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;
1342 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1343 /* Never send packets back to the socket
1344 * they originated from - MvS (miquels@drinkel.ow.org)
1346 if ((ptype->dev == dev || !ptype->dev) &&
1347 (ptype->af_packet_priv == NULL ||
1348 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1349 struct sk_buff *skb2= skb_clone(skb, GFP_ATOMIC);
1353 /* skb->nh should be correctly
1354 set by sender, so that the second statement is
1355 just protection against buggy protocols.
1357 skb_reset_mac_header(skb2);
1359 if (skb_network_header(skb2) < skb2->data ||
1360 skb2->network_header > skb2->tail) {
1361 if (net_ratelimit())
1362 printk(KERN_CRIT "protocol %04x is "
1364 skb2->protocol, dev->name);
1365 skb_reset_network_header(skb2);
1368 skb2->transport_header = skb2->network_header;
1369 skb2->pkt_type = PACKET_OUTGOING;
1370 ptype->func(skb2, skb->dev, ptype, skb->dev);
1377 static inline void __netif_reschedule(struct Qdisc *q)
1379 struct softnet_data *sd;
1380 unsigned long flags;
1382 local_irq_save(flags);
1383 sd = &__get_cpu_var(softnet_data);
1384 q->next_sched = sd->output_queue;
1385 sd->output_queue = q;
1386 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1387 local_irq_restore(flags);
1390 void __netif_schedule(struct Qdisc *q)
1392 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1393 __netif_reschedule(q);
1395 EXPORT_SYMBOL(__netif_schedule);
1397 void dev_kfree_skb_irq(struct sk_buff *skb)
1399 if (atomic_dec_and_test(&skb->users)) {
1400 struct softnet_data *sd;
1401 unsigned long flags;
1403 local_irq_save(flags);
1404 sd = &__get_cpu_var(softnet_data);
1405 skb->next = sd->completion_queue;
1406 sd->completion_queue = skb;
1407 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1408 local_irq_restore(flags);
1411 EXPORT_SYMBOL(dev_kfree_skb_irq);
1413 void dev_kfree_skb_any(struct sk_buff *skb)
1415 if (in_irq() || irqs_disabled())
1416 dev_kfree_skb_irq(skb);
1420 EXPORT_SYMBOL(dev_kfree_skb_any);
1424 * netif_device_detach - mark device as removed
1425 * @dev: network device
1427 * Mark device as removed from system and therefore no longer available.
1429 void netif_device_detach(struct net_device *dev)
1431 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1432 netif_running(dev)) {
1433 netif_tx_stop_all_queues(dev);
1436 EXPORT_SYMBOL(netif_device_detach);
1439 * netif_device_attach - mark device as attached
1440 * @dev: network device
1442 * Mark device as attached from system and restart if needed.
1444 void netif_device_attach(struct net_device *dev)
1446 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1447 netif_running(dev)) {
1448 netif_tx_wake_all_queues(dev);
1449 __netdev_watchdog_up(dev);
1452 EXPORT_SYMBOL(netif_device_attach);
1454 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1456 return ((features & NETIF_F_GEN_CSUM) ||
1457 ((features & NETIF_F_IP_CSUM) &&
1458 protocol == htons(ETH_P_IP)) ||
1459 ((features & NETIF_F_IPV6_CSUM) &&
1460 protocol == htons(ETH_P_IPV6)) ||
1461 ((features & NETIF_F_FCOE_CRC) &&
1462 protocol == htons(ETH_P_FCOE)));
1465 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1467 if (can_checksum_protocol(dev->features, skb->protocol))
1470 if (skb->protocol == htons(ETH_P_8021Q)) {
1471 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1472 if (can_checksum_protocol(dev->features & dev->vlan_features,
1473 veh->h_vlan_encapsulated_proto))
1481 * Invalidate hardware checksum when packet is to be mangled, and
1482 * complete checksum manually on outgoing path.
1484 int skb_checksum_help(struct sk_buff *skb)
1487 int ret = 0, offset;
1489 if (skb->ip_summed == CHECKSUM_COMPLETE)
1490 goto out_set_summed;
1492 if (unlikely(skb_shinfo(skb)->gso_size)) {
1493 /* Let GSO fix up the checksum. */
1494 goto out_set_summed;
1497 offset = skb->csum_start - skb_headroom(skb);
1498 BUG_ON(offset >= skb_headlen(skb));
1499 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1501 offset += skb->csum_offset;
1502 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1504 if (skb_cloned(skb) &&
1505 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1506 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1511 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1513 skb->ip_summed = CHECKSUM_NONE;
1519 * skb_gso_segment - Perform segmentation on skb.
1520 * @skb: buffer to segment
1521 * @features: features for the output path (see dev->features)
1523 * This function segments the given skb and returns a list of segments.
1525 * It may return NULL if the skb requires no segmentation. This is
1526 * only possible when GSO is used for verifying header integrity.
1528 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1530 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1531 struct packet_type *ptype;
1532 __be16 type = skb->protocol;
1535 skb_reset_mac_header(skb);
1536 skb->mac_len = skb->network_header - skb->mac_header;
1537 __skb_pull(skb, skb->mac_len);
1539 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1540 struct net_device *dev = skb->dev;
1541 struct ethtool_drvinfo info = {};
1543 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1544 dev->ethtool_ops->get_drvinfo(dev, &info);
1546 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1548 info.driver, dev ? dev->features : 0L,
1549 skb->sk ? skb->sk->sk_route_caps : 0L,
1550 skb->len, skb->data_len, skb->ip_summed);
1552 if (skb_header_cloned(skb) &&
1553 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1554 return ERR_PTR(err);
1558 list_for_each_entry_rcu(ptype,
1559 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1560 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1561 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1562 err = ptype->gso_send_check(skb);
1563 segs = ERR_PTR(err);
1564 if (err || skb_gso_ok(skb, features))
1566 __skb_push(skb, (skb->data -
1567 skb_network_header(skb)));
1569 segs = ptype->gso_segment(skb, features);
1575 __skb_push(skb, skb->data - skb_mac_header(skb));
1580 EXPORT_SYMBOL(skb_gso_segment);
1582 /* Take action when hardware reception checksum errors are detected. */
1584 void netdev_rx_csum_fault(struct net_device *dev)
1586 if (net_ratelimit()) {
1587 printk(KERN_ERR "%s: hw csum failure.\n",
1588 dev ? dev->name : "<unknown>");
1592 EXPORT_SYMBOL(netdev_rx_csum_fault);
1595 /* Actually, we should eliminate this check as soon as we know, that:
1596 * 1. IOMMU is present and allows to map all the memory.
1597 * 2. No high memory really exists on this machine.
1600 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1602 #ifdef CONFIG_HIGHMEM
1605 if (dev->features & NETIF_F_HIGHDMA)
1608 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1609 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1617 void (*destructor)(struct sk_buff *skb);
1620 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1622 static void dev_gso_skb_destructor(struct sk_buff *skb)
1624 struct dev_gso_cb *cb;
1627 struct sk_buff *nskb = skb->next;
1629 skb->next = nskb->next;
1632 } while (skb->next);
1634 cb = DEV_GSO_CB(skb);
1636 cb->destructor(skb);
1640 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1641 * @skb: buffer to segment
1643 * This function segments the given skb and stores the list of segments
1646 static int dev_gso_segment(struct sk_buff *skb)
1648 struct net_device *dev = skb->dev;
1649 struct sk_buff *segs;
1650 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1653 segs = skb_gso_segment(skb, features);
1655 /* Verifying header integrity only. */
1660 return PTR_ERR(segs);
1663 DEV_GSO_CB(skb)->destructor = skb->destructor;
1664 skb->destructor = dev_gso_skb_destructor;
1669 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1670 struct netdev_queue *txq)
1672 const struct net_device_ops *ops = dev->netdev_ops;
1675 if (likely(!skb->next)) {
1676 if (!list_empty(&ptype_all))
1677 dev_queue_xmit_nit(skb, dev);
1679 if (netif_needs_gso(dev, skb)) {
1680 if (unlikely(dev_gso_segment(skb)))
1686 rc = ops->ndo_start_xmit(skb, dev);
1688 * TODO: if skb_orphan() was called by
1689 * dev->hard_start_xmit() (for example, the unmodified
1690 * igb driver does that; bnx2 doesn't), then
1691 * skb_tx_software_timestamp() will be unable to send
1692 * back the time stamp.
1694 * How can this be prevented? Always create another
1695 * reference to the socket before calling
1696 * dev->hard_start_xmit()? Prevent that skb_orphan()
1697 * does anything in dev->hard_start_xmit() by clearing
1698 * the skb destructor before the call and restoring it
1699 * afterwards, then doing the skb_orphan() ourselves?
1706 struct sk_buff *nskb = skb->next;
1708 skb->next = nskb->next;
1710 rc = ops->ndo_start_xmit(nskb, dev);
1712 nskb->next = skb->next;
1716 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1717 return NETDEV_TX_BUSY;
1718 } while (skb->next);
1720 skb->destructor = DEV_GSO_CB(skb)->destructor;
1727 static u32 skb_tx_hashrnd;
1729 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1733 if (skb_rx_queue_recorded(skb)) {
1734 hash = skb_get_rx_queue(skb);
1735 } else if (skb->sk && skb->sk->sk_hash) {
1736 hash = skb->sk->sk_hash;
1738 hash = skb->protocol;
1740 hash = jhash_1word(hash, skb_tx_hashrnd);
1742 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1744 EXPORT_SYMBOL(skb_tx_hash);
1746 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1747 struct sk_buff *skb)
1749 const struct net_device_ops *ops = dev->netdev_ops;
1750 u16 queue_index = 0;
1752 if (ops->ndo_select_queue)
1753 queue_index = ops->ndo_select_queue(dev, skb);
1754 else if (dev->real_num_tx_queues > 1)
1755 queue_index = skb_tx_hash(dev, skb);
1757 skb_set_queue_mapping(skb, queue_index);
1758 return netdev_get_tx_queue(dev, queue_index);
1762 * dev_queue_xmit - transmit a buffer
1763 * @skb: buffer to transmit
1765 * Queue a buffer for transmission to a network device. The caller must
1766 * have set the device and priority and built the buffer before calling
1767 * this function. The function can be called from an interrupt.
1769 * A negative errno code is returned on a failure. A success does not
1770 * guarantee the frame will be transmitted as it may be dropped due
1771 * to congestion or traffic shaping.
1773 * -----------------------------------------------------------------------------------
1774 * I notice this method can also return errors from the queue disciplines,
1775 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1778 * Regardless of the return value, the skb is consumed, so it is currently
1779 * difficult to retry a send to this method. (You can bump the ref count
1780 * before sending to hold a reference for retry if you are careful.)
1782 * When calling this method, interrupts MUST be enabled. This is because
1783 * the BH enable code must have IRQs enabled so that it will not deadlock.
1786 int dev_queue_xmit(struct sk_buff *skb)
1788 struct net_device *dev = skb->dev;
1789 struct netdev_queue *txq;
1793 /* GSO will handle the following emulations directly. */
1794 if (netif_needs_gso(dev, skb))
1797 if (skb_shinfo(skb)->frag_list &&
1798 !(dev->features & NETIF_F_FRAGLIST) &&
1799 __skb_linearize(skb))
1802 /* Fragmented skb is linearized if device does not support SG,
1803 * or if at least one of fragments is in highmem and device
1804 * does not support DMA from it.
