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)
139 * The list of packet types we will receive (as opposed to discard)
140 * and the routines to invoke.
142 * Why 16. Because with 16 the only overlap we get on a hash of the
143 * low nibble of the protocol value is RARP/SNAP/X.25.
145 * NOTE: That is no longer true with the addition of VLAN tags. Not
146 * sure which should go first, but I bet it won't make much
147 * difference if we are running VLANs. The good news is that
148 * this protocol won't be in the list unless compiled in, so
149 * the average user (w/out VLANs) will not be adversely affected.
166 #define PTYPE_HASH_SIZE (16)
167 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
169 static DEFINE_SPINLOCK(ptype_lock);
170 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
171 static struct list_head ptype_all __read_mostly; /* Taps */
174 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
177 * Pure readers hold dev_base_lock for reading.
179 * Writers must hold the rtnl semaphore while they loop through the
180 * dev_base_head list, and hold dev_base_lock for writing when they do the
181 * actual updates. This allows pure readers to access the list even
182 * while a writer is preparing to update it.
184 * To put it another way, dev_base_lock is held for writing only to
185 * protect against pure readers; the rtnl semaphore provides the
186 * protection against other writers.
188 * See, for example usages, register_netdevice() and
189 * unregister_netdevice(), which must be called with the rtnl
192 DEFINE_RWLOCK(dev_base_lock);
194 EXPORT_SYMBOL(dev_base_lock);
196 #define NETDEV_HASHBITS 8
197 #define NETDEV_HASHENTRIES (1 << NETDEV_HASHBITS)
199 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
201 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
202 return &net->dev_name_head[hash & ((1 << NETDEV_HASHBITS) - 1)];
205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 return &net->dev_index_head[ifindex & ((1 << NETDEV_HASHBITS) - 1)];
210 /* Device list insertion */
211 static int list_netdevice(struct net_device *dev)
213 struct net *net = dev_net(dev);
217 write_lock_bh(&dev_base_lock);
218 list_add_tail(&dev->dev_list, &net->dev_base_head);
219 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
220 hlist_add_head(&dev->index_hlist, dev_index_hash(net, dev->ifindex));
221 write_unlock_bh(&dev_base_lock);
225 /* Device list removal */
226 static void unlist_netdevice(struct net_device *dev)
230 /* Unlink dev from the device chain */
231 write_lock_bh(&dev_base_lock);
232 list_del(&dev->dev_list);
233 hlist_del(&dev->name_hlist);
234 hlist_del(&dev->index_hlist);
235 write_unlock_bh(&dev_base_lock);
242 static RAW_NOTIFIER_HEAD(netdev_chain);
245 * Device drivers call our routines to queue packets here. We empty the
246 * queue in the local softnet handler.
249 DEFINE_PER_CPU(struct softnet_data, softnet_data);
251 #ifdef CONFIG_LOCKDEP
253 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
254 * according to dev->type
256 static const unsigned short netdev_lock_type[] =
257 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
258 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
259 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
260 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
261 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
262 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
263 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
264 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
265 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
266 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
267 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
268 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
269 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
270 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
271 ARPHRD_PHONET_PIPE, ARPHRD_VOID, ARPHRD_NONE};
273 static const char *netdev_lock_name[] =
274 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
275 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
276 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
277 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
278 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
279 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
280 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
281 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
282 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
283 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
284 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
285 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
286 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
287 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
288 "_xmit_PHONET_PIPE", "_xmit_VOID", "_xmit_NONE"};
290 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
291 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
293 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
297 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
298 if (netdev_lock_type[i] == dev_type)
300 /* the last key is used by default */
301 return ARRAY_SIZE(netdev_lock_type) - 1;
304 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
305 unsigned short dev_type)
309 i = netdev_lock_pos(dev_type);
310 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
311 netdev_lock_name[i]);
314 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
318 i = netdev_lock_pos(dev->type);
319 lockdep_set_class_and_name(&dev->addr_list_lock,
320 &netdev_addr_lock_key[i],
321 netdev_lock_name[i]);
324 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
325 unsigned short dev_type)
328 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
333 /*******************************************************************************
335 Protocol management and registration routines
337 *******************************************************************************/
340 * Add a protocol ID to the list. Now that the input handler is
341 * smarter we can dispense with all the messy stuff that used to be
344 * BEWARE!!! Protocol handlers, mangling input packets,
345 * MUST BE last in hash buckets and checking protocol handlers
346 * MUST start from promiscuous ptype_all chain in net_bh.
347 * It is true now, do not change it.
348 * Explanation follows: if protocol handler, mangling packet, will
349 * be the first on list, it is not able to sense, that packet
350 * is cloned and should be copied-on-write, so that it will
351 * change it and subsequent readers will get broken packet.
356 * dev_add_pack - add packet handler
357 * @pt: packet type declaration
359 * Add a protocol handler to the networking stack. The passed &packet_type
360 * is linked into kernel lists and may not be freed until it has been
361 * removed from the kernel lists.
363 * This call does not sleep therefore it can not
364 * guarantee all CPU's that are in middle of receiving packets
365 * will see the new packet type (until the next received packet).
368 void dev_add_pack(struct packet_type *pt)
372 spin_lock_bh(&ptype_lock);
373 if (pt->type == htons(ETH_P_ALL))
374 list_add_rcu(&pt->list, &ptype_all);
376 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
377 list_add_rcu(&pt->list, &ptype_base[hash]);
379 spin_unlock_bh(&ptype_lock);
383 * __dev_remove_pack - remove packet handler
384 * @pt: packet type declaration
386 * Remove a protocol handler that was previously added to the kernel
387 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
388 * from the kernel lists and can be freed or reused once this function
391 * The packet type might still be in use by receivers
392 * and must not be freed until after all the CPU's have gone
393 * through a quiescent state.
395 void __dev_remove_pack(struct packet_type *pt)
397 struct list_head *head;
398 struct packet_type *pt1;
400 spin_lock_bh(&ptype_lock);
402 if (pt->type == htons(ETH_P_ALL))
405 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
407 list_for_each_entry(pt1, head, list) {
409 list_del_rcu(&pt->list);
414 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
416 spin_unlock_bh(&ptype_lock);
419 * dev_remove_pack - remove packet handler
420 * @pt: packet type declaration
422 * Remove a protocol handler that was previously added to the kernel
423 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
424 * from the kernel lists and can be freed or reused once this function
427 * This call sleeps to guarantee that no CPU is looking at the packet
430 void dev_remove_pack(struct packet_type *pt)
432 __dev_remove_pack(pt);
437 /******************************************************************************
439 Device Boot-time Settings Routines
441 *******************************************************************************/
443 /* Boot time configuration table */
444 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
447 * netdev_boot_setup_add - add new setup entry
448 * @name: name of the device
449 * @map: configured settings for the device
451 * Adds new setup entry to the dev_boot_setup list. The function
452 * returns 0 on error and 1 on success. This is a generic routine to
455 static int netdev_boot_setup_add(char *name, struct ifmap *map)
457 struct netdev_boot_setup *s;
461 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
462 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
463 memset(s[i].name, 0, sizeof(s[i].name));
464 strlcpy(s[i].name, name, IFNAMSIZ);
465 memcpy(&s[i].map, map, sizeof(s[i].map));
470 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
474 * netdev_boot_setup_check - check boot time settings
475 * @dev: the netdevice
477 * Check boot time settings for the device.
478 * The found settings are set for the device to be used
479 * later in the device probing.
480 * Returns 0 if no settings found, 1 if they are.
482 int netdev_boot_setup_check(struct net_device *dev)
484 struct netdev_boot_setup *s = dev_boot_setup;
487 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
488 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
489 !strcmp(dev->name, s[i].name)) {
490 dev->irq = s[i].map.irq;
491 dev->base_addr = s[i].map.base_addr;
492 dev->mem_start = s[i].map.mem_start;
493 dev->mem_end = s[i].map.mem_end;
502 * netdev_boot_base - get address from boot time settings
503 * @prefix: prefix for network device
504 * @unit: id for network device
506 * Check boot time settings for the base address of device.
507 * The found settings are set for the device to be used
508 * later in the device probing.
509 * Returns 0 if no settings found.
511 unsigned long netdev_boot_base(const char *prefix, int unit)
513 const struct netdev_boot_setup *s = dev_boot_setup;
517 sprintf(name, "%s%d", prefix, unit);
520 * If device already registered then return base of 1
521 * to indicate not to probe for this interface
523 if (__dev_get_by_name(&init_net, name))
526 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
527 if (!strcmp(name, s[i].name))
528 return s[i].map.base_addr;
533 * Saves at boot time configured settings for any netdevice.
535 int __init netdev_boot_setup(char *str)
540 str = get_options(str, ARRAY_SIZE(ints), ints);
545 memset(&map, 0, sizeof(map));
549 map.base_addr = ints[2];
551 map.mem_start = ints[3];
553 map.mem_end = ints[4];
555 /* Add new entry to the list */
556 return netdev_boot_setup_add(str, &map);
559 __setup("netdev=", netdev_boot_setup);
561 /*******************************************************************************
563 Device Interface Subroutines
565 *******************************************************************************/
568 * __dev_get_by_name - find a device by its name
569 * @net: the applicable net namespace
570 * @name: name to find
572 * Find an interface by name. Must be called under RTNL semaphore
573 * or @dev_base_lock. If the name is found a pointer to the device
574 * is returned. If the name is not found then %NULL is returned. The
575 * reference counters are not incremented so the caller must be
576 * careful with locks.
579 struct net_device *__dev_get_by_name(struct net *net, const char *name)
581 struct hlist_node *p;
583 hlist_for_each(p, dev_name_hash(net, name)) {
584 struct net_device *dev
585 = hlist_entry(p, struct net_device, name_hlist);
586 if (!strncmp(dev->name, name, IFNAMSIZ))
593 * dev_get_by_name - find a device by its name
594 * @net: the applicable net namespace
595 * @name: name to find
597 * Find an interface by name. This can be called from any
598 * context and does its own locking. The returned handle has
599 * the usage count incremented and the caller must use dev_put() to
600 * release it when it is no longer needed. %NULL is returned if no
601 * matching device is found.
604 struct net_device *dev_get_by_name(struct net *net, const char *name)
606 struct net_device *dev;
608 read_lock(&dev_base_lock);
609 dev = __dev_get_by_name(net, name);
612 read_unlock(&dev_base_lock);
617 * __dev_get_by_index - find a device by its ifindex
618 * @net: the applicable net namespace
619 * @ifindex: index of device
621 * Search for an interface by index. Returns %NULL if the device
622 * is not found or a pointer to the device. The device has not
623 * had its reference counter increased so the caller must be careful
624 * about locking. The caller must hold either the RTNL semaphore
628 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
630 struct hlist_node *p;
632 hlist_for_each(p, dev_index_hash(net, ifindex)) {
633 struct net_device *dev
634 = hlist_entry(p, struct net_device, index_hlist);
635 if (dev->ifindex == ifindex)
643 * dev_get_by_index - find a device by its ifindex
644 * @net: the applicable net namespace
645 * @ifindex: index of device
647 * Search for an interface by index. Returns NULL if the device
648 * is not found or a pointer to the device. The device returned has
649 * had a reference added and the pointer is safe until the user calls
650 * dev_put to indicate they have finished with it.
653 struct net_device *dev_get_by_index(struct net *net, int ifindex)
655 struct net_device *dev;
657 read_lock(&dev_base_lock);
658 dev = __dev_get_by_index(net, ifindex);
661 read_unlock(&dev_base_lock);
666 * dev_getbyhwaddr - find a device by its hardware address
667 * @net: the applicable net namespace
668 * @type: media type of device
669 * @ha: hardware address
671 * Search for an interface by MAC address. Returns NULL if the device
672 * is not found or a pointer to the device. The caller must hold the
673 * rtnl semaphore. The returned device has not had its ref count increased
674 * and the caller must therefore be careful about locking
677 * If the API was consistent this would be __dev_get_by_hwaddr
680 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
682 struct net_device *dev;
686 for_each_netdev(net, dev)
687 if (dev->type == type &&
688 !memcmp(dev->dev_addr, ha, dev->addr_len))
694 EXPORT_SYMBOL(dev_getbyhwaddr);
696 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
698 struct net_device *dev;
701 for_each_netdev(net, dev)
702 if (dev->type == type)
708 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
710 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
712 struct net_device *dev;
715 dev = __dev_getfirstbyhwtype(net, type);
722 EXPORT_SYMBOL(dev_getfirstbyhwtype);
725 * dev_get_by_flags - find any device with given flags
726 * @net: the applicable net namespace
727 * @if_flags: IFF_* values
728 * @mask: bitmask of bits in if_flags to check
730 * Search for any interface with the given flags. Returns NULL if a device
731 * is not found or a pointer to the device. The device returned has
732 * had a reference added and the pointer is safe until the user calls
733 * dev_put to indicate they have finished with it.
