent, this will be fatal. */
3588 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
3589 BUG_ON(!dev->nd_net);
3592 spin_lock_init(&dev->queue_lock);
3593 spin_lock_init(&dev->_xmit_lock);
3594 netdev_set_lockdep_class(&dev->_xmit_lock, dev->type);
3595 dev->xmit_lock_owner = -1;
3596 spin_lock_init(&dev->ingress_lock);
3600 /* Init, if this function is available */
3602 ret = dev->init(dev);
3610 if (!dev_valid_name(dev->name)) {
3615 dev->ifindex = dev_new_index(net);
3616 if (dev->iflink == -1)
3617 dev->iflink = dev->ifindex;
3619 /* Check for existence of name */
3620 head = dev_name_hash(net, dev->name);
3621 hlist_for_each(p, head) {
3622 struct net_device *d
3623 = hlist_entry(p, struct net_device, name_hlist);
3624 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
3630 /* Fix illegal checksum combinations */
3631 if ((dev->features & NETIF_F_HW_CSUM) &&
3632 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3633 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
3635 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
3638 if ((dev->features & NETIF_F_NO_CSUM) &&
3639 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
3640 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
3642 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
3646 /* Fix illegal SG+CSUM combinations. */
3647 if ((dev->features & NETIF_F_SG) &&
3648 !(dev->features & NETIF_F_ALL_CSUM)) {
3649 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no checksum feature.\n",
3651 dev->features &= ~NETIF_F_SG;
3654 /* TSO requires that SG is present as well. */
3655 if ((dev->features & NETIF_F_TSO) &&
3656 !(dev->features & NETIF_F_SG)) {
3657 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no SG feature.\n",
3659 dev->features &= ~NETIF_F_TSO;
3661 if (dev->features & NETIF_F_UFO) {
3662 if (!(dev->features & NETIF_F_HW_CSUM)) {
3663 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3664 "NETIF_F_HW_CSUM feature.\n",
3666 dev->features &= ~NETIF_F_UFO;
3668 if (!(dev->features & NETIF_F_SG)) {
3669 printk(KERN_ERR "%s: Dropping NETIF_F_UFO since no "
3670 "NETIF_F_SG feature.\n",
3672 dev->features &= ~NETIF_F_UFO;
3676 ret = netdev_register_kobject(dev);
3679 dev->reg_state = NETREG_REGISTERED;
3682 * Default initial state at registry is that the
3683 * device is present.
3686 set_bit(__LINK_STATE_PRESENT, &dev->state);
3688 dev_init_scheduler(dev);
3690 list_netdevice(dev);
3692 /* Notify protocols, that a new device appeared. */
3693 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
3694 ret = notifier_to_errno(ret);
3696 rollback_registered(dev);
3697 dev->reg_state = NETREG_UNREGISTERED;
3710 * register_netdev - register a network device
3711 * @dev: device to register
3713 * Take a completed network device structure and add it to the kernel
3714 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
3715 * chain. 0 is returned on success. A negative errno code is returned
3716 * on a failure to set up the device, or if the name is a duplicate.
3718 * This is a wrapper around register_netdevice that takes the rtnl semaphore
3719 * and expands the device name if you passed a format string to
3722 int register_netdev(struct net_device *dev)
3729 * If the name is a format string the caller wants us to do a
3732 if (strchr(dev->name, '%')) {
3733 err = dev_alloc_name(dev, dev->name);
3738 err = register_netdevice(dev);
3743 EXPORT_SYMBOL(register_netdev);
3746 * netdev_wait_allrefs - wait until all references are gone.
3748 * This is called when unregistering network devices.
3750 * Any protocol or device that holds a reference should register
3751 * for netdevice notification, and cleanup and put back the
3752 * reference if they receive an UNREGISTER event.