1806 if (skb_shinfo(skb)->nr_frags &&
1807 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1808 __skb_linearize(skb))
1811 /* If packet is not checksummed and device does not support
1812 * checksumming for this protocol, complete checksumming here.
1814 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1815 skb_set_transport_header(skb, skb->csum_start -
1817 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1822 /* Disable soft irqs for various locks below. Also
1823 * stops preemption for RCU.
1827 txq = dev_pick_tx(dev, skb);
1828 q = rcu_dereference(txq->qdisc);
1830 #ifdef CONFIG_NET_CLS_ACT
1831 skb->tc_verd = SET_TC_AT(skb->tc_verd,AT_EGRESS);
1834 spinlock_t *root_lock = qdisc_lock(q);
1836 spin_lock(root_lock);
1838 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1842 rc = qdisc_enqueue_root(skb, q);
1845 spin_unlock(root_lock);
1850 /* The device has no queue. Common case for software devices:
1851 loopback, all the sorts of tunnels...
1853 Really, it is unlikely that netif_tx_lock protection is necessary
1854 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1856 However, it is possible, that they rely on protection
1859 Check this and shot the lock. It is not prone from deadlocks.
1860 Either shot noqueue qdisc, it is even simpler 8)
1862 if (dev->flags & IFF_UP) {
1863 int cpu = smp_processor_id(); /* ok because BHs are off */
1865 if (txq->xmit_lock_owner != cpu) {
1867 HARD_TX_LOCK(dev, txq, cpu);
1869 if (!netif_tx_queue_stopped(txq)) {
1871 if (!dev_hard_start_xmit(skb, dev, txq)) {
1872 HARD_TX_UNLOCK(dev, txq);
1876 HARD_TX_UNLOCK(dev, txq);
1877 if (net_ratelimit())
1878 printk(KERN_CRIT "Virtual device %s asks to "
1879 "queue packet!\n", dev->name);
1881 /* Recursion is detected! It is possible,
1883 if (net_ratelimit())
1884 printk(KERN_CRIT "Dead loop on virtual device "
1885 "%s, fix it urgently!\n", dev->name);
1890 rcu_read_unlock_bh();
1896 rcu_read_unlock_bh();
1901 /*=======================================================================
1903 =======================================================================*/
1905 int netdev_max_backlog __read_mostly = 1000;
1906 int netdev_budget __read_mostly = 300;
1907 int weight_p __read_mostly = 64; /* old backlog weight */
1909 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1913 * netif_rx - post buffer to the network code
1914 * @skb: buffer to post
1916 * This function receives a packet from a device driver and queues it for
1917 * the upper (protocol) levels to process. It always succeeds. The buffer
1918 * may be dropped during processing for congestion control or by the
1922 * NET_RX_SUCCESS (no congestion)
1923 * NET_RX_DROP (packet was dropped)
1927 int netif_rx(struct sk_buff *skb)
1929 struct softnet_data *queue;
1930 unsigned long flags;
1932 /* if netpoll wants it, pretend we never saw it */
1933 if (netpoll_rx(skb))
1936 if (!skb->tstamp.tv64)
1940 * The code is rearranged so that the path is the most
1941 * short when CPU is congested, but is still operating.
1943 local_irq_save(flags);
1944 queue = &__get_cpu_var(softnet_data);
1946 __get_cpu_var(netdev_rx_stat).total++;
1947 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
1948 if (queue->input_pkt_queue.qlen) {
1950 __skb_queue_tail(&queue->input_pkt_queue, skb);
1951 local_irq_restore(flags);
1952 return NET_RX_SUCCESS;
1955 napi_schedule(&queue->backlog);
1959 __get_cpu_var(netdev_rx_stat).dropped++;
1960 local_irq_restore(flags);
1966 int netif_rx_ni(struct sk_buff *skb)
1971 err = netif_rx(skb);
1972 if (local_softirq_pending())
1979 EXPORT_SYMBOL(netif_rx_ni);
1981 static void net_tx_action(struct softirq_action *h)
1983 struct softnet_data *sd = &__get_cpu_var(softnet_data);
1985 if (sd->completion_queue) {
1986 struct sk_buff *clist;
1988 local_irq_disable();
1989 clist = sd->completion_queue;
1990 sd->completion_queue = NULL;
1994 struct sk_buff *skb = clist;
1995 clist = clist->next;
1997 WARN_ON(atomic_read(&skb->users));
2002 if (sd->output_queue) {
2005 local_irq_disable();
2006 head = sd->output_queue;
2007 sd->output_queue = NULL;
2011 struct Qdisc *q = head;
2012 spinlock_t *root_lock;
2014 head = head->next_sched;
2016 root_lock = qdisc_lock(q);
2017 if (spin_trylock(root_lock)) {
2018 smp_mb__before_clear_bit();
2019 clear_bit(__QDISC_STATE_SCHED,
2022 spin_unlock(root_lock);
2024 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2026 __netif_reschedule(q);
2028 smp_mb__before_clear_bit();
2029 clear_bit(__QDISC_STATE_SCHED,
2037 static inline int deliver_skb(struct sk_buff *skb,
2038 struct packet_type *pt_prev,
2039 struct net_device *orig_dev)
2041 atomic_inc(&skb->users);
2042 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2045 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2046 /* These hooks defined here for ATM */
2048 struct net_bridge_fdb_entry *(*br_fdb_get_hook)(struct net_bridge *br,
2049 unsigned char *addr);
2050 void (*br_fdb_put_hook)(struct net_bridge_fdb_entry *ent) __read_mostly;
2053 * If bridge module is loaded call bridging hook.
2054 * returns NULL if packet was consumed.
2056 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2057 struct sk_buff *skb) __read_mostly;
2058 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2059 struct packet_type **pt_prev, int *ret,
2060 struct net_device *orig_dev)
2062 struct net_bridge_port *port;
2064 if (skb->pkt_type == PACKET_LOOPBACK ||
2065 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2069 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2073 return br_handle_frame_hook(port, skb);
2076 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2079 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2080 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2081 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2083 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2084 struct packet_type **pt_prev,
2086 struct net_device *orig_dev)
2088 if (skb->dev->macvlan_port == NULL)
2092 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2095 return macvlan_handle_frame_hook(skb);
2098 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2101 #ifdef CONFIG_NET_CLS_ACT
2102 /* TODO: Maybe we should just force sch_ingress to be compiled in
2103 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2104 * a compare and 2 stores extra right now if we dont have it on
2105 * but have CONFIG_NET_CLS_ACT
2106 * NOTE: This doesnt stop any functionality; if you dont have
2107 * the ingress scheduler, you just cant add policies on ingress.
2110 static int ing_filter(struct sk_buff *skb)
2112 struct net_device *dev = skb->dev;
2113 u32 ttl = G_TC_RTTL(skb->tc_verd);
2114 struct netdev_queue *rxq;
2115 int result = TC_ACT_OK;
2118 if (MAX_RED_LOOP < ttl++) {
2120 "Redir loop detected Dropping packet (%d->%d)\n",
2121 skb->iif, dev->ifindex);
2125 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2126 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2128 rxq = &dev->rx_queue;
2131 if (q != &noop_qdisc) {
2132 spin_lock(qdisc_lock(q));
2133 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2134 result = qdisc_enqueue_root(skb, q);
2135 spin_unlock(qdisc_lock(q));
2141 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2142 struct packet_type **pt_prev,
2143 int *ret, struct net_device *orig_dev)
2145 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2149 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2152 /* Huh? Why does turning on AF_PACKET affect this? */
2153 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2156 switch (ing_filter(skb)) {
2170 * netif_nit_deliver - deliver received packets to network taps
2173 * This function is used to deliver incoming packets to network
2174 * taps. It should be used when the normal netif_receive_skb path
2175 * is bypassed, for example because of VLAN acceleration.
2177 void netif_nit_deliver(struct sk_buff *skb)
2179 struct packet_type *ptype;
2181 if (list_empty(&ptype_all))
2184 skb_reset_network_header(skb);
2185 skb_reset_transport_header(skb);
2186 skb->mac_len = skb->network_header - skb->mac_header;
2189 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2190 if (!ptype->dev || ptype->dev == skb->dev)
2191 deliver_skb(skb, ptype, skb->dev);
2197 * netif_receive_skb - process receive buffer from network
2198 * @skb: buffer to process
2200 * netif_receive_skb() is the main receive data processing function.
2201 * It always succeeds. The buffer may be dropped during processing
2202 * for congestion control or by the protocol layers.
2204 * This function may only be called from softirq context and interrupts
2205 * should be enabled.
