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/ethtool.h>
94 #include <linux/notifier.h>
95 #include <linux/skbuff.h>
96 #include <net/net_namespace.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/stat.h>
102 #include <linux/if_bridge.h>
103 #include <linux/if_macvlan.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
130 #include "net-sysfs.h"
132 /* Instead of increasing this, you should create a hash table. */
133 #define MAX_GRO_SKBS 8
135 /* This should be increased if a protocol with a bigger head is added. */
136 #define GRO_MAX_HEAD (MAX_HEADER + 128)
147 * The list of packet types we will receive (as opposed to discard)
148 * and the routines to invoke.
150 * Why 16. Because with 16 the only overlap we get on a hash of the
151 * low nibble of the protocol value is RARP/SNAP/X.25.
153 * NOTE: That is no longer true with the addition of VLAN tags. Not
154 * sure which should go first, but I bet it won't make much
155 * difference if we are running VLANs. The good news is that
156 * this protocol won't be in the list unless compiled in, so
157 * the average user (w/out VLANs) will not be adversely affected.
174 #define PTYPE_HASH_SIZE (16)
175 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
177 static DEFINE_SPINLOCK(ptype_lock);
178 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
179 static struct list_head ptype_all __read_mostly; /* Taps */
182 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
185 * Pure readers hold dev_base_lock for reading.
187 * Writers must hold the rtnl semaphore while they loop through the
188 * dev_base_head list, and hold dev_base_lock for writing when they do the
189 * actual updates. This allows pure readers to access the list even
190 * while a writer is preparing to update it.
192 * To put it another way, dev_base_lock is held for writing only to
193 * protect against pure readers; the rtnl semaphore provides the
194 * protection against other writers.
196 * See, for example usages, register_netdevice() and
197 * unregister_netdevice(), which must be called with the rtnl
200 DEFINE_RWLOCK(dev_base_lock);
202 EXPORT_SYMBOL(dev_base_lock);
204 #define NETDEV_HASHBITS 8
205 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
207 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
209 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
210 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
213 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
215 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
218 /* Device list insertion */
219 static int list_netdevice(struct net_device *dev)
221 struct net *net = dev_net(dev);
225 write_lock_bh(&dev_base_lock);
226 list_add_tail(&dev->dev_list, &net->dev_base_head);
227 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
228 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
229 write_unlock_bh(&dev_base_lock);
233 /* Device list removal */
234 static void unlist_netdevice(struct net_device *dev)
238 /* Unlink dev from the device chain */
239 write_lock_bh(&dev_base_lock);
240 list_del(&dev->dev_list);
241 hlist_del(&dev->name_hlist);
242 hlist_del(&dev->index_hlist);
243 write_unlock_bh(&dev_base_lock);
250 static RAW_NOTIFIER_HEAD(netdev_chain);
253 * Device drivers call our routines to queue packets here. We empty the
254 * queue in the local softnet handler.
257 DEFINE_PER_CPU(struct softnet_data, softnet_data);
259 #ifdef CONFIG_LOCKDEP
261 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
262 * according to dev->type
264 static const unsigned short netdev_lock_type[] =
265 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
266 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
267 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
268 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
269 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
270 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
271 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
272 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
273 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
274 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
275 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
276 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
277 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
278 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
279 ARPHRD_PHONET_PIPE, ARPHRD_VOID, ARPHRD_NONE};
281 static const char *netdev_lock_name[] =
282 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
283 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
284 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
285 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
286 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
287 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
288 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
289 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
290 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
291 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
292 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
293 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
294 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
295 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
296 "_xmit_PHONET_PIPE", "_xmit_VOID", "_xmit_NONE"};
298 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
299 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
301 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
305 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
306 if (netdev_lock_type[i] == dev_type)
308 /* the last key is used by default */
309 return ARRAY_SIZE(netdev_lock_type) - 1;
312 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
313 unsigned short dev_type)
317 i = netdev_lock_pos(dev_type);
318 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
319 netdev_lock_name[i]);
322 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
326 i = netdev_lock_pos(dev->type);
327 lockdep_set_class_and_name(&dev->addr_list_lock,
328 &netdev_addr_lock_key[i],
329 netdev_lock_name[i]);
332 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
333 unsigned short dev_type)
336 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
341 /*******************************************************************************
343 Protocol management and registration routines
345 *******************************************************************************/
348 * Add a protocol ID to the list. Now that the input handler is
349 * smarter we can dispense with all the messy stuff that used to be
352 * BEWARE!!! Protocol handlers, mangling input packets,
353 * MUST BE last in hash buckets and checking protocol handlers
354 * MUST start from promiscuous ptype_all chain in net_bh.
355 * It is true now, do not change it.
356 * Explanation follows: if protocol handler, mangling packet, will
357 * be the first on list, it is not able to sense, that packet
358 * is cloned and should be copied-on-write, so that it will
359 * change it and subsequent readers will get broken packet.
364 * dev_add_pack - add packet handler
365 * @pt: packet type declaration
367 * Add a protocol handler to the networking stack. The passed &packet_type
368 * is linked into kernel lists and may not be freed until it has been
369 * removed from the kernel lists.
371 * This call does not sleep therefore it can not
372 * guarantee all CPU's that are in middle of receiving packets
373 * will see the new packet type (until the next received packet).
376 void dev_add_pack(struct packet_type *pt)
380 spin_lock_bh(&ptype_lock);
381 if (pt->type == htons(ETH_P_ALL))
382 list_add_rcu(&pt->list, &ptype_all);
384 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
385 list_add_rcu(&pt->list, &ptype_base[hash]);
387 spin_unlock_bh(&ptype_lock);
391 * __dev_remove_pack - remove packet handler
392 * @pt: packet type declaration
394 * Remove a protocol handler that was previously added to the kernel
395 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
396 * from the kernel lists and can be freed or reused once this function
399 * The packet type might still be in use by receivers
400 * and must not be freed until after all the CPU's have gone
401 * through a quiescent state.
403 void __dev_remove_pack(struct packet_type *pt)
405 struct list_head *head;
406 struct packet_type *pt1;
408 spin_lock_bh(&ptype_lock);
410 if (pt->type == htons(ETH_P_ALL))
413 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
415 list_for_each_entry(pt1, head, list) {
417 list_del_rcu(&pt->list);
422 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
424 spin_unlock_bh(&ptype_lock);
427 * dev_remove_pack - remove packet handler
428 * @pt: packet type declaration
430 * Remove a protocol handler that was previously added to the kernel
431 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
432 * from the kernel lists and can be freed or reused once this function
435 * This call sleeps to guarantee that no CPU is looking at the packet
438 void dev_remove_pack(struct packet_type *pt)
440 __dev_remove_pack(pt);
445 /******************************************************************************
447 Device Boot-time Settings Routines
449 *******************************************************************************/
451 /* Boot time configuration table */
452 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
455 * netdev_boot_setup_add - add new setup entry
456 * @name: name of the device
457 * @map: configured settings for the device
459 * Adds new setup entry to the dev_boot_setup list. The function
460 * returns 0 on error and 1 on success. This is a generic routine to
463 static int netdev_boot_setup_add(char *name, struct ifmap *map)
465 struct netdev_boot_setup *s;
469 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
470 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
471 memset(s[i].name, 0, sizeof(s[i].name));
472 strlcpy(s[i].name, name, IFNAMSIZ);
473 memcpy(&s[i].map, map, sizeof(s[i].map));
478 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
482 * netdev_boot_setup_check - check boot time settings
483 * @dev: the netdevice
485 * Check boot time settings for the device.
486 * The found settings are set for the device to be used
487 * later in the device probing.
488 * Returns 0 if no settings found, 1 if they are.
490 int netdev_boot_setup_check(struct net_device *dev)
492 struct netdev_boot_setup *s = dev_boot_setup;
495 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
496 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
497 !strcmp(dev->name, s[i].name)) {
498 dev->irq = s[i].map.irq;
499 dev->base_addr = s[i].map.base_addr;
500 dev->mem_start = s[i].map.mem_start;
501 dev->mem_end = s[i].map.mem_end;
510 * netdev_boot_base - get address from boot time settings
511 * @prefix: prefix for network device
512 * @unit: id for network device
514 * Check boot time settings for the base address of device.
515 * The found settings are set for the device to be used
516 * later in the device probing.
517 * Returns 0 if no settings found.
519 unsigned long netdev_boot_base(const char *prefix, int unit)
521 const struct netdev_boot_setup *s = dev_boot_setup;
525 sprintf(name, "%s%d", prefix, unit);
528 * If device already registered then return base of 1
529 * to indicate not to probe for this interface
531 if (__dev_get_by_name(&init_net, name))
534 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
535 if (!strcmp(name, s[i].name))
536 return s[i].map.base_addr;
541 * Saves at boot time configured settings for any netdevice.
543 int __init netdev_boot_setup(char *str)
548 str = get_options(str, ARRAY_SIZE(ints), ints);
553 memset(&map, 0, sizeof(map));
557 map.base_addr = ints[2];
559 map.mem_start = ints[3];
561 map.mem_end = ints[4];
563 /* Add new entry to the list */
564 return netdev_boot_setup_add(str, &map);
567 __setup("netdev=", netdev_boot_setup);
569 /*******************************************************************************
571 Device Interface Subroutines
573 *******************************************************************************/
576 * __dev_get_by_name - find a device by its name
577 * @net: the applicable net namespace
578 * @name: name to find
580 * Find an interface by name. Must be called under RTNL semaphore
581 * or @dev_base_lock. If the name is found a pointer to the device
582 * is returned. If the name is not found then %NULL is returned. The
583 * reference counters are not incremented so the caller must be
584 * careful with locks.
587 struct net_device *__dev_get_by_name(struct net *net, const char *name)
589 struct hlist_node *p;
591 hlist_for_each(p, dev_name_hash(net, name)) {
592 struct net_device *dev
593 = hlist_entry(p, struct net_device, name_hlist);
594 if (!strncmp(dev->name, name, IFNAMSIZ))
601 * dev_get_by_name - find a device by its name
602 * @net: the applicable net namespace
603 * @name: name to find
605 * Find an interface by name. This can be called from any
606 * context and does its own locking. The returned handle has
607 * the usage count incremented and the caller must use dev_put() to
608 * release it when it is no longer needed. %NULL is returned if no
609 * matching device is found.
612 struct net_device *dev_get_by_name(struct net *net, const char *name)
614 struct net_device *dev;
616 read_lock(&dev_base_lock);
617 dev = __dev_get_by_name(net, name);
620 read_unlock(&dev_base_lock);
625 * __dev_get_by_index - find a device by its ifindex
626 * @net: the applicable net namespace
627 * @ifindex: index of device
629 * Search for an interface by index. Returns %NULL if the device
630 * is not found or a pointer to the device. The device has not
631 * had its reference counter increased so the caller must be careful
632 * about locking. The caller must hold either the RTNL semaphore
636 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
638 struct hlist_node *p;
640 hlist_for_each(p, dev_index_hash(net, ifindex)) {
641 struct net_device *dev
642 = hlist_entry(p, struct net_device, index_hlist);
643 if (dev->ifindex == ifindex)
651 * dev_get_by_index - find a device by its ifindex
652 * @net: the applicable net namespace
653 * @ifindex: index of device
655 * Search for an interface by index. Returns NULL if the device
656 * is not found or a pointer to the device. The device returned has
657 * had a reference added and the pointer is safe until the user calls
658 * dev_put to indicate they have finished with it.
661 struct net_device *dev_get_by_index(struct net *net, int ifindex)
663 struct net_device *dev;
665 read_lock(&dev_base_lock);
666 dev = __dev_get_by_index(net, ifindex);
669 read_unlock(&dev_base_lock);
674 * dev_getbyhwaddr - find a device by its hardware address
675 * @net: the applicable net namespace
676 * @type: media type of device
677 * @ha: hardware address
679 * Search for an interface by MAC address. Returns NULL if the device
680 * is not found or a pointer to the device. The caller must hold the
681 * rtnl semaphore. The returned device has not had its ref count increased
682 * and the caller must therefore be careful about locking
685 * If the API was consistent this would be __dev_get_by_hwaddr
688 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
690 struct net_device *dev;
694 for_each_netdev(net, dev)
695 if (dev->type == type &&
696 !memcmp(dev->dev_addr, ha, dev->addr_len))
702 EXPORT_SYMBOL(dev_getbyhwaddr);
704 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
706 struct net_device *dev;
709 for_each_netdev(net, dev)
710 if (dev->type == type)
716 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
718 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
720 struct net_device *dev;
723 dev = __dev_getfirstbyhwtype(net, type);
730 EXPORT_SYMBOL(dev_getfirstbyhwtype);
733 * dev_get_by_flags - find any device with given flags
734 * @net: the applicable net namespace
735 * @if_flags: IFF_* values
736 * @mask: bitmask of bits in if_flags to check
738 * Search for any interface with the given flags. Returns NULL if a device
739 * is not found or a pointer to the device. The device returned has
740 * had a reference added and the pointer is safe until the user calls
741 * dev_put to indicate they have finished with it.
744 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
746 struct net_device *dev, *ret;
749 read_lock(&dev_base_lock);
750 for_each_netdev(net, dev) {
751 if (((dev->flags ^ if_flags) & mask) == 0) {
757 read_unlock(&dev_base_lock);
762 * dev_valid_name - check if name is okay for network device
765 * Network device names need to be valid file names to
766 * to allow sysfs to work. We also disallow any kind of
769 int dev_valid_name(const char *name)
773 if (strlen(name) >= IFNAMSIZ)
775 if (!strcmp(name, ".") || !strcmp(name, ".."))