736 struct net_device * dev_get_by_flags(struct net *net, unsigned short if_flags, unsigned short mask)
738 struct net_device *dev, *ret;
741 read_lock(&dev_base_lock);
742 for_each_netdev(net, dev) {
743 if (((dev->flags ^ if_flags) & mask) == 0) {
749 read_unlock(&dev_base_lock);
754 * dev_valid_name - check if name is okay for network device
757 * Network device names need to be valid file names to
758 * to allow sysfs to work. We also disallow any kind of
761 int dev_valid_name(const char *name)
765 if (strlen(name) >= IFNAMSIZ)
767 if (!strcmp(name, ".") || !strcmp(name, ".."))
771 if (*name == '/' || isspace(*name))
779 * __dev_alloc_name - allocate a name for a device
780 * @net: network namespace to allocate the device name in
781 * @name: name format string
782 * @buf: scratch buffer and result name string
784 * Passed a format string - eg "lt%d" it will try and find a suitable
785 * id. It scans list of devices to build up a free map, then chooses
786 * the first empty slot. The caller must hold the dev_base or rtnl lock
787 * while allocating the name and adding the device in order to avoid
789 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
790 * Returns the number of the unit assigned or a negative errno code.
793 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
797 const int max_netdevices = 8*PAGE_SIZE;
798 unsigned long *inuse;
799 struct net_device *d;
801 p = strnchr(name, IFNAMSIZ-1, '%');
804 * Verify the string as this thing may have come from
805 * the user. There must be either one "%d" and no other "%"
808 if (p[1] != 'd' || strchr(p + 2, '%'))
811 /* Use one page as a bit array of possible slots */
812 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
816 for_each_netdev(net, d) {
817 if (!sscanf(d->name, name, &i))
819 if (i < 0 || i >= max_netdevices)
822 /* avoid cases where sscanf is not exact inverse of printf */
823 snprintf(buf, IFNAMSIZ, name, i);
824 if (!strncmp(buf, d->name, IFNAMSIZ))
828 i = find_first_zero_bit(inuse, max_netdevices);
829 free_page((unsigned long) inuse);
832 snprintf(buf, IFNAMSIZ, name, i);
833 if (!__dev_get_by_name(net, buf))
836 /* It is possible to run out of possible slots
837 * when the name is long and there isn't enough space left
838 * for the digits, or if all bits are used.
844 * dev_alloc_name - allocate a name for a device
846 * @name: name format string
848 * Passed a format string - eg "lt%d" it will try and find a suitable
849 * id. It scans list of devices to build up a free map, then chooses
850 * the first empty slot. The caller must hold the dev_base or rtnl lock
851 * while allocating the name and adding the device in order to avoid
853 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
854 * Returns the number of the unit assigned or a negative errno code.
857 int dev_alloc_name(struct net_device *dev, const char *name)
863 BUG_ON(!dev_net(dev));
865 ret = __dev_alloc_name(net, name, buf);
867 strlcpy(dev->name, buf, IFNAMSIZ);
873 * dev_change_name - change name of a device
875 * @newname: name (or format string) must be at least IFNAMSIZ
877 * Change name of a device, can pass format strings "eth%d".
880 int dev_change_name(struct net_device *dev, const char *newname)
882 char oldname[IFNAMSIZ];
888 BUG_ON(!dev_net(dev));
891 if (dev->flags & IFF_UP)
894 if (!dev_valid_name(newname))
897 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
900 memcpy(oldname, dev->name, IFNAMSIZ);
902 if (strchr(newname, '%')) {
903 err = dev_alloc_name(dev, newname);
907 else if (__dev_get_by_name(net, newname))
910 strlcpy(dev->name, newname, IFNAMSIZ);
913 /* For now only devices in the initial network namespace
916 if (net == &init_net) {
917 ret = device_rename(&dev->dev, dev->name);
919 memcpy(dev->name, oldname, IFNAMSIZ);
924 write_lock_bh(&dev_base_lock);
925 hlist_del(&dev->name_hlist);
926 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
927 write_unlock_bh(&dev_base_lock);
929 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
930 ret = notifier_to_errno(ret);
935 "%s: name change rollback failed: %d.\n",
939 memcpy(dev->name, oldname, IFNAMSIZ);
948 * dev_set_alias - change ifalias of a device
950 * @alias: name up to IFALIASZ
951 * @len: limit of bytes to copy from info
953 * Set ifalias for a device,
955 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
970 dev->ifalias = krealloc(dev->ifalias, len+1, GFP_KERNEL);
974 strlcpy(dev->ifalias, alias, len+1);
980 * netdev_features_change - device changes features
981 * @dev: device to cause notification
983 * Called to indicate a device has changed features.
985 void netdev_features_change(struct net_device *dev)
987 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
989 EXPORT_SYMBOL(netdev_features_change);
992 * netdev_state_change - device changes state
993 * @dev: device to cause notification
995 * Called to indicate a device has changed state. This function calls
996 * the notifier chains for netdev_chain and sends a NEWLINK message
997 * to the routing socket.
999 void netdev_state_change(struct net_device *dev)
1001 if (dev->flags & IFF_UP) {
1002 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1003 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1007 void netdev_bonding_change(struct net_device *dev)
1009 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, dev);
1011 EXPORT_SYMBOL(netdev_bonding_change);
1014 * dev_load - load a network module
1015 * @net: the applicable net namespace
1016 * @name: name of interface
1018 * If a network interface is not present and the process has suitable
1019 * privileges this function loads the module. If module loading is not
1020 * available in this kernel then it becomes a nop.
1023 void dev_load(struct net *net, const char *name)
1025 struct net_device *dev;
1027 read_lock(&dev_base_lock);
1028 dev = __dev_get_by_name(net, name);
1029 read_unlock(&dev_base_lock);
1031 if (!dev && capable(CAP_SYS_MODULE))
1032 request_module("%s", name);
1036 * dev_open - prepare an interface for use.
1037 * @dev: device to open
1039 * Takes a device from down to up state. The device's private open
1040 * function is invoked and then the multicast lists are loaded. Finally
1041 * the device is moved into the up state and a %NETDEV_UP message is
1042 * sent to the netdev notifier chain.
1044 * Calling this function on an active interface is a nop. On a failure
1045 * a negative errno code is returned.
1047 int dev_open(struct net_device *dev)
1049 const struct net_device_ops *ops = dev->netdev_ops;
1058 if (dev->flags & IFF_UP)
1062 * Is it even present?
1064 if (!netif_device_present(dev))
1068 * Call device private open method
1070 set_bit(__LINK_STATE_START, &dev->state);
1072 if (ops->ndo_validate_addr)
1073 ret = ops->ndo_validate_addr(dev);
1075 if (!ret && ops->ndo_open)
1076 ret = ops->ndo_open(dev);
1079 * If it went open OK then:
1083 clear_bit(__LINK_STATE_START, &dev->state);
1088 dev->flags |= IFF_UP;
1093 net_dmaengine_get();
1096 * Initialize multicasting status
1098 dev_set_rx_mode(dev);
1101 * Wakeup transmit queue engine
1106 * ... and announce new interface.
1108 call_netdevice_notifiers(NETDEV_UP, dev);
1115 * dev_close - shutdown an interface.
1116 * @dev: device to shutdown
1118 * This function moves an active device into down state. A
1119 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1120 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1123 int dev_close(struct net_device *dev)
1125 const struct net_device_ops *ops = dev->netdev_ops;
1130 if (!(dev->flags & IFF_UP))
1134 * Tell people we are going down, so that they can
1135 * prepare to death, when device is still operating.
1137 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1139 clear_bit(__LINK_STATE_START, &dev->state);
1141 /* Synchronize to scheduled poll. We cannot touch poll list,
1142 * it can be even on different cpu. So just clear netif_running().
1144 * dev->stop() will invoke napi_disable() on all of it's
1145 * napi_struct instances on this device.
1147 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1149 dev_deactivate(dev);
1152 * Call the device specific close. This cannot fail.
1153 * Only if device is UP
1155 * We allow it to be called even after a DETACH hot-plug
1162 * Device is now down.
1165 dev->flags &= ~IFF_UP;
1168 * Tell people we are down
1170 call_netdevice_notifiers(NETDEV_DOWN, dev);
1175 net_dmaengine_put();
1182 * dev_disable_lro - disable Large Receive Offload on a device
1185 * Disable Large Receive Offload (LRO) on a net device. Must be
1186 * called under RTNL. This is needed if received packets may be
1187 * forwarded to another interface.
1189 void dev_disable_lro(struct net_device *dev)
1191 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1192 dev->ethtool_ops->set_flags) {
1193 u32 flags = dev->ethtool_ops->get_flags(dev);
1194 if (flags & ETH_FLAG_LRO) {
1195 flags &= ~ETH_FLAG_LRO;
1196 dev->ethtool_ops->set_flags(dev, flags);
1199 WARN_ON(dev->features & NETIF_F_LRO);
1201 EXPORT_SYMBOL(dev_disable_lro);
1204 static int dev_boot_phase = 1;
1207 * Device change register/unregister. These are not inline or static
1208 * as we export them to the world.
1212 * register_netdevice_notifier - register a network notifier block
1215 * Register a notifier to be called when network device events occur.
1216 * The notifier passed is linked into the kernel structures and must
1217 * not be reused until it has been unregistered. A negative errno code
1218 * is returned on a failure.
1220 * When registered all registration and up events are replayed
1221 * to the new notifier to allow device to have a race free
1222 * view of the network device list.
1225 int register_netdevice_notifier(struct notifier_block *nb)
1227 struct net_device *dev;
1228 struct net_device *last;
1233 err = raw_notifier_chain_register(&netdev_chain, nb);
1239 for_each_netdev(net, dev) {
1240 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1241 err = notifier_to_errno(err);
1245 if (!(dev->flags & IFF_UP))
1248 nb->notifier_call(nb, NETDEV_UP, dev);
1259 for_each_netdev(net, dev) {
1263 if (dev->flags & IFF_UP) {
1264 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1265 nb->notifier_call(nb, NETDEV_DOWN, dev);
1267 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1271 raw_notifier_chain_unregister(&netdev_chain, nb);
1276 * unregister_netdevice_notifier - unregister a network notifier block
1279 * Unregister a notifier previously registered by
1280 * register_netdevice_notifier(). The notifier is unlinked into the
1281 * kernel structures and may then be reused. A negative errno code
1282 * is returned on a failure.
1285 int unregister_netdevice_notifier(struct notifier_block *nb)
1290 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1296 * call_netdevice_notifiers - call all network notifier blocks
1297 * @val: value passed unmodified to notifier function
1298 * @dev: net_device pointer passed unmodified to notifier function
1300 * Call all network notifier blocks. Parameters and return value
1301 * are as for raw_notifier_call_chain().
1304 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1306 return raw_notifier_call_chain(&netdev_chain, val, dev);
1309 /* When > 0 there are consumers of rx skb time stamps */
1310 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1312 void net_enable_timestamp(void)
1314 atomic_inc(&netstamp_needed);
1317 void net_disable_timestamp(void)
1319 atomic_dec(&netstamp_needed);
1322 static inline void net_timestamp(struct sk_buff *skb)
1324 if (atomic_read(&netstamp_needed))
1325 __net_timestamp(skb);
1327 skb->tstamp.tv64 = 0;
1331 * Support routine. Sends outgoing frames to any network
1332 * taps currently in use.