2207 * Return values (usually ignored):
2208 * NET_RX_SUCCESS: no congestion
2209 * NET_RX_DROP: packet was dropped
2211 int netif_receive_skb(struct sk_buff *skb)
2213 struct packet_type *ptype, *pt_prev;
2214 struct net_device *orig_dev;
2215 struct net_device *null_or_orig;
2216 int ret = NET_RX_DROP;
2219 if (skb->vlan_tci && vlan_hwaccel_do_receive(skb))
2220 return NET_RX_SUCCESS;
2222 /* if we've gotten here through NAPI, check netpoll */
2223 if (netpoll_receive_skb(skb))
2226 if (!skb->tstamp.tv64)
2230 skb->iif = skb->dev->ifindex;
2232 null_or_orig = NULL;
2233 orig_dev = skb->dev;
2234 if (orig_dev->master) {
2235 if (skb_bond_should_drop(skb))
2236 null_or_orig = orig_dev; /* deliver only exact match */
2238 skb->dev = orig_dev->master;
2241 __get_cpu_var(netdev_rx_stat).total++;
2243 skb_reset_network_header(skb);
2244 skb_reset_transport_header(skb);
2245 skb->mac_len = skb->network_header - skb->mac_header;
2251 #ifdef CONFIG_NET_CLS_ACT
2252 if (skb->tc_verd & TC_NCLS) {
2253 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2258 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2259 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2260 ptype->dev == orig_dev) {
2262 ret = deliver_skb(skb, pt_prev, orig_dev);
2267 #ifdef CONFIG_NET_CLS_ACT
2268 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2274 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2277 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2283 type = skb->protocol;
2284 list_for_each_entry_rcu(ptype,
2285 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2286 if (ptype->type == type &&
2287 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2288 ptype->dev == orig_dev)) {
2290 ret = deliver_skb(skb, pt_prev, orig_dev);
2296 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2299 /* Jamal, now you will not able to escape explaining
2300 * me how you were going to use this. :-)
2310 /* Network device is going away, flush any packets still pending */
2311 static void flush_backlog(void *arg)
2313 struct net_device *dev = arg;
2314 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2315 struct sk_buff *skb, *tmp;
2317 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2318 if (skb->dev == dev) {
2319 __skb_unlink(skb, &queue->input_pkt_queue);
2324 static int napi_gro_complete(struct sk_buff *skb)
2326 struct packet_type *ptype;
2327 __be16 type = skb->protocol;
2328 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2331 if (NAPI_GRO_CB(skb)->count == 1) {
2332 skb_shinfo(skb)->gso_size = 0;
2337 list_for_each_entry_rcu(ptype, head, list) {
2338 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2341 err = ptype->gro_complete(skb);
2347 WARN_ON(&ptype->list == head);
2349 return NET_RX_SUCCESS;
2353 return netif_receive_skb(skb);
2356 void napi_gro_flush(struct napi_struct *napi)
2358 struct sk_buff *skb, *next;
2360 for (skb = napi->gro_list; skb; skb = next) {
2363 napi_gro_complete(skb);
2366 napi->gro_count = 0;
2367 napi->gro_list = NULL;
2369 EXPORT_SYMBOL(napi_gro_flush);
2371 void *skb_gro_header(struct sk_buff *skb, unsigned int hlen)
2373 unsigned int offset = skb_gro_offset(skb);
2376 if (hlen <= skb_headlen(skb))
2377 return skb->data + offset;
2379 if (unlikely(!skb_shinfo(skb)->nr_frags ||
2380 skb_shinfo(skb)->frags[0].size <=
2381 hlen - skb_headlen(skb) ||
2382 PageHighMem(skb_shinfo(skb)->frags[0].page)))
2383 return pskb_may_pull(skb, hlen) ? skb->data + offset : NULL;
2385 return page_address(skb_shinfo(skb)->frags[0].page) +
2386 skb_shinfo(skb)->frags[0].page_offset +
2387 offset - skb_headlen(skb);
2389 EXPORT_SYMBOL(skb_gro_header);
2391 int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2393 struct sk_buff **pp = NULL;
2394 struct packet_type *ptype;
2395 __be16 type = skb->protocol;
2396 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2401 if (!(skb->dev->features & NETIF_F_GRO))
2404 if (skb_is_gso(skb) || skb_shinfo(skb)->frag_list)
2408 list_for_each_entry_rcu(ptype, head, list) {
2409 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2412 skb_set_network_header(skb, skb_gro_offset(skb));
2413 mac_len = skb->network_header - skb->mac_header;
2414 skb->mac_len = mac_len;
2415 NAPI_GRO_CB(skb)->same_flow = 0;
2416 NAPI_GRO_CB(skb)->flush = 0;
2417 NAPI_GRO_CB(skb)->free = 0;
2419 pp = ptype->gro_receive(&napi->gro_list, skb);
2424 if (&ptype->list == head)
2427 same_flow = NAPI_GRO_CB(skb)->same_flow;
2428 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2431 struct sk_buff *nskb = *pp;
2435 napi_gro_complete(nskb);
2442 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2446 NAPI_GRO_CB(skb)->count = 1;
2447 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2448 skb->next = napi->gro_list;
2449 napi->gro_list = skb;
2453 if (unlikely(!pskb_may_pull(skb, skb_gro_offset(skb)))) {
2454 if (napi->gro_list == skb)
2455 napi->gro_list = skb->next;
2466 EXPORT_SYMBOL(dev_gro_receive);
2468 static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2472 if (netpoll_rx_on(skb))
2475 for (p = napi->gro_list; p; p = p->next) {
2476 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2477 && !compare_ether_header(skb_mac_header(p),
2478 skb_gro_mac_header(skb));
2479 NAPI_GRO_CB(p)->flush = 0;
2482 return dev_gro_receive(napi, skb);
2485 int napi_skb_finish(int ret, struct sk_buff *skb)
2487 int err = NET_RX_SUCCESS;
2491 return netif_receive_skb(skb);
2497 case GRO_MERGED_FREE:
2504 EXPORT_SYMBOL(napi_skb_finish);
2506 int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2508 skb_gro_reset_offset(skb);
2510 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2512 EXPORT_SYMBOL(napi_gro_receive);
2514 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2516 __skb_pull(skb, skb_headlen(skb));
2517 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2521 EXPORT_SYMBOL(napi_reuse_skb);
2523 struct sk_buff *napi_fraginfo_skb(struct napi_struct *napi,
2524 struct napi_gro_fraginfo *info)
2526 struct net_device *dev = napi->dev;
2527 struct sk_buff *skb = napi->skb;
2535 skb = netdev_alloc_skb(dev, GRO_MAX_HEAD + NET_IP_ALIGN);
2539 skb_reserve(skb, NET_IP_ALIGN);
2542 BUG_ON(info->nr_frags > MAX_SKB_FRAGS);
2543 frag = &info->frags[info->nr_frags - 1];
2545 for (i = skb_shinfo(skb)->nr_frags; i < info->nr_frags; i++) {
2546 skb_fill_page_desc(skb, i, frag->page, frag->page_offset,
2550 skb_shinfo(skb)->nr_frags = info->nr_frags;
2552 skb->data_len = info->len;
2553 skb->len += info->len;
2554 skb->truesize += info->len;
2556 skb_reset_mac_header(skb);
2557 skb_gro_reset_offset(skb);
2559 eth = skb_gro_header(skb, sizeof(*eth));
2561 napi_reuse_skb(napi, skb);
2566 skb_gro_pull(skb, sizeof(*eth));
2569 * This works because the only protocols we care about don't require
2570 * special handling. We'll fix it up properly at the end.
2572 skb->protocol = eth->h_proto;
2574 skb->ip_summed = info->ip_summed;
2575 skb->csum = info->csum;
2580 EXPORT_SYMBOL(napi_fraginfo_skb);
2582 int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret)
2584 int err = NET_RX_SUCCESS;
2589 skb->protocol = eth_type_trans(skb, napi->dev);
2591 if (ret == GRO_NORMAL)
2592 return netif_receive_skb(skb);
2594 skb_gro_pull(skb, -ETH_HLEN);
2601 case GRO_MERGED_FREE:
2602 napi_reuse_skb(napi, skb);
2608 EXPORT_SYMBOL(napi_frags_finish);
2610 int napi_gro_frags(struct napi_struct *napi, struct napi_gro_fraginfo *info)
2612 struct sk_buff *skb = napi_fraginfo_skb(napi, info);
2617 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2619 EXPORT_SYMBOL(napi_gro_frags);
2621 static int process_backlog(struct napi_struct *napi, int quota)
2624 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2625 unsigned long start_time = jiffies;
2627 napi->weight = weight_p;
2629 struct sk_buff *skb;
2631 local_irq_disable();
2632 skb = __skb_dequeue(&queue->input_pkt_queue);
2634 __napi_complete(napi);
2640 netif_receive_skb(skb);
2641 } while (++work < quota && jiffies == start_time);
2647 * __napi_schedule - schedule for receive
2648 * @n: entry to schedule
2650 * The entry's receive function will be scheduled to run
2652 void __napi_schedule(struct napi_struct *n)
2654 unsigned long flags;
2656 local_irq_save(flags);
2657 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2658 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2659 local_irq_restore(flags);
2661 EXPORT_SYMBOL(__napi_schedule);
2663 void __napi_complete(struct napi_struct *n)
2665 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2666 BUG_ON(n->gro_list);
2668 list_del(&n->poll_list);
2669 smp_mb__before_clear_bit();
2670 clear_bit(NAPI_STATE_SCHED, &n->state);
2672 EXPORT_SYMBOL(__napi_complete);
2674 void napi_complete(struct napi_struct *n)
2676 unsigned long flags;
2679 * don't let napi dequeue from the cpu poll list
2680 * just in case its running on a different cpu
2682 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2686 local_irq_save(flags);
2688 local_irq_restore(flags);
2690 EXPORT_SYMBOL(napi_complete);
2692 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2693 int (*poll)(struct napi_struct *, int), int weight)
2695 INIT_LIST_HEAD(&napi->poll_list);
2696 napi->gro_count = 0;
2697 napi->gro_list = NULL;
2700 napi->weight = weight;
2701 list_add(&napi->dev_list, &dev->napi_list);
2703 #ifdef CONFIG_NETPOLL
2704 spin_lock_init(&napi->poll_lock);
2705 napi->poll_owner = -1;
2707 set_bit(NAPI_STATE_SCHED, &napi->state);
2709 EXPORT_SYMBOL(netif_napi_add);
2711 void netif_napi_del(struct napi_struct *napi)
2713 struct sk_buff *skb, *next;
2715 list_del_init(&napi->dev_list);
2716 kfree_skb(napi->skb);
2718 for (skb = napi->gro_list; skb; skb = next) {
2724 napi->gro_list = NULL;
2725 napi->gro_count = 0;
2727 EXPORT_SYMBOL(netif_napi_del);
2730 static void net_rx_action(struct softirq_action *h)
2732 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2733 unsigned long time_limit = jiffies + 2;
2734 int budget = netdev_budget;
2737 local_irq_disable();
2739 while (!list_empty(list)) {
2740 struct napi_struct *n;
2743 /* If softirq window is exhuasted then punt.
2744 * Allow this to run for 2 jiffies since which will allow
2745 * an average latency of 1.5/HZ.
2747 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2752 /* Even though interrupts have been re-enabled, this
2753 * access is safe because interrupts can only add new
2754 * entries to the tail of this list, and only ->poll()
2755 * calls can remove this head entry from the list.
2757 n = list_entry(list->next, struct napi_struct, poll_list);
2759 have = netpoll_poll_lock(n);
2763 /* This NAPI_STATE_SCHED test is for avoiding a race
2764 * with netpoll's poll_napi(). Only the entity which
2765 * obtains the lock and sees NAPI_STATE_SCHED set will
2766 * actually make the ->poll() call. Therefore we avoid
2767 * accidently calling ->poll() when NAPI is not scheduled.
2770 if (test_bit(NAPI_STATE_SCHED, &n->state))
2771 work = n->poll(n, weight);
2773 WARN_ON_ONCE(work > weight);
2777 local_irq_disable();
2779 /* Drivers must not modify the NAPI state if they
2780 * consume the entire weight. In such cases this code
2781 * still "owns" the NAPI instance and therefore can
2782 * move the instance around on the list at-will.
2784 if (unlikely(work == weight)) {
2785 if (unlikely(napi_disable_pending(n)))
2788 list_move_tail(&n->poll_list, list);
2791 netpoll_poll_unlock(have);
2796 #ifdef CONFIG_NET_DMA
2798 * There may not be any more sk_buffs coming right now, so push
2799 * any pending DMA copies to hardware
2801 dma_issue_pending_all();
2807 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2808 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2812 static gifconf_func_t * gifconf_list [NPROTO];
2815 * register_gifconf - register a SIOCGIF handler
2816 * @family: Address family
2817 * @gifconf: Function handler
2819 * Register protocol dependent address dumping routines. The handler
2820 * that is passed must not be freed or reused until it has been replaced
2821 * by another handler.
2823 int register_gifconf(unsigned int family, gifconf_func_t * gifconf)
2825 if (family >= NPROTO)
2827 gifconf_list[family] = gifconf;
2833 * Map an interface index to its name (SIOCGIFNAME)
2837 * We need this ioctl for efficient implementation of the
2838 * if_indextoname() function required by the IPv6 API. Without
2839 * it, we would have to search all the interfaces to find a
2843 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2845 struct net_device *dev;
2849 * Fetch the caller's info block.