779 if (*name == '/' || isspace(*name))
787 * __dev_alloc_name - allocate a name for a device
788 * @net: network namespace to allocate the device name in
789 * @name: name format string
790 * @buf: scratch buffer and result name string
792 * Passed a format string - eg "lt%d" it will try and find a suitable
793 * id. It scans list of devices to build up a free map, then chooses
794 * the first empty slot. The caller must hold the dev_base or rtnl lock
795 * while allocating the name and adding the device in order to avoid
797 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
798 * Returns the number of the unit assigned or a negative errno code.
801 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
805 const int max_netdevices = 8*PAGE_SIZE;
806 unsigned long *inuse;
807 struct net_device *d;
809 p = strnchr(name, IFNAMSIZ-1, '%');
812 * Verify the string as this thing may have come from
813 * the user. There must be either one "%d" and no other "%"
816 if (p[1] != 'd' || strchr(p + 2, '%'))
819 /* Use one page as a bit array of possible slots */
820 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
824 for_each_netdev(net, d) {
825 if (!sscanf(d->name, name, &i))
827 if (i < 0 || i >= max_netdevices)
830 /* avoid cases where sscanf is not exact inverse of printf */
831 snprintf(buf, IFNAMSIZ, name, i);
832 if (!strncmp(buf, d->name, IFNAMSIZ))
836 i = find_first_zero_bit(inuse, max_netdevices);
837 free_page((unsigned long) inuse);
840 snprintf(buf, IFNAMSIZ, name, i);
841 if (!__dev_get_by_name(net, buf))
844 /* It is possible to run out of possible slots
845 * when the name is long and there isn't enough space left
846 * for the digits, or if all bits are used.
852 * dev_alloc_name - allocate a name for a device
854 * @name: name format string
856 * Passed a format string - eg "lt%d" it will try and find a suitable
857 * id. It scans list of devices to build up a free map, then chooses
858 * the first empty slot. The caller must hold the dev_base or rtnl lock
859 * while allocating the name and adding the device in order to avoid
861 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
862 * Returns the number of the unit assigned or a negative errno code.
865 int dev_alloc_name(struct net_device *dev, const char *name)
871 BUG_ON(!dev_net(dev));
873 ret = __dev_alloc_name(net, name, buf);
875 strlcpy(dev->name, buf, IFNAMSIZ);
881 * dev_change_name - change name of a device
883 * @newname: name (or format string) must be at least IFNAMSIZ
885 * Change name of a device, can pass format strings "eth%d".
888 int dev_change_name(struct net_device *dev, const char *newname)
890 char oldname[IFNAMSIZ];
896 BUG_ON(!dev_net(dev));
899 if (dev->flags & IFF_UP)
902 if (!dev_valid_name(newname))
905 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
908 memcpy(oldname, dev->name, IFNAMSIZ);
910 if (strchr(newname, '%')) {
911 err = dev_alloc_name(dev, newname);
915 else if (__dev_get_by_name(net, newname))
918 strlcpy(dev->name, newname, IFNAMSIZ);
921 /* For now only devices in the initial network namespace
924 if (net == &init_net) {
925 ret = device_rename(&dev->dev, dev->name);
927 memcpy(dev->name, oldname, IFNAMSIZ);
932 write_lock_bh(&dev_base_lock);
933 hlist_del(&dev->name_hlist);
934 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
935 write_unlock_bh(&dev_base_lock);
937 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
938 ret = notifier_to_errno(ret);
943 "%s: name change rollback failed: %d.\n",
947 memcpy(dev->name, oldname, IFNAMSIZ);
956 * dev_set_alias - change ifalias of a device
958 * @alias: name up to IFALIASZ
959 * @len: limit of bytes to copy from info
961 * Set ifalias for a device,
963 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
978 dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL);
982 strlcpy(dev->ifalias, alias, len+1);
988 * netdev_features_change - device changes features
989 * @dev: device to cause notification
991 * Called to indicate a device has changed features.
993 void netdev_features_change(struct net_device *dev)
995 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
997 EXPORT_SYMBOL(netdev_features_change);
1000 * netdev_state_change - device changes state
1001 * @dev: device to cause notification
1003 * Called to indicate a device has changed state. This function calls
1004 * the notifier chains for netdev_chain and sends a NEWLINK message
1005 * to the routing socket.
1007 void netdev_state_change(struct net_device *dev)
1009 if (dev->flags & IFF_UP) {
1010 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1011 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1015 void netdev_bonding_change(struct net_device *dev)
1017 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
1019 EXPORT_SYMBOL(netdev_bonding_change);
1022 * dev_load - load a network module
1023 * @net: the applicable net namespace
1024 * @name: name of interface
1026 * If a network interface is not present and the process has suitable
1027 * privileges this function loads the module. If module loading is not
1028 * available in this kernel then it becomes a nop.
1031 void dev_load(struct net *net, const char *name)
1033 struct net_device *dev;
1035 read_lock(&dev_base_lock);
1036 dev = __dev_get_by_name(net, name);
1037 read_unlock(&dev_base_lock);
1039 if (!dev && capable(CAP_SYS_MODULE))
1040 request_module("%s", name);
1044 * dev_open - prepare an interface for use.
1045 * @dev: device to open
1047 * Takes a device from down to up state. The device's private open
1048 * function is invoked and then the multicast lists are loaded. Finally
1049 * the device is moved into the up state and a %NETDEV_UP message is
1050 * sent to the netdev notifier chain.
1052 * Calling this function on an active interface is a nop. On a failure
1053 * a negative errno code is returned.
1055 int dev_open(struct net_device *dev)
1057 const struct net_device_ops *ops = dev->netdev_ops;
1066 if (dev->flags & IFF_UP)
1070 * Is it even present?
1072 if (!netif_device_present(dev))
1076 * Call device private open method
1078 set_bit(__LINK_STATE_START, &dev->state);
1080 if (ops->ndo_validate_addr)
1081 ret = ops->ndo_validate_addr(dev);
1083 if (!ret && ops->ndo_open)
1084 ret = ops->ndo_open(dev);
1087 * If it went open OK then:
1091 clear_bit(__LINK_STATE_START, &dev->state);
1096 dev->flags |= IFF_UP;
1101 net_dmaengine_get();
1104 * Initialize multicasting status
1106 dev_set_rx_mode(dev);
1109 * Wakeup transmit queue engine
1114 * ... and announce new interface.
1116 call_netdevice_notifiers(NETDEV_UP, dev);
1123 * dev_close - shutdown an interface.
1124 * @dev: device to shutdown
1126 * This function moves an active device into down state. A
1127 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1128 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1131 int dev_close(struct net_device *dev)
1133 const struct net_device_ops *ops = dev->netdev_ops;
1138 if (!(dev->flags & IFF_UP))
1142 * Tell people we are going down, so that they can
1143 * prepare to death, when device is still operating.
1145 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1147 clear_bit(__LINK_STATE_START, &dev->state);
1149 /* Synchronize to scheduled poll. We cannot touch poll list,
1150 * it can be even on different cpu. So just clear netif_running().
1152 * dev->stop() will invoke napi_disable() on all of it's
1153 * napi_struct instances on this device.
1155 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1157 dev_deactivate(dev);
1160 * Call the device specific close. This cannot fail.
1161 * Only if device is UP
1163 * We allow it to be called even after a DETACH hot-plug
1170 * Device is now down.
1173 dev->flags &= ~IFF_UP;
1176 * Tell people we are down
1178 call_netdevice_notifiers(NETDEV_DOWN, dev);
1183 net_dmaengine_put();
1190 * dev_disable_lro - disable Large Receive Offload on a device
1193 * Disable Large Receive Offload (LRO) on a net device. Must be
1194 * called under RTNL. This is needed if received packets may be
1195 * forwarded to another interface.
1197 void dev_disable_lro(struct net_device *dev)
1199 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1200 dev->ethtool_ops->set_flags) {
1201 u32 flags = dev->ethtool_ops->get_flags(dev);
1202 if (flags & ETH_FLAG_LRO) {
1203 flags &= ~ETH_FLAG_LRO;
1204 dev->ethtool_ops->set_flags(dev, flags);
1207 WARN_ON(dev->features & NETIF_F_LRO);
1209 EXPORT_SYMBOL(dev_disable_lro);
1212 static int dev_boot_phase = 1;
1215 * Device change register/unregister. These are not inline or static
1216 * as we export them to the world.
1220 * register_netdevice_notifier - register a network notifier block
1223 * Register a notifier to be called when network device events occur.
1224 * The notifier passed is linked into the kernel structures and must
1225 * not be reused until it has been unregistered. A negative errno code
1226 * is returned on a failure.
1228 * When registered all registration and up events are replayed
1229 * to the new notifier to allow device to have a race free
1230 * view of the network device list.
1233 int register_netdevice_notifier(struct notifier_block *nb)
1235 struct net_device *dev;
1236 struct net_device *last;
1241 err = raw_notifier_chain_register(&netdev_chain, nb);
1247 for_each_netdev(net, dev) {
1248 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1249 err = notifier_to_errno(err);
1253 if (!(dev->flags & IFF_UP))
1256 nb->notifier_call(nb, NETDEV_UP, dev);
1267 for_each_netdev(net, dev) {
1271 if (dev->flags & IFF_UP) {
1272 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1273 nb->notifier_call(nb, NETDEV_DOWN, dev);
1275 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1279 raw_notifier_chain_unregister(&netdev_chain, nb);
1284 * unregister_netdevice_notifier - unregister a network notifier block
1287 * Unregister a notifier previously registered by
1288 * register_netdevice_notifier(). The notifier is unlinked into the
1289 * kernel structures and may then be reused. A negative errno code
1290 * is returned on a failure.
1293 int unregister_netdevice_notifier(struct notifier_block *nb)
1298 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1304 * call_netdevice_notifiers - call all network notifier blocks
1305 * @val: value passed unmodified to notifier function
1306 * @dev: net_device pointer passed unmodified to notifier function
1308 * Call all network notifier blocks. Parameters and return value
1309 * are as for raw_notifier_call_chain().
1312 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1314 return raw_notifier_call_chain(&netdev_chain, val, dev);
1317 /* When > 0 there are consumers of rx skb time stamps */
1318 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1320 void net_enable_timestamp(void)
1322 atomic_inc(&netstamp_needed);
1325 void net_disable_timestamp(void)
1327 atomic_dec(&netstamp_needed);
1330 static inline void net_timestamp(struct sk_buff *skb)
1332 if (atomic_read(&netstamp_needed))
1333 __net_timestamp(skb);
1335 skb->tstamp.tv64 = 0;
1339 * Support routine. Sends outgoing frames to any network
1340 * taps currently in use.
1343 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1345 struct packet_type *ptype;
1350 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1351 /* Never send packets back to the socket
1352 * they originated from - MvS (miquels@drinkel.ow.org)
1354 if ((ptype->dev == dev || !ptype->dev) &&
1355 (ptype->af_packet_priv == NULL ||
1356 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1357 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1361 /* skb->nh should be correctly
1362 set by sender, so that the second statement is
1363 just protection against buggy protocols.
1365 skb_reset_mac_header(skb2);
1367 if (skb_network_header(skb2) < skb2->data ||
1368 skb2->network_header > skb2->tail) {
1369 if (net_ratelimit())
1370 printk(KERN_CRIT "protocol %04x is "
1372 skb2->protocol, dev->name);
1373 skb_reset_network_header(skb2);
1376 skb2->transport_header = skb2->network_header;
1377 skb2->pkt_type = PACKET_OUTGOING;
1378 ptype->func(skb2, skb->dev, ptype, skb->dev);
1385 static inline void __netif_reschedule(struct Qdisc *q)
1387 struct softnet_data *sd;
1388 unsigned long flags;
1390 local_irq_save(flags);
1391 sd = &__get_cpu_var(softnet_data);
1392 q->next_sched = sd->output_queue;
1393 sd->output_queue = q;
1394 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1395 local_irq_restore(flags);
1398 void __netif_schedule(struct Qdisc *q)
1400 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1401 __netif_reschedule(q);
1403 EXPORT_SYMBOL(__netif_schedule);
1405 void dev_kfree_skb_irq(struct sk_buff *skb)
1407 if (atomic_dec_and_test(&skb->users)) {
1408 struct softnet_data *sd;
1409 unsigned long flags;
1411 local_irq_save(flags);
1412 sd = &__get_cpu_var(softnet_data);
1413 skb->next = sd->completion_queue;
1414 sd->completion_queue = skb;
1415 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1416 local_irq_restore(flags);
1419 EXPORT_SYMBOL(dev_kfree_skb_irq);
1421 void dev_kfree_skb_any(struct sk_buff *skb)
1423 if (in_irq() || irqs_disabled())
1424 dev_kfree_skb_irq(skb);
1428 EXPORT_SYMBOL(dev_kfree_skb_any);
1432 * netif_device_detach - mark device as removed
1433 * @dev: network device
1435 * Mark device as removed from system and therefore no longer available.
1437 void netif_device_detach(struct net_device *dev)
1439 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1440 netif_running(dev)) {
1441 netif_stop_queue(dev);
1444 EXPORT_SYMBOL(netif_device_detach);
1447 * netif_device_attach - mark device as attached
1448 * @dev: network device
1450 * Mark device as attached from system and restart if needed.
1452 void netif_device_attach(struct net_device *dev)
1454 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1455 netif_running(dev)) {
1456 netif_wake_queue(dev);
1457 __netdev_watchdog_up(dev);
1460 EXPORT_SYMBOL(netif_device_attach);
1462 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1464 return ((features & NETIF_F_GEN_CSUM) ||
1465 ((features & NETIF_F_IP_CSUM) &&
1466 protocol == htons(ETH_P_IP)) ||
1467 ((features & NETIF_F_IPV6_CSUM) &&
1468 protocol == htons(ETH_P_IPV6)));
1471 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1473 if (can_checksum_protocol(dev->features, skb->protocol))
1476 if (skb->protocol == htons(ETH_P_8021Q)) {
1477 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1478 if (can_checksum_protocol(dev->features & dev->vlan_features,
1479 veh->h_vlan_encapsulated_proto))
1487 * Invalidate hardware checksum when packet is to be mangled, and
1488 * complete checksum manually on outgoing path.