1335 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1337 struct packet_type *ptype;
1339 #ifdef CONFIG_NET_CLS_ACT
1340 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1347 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1348 /* Never send packets back to the socket
1349 * they originated from - MvS (miquels@drinkel.ow.org)
1351 if ((ptype->dev == dev || !ptype->dev) &&
1352 (ptype->af_packet_priv == NULL ||
1353 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1354 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1358 /* skb->nh should be correctly
1359 set by sender, so that the second statement is
1360 just protection against buggy protocols.
1362 skb_reset_mac_header(skb2);
1364 if (skb_network_header(skb2) < skb2->data ||
1365 skb2->network_header > skb2->tail) {
1366 if (net_ratelimit())
1367 printk(KERN_CRIT "protocol %04x is "
1369 skb2->protocol, dev->name);
1370 skb_reset_network_header(skb2);
1373 skb2->transport_header = skb2->network_header;
1374 skb2->pkt_type = PACKET_OUTGOING;
1375 ptype->func(skb2, skb->dev, ptype, skb->dev);
1382 static inline void __netif_reschedule(struct Qdisc *q)
1384 struct softnet_data *sd;
1385 unsigned long flags;
1387 local_irq_save(flags);
1388 sd = &__get_cpu_var(softnet_data);
1389 q->next_sched = sd->output_queue;
1390 sd->output_queue = q;
1391 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1392 local_irq_restore(flags);
1395 void __netif_schedule(struct Qdisc *q)
1397 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1398 __netif_reschedule(q);
1400 EXPORT_SYMBOL(__netif_schedule);
1402 void dev_kfree_skb_irq(struct sk_buff *skb)
1404 if (atomic_dec_and_test(&skb->users)) {
1405 struct softnet_data *sd;
1406 unsigned long flags;
1408 local_irq_save(flags);
1409 sd = &__get_cpu_var(softnet_data);
1410 skb->next = sd->completion_queue;
1411 sd->completion_queue = skb;
1412 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1413 local_irq_restore(flags);
1416 EXPORT_SYMBOL(dev_kfree_skb_irq);
1418 void dev_kfree_skb_any(struct sk_buff *skb)
1420 if (in_irq() || irqs_disabled())
1421 dev_kfree_skb_irq(skb);
1425 EXPORT_SYMBOL(dev_kfree_skb_any);
1429 * netif_device_detach - mark device as removed
1430 * @dev: network device
1432 * Mark device as removed from system and therefore no longer available.
1434 void netif_device_detach(struct net_device *dev)
1436 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1437 netif_running(dev)) {
1438 netif_tx_stop_all_queues(dev);
1441 EXPORT_SYMBOL(netif_device_detach);
1444 * netif_device_attach - mark device as attached
1445 * @dev: network device
1447 * Mark device as attached from system and restart if needed.
1449 void netif_device_attach(struct net_device *dev)
1451 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1452 netif_running(dev)) {
1453 netif_tx_wake_all_queues(dev);
1454 __netdev_watchdog_up(dev);
1457 EXPORT_SYMBOL(netif_device_attach);
1459 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1461 return ((features & NETIF_F_GEN_CSUM) ||
1462 ((features & NETIF_F_IP_CSUM) &&
1463 protocol == htons(ETH_P_IP)) ||
1464 ((features & NETIF_F_IPV6_CSUM) &&
1465 protocol == htons(ETH_P_IPV6)) ||
1466 ((features & NETIF_F_FCOE_CRC) &&
1467 protocol == htons(ETH_P_FCOE)));
1470 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1472 if (can_checksum_protocol(dev->features, skb->protocol))
1475 if (skb->protocol == htons(ETH_P_8021Q)) {
1476 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1477 if (can_checksum_protocol(dev->features & dev->vlan_features,
1478 veh->h_vlan_encapsulated_proto))
1486 * Invalidate hardware checksum when packet is to be mangled, and
1487 * complete checksum manually on outgoing path.
1489 int skb_checksum_help(struct sk_buff *skb)
1492 int ret = 0, offset;
1494 if (skb->ip_summed == CHECKSUM_COMPLETE)
1495 goto out_set_summed;
1497 if (unlikely(skb_shinfo(skb)->gso_size)) {
1498 /* Let GSO fix up the checksum. */
1499 goto out_set_summed;
1502 offset = skb->csum_start - skb_headroom(skb);
1503 BUG_ON(offset >= skb_headlen(skb));
1504 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1506 offset += skb->csum_offset;
1507 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1509 if (skb_cloned(skb) &&
1510 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1511 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1516 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1518 skb->ip_summed = CHECKSUM_NONE;
1524 * skb_gso_segment - Perform segmentation on skb.
1525 * @skb: buffer to segment
1526 * @features: features for the output path (see dev->features)
1528 * This function segments the given skb and returns a list of segments.
1530 * It may return NULL if the skb requires no segmentation. This is
1531 * only possible when GSO is used for verifying header integrity.
1533 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1535 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1536 struct packet_type *ptype;
1537 __be16 type = skb->protocol;
1540 skb_reset_mac_header(skb);
1541 skb->mac_len = skb->network_header - skb->mac_header;
1542 __skb_pull(skb, skb->mac_len);
1544 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1545 struct net_device *dev = skb->dev;
1546 struct ethtool_drvinfo info = {};
1548 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1549 dev->ethtool_ops->get_drvinfo(dev, &info);
1551 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1553 info.driver, dev ? dev->features : 0L,
1554 skb->sk ? skb->sk->sk_route_caps : 0L,
1555 skb->len, skb->data_len, skb->ip_summed);
1557 if (skb_header_cloned(skb) &&
1558 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1559 return ERR_PTR(err);
1563 list_for_each_entry_rcu(ptype,
1564 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1565 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1566 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1567 err = ptype->gso_send_check(skb);
1568 segs = ERR_PTR(err);
1569 if (err || skb_gso_ok(skb, features))
1571 __skb_push(skb, (skb->data -
1572 skb_network_header(skb)));
1574 segs = ptype->gso_segment(skb, features);
1580 __skb_push(skb, skb->data - skb_mac_header(skb));
1585 EXPORT_SYMBOL(skb_gso_segment);
1587 /* Take action when hardware reception checksum errors are detected. */
1589 void netdev_rx_csum_fault(struct net_device *dev)
1591 if (net_ratelimit()) {
1592 printk(KERN_ERR "%s: hw csum failure.\n",
1593 dev ? dev->name : "<unknown>");
1597 EXPORT_SYMBOL(netdev_rx_csum_fault);
1600 /* Actually, we should eliminate this check as soon as we know, that:
1601 * 1. IOMMU is present and allows to map all the memory.
1602 * 2. No high memory really exists on this machine.
1605 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1607 #ifdef CONFIG_HIGHMEM
1610 if (dev->features & NETIF_F_HIGHDMA)
1613 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1614 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1622 void (*destructor)(struct sk_buff *skb);
1625 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1627 static void dev_gso_skb_destructor(struct sk_buff *skb)
1629 struct dev_gso_cb *cb;
1632 struct sk_buff *nskb = skb->next;
1634 skb->next = nskb->next;
1637 } while (skb->next);
1639 cb = DEV_GSO_CB(skb);
1641 cb->destructor(skb);
1645 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1646 * @skb: buffer to segment
1648 * This function segments the given skb and stores the list of segments
1651 static int dev_gso_segment(struct sk_buff *skb)
1653 struct net_device *dev = skb->dev;
1654 struct sk_buff *segs;
1655 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1658 segs = skb_gso_segment(skb, features);
1660 /* Verifying header integrity only. */
1665 return PTR_ERR(segs);
1668 DEV_GSO_CB(skb)->destructor = skb->destructor;
1669 skb->destructor = dev_gso_skb_destructor;
1674 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1675 struct netdev_queue *txq)
1677 const struct net_device_ops *ops = dev->netdev_ops;
1680 if (likely(!skb->next)) {
1681 if (!list_empty(&ptype_all))
1682 dev_queue_xmit_nit(skb, dev);
1684 if (netif_needs_gso(dev, skb)) {
1685 if (unlikely(dev_gso_segment(skb)))
1691 rc = ops->ndo_start_xmit(skb, dev);
1693 * TODO: if skb_orphan() was called by
1694 * dev->hard_start_xmit() (for example, the unmodified
1695 * igb driver does that; bnx2 doesn't), then
1696 * skb_tx_software_timestamp() will be unable to send
1697 * back the time stamp.
1699 * How can this be prevented? Always create another
1700 * reference to the socket before calling
1701 * dev->hard_start_xmit()? Prevent that skb_orphan()
1702 * does anything in dev->hard_start_xmit() by clearing
1703 * the skb destructor before the call and restoring it
1704 * afterwards, then doing the skb_orphan() ourselves?
1711 struct sk_buff *nskb = skb->next;
1713 skb->next = nskb->next;
1715 rc = ops->ndo_start_xmit(nskb, dev);
1717 nskb->next = skb->next;
1721 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1722 return NETDEV_TX_BUSY;
1723 } while (skb->next);
1725 skb->destructor = DEV_GSO_CB(skb)->destructor;
1732 static u32 skb_tx_hashrnd;
1734 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1738 if (skb_rx_queue_recorded(skb))
1739 return skb_get_rx_queue(skb) % dev->real_num_tx_queues;
1741 if (skb->sk && skb->sk->sk_hash)
1742 hash = skb->sk->sk_hash;
1744 hash = skb->protocol;
1746 hash = jhash_1word(hash, skb_tx_hashrnd);
1748 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1750 EXPORT_SYMBOL(skb_tx_hash);
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) {
2338 skb_shinfo(skb)->gso_size = 0;
2343 list_for_each_entry_rcu(ptype, head, list) {
2344 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2347 err = ptype->gro_complete(skb);
2353 WARN_ON(&ptype->list == head);
2355 return NET_RX_SUCCESS;
2359 return netif_receive_skb(skb);
2362 void napi_gro_flush(struct napi_struct *napi)
2364 struct sk_buff *skb, *next;
2366 for (skb = napi->gro_list; skb; skb = next) {
2369 napi_gro_complete(skb);
2372 napi->gro_count = 0;
2373 napi->gro_list = NULL;
2375 EXPORT_SYMBOL(napi_gro_flush);
2377 void *skb_gro_header(struct sk_buff *skb, unsigned int hlen)
2379 unsigned int offset = skb_gro_offset(skb);
2382 if (unlikely(skb_headlen(skb) ||
2383 skb_shinfo(skb)->frags[0].size < hlen ||
2384 PageHighMem(skb_shinfo(skb)->frags[0].page)))
2385 return pskb_may_pull(skb, hlen) ? skb->data + offset : NULL;
2387 return page_address(skb_shinfo(skb)->frags[0].page) +
2388 skb_shinfo(skb)->frags[0].page_offset + offset;
2390 EXPORT_SYMBOL(skb_gro_header);
2392 int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2394 struct sk_buff **pp = NULL;
2395 struct packet_type *ptype;
2396 __be16 type = skb->protocol;
2397 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2402 if (!(skb->dev->features & NETIF_F_GRO))
2405 if (skb_is_gso(skb) || skb_shinfo(skb)->frag_list)
2409 list_for_each_entry_rcu(ptype, head, list) {
2410 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2413 skb_set_network_header(skb, skb_gro_offset(skb));
2414 mac_len = skb->network_header - skb->mac_header;
2415 skb->mac_len = mac_len;
2416 NAPI_GRO_CB(skb)->same_flow = 0;
2417 NAPI_GRO_CB(skb)->flush = 0;
2418 NAPI_GRO_CB(skb)->free = 0;
2420 pp = ptype->gro_receive(&napi->gro_list, skb);
2425 if (&ptype->list == head)
2428 same_flow = NAPI_GRO_CB(skb)->same_flow;
2429 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2432 struct sk_buff *nskb = *pp;
2436 napi_gro_complete(nskb);
2443 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2447 NAPI_GRO_CB(skb)->count = 1;
2448 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2449 skb->next = napi->gro_list;
2450 napi->gro_list = skb;
2454 if (unlikely(!pskb_may_pull(skb, skb_gro_offset(skb)))) {
2455 if (napi->gro_list == skb)
2456 napi->gro_list = skb->next;
2467 EXPORT_SYMBOL(dev_gro_receive);
2469 static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2473 if (netpoll_rx_on(skb))
2476 for (p = napi->gro_list; p; p = p->next) {
2477 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2478 && !compare_ether_header(skb_mac_header(p),
2479 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_get_frags(struct napi_struct *napi)
2526 struct net_device *dev = napi->dev;
2527 struct sk_buff *skb = napi->skb;
2530 skb = netdev_alloc_skb(dev, GRO_MAX_HEAD + NET_IP_ALIGN);
2534 skb_reserve(skb, NET_IP_ALIGN);
2542 EXPORT_SYMBOL(napi_get_frags);
2544 int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret)
2546 int err = NET_RX_SUCCESS;
2551 skb->protocol = eth_type_trans(skb, napi->dev);
2553 if (ret == GRO_NORMAL)
2554 return netif_receive_skb(skb);
2556 skb_gro_pull(skb, -ETH_HLEN);
2563 case GRO_MERGED_FREE:
2564 napi_reuse_skb(napi, skb);
2570 EXPORT_SYMBOL(napi_frags_finish);
2572 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2574 struct sk_buff *skb = napi->skb;
2579 skb_reset_mac_header(skb);
2580 skb_gro_reset_offset(skb);
2582 eth = skb_gro_header(skb, sizeof(*eth));
2584 napi_reuse_skb(napi, skb);
2589 skb_gro_pull(skb, sizeof(*eth));
2592 * This works because the only protocols we care about don't require
2593 * special handling. We'll fix it up properly at the end.