2852 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2855 read_lock(&dev_base_lock);
2856 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2858 read_unlock(&dev_base_lock);
2862 strcpy(ifr.ifr_name, dev->name);
2863 read_unlock(&dev_base_lock);
2865 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2871 * Perform a SIOCGIFCONF call. This structure will change
2872 * size eventually, and there is nothing I can do about it.
2873 * Thus we will need a 'compatibility mode'.
2876 static int dev_ifconf(struct net *net, char __user *arg)
2879 struct net_device *dev;
2886 * Fetch the caller's info block.
2889 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2896 * Loop over the interfaces, and write an info block for each.
2900 for_each_netdev(net, dev) {
2901 for (i = 0; i < NPROTO; i++) {
2902 if (gifconf_list[i]) {
2905 done = gifconf_list[i](dev, NULL, 0);
2907 done = gifconf_list[i](dev, pos + total,
2917 * All done. Write the updated control block back to the caller.
2919 ifc.ifc_len = total;
2922 * Both BSD and Solaris return 0 here, so we do too.
2924 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
2927 #ifdef CONFIG_PROC_FS
2929 * This is invoked by the /proc filesystem handler to display a device
2932 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
2933 __acquires(dev_base_lock)
2935 struct net *net = seq_file_net(seq);
2937 struct net_device *dev;
2939 read_lock(&dev_base_lock);
2941 return SEQ_START_TOKEN;
2944 for_each_netdev(net, dev)
2951 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2953 struct net *net = seq_file_net(seq);
2955 return v == SEQ_START_TOKEN ?
2956 first_net_device(net) : next_net_device((struct net_device *)v);
2959 void dev_seq_stop(struct seq_file *seq, void *v)
2960 __releases(dev_base_lock)
2962 read_unlock(&dev_base_lock);
2965 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
2967 const struct net_device_stats *stats = dev_get_stats(dev);
2969 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
2970 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
2971 dev->name, stats->rx_bytes, stats->rx_packets,
2973 stats->rx_dropped + stats->rx_missed_errors,
2974 stats->rx_fifo_errors,
2975 stats->rx_length_errors + stats->rx_over_errors +
2976 stats->rx_crc_errors + stats->rx_frame_errors,
2977 stats->rx_compressed, stats->multicast,
2978 stats->tx_bytes, stats->tx_packets,
2979 stats->tx_errors, stats->tx_dropped,
2980 stats->tx_fifo_errors, stats->collisions,
2981 stats->tx_carrier_errors +
2982 stats->tx_aborted_errors +
2983 stats->tx_window_errors +
2984 stats->tx_heartbeat_errors,
2985 stats->tx_compressed);
2989 * Called from the PROCfs module. This now uses the new arbitrary sized
2990 * /proc/net interface to create /proc/net/dev
2992 static int dev_seq_show(struct seq_file *seq, void *v)
2994 if (v == SEQ_START_TOKEN)
2995 seq_puts(seq, "Inter-| Receive "
2997 " face |bytes packets errs drop fifo frame "
2998 "compressed multicast|bytes packets errs "
2999 "drop fifo colls carrier compressed\n");
3001 dev_seq_printf_stats(seq, v);
3005 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3007 struct netif_rx_stats *rc = NULL;
3009 while (*pos < nr_cpu_ids)
3010 if (cpu_online(*pos)) {
3011 rc = &per_cpu(netdev_rx_stat, *pos);
3018 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3020 return softnet_get_online(pos);
3023 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3026 return softnet_get_online(pos);
3029 static void softnet_seq_stop(struct seq_file *seq, void *v)
3033 static int softnet_seq_show(struct seq_file *seq, void *v)
3035 struct netif_rx_stats *s = v;
3037 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3038 s->total, s->dropped, s->time_squeeze, 0,
3039 0, 0, 0, 0, /* was fastroute */
3044 static const struct seq_operations dev_seq_ops = {
3045 .start = dev_seq_start,
3046 .next = dev_seq_next,
3047 .stop = dev_seq_stop,
3048 .show = dev_seq_show,
3051 static int dev_seq_open(struct inode *inode, struct file *file)
3053 return seq_open_net(inode, file, &dev_seq_ops,
3054 sizeof(struct seq_net_private));
3057 static const struct file_operations dev_seq_fops = {
3058 .owner = THIS_MODULE,
3059 .open = dev_seq_open,
3061 .llseek = seq_lseek,
3062 .release = seq_release_net,
3065 static const struct seq_operations softnet_seq_ops = {
3066 .start = softnet_seq_start,
3067 .next = softnet_seq_next,
3068 .stop = softnet_seq_stop,
3069 .show = softnet_seq_show,
3072 static int softnet_seq_open(struct inode *inode, struct file *file)
3074 return seq_open(file, &softnet_seq_ops);
3077 static const struct file_operations softnet_seq_fops = {
3078 .owner = THIS_MODULE,
3079 .open = softnet_seq_open,
3081 .llseek = seq_lseek,
3082 .release = seq_release,
3085 static void *ptype_get_idx(loff_t pos)
3087 struct packet_type *pt = NULL;
3091 list_for_each_entry_rcu(pt, &ptype_all, list) {
3097 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3098 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3107 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3111 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3114 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3116 struct packet_type *pt;
3117 struct list_head *nxt;
3121 if (v == SEQ_START_TOKEN)
3122 return ptype_get_idx(0);
3125 nxt = pt->list.next;
3126 if (pt->type == htons(ETH_P_ALL)) {
3127 if (nxt != &ptype_all)
3130 nxt = ptype_base[0].next;
3132 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3134 while (nxt == &ptype_base[hash]) {
3135 if (++hash >= PTYPE_HASH_SIZE)
3137 nxt = ptype_base[hash].next;
3140 return list_entry(nxt, struct packet_type, list);
3143 static void ptype_seq_stop(struct seq_file *seq, void *v)
3149 static int ptype_seq_show(struct seq_file *seq, void *v)
3151 struct packet_type *pt = v;
3153 if (v == SEQ_START_TOKEN)
3154 seq_puts(seq, "Type Device Function\n");
3155 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3156 if (pt->type == htons(ETH_P_ALL))
3157 seq_puts(seq, "ALL ");
3159 seq_printf(seq, "%04x", ntohs(pt->type));
3161 seq_printf(seq, " %-8s %pF\n",
3162 pt->dev ? pt->dev->name : "", pt->func);
3168 static const struct seq_operations ptype_seq_ops = {
3169 .start = ptype_seq_start,
3170 .next = ptype_seq_next,
3171 .stop = ptype_seq_stop,
3172 .show = ptype_seq_show,
3175 static int ptype_seq_open(struct inode *inode, struct file *file)
3177 return seq_open_net(inode, file, &ptype_seq_ops,
3178 sizeof(struct seq_net_private));
3181 static const struct file_operations ptype_seq_fops = {
3182 .owner = THIS_MODULE,
3183 .open = ptype_seq_open,
3185 .llseek = seq_lseek,
3186 .release = seq_release_net,
3190 static int __net_init dev_proc_net_init(struct net *net)
3194 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3196 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3198 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3201 if (wext_proc_init(net))
3207 proc_net_remove(net, "ptype");
3209 proc_net_remove(net, "softnet_stat");
3211 proc_net_remove(net, "dev");
3215 static void __net_exit dev_proc_net_exit(struct net *net)
3217 wext_proc_exit(net);
3219 proc_net_remove(net, "ptype");
3220 proc_net_remove(net, "softnet_stat");
3221 proc_net_remove(net, "dev");
3224 static struct pernet_operations __net_initdata dev_proc_ops = {
3225 .init = dev_proc_net_init,
3226 .exit = dev_proc_net_exit,
3229 static int __init dev_proc_init(void)
3231 return register_pernet_subsys(&dev_proc_ops);
3234 #define dev_proc_init() 0
3235 #endif /* CONFIG_PROC_FS */
3239 * netdev_set_master - set up master/slave pair
3240 * @slave: slave device
3241 * @master: new master device
3243 * Changes the master device of the slave. Pass %NULL to break the
3244 * bonding. The caller must hold the RTNL semaphore. On a failure
3245 * a negative errno code is returned. On success the reference counts
3246 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3247 * function returns zero.
3249 int netdev_set_master(struct net_device *slave, struct net_device *master)
3251 struct net_device *old = slave->master;
3261 slave->master = master;
3269 slave->flags |= IFF_SLAVE;
3271 slave->flags &= ~IFF_SLAVE;
3273 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3277 static void dev_change_rx_flags(struct net_device *dev, int flags)
3279 const struct net_device_ops *ops = dev->netdev_ops;
3281 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3282 ops->ndo_change_rx_flags(dev, flags);
3285 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3287 unsigned short old_flags = dev->flags;
3293 dev->flags |= IFF_PROMISC;
3294 dev->promiscuity += inc;
3295 if (dev->promiscuity == 0) {
3298 * If inc causes overflow, untouch promisc and return error.
3301 dev->flags &= ~IFF_PROMISC;
3303 dev->promiscuity -= inc;
3304 printk(KERN_WARNING "%s: promiscuity touches roof, "
3305 "set promiscuity failed, promiscuity feature "
3306 "of device might be broken.\n", dev->name);
3310 if (dev->flags != old_flags) {
3311 printk(KERN_INFO "device %s %s promiscuous mode\n",
3312 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3314 if (audit_enabled) {
3315 current_uid_gid(&uid, &gid);
3316 audit_log(current->audit_context, GFP_ATOMIC,
3317 AUDIT_ANOM_PROMISCUOUS,
3318 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3319 dev->name, (dev->flags & IFF_PROMISC),
3320 (old_flags & IFF_PROMISC),
3321 audit_get_loginuid(current),
3323 audit_get_sessionid(current));
3326 dev_change_rx_flags(dev, IFF_PROMISC);
3332 * dev_set_promiscuity - update promiscuity count on a device
3336 * Add or remove promiscuity from a device. While the count in the device
3337 * remains above zero the interface remains promiscuous. Once it hits zero
3338 * the device reverts back to normal filtering operation. A negative inc
3339 * value is used to drop promiscuity on the device.
3340 * Return 0 if successful or a negative errno code on error.
3342 int dev_set_promiscuity(struct net_device *dev, int inc)
3344 unsigned short old_flags = dev->flags;
3347 err = __dev_set_promiscuity(dev, inc);
3350 if (dev->flags != old_flags)
3351 dev_set_rx_mode(dev);
3356 * dev_set_allmulti - update allmulti count on a device
3360 * Add or remove reception of all multicast frames to a device. While the
3361 * count in the device remains above zero the interface remains listening
3362 * to all interfaces. Once it hits zero the device reverts back to normal
3363 * filtering operation. A negative @inc value is used to drop the counter
3364 * when releasing a resource needing all multicasts.
3365 * Return 0 if successful or a negative errno code on error.
3368 int dev_set_allmulti(struct net_device *dev, int inc)
3370 unsigned short old_flags = dev->flags;
3374 dev->flags |= IFF_ALLMULTI;
3375 dev->allmulti += inc;
3376 if (dev->allmulti == 0) {
3379 * If inc causes overflow, untouch allmulti and return error.