3753 * We can get stuck here if buggy protocols don't correctly
3756 static void netdev_wait_allrefs(struct net_device *dev)
3758 unsigned long rebroadcast_time, warning_time;
3760 rebroadcast_time = warning_time = jiffies;
3761 while (atomic_read(&dev->refcnt) != 0) {
3762 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
3765 /* Rebroadcast unregister notification */
3766 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
3768 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
3770 /* We must not have linkwatch events
3771 * pending on unregister. If this
3772 * happens, we simply run the queue
3773 * unscheduled, resulting in a noop
3776 linkwatch_run_queue();
3781 rebroadcast_time = jiffies;
3786 if (time_after(jiffies, warning_time + 10 * HZ)) {
3787 printk(KERN_EMERG "unregister_netdevice: "
3788 "waiting for %s to become free. Usage "
3790 dev->name, atomic_read(&dev->refcnt));
3791 warning_time = jiffies;
3800 * register_netdevice(x1);
3801 * register_netdevice(x2);
3803 * unregister_netdevice(y1);
3804 * unregister_netdevice(y2);
3810 * We are invoked by rtnl_unlock() after it drops the semaphore.
3811 * This allows us to deal with problems:
3812 * 1) We can delete sysfs objects which invoke hotplug
3813 * without deadlocking with linkwatch via keventd.
3814 * 2) Since we run with the RTNL semaphore not held, we can sleep
3815 * safely in order to wait for the netdev refcnt to drop to zero.
3817 static DEFINE_MUTEX(net_todo_run_mutex);
3818 void netdev_run_todo(void)
3820 struct list_head list;
3822 /* Need to guard against multiple cpu's getting out of order. */
3823 mutex_lock(&net_todo_run_mutex);
3825 /* Not safe to do outside the semaphore. We must not return
3826 * until all unregister events invoked by the local processor
3827 * have been completed (either by this todo run, or one on
3830 if (list_empty(&net_todo_list))
3833 /* Snapshot list, allow later requests */
3834 spin_lock(&net_todo_list_lock);
3835 list_replace_init(&net_todo_list, &list);
3836 spin_unlock(&net_todo_list_lock);
3838 while (!list_empty(&list)) {
3839 struct net_device *dev
3840 = list_entry(list.next, struct net_device, todo_list);
3841 list_del(&dev->todo_list);
3843 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
3844 printk(KERN_ERR "network todo '%s' but state %d\n",
3845 dev->name, dev->reg_state);
3850 dev->reg_state = NETREG_UNREGISTERED;
3852 netdev_wait_allrefs(dev);
3855 BUG_ON(atomic_read(&dev->refcnt));
3856 BUG_TRAP(!dev->ip_ptr);
3857 BUG_TRAP(!dev->ip6_ptr);
3858 BUG_TRAP(!dev->dn_ptr);
3860 if (dev->destructor)
3861 dev->destructor(dev);
3863 /* Free network device */
3864 kobject_put(&dev->dev.kobj);
3868 mutex_unlock(&net_todo_run_mutex);
3871 static struct net_device_stats *internal_stats(struct net_device *dev)
3877 * alloc_netdev_mq - allocate network device
3878 * @sizeof_priv: size of private data to allocate space for
3879 * @name: device name format string
3880 * @setup: callback to initialize device
3881 * @queue_count: the number of subqueues to allocate
3883 * Allocates a struct net_device with private data area for driver use
3884 * and performs basic initialization. Also allocates subquue structs
3885 * for each queue on the device at the end of the netdevice.
3887 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
3888 void (*setup)(struct net_device *), unsigned int queue_count)
3891 struct net_device *dev;
3894 BUG_ON(strlen(name) >= sizeof(dev->name));
3896 /* ensure 32-byte alignment of both the device and private area */
3897 alloc_size = (sizeof(*dev) + NETDEV_ALIGN_CONST +
3898 (sizeof(struct net_device_subqueue) * (queue_count - 1))) &
3899 ~NETDEV_ALIGN_CONST;
3900 alloc_size += sizeof_priv + NETDEV_ALIGN_CONST;
3902 p = kzalloc(alloc_size, GFP_KERNEL);
3904 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
3908 dev = (struct net_device *)
3909 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
3910 dev->padded = (char *)dev - (char *)p;
3911 dev->nd_net = &init_net;
3914 dev->priv = ((char *)dev +
3915 ((sizeof(struct net_device) +
3916 (sizeof(struct net_device_subqueue) *
3917 (queue_count - 1)) + NETDEV_ALIGN_CONST)
3918 & ~NETDEV_ALIGN_CONST));
3921 dev->egress_subqueue_count = queue_count;
3923 dev->get_stats = internal_stats;
3924 netpoll_netdev_init(dev);
3926 strcpy(dev->name, name);
3929 EXPORT_SYMBOL(alloc_netdev_mq);
3932 * free_netdev - free network device
3935 * This function does the last stage of destroying an allocated device
3936 * interface. The reference to the device object is released.