1490 int skb_checksum_help(struct sk_buff *skb)
1493 int ret = 0, offset;
1495 if (skb->ip_summed == CHECKSUM_COMPLETE)
1496 goto out_set_summed;
1498 if (unlikely(skb_shinfo(skb)->gso_size)) {
1499 /* Let GSO fix up the checksum. */
1500 goto out_set_summed;
1503 offset = skb->csum_start - skb_headroom(skb);
1504 BUG_ON(offset >= skb_headlen(skb));
1505 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1507 offset += skb->csum_offset;
1508 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1510 if (skb_cloned(skb) &&
1511 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1512 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1517 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1519 skb->ip_summed = CHECKSUM_NONE;
1525 * skb_gso_segment - Perform segmentation on skb.
1526 * @skb: buffer to segment
1527 * @features: features for the output path (see dev->features)
1529 * This function segments the given skb and returns a list of segments.
1531 * It may return NULL if the skb requires no segmentation. This is
1532 * only possible when GSO is used for verifying header integrity.
1534 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1536 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1537 struct packet_type *ptype;
1538 __be16 type = skb->protocol;
1541 skb_reset_mac_header(skb);
1542 skb->mac_len = skb->network_header - skb->mac_header;
1543 __skb_pull(skb, skb->mac_len);
1545 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1546 struct net_device *dev = skb->dev;
1547 struct ethtool_drvinfo info = {};
1549 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1550 dev->ethtool_ops->get_drvinfo(dev, &info);
1552 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1554 info.driver, dev ? dev->features : 0L,
1555 skb->sk ? skb->sk->sk_route_caps : 0L,
1556 skb->len, skb->data_len, skb->ip_summed);
1558 if (skb_header_cloned(skb) &&
1559 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1560 return ERR_PTR(err);
1564 list_for_each_entry_rcu(ptype,
1565 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1566 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1567 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1568 err = ptype->gso_send_check(skb);
1569 segs = ERR_PTR(err);
1570 if (err || skb_gso_ok(skb, features))
1572 __skb_push(skb, (skb->data -
1573 skb_network_header(skb)));
1575 segs = ptype->gso_segment(skb, features);
1581 __skb_push(skb, skb->data - skb_mac_header(skb));
1586 EXPORT_SYMBOL(skb_gso_segment);
1588 /* Take action when hardware reception checksum errors are detected. */
1590 void netdev_rx_csum_fault(struct net_device *dev)
1592 if (net_ratelimit()) {
1593 printk(KERN_ERR "%s: hw csum failure.\n",
1594 dev ? dev->name : "<unknown>");
1598 EXPORT_SYMBOL(netdev_rx_csum_fault);
1601 /* Actually, we should eliminate this check as soon as we know, that:
1602 * 1. IOMMU is present and allows to map all the memory.
1603 * 2. No high memory really exists on this machine.
1606 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1608 #ifdef CONFIG_HIGHMEM
1611 if (dev->features & NETIF_F_HIGHDMA)
1614 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1615 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1623 void (*destructor)(struct sk_buff *skb);
1626 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1628 static void dev_gso_skb_destructor(struct sk_buff *skb)
1630 struct dev_gso_cb *cb;
1633 struct sk_buff *nskb = skb->next;
1635 skb->next = nskb->next;
1638 } while (skb->next);
1640 cb = DEV_GSO_CB(skb);
1642 cb->destructor(skb);
1646 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1647 * @skb: buffer to segment
1649 * This function segments the given skb and stores the list of segments
1652 static int dev_gso_segment(struct sk_buff *skb)
1654 struct net_device *dev = skb->dev;
1655 struct sk_buff *segs;
1656 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1659 segs = skb_gso_segment(skb, features);
1661 /* Verifying header integrity only. */
1666 return PTR_ERR(segs);
1669 DEV_GSO_CB(skb)->destructor = skb->destructor;
1670 skb->destructor = dev_gso_skb_destructor;
1675 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1676 struct netdev_queue *txq)
1678 const struct net_device_ops *ops = dev->netdev_ops;
1681 prefetch(&dev->netdev_ops->ndo_start_xmit);
1682 if (likely(!skb->next)) {
1683 if (!list_empty(&ptype_all))
1684 dev_queue_xmit_nit(skb, dev);
1686 if (netif_needs_gso(dev, skb)) {
1687 if (unlikely(dev_gso_segment(skb)))
1693 rc = ops->ndo_start_xmit(skb, dev);
1695 * TODO: if skb_orphan() was called by
1696 * dev->hard_start_xmit() (for example, the unmodified
1697 * igb driver does that; bnx2 doesn't), then
1698 * skb_tx_software_timestamp() will be unable to send
1699 * back the time stamp.
1701 * How can this be prevented? Always create another
1702 * reference to the socket before calling
1703 * dev->hard_start_xmit()? Prevent that skb_orphan()
1704 * does anything in dev->hard_start_xmit() by clearing
1705 * the skb destructor before the call and restoring it
1706 * afterwards, then doing the skb_orphan() ourselves?
1713 struct sk_buff *nskb = skb->next;
1715 skb->next = nskb->next;
1717 rc = ops->ndo_start_xmit(nskb, dev);
1719 nskb->next = skb->next;
1723 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1724 return NETDEV_TX_BUSY;
1725 } while (skb->next);
1727 skb->destructor = DEV_GSO_CB(skb)->destructor;
1734 static u32 skb_tx_hashrnd;
1736 static u16 skb_tx_hash(struct net_device *dev, struct sk_buff *skb)
1740 if (skb_rx_queue_recorded(skb)) {
1741 hash = skb_get_rx_queue(skb);
1742 } else if (skb->sk && skb->sk->sk_hash) {
1743 hash = skb->sk->sk_hash;
1745 hash = skb->protocol;
1747 hash = jhash_1word(hash, skb_tx_hashrnd);
1749 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1752 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1753 struct sk_buff *skb)
1755 const struct net_device_ops *ops = dev->netdev_ops;
1756 u16 queue_index = 0;
1758 if (ops->ndo_select_queue)
1759 queue_index = ops->ndo_select_queue(dev, skb);
1760 else if (dev->real_num_tx_queues > 1)
1761 queue_index = skb_tx_hash(dev, skb);
1763 skb_set_queue_mapping(skb, queue_index);
1764 return netdev_get_tx_queue(dev, queue_index);
1768 * dev_queue_xmit - transmit a buffer
1769 * @skb: buffer to transmit
1771 * Queue a buffer for transmission to a network device. The caller must
1772 * have set the device and priority and built the buffer before calling
1773 * this function. The function can be called from an interrupt.
1775 * A negative errno code is returned on a failure. A success does not
1776 * guarantee the frame will be transmitted as it may be dropped due
1777 * to congestion or traffic shaping.
1779 * -----------------------------------------------------------------------------------
1780 * I notice this method can also return errors from the queue disciplines,
1781 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1784 * Regardless of the return value, the skb is consumed, so it is currently
1785 * difficult to retry a send to this method. (You can bump the ref count
1786 * before sending to hold a reference for retry if you are careful.)
1788 * When calling this method, interrupts MUST be enabled. This is because
1789 * the BH enable code must have IRQs enabled so that it will not deadlock.
1792 int dev_queue_xmit(struct sk_buff *skb)
1794 struct net_device *dev = skb->dev;
1795 struct netdev_queue *txq;
1799 /* GSO will handle the following emulations directly. */
1800 if (netif_needs_gso(dev, skb))
1803 if (skb_shinfo(skb)->frag_list &&
1804 !(dev->features & NETIF_F_FRAGLIST) &&
1805 __skb_linearize(skb))
1808 /* Fragmented skb is linearized if device does not support SG,
1809 * or if at least one of fragments is in highmem and device
1810 * does not support DMA from it.
1812 if (skb_shinfo(skb)->nr_frags &&
1813 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1814 __skb_linearize(skb))
1817 /* If packet is not checksummed and device does not support
1818 * checksumming for this protocol, complete checksumming here.
1820 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1821 skb_set_transport_header(skb, skb->csum_start -
1823 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1828 /* Disable soft irqs for various locks below. Also
1829 * stops preemption for RCU.
1833 txq = dev_pick_tx(dev, skb);
1834 q = rcu_dereference(txq->qdisc);
1836 #ifdef CONFIG_NET_CLS_ACT
1837 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1840 spinlock_t *root_lock = qdisc_lock(q);
1842 spin_lock(root_lock);
1844 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1848 rc = qdisc_enqueue_root(skb, q);
1851 spin_unlock(root_lock);
1856 /* The device has no queue. Common case for software devices:
1857 loopback, all the sorts of tunnels...
1859 Really, it is unlikely that netif_tx_lock protection is necessary
1860 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1862 However, it is possible, that they rely on protection
1865 Check this and shot the lock. It is not prone from deadlocks.
1866 Either shot noqueue qdisc, it is even simpler 8)
1868 if (dev->flags & IFF_UP) {
1869 int cpu = smp_processor_id(); /* ok because BHs are off */
1871 if (txq->xmit_lock_owner != cpu) {
1873 HARD_TX_LOCK(dev, txq, cpu);
1875 if (!netif_tx_queue_stopped(txq)) {
1877 if (!dev_hard_start_xmit(skb, dev, txq)) {
1878 HARD_TX_UNLOCK(dev, txq);
1882 HARD_TX_UNLOCK(dev, txq);
1883 if (net_ratelimit())
1884 printk(KERN_CRIT "Virtual device %s asks to "
1885 "queue packet!\n", dev->name);
1887 /* Recursion is detected! It is possible,
1889 if (net_ratelimit())
1890 printk(KERN_CRIT "Dead loop on virtual device "
1891 "%s, fix it urgently!\n", dev->name);
1896 rcu_read_unlock_bh();
1902 rcu_read_unlock_bh();
1907 /*=======================================================================
1909 =======================================================================*/
1911 int netdev_max_backlog __read_mostly = 1000;
1912 int netdev_budget __read_mostly = 300;
1913 int weight_p __read_mostly = 64; /* old backlog weight */
1915 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1919 * netif_rx - post buffer to the network code
1920 * @skb: buffer to post
1922 * This function receives a packet from a device driver and queues it for
1923 * the upper (protocol) levels to process. It always succeeds. The buffer
1924 * may be dropped during processing for congestion control or by the
1928 * NET_RX_SUCCESS (no congestion)
1929 * NET_RX_DROP (packet was dropped)
1933 int netif_rx(struct sk_buff *skb)
1935 struct softnet_data *queue;
1936 unsigned long flags;
1938 /* if netpoll wants it, pretend we never saw it */
1939 if (netpoll_rx(skb))
1942 if (!skb->tstamp.tv64)
1946 * The code is rearranged so that the path is the most
1947 * short when CPU is congested, but is still operating.
1949 local_irq_save(flags);
1950 queue = &__get_cpu_var(softnet_data);
1952 __get_cpu_var(netdev_rx_stat).total++;
1953 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1954 if (queue->input_pkt_queue.qlen) {
1956 __skb_queue_tail(&queue->input_pkt_queue, skb);
1957 local_irq_restore(flags);
1958 return NET_RX_SUCCESS;
1961 napi_schedule(&queue->backlog);
1965 __get_cpu_var(netdev_rx_stat).dropped++;
1966 local_irq_restore(flags);
1972 int netif_rx_ni(struct sk_buff *skb)
1977 err = netif_rx(skb);
1978 if (local_softirq_pending())
1985 EXPORT_SYMBOL(netif_rx_ni);
1987 static void net_tx_action(struct softirq_action *h)
1989 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1991 if (sd->completion_queue) {
1992 struct sk_buff *clist;
1994 local_irq_disable();
1995 clist = sd->completion_queue;
1996 sd->completion_queue = NULL;
2000 struct sk_buff *skb = clist;
2001 clist = clist->next;
2003 WARN_ON(atomic_read(&skb->users));
2008 if (sd->output_queue) {
2011 local_irq_disable();
2012 head = sd->output_queue;
2013 sd->output_queue = NULL;
2017 struct Qdisc *q = head;
2018 spinlock_t *root_lock;
2020 head = head->next_sched;
2022 root_lock = qdisc_lock(q);
2023 if (spin_trylock(root_lock)) {
2024 smp_mb__before_clear_bit();
2025 clear_bit(__QDISC_STATE_SCHED,
2028 spin_unlock(root_lock);
2030 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2032 __netif_reschedule(q);
2034 smp_mb__before_clear_bit();
2035 clear_bit(__QDISC_STATE_SCHED,
2043 static inline int deliver_skb(struct sk_buff *skb,
2044 struct packet_type *pt_prev,
2045 struct net_device *orig_dev)
2047 atomic_inc(&skb->users);
2048 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2051 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2052 /* These hooks defined here for ATM */
2054 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2055 unsigned char *addr);
2056 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2059 * If bridge module is loaded call bridging hook.
2060 * returns NULL if packet was consumed.
2062 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2063 struct sk_buff *skb) __read_mostly;
2064 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2065 struct packet_type **pt_prev, int *ret,
2066 struct net_device *orig_dev)
2068 struct net_bridge_port *port;
2070 if (skb->pkt_type == PACKET_LOOPBACK ||
2071 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2075 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2079 return br_handle_frame_hook(port, skb);
2082 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2085 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2086 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2087 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2089 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2090 struct packet_type **pt_prev,
2092 struct net_device *orig_dev)
2094 if (skb->dev->macvlan_port == NULL)
2098 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2101 return macvlan_handle_frame_hook(skb);
2104 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2107 #ifdef CONFIG_NET_CLS_ACT
2108 /* TODO: Maybe we should just force sch_ingress to be compiled in
2109 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2110 * a compare and 2 stores extra right now if we dont have it on
2111 * but have CONFIG_NET_CLS_ACT
2112 * NOTE: This doesnt stop any functionality; if you dont have
2113 * the ingress scheduler, you just cant add policies on ingress.