3382 dev->flags &= ~IFF_ALLMULTI;
3384 dev->allmulti -= inc;
3385 printk(KERN_WARNING "%s: allmulti touches roof, "
3386 "set allmulti failed, allmulti feature of "
3387 "device might be broken.\n", dev->name);
3391 if (dev->flags ^ old_flags) {
3392 dev_change_rx_flags(dev, IFF_ALLMULTI);
3393 dev_set_rx_mode(dev);
3399 * Upload unicast and multicast address lists to device and
3400 * configure RX filtering. When the device doesn't support unicast
3401 * filtering it is put in promiscuous mode while unicast addresses
3404 void __dev_set_rx_mode(struct net_device *dev)
3406 const struct net_device_ops *ops = dev->netdev_ops;
3408 /* dev_open will call this function so the list will stay sane. */
3409 if (!(dev->flags&IFF_UP))
3412 if (!netif_device_present(dev))
3415 if (ops->ndo_set_rx_mode)
3416 ops->ndo_set_rx_mode(dev);
3418 /* Unicast addresses changes may only happen under the rtnl,
3419 * therefore calling __dev_set_promiscuity here is safe.
3421 if (dev->uc_count > 0 && !dev->uc_promisc) {
3422 __dev_set_promiscuity(dev, 1);
3423 dev->uc_promisc = 1;
3424 } else if (dev->uc_count == 0 && dev->uc_promisc) {
3425 __dev_set_promiscuity(dev, -1);
3426 dev->uc_promisc = 0;
3429 if (ops->ndo_set_multicast_list)
3430 ops->ndo_set_multicast_list(dev);
3434 void dev_set_rx_mode(struct net_device *dev)
3436 netif_addr_lock_bh(dev);
3437 __dev_set_rx_mode(dev);
3438 netif_addr_unlock_bh(dev);
3441 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3442 void *addr, int alen, int glbl)
3444 struct dev_addr_list *da;
3446 for (; (da = *list) != NULL; list = &da->next) {
3447 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3448 alen == da->da_addrlen) {
3450 int old_glbl = da->da_gusers;
3467 int __dev_addr_add(struct dev_addr_list **list, int *count,
3468 void *addr, int alen, int glbl)
3470 struct dev_addr_list *da;
3472 for (da = *list; da != NULL; da = da->next) {
3473 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3474 da->da_addrlen == alen) {
3476 int old_glbl = da->da_gusers;
3486 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3489 memcpy(da->da_addr, addr, alen);
3490 da->da_addrlen = alen;
3492 da->da_gusers = glbl ? 1 : 0;
3500 * dev_unicast_delete - Release secondary unicast address.
3502 * @addr: address to delete
3503 * @alen: length of @addr
3505 * Release reference to a secondary unicast address and remove it
3506 * from the device if the reference count drops to zero.
3508 * The caller must hold the rtnl_mutex.
3510 int dev_unicast_delete(struct net_device *dev, void *addr, int alen)
3516 netif_addr_lock_bh(dev);
3517 err = __dev_addr_delete(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3519 __dev_set_rx_mode(dev);
3520 netif_addr_unlock_bh(dev);
3523 EXPORT_SYMBOL(dev_unicast_delete);
3526 * dev_unicast_add - add a secondary unicast address
3528 * @addr: address to add
3529 * @alen: length of @addr
3531 * Add a secondary unicast address to the device or increase
3532 * the reference count if it already exists.
3534 * The caller must hold the rtnl_mutex.
3536 int dev_unicast_add(struct net_device *dev, void *addr, int alen)
3542 netif_addr_lock_bh(dev);
3543 err = __dev_addr_add(&dev->uc_list, &dev->uc_count, addr, alen, 0);
3545 __dev_set_rx_mode(dev);
3546 netif_addr_unlock_bh(dev);
3549 EXPORT_SYMBOL(dev_unicast_add);
3551 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3552 struct dev_addr_list **from, int *from_count)
3554 struct dev_addr_list *da, *next;
3558 while (da != NULL) {
3560 if (!da->da_synced) {
3561 err = __dev_addr_add(to, to_count,
3562 da->da_addr, da->da_addrlen, 0);
3567 } else if (da->da_users == 1) {
3568 __dev_addr_delete(to, to_count,
3569 da->da_addr, da->da_addrlen, 0);
3570 __dev_addr_delete(from, from_count,
3571 da->da_addr, da->da_addrlen, 0);
3578 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3579 struct dev_addr_list **from, int *from_count)
3581 struct dev_addr_list *da, *next;
3584 while (da != NULL) {
3586 if (da->da_synced) {
3587 __dev_addr_delete(to, to_count,
3588 da->da_addr, da->da_addrlen, 0);
3590 __dev_addr_delete(from, from_count,
3591 da->da_addr, da->da_addrlen, 0);
3598 * dev_unicast_sync - Synchronize device's unicast list to another device
3599 * @to: destination device
3600 * @from: source device
3602 * Add newly added addresses to the destination device and release
3603 * addresses that have no users left. The source device must be
3604 * locked by netif_tx_lock_bh.
3606 * This function is intended to be called from the dev->set_rx_mode
3607 * function of layered software devices.
3609 int dev_unicast_sync(struct net_device *to, struct net_device *from)
3613 netif_addr_lock_bh(to);
3614 err = __dev_addr_sync(&to->uc_list, &to->uc_count,
3615 &from->uc_list, &from->uc_count);
3617 __dev_set_rx_mode(to);
3618 netif_addr_unlock_bh(to);
3621 EXPORT_SYMBOL(dev_unicast_sync);
3624 * dev_unicast_unsync - Remove synchronized addresses from the destination device
3625 * @to: destination device
3626 * @from: source device
3628 * Remove all addresses that were added to the destination device by
3629 * dev_unicast_sync(). This function is intended to be called from the
3630 * dev->stop function of layered software devices.
3632 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
3634 netif_addr_lock_bh(from);
3635 netif_addr_lock(to);
3637 __dev_addr_unsync(&to->uc_list, &to->uc_count,
3638 &from->uc_list, &from->uc_count);
3639 __dev_set_rx_mode(to);
3641 netif_addr_unlock(to);
3642 netif_addr_unlock_bh(from);
3644 EXPORT_SYMBOL(dev_unicast_unsync);
3646 static void __dev_addr_discard(struct dev_addr_list **list)
3648 struct dev_addr_list *tmp;
3650 while (*list != NULL) {
3653 if (tmp->da_users > tmp->da_gusers)
3654 printk("__dev_addr_discard: address leakage! "
3655 "da_users=%d\n", tmp->da_users);
3660 static void dev_addr_discard(struct net_device *dev)
3662 netif_addr_lock_bh(dev);
3664 __dev_addr_discard(&dev->uc_list);
3667 __dev_addr_discard(&dev->mc_list);
3670 netif_addr_unlock_bh(dev);
3674 * dev_get_flags - get flags reported to userspace
3677 * Get the combination of flag bits exported through APIs to userspace.
3679 unsigned dev_get_flags(const struct net_device *dev)
3683 flags = (dev->flags & ~(IFF_PROMISC |
3688 (dev->gflags & (IFF_PROMISC |
3691 if (netif_running(dev)) {
3692 if (netif_oper_up(dev))
3693 flags |= IFF_RUNNING;
3694 if (netif_carrier_ok(dev))
3695 flags |= IFF_LOWER_UP;
3696 if (netif_dormant(dev))
3697 flags |= IFF_DORMANT;
3704 * dev_change_flags - change device settings
3706 * @flags: device state flags
3708 * Change settings on device based state flags. The flags are
3709 * in the userspace exported format.
3711 int dev_change_flags(struct net_device *dev, unsigned flags)
3714 int old_flags = dev->flags;
3719 * Set the flags on our device.
3722 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
3723 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
3725 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
3729 * Load in the correct multicast list now the flags have changed.
3732 if ((old_flags ^ flags) & IFF_MULTICAST)
3733 dev_change_rx_flags(dev, IFF_MULTICAST);
3735 dev_set_rx_mode(dev);
3738 * Have we downed the interface. We handle IFF_UP ourselves
3739 * according to user attempts to set it, rather than blindly
3744 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
3745 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
3748 dev_set_rx_mode(dev);
3751 if (dev->flags & IFF_UP &&
3752 ((old_flags ^ dev->flags) &~ (IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
3754 call_netdevice_notifiers(NETDEV_CHANGE, dev);
3756 if ((flags ^ dev->gflags) & IFF_PROMISC) {
3757 int inc = (flags & IFF_PROMISC) ? +1 : -1;
3758 dev->gflags ^= IFF_PROMISC;
3759 dev_set_promiscuity(dev, inc);
3762 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
3763 is important. Some (broken) drivers set IFF_PROMISC, when
3764 IFF_ALLMULTI is requested not asking us and not reporting.
3766 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
3767 int inc = (flags & IFF_ALLMULTI) ? +1 : -1;
3768 dev->gflags ^= IFF_ALLMULTI;
3769 dev_set_allmulti(dev, inc);
3772 /* Exclude state transition flags, already notified */
3773 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
3775 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
3781 * dev_set_mtu - Change maximum transfer unit
3783 * @new_mtu: new transfer unit
3785 * Change the maximum transfer size of the network device.