3937 * If this is the last reference then it will be freed.
3939 void free_netdev(struct net_device *dev)
3941 /* Compatibility with error handling in drivers */
3942 if (dev->reg_state == NETREG_UNINITIALIZED) {
3943 kfree((char *)dev - dev->padded);
3947 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
3948 dev->reg_state = NETREG_RELEASED;
3950 /* will free via device release */
3951 put_device(&dev->dev);
3954 /* Synchronize with packet receive processing. */
3955 void synchronize_net(void)
3962 * unregister_netdevice - remove device from the kernel
3965 * This function shuts down a device interface and removes it
3966 * from the kernel tables.
3968 * Callers must hold the rtnl semaphore. You may want
3969 * unregister_netdev() instead of this.
3972 void unregister_netdevice(struct net_device *dev)
3976 rollback_registered(dev);
3977 /* Finish processing unregister after unlock */
3982 * unregister_netdev - remove device from the kernel
3985 * This function shuts down a device interface and removes it
3986 * from the kernel tables.
3988 * This is just a wrapper for unregister_netdevice that takes
3989 * the rtnl semaphore. In general you want to use this and not
3990 * unregister_netdevice.
3992 void unregister_netdev(struct net_device *dev)
3995 unregister_netdevice(dev);
3999 EXPORT_SYMBOL(unregister_netdev);
4002 * dev_change_net_namespace - move device to different nethost namespace
4004 * @net: network namespace
4005 * @pat: If not NULL name pattern to try if the current device name
4006 * is already taken in the destination network namespace.
4008 * This function shuts down a device interface and moves it
4009 * to a new network namespace. On success 0 is returned, on
4010 * a failure a netagive errno code is returned.
4012 * Callers must hold the rtnl semaphore.
4015 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4018 const char *destname;
4023 /* Don't allow namespace local devices to be moved. */
4025 if (dev->features & NETIF_F_NETNS_LOCAL)
4028 /* Ensure the device has been registrered */
4030 if (dev->reg_state != NETREG_REGISTERED)
4033 /* Get out if there is nothing todo */
4035 if (dev->nd_net == net)
4038 /* Pick the destination device name, and ensure
4039 * we can use it in the destination network namespace.
4042 destname = dev->name;
4043 if (__dev_get_by_name(net, destname)) {
4044 /* We get here if we can't use the current device name */
4047 if (!dev_valid_name(pat))
4049 if (strchr(pat, '%')) {
4050 if (__dev_alloc_name(net, pat, buf) < 0)
4055 if (__dev_get_by_name(net, destname))
4060 * And now a mini version of register_netdevice unregister_netdevice.
4063 /* If device is running close it first. */
4066 /* And unlink it from device chain */
4068 unlist_netdevice(dev);
4072 /* Shutdown queueing discipline. */
4075 /* Notify protocols, that we are about to destroy
4076 this device. They should clean all the things.