2595 skb->protocol = eth->h_proto;
2600 EXPORT_SYMBOL(napi_frags_skb);
2602 int napi_gro_frags(struct napi_struct *napi)
2604 struct sk_buff *skb = napi_frags_skb(napi);
2609 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2611 EXPORT_SYMBOL(napi_gro_frags);
2613 static int process_backlog(struct napi_struct *napi, int quota)
2616 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2617 unsigned long start_time = jiffies;
2619 napi->weight = weight_p;
2621 struct sk_buff *skb;
2623 local_irq_disable();
2624 skb = __skb_dequeue(&queue->input_pkt_queue);
2626 __napi_complete(napi);
2632 netif_receive_skb(skb);
2633 } while (++work < quota && jiffies == start_time);
2639 * __napi_schedule - schedule for receive
2640 * @n: entry to schedule
2642 * The entry's receive function will be scheduled to run
2644 void __napi_schedule(struct napi_struct *n)
2646 unsigned long flags;
2648 local_irq_save(flags);
2649 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2650 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2651 local_irq_restore(flags);
2653 EXPORT_SYMBOL(__napi_schedule);
2655 void __napi_complete(struct napi_struct *n)
2657 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2658 BUG_ON(n->gro_list);
2660 list_del(&n->poll_list);
2661 smp_mb__before_clear_bit();
2662 clear_bit(NAPI_STATE_SCHED, &n->state);
2664 EXPORT_SYMBOL(__napi_complete);
2666 void napi_complete(struct napi_struct *n)
2668 unsigned long flags;
2671 * don't let napi dequeue from the cpu poll list
2672 * just in case its running on a different cpu
2674 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2678 local_irq_save(flags);
2680 local_irq_restore(flags);
2682 EXPORT_SYMBOL(napi_complete);
2684 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2685 int (*poll)(struct napi_struct *, int), int weight)
2687 INIT_LIST_HEAD(&napi->poll_list);
2688 napi->gro_count = 0;
2689 napi->gro_list = NULL;
2692 napi->weight = weight;
2693 list_add(&napi->dev_list, &dev->napi_list);
2695 #ifdef CONFIG_NETPOLL
2696 spin_lock_init(&napi->poll_lock);
2697 napi->poll_owner = -1;
2699 set_bit(NAPI_STATE_SCHED, &napi->state);
2701 EXPORT_SYMBOL(netif_napi_add);
2703 void netif_napi_del(struct napi_struct *napi)
2705 struct sk_buff *skb, *next;
2707 list_del_init(&napi->dev_list);
2708 napi_free_frags(napi);
2710 for (skb = napi->gro_list; skb; skb = next) {
2716 napi->gro_list = NULL;
2717 napi->gro_count = 0;
2719 EXPORT_SYMBOL(netif_napi_del);
2722 static void net_rx_action(struct softirq_action *h)
2724 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2725 unsigned long time_limit = jiffies + 2;
2726 int budget = netdev_budget;
2729 local_irq_disable();
2731 while (!list_empty(list)) {
2732 struct napi_struct *n;
2735 /* If softirq window is exhuasted then punt.
2736 * Allow this to run for 2 jiffies since which will allow
2737 * an average latency of 1.5/HZ.
2739 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2744 /* Even though interrupts have been re-enabled, this
2745 * access is safe because interrupts can only add new
2746 * entries to the tail of this list, and only ->poll()
2747 * calls can remove this head entry from the list.
2749 n = list_entry(list->next, struct napi_struct, poll_list);
2751 have = netpoll_poll_lock(n);
2755 /* This NAPI_STATE_SCHED test is for avoiding a race
2756 * with netpoll's poll_napi(). Only the entity which
2757 * obtains the lock and sees NAPI_STATE_SCHED set will
2758 * actually make the ->poll() call. Therefore we avoid
2759 * accidently calling ->poll() when NAPI is not scheduled.
2762 if (test_bit(NAPI_STATE_SCHED, &n->state))
2763 work = n->poll(n, weight);
2765 WARN_ON_ONCE(work > weight);
2769 local_irq_disable();
2771 /* Drivers must not modify the NAPI state if they
2772 * consume the entire weight. In such cases this code
2773 * still "owns" the NAPI instance and therefore can
2774 * move the instance around on the list at-will.
2776 if (unlikely(work == weight)) {
2777 if (unlikely(napi_disable_pending(n)))
2780 list_move_tail(&n->poll_list, list);
2783 netpoll_poll_unlock(have);
2788 #ifdef CONFIG_NET_DMA
2790 * There may not be any more sk_buffs coming right now, so push
2791 * any pending DMA copies to hardware
2793 dma_issue_pending_all();
2799 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2800 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2804 static gifconf_func_t * gifconf_list [NPROTO];
2807 * register_gifconf - register a SIOCGIF handler
2808 * @family: Address family
2809 * @gifconf: Function handler
2811 * Register protocol dependent address dumping routines. The handler
2812 * that is passed must not be freed or reused until it has been replaced
2813 * by another handler.
2815 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2817 if (family >= NPROTO)
2819 gifconf_list[family] = gifconf;
2825 * Map an interface index to its name (SIOCGIFNAME)
2829 * We need this ioctl for efficient implementation of the
2830 * if_indextoname() function required by the IPv6 API. Without
2831 * it, we would have to search all the interfaces to find a
2835 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2837 struct net_device *dev;
2841 * Fetch the caller's info block.
2844 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2847 read_lock(&dev_base_lock);
2848 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2850 read_unlock(&dev_base_lock);
2854 strcpy(ifr.ifr_name, dev->name);
2855 read_unlock(&dev_base_lock);
2857 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2863 * Perform a SIOCGIFCONF call. This structure will change
2864 * size eventually, and there is nothing I can do about it.
2865 * Thus we will need a 'compatibility mode'.
2868 static int dev_ifconf(struct net *net, char __user *arg)
2871 struct net_device *dev;
2878 * Fetch the caller's info block.
2881 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2888 * Loop over the interfaces, and write an info block for each.
2892 for_each_netdev(net, dev) {
2893 for (i = 0; i < NPROTO; i++) {
2894 if (gifconf_list[i]) {
2897 done = gifconf_list[i](dev, NULL, 0);
2899 done = gifconf_list[i](dev, pos + total,
2909 * All done. Write the updated control block back to the caller.
2911 ifc.ifc_len = total;
2914 * Both BSD and Solaris return 0 here, so we do too.
2916 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2919 #ifdef CONFIG_PROC_FS
2921 * This is invoked by the /proc filesystem handler to display a device
2924 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2925 __acquires(dev_base_lock)
2927 struct net *net = seq_file_net(seq);
2929 struct net_device *dev;
2931 read_lock(&dev_base_lock);
2933 return SEQ_START_TOKEN;
2936 for_each_netdev(net, dev)
2943 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2945 struct net *net = seq_file_net(seq);
2947 return v == SEQ_START_TOKEN ?
2948 first_net_device(net) : next_net_device((struct net_device *)v);
2951 void dev_seq_stop(struct seq_file *seq, void *v)
2952 __releases(dev_base_lock)
2954 read_unlock(&dev_base_lock);
2957 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2959 const struct net_device_stats *stats = dev_get_stats(dev);
2961 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2962 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2963 dev->name, stats->rx_bytes, stats->rx_packets,
2965 stats->rx_dropped + stats->rx_missed_errors,
2966 stats->rx_fifo_errors,
2967 stats->rx_length_errors + stats->rx_over_errors +
2968 stats->rx_crc_errors + stats->rx_frame_errors,
2969 stats->rx_compressed, stats->multicast,
2970 stats->tx_bytes, stats->tx_packets,
2971 stats->tx_errors, stats->tx_dropped,
2972 stats->tx_fifo_errors, stats->collisions,
2973 stats->tx_carrier_errors +
2974 stats->tx_aborted_errors +
2975 stats->tx_window_errors +
2976 stats->tx_heartbeat_errors,
2977 stats->tx_compressed);
2981 * Called from the PROCfs module. This now uses the new arbitrary sized
2982 * /proc/net interface to create /proc/net/dev
2984 static int dev_seq_show(struct seq_file *seq, void *v)
2986 if (v == SEQ_START_TOKEN)
2987 seq_puts(seq, "Inter-| Receive "
2989 " face |bytes packets errs drop fifo frame "
2990 "compressed multicast|bytes packets errs "
2991 "drop fifo colls carrier compressed\n");
2993 dev_seq_printf_stats(seq, v);
2997 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
2999 struct netif_rx_stats *rc = NULL;
3001 while (*pos < nr_cpu_ids)
3002 if (cpu_online(*pos)) {
3003 rc = &per_cpu(netdev_rx_stat, *pos);
3010 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3012 return softnet_get_online(pos);
3015 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3018 return softnet_get_online(pos);
3021 static void softnet_seq_stop(struct seq_file *seq, void *v)
3025 static int softnet_seq_show(struct seq_file *seq, void *v)
3027 struct netif_rx_stats *s = v;
3029 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3030 s->total, s->dropped, s->time_squeeze, 0,
3031 0, 0, 0, 0, /* was fastroute */
3036 static const struct seq_operations dev_seq_ops = {
3037 .start = dev_seq_start,
3038 .next = dev_seq_next,
3039 .stop = dev_seq_stop,
3040 .show = dev_seq_show,
3043 static int dev_seq_open(struct inode *inode, struct file *file)
3045 return seq_open_net(inode, file, &dev_seq_ops,
3046 sizeof(struct seq_net_private));
3049 static const struct file_operations dev_seq_fops = {
3050 .owner = THIS_MODULE,
3051 .open = dev_seq_open,
3053 .llseek = seq_lseek,
3054 .release = seq_release_net,
3057 static const struct seq_operations softnet_seq_ops = {
3058 .start = softnet_seq_start,
3059 .next = softnet_seq_next,
3060 .stop = softnet_seq_stop,
3061 .show = softnet_seq_show,
3064 static int softnet_seq_open(struct inode *inode, struct file *file)
3066 return seq_open(file, &softnet_seq_ops);
3069 static const struct file_operations softnet_seq_fops = {
3070 .owner = THIS_MODULE,
3071 .open = softnet_seq_open,
3073 .llseek = seq_lseek,
3074 .release = seq_release,
3077 static void *ptype_get_idx(loff_t pos)
3079 struct packet_type *pt = NULL;
3083 list_for_each_entry_rcu(pt, &ptype_all, list) {
3089 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3090 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3099 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3103 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3106 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3108 struct packet_type *pt;
3109 struct list_head *nxt;
3113 if (v == SEQ_START_TOKEN)
3114 return ptype_get_idx(0);
3117 nxt = pt->list.next;
3118 if (pt->type == htons(ETH_P_ALL)) {
3119 if (nxt != &ptype_all)
3122 nxt = ptype_base[0].next;
3124 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3126 while (nxt == &ptype_base[hash]) {
3127 if (++hash >= PTYPE_HASH_SIZE)
3129 nxt = ptype_base[hash].next;
3132 return list_entry(nxt, struct packet_type, list);
3135 static void ptype_seq_stop(struct seq_file *seq, void *v)
3141 static int ptype_seq_show(struct seq_file *seq, void *v)
3143 struct packet_type *pt = v;
3145 if (v == SEQ_START_TOKEN)
3146 seq_puts(seq, "Type Device Function\n");
3147 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3148 if (pt->type == htons(ETH_P_ALL))
3149 seq_puts(seq, "ALL ");
3151 seq_printf(seq, "%04x", ntohs(pt->type));
3153 seq_printf(seq, " %-8s %pF\n",
3154 pt->dev ? pt->dev->name : "", pt->func);
3160 static const struct seq_operations ptype_seq_ops = {
3161 .start = ptype_seq_start,
3162 .next = ptype_seq_next,
3163 .stop = ptype_seq_stop,
3164 .show = ptype_seq_show,
3167 static int ptype_seq_open(struct inode *inode, struct file *file)
3169 return seq_open_net(inode, file, &ptype_seq_ops,
3170 sizeof(struct seq_net_private));
3173 static const struct file_operations ptype_seq_fops = {
3174 .owner = THIS_MODULE,
3175 .open = ptype_seq_open,
3177 .llseek = seq_lseek,
3178 .release = seq_release_net,
3182 static int __net_init dev_proc_net_init(struct net *net)
3186 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3188 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3190 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3193 if (wext_proc_init(net))
3199 proc_net_remove(net, "ptype");
3201 proc_net_remove(net, "softnet_stat");
3203 proc_net_remove(net, "dev");
3207 static void __net_exit dev_proc_net_exit(struct net *net)
3209 wext_proc_exit(net);
3211 proc_net_remove(net, "ptype");
3212 proc_net_remove(net, "softnet_stat");
3213 proc_net_remove(net, "dev");
3216 static struct pernet_operations __net_initdata dev_proc_ops = {
3217 .init = dev_proc_net_init,
3218 .exit = dev_proc_net_exit,
3221 static int __init dev_proc_init(void)
3223 return register_pernet_subsys(&dev_proc_ops);
3226 #define dev_proc_init() 0
3227 #endif /* CONFIG_PROC_FS */
3231 * netdev_set_master - set up master/slave pair
3232 * @slave: slave device
3233 * @master: new master device
3235 * Changes the master device of the slave. Pass %NULL to break the
3236 * bonding. The caller must hold the RTNL semaphore. On a failure
3237 * a negative errno code is returned. On success the reference counts
3238 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3239 * function returns zero.