3787 int dev_set_mtu(struct net_device *dev, int new_mtu)
3789 const struct net_device_ops *ops = dev->netdev_ops;
3792 if (new_mtu == dev->mtu)
3795 /* MTU must be positive. */
3799 if (!netif_device_present(dev))
3803 if (ops->ndo_change_mtu)
3804 err = ops->ndo_change_mtu(dev, new_mtu);
3808 if (!err && dev->flags & IFF_UP)
3809 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
3814 * dev_set_mac_address - Change Media Access Control Address
3818 * Change the hardware (MAC) address of the device
3820 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
3822 const struct net_device_ops *ops = dev->netdev_ops;
3825 if (!ops->ndo_set_mac_address)
3827 if (sa->sa_family != dev->type)
3829 if (!netif_device_present(dev))
3831 err = ops->ndo_set_mac_address(dev, sa);
3833 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3838 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
3840 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
3843 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3849 case SIOCGIFFLAGS: /* Get interface flags */
3850 ifr->ifr_flags = dev_get_flags(dev);
3853 case SIOCGIFMETRIC: /* Get the metric on the interface
3854 (currently unused) */
3855 ifr->ifr_metric = 0;
3858 case SIOCGIFMTU: /* Get the MTU of a device */
3859 ifr->ifr_mtu = dev->mtu;
3864 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
3866 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
3867 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3868 ifr->ifr_hwaddr.sa_family = dev->type;
3876 ifr->ifr_map.mem_start = dev->mem_start;
3877 ifr->ifr_map.mem_end = dev->mem_end;
3878 ifr->ifr_map.base_addr = dev->base_addr;
3879 ifr->ifr_map.irq = dev->irq;
3880 ifr->ifr_map.dma = dev->dma;
3881 ifr->ifr_map.port = dev->if_port;
3885 ifr->ifr_ifindex = dev->ifindex;
3889 ifr->ifr_qlen = dev->tx_queue_len;
3893 /* dev_ioctl() should ensure this case
3905 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
3907 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
3910 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
3911 const struct net_device_ops *ops;
3916 ops = dev->netdev_ops;
3919 case SIOCSIFFLAGS: /* Set interface flags */
3920 return dev_change_flags(dev, ifr->ifr_flags);
3922 case SIOCSIFMETRIC: /* Set the metric on the interface
3923 (currently unused) */
3926 case SIOCSIFMTU: /* Set the MTU of a device */
3927 return dev_set_mtu(dev, ifr->ifr_mtu);
3930 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
3932 case SIOCSIFHWBROADCAST:
3933 if (ifr->ifr_hwaddr.sa_family != dev->type)
3935 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
3936 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
3937 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3941 if (ops->ndo_set_config) {
3942 if (!netif_device_present(dev))
3944 return ops->ndo_set_config(dev, &ifr->ifr_map);
3949 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
3950 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3952 if (!netif_device_present(dev))
3954 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
3958 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
3959 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
3961 if (!netif_device_present(dev))
3963 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
3967 if (ifr->ifr_qlen < 0)
3969 dev->tx_queue_len = ifr->ifr_qlen;
3973 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
3974 return dev_change_name(dev, ifr->ifr_newname);
3977 * Unknown or private ioctl
3981 if ((cmd >= SIOCDEVPRIVATE &&
3982 cmd <= SIOCDEVPRIVATE + 15) ||
3983 cmd == SIOCBONDENSLAVE ||
3984 cmd == SIOCBONDRELEASE ||
3985 cmd == SIOCBONDSETHWADDR ||
3986 cmd == SIOCBONDSLAVEINFOQUERY ||
3987 cmd == SIOCBONDINFOQUERY ||
3988 cmd == SIOCBONDCHANGEACTIVE ||
3989 cmd == SIOCGMIIPHY ||
3990 cmd == SIOCGMIIREG ||
3991 cmd == SIOCSMIIREG ||
3992 cmd == SIOCBRADDIF ||
3993 cmd == SIOCBRDELIF ||
3994 cmd == SIOCSHWTSTAMP ||
3995 cmd == SIOCWANDEV) {
3997 if (ops->ndo_do_ioctl) {
3998 if (netif_device_present(dev))
3999 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4011 * This function handles all "interface"-type I/O control requests. The actual
4012 * 'doing' part of this is dev_ifsioc above.
4016 * dev_ioctl - network device ioctl
4017 * @net: the applicable net namespace
4018 * @cmd: command to issue
4019 * @arg: pointer to a struct ifreq in user space
4021 * Issue ioctl functions to devices. This is normally called by the
4022 * user space syscall interfaces but can sometimes be useful for
4023 * other purposes. The return value is the return from the syscall if
4024 * positive or a negative errno code on error.
4027 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4033 /* One special case: SIOCGIFCONF takes ifconf argument
4034 and requires shared lock, because it sleeps writing
4038 if (cmd == SIOCGIFCONF) {
4040 ret = dev_ifconf(net, (char __user *) arg);
4044 if (cmd == SIOCGIFNAME)
4045 return dev_ifname(net, (struct ifreq __user *)arg);
4047 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4050 ifr.ifr_name[IFNAMSIZ-1] = 0;
4052 colon = strchr(ifr.ifr_name, ':');
4057 * See which interface the caller is talking about.
4062 * These ioctl calls:
4063 * - can be done by all.
4064 * - atomic and do not require locking.
4075 dev_load(net, ifr.ifr_name);
4076 read_lock(&dev_base_lock);
4077 ret = dev_ifsioc_locked(net, &ifr, cmd);
4078 read_unlock(&dev_base_lock);
4082 if (copy_to_user(arg, &ifr,
4083 sizeof(struct ifreq)))
4089 dev_load(net, ifr.ifr_name);
4091 ret = dev_ethtool(net, &ifr);
4096 if (copy_to_user(arg, &ifr,
4097 sizeof(struct ifreq)))
4103 * These ioctl calls:
4104 * - require superuser power.
4105 * - require strict serialization.
4111 if (!capable(CAP_NET_ADMIN))
4113 dev_load(net, ifr.ifr_name);
4115 ret = dev_ifsioc(net, &ifr, cmd);
4120 if (copy_to_user(arg, &ifr,
4121 sizeof(struct ifreq)))
4127 * These ioctl calls:
4128 * - require superuser power.
4129 * - require strict serialization.
4130 * - do not return a value
4140 case SIOCSIFHWBROADCAST:
4143 case SIOCBONDENSLAVE:
4144 case SIOCBONDRELEASE:
4145 case SIOCBONDSETHWADDR:
4146 case SIOCBONDCHANGEACTIVE:
4150 if (!capable(CAP_NET_ADMIN))
4153 case SIOCBONDSLAVEINFOQUERY:
4154 case SIOCBONDINFOQUERY:
4155 dev_load(net, ifr.ifr_name);
4157 ret = dev_ifsioc(net, &ifr, cmd);
4162 /* Get the per device memory space. We can add this but
4163 * currently do not support it */
4165 /* Set the per device memory buffer space.
4166 * Not applicable in our case */
4171 * Unknown or private ioctl.
4174 if (cmd == SIOCWANDEV ||
4175 (cmd >= SIOCDEVPRIVATE &&
4176 cmd <= SIOCDEVPRIVATE + 15)) {
4177 dev_load(net, ifr.ifr_name);
4179 ret = dev_ifsioc(net, &ifr, cmd);
4181 if (!ret && copy_to_user(arg, &ifr,
4182 sizeof(struct ifreq)))
4186 /* Take care of Wireless Extensions */
4187 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4188 return wext_handle_ioctl(net, &ifr, cmd, arg);
4195 * dev_new_index - allocate an ifindex
4196 * @net: the applicable net namespace
4198 * Returns a suitable unique value for a new device interface
4199 * number. The caller must hold the rtnl semaphore or the
4200 * dev_base_lock to be sure it remains unique.
4202 static int dev_new_index(struct net *net)
4208 if (!__dev_get_by_index(net, ifindex))
4213 /* Delayed registration/unregisteration */
4214 static LIST_HEAD(net_todo_list);
4216 static void net_set_todo(struct net_device *dev)
4218 list_add_tail(&dev->todo_list, &net_todo_list);
4221 static void rollback_registered(struct net_device *dev)
4223 BUG_ON(dev_boot_phase);
4226 /* Some devices call without registering for initialization unwind. */
4227 if (dev->reg_state == NETREG_UNINITIALIZED) {
4228 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
4229 "was registered\n", dev->name, dev);
4235 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4237 /* If device is running, close it first. */
4240 /* And unlink it from device chain. */
4241 unlist_netdevice(dev);
4243 dev->reg_state = NETREG_UNREGISTERING;
4247 /* Shutdown queueing discipline. */
4251 /* Notify protocols, that we are about to destroy
4252 this device. They should clean all the things.
4254 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4257 * Flush the unicast and multicast chains
4259 dev_addr_discard(dev);
4261 if (dev->netdev_ops->ndo_uninit)
4262 dev->netdev_ops->ndo_uninit(dev);
4264 /* Notifier chain MUST detach us from master device. */
4265 WARN_ON(dev->master);
4267 /* Remove entries from kobject tree */
4268 netdev_unregister_kobject(dev);
4275 static void __netdev_init_queue_locks_one(struct net_device *dev,
4276 struct netdev_queue *dev_queue,
4279 spin_lock_init(&dev_queue->_xmit_lock);
4280 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4281 dev_queue->xmit_lock_owner = -1;
4284 static void netdev_init_queue_locks(struct net_device *dev)
4286 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4287 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4290 unsigned long netdev_fix_features(unsigned long features, const char *name)
4292 /* Fix illegal SG+CSUM combinations. */
4293 if ((features & NETIF_F_SG) &&
4294 !(features & NETIF_F_ALL_CSUM)) {
4296 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4297 "checksum feature.\n", name);
4298 features &= ~NETIF_F_SG;
4301 /* TSO requires that SG is present as well. */
4302 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4304 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4305 "SG feature.\n", name);
4306 features &= ~NETIF_F_TSO;
4309 if (features & NETIF_F_UFO) {
4310 if (!(features & NETIF_F_GEN_CSUM)) {
4312 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4313 "since no NETIF_F_HW_CSUM feature.\n",
4315 features &= ~NETIF_F_UFO;
4318 if (!(features & NETIF_F_SG)) {
4320 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4321 "since no NETIF_F_SG feature.\n", name);
4322 features &= ~NETIF_F_UFO;
4328 EXPORT_SYMBOL(netdev_fix_features);
4330 /* Some devices need to (re-)set their netdev_ops inside
4331 * ->init() or similar. If that happens, we have to setup
4332 * the compat pointers again.
4334 void netdev_resync_ops(struct net_device *dev)
4336 #ifdef CONFIG_COMPAT_NET_DEV_OPS
4337 const struct net_device_ops *ops = dev->netdev_ops;
4339 dev->init = ops->ndo_init;
4340 dev->uninit = ops->ndo_uninit;
4341 dev->open = ops->ndo_open;
4342 dev->change_rx_flags = ops->ndo_change_rx_flags;
4343 dev->set_rx_mode = ops->ndo_set_rx_mode;
4344 dev->set_multicast_list = ops->ndo_set_multicast_list;
4345 dev->set_mac_address = ops->ndo_set_mac_address;
4346 dev->validate_addr = ops->ndo_validate_addr;
4347 dev->do_ioctl = ops->ndo_do_ioctl;
4348 dev->set_config = ops->ndo_set_config;
4349 dev->change_mtu = ops->ndo_change_mtu;
4350 dev->neigh_setup = ops->ndo_neigh_setup;
4351 dev->tx_timeout = ops->ndo_tx_timeout;
4352 dev->get_stats = ops->ndo_get_stats;
4353 dev->vlan_rx_register = ops->ndo_vlan_rx_register;
4354 dev->vlan_rx_add_vid = ops->ndo_vlan_rx_add_vid;
4355 dev->vlan_rx_kill_vid = ops->ndo_vlan_rx_kill_vid;
4356 #ifdef CONFIG_NET_POLL_CONTROLLER
4357 dev->poll_controller = ops->ndo_poll_controller;
4361 EXPORT_SYMBOL(netdev_resync_ops);
4364 * register_netdevice - register a network device
4365 * @dev: device to register
4367 * Take a completed network device structure and add it to the kernel
4368 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4369 * chain. 0 is returned on success. A negative errno code is returned
4370 * on a failure to set up the device, or if the name is a duplicate.
4372 * Callers must hold the rtnl semaphore. You may want
4373 * register_netdev() instead of this.
4376 * The locking appears insufficient to guarantee two parallel registers
4377 * will not get the same name.
4380 int register_netdevice(struct net_device *dev)
4382 struct hlist_head *head;
4383 struct hlist_node *p;
4385 struct net *net = dev_net(dev);
4387 BUG_ON(dev_boot_phase);
4392 /* When net_device's are persistent, this will be fatal. */
4393 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4396 spin_lock_init(&dev->addr_list_lock);
4397 netdev_set_addr_lockdep_class(dev);
4398 netdev_init_queue_locks(dev);
4402 #ifdef CONFIG_COMPAT_NET_DEV_OPS
4403 /* Netdevice_ops API compatiability support.