4078 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4081 * Flush the unicast and multicast chains
4083 dev_addr_discard(dev);
4085 /* Actually switch the network namespace */
4088 /* Assign the new device name */
4089 if (destname != dev->name)
4090 strcpy(dev->name, destname);
4092 /* If there is an ifindex conflict assign a new one */
4093 if (__dev_get_by_index(net, dev->ifindex)) {
4094 int iflink = (dev->iflink == dev->ifindex);
4095 dev->ifindex = dev_new_index(net);
4097 dev->iflink = dev->ifindex;
4100 /* Fixup kobjects */
4101 err = device_rename(&dev->dev, dev->name);
4104 /* Add the device back in the hashes */
4105 list_netdevice(dev);
4107 /* Notify protocols, that a new device appeared. */
4108 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4116 static int dev_cpu_callback(struct notifier_block *nfb,
4117 unsigned long action,
4120 struct sk_buff **list_skb;
4121 struct net_device **list_net;
4122 struct sk_buff *skb;
4123 unsigned int cpu, oldcpu = (unsigned long)ocpu;
4124 struct softnet_data *sd, *oldsd;
4126 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
4129 local_irq_disable();
4130 cpu = smp_processor_id();
4131 sd = &per_cpu(softnet_data, cpu);
4132 oldsd = &per_cpu(softnet_data, oldcpu);
4134 /* Find end of our completion_queue. */
4135 list_skb = &sd->completion_queue;
4137 list_skb = &(*list_skb)->next;
4138 /* Append completion queue from offline CPU. */
4139 *list_skb = oldsd->completion_queue;
4140 oldsd->completion_queue = NULL;
4142 /* Find end of our output_queue. */
4143 list_net = &sd->output_queue;
4145 list_net = &(*list_net)->next_sched;
4146 /* Append output queue from offline CPU. */
4147 *list_net = oldsd->output_queue;
4148 oldsd->output_queue = NULL;
4150 raise_softirq_irqoff(NET_TX_SOFTIRQ);
4153 /* Process offline CPU's input_pkt_queue */
4154 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
4160 #ifdef CONFIG_NET_DMA
4162 * net_dma_rebalance - try to maintain one DMA channel per CPU
4163 * @net_dma: DMA client and associated data (lock, channels, channel_mask)
4165 * This is called when the number of channels allocated to the net_dma client
4166 * changes. The net_dma client tries to have one DMA channel per CPU.
4169 static void net_dma_rebalance(struct net_dma *net_dma)
4171 unsigned int cpu, i, n, chan_idx;
4172 struct dma_chan *chan;
4174 if (cpus_empty(net_dma->channel_mask)) {
4175 for_each_online_cpu(cpu)
4176 rcu_assign_pointer(per_cpu(softnet_data, cpu).net_dma, NULL);
4181 cpu = first_cpu(cpu_online_map);
4183 for_each_cpu_mask(chan_idx, net_dma->channel_mask) {
4184 chan = net_dma->channels[chan_idx];
4186 n = ((num_online_cpus() / cpus_weight(net_dma->channel_mask))
4187 + (i < (num_online_cpus() %
4188 cpus_weight(net_dma->channel_mask)) ? 1 : 0));
4191 per_cpu(softnet_data, cpu).net_dma = chan;
4192 cpu = next_cpu(cpu, cpu_online_map);
4200 * netdev_dma_event - event callback for the net_dma_client
4201 * @client: should always be net_dma_client
4202 * @chan: DMA channel for the event
4203 * @state: DMA state to be handled
4205 static enum dma_state_client
4206 netdev_dma_event(struct dma_client *client, struct dma_chan *chan,
4207 enum dma_state state)
4209 int i, found = 0, pos = -1;
4210 struct net_dma *net_dma =
4211 container_of(client, struct net_dma, client);
4212 enum dma_state_client ack = DMA_DUP; /* default: take no action */
4214 spin_lock(&net_dma->lock);
4216 case DMA_RESOURCE_AVAILABLE:
4217 for (i = 0; i < NR_CPUS; i++)
4218 if (net_dma->channels[i] == chan) {
4221 } else if (net_dma->channels[i] == NULL && pos < 0)
4224 if (!found && pos >= 0) {
4226 net_dma->channels[pos] = chan;
4227 cpu_set(pos, net_dma->channel_mask);
4228 net_dma_rebalance(net_dma);
4231 case DMA_RESOURCE_REMOVED:
4232 for (i = 0; i < NR_CPUS; i++)
4233 if (net_dma->channels[i] == chan) {
4241 cpu_clear(pos, net_dma->channel_mask);
4242 net_dma->channels[i] = NULL;
4243 net_dma_rebalance(net_dma);
4249 spin_unlock(&net_dma->lock);
4255 * netdev_dma_regiser - register the networking subsystem as a DMA client
4257 static int __init netdev_dma_register(void)
4259 spin_lock_init(&net_dma.lock);
4260 dma_cap_set(DMA_MEMCPY, net_dma.client.cap_mask);
4261 dma_async_client_register(&net_dma.client);
4262 dma_async_client_chan_request(&net_dma.client);
4267 static int __init netdev_dma_register(void) { return -ENODEV; }
4268 #endif /* CONFIG_NET_DMA */
4271 * netdev_compute_feature - compute conjunction of two feature sets
4272 * @all: first feature set
4273 * @one: second feature set
4275 * Computes a new feature set after adding a device with feature set
4276 * @one to the master device with current feature set @all. Returns
4277 * the new feature set.