2116 static int ing_filter(struct sk_buff *skb)
2118 struct net_device *dev = skb->dev;
2119 u32 ttl = G_TC_RTTL(skb->tc_verd);
2120 struct netdev_queue *rxq;
2121 int result = TC_ACT_OK;
2124 if (MAX_RED_LOOP < ttl++) {
2126 "Redir loop detected Dropping packet (%d->%d)\n",
2127 skb->iif, dev->ifindex);
2131 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2132 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2134 rxq = &dev->rx_queue;
2137 if (q != &noop_qdisc) {
2138 spin_lock(qdisc_lock(q));
2139 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2140 result = qdisc_enqueue_root(skb, q);
2141 spin_unlock(qdisc_lock(q));
2147 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2148 struct packet_type **pt_prev,
2149 int *ret, struct net_device *orig_dev)
2151 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2155 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2158 /* Huh? Why does turning on AF_PACKET affect this? */
2159 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2162 switch (ing_filter(skb)) {
2176 * netif_nit_deliver - deliver received packets to network taps
2179 * This function is used to deliver incoming packets to network
2180 * taps. It should be used when the normal netif_receive_skb path
2181 * is bypassed, for example because of VLAN acceleration.
2183 void netif_nit_deliver(struct sk_buff *skb)
2185 struct packet_type *ptype;
2187 if (list_empty(&ptype_all))
2190 skb_reset_network_header(skb);
2191 skb_reset_transport_header(skb);
2192 skb->mac_len = skb->network_header - skb->mac_header;
2195 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2196 if (!ptype->dev || ptype->dev == skb->dev)
2197 deliver_skb(skb, ptype, skb->dev);
2203 * netif_receive_skb - process receive buffer from network
2204 * @skb: buffer to process
2206 * netif_receive_skb() is the main receive data processing function.
2207 * It always succeeds. The buffer may be dropped during processing
2208 * for congestion control or by the protocol layers.
2210 * This function may only be called from softirq context and interrupts
2211 * should be enabled.
2213 * Return values (usually ignored):
2214 * NET_RX_SUCCESS: no congestion
2215 * NET_RX_DROP: packet was dropped
2217 int netif_receive_skb(struct sk_buff *skb)
2219 struct packet_type *ptype, *pt_prev;
2220 struct net_device *orig_dev;
2221 struct net_device *null_or_orig;
2222 int ret = NET_RX_DROP;
2225 if (skb->vlan_tci && vlan_hwaccel_do_receive(skb))
2226 return NET_RX_SUCCESS;
2228 /* if we've gotten here through NAPI, check netpoll */
2229 if (netpoll_receive_skb(skb))
2232 if (!skb->tstamp.tv64)
2236 skb->iif = skb->dev->ifindex;
2238 null_or_orig = NULL;
2239 orig_dev = skb->dev;
2240 if (orig_dev->master) {
2241 if (skb_bond_should_drop(skb))
2242 null_or_orig = orig_dev; /* deliver only exact match */
2244 skb->dev = orig_dev->master;
2247 __get_cpu_var(netdev_rx_stat).total++;
2249 skb_reset_network_header(skb);
2250 skb_reset_transport_header(skb);
2251 skb->mac_len = skb->network_header - skb->mac_header;
2257 #ifdef CONFIG_NET_CLS_ACT
2258 if (skb->tc_verd & TC_NCLS) {
2259 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2264 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2265 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2266 ptype->dev == orig_dev) {
2268 ret = deliver_skb(skb, pt_prev, orig_dev);
2273 #ifdef CONFIG_NET_CLS_ACT
2274 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2280 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2283 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2289 type = skb->protocol;
2290 list_for_each_entry_rcu(ptype,
2291 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2292 if (ptype->type == type &&
2293 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2294 ptype->dev == orig_dev)) {
2296 ret = deliver_skb(skb, pt_prev, orig_dev);
2302 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2305 /* Jamal, now you will not able to escape explaining
2306 * me how you were going to use this. :-)
2316 /* Network device is going away, flush any packets still pending */
2317 static void flush_backlog(void *arg)
2319 struct net_device *dev = arg;
2320 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2321 struct sk_buff *skb, *tmp;
2323 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2324 if (skb->dev == dev) {
2325 __skb_unlink(skb, &queue->input_pkt_queue);
2330 static int napi_gro_complete(struct sk_buff *skb)
2332 struct packet_type *ptype;
2333 __be16 type = skb->protocol;
2334 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2337 if (NAPI_GRO_CB(skb)->count == 1)
2341 list_for_each_entry_rcu(ptype, head, list) {
2342 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2345 err = ptype->gro_complete(skb);
2351 WARN_ON(&ptype->list == head);
2353 return NET_RX_SUCCESS;
2357 skb_shinfo(skb)->gso_size = 0;
2358 return netif_receive_skb(skb);
2361 void napi_gro_flush(struct napi_struct *napi)
2363 struct sk_buff *skb, *next;
2365 for (skb = napi->gro_list; skb; skb = next) {
2368 napi_gro_complete(skb);
2371 napi->gro_count = 0;
2372 napi->gro_list = NULL;
2374 EXPORT_SYMBOL(napi_gro_flush);
2376 void *skb_gro_header(struct sk_buff *skb, unsigned int hlen)
2378 unsigned int offset = skb_gro_offset(skb);
2381 if (hlen <= skb_headlen(skb))
2382 return skb->data + offset;
2384 if (unlikely(!skb_shinfo(skb)->nr_frags ||
2385 skb_shinfo(skb)->frags[0].size <=
2386 hlen - skb_headlen(skb) ||
2387 PageHighMem(skb_shinfo(skb)->frags[0].page)))
2388 return pskb_may_pull(skb, hlen) ? skb->data + offset : NULL;
2390 return page_address(skb_shinfo(skb)->frags[0].page) +
2391 skb_shinfo(skb)->frags[0].page_offset +
2392 offset - skb_headlen(skb);
2394 EXPORT_SYMBOL(skb_gro_header);
2396 int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2398 struct sk_buff **pp = NULL;
2399 struct packet_type *ptype;
2400 __be16 type = skb->protocol;
2401 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2406 if (!(skb->dev->features & NETIF_F_GRO))
2409 if (skb_is_gso(skb) || skb_shinfo(skb)->frag_list)
2413 list_for_each_entry_rcu(ptype, head, list) {
2414 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2417 skb_set_network_header(skb, skb_gro_offset(skb));
2418 mac_len = skb->network_header - skb->mac_header;
2419 skb->mac_len = mac_len;
2420 NAPI_GRO_CB(skb)->same_flow = 0;
2421 NAPI_GRO_CB(skb)->flush = 0;
2422 NAPI_GRO_CB(skb)->free = 0;
2424 pp = ptype->gro_receive(&napi->gro_list, skb);
2429 if (&ptype->list == head)
2432 same_flow = NAPI_GRO_CB(skb)->same_flow;
2433 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2436 struct sk_buff *nskb = *pp;
2440 napi_gro_complete(nskb);
2447 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2451 NAPI_GRO_CB(skb)->count = 1;
2452 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2453 skb->next = napi->gro_list;
2454 napi->gro_list = skb;
2458 if (unlikely(!pskb_may_pull(skb, skb_gro_offset(skb)))) {
2459 if (napi->gro_list == skb)
2460 napi->gro_list = skb->next;
2471 EXPORT_SYMBOL(dev_gro_receive);
2473 static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2477 for (p = napi->gro_list; p; p = p->next) {
2478 NAPI_GRO_CB(p)->same_flow = !compare_ether_header(
2479 skb_mac_header(p), skb_gro_mac_header(skb));
2480 NAPI_GRO_CB(p)->flush = 0;
2483 return dev_gro_receive(napi, skb);
2486 int napi_skb_finish(int ret, struct sk_buff *skb)
2488 int err = NET_RX_SUCCESS;
2492 return netif_receive_skb(skb);
2498 case GRO_MERGED_FREE:
2505 EXPORT_SYMBOL(napi_skb_finish);
2507 int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2509 skb_gro_reset_offset(skb);
2511 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2513 EXPORT_SYMBOL(napi_gro_receive);
2515 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2517 __skb_pull(skb, skb_headlen(skb));
2518 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2522 EXPORT_SYMBOL(napi_reuse_skb);
2524 struct sk_buff *napi_fraginfo_skb(struct napi_struct *napi,
2525 struct napi_gro_fraginfo *info)
2527 struct net_device *dev = napi->dev;
2528 struct sk_buff *skb = napi->skb;
2536 skb = netdev_alloc_skb(dev, GRO_MAX_HEAD + NET_IP_ALIGN);
2540 skb_reserve(skb, NET_IP_ALIGN);
2543 BUG_ON(info->nr_frags > MAX_SKB_FRAGS);
2544 frag = &info->frags[info->nr_frags - 1];
2546 for (i = skb_shinfo(skb)->nr_frags; i < info->nr_frags; i++) {
2547 skb_fill_page_desc(skb, i, frag->page, frag->page_offset,
2551 skb_shinfo(skb)->nr_frags = info->nr_frags;
2553 skb->data_len = info->len;
2554 skb->len += info->len;
2555 skb->truesize += info->len;
2557 skb_reset_mac_header(skb);
2558 skb_gro_reset_offset(skb);
2560 eth = skb_gro_header(skb, sizeof(*eth));
2562 napi_reuse_skb(napi, skb);
2567 skb_gro_pull(skb, sizeof(*eth));
2570 * This works because the only protocols we care about don't require
2571 * special handling. We'll fix it up properly at the end.
2573 skb->protocol = eth->h_proto;
2575 skb->ip_summed = info->ip_summed;
2576 skb->csum = info->csum;
2581 EXPORT_SYMBOL(napi_fraginfo_skb);
2583 int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret)
2585 int err = NET_RX_SUCCESS;
2590 skb->protocol = eth_type_trans(skb, napi->dev);
2592 if (ret == GRO_NORMAL)
2593 return netif_receive_skb(skb);
2595 skb_gro_pull(skb, -ETH_HLEN);
2602 case GRO_MERGED_FREE:
2603 napi_reuse_skb(napi, skb);
2609 EXPORT_SYMBOL(napi_frags_finish);
2611 int napi_gro_frags(struct napi_struct *napi, struct napi_gro_fraginfo *info)
2613 struct sk_buff *skb = napi_fraginfo_skb(napi, info);
2618 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2620 EXPORT_SYMBOL(napi_gro_frags);
2622 static int process_backlog(struct napi_struct *napi, int quota)
2625 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2626 unsigned long start_time = jiffies;
2628 napi->weight = weight_p;
2630 struct sk_buff *skb;
2632 local_irq_disable();
2633 skb = __skb_dequeue(&queue->input_pkt_queue);
2635 __napi_complete(napi);
2641 napi_gro_receive(napi, skb);
2642 } while (++work < quota && jiffies == start_time);
2644 napi_gro_flush(napi);
2650 * __napi_schedule - schedule for receive
2651 * @n: entry to schedule
2653 * The entry's receive function will be scheduled to run
2655 void __napi_schedule(struct napi_struct *n)
2657 unsigned long flags;
2659 local_irq_save(flags);
2660 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2661 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2662 local_irq_restore(flags);
2664 EXPORT_SYMBOL(__napi_schedule);
2666 void __napi_complete(struct napi_struct *n)
2668 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2669 BUG_ON(n->gro_list);
2671 list_del(&n->poll_list);
2672 smp_mb__before_clear_bit();
2673 clear_bit(NAPI_STATE_SCHED, &n->state);
2675 EXPORT_SYMBOL(__napi_complete);
2677 void napi_complete(struct napi_struct *n)
2679 unsigned long flags;
2682 * don't let napi dequeue from the cpu poll list
2683 * just in case its running on a different cpu
2685 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2689 local_irq_save(flags);
2691 local_irq_restore(flags);
2693 EXPORT_SYMBOL(napi_complete);
2695 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2696 int (*poll)(struct napi_struct *, int), int weight)
2698 INIT_LIST_HEAD(&napi->poll_list);
2699 napi->gro_count = 0;
2700 napi->gro_list = NULL;
2703 napi->weight = weight;
2704 list_add(&napi->dev_list, &dev->napi_list);
2706 #ifdef CONFIG_NETPOLL
2707 spin_lock_init(&napi->poll_lock);
2708 napi->poll_owner = -1;
2710 set_bit(NAPI_STATE_SCHED, &napi->state);
2712 EXPORT_SYMBOL(netif_napi_add);
2714 void netif_napi_del(struct napi_struct *napi)
2716 struct sk_buff *skb, *next;
2718 list_del_init(&napi->dev_list);
2721 for (skb = napi->gro_list; skb; skb = next) {
2727 napi->gro_list = NULL;
2728 napi->gro_count = 0;
2730 EXPORT_SYMBOL(netif_napi_del);
2733 static void net_rx_action(struct softirq_action *h)
2735 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2736 unsigned long time_limit = jiffies + 2;
2737 int budget = netdev_budget;
2740 local_irq_disable();
2742 while (!list_empty(list)) {
2743 struct napi_struct *n;
2746 /* If softirq window is exhuasted then punt.
2747 * Allow this to run for 2 jiffies since which will allow
2748 * an average latency of 1.5/HZ.
2750 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2755 /* Even though interrupts have been re-enabled, this
2756 * access is safe because interrupts can only add new
2757 * entries to the tail of this list, and only ->poll()
2758 * calls can remove this head entry from the list.
2760 n = list_entry(list->next, struct napi_struct, poll_list);
2762 have = netpoll_poll_lock(n);
2766 /* This NAPI_STATE_SCHED test is for avoiding a race
2767 * with netpoll's poll_napi(). Only the entity which
2768 * obtains the lock and sees NAPI_STATE_SCHED set will
2769 * actually make the ->poll() call. Therefore we avoid
2770 * accidently calling ->poll() when NAPI is not scheduled.