4404 * This is temporary until all network devices are converted.
4406 if (dev->netdev_ops) {
4407 netdev_resync_ops(dev);
4409 char drivername[64];
4410 pr_info("%s (%s): not using net_device_ops yet\n",
4411 dev->name, netdev_drivername(dev, drivername, 64));
4413 /* This works only because net_device_ops and the
4414 compatiablity structure are the same. */
4415 dev->netdev_ops = (void *) &(dev->init);
4419 /* Init, if this function is available */
4420 if (dev->netdev_ops->ndo_init) {
4421 ret = dev->netdev_ops->ndo_init(dev);
4429 if (!dev_valid_name(dev->name)) {
4434 dev->ifindex = dev_new_index(net);
4435 if (dev->iflink == -1)
4436 dev->iflink = dev->ifindex;
4438 /* Check for existence of name */
4439 head = dev_name_hash(net, dev->name);
4440 hlist_for_each(p, head) {
4441 struct net_device *d
4442 = hlist_entry(p, struct net_device, name_hlist);
4443 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4449 /* Fix illegal checksum combinations */
4450 if ((dev->features & NETIF_F_HW_CSUM) &&
4451 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4452 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4454 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4457 if ((dev->features & NETIF_F_NO_CSUM) &&
4458 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4459 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4461 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4464 dev->features = netdev_fix_features(dev->features, dev->name);
4466 /* Enable software GSO if SG is supported. */
4467 if (dev->features & NETIF_F_SG)
4468 dev->features |= NETIF_F_GSO;
4470 netdev_initialize_kobject(dev);
4471 ret = netdev_register_kobject(dev);
4474 dev->reg_state = NETREG_REGISTERED;
4477 * Default initial state at registry is that the
4478 * device is present.
4481 set_bit(__LINK_STATE_PRESENT, &dev->state);
4483 dev_init_scheduler(dev);
4485 list_netdevice(dev);
4487 /* Notify protocols, that a new device appeared. */
4488 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4489 ret = notifier_to_errno(ret);
4491 rollback_registered(dev);
4492 dev->reg_state = NETREG_UNREGISTERED;
4499 if (dev->netdev_ops->ndo_uninit)
4500 dev->netdev_ops->ndo_uninit(dev);
4505 * init_dummy_netdev - init a dummy network device for NAPI
4506 * @dev: device to init
4508 * This takes a network device structure and initialize the minimum
4509 * amount of fields so it can be used to schedule NAPI polls without
4510 * registering a full blown interface. This is to be used by drivers
4511 * that need to tie several hardware interfaces to a single NAPI
4512 * poll scheduler due to HW limitations.
4514 int init_dummy_netdev(struct net_device *dev)
4516 /* Clear everything. Note we don't initialize spinlocks
4517 * are they aren't supposed to be taken by any of the
4518 * NAPI code and this dummy netdev is supposed to be
4519 * only ever used for NAPI polls
4521 memset(dev, 0, sizeof(struct net_device));
4523 /* make sure we BUG if trying to hit standard
4524 * register/unregister code path
4526 dev->reg_state = NETREG_DUMMY;
4528 /* initialize the ref count */
4529 atomic_set(&dev->refcnt, 1);
4531 /* NAPI wants this */
4532 INIT_LIST_HEAD(&dev->napi_list);
4534 /* a dummy interface is started by default */
4535 set_bit(__LINK_STATE_PRESENT, &dev->state);
4536 set_bit(__LINK_STATE_START, &dev->state);
4540 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4544 * register_netdev - register a network device
4545 * @dev: device to register
4547 * Take a completed network device structure and add it to the kernel
4548 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4549 * chain. 0 is returned on success. A negative errno code is returned
4550 * on a failure to set up the device, or if the name is a duplicate.
4552 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4553 * and expands the device name if you passed a format string to
4556 int register_netdev(struct net_device *dev)
4563 * If the name is a format string the caller wants us to do a
4566 if (strchr(dev->name, '%')) {
4567 err = dev_alloc_name(dev, dev->name);
4572 err = register_netdevice(dev);
4577 EXPORT_SYMBOL(register_netdev);
4580 * netdev_wait_allrefs - wait until all references are gone.
4582 * This is called when unregistering network devices.
4584 * Any protocol or device that holds a reference should register
4585 * for netdevice notification, and cleanup and put back the
4586 * reference if they receive an UNREGISTER event.
4587 * We can get stuck here if buggy protocols don't correctly
4590 static void netdev_wait_allrefs(struct net_device *dev)
4592 unsigned long rebroadcast_time, warning_time;
4594 rebroadcast_time = warning_time = jiffies;
4595 while (atomic_read(&dev->refcnt) != 0) {
4596 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4599 /* Rebroadcast unregister notification */
4600 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4602 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4604 /* We must not have linkwatch events
4605 * pending on unregister. If this
4606 * happens, we simply run the queue
4607 * unscheduled, resulting in a noop
4610 linkwatch_run_queue();
4615 rebroadcast_time = jiffies;
4620 if (time_after(jiffies, warning_time + 10 * HZ)) {
4621 printk(KERN_EMERG "unregister_netdevice: "
4622 "waiting for %s to become free. Usage "
4624 dev->name, atomic_read(&dev->refcnt));
4625 warning_time = jiffies;
4634 * register_netdevice(x1);
4635 * register_netdevice(x2);
4637 * unregister_netdevice(y1);
4638 * unregister_netdevice(y2);
4644 * We are invoked by rtnl_unlock().
4645 * This allows us to deal with problems:
4646 * 1) We can delete sysfs objects which invoke hotplug
4647 * without deadlocking with linkwatch via keventd.
4648 * 2) Since we run with the RTNL semaphore not held, we can sleep
4649 * safely in order to wait for the netdev refcnt to drop to zero.
4651 * We must not return until all unregister events added during
4652 * the interval the lock was held have been completed.
4654 void netdev_run_todo(void)
4656 struct list_head list;
4658 /* Snapshot list, allow later requests */
4659 list_replace_init(&net_todo_list, &list);
4663 while (!list_empty(&list)) {
4664 struct net_device *dev
4665 = list_entry(list.next, struct net_device, todo_list);
4666 list_del(&dev->todo_list);
4668 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
4669 printk(KERN_ERR "network todo '%s' but state %d\n",
4670 dev->name, dev->reg_state);
4675 dev->reg_state = NETREG_UNREGISTERED;
4677 on_each_cpu(flush_backlog, dev, 1);
4679 netdev_wait_allrefs(dev);
4682 BUG_ON(atomic_read(&dev->refcnt));
4683 WARN_ON(dev->ip_ptr);
4684 WARN_ON(dev->ip6_ptr);
4685 WARN_ON(dev->dn_ptr);
4687 if (dev->destructor)
4688 dev->destructor(dev);
4690 /* Free network device */
4691 kobject_put(&dev->dev.kobj);
4696 * dev_get_stats - get network device statistics
4697 * @dev: device to get statistics from
4699 * Get network statistics from device. The device driver may provide
4700 * its own method by setting dev->netdev_ops->get_stats; otherwise
4701 * the internal statistics structure is used.
4703 const struct net_device_stats *dev_get_stats(struct net_device *dev)
4705 const struct net_device_ops *ops = dev->netdev_ops;
4707 if (ops->ndo_get_stats)
4708 return ops->ndo_get_stats(dev);
4712 EXPORT_SYMBOL(dev_get_stats);
4714 static void netdev_init_one_queue(struct net_device *dev,
4715 struct netdev_queue *queue,
4721 static void netdev_init_queues(struct net_device *dev)
4723 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
4724 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
4725 spin_lock_init(&dev->tx_global_lock);
4729 * alloc_netdev_mq - allocate network device
4730 * @sizeof_priv: size of private data to allocate space for
4731 * @name: device name format string
4732 * @setup: callback to initialize device
4733 * @queue_count: the number of subqueues to allocate
4735 * Allocates a struct net_device with private data area for driver use
4736 * and performs basic initialization. Also allocates subquue structs
4737 * for each queue on the device at the end of the netdevice.
4739 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
4740 void (*setup)(struct net_device *), unsigned int queue_count)
4742 struct netdev_queue *tx;
4743 struct net_device *dev;
4747 BUG_ON(strlen(name) >= sizeof(dev->name));
4749 alloc_size = sizeof(struct net_device);
4751 /* ensure 32-byte alignment of private area */
4752 alloc_size = (alloc_size + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST;
4753 alloc_size += sizeof_priv;
4755 /* ensure 32-byte alignment of whole construct */
4756 alloc_size += NETDEV_ALIGN_CONST;
4758 p = kzalloc(alloc_size, GFP_KERNEL);
4760 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
4764 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
4766 printk(KERN_ERR "alloc_netdev: Unable to allocate "
4772 dev = (struct net_device *)
4773 (((long)p + NETDEV_ALIGN_CONST) & ~NETDEV_ALIGN_CONST);
4774 dev->padded = (char *)dev - (char *)p;
4775 dev_net_set(dev, &init_net);
4778 dev->num_tx_queues = queue_count;
4779 dev->real_num_tx_queues = queue_count;
4781 dev->gso_max_size = GSO_MAX_SIZE;
4783 netdev_init_queues(dev);
4785 INIT_LIST_HEAD(&dev->napi_list);
4787 strcpy(dev->name, name);
4790 EXPORT_SYMBOL(alloc_netdev_mq);
4793 * free_netdev - free network device
4796 * This function does the last stage of destroying an allocated device
4797 * interface. The reference to the device object is released.
4798 * If this is the last reference then it will be freed.
4800 void free_netdev(struct net_device *dev)
4802 struct napi_struct *p, *n;
4804 release_net(dev_net(dev));
4808 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
4811 /* Compatibility with error handling in drivers */
4812 if (dev->reg_state == NETREG_UNINITIALIZED) {
4813 kfree((char *)dev - dev->padded);
4817 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
4818 dev->reg_state = NETREG_RELEASED;
4820 /* will free via device release */
4821 put_device(&dev->dev);
4825 * synchronize_net - Synchronize with packet receive processing
4827 * Wait for packets currently being received to be done.
4828 * Does not block later packets from starting.
4830 void synchronize_net(void)
4837 * unregister_netdevice - remove device from the kernel
4840 * This function shuts down a device interface and removes it
4841 * from the kernel tables.
4843 * Callers must hold the rtnl semaphore. You may want
4844 * unregister_netdev() instead of this.
4847 void unregister_netdevice(struct net_device *dev)
4851 rollback_registered(dev);
4852 /* Finish processing unregister after unlock */
4857 * unregister_netdev - remove device from the kernel
4860 * This function shuts down a device interface and removes it
4861 * from the kernel tables.
4863 * This is just a wrapper for unregister_netdevice that takes
4864 * the rtnl semaphore. In general you want to use this and not
4865 * unregister_netdevice.
4867 void unregister_netdev(struct net_device *dev)
4870 unregister_netdevice(dev);
4874 EXPORT_SYMBOL(unregister_netdev);
4877 * dev_change_net_namespace - move device to different nethost namespace
4879 * @net: network namespace
4880 * @pat: If not NULL name pattern to try if the current device name
4881 * is already taken in the destination network namespace.