3241 int netdev_set_master(struct net_device *slave, struct net_device *master)
3243 struct net_device *old = slave->master;
3253 slave->master = master;
3261 slave->flags |= IFF_SLAVE;
3263 slave->flags &= ~IFF_SLAVE;
3265 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3269 static void dev_change_rx_flags(struct net_device *dev, int flags)
3271 const struct net_device_ops *ops = dev->netdev_ops;
3273 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3274 ops->ndo_change_rx_flags(dev, flags);
3277 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3279 unsigned short old_flags = dev->flags;
3285 dev->flags |= IFF_PROMISC;
3286 dev->promiscuity += inc;
3287 if (dev->promiscuity == 0) {
3290 * If inc causes overflow, untouch promisc and return error.
3293 dev->flags &= ~IFF_PROMISC;
3295 dev->promiscuity -= inc;
3296 printk(KERN_WARNING "%s: promiscuity touches roof, "
3297 "set promiscuity failed, promiscuity feature "
3298 "of device might be broken.\n", dev->name);
3302 if (dev->flags != old_flags) {
3303 printk(KERN_INFO "device %s %s promiscuous mode\n",
3304 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3306 if (audit_enabled) {
3307 current_uid_gid(&uid, &gid);
3308 audit_log(current->audit_context, GFP_ATOMIC,
3309 AUDIT_ANOM_PROMISCUOUS,
3310 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3311 dev->name, (dev->flags & IFF_PROMISC),
3312 (old_flags & IFF_PROMISC),
3313 audit_get_loginuid(current),
3315 audit_get_sessionid(current));
3318 dev_change_rx_flags(dev, IFF_PROMISC);
3324 * dev_set_promiscuity - update promiscuity count on a device
3328 * Add or remove promiscuity from a device. While the count in the device
3329 * remains above zero the interface remains promiscuous. Once it hits zero
3330 * the device reverts back to normal filtering operation. A negative inc
3331 * value is used to drop promiscuity on the device.
3332 * Return 0 if successful or a negative errno code on error.
3334 int dev_set_promiscuity(struct net_device *dev, int inc)
3336 unsigned short old_flags = dev->flags;
3339 err = __dev_set_promiscuity(dev, inc);
3342 if (dev->flags != old_flags)
3343 dev_set_rx_mode(dev);
3348 * dev_set_allmulti - update allmulti count on a device
3352 * Add or remove reception of all multicast frames to a device. While the
3353 * count in the device remains above zero the interface remains listening
3354 * to all interfaces. Once it hits zero the device reverts back to normal
3355 * filtering operation. A negative @inc value is used to drop the counter
3356 * when releasing a resource needing all multicasts.
3357 * Return 0 if successful or a negative errno code on error.
3360 int dev_set_allmulti(struct net_device *dev, int inc)
3362 unsigned short old_flags = dev->flags;
3366 dev->flags |= IFF_ALLMULTI;
3367 dev->allmulti += inc;
3368 if (dev->allmulti == 0) {
3371 * If inc causes overflow, untouch allmulti and return error.
3374 dev->flags &= ~IFF_ALLMULTI;
3376 dev->allmulti -= inc;
3377 printk(KERN_WARNING "%s: allmulti touches roof, "
3378 "set allmulti failed, allmulti feature of "
3379 "device might be broken.\n", dev->name);
3383 if (dev->flags ^ old_flags) {
3384 dev_change_rx_flags(dev, IFF_ALLMULTI);
3385 dev_set_rx_mode(dev);
3391 * Upload unicast and multicast address lists to device and
3392 * configure RX filtering. When the device doesn't support unicast
3393 * filtering it is put in promiscuous mode while unicast addresses
3396 void __dev_set_rx_mode(struct net_device *dev)
3398 const struct net_device_ops *ops = dev->netdev_ops;
3400 /* dev_open will call this function so the list will stay sane. */
3401 if (!(dev->flags&IFF_UP))
3404 if (!netif_device_present(dev))
3407 if (ops->ndo_set_rx_mode)
3408 ops->ndo_set_rx_mode(dev);
3410 /* Unicast addresses changes may only happen under the rtnl,
3411 * therefore calling __dev_set_promiscuity here is safe.
3413 if (dev->uc_count > 0 && !dev->uc_promisc) {
3414 __dev_set_promiscuity(dev, 1);
3415 dev->uc_promisc = 1;
3416 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3417 __dev_set_promiscuity(dev, -1);
3418 dev->uc_promisc = 0;
3421 if (ops->ndo_set_multicast_list)
3422 ops->ndo_set_multicast_list(dev);
3426 void dev_set_rx_mode(struct net_device *dev)
3428 netif_addr_lock_bh(dev);
3429 __dev_set_rx_mode(dev);
3430 netif_addr_unlock_bh(dev);
3433 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3434 void *addr, int alen, int glbl)
3436 struct dev_addr_list *da;
3438 for (; (da = *list) != NULL; list = &da->next) {
3439 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3440 alen == da->da_addrlen) {
3442 int old_glbl = da->da_gusers;
3459 int __dev_addr_add(struct dev_addr_list **list, int *count,
3460 void *addr, int alen, int glbl)
3462 struct dev_addr_list *da;
3464 for (da = *list; da != NULL; da = da->next) {
3465 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3466 da->da_addrlen == alen) {
3468 int old_glbl = da->da_gusers;
3478 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3481 memcpy(da->da_addr, addr, alen);
3482 da->da_addrlen = alen;
3484 da->da_gusers = glbl ? 1 : 0;
3492 * dev_unicast_delete - Release secondary unicast address.
3494 * @addr: address to delete
3495 * @alen: length of @addr
3497 * Release reference to a secondary unicast address and remove it
3498 * from the device if the reference count drops to zero.
3500 * The caller must hold the rtnl_mutex.
3502 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3508 netif_addr_lock_bh(dev);
3509 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3511 __dev_set_rx_mode(dev);
3512 netif_addr_unlock_bh(dev);
3515 EXPORT_SYMBOL(dev_unicast_delete);
3518 * dev_unicast_add - add a secondary unicast address
3520 * @addr: address to add
3521 * @alen: length of @addr
3523 * Add a secondary unicast address to the device or increase
3524 * the reference count if it already exists.
3526 * The caller must hold the rtnl_mutex.
3528 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3534 netif_addr_lock_bh(dev);
3535 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3537 __dev_set_rx_mode(dev);
3538 netif_addr_unlock_bh(dev);
3541 EXPORT_SYMBOL(dev_unicast_add);
3543 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3544 struct dev_addr_list **from, int *from_count)
3546 struct dev_addr_list *da, *next;
3550 while (da != NULL) {
3552 if (!da->da_synced) {
3553 err = __dev_addr_add(to, to_count,
3554 da->da_addr, da->da_addrlen, 0);
3559 } else if (da->da_users == 1) {
3560 __dev_addr_delete(to, to_count,
3561 da->da_addr, da->da_addrlen, 0);
3562 __dev_addr_delete(from, from_count,
3563 da->da_addr, da->da_addrlen, 0);
3570 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3571 struct dev_addr_list **from, int *from_count)
3573 struct dev_addr_list *da, *next;
3576 while (da != NULL) {
3578 if (da->da_synced) {
3579 __dev_addr_delete(to, to_count,
3580 da->da_addr, da->da_addrlen, 0);
3582 __dev_addr_delete(from, from_count,
3583 da->da_addr, da->da_addrlen, 0);
3590 * dev_unicast_sync - Synchronize device's unicast list to another device
3591 * @to: destination device
3592 * @from: source device
3594 * Add newly added addresses to the destination device and release
3595 * addresses that have no users left. The source device must be
3596 * locked by netif_tx_lock_bh.
3598 * This function is intended to be called from the dev->set_rx_mode
3599 * function of layered software devices.
3601 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3605 netif_addr_lock_bh(to);
3606 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3607 &from->uc_list, &from->uc_count);
3609 __dev_set_rx_mode(to);
3610 netif_addr_unlock_bh(to);
3613 EXPORT_SYMBOL(dev_unicast_sync);
3616 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3617 * @to: destination device
3618 * @from: source device
3620 * Remove all addresses that were added to the destination device by
3621 * dev_unicast_sync(). This function is intended to be called from the
3622 * dev->stop function of layered software devices.
3624 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3626 netif_addr_lock_bh(from);
3627 netif_addr_lock(to);
3629 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3630 &from->uc_list, &from->uc_count);
3631 __dev_set_rx_mode(to);
3633 netif_addr_unlock(to);
3634 netif_addr_unlock_bh(from);
3636 EXPORT_SYMBOL(dev_unicast_unsync);
3638 static void __dev_addr_discard(struct dev_addr_list **list)
3640 struct dev_addr_list *tmp;
3642 while (*list != NULL) {
3645 if (tmp->da_users > tmp->da_gusers)
3646 printk("__dev_addr_discard: address leakage! "
3647 "da_users=%d\n", tmp->da_users);
3652 static void dev_addr_discard(struct net_device *dev)
3654 netif_addr_lock_bh(dev);
3656 __dev_addr_discard(&dev->uc_list);
3659 __dev_addr_discard(&dev->mc_list);
3662 netif_addr_unlock_bh(dev);
3666 * dev_get_flags - get flags reported to userspace
3669 * Get the combination of flag bits exported through APIs to userspace.
3671 unsigned dev_get_flags(const struct net_device *dev)
3675 flags = (dev->flags & ~(IFF_PROMISC |
3680 (dev->gflags & (IFF_PROMISC |
3683 if (netif_running(dev)) {
3684 if (netif_oper_up(dev))
3685 flags |= IFF_RUNNING;
3686 if (netif_carrier_ok(dev))
3687 flags |= IFF_LOWER_UP;
3688 if (netif_dormant(dev))
3689 flags |= IFF_DORMANT;
3696 * dev_change_flags - change device settings
3698 * @flags: device state flags
3700 * Change settings on device based state flags. The flags are
3701 * in the userspace exported format.