2773 if (test_bit(NAPI_STATE_SCHED, &n->state))
2774 work = n->poll(n, weight);
2776 WARN_ON_ONCE(work > weight);
2780 local_irq_disable();
2782 /* Drivers must not modify the NAPI state if they
2783 * consume the entire weight. In such cases this code
2784 * still "owns" the NAPI instance and therefore can
2785 * move the instance around on the list at-will.
2787 if (unlikely(work == weight)) {
2788 if (unlikely(napi_disable_pending(n)))
2791 list_move_tail(&n->poll_list, list);
2794 netpoll_poll_unlock(have);
2799 #ifdef CONFIG_NET_DMA
2801 * There may not be any more sk_buffs coming right now, so push
2802 * any pending DMA copies to hardware
2804 dma_issue_pending_all();
2810 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2811 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2815 static gifconf_func_t * gifconf_list [NPROTO];
2818 * register_gifconf - register a SIOCGIF handler
2819 * @family: Address family
2820 * @gifconf: Function handler
2822 * Register protocol dependent address dumping routines. The handler
2823 * that is passed must not be freed or reused until it has been replaced
2824 * by another handler.
2826 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2828 if (family >= NPROTO)
2830 gifconf_list[family] = gifconf;
2836 * Map an interface index to its name (SIOCGIFNAME)
2840 * We need this ioctl for efficient implementation of the
2841 * if_indextoname() function required by the IPv6 API. Without
2842 * it, we would have to search all the interfaces to find a
2846 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2848 struct net_device *dev;
2852 * Fetch the caller's info block.
2855 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2858 read_lock(&dev_base_lock);
2859 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2861 read_unlock(&dev_base_lock);
2865 strcpy(ifr.ifr_name, dev->name);
2866 read_unlock(&dev_base_lock);
2868 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2874 * Perform a SIOCGIFCONF call. This structure will change
2875 * size eventually, and there is nothing I can do about it.
2876 * Thus we will need a 'compatibility mode'.
2879 static int dev_ifconf(struct net *net, char __user *arg)
2882 struct net_device *dev;
2889 * Fetch the caller's info block.
2892 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2899 * Loop over the interfaces, and write an info block for each.
2903 for_each_netdev(net, dev) {
2904 for (i = 0; i < NPROTO; i++) {
2905 if (gifconf_list[i]) {
2908 done = gifconf_list[i](dev, NULL, 0);
2910 done = gifconf_list[i](dev, pos + total,
2920 * All done. Write the updated control block back to the caller.
2922 ifc.ifc_len = total;
2925 * Both BSD and Solaris return 0 here, so we do too.
2927 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2930 #ifdef CONFIG_PROC_FS
2932 * This is invoked by the /proc filesystem handler to display a device
2935 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2936 __acquires(dev_base_lock)
2938 struct net *net = seq_file_net(seq);
2940 struct net_device *dev;
2942 read_lock(&dev_base_lock);
2944 return SEQ_START_TOKEN;
2947 for_each_netdev(net, dev)
2954 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2956 struct net *net = seq_file_net(seq);
2958 return v == SEQ_START_TOKEN ?
2959 first_net_device(net) : next_net_device((struct net_device *)v);
2962 void dev_seq_stop(struct seq_file *seq, void *v)
2963 __releases(dev_base_lock)
2965 read_unlock(&dev_base_lock);
2968 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2970 const struct net_device_stats *stats = dev_get_stats(dev);
2972 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2973 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2974 dev->name, stats->rx_bytes, stats->rx_packets,
2976 stats->rx_dropped + stats->rx_missed_errors,
2977 stats->rx_fifo_errors,
2978 stats->rx_length_errors + stats->rx_over_errors +
2979 stats->rx_crc_errors + stats->rx_frame_errors,
2980 stats->rx_compressed, stats->multicast,
2981 stats->tx_bytes, stats->tx_packets,
2982 stats->tx_errors, stats->tx_dropped,
2983 stats->tx_fifo_errors, stats->collisions,
2984 stats->tx_carrier_errors +
2985 stats->tx_aborted_errors +
2986 stats->tx_window_errors +
2987 stats->tx_heartbeat_errors,
2988 stats->tx_compressed);
2992 * Called from the PROCfs module. This now uses the new arbitrary sized
2993 * /proc/net interface to create /proc/net/dev
2995 static int dev_seq_show(struct seq_file *seq, void *v)
2997 if (v == SEQ_START_TOKEN)
2998 seq_puts(seq, "Inter-| Receive "
3000 " face |bytes packets errs drop fifo frame "
3001 "compressed multicast|bytes packets errs "
3002 "drop fifo colls carrier compressed\n");
3004 dev_seq_printf_stats(seq, v);
3008 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3010 struct netif_rx_stats *rc = NULL;
3012 while (*pos < nr_cpu_ids)
3013 if (cpu_online(*pos)) {
3014 rc = &per_cpu(netdev_rx_stat, *pos);
3021 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3023 return softnet_get_online(pos);
3026 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3029 return softnet_get_online(pos);
3032 static void softnet_seq_stop(struct seq_file *seq, void *v)
3036 static int softnet_seq_show(struct seq_file *seq, void *v)
3038 struct netif_rx_stats *s = v;
3040 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3041 s->total, s->dropped, s->time_squeeze, 0,
3042 0, 0, 0, 0, /* was fastroute */
3047 static const struct seq_operations dev_seq_ops = {
3048 .start = dev_seq_start,
3049 .next = dev_seq_next,
3050 .stop = dev_seq_stop,
3051 .show = dev_seq_show,
3054 static int dev_seq_open(struct inode *inode, struct file *file)
3056 return seq_open_net(inode, file, &dev_seq_ops,
3057 sizeof(struct seq_net_private));
3060 static const struct file_operations dev_seq_fops = {
3061 .owner = THIS_MODULE,
3062 .open = dev_seq_open,
3064 .llseek = seq_lseek,
3065 .release = seq_release_net,
3068 static const struct seq_operations softnet_seq_ops = {
3069 .start = softnet_seq_start,
3070 .next = softnet_seq_next,
3071 .stop = softnet_seq_stop,
3072 .show = softnet_seq_show,
3075 static int softnet_seq_open(struct inode *inode, struct file *file)
3077 return seq_open(file, &softnet_seq_ops);
3080 static const struct file_operations softnet_seq_fops = {
3081 .owner = THIS_MODULE,
3082 .open = softnet_seq_open,
3084 .llseek = seq_lseek,
3085 .release = seq_release,
3088 static void *ptype_get_idx(loff_t pos)
3090 struct packet_type *pt = NULL;
3094 list_for_each_entry_rcu(pt, &ptype_all, list) {
3100 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3101 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3110 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3114 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3117 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3119 struct packet_type *pt;
3120 struct list_head *nxt;
3124 if (v == SEQ_START_TOKEN)
3125 return ptype_get_idx(0);
3128 nxt = pt->list.next;
3129 if (pt->type == htons(ETH_P_ALL)) {
3130 if (nxt != &ptype_all)
3133 nxt = ptype_base[0].next;
3135 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3137 while (nxt == &ptype_base[hash]) {
3138 if (++hash >= PTYPE_HASH_SIZE)
3140 nxt = ptype_base[hash].next;
3143 return list_entry(nxt, struct packet_type, list);
3146 static void ptype_seq_stop(struct seq_file *seq, void *v)
3152 static int ptype_seq_show(struct seq_file *seq, void *v)
3154 struct packet_type *pt = v;
3156 if (v == SEQ_START_TOKEN)
3157 seq_puts(seq, "Type Device Function\n");
3158 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3159 if (pt->type == htons(ETH_P_ALL))
3160 seq_puts(seq, "ALL ");
3162 seq_printf(seq, "%04x", ntohs(pt->type));
3164 seq_printf(seq, " %-8s %pF\n",
3165 pt->dev ? pt->dev->name : "", pt->func);
3171 static const struct seq_operations ptype_seq_ops = {
3172 .start = ptype_seq_start,
3173 .next = ptype_seq_next,
3174 .stop = ptype_seq_stop,
3175 .show = ptype_seq_show,
3178 static int ptype_seq_open(struct inode *inode, struct file *file)
3180 return seq_open_net(inode, file, &ptype_seq_ops,
3181 sizeof(struct seq_net_private));
3184 static const struct file_operations ptype_seq_fops = {
3185 .owner = THIS_MODULE,
3186 .open = ptype_seq_open,
3188 .llseek = seq_lseek,
3189 .release = seq_release_net,
3193 static int __net_init dev_proc_net_init(struct net *net)
3197 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3199 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3201 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3204 if (wext_proc_init(net))
3210 proc_net_remove(net, "ptype");
3212 proc_net_remove(net, "softnet_stat");
3214 proc_net_remove(net, "dev");
3218 static void __net_exit dev_proc_net_exit(struct net *net)
3220 wext_proc_exit(net);
3222 proc_net_remove(net, "ptype");
3223 proc_net_remove(net, "softnet_stat");
3224 proc_net_remove(net, "dev");
3227 static struct pernet_operations __net_initdata dev_proc_ops = {
3228 .init = dev_proc_net_init,
3229 .exit = dev_proc_net_exit,
3232 static int __init dev_proc_init(void)
3234 return register_pernet_subsys(&dev_proc_ops);
3237 #define dev_proc_init() 0
3238 #endif /* CONFIG_PROC_FS */
3242 * netdev_set_master - set up master/slave pair
3243 * @slave: slave device
3244 * @master: new master device
3246 * Changes the master device of the slave. Pass %NULL to break the
3247 * bonding. The caller must hold the RTNL semaphore. On a failure
3248 * a negative errno code is returned. On success the reference counts
3249 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3250 * function returns zero.
3252 int netdev_set_master(struct net_device *slave, struct net_device *master)
3254 struct net_device *old = slave->master;
3264 slave->master = master;
3272 slave->flags |= IFF_SLAVE;
3274 slave->flags &= ~IFF_SLAVE;
3276 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3280 static void dev_change_rx_flags(struct net_device *dev, int flags)
3282 const struct net_device_ops *ops = dev->netdev_ops;
3284 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3285 ops->ndo_change_rx_flags(dev, flags);
3288 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3290 unsigned short old_flags = dev->flags;
3296 dev->flags |= IFF_PROMISC;
3297 dev->promiscuity += inc;
3298 if (dev->promiscuity == 0) {
3301 * If inc causes overflow, untouch promisc and return error.
3304 dev->flags &= ~IFF_PROMISC;
3306 dev->promiscuity -= inc;
3307 printk(KERN_WARNING "%s: promiscuity touches roof, "
3308 "set promiscuity failed, promiscuity feature "
3309 "of device might be broken.\n", dev->name);
3313 if (dev->flags != old_flags) {
3314 printk(KERN_INFO "device %s %s promiscuous mode\n",
3315 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3317 if (audit_enabled) {
3318 current_uid_gid(&uid, &gid);
3319 audit_log(current->audit_context, GFP_ATOMIC,
3320 AUDIT_ANOM_PROMISCUOUS,
3321 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3322 dev->name, (dev->flags & IFF_PROMISC),
3323 (old_flags & IFF_PROMISC),
3324 audit_get_loginuid(current),
3326 audit_get_sessionid(current));
3329 dev_change_rx_flags(dev, IFF_PROMISC);
3335 * dev_set_promiscuity - update promiscuity count on a device
3339 * Add or remove promiscuity from a device. While the count in the device
3340 * remains above zero the interface remains promiscuous. Once it hits zero
3341 * the device reverts back to normal filtering operation. A negative inc
3342 * value is used to drop promiscuity on the device.
3343 * Return 0 if successful or a negative errno code on error.
3345 int dev_set_promiscuity(struct net_device *dev, int inc)
3347 unsigned short old_flags = dev->flags;
3350 err = __dev_set_promiscuity(dev, inc);
3353 if (dev->flags != old_flags)
3354 dev_set_rx_mode(dev);
3359 * dev_set_allmulti - update allmulti count on a device
3363 * Add or remove reception of all multicast frames to a device. While the
3364 * count in the device remains above zero the interface remains listening
3365 * to all interfaces. Once it hits zero the device reverts back to normal
3366 * filtering operation. A negative @inc value is used to drop the counter
3367 * when releasing a resource needing all multicasts.
3368 * Return 0 if successful or a negative errno code on error.
3371 int dev_set_allmulti(struct net_device *dev, int inc)
3373 unsigned short old_flags = dev->flags;
3377 dev->flags |= IFF_ALLMULTI;
3378 dev->allmulti += inc;
3379 if (dev->allmulti == 0) {
3382 * If inc causes overflow, untouch allmulti and return error.
3385 dev->flags &= ~IFF_ALLMULTI;
3387 dev->allmulti -= inc;
3388 printk(KERN_WARNING "%s: allmulti touches roof, "
3389 "set allmulti failed, allmulti feature of "
3390 "device might be broken.\n", dev->name);
3394 if (dev->flags ^ old_flags) {
3395 dev_change_rx_flags(dev, IFF_ALLMULTI);
3396 dev_set_rx_mode(dev);
3402 * Upload unicast and multicast address lists to device and
3403 * configure RX filtering. When the device doesn't support unicast
3404 * filtering it is put in promiscuous mode while unicast addresses
3407 void __dev_set_rx_mode(struct net_device *dev)
3409 const struct net_device_ops *ops = dev->netdev_ops;
3411 /* dev_open will call this function so the list will stay sane. */
3412 if (!(dev->flags&IFF_UP))
3415 if (!netif_device_present(dev))
3418 if (ops->ndo_set_rx_mode)
3419 ops->ndo_set_rx_mode(dev);
3421 /* Unicast addresses changes may only happen under the rtnl,
3422 * therefore calling __dev_set_promiscuity here is safe.