4883 * This function shuts down a device interface and moves it
4884 * to a new network namespace. On success 0 is returned, on
4885 * a failure a netagive errno code is returned.
4887 * Callers must hold the rtnl semaphore.
4890 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
4893 const char *destname;
4898 /* Don't allow namespace local devices to be moved. */
4900 if (dev->features & NETIF_F_NETNS_LOCAL)
4904 /* Don't allow real devices to be moved when sysfs
4908 if (dev->dev.parent)
4912 /* Ensure the device has been registrered */
4914 if (dev->reg_state != NETREG_REGISTERED)
4917 /* Get out if there is nothing todo */
4919 if (net_eq(dev_net(dev), net))
4922 /* Pick the destination device name, and ensure
4923 * we can use it in the destination network namespace.
4926 destname = dev->name;
4927 if (__dev_get_by_name(net, destname)) {
4928 /* We get here if we can't use the current device name */
4931 if (!dev_valid_name(pat))
4933 if (strchr(pat, '%')) {
4934 if (__dev_alloc_name(net, pat, buf) < 0)
4939 if (__dev_get_by_name(net, destname))
4944 * And now a mini version of register_netdevice unregister_netdevice.
4947 /* If device is running close it first. */
4950 /* And unlink it from device chain */
4952 unlist_netdevice(dev);
4956 /* Shutdown queueing discipline. */
4959 /* Notify protocols, that we are about to destroy
4960 this device. They should clean all the things.
4962 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4965 * Flush the unicast and multicast chains
4967 dev_addr_discard(dev);
4969 netdev_unregister_kobject(dev);
4971 /* Actually switch the network namespace */
4972 dev_net_set(dev, net);
4974 /* Assign the new device name */
4975 if (destname != dev->name)
4976 strcpy(dev->name, destname);
4978 /* If there is an ifindex conflict assign a new one */
4979 if (__dev_get_by_index(net, dev->ifindex)) {
4980 int iflink = (dev->iflink == dev->ifindex);
4981 dev->ifindex = dev_new_index(net);
4983 dev->iflink = dev->ifindex;
4986 /* Fixup kobjects */
4987 err = netdev_register_kobject(dev);
4990 /* Add the device back in the hashes */
4991 list_netdevice(dev);
4993 /* Notify protocols, that a new device appeared. */
4994 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5002 static int dev_cpu_callback(struct notifier_block *nfb,
5003 unsigned long action,
5006 struct sk_buff **list_skb;
5007 struct Qdisc **list_net;
5008 struct sk_buff *skb;
5009 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5010 struct softnet_data *sd, *oldsd;
5012 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5015 local_irq_disable();
5016 cpu = smp_processor_id();
5017 sd = &per_cpu(softnet_data, cpu);
5018 oldsd = &per_cpu(softnet_data, oldcpu);
5020 /* Find end of our completion_queue. */
5021 list_skb = &sd->completion_queue;
5023 list_skb = &(*list_skb)->next;
5024 /* Append completion queue from offline CPU. */
5025 *list_skb = oldsd->completion_queue;
5026 oldsd->completion_queue = NULL;
5028 /* Find end of our output_queue. */
5029 list_net = &sd->output_queue;
5031 list_net = &(*list_net)->next_sched;
5032 /* Append output queue from offline CPU. */
5033 *list_net = oldsd->output_queue;
5034 oldsd->output_queue = NULL;
5036 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5039 /* Process offline CPU's input_pkt_queue */
5040 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5048 * netdev_increment_features - increment feature set by one
5049 * @all: current feature set
5050 * @one: new feature set
5051 * @mask: mask feature set
5053 * Computes a new feature set after adding a device with feature set
5054 * @one to the master device with current feature set @all. Will not
5055 * enable anything that is off in @mask. Returns the new feature set.
5057 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5060 /* If device needs checksumming, downgrade to it. */
5061 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5062 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5063 else if (mask & NETIF_F_ALL_CSUM) {
5064 /* If one device supports v4/v6 checksumming, set for all. */
5065 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5066 !(all & NETIF_F_GEN_CSUM)) {
5067 all &= ~NETIF_F_ALL_CSUM;
5068 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5071 /* If one device supports hw checksumming, set for all. */
5072 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5073 all &= ~NETIF_F_ALL_CSUM;
5074 all |= NETIF_F_HW_CSUM;
5078 one |= NETIF_F_ALL_CSUM;
5080 one |= all & NETIF_F_ONE_FOR_ALL;
5081 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
5082 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5086 EXPORT_SYMBOL(netdev_increment_features);
5088 static struct hlist_head *netdev_create_hash(void)
5091 struct hlist_head *hash;
5093 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5095 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5096 INIT_HLIST_HEAD(&hash[i]);
5101 /* Initialize per network namespace state */
5102 static int __net_init netdev_init(struct net *net)
5104 INIT_LIST_HEAD(&net->dev_base_head);
5106 net->dev_name_head = netdev_create_hash();
5107 if (net->dev_name_head == NULL)
5110 net->dev_index_head = netdev_create_hash();
5111 if (net->dev_index_head == NULL)
5117 kfree(net->dev_name_head);
5123 * netdev_drivername - network driver for the device
5124 * @dev: network device
5125 * @buffer: buffer for resulting name
5126 * @len: size of buffer
5128 * Determine network driver for device.
5130 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5132 const struct device_driver *driver;
5133 const struct device *parent;
5135 if (len <= 0 || !buffer)
5139 parent = dev->dev.parent;
5144 driver = parent->driver;
5145 if (driver && driver->name)
5146 strlcpy(buffer, driver->name, len);
5150 static void __net_exit netdev_exit(struct net *net)
5152 kfree(net->dev_name_head);
5153 kfree(net->dev_index_head);
5156 static struct pernet_operations __net_initdata netdev_net_ops = {
5157 .init = netdev_init,
5158 .exit = netdev_exit,
5161 static void __net_exit default_device_exit(struct net *net)
5163 struct net_device *dev;
5165 * Push all migratable of the network devices back to the
5166 * initial network namespace
5170 for_each_netdev(net, dev) {
5172 char fb_name[IFNAMSIZ];
5174 /* Ignore unmoveable devices (i.e. loopback) */
5175 if (dev->features & NETIF_F_NETNS_LOCAL)
5178 /* Delete virtual devices */
5179 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5180 dev->rtnl_link_ops->dellink(dev);
5184 /* Push remaing network devices to init_net */
5185 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5186 err = dev_change_net_namespace(dev, &init_net, fb_name);
5188 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5189 __func__, dev->name, err);
5197 static struct pernet_operations __net_initdata default_device_ops = {
5198 .exit = default_device_exit,
5202 * Initialize the DEV module. At boot time this walks the device list and
5203 * unhooks any devices that fail to initialise (normally hardware not
5204 * present) and leaves us with a valid list of present and active devices.
5209 * This is called single threaded during boot, so no need
5210 * to take the rtnl semaphore.
5212 static int __init net_dev_init(void)
5214 int i, rc = -ENOMEM;
5216 BUG_ON(!dev_boot_phase);
5218 if (dev_proc_init())
5221 if (netdev_kobject_init())
5224 INIT_LIST_HEAD(&ptype_all);
5225 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5226 INIT_LIST_HEAD(&ptype_base[i]);
5228 if (register_pernet_subsys(&netdev_net_ops))
5232 * Initialise the packet receive queues.
5235 for_each_possible_cpu(i) {
5236 struct softnet_data *queue;
5238 queue = &per_cpu(softnet_data, i);
5239 skb_queue_head_init(&queue->input_pkt_queue);
5240 queue->completion_queue = NULL;
5241 INIT_LIST_HEAD(&queue->poll_list);
5243 queue->backlog.poll = process_backlog;
5244 queue->backlog.weight = weight_p;
5245 queue->backlog.gro_list = NULL;
5246 queue->backlog.gro_count = 0;
5251 /* The loopback device is special if any other network devices
5252 * is present in a network namespace the loopback device must
5253 * be present. Since we now dynamically allocate and free the
5254 * loopback device ensure this invariant is maintained by
5255 * keeping the loopback device as the first device on the
5256 * list of network devices. Ensuring the loopback devices
5257 * is the first device that appears and the last network device
5260 if (register_pernet_device(&loopback_net_ops))
5263 if (register_pernet_device(&default_device_ops))
5266 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5267 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5269 hotcpu_notifier(dev_cpu_callback, 0);
5277 subsys_initcall(net_dev_init);
5279 static int __init initialize_hashrnd(void)
5281 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5285 late_initcall_sync(initialize_hashrnd);
5287 EXPORT_SYMBOL(__dev_get_by_index);
5288 EXPORT_SYMBOL(__dev_get_by_name);
5289 EXPORT_SYMBOL(__dev_remove_pack);
5290 EXPORT_SYMBOL(dev_valid_name);
5291 EXPORT_SYMBOL(dev_add_pack);
5292 EXPORT_SYMBOL(dev_alloc_name);
5293 EXPORT_SYMBOL(dev_close);
5294 EXPORT_SYMBOL(dev_get_by_flags);
5295 EXPORT_SYMBOL(dev_get_by_index);
5296 EXPORT_SYMBOL(dev_get_by_name);
5297 EXPORT_SYMBOL(dev_open);
5298 EXPORT_SYMBOL(dev_queue_xmit);
5299 EXPORT_SYMBOL(dev_remove_pack);
5300 EXPORT_SYMBOL(dev_set_allmulti);
5301 EXPORT_SYMBOL(dev_set_promiscuity);
5302 EXPORT_SYMBOL(dev_change_flags);
5303 EXPORT_SYMBOL(dev_set_mtu);
5304 EXPORT_SYMBOL(dev_set_mac_address);
5305 EXPORT_SYMBOL(free_netdev);
5306 EXPORT_SYMBOL(netdev_boot_setup_check);
5307 EXPORT_SYMBOL(netdev_set_master);
5308 EXPORT_SYMBOL(netdev_state_change);
5309 EXPORT_SYMBOL(netif_receive_skb);
5310 EXPORT_SYMBOL(netif_rx);
5311 EXPORT_SYMBOL(register_gifconf);
5312 EXPORT_SYMBOL(register_netdevice);
5313 EXPORT_SYMBOL(register_netdevice_notifier);
5314 EXPORT_SYMBOL(skb_checksum_help);
5315 EXPORT_SYMBOL(synchronize_net);
5316 EXPORT_SYMBOL(unregister_netdevice);
5317 EXPORT_SYMBOL(unregister_netdevice_notifier);
5318 EXPORT_SYMBOL(net_enable_timestamp);
5319 EXPORT_SYMBOL(net_disable_timestamp);
5320 EXPORT_SYMBOL(dev_get_flags);
5322 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
5323 EXPORT_SYMBOL(br_handle_frame_hook);
5324 EXPORT_SYMBOL(br_fdb_get_hook);
5325 EXPORT_SYMBOL(br_fdb_put_hook);
5328 EXPORT_SYMBOL(dev_load);
5330 EXPORT_PER_CPU_SYMBOL(softnet_data);