3703 int dev_change_flags(struct net_device *dev, unsigned flags)
3706 int old_flags = dev->flags;
3711 * Set the flags on our device.
3714 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3715 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3717 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3721 * Load in the correct multicast list now the flags have changed.
3724 if ((old_flags ^ flags) & IFF_MULTICAST)
3725 dev_change_rx_flags(dev, IFF_MULTICAST);
3727 dev_set_rx_mode(dev);
3730 * Have we downed the interface. We handle IFF_UP ourselves
3731 * according to user attempts to set it, rather than blindly
3736 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3737 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3740 dev_set_rx_mode(dev);
3743 if (dev->flags & IFF_UP &&
3744 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3746 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3748 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3749 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3750 dev->gflags ^= IFF_PROMISC;
3751 dev_set_promiscuity(dev, inc);
3754 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3755 is important. Some (broken) drivers set IFF_PROMISC, when
3756 IFF_ALLMULTI is requested not asking us and not reporting.
3758 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3759 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3760 dev->gflags ^= IFF_ALLMULTI;
3761 dev_set_allmulti(dev, inc);
3764 /* Exclude state transition flags, already notified */
3765 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3767 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3773 * dev_set_mtu - Change maximum transfer unit
3775 * @new_mtu: new transfer unit
3777 * Change the maximum transfer size of the network device.
3779 int dev_set_mtu(struct net_device *dev, int new_mtu)
3781 const struct net_device_ops *ops = dev->netdev_ops;
3784 if (new_mtu == dev->mtu)
3787 /* MTU must be positive. */
3791 if (!netif_device_present(dev))
3795 if (ops->ndo_change_mtu)
3796 err = ops->ndo_change_mtu(dev, new_mtu);
3800 if (!err && dev->flags & IFF_UP)
3801 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3806 * dev_set_mac_address - Change Media Access Control Address
3810 * Change the hardware (MAC) address of the device
3812 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3814 const struct net_device_ops *ops = dev->netdev_ops;
3817 if (!ops->ndo_set_mac_address)
3819 if (sa->sa_family != dev->type)
3821 if (!netif_device_present(dev))
3823 err = ops->ndo_set_mac_address(dev, sa);
3825 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3830 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3832 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3835 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3841 case SIOCGIFFLAGS: /* Get interface flags */
3842 ifr->ifr_flags = dev_get_flags(dev);
3845 case SIOCGIFMETRIC: /* Get the metric on the interface
3846 (currently unused) */
3847 ifr->ifr_metric = 0;
3850 case SIOCGIFMTU: /* Get the MTU of a device */
3851 ifr->ifr_mtu = dev->mtu;
3856 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3858 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3859 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3860 ifr->ifr_hwaddr.sa_family = dev->type;
3868 ifr->ifr_map.mem_start = dev->mem_start;
3869 ifr->ifr_map.mem_end = dev->mem_end;
3870 ifr->ifr_map.base_addr = dev->base_addr;
3871 ifr->ifr_map.irq = dev->irq;
3872 ifr->ifr_map.dma = dev->dma;
3873 ifr->ifr_map.port = dev->if_port;
3877 ifr->ifr_ifindex = dev->ifindex;
3881 ifr->ifr_qlen = dev->tx_queue_len;
3885 /* dev_ioctl() should ensure this case
3897 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3899 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3902 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3903 const struct net_device_ops *ops;
3908 ops = dev->netdev_ops;
3911 case SIOCSIFFLAGS: /* Set interface flags */
3912 return dev_change_flags(dev, ifr->ifr_flags);
3914 case SIOCSIFMETRIC: /* Set the metric on the interface
3915 (currently unused) */
3918 case SIOCSIFMTU: /* Set the MTU of a device */
3919 return dev_set_mtu(dev, ifr->ifr_mtu);
3922 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3924 case SIOCSIFHWBROADCAST:
3925 if (ifr->ifr_hwaddr.sa_family != dev->type)
3927 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3928 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3929 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3933 if (ops->ndo_set_config) {
3934 if (!netif_device_present(dev))
3936 return ops->ndo_set_config(dev, &ifr->ifr_map);
3941 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
3942 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3944 if (!netif_device_present(dev))
3946 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3950 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
3951 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3953 if (!netif_device_present(dev))
3955 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3959 if (ifr->ifr_qlen < 0)
3961 dev->tx_queue_len = ifr->ifr_qlen;
3965 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3966 return dev_change_name(dev, ifr->ifr_newname);
3969 * Unknown or private ioctl
3973 if ((cmd >= SIOCDEVPRIVATE &&
3974 cmd <= SIOCDEVPRIVATE + 15) ||
3975 cmd == SIOCBONDENSLAVE ||
3976 cmd == SIOCBONDRELEASE ||
3977 cmd == SIOCBONDSETHWADDR ||
3978 cmd == SIOCBONDSLAVEINFOQUERY ||
3979 cmd == SIOCBONDINFOQUERY ||
3980 cmd == SIOCBONDCHANGEACTIVE ||
3981 cmd == SIOCGMIIPHY ||
3982 cmd == SIOCGMIIREG ||
3983 cmd == SIOCSMIIREG ||
3984 cmd == SIOCBRADDIF ||
3985 cmd == SIOCBRDELIF ||
3986 cmd == SIOCSHWTSTAMP ||
3987 cmd == SIOCWANDEV) {
3989 if (ops->ndo_do_ioctl) {
3990 if (netif_device_present(dev))
3991 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4003 * This function handles all "interface"-type I/O control requests. The actual
4004 * 'doing' part of this is dev_ifsioc above.
4008 * dev_ioctl - network device ioctl
4009 * @net: the applicable net namespace
4010 * @cmd: command to issue
4011 * @arg: pointer to a struct ifreq in user space
4013 * Issue ioctl functions to devices. This is normally called by the
4014 * user space syscall interfaces but can sometimes be useful for
4015 * other purposes. The return value is the return from the syscall if
4016 * positive or a negative errno code on error.
4019 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4025 /* One special case: SIOCGIFCONF takes ifconf argument
4026 and requires shared lock, because it sleeps writing
4030 if (cmd == SIOCGIFCONF) {
4032 ret = dev_ifconf(net, (char __user *) arg);
4036 if (cmd == SIOCGIFNAME)
4037 return dev_ifname(net, (struct ifreq __user *)arg);
4039 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4042 ifr.ifr_name[IFNAMSIZ-1] = 0;
4044 colon = strchr(ifr.ifr_name, ':');
4049 * See which interface the caller is talking about.
4054 * These ioctl calls:
4055 * - can be done by all.
4056 * - atomic and do not require locking.
4067 dev_load(net, ifr.ifr_name);
4068 read_lock(&dev_base_lock);
4069 ret = dev_ifsioc_locked(net, &ifr, cmd);
4070 read_unlock(&dev_base_lock);
4074 if (copy_to_user(arg, &ifr,
4075 sizeof(struct ifreq)))
4081 dev_load(net, ifr.ifr_name);
4083 ret = dev_ethtool(net, &ifr);
4088 if (copy_to_user(arg, &ifr,
4089 sizeof(struct ifreq)))
4095 * These ioctl calls:
4096 * - require superuser power.
4097 * - require strict serialization.
4103 if (!capable(CAP_NET_ADMIN))
4105 dev_load(net, ifr.ifr_name);
4107 ret = dev_ifsioc(net, &ifr, cmd);
4112 if (copy_to_user(arg, &ifr,
4113 sizeof(struct ifreq)))
4119 * These ioctl calls:
4120 * - require superuser power.
4121 * - require strict serialization.
4122 * - do not return a value
4132 case SIOCSIFHWBROADCAST:
4135 case SIOCBONDENSLAVE:
4136 case SIOCBONDRELEASE:
4137 case SIOCBONDSETHWADDR:
4138 case SIOCBONDCHANGEACTIVE:
4142 if (!capable(CAP_NET_ADMIN))
4145 case SIOCBONDSLAVEINFOQUERY:
4146 case SIOCBONDINFOQUERY:
4147 dev_load(net, ifr.ifr_name);
4149 ret = dev_ifsioc(net, &ifr, cmd);
4154 /* Get the per device memory space. We can add this but
4155 * currently do not support it */
4157 /* Set the per device memory buffer space.
4158 * Not applicable in our case */
4163 * Unknown or private ioctl.
4166 if (cmd == SIOCWANDEV ||
4167 (cmd >= SIOCDEVPRIVATE &&
4168 cmd <= SIOCDEVPRIVATE + 15)) {
4169 dev_load(net, ifr.ifr_name);
4171 ret = dev_ifsioc(net, &ifr, cmd);
4173 if (!ret && copy_to_user(arg, &ifr,
4174 sizeof(struct ifreq)))
4178 /* Take care of Wireless Extensions */
4179 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4180 return wext_handle_ioctl(net, &ifr, cmd, arg);
4187 * dev_new_index - allocate an ifindex
4188 * @net: the applicable net namespace
4190 * Returns a suitable unique value for a new device interface
4191 * number. The caller must hold the rtnl semaphore or the
4192 * dev_base_lock to be sure it remains unique.
4194 static int dev_new_index(struct net *net)
4200 if (!__dev_get_by_index(net, ifindex))
4205 /* Delayed registration/unregisteration */
4206 static LIST_HEAD(net_todo_list);
4208 static void net_set_todo(struct net_device *dev)
4210 list_add_tail(&dev->todo_list, &net_todo_list);
4213 static void rollback_registered(struct net_device *dev)
4215 BUG_ON(dev_boot_phase);
4218 /* Some devices call without registering for initialization unwind. */
4219 if (dev->reg_state == NETREG_UNINITIALIZED) {
4220 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
4221 "was registered\n", dev->name, dev);
4227 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4229 /* If device is running, close it first. */
4232 /* And unlink it from device chain. */
4233 unlist_netdevice(dev);
4235 dev->reg_state = NETREG_UNREGISTERING;
4239 /* Shutdown queueing discipline. */
4243 /* Notify protocols, that we are about to destroy
4244 this device. They should clean all the things.
4246 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4249 * Flush the unicast and multicast chains
4251 dev_addr_discard(dev);
4253 if (dev->netdev_ops->ndo_uninit)
4254 dev->netdev_ops->ndo_uninit(dev);
4256 /* Notifier chain MUST detach us from master device. */
4257 WARN_ON(dev->master);
4259 /* Remove entries from kobject tree */
4260 netdev_unregister_kobject(dev);
4267 static void __netdev_init_queue_locks_one(struct net_device *dev,
4268 struct netdev_queue *dev_queue,
4271 spin_lock_init(&dev_queue->_xmit_lock);
4272 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4273 dev_queue->xmit_lock_owner = -1;
4276 static void netdev_init_queue_locks(struct net_device *dev)
4278 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4279 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4282 unsigned long netdev_fix_features(unsigned long features, const char *name)
4284 /* Fix illegal SG+CSUM combinations. */
4285 if ((features & NETIF_F_SG) &&
4286 !(features & NETIF_F_ALL_CSUM)) {
4288 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4289 "checksum feature.\n", name);
4290 features &= ~NETIF_F_SG;
4293 /* TSO requires that SG is present as well. */
4294 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4296 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4297 "SG feature.\n", name);
4298 features &= ~NETIF_F_TSO;
4301 if (features & NETIF_F_UFO) {
4302 if (!(features & NETIF_F_GEN_CSUM)) {
4304 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4305 "since no NETIF_F_HW_CSUM feature.\n",
4307 features &= ~NETIF_F_UFO;
4310 if (!(features & NETIF_F_SG)) {
4312 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4313 "since no NETIF_F_SG feature.\n", name);
4314 features &= ~NETIF_F_UFO;
4320 EXPORT_SYMBOL(netdev_fix_features);
4322 /* Some devices need to (re-)set their netdev_ops inside
4323 * ->init() or similar. If that happens, we have to setup
4324 * the compat pointers again.