4279 int netdev_compute_features(unsigned long all, unsigned long one)
4281 /* if device needs checksumming, downgrade to hw checksumming */
4282 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
4283 all ^= NETIF_F_NO_CSUM | NETIF_F_HW_CSUM;
4285 /* if device can't do all checksum, downgrade to ipv4/ipv6 */
4286 if (all & NETIF_F_HW_CSUM && !(one & NETIF_F_HW_CSUM))
4287 all ^= NETIF_F_HW_CSUM
4288 | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
4290 if (one & NETIF_F_GSO)
4291 one |= NETIF_F_GSO_SOFTWARE;
4294 /* If even one device supports robust GSO, enable it for all. */
4295 if (one & NETIF_F_GSO_ROBUST)
4296 all |= NETIF_F_GSO_ROBUST;
4298 all &= one | NETIF_F_LLTX;
4300 if (!(all & NETIF_F_ALL_CSUM))
4302 if (!(all & NETIF_F_SG))
4303 all &= ~NETIF_F_GSO_MASK;
4307 EXPORT_SYMBOL(netdev_compute_features);
4309 static struct hlist_head *netdev_create_hash(void)
4312 struct hlist_head *hash;
4314 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
4316 for (i = 0; i < NETDEV_HASHENTRIES; i++)
4317 INIT_HLIST_HEAD(&hash[i]);
4322 /* Initialize per network namespace state */
4323 static int __net_init netdev_init(struct net *net)
4325 INIT_LIST_HEAD(&net->dev_base_head);
4327 net->dev_name_head = netdev_create_hash();
4328 if (net->dev_name_head == NULL)
4331 net->dev_index_head = netdev_create_hash();
4332 if (net->dev_index_head == NULL)
4338 kfree(net->dev_name_head);
4343 static void __net_exit netdev_exit(struct net *net)
4345 kfree(net->dev_name_head);
4346 kfree(net->dev_index_head);
4349 static struct pernet_operations __net_initdata netdev_net_ops = {
4350 .init = netdev_init,
4351 .exit = netdev_exit,
4354 static void __net_exit default_device_exit(struct net *net)
4356 struct net_device *dev, *next;
4358 * Push all migratable of the network devices back to the
4359 * initial network namespace
4362 for_each_netdev_safe(net, dev, next) {
4365 /* Ignore unmoveable devices (i.e. loopback) */
4366 if (dev->features & NETIF_F_NETNS_LOCAL)
4369 /* Push remaing network devices to init_net */
4370 err = dev_change_net_namespace(dev, &init_net, "dev%d");
4372 printk(KERN_WARNING "%s: failed to move %s to init_net: %d\n",
4373 __func__, dev->name, err);
4374 unregister_netdevice(dev);
4380 static struct pernet_operations __net_initdata default_device_ops = {
4381 .exit = default_device_exit,
4385 * Initialize the DEV module. At boot time this walks the device list and
4386 * unhooks any devices that fail to initialise (normally hardware not
4387 * present) and leaves us with a valid list of present and active devices.