3424 if (dev->uc_count > 0 && !dev->uc_promisc) {
3425 __dev_set_promiscuity(dev, 1);
3426 dev->uc_promisc = 1;
3427 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3428 __dev_set_promiscuity(dev, -1);
3429 dev->uc_promisc = 0;
3432 if (ops->ndo_set_multicast_list)
3433 ops->ndo_set_multicast_list(dev);
3437 void dev_set_rx_mode(struct net_device *dev)
3439 netif_addr_lock_bh(dev);
3440 __dev_set_rx_mode(dev);
3441 netif_addr_unlock_bh(dev);
3444 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3445 void *addr, int alen, int glbl)
3447 struct dev_addr_list *da;
3449 for (; (da = *list) != NULL; list = &da->next) {
3450 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3451 alen == da->da_addrlen) {
3453 int old_glbl = da->da_gusers;
3470 int __dev_addr_add(struct dev_addr_list **list, int *count,
3471 void *addr, int alen, int glbl)
3473 struct dev_addr_list *da;
3475 for (da = *list; da != NULL; da = da->next) {
3476 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3477 da->da_addrlen == alen) {
3479 int old_glbl = da->da_gusers;
3489 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3492 memcpy(da->da_addr, addr, alen);
3493 da->da_addrlen = alen;
3495 da->da_gusers = glbl ? 1 : 0;
3503 * dev_unicast_delete - Release secondary unicast address.
3505 * @addr: address to delete
3506 * @alen: length of @addr
3508 * Release reference to a secondary unicast address and remove it
3509 * from the device if the reference count drops to zero.
3511 * The caller must hold the rtnl_mutex.
3513 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3519 netif_addr_lock_bh(dev);
3520 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3522 __dev_set_rx_mode(dev);
3523 netif_addr_unlock_bh(dev);
3526 EXPORT_SYMBOL(dev_unicast_delete);
3529 * dev_unicast_add - add a secondary unicast address
3531 * @addr: address to add
3532 * @alen: length of @addr
3534 * Add a secondary unicast address to the device or increase
3535 * the reference count if it already exists.
3537 * The caller must hold the rtnl_mutex.
3539 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3545 netif_addr_lock_bh(dev);
3546 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3548 __dev_set_rx_mode(dev);
3549 netif_addr_unlock_bh(dev);
3552 EXPORT_SYMBOL(dev_unicast_add);
3554 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3555 struct dev_addr_list **from, int *from_count)
3557 struct dev_addr_list *da, *next;
3561 while (da != NULL) {
3563 if (!da->da_synced) {
3564 err = __dev_addr_add(to, to_count,
3565 da->da_addr, da->da_addrlen, 0);
3570 } else if (da->da_users == 1) {
3571 __dev_addr_delete(to, to_count,
3572 da->da_addr, da->da_addrlen, 0);
3573 __dev_addr_delete(from, from_count,
3574 da->da_addr, da->da_addrlen, 0);
3581 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3582 struct dev_addr_list **from, int *from_count)
3584 struct dev_addr_list *da, *next;
3587 while (da != NULL) {
3589 if (da->da_synced) {
3590 __dev_addr_delete(to, to_count,
3591 da->da_addr, da->da_addrlen, 0);
3593 __dev_addr_delete(from, from_count,
3594 da->da_addr, da->da_addrlen, 0);
3601 * dev_unicast_sync - Synchronize device's unicast list to another device
3602 * @to: destination device
3603 * @from: source device
3605 * Add newly added addresses to the destination device and release
3606 * addresses that have no users left. The source device must be
3607 * locked by netif_tx_lock_bh.
3609 * This function is intended to be called from the dev->set_rx_mode
3610 * function of layered software devices.
3612 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3616 netif_addr_lock_bh(to);
3617 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3618 &from->uc_list, &from->uc_count);
3620 __dev_set_rx_mode(to);
3621 netif_addr_unlock_bh(to);
3624 EXPORT_SYMBOL(dev_unicast_sync);
3627 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3628 * @to: destination device
3629 * @from: source device
3631 * Remove all addresses that were added to the destination device by
3632 * dev_unicast_sync(). This function is intended to be called from the
3633 * dev->stop function of layered software devices.
3635 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3637 netif_addr_lock_bh(from);
3638 netif_addr_lock(to);
3640 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3641 &from->uc_list, &from->uc_count);
3642 __dev_set_rx_mode(to);
3644 netif_addr_unlock(to);
3645 netif_addr_unlock_bh(from);
3647 EXPORT_SYMBOL(dev_unicast_unsync);
3649 static void __dev_addr_discard(struct dev_addr_list **list)
3651 struct dev_addr_list *tmp;
3653 while (*list != NULL) {
3656 if (tmp->da_users > tmp->da_gusers)
3657 printk("__dev_addr_discard: address leakage! "
3658 "da_users=%d\n", tmp->da_users);
3663 static void dev_addr_discard(struct net_device *dev)
3665 netif_addr_lock_bh(dev);
3667 __dev_addr_discard(&dev->uc_list);
3670 __dev_addr_discard(&dev->mc_list);
3673 netif_addr_unlock_bh(dev);
3677 * dev_get_flags - get flags reported to userspace
3680 * Get the combination of flag bits exported through APIs to userspace.
3682 unsigned dev_get_flags(const struct net_device *dev)
3686 flags = (dev->flags & ~(IFF_PROMISC |
3691 (dev->gflags & (IFF_PROMISC |
3694 if (netif_running(dev)) {
3695 if (netif_oper_up(dev))
3696 flags |= IFF_RUNNING;
3697 if (netif_carrier_ok(dev))
3698 flags |= IFF_LOWER_UP;
3699 if (netif_dormant(dev))
3700 flags |= IFF_DORMANT;
3707 * dev_change_flags - change device settings
3709 * @flags: device state flags
3711 * Change settings on device based state flags. The flags are
3712 * in the userspace exported format.
3714 int dev_change_flags(struct net_device *dev, unsigned flags)
3717 int old_flags = dev->flags;
3722 * Set the flags on our device.
3725 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3726 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3728 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3732 * Load in the correct multicast list now the flags have changed.
3735 if ((old_flags ^ flags) & IFF_MULTICAST)
3736 dev_change_rx_flags(dev, IFF_MULTICAST);
3738 dev_set_rx_mode(dev);
3741 * Have we downed the interface. We handle IFF_UP ourselves
3742 * according to user attempts to set it, rather than blindly
3747 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3748 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3751 dev_set_rx_mode(dev);
3754 if (dev->flags & IFF_UP &&
3755 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3757 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3759 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3760 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3761 dev->gflags ^= IFF_PROMISC;
3762 dev_set_promiscuity(dev, inc);
3765 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3766 is important. Some (broken) drivers set IFF_PROMISC, when
3767 IFF_ALLMULTI is requested not asking us and not reporting.
3769 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3770 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3771 dev->gflags ^= IFF_ALLMULTI;
3772 dev_set_allmulti(dev, inc);
3775 /* Exclude state transition flags, already notified */
3776 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3778 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3784 * dev_set_mtu - Change maximum transfer unit
3786 * @new_mtu: new transfer unit
3788 * Change the maximum transfer size of the network device.
3790 int dev_set_mtu(struct net_device *dev, int new_mtu)
3792 const struct net_device_ops *ops = dev->netdev_ops;
3795 if (new_mtu == dev->mtu)
3798 /* MTU must be positive. */
3802 if (!netif_device_present(dev))
3806 if (ops->ndo_change_mtu)
3807 err = ops->ndo_change_mtu(dev, new_mtu);
3811 if (!err && dev->flags & IFF_UP)
3812 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3817 * dev_set_mac_address - Change Media Access Control Address
3821 * Change the hardware (MAC) address of the device
3823 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3825 const struct net_device_ops *ops = dev->netdev_ops;
3828 if (!ops->ndo_set_mac_address)
3830 if (sa->sa_family != dev->type)
3832 if (!netif_device_present(dev))
3834 err = ops->ndo_set_mac_address(dev, sa);
3836 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3841 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3843 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3846 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3852 case SIOCGIFFLAGS: /* Get interface flags */
3853 ifr->ifr_flags = dev_get_flags(dev);
3856 case SIOCGIFMETRIC: /* Get the metric on the interface
3857 (currently unused) */
3858 ifr->ifr_metric = 0;
3861 case SIOCGIFMTU: /* Get the MTU of a device */
3862 ifr->ifr_mtu = dev->mtu;
3867 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3869 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3870 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3871 ifr->ifr_hwaddr.sa_family = dev->type;
3879 ifr->ifr_map.mem_start = dev->mem_start;
3880 ifr->ifr_map.mem_end = dev->mem_end;
3881 ifr->ifr_map.base_addr = dev->base_addr;
3882 ifr->ifr_map.irq = dev->irq;
3883 ifr->ifr_map.dma = dev->dma;
3884 ifr->ifr_map.port = dev->if_port;
3888 ifr->ifr_ifindex = dev->ifindex;
3892 ifr->ifr_qlen = dev->tx_queue_len;
3896 /* dev_ioctl() should ensure this case
3908 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3910 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3913 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3914 const struct net_device_ops *ops;
3919 ops = dev->netdev_ops;
3922 case SIOCSIFFLAGS: /* Set interface flags */
3923 return dev_change_flags(dev, ifr->ifr_flags);
3925 case SIOCSIFMETRIC: /* Set the metric on the interface
3926 (currently unused) */
3929 case SIOCSIFMTU: /* Set the MTU of a device */
3930 return dev_set_mtu(dev, ifr->ifr_mtu);
3933 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3935 case SIOCSIFHWBROADCAST:
3936 if (ifr->ifr_hwaddr.sa_family != dev->type)
3938 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3939 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3940 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3944 if (ops->ndo_set_config) {
3945 if (!netif_device_present(dev))
3947 return ops->ndo_set_config(dev, &ifr->ifr_map);
3952 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
3953 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3955 if (!netif_device_present(dev))
3957 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3961 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
3962 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3964 if (!netif_device_present(dev))
3966 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3970 if (ifr->ifr_qlen < 0)
3972 dev->tx_queue_len = ifr->ifr_qlen;
3976 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3977 return dev_change_name(dev, ifr->ifr_newname);
3980 * Unknown or private ioctl
3984 if ((cmd >= SIOCDEVPRIVATE &&
3985 cmd <= SIOCDEVPRIVATE + 15) ||
3986 cmd == SIOCBONDENSLAVE ||
3987 cmd == SIOCBONDRELEASE ||
3988 cmd == SIOCBONDSETHWADDR ||
3989 cmd == SIOCBONDSLAVEINFOQUERY ||
3990 cmd == SIOCBONDINFOQUERY ||
3991 cmd == SIOCBONDCHANGEACTIVE ||
3992 cmd == SIOCGMIIPHY ||
3993 cmd == SIOCGMIIREG ||
3994 cmd == SIOCSMIIREG ||
3995 cmd == SIOCBRADDIF ||
3996 cmd == SIOCBRDELIF ||
3997 cmd == SIOCSHWTSTAMP ||
3998 cmd == SIOCWANDEV) {
4000 if (ops->ndo_do_ioctl) {
4001 if (netif_device_present(dev))
4002 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4014 * This function handles all "interface"-type I/O control requests. The actual
4015 * 'doing' part of this is dev_ifsioc above.
4019 * dev_ioctl - network device ioctl
4020 * @net: the applicable net namespace
4021 * @cmd: command to issue
4022 * @arg: pointer to a struct ifreq in user space
4024 * Issue ioctl functions to devices. This is normally called by the
4025 * user space syscall interfaces but can sometimes be useful for
4026 * other purposes. The return value is the return from the syscall if
4027 * positive or a negative errno code on error.
4030 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4036 /* One special case: SIOCGIFCONF takes ifconf argument
4037 and requires shared lock, because it sleeps writing
4041 if (cmd == SIOCGIFCONF) {
4043 ret = dev_ifconf(net, (char __user *) arg);
4047 if (cmd == SIOCGIFNAME)
4048 return dev_ifname(net, (struct ifreq __user *)arg);
4050 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4053 ifr.ifr_name[IFNAMSIZ-1] = 0;
4055 colon = strchr(ifr.ifr_name, ':');
4060 * See which interface the caller is talking about.
4065 * These ioctl calls:
4066 * - can be done by all.
4067 * - atomic and do not require locking.
4078 dev_load(net, ifr.ifr_name);
4079 read_lock(&dev_base_lock);
4080 ret = dev_ifsioc_locked(net, &ifr, cmd);
4081 read_unlock(&dev_base_lock);
4085 if (copy_to_user(arg, &ifr,
4086 sizeof(struct ifreq)))
4092 dev_load(net, ifr.ifr_name);
4094 ret = dev_ethtool(net, &ifr);
4099 if (copy_to_user(arg, &ifr,
4100 sizeof(struct ifreq)))
4106 * These ioctl calls:
4107 * - require superuser power.
4108 * - require strict serialization.
4114 if (!capable(CAP_NET_ADMIN))
4116 dev_load(net, ifr.ifr_name);
4118 ret = dev_ifsioc(net, &ifr, cmd);
4123 if (copy_to_user(arg, &ifr,
4124 sizeof(struct ifreq)))
4130 * These ioctl calls:
4131 * - require superuser power.
4132 * - require strict serialization.
4133 * - do not return a value
4143 case SIOCSIFHWBROADCAST:
4146 case SIOCBONDENSLAVE:
4147 case SIOCBONDRELEASE:
4148 case SIOCBONDSETHWADDR:
4149 case SIOCBONDCHANGEACTIVE:
4153 if (!capable(CAP_NET_ADMIN))
4156 case SIOCBONDSLAVEINFOQUERY:
4157 case SIOCBONDINFOQUERY:
4158 dev_load(net, ifr.ifr_name);
4160 ret = dev_ifsioc(net, &ifr, cmd);
4165 /* Get the per device memory space. We can add this but
4166 * currently do not support it */
4168 /* Set the per device memory buffer space.