4326 void netdev_resync_ops(struct net_device *dev)
4328 #ifdef CONFIG_COMPAT_NET_DEV_OPS
4329 const struct net_device_ops *ops = dev->netdev_ops;
4331 dev->init = ops->ndo_init;
4332 dev->uninit = ops->ndo_uninit;
4333 dev->open = ops->ndo_open;
4334 dev->change_rx_flags = ops->ndo_change_rx_flags;
4335 dev->set_rx_mode = ops->ndo_set_rx_mode;
4336 dev->set_multicast_list = ops->ndo_set_multicast_list;
4337 dev->set_mac_address = ops->ndo_set_mac_address;
4338 dev->validate_addr = ops->ndo_validate_addr;
4339 dev->do_ioctl = ops->ndo_do_ioctl;
4340 dev->set_config = ops->ndo_set_config;
4341 dev->change_mtu = ops->ndo_change_mtu;
4342 dev->neigh_setup = ops->ndo_neigh_setup;
4343 dev->tx_timeout = ops->ndo_tx_timeout;
4344 dev->get_stats = ops->ndo_get_stats;
4345 dev->vlan_rx_register = ops->ndo_vlan_rx_register;
4346 dev->vlan_rx_add_vid = ops->ndo_vlan_rx_add_vid;
4347 dev->vlan_rx_kill_vid = ops->ndo_vlan_rx_kill_vid;
4348 #ifdef CONFIG_NET_POLL_CONTROLLER
4349 dev->poll_controller = ops->ndo_poll_controller;
4353 EXPORT_SYMBOL(netdev_resync_ops);
4356 * register_netdevice - register a network device
4357 * @dev: device to register
4359 * Take a completed network device structure and add it to the kernel
4360 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4361 * chain. 0 is returned on success. A negative errno code is returned
4362 * on a failure to set up the device, or if the name is a duplicate.
4364 * Callers must hold the rtnl semaphore. You may want
4365 * register_netdev() instead of this.
4368 * The locking appears insufficient to guarantee two parallel registers
4369 * will not get the same name.
4372 int register_netdevice(struct net_device *dev)
4374 struct hlist_head *head;
4375 struct hlist_node *p;
4377 struct net *net = dev_net(dev);
4379 BUG_ON(dev_boot_phase);
4384 /* When net_device's are persistent, this will be fatal. */
4385 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4388 spin_lock_init(&dev->addr_list_lock);
4389 netdev_set_addr_lockdep_class(dev);
4390 netdev_init_queue_locks(dev);
4394 #ifdef CONFIG_COMPAT_NET_DEV_OPS
4395 /* Netdevice_ops API compatibility support.
4396 * This is temporary until all network devices are converted.
4398 if (dev->netdev_ops) {
4399 netdev_resync_ops(dev);
4401 char drivername[64];
4402 pr_info("%s (%s): not using net_device_ops yet\n",
4403 dev->name, netdev_drivername(dev, drivername, 64));
4405 /* This works only because net_device_ops and the
4406 compatibility structure are the same. */
4407 dev->netdev_ops = (void *) &(dev->init);
4411 /* Init, if this function is available */
4412 if (dev->netdev_ops->ndo_init) {
4413 ret = dev->netdev_ops->ndo_init(dev);
4421 if (!dev_valid_name(dev->name)) {
4426 dev->ifindex = dev_new_index(net);
4427 if (dev->iflink == -1)
4428 dev->iflink = dev->ifindex;
4430 /* Check for existence of name */
4431 head = dev_name_hash(net, dev->name);
4432 hlist_for_each(p, head) {
4433 struct net_device *d
4434 = hlist_entry(p, struct net_device, name_hlist);
4435 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4441 /* Fix illegal checksum combinations */
4442 if ((dev->features & NETIF_F_HW_CSUM) &&
4443 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4444 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4446 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4449 if ((dev->features & NETIF_F_NO_CSUM) &&
4450 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4451 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4453 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4456 dev->features = netdev_fix_features(dev->features, dev->name);
4458 /* Enable software GSO if SG is supported. */
4459 if (dev->features & NETIF_F_SG)
4460 dev->features |= NETIF_F_GSO;
4462 netdev_initialize_kobject(dev);
4463 ret = netdev_register_kobject(dev);
4466 dev->reg_state = NETREG_REGISTERED;
4469 * Default initial state at registry is that the
4470 * device is present.
4473 set_bit(__LINK_STATE_PRESENT, &dev->state);
4475 dev_init_scheduler(dev);
4477 list_netdevice(dev);
4479 /* Notify protocols, that a new device appeared. */
4480 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4481 ret = notifier_to_errno(ret);
4483 rollback_registered(dev);
4484 dev->reg_state = NETREG_UNREGISTERED;
4491 if (dev->netdev_ops->ndo_uninit)
4492 dev->netdev_ops->ndo_uninit(dev);
4497 * init_dummy_netdev - init a dummy network device for NAPI
4498 * @dev: device to init
4500 * This takes a network device structure and initialize the minimum
4501 * amount of fields so it can be used to schedule NAPI polls without
4502 * registering a full blown interface. This is to be used by drivers
4503 * that need to tie several hardware interfaces to a single NAPI
4504 * poll scheduler due to HW limitations.
4506 int init_dummy_netdev(struct net_device *dev)
4508 /* Clear everything. Note we don't initialize spinlocks
4509 * are they aren't supposed to be taken by any of the
4510 * NAPI code and this dummy netdev is supposed to be
4511 * only ever used for NAPI polls
4513 memset(dev, 0, sizeof(struct net_device));
4515 /* make sure we BUG if trying to hit standard
4516 * register/unregister code path
4518 dev->reg_state = NETREG_DUMMY;
4520 /* initialize the ref count */
4521 atomic_set(&dev->refcnt, 1);
4523 /* NAPI wants this */
4524 INIT_LIST_HEAD(&dev->napi_list);
4526 /* a dummy interface is started by default */
4527 set_bit(__LINK_STATE_PRESENT, &dev->state);
4528 set_bit(__LINK_STATE_START, &dev->state);
4532 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4536 * register_netdev - register a network device
4537 * @dev: device to register
4539 * Take a completed network device structure and add it to the kernel
4540 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4541 * chain. 0 is returned on success. A negative errno code is returned
4542 * on a failure to set up the device, or if the name is a duplicate.
4544 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4545 * and expands the device name if you passed a format string to
4548 int register_netdev(struct net_device *dev)
4555 * If the name is a format string the caller wants us to do a
4558 if (strchr(dev->name, '%')) {
4559 err = dev_alloc_name(dev, dev->name);
4564 err = register_netdevice(dev);
4569 EXPORT_SYMBOL(register_netdev);
4572 * netdev_wait_allrefs - wait until all references are gone.
4574 * This is called when unregistering network devices.
4576 * Any protocol or device that holds a reference should register
4577 * for netdevice notification, and cleanup and put back the
4578 * reference if they receive an UNREGISTER event.
4579 * We can get stuck here if buggy protocols don't correctly
4582 static void netdev_wait_allrefs(struct net_device *dev)
4584 unsigned long rebroadcast_time, warning_time;
4586 rebroadcast_time = warning_time = jiffies;
4587 while (atomic_read(&dev->refcnt) != 0) {
4588 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4591 /* Rebroadcast unregister notification */
4592 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4594 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4596 /* We must not have linkwatch events
4597 * pending on unregister. If this
4598 * happens, we simply run the queue
4599 * unscheduled, resulting in a noop
4602 linkwatch_run_queue();
4607 rebroadcast_time = jiffies;
4612 if (time_after(jiffies, warning_time + 10 * HZ)) {
4613 printk(KERN_EMERG "unregister_netdevice: "
4614 "waiting for %s to become free. Usage "
4616 dev->name, atomic_read(&dev->refcnt));
4617 warning_time = jiffies;
4626 * register_netdevice(x1);
4627 * register_netdevice(x2);
4629 * unregister_netdevice(y1);
4630 * unregister_netdevice(y2);
4636 * We are invoked by rtnl_unlock().
4637 * This allows us to deal with problems:
4638 * 1) We can delete sysfs objects which invoke hotplug
4639 * without deadlocking with linkwatch via keventd.
4640 * 2) Since we run with the RTNL semaphore not held, we can sleep
4641 * safely in order to wait for the netdev refcnt to drop to zero.
4643 * We must not return until all unregister events added during
4644 * the interval the lock was held have been completed.
4646 void netdev_run_todo(void)
4648 struct list_head list;
4650 /* Snapshot list, allow later requests */
4651 list_replace_init(&net_todo_list, &list);
4655 while (!list_empty(&list)) {
4656 struct net_device *dev
4657 = list_entry(list.next, struct net_device, todo_list);
4658 list_del(&dev->todo_list);
4660 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4661 printk(KERN_ERR "network todo '%s' but state %d\n",
4662 dev->name, dev->reg_state);
4667 dev->reg_state = NETREG_UNREGISTERED;
4669 on_each_cpu(flush_backlog, dev, 1);
4671 netdev_wait_allrefs(dev);
4674 BUG_ON(atomic_read(&dev->refcnt));
4675 WARN_ON(dev->ip_ptr);
4676 WARN_ON(dev->ip6_ptr);
4677 WARN_ON(dev->dn_ptr);
4679 if (dev->destructor)
4680 dev->destructor(dev);
4682 /* Free network device */
4683 kobject_put(&dev->dev.kobj);
4688 * dev_get_stats - get network device statistics
4689 * @dev: device to get statistics from
4691 * Get network statistics from device. The device driver may provide
4692 * its own method by setting dev->netdev_ops->get_stats; otherwise
4693 * the internal statistics structure is used.
4695 const struct net_device_stats *dev_get_stats(struct net_device *dev)
4697 const struct net_device_ops *ops = dev->netdev_ops;
4699 if (ops->ndo_get_stats)
4700 return ops->ndo_get_stats(dev);
4704 EXPORT_SYMBOL(dev_get_stats);
4706 static void netdev_init_one_queue(struct net_device *dev,
4707 struct netdev_queue *queue,
4713 static void netdev_init_queues(struct net_device *dev)
4715 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4716 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4717 spin_lock_init(&dev->tx_global_lock);
4721 * alloc_netdev_mq - allocate network device
4722 * @sizeof_priv: size of private data to allocate space for
4723 * @name: device name format string
4724 * @setup: callback to initialize device
4725 * @queue_count: the number of subqueues to allocate
4727 * Allocates a struct net_device with private data area for driver use
4728 * and performs basic initialization. Also allocates subquue structs
4729 * for each queue on the device at the end of the netdevice.
4731 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4732 void (*setup)(struct net_device *), unsigned int queue_count)
4734 struct netdev_queue *tx;
4735 struct net_device *dev;
4739 BUG_ON(strlen(name) >= sizeof(dev->name));
4741 alloc_size = sizeof(struct net_device);
4743 /* ensure 32-byte alignment of private area */
4744 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4745 alloc_size += sizeof_priv;
4747 /* ensure 32-byte alignment of whole construct */
4748 alloc_size += NETDEV_ALIGN_CONST;
4750 p = kzalloc(alloc_size, GFP_KERNEL);
4752 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4756 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4758 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4764 dev = (struct net_device *)
4765 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4766 dev->padded = (char *)dev - (char *)p;
4767 dev_net_set(dev, &init_net);
4770 dev->num_tx_queues = queue_count;
4771 dev->real_num_tx_queues = queue_count;
4773 dev->gso_max_size = GSO_MAX_SIZE;
4775 netdev_init_queues(dev);
4777 INIT_LIST_HEAD(&dev->napi_list);
4779 strcpy(dev->name, name);
4782 EXPORT_SYMBOL(alloc_netdev_mq);
4785 * free_netdev - free network device
4788 * This function does the last stage of destroying an allocated device
4789 * interface. The reference to the device object is released.
4790 * If this is the last reference then it will be freed.
4792 void free_netdev(struct net_device *dev)
4794 struct napi_struct *p, *n;
4796 release_net(dev_net(dev));
4800 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
4803 /* Compatibility with error handling in drivers */
4804 if (dev->reg_state == NETREG_UNINITIALIZED) {
4805 kfree((char *)dev - dev->padded);
4809 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4810 dev->reg_state = NETREG_RELEASED;
4812 /* will free via device release */
4813 put_device(&dev->dev);
4817 * synchronize_net - Synchronize with packet receive processing
4819 * Wait for packets currently being received to be done.