4392 * This is called single threaded during boot, so no need
4393 * to take the rtnl semaphore.
4395 static int __init net_dev_init(void)
4397 int i, rc = -ENOMEM;
4399 BUG_ON(!dev_boot_phase);
4401 if (dev_proc_init())
4404 if (netdev_kobject_init())
4407 INIT_LIST_HEAD(&ptype_all);
4408 for (i = 0; i < PTYPE_HASH_SIZE; i++)
4409 INIT_LIST_HEAD(&ptype_base[i]);
4411 if (register_pernet_subsys(&netdev_net_ops))
4414 if (register_pernet_device(&default_device_ops))
4418 * Initialise the packet receive queues.
4421 for_each_possible_cpu(i) {
4422 struct softnet_data *queue;
4424 queue = &per_cpu(softnet_data, i);
4425 skb_queue_head_init(&queue->input_pkt_queue);
4426 queue->completion_queue = NULL;
4427 INIT_LIST_HEAD(&queue->poll_list);
4429 queue->backlog.poll = process_backlog;
4430 queue->backlog.weight = weight_p;
4433 netdev_dma_register();
4437 open_softirq(NET_TX_SOFTIRQ, net_tx_action, NULL);
4438 open_softirq(NET_RX_SOFTIRQ, net_rx_action, NULL);
4440 hotcpu_notifier(dev_cpu_callback, 0);
4448 subsys_initcall(net_dev_init);
4450 EXPORT_SYMBOL(__dev_get_by_index);
4451 EXPORT_SYMBOL(__dev_get_by_name);
4452 EXPORT_SYMBOL(__dev_remove_pack);
4453 EXPORT_SYMBOL(dev_valid_name);
4454 EXPORT_SYMBOL(dev_add_pack);
4455 EXPORT_SYMBOL(dev_alloc_name);
4456 EXPORT_SYMBOL(dev_close);
4457 EXPORT_SYMBOL(dev_get_by_flags);
4458 EXPORT_SYMBOL(dev_get_by_index);
4459 EXPORT_SYMBOL(dev_get_by_name);
4460 EXPORT_SYMBOL(dev_open);
4461 EXPORT_SYMBOL(dev_queue_xmit);
4462 EXPORT_SYMBOL(dev_remove_pack);
4463 EXPORT_SYMBOL(dev_set_allmulti);
4464 EXPORT_SYMBOL(dev_set_promiscuity);
4465 EXPORT_SYMBOL(dev_change_flags);
4466 EXPORT_SYMBOL(dev_set_mtu);
4467 EXPORT_SYMBOL(dev_set_mac_address);
4468 EXPORT_SYMBOL(free_netdev);
4469 EXPORT_SYMBOL(netdev_boot_setup_check);
4470 EXPORT_SYMBOL(netdev_set_master);
4471 EXPORT_SYMBOL(netdev_state_change);
4472 EXPORT_SYMBOL(netif_receive_skb);
4473 EXPORT_SYMBOL(netif_rx);
4474 EXPORT_SYMBOL(register_gifconf);
4475 EXPORT_SYMBOL(register_netdevice);
4476 EXPORT_SYMBOL(register_netdevice_notifier);
4477 EXPORT_SYMBOL(skb_checksum_help);
4478 EXPORT_SYMBOL(synchronize_net);
4479 EXPORT_SYMBOL(unregister_netdevice);
4480 EXPORT_SYMBOL(unregister_netdevice_notifier);
4481 EXPORT_SYMBOL(net_enable_timestamp);
4482 EXPORT_SYMBOL(net_disable_timestamp);
4483 EXPORT_SYMBOL(dev_get_flags);
4485 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
4486 EXPORT_SYMBOL(br_handle_frame_hook);
4487 EXPORT_SYMBOL(br_fdb_get_hook);
4488 EXPORT_SYMBOL(br_fdb_put_hook);
4492 EXPORT_SYMBOL(dev_load);
4495 EXPORT_PER_CPU_SYMBOL(softnet_data);