4169 * Not applicable in our case */
4174 * Unknown or private ioctl.
4177 if (cmd == SIOCWANDEV ||
4178 (cmd >= SIOCDEVPRIVATE &&
4179 cmd <= SIOCDEVPRIVATE + 15)) {
4180 dev_load(net, ifr.ifr_name);
4182 ret = dev_ifsioc(net, &ifr, cmd);
4184 if (!ret && copy_to_user(arg, &ifr,
4185 sizeof(struct ifreq)))
4189 /* Take care of Wireless Extensions */
4190 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4191 return wext_handle_ioctl(net, &ifr, cmd, arg);
4198 * dev_new_index - allocate an ifindex
4199 * @net: the applicable net namespace
4201 * Returns a suitable unique value for a new device interface
4202 * number. The caller must hold the rtnl semaphore or the
4203 * dev_base_lock to be sure it remains unique.
4205 static int dev_new_index(struct net *net)
4211 if (!__dev_get_by_index(net, ifindex))
4216 /* Delayed registration/unregisteration */
4217 static LIST_HEAD(net_todo_list);
4219 static void net_set_todo(struct net_device *dev)
4221 list_add_tail(&dev->todo_list, &net_todo_list);
4224 static void rollback_registered(struct net_device *dev)
4226 BUG_ON(dev_boot_phase);
4229 /* Some devices call without registering for initialization unwind. */
4230 if (dev->reg_state == NETREG_UNINITIALIZED) {
4231 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
4232 "was registered\n", dev->name, dev);
4238 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4240 /* If device is running, close it first. */
4243 /* And unlink it from device chain. */
4244 unlist_netdevice(dev);
4246 dev->reg_state = NETREG_UNREGISTERING;
4250 /* Shutdown queueing discipline. */
4254 /* Notify protocols, that we are about to destroy
4255 this device. They should clean all the things.
4257 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4260 * Flush the unicast and multicast chains
4262 dev_addr_discard(dev);
4264 if (dev->netdev_ops->ndo_uninit)
4265 dev->netdev_ops->ndo_uninit(dev);
4267 /* Notifier chain MUST detach us from master device. */
4268 WARN_ON(dev->master);
4270 /* Remove entries from kobject tree */
4271 netdev_unregister_kobject(dev);
4278 static void __netdev_init_queue_locks_one(struct net_device *dev,
4279 struct netdev_queue *dev_queue,
4282 spin_lock_init(&dev_queue->_xmit_lock);
4283 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4284 dev_queue->xmit_lock_owner = -1;
4287 static void netdev_init_queue_locks(struct net_device *dev)
4289 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4290 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4293 unsigned long netdev_fix_features(unsigned long features, const char *name)
4295 /* Fix illegal SG+CSUM combinations. */
4296 if ((features & NETIF_F_SG) &&
4297 !(features & NETIF_F_ALL_CSUM)) {
4299 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4300 "checksum feature.\n", name);
4301 features &= ~NETIF_F_SG;
4304 /* TSO requires that SG is present as well. */
4305 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4307 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4308 "SG feature.\n", name);
4309 features &= ~NETIF_F_TSO;
4312 if (features & NETIF_F_UFO) {
4313 if (!(features & NETIF_F_GEN_CSUM)) {
4315 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4316 "since no NETIF_F_HW_CSUM feature.\n",
4318 features &= ~NETIF_F_UFO;
4321 if (!(features & NETIF_F_SG)) {
4323 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4324 "since no NETIF_F_SG feature.\n", name);
4325 features &= ~NETIF_F_UFO;
4331 EXPORT_SYMBOL(netdev_fix_features);
4334 * register_netdevice - register a network device
4335 * @dev: device to register
4337 * Take a completed network device structure and add it to the kernel
4338 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4339 * chain. 0 is returned on success. A negative errno code is returned
4340 * on a failure to set up the device, or if the name is a duplicate.
4342 * Callers must hold the rtnl semaphore. You may want
4343 * register_netdev() instead of this.
4346 * The locking appears insufficient to guarantee two parallel registers
4347 * will not get the same name.
4350 int register_netdevice(struct net_device *dev)
4352 struct hlist_head *head;
4353 struct hlist_node *p;
4355 struct net *net = dev_net(dev);
4357 BUG_ON(dev_boot_phase);
4362 /* When net_device's are persistent, this will be fatal. */
4363 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4366 spin_lock_init(&dev->addr_list_lock);
4367 netdev_set_addr_lockdep_class(dev);
4368 netdev_init_queue_locks(dev);
4372 #ifdef CONFIG_COMPAT_NET_DEV_OPS
4373 /* Netdevice_ops API compatiability support.
4374 * This is temporary until all network devices are converted.
4376 if (dev->netdev_ops) {
4377 const struct net_device_ops *ops = dev->netdev_ops;
4379 dev->init = ops->ndo_init;
4380 dev->uninit = ops->ndo_uninit;
4381 dev->open = ops->ndo_open;
4382 dev->change_rx_flags = ops->ndo_change_rx_flags;
4383 dev->set_rx_mode = ops->ndo_set_rx_mode;
4384 dev->set_multicast_list = ops->ndo_set_multicast_list;
4385 dev->set_mac_address = ops->ndo_set_mac_address;
4386 dev->validate_addr = ops->ndo_validate_addr;
4387 dev->do_ioctl = ops->ndo_do_ioctl;
4388 dev->set_config = ops->ndo_set_config;
4389 dev->change_mtu = ops->ndo_change_mtu;
4390 dev->tx_timeout = ops->ndo_tx_timeout;
4391 dev->get_stats = ops->ndo_get_stats;
4392 dev->vlan_rx_register = ops->ndo_vlan_rx_register;
4393 dev->vlan_rx_add_vid = ops->ndo_vlan_rx_add_vid;
4394 dev->vlan_rx_kill_vid = ops->ndo_vlan_rx_kill_vid;
4395 #ifdef CONFIG_NET_POLL_CONTROLLER
4396 dev->poll_controller = ops->ndo_poll_controller;
4399 char drivername[64];
4400 pr_info("%s (%s): not using net_device_ops yet\n",
4401 dev->name, netdev_drivername(dev, drivername, 64));
4403 /* This works only because net_device_ops and the
4404 compatiablity structure are the same. */
4405 dev->netdev_ops = (void *) &(dev->init);
4409 /* Init, if this function is available */
4410 if (dev->netdev_ops->ndo_init) {
4411 ret = dev->netdev_ops->ndo_init(dev);
4419 if (!dev_valid_name(dev->name)) {
4424 dev->ifindex = dev_new_index(net);
4425 if (dev->iflink == -1)
4426 dev->iflink = dev->ifindex;
4428 /* Check for existence of name */
4429 head = dev_name_hash(net, dev->name);
4430 hlist_for_each(p, head) {
4431 struct net_device *d
4432 = hlist_entry(p, struct net_device, name_hlist);
4433 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4439 /* Fix illegal checksum combinations */
4440 if ((dev->features & NETIF_F_HW_CSUM) &&
4441 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4442 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4444 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4447 if ((dev->features & NETIF_F_NO_CSUM) &&
4448 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4449 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4451 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4454 dev->features = netdev_fix_features(dev->features, dev->name);
4456 /* Enable software GSO if SG is supported. */
4457 if (dev->features & NETIF_F_SG)
4458 dev->features |= NETIF_F_GSO;
4460 netdev_initialize_kobject(dev);
4461 ret = netdev_register_kobject(dev);
4464 dev->reg_state = NETREG_REGISTERED;
4467 * Default initial state at registry is that the
4468 * device is present.
4471 set_bit(__LINK_STATE_PRESENT, &dev->state);
4473 dev_init_scheduler(dev);
4475 list_netdevice(dev);
4477 /* Notify protocols, that a new device appeared. */
4478 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4479 ret = notifier_to_errno(ret);
4481 rollback_registered(dev);
4482 dev->reg_state = NETREG_UNREGISTERED;
4489 if (dev->netdev_ops->ndo_uninit)
4490 dev->netdev_ops->ndo_uninit(dev);
4495 * init_dummy_netdev - init a dummy network device for NAPI
4496 * @dev: device to init
4498 * This takes a network device structure and initialize the minimum
4499 * amount of fields so it can be used to schedule NAPI polls without
4500 * registering a full blown interface. This is to be used by drivers
4501 * that need to tie several hardware interfaces to a single NAPI
4502 * poll scheduler due to HW limitations.
4504 int init_dummy_netdev(struct net_device *dev)
4506 /* Clear everything. Note we don't initialize spinlocks
4507 * are they aren't supposed to be taken by any of the
4508 * NAPI code and this dummy netdev is supposed to be
4509 * only ever used for NAPI polls
4511 memset(dev, 0, sizeof(struct net_device));
4513 /* make sure we BUG if trying to hit standard
4514 * register/unregister code path
4516 dev->reg_state = NETREG_DUMMY;
4518 /* initialize the ref count */
4519 atomic_set(&dev->refcnt, 1);
4521 /* NAPI wants this */
4522 INIT_LIST_HEAD(&dev->napi_list);
4524 /* a dummy interface is started by default */
4525 set_bit(__LINK_STATE_PRESENT, &dev->state);
4526 set_bit(__LINK_STATE_START, &dev->state);
4530 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4534 * register_netdev - register a network device
4535 * @dev: device to register
4537 * Take a completed network device structure and add it to the kernel
4538 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4539 * chain. 0 is returned on success. A negative errno code is returned
4540 * on a failure to set up the device, or if the name is a duplicate.
4542 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4543 * and expands the device name if you passed a format string to
4546 int register_netdev(struct net_device *dev)
4553 * If the name is a format string the caller wants us to do a
4556 if (strchr(dev->name, '%')) {
4557 err = dev_alloc_name(dev, dev->name);
4562 err = register_netdevice(dev);
4567 EXPORT_SYMBOL(register_netdev);
4570 * netdev_wait_allrefs - wait until all references are gone.
4572 * This is called when unregistering network devices.
4574 * Any protocol or device that holds a reference should register
4575 * for netdevice notification, and cleanup and put back the
4576 * reference if they receive an UNREGISTER event.
4577 * We can get stuck here if buggy protocols don't correctly
4580 static void netdev_wait_allrefs(struct net_device *dev)
4582 unsigned long rebroadcast_time, warning_time;
4584 rebroadcast_time = warning_time = jiffies;
4585 while (atomic_read(&dev->refcnt) != 0) {
4586 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4589 /* Rebroadcast unregister notification */
4590 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4592 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4594 /* We must not have linkwatch events
4595 * pending on unregister. If this
4596 * happens, we simply run the queue
4597 * unscheduled, resulting in a noop
4600 linkwatch_run_queue();
4605 rebroadcast_time = jiffies;
4610 if (time_after(jiffies, warning_time + 10 * HZ)) {
4611 printk(KERN_EMERG "unregister_netdevice: "
4612 "waiting for %s to become free. Usage "
4614 dev->name, atomic_read(&dev->refcnt));
4615 warning_time = jiffies;
4624 * register_netdevice(x1);
4625 * register_netdevice(x2);
4627 * unregister_netdevice(y1);
4628 * unregister_netdevice(y2);
4634 * We are invoked by rtnl_unlock().
4635 * This allows us to deal with problems:
4636 * 1) We can delete sysfs objects which invoke hotplug
4637 * without deadlocking with linkwatch via keventd.
4638 * 2) Since we run with the RTNL semaphore not held, we can sleep
4639 * safely in order to wait for the netdev refcnt to drop to zero.
4641 * We must not return until all unregister events added during
4642 * the interval the lock was held have been completed.
4644 void netdev_run_todo(void)
4646 struct list_head list;
4648 /* Snapshot list, allow later requests */
4649 list_replace_init(&net_todo_list, &list);
4653 while (!list_empty(&list)) {
4654 struct net_device *dev
4655 = list_entry(list.next, struct net_device, todo_list);
4656 list_del(&dev->todo_list);
4658 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4659 printk(KERN_ERR "network todo '%s' but state %d\n",
4660 dev->name, dev->reg_state);
4665 dev->reg_state = NETREG_UNREGISTERED;
4667 on_each_cpu(flush_backlog, dev, 1);
4669 netdev_wait_allrefs(dev);
4672 BUG_ON(atomic_read(&dev->refcnt));
4673 WARN_ON(dev->ip_ptr);
4674 WARN_ON(dev->ip6_ptr);
4675 WARN_ON(dev->dn_ptr);
4677 if (dev->destructor)
4678 dev->destructor(dev);
4680 /* Free network device */
4681 kobject_put(&dev->dev.kobj);
4686 * dev_get_stats - get network device statistics
4687 * @dev: device to get statistics from
4689 * Get network statistics from device. The device driver may provide
4690 * its own method by setting dev->netdev_ops->get_stats; otherwise
4691 * the internal statistics structure is used.
4693 const struct net_device_stats *dev_get_stats(struct net_device *dev)
4695 const struct net_device_ops *ops = dev->netdev_ops;
4697 if (ops->ndo_get_stats)
4698 return ops->ndo_get_stats(dev);
4702 EXPORT_SYMBOL(dev_get_stats);
4704 static void netdev_init_one_queue(struct net_device *dev,
4705 struct netdev_queue *queue,
4711 static void netdev_init_queues(struct net_device *dev)
4713 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4714 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4715 spin_lock_init(&dev->tx_global_lock);
4719 * alloc_netdev_mq - allocate network device
4720 * @sizeof_priv: size of private data to allocate space for
4721 * @name: device name format string
4722 * @setup: callback to initialize device
4723 * @queue_count: the number of subqueues to allocate
4725 * Allocates a struct net_device with private data area for driver use
4726 * and performs basic initialization. Also allocates subquue structs
4727 * for each queue on the device at the end of the netdevice.