4820 * Does not block later packets from starting.
4822 void synchronize_net(void)
4829 * unregister_netdevice - remove device from the kernel
4832 * This function shuts down a device interface and removes it
4833 * from the kernel tables.
4835 * Callers must hold the rtnl semaphore. You may want
4836 * unregister_netdev() instead of this.
4839 void unregister_netdevice(struct net_device *dev)
4843 rollback_registered(dev);
4844 /* Finish processing unregister after unlock */
4849 * unregister_netdev - remove device from the kernel
4852 * This function shuts down a device interface and removes it
4853 * from the kernel tables.
4855 * This is just a wrapper for unregister_netdevice that takes
4856 * the rtnl semaphore. In general you want to use this and not
4857 * unregister_netdevice.
4859 void unregister_netdev(struct net_device *dev)
4862 unregister_netdevice(dev);
4866 EXPORT_SYMBOL(unregister_netdev);
4869 * dev_change_net_namespace - move device to different nethost namespace
4871 * @net: network namespace
4872 * @pat: If not NULL name pattern to try if the current device name
4873 * is already taken in the destination network namespace.
4875 * This function shuts down a device interface and moves it
4876 * to a new network namespace. On success 0 is returned, on
4877 * a failure a netagive errno code is returned.
4879 * Callers must hold the rtnl semaphore.
4882 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4885 const char *destname;
4890 /* Don't allow namespace local devices to be moved. */
4892 if (dev->features & NETIF_F_NETNS_LOCAL)
4896 /* Don't allow real devices to be moved when sysfs
4900 if (dev->dev.parent)
4904 /* Ensure the device has been registrered */
4906 if (dev->reg_state != NETREG_REGISTERED)
4909 /* Get out if there is nothing todo */
4911 if (net_eq(dev_net(dev), net))
4914 /* Pick the destination device name, and ensure
4915 * we can use it in the destination network namespace.
4918 destname = dev->name;
4919 if (__dev_get_by_name(net, destname)) {
4920 /* We get here if we can't use the current device name */
4923 if (!dev_valid_name(pat))
4925 if (strchr(pat, '%')) {
4926 if (__dev_alloc_name(net, pat, buf) < 0)
4931 if (__dev_get_by_name(net, destname))
4936 * And now a mini version of register_netdevice unregister_netdevice.
4939 /* If device is running close it first. */
4942 /* And unlink it from device chain */
4944 unlist_netdevice(dev);
4948 /* Shutdown queueing discipline. */
4951 /* Notify protocols, that we are about to destroy
4952 this device. They should clean all the things.
4954 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4957 * Flush the unicast and multicast chains
4959 dev_addr_discard(dev);
4961 netdev_unregister_kobject(dev);
4963 /* Actually switch the network namespace */
4964 dev_net_set(dev, net);
4966 /* Assign the new device name */
4967 if (destname != dev->name)
4968 strcpy(dev->name, destname);
4970 /* If there is an ifindex conflict assign a new one */
4971 if (__dev_get_by_index(net, dev->ifindex)) {
4972 int iflink = (dev->iflink == dev->ifindex);
4973 dev->ifindex = dev_new_index(net);
4975 dev->iflink = dev->ifindex;
4978 /* Fixup kobjects */
4979 err = netdev_register_kobject(dev);
4982 /* Add the device back in the hashes */
4983 list_netdevice(dev);
4985 /* Notify protocols, that a new device appeared. */
4986 call_netdevice_notifiers(NETDEV_REGISTER, dev);
4994 static int dev_cpu_callback(struct notifier_block *nfb,
4995 unsigned long action,
4998 struct sk_buff **list_skb;
4999 struct Qdisc **list_net;
5000 struct sk_buff *skb;
5001 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5002 struct softnet_data *sd, *oldsd;
5004 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5007 local_irq_disable();
5008 cpu = smp_processor_id();
5009 sd = &per_cpu(softnet_data, cpu);
5010 oldsd = &per_cpu(softnet_data, oldcpu);
5012 /* Find end of our completion_queue. */
5013 list_skb = &sd->completion_queue;
5015 list_skb = &(*list_skb)->next;
5016 /* Append completion queue from offline CPU. */
5017 *list_skb = oldsd->completion_queue;
5018 oldsd->completion_queue = NULL;
5020 /* Find end of our output_queue. */
5021 list_net = &sd->output_queue;
5023 list_net = &(*list_net)->next_sched;
5024 /* Append output queue from offline CPU. */
5025 *list_net = oldsd->output_queue;
5026 oldsd->output_queue = NULL;
5028 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5031 /* Process offline CPU's input_pkt_queue */
5032 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5040 * netdev_increment_features - increment feature set by one
5041 * @all: current feature set
5042 * @one: new feature set
5043 * @mask: mask feature set
5045 * Computes a new feature set after adding a device with feature set
5046 * @one to the master device with current feature set @all. Will not
5047 * enable anything that is off in @mask. Returns the new feature set.
5049 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5052 /* If device needs checksumming, downgrade to it. */
5053 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5054 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5055 else if (mask & NETIF_F_ALL_CSUM) {
5056 /* If one device supports v4/v6 checksumming, set for all. */
5057 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5058 !(all & NETIF_F_GEN_CSUM)) {
5059 all &= ~NETIF_F_ALL_CSUM;
5060 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5063 /* If one device supports hw checksumming, set for all. */
5064 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5065 all &= ~NETIF_F_ALL_CSUM;
5066 all |= NETIF_F_HW_CSUM;
5070 one |= NETIF_F_ALL_CSUM;
5072 one |= all & NETIF_F_ONE_FOR_ALL;
5073 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
5074 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5078 EXPORT_SYMBOL(netdev_increment_features);
5080 static struct hlist_head *netdev_create_hash(void)
5083 struct hlist_head *hash;
5085 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5087 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5088 INIT_HLIST_HEAD(&hash[i]);
5093 /* Initialize per network namespace state */
5094 static int __net_init netdev_init(struct net *net)
5096 INIT_LIST_HEAD(&net->dev_base_head);
5098 net->dev_name_head = netdev_create_hash();
5099 if (net->dev_name_head == NULL)
5102 net->dev_index_head = netdev_create_hash();
5103 if (net->dev_index_head == NULL)
5109 kfree(net->dev_name_head);
5115 * netdev_drivername - network driver for the device
5116 * @dev: network device
5117 * @buffer: buffer for resulting name
5118 * @len: size of buffer
5120 * Determine network driver for device.
5122 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5124 const struct device_driver *driver;
5125 const struct device *parent;
5127 if (len <= 0 || !buffer)
5131 parent = dev->dev.parent;
5136 driver = parent->driver;
5137 if (driver && driver->name)
5138 strlcpy(buffer, driver->name, len);
5142 static void __net_exit netdev_exit(struct net *net)
5144 kfree(net->dev_name_head);
5145 kfree(net->dev_index_head);
5148 static struct pernet_operations __net_initdata netdev_net_ops = {
5149 .init = netdev_init,
5150 .exit = netdev_exit,
5153 static void __net_exit default_device_exit(struct net *net)
5155 struct net_device *dev;
5157 * Push all migratable of the network devices back to the
5158 * initial network namespace
5162 for_each_netdev(net, dev) {
5164 char fb_name[IFNAMSIZ];
5166 /* Ignore unmoveable devices (i.e. loopback) */
5167 if (dev->features & NETIF_F_NETNS_LOCAL)
5170 /* Delete virtual devices */
5171 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5172 dev->rtnl_link_ops->dellink(dev);
5176 /* Push remaing network devices to init_net */
5177 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5178 err = dev_change_net_namespace(dev, &init_net, fb_name);
5180 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5181 __func__, dev->name, err);
5189 static struct pernet_operations __net_initdata default_device_ops = {
5190 .exit = default_device_exit,
5194 * Initialize the DEV module. At boot time this walks the device list and
5195 * unhooks any devices that fail to initialise (normally hardware not
5196 * present) and leaves us with a valid list of present and active devices.
5201 * This is called single threaded during boot, so no need
5202 * to take the rtnl semaphore.
5204 static int __init net_dev_init(void)
5206 int i, rc = -ENOMEM;
5208 BUG_ON(!dev_boot_phase);
5210 if (dev_proc_init())
5213 if (netdev_kobject_init())
5216 INIT_LIST_HEAD(&ptype_all);
5217 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5218 INIT_LIST_HEAD(&ptype_base[i]);
5220 if (register_pernet_subsys(&netdev_net_ops))
5224 * Initialise the packet receive queues.
5227 for_each_possible_cpu(i) {
5228 struct softnet_data *queue;
5230 queue = &per_cpu(softnet_data, i);
5231 skb_queue_head_init(&queue->input_pkt_queue);
5232 queue->completion_queue = NULL;
5233 INIT_LIST_HEAD(&queue->poll_list);
5235 queue->backlog.poll = process_backlog;
5236 queue->backlog.weight = weight_p;
5237 queue->backlog.gro_list = NULL;
5238 queue->backlog.gro_count = 0;
5243 /* The loopback device is special if any other network devices
5244 * is present in a network namespace the loopback device must
5245 * be present. Since we now dynamically allocate and free the
5246 * loopback device ensure this invariant is maintained by
5247 * keeping the loopback device as the first device on the
5248 * list of network devices. Ensuring the loopback devices
5249 * is the first device that appears and the last network device
5252 if (register_pernet_device(&loopback_net_ops))
5255 if (register_pernet_device(&default_device_ops))
5258 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5259 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5261 hotcpu_notifier(dev_cpu_callback, 0);
5269 subsys_initcall(net_dev_init);
5271 static int __init initialize_hashrnd(void)
5273 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5277 late_initcall_sync(initialize_hashrnd);
5279 EXPORT_SYMBOL(__dev_get_by_index);
5280 EXPORT_SYMBOL(__dev_get_by_name);
5281 EXPORT_SYMBOL(__dev_remove_pack);
5282 EXPORT_SYMBOL(dev_valid_name);
5283 EXPORT_SYMBOL(dev_add_pack);
5284 EXPORT_SYMBOL(dev_alloc_name);
5285 EXPORT_SYMBOL(dev_close);
5286 EXPORT_SYMBOL(dev_get_by_flags);
5287 EXPORT_SYMBOL(dev_get_by_index);
5288 EXPORT_SYMBOL(dev_get_by_name);
5289 EXPORT_SYMBOL(dev_open);
5290 EXPORT_SYMBOL(dev_queue_xmit);
5291 EXPORT_SYMBOL(dev_remove_pack);
5292 EXPORT_SYMBOL(dev_set_allmulti);
5293 EXPORT_SYMBOL(dev_set_promiscuity);
5294 EXPORT_SYMBOL(dev_change_flags);
5295 EXPORT_SYMBOL(dev_set_mtu);
5296 EXPORT_SYMBOL(dev_set_mac_address);
5297 EXPORT_SYMBOL(free_netdev);
5298 EXPORT_SYMBOL(netdev_boot_setup_check);
5299 EXPORT_SYMBOL(netdev_set_master);
5300 EXPORT_SYMBOL(netdev_state_change);
5301 EXPORT_SYMBOL(netif_receive_skb);
5302 EXPORT_SYMBOL(netif_rx);
5303 EXPORT_SYMBOL(register_gifconf);
5304 EXPORT_SYMBOL(register_netdevice);
5305 EXPORT_SYMBOL(register_netdevice_notifier);
5306 EXPORT_SYMBOL(skb_checksum_help);
5307 EXPORT_SYMBOL(synchronize_net);
5308 EXPORT_SYMBOL(unregister_netdevice);
5309 EXPORT_SYMBOL(unregister_netdevice_notifier);
5310 EXPORT_SYMBOL(net_enable_timestamp);
5311 EXPORT_SYMBOL(net_disable_timestamp);
5312 EXPORT_SYMBOL(dev_get_flags);
5314 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
5315 EXPORT_SYMBOL(br_handle_frame_hook);
5316 EXPORT_SYMBOL(br_fdb_get_hook);
5317 EXPORT_SYMBOL(br_fdb_put_hook);
5320 EXPORT_SYMBOL(dev_load);
5322 EXPORT_PER_CPU_SYMBOL(softnet_data);