4729 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4730 void (*setup)(struct net_device *), unsigned int queue_count)
4732 struct netdev_queue *tx;
4733 struct net_device *dev;
4737 BUG_ON(strlen(name) >= sizeof(dev->name));
4739 alloc_size = sizeof(struct net_device);
4741 /* ensure 32-byte alignment of private area */
4742 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4743 alloc_size += sizeof_priv;
4745 /* ensure 32-byte alignment of whole construct */
4746 alloc_size += NETDEV_ALIGN_CONST;
4748 p = kzalloc(alloc_size, GFP_KERNEL);
4750 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4754 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4756 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4762 dev = (struct net_device *)
4763 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4764 dev->padded = (char *)dev - (char *)p;
4765 dev_net_set(dev, &init_net);
4768 dev->num_tx_queues = queue_count;
4769 dev->real_num_tx_queues = queue_count;
4771 dev->gso_max_size = GSO_MAX_SIZE;
4773 netdev_init_queues(dev);
4775 INIT_LIST_HEAD(&dev->napi_list);
4777 strcpy(dev->name, name);
4780 EXPORT_SYMBOL(alloc_netdev_mq);
4783 * free_netdev - free network device
4786 * This function does the last stage of destroying an allocated device
4787 * interface. The reference to the device object is released.
4788 * If this is the last reference then it will be freed.
4790 void free_netdev(struct net_device *dev)
4792 struct napi_struct *p, *n;
4794 release_net(dev_net(dev));
4798 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
4801 /* Compatibility with error handling in drivers */
4802 if (dev->reg_state == NETREG_UNINITIALIZED) {
4803 kfree((char *)dev - dev->padded);
4807 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4808 dev->reg_state = NETREG_RELEASED;
4810 /* will free via device release */
4811 put_device(&dev->dev);
4815 * synchronize_net - Synchronize with packet receive processing
4817 * Wait for packets currently being received to be done.
4818 * Does not block later packets from starting.
4820 void synchronize_net(void)
4827 * unregister_netdevice - remove device from the kernel
4830 * This function shuts down a device interface and removes it
4831 * from the kernel tables.
4833 * Callers must hold the rtnl semaphore. You may want
4834 * unregister_netdev() instead of this.
4837 void unregister_netdevice(struct net_device *dev)
4841 rollback_registered(dev);
4842 /* Finish processing unregister after unlock */
4847 * unregister_netdev - remove device from the kernel
4850 * This function shuts down a device interface and removes it
4851 * from the kernel tables.
4853 * This is just a wrapper for unregister_netdevice that takes
4854 * the rtnl semaphore. In general you want to use this and not
4855 * unregister_netdevice.
4857 void unregister_netdev(struct net_device *dev)
4860 unregister_netdevice(dev);
4864 EXPORT_SYMBOL(unregister_netdev);
4867 * dev_change_net_namespace - move device to different nethost namespace
4869 * @net: network namespace
4870 * @pat: If not NULL name pattern to try if the current device name
4871 * is already taken in the destination network namespace.
4873 * This function shuts down a device interface and moves it
4874 * to a new network namespace. On success 0 is returned, on
4875 * a failure a netagive errno code is returned.
4877 * Callers must hold the rtnl semaphore.
4880 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4883 const char *destname;
4888 /* Don't allow namespace local devices to be moved. */
4890 if (dev->features & NETIF_F_NETNS_LOCAL)
4894 /* Don't allow real devices to be moved when sysfs
4898 if (dev->dev.parent)
4902 /* Ensure the device has been registrered */
4904 if (dev->reg_state != NETREG_REGISTERED)
4907 /* Get out if there is nothing todo */
4909 if (net_eq(dev_net(dev), net))
4912 /* Pick the destination device name, and ensure
4913 * we can use it in the destination network namespace.
4916 destname = dev->name;
4917 if (__dev_get_by_name(net, destname)) {
4918 /* We get here if we can't use the current device name */
4921 if (!dev_valid_name(pat))
4923 if (strchr(pat, '%')) {
4924 if (__dev_alloc_name(net, pat, buf) < 0)
4929 if (__dev_get_by_name(net, destname))
4934 * And now a mini version of register_netdevice unregister_netdevice.
4937 /* If device is running close it first. */
4940 /* And unlink it from device chain */
4942 unlist_netdevice(dev);
4946 /* Shutdown queueing discipline. */
4949 /* Notify protocols, that we are about to destroy
4950 this device. They should clean all the things.
4952 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4955 * Flush the unicast and multicast chains
4957 dev_addr_discard(dev);
4959 netdev_unregister_kobject(dev);
4961 /* Actually switch the network namespace */
4962 dev_net_set(dev, net);
4964 /* Assign the new device name */
4965 if (destname != dev->name)
4966 strcpy(dev->name, destname);
4968 /* If there is an ifindex conflict assign a new one */
4969 if (__dev_get_by_index(net, dev->ifindex)) {
4970 int iflink = (dev->iflink == dev->ifindex);
4971 dev->ifindex = dev_new_index(net);
4973 dev->iflink = dev->ifindex;
4976 /* Fixup kobjects */
4977 err = netdev_register_kobject(dev);
4980 /* Add the device back in the hashes */
4981 list_netdevice(dev);
4983 /* Notify protocols, that a new device appeared. */
4984 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4992 static int dev_cpu_callback(struct notifier_block *nfb,
4993 unsigned long action,
4996 struct sk_buff **list_skb;
4997 struct Qdisc **list_net;
4998 struct sk_buff *skb;
4999 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5000 struct softnet_data *sd, *oldsd;
5002 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5005 local_irq_disable();
5006 cpu = smp_processor_id();
5007 sd = &per_cpu(softnet_data, cpu);
5008 oldsd = &per_cpu(softnet_data, oldcpu);
5010 /* Find end of our completion_queue. */
5011 list_skb = &sd->completion_queue;
5013 list_skb = &(*list_skb)->next;
5014 /* Append completion queue from offline CPU. */
5015 *list_skb = oldsd->completion_queue;
5016 oldsd->completion_queue = NULL;
5018 /* Find end of our output_queue. */
5019 list_net = &sd->output_queue;
5021 list_net = &(*list_net)->next_sched;
5022 /* Append output queue from offline CPU. */
5023 *list_net = oldsd->output_queue;
5024 oldsd->output_queue = NULL;
5026 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5029 /* Process offline CPU's input_pkt_queue */
5030 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5038 * netdev_increment_features - increment feature set by one
5039 * @all: current feature set
5040 * @one: new feature set
5041 * @mask: mask feature set
5043 * Computes a new feature set after adding a device with feature set
5044 * @one to the master device with current feature set @all. Will not
5045 * enable anything that is off in @mask. Returns the new feature set.
5047 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5050 /* If device needs checksumming, downgrade to it. */
5051 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5052 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5053 else if (mask & NETIF_F_ALL_CSUM) {
5054 /* If one device supports v4/v6 checksumming, set for all. */
5055 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5056 !(all & NETIF_F_GEN_CSUM)) {
5057 all &= ~NETIF_F_ALL_CSUM;
5058 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5061 /* If one device supports hw checksumming, set for all. */
5062 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5063 all &= ~NETIF_F_ALL_CSUM;
5064 all |= NETIF_F_HW_CSUM;
5068 one |= NETIF_F_ALL_CSUM;
5070 one |= all & NETIF_F_ONE_FOR_ALL;
5071 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
5072 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5076 EXPORT_SYMBOL(netdev_increment_features);
5078 static struct hlist_head *netdev_create_hash(void)
5081 struct hlist_head *hash;
5083 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5085 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5086 INIT_HLIST_HEAD(&hash[i]);
5091 /* Initialize per network namespace state */
5092 static int __net_init netdev_init(struct net *net)
5094 INIT_LIST_HEAD(&net->dev_base_head);
5096 net->dev_name_head = netdev_create_hash();
5097 if (net->dev_name_head == NULL)
5100 net->dev_index_head = netdev_create_hash();
5101 if (net->dev_index_head == NULL)
5107 kfree(net->dev_name_head);
5113 * netdev_drivername - network driver for the device
5114 * @dev: network device
5115 * @buffer: buffer for resulting name
5116 * @len: size of buffer
5118 * Determine network driver for device.
5120 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5122 const struct device_driver *driver;
5123 const struct device *parent;
5125 if (len <= 0 || !buffer)
5129 parent = dev->dev.parent;
5134 driver = parent->driver;
5135 if (driver && driver->name)
5136 strlcpy(buffer, driver->name, len);
5140 static void __net_exit netdev_exit(struct net *net)
5142 kfree(net->dev_name_head);
5143 kfree(net->dev_index_head);
5146 static struct pernet_operations __net_initdata netdev_net_ops = {
5147 .init = netdev_init,
5148 .exit = netdev_exit,
5151 static void __net_exit default_device_exit(struct net *net)
5153 struct net_device *dev;
5155 * Push all migratable of the network devices back to the
5156 * initial network namespace
5160 for_each_netdev(net, dev) {
5162 char fb_name[IFNAMSIZ];
5164 /* Ignore unmoveable devices (i.e. loopback) */
5165 if (dev->features & NETIF_F_NETNS_LOCAL)
5168 /* Delete virtual devices */
5169 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5170 dev->rtnl_link_ops->dellink(dev);
5174 /* Push remaing network devices to init_net */
5175 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5176 err = dev_change_net_namespace(dev, &init_net, fb_name);
5178 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5179 __func__, dev->name, err);
5187 static struct pernet_operations __net_initdata default_device_ops = {
5188 .exit = default_device_exit,
5192 * Initialize the DEV module. At boot time this walks the device list and
5193 * unhooks any devices that fail to initialise (normally hardware not
5194 * present) and leaves us with a valid list of present and active devices.
5199 * This is called single threaded during boot, so no need
5200 * to take the rtnl semaphore.
5202 static int __init net_dev_init(void)
5204 int i, rc = -ENOMEM;
5206 BUG_ON(!dev_boot_phase);
5208 if (dev_proc_init())
5211 if (netdev_kobject_init())
5214 INIT_LIST_HEAD(&ptype_all);
5215 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5216 INIT_LIST_HEAD(&ptype_base[i]);
5218 if (register_pernet_subsys(&netdev_net_ops))
5222 * Initialise the packet receive queues.
5225 for_each_possible_cpu(i) {
5226 struct softnet_data *queue;
5228 queue = &per_cpu(softnet_data, i);
5229 skb_queue_head_init(&queue->input_pkt_queue);
5230 queue->completion_queue = NULL;
5231 INIT_LIST_HEAD(&queue->poll_list);
5233 queue->backlog.poll = process_backlog;
5234 queue->backlog.weight = weight_p;
5235 queue->backlog.gro_list = NULL;
5236 queue->backlog.gro_count = 0;
5241 /* The loopback device is special if any other network devices
5242 * is present in a network namespace the loopback device must
5243 * be present. Since we now dynamically allocate and free the
5244 * loopback device ensure this invariant is maintained by
5245 * keeping the loopback device as the first device on the
5246 * list of network devices. Ensuring the loopback devices
5247 * is the first device that appears and the last network device
5250 if (register_pernet_device(&loopback_net_ops))
5253 if (register_pernet_device(&default_device_ops))
5256 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5257 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5259 hotcpu_notifier(dev_cpu_callback, 0);
5267 subsys_initcall(net_dev_init);
5269 static int __init initialize_hashrnd(void)
5271 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5275 late_initcall_sync(initialize_hashrnd);
5277 EXPORT_SYMBOL(__dev_get_by_index);
5278 EXPORT_SYMBOL(__dev_get_by_name);
5279 EXPORT_SYMBOL(__dev_remove_pack);
5280 EXPORT_SYMBOL(dev_valid_name);
5281 EXPORT_SYMBOL(dev_add_pack);
5282 EXPORT_SYMBOL(dev_alloc_name);
5283 EXPORT_SYMBOL(dev_close);
5284 EXPORT_SYMBOL(dev_get_by_flags);
5285 EXPORT_SYMBOL(dev_get_by_index);
5286 EXPORT_SYMBOL(dev_get_by_name);
5287 EXPORT_SYMBOL(dev_open);
5288 EXPORT_SYMBOL(dev_queue_xmit);
5289 EXPORT_SYMBOL(dev_remove_pack);
5290 EXPORT_SYMBOL(dev_set_allmulti);
5291 EXPORT_SYMBOL(dev_set_promiscuity);
5292 EXPORT_SYMBOL(dev_change_flags);
5293 EXPORT_SYMBOL(dev_set_mtu);
5294 EXPORT_SYMBOL(dev_set_mac_address);
5295 EXPORT_SYMBOL(free_netdev);
5296 EXPORT_SYMBOL(netdev_boot_setup_check);
5297 EXPORT_SYMBOL(netdev_set_master);
5298 EXPORT_SYMBOL(netdev_state_change);
5299 EXPORT_SYMBOL(netif_receive_skb);
5300 EXPORT_SYMBOL(netif_rx);
5301 EXPORT_SYMBOL(register_gifconf);
5302 EXPORT_SYMBOL(register_netdevice);
5303 EXPORT_SYMBOL(register_netdevice_notifier);
5304 EXPORT_SYMBOL(skb_checksum_help);
5305 EXPORT_SYMBOL(synchronize_net);
5306 EXPORT_SYMBOL(unregister_netdevice);
5307 EXPORT_SYMBOL(unregister_netdevice_notifier);
5308 EXPORT_SYMBOL(net_enable_timestamp);
5309 EXPORT_SYMBOL(net_disable_timestamp);
5310 EXPORT_SYMBOL(dev_get_flags);
5312 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
5313 EXPORT_SYMBOL(br_handle_frame_hook);
5314 EXPORT_SYMBOL(br_fdb_get_hook);
5315 EXPORT_SYMBOL(br_fdb_put_hook);
5318 EXPORT_SYMBOL(dev_load);
5320 EXPORT_PER_CPU_SYMBOL(softnet_data);