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>
129 #include <trace/events/napi.h>
131 #include "net-sysfs.h"
133 /* Instead of increasing this, you should create a hash table. */
134 #define MAX_GRO_SKBS 8
136 /* This should be increased if a protocol with a bigger head is added. */
137 #define GRO_MAX_HEAD (MAX_HEADER + 128)
140 * The list of packet types we will receive (as opposed to discard)
141 * and the routines to invoke.
143 * Why 16. Because with 16 the only overlap we get on a hash of the
144 * low nibble of the protocol value is RARP/SNAP/X.25.
146 * NOTE: That is no longer true with the addition of VLAN tags. Not
147 * sure which should go first, but I bet it won't make much
148 * difference if we are running VLANs. The good news is that
149 * this protocol won't be in the list unless compiled in, so
150 * the average user (w/out VLANs) will not be adversely affected.
167 #define PTYPE_HASH_SIZE (16)
168 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
170 static DEFINE_SPINLOCK(ptype_lock);
171 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
172 static struct list_head ptype_all __read_mostly; /* Taps */
175 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
178 * Pure readers hold dev_base_lock for reading.
180 * Writers must hold the rtnl semaphore while they loop through the
181 * dev_base_head list, and hold dev_base_lock for writing when they do the
182 * actual updates. This allows pure readers to access the list even
183 * while a writer is preparing to update it.
185 * To put it another way, dev_base_lock is held for writing only to
186 * protect against pure readers; the rtnl semaphore provides the
187 * protection against other writers.
189 * See, for example usages, register_netdevice() and
190 * unregister_netdevice(), which must be called with the rtnl
193 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);
250 EXPORT_PER_CPU_SYMBOL(softnet_data);
252 #ifdef CONFIG_LOCKDEP
254 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
255 * according to dev->type
257 static const unsigned short netdev_lock_type[] =
258 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
259 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
260 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
261 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
262 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
263 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
264 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
265 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
266 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
267 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
268 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
269 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
270 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
271 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
272 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
273 ARPHRD_VOID, ARPHRD_NONE};
275 static const char *const netdev_lock_name[] =
276 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
277 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
278 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
279 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
280 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
281 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
282 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
283 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
284 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
285 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
286 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
287 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
288 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
289 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
290 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
291 "_xmit_VOID", "_xmit_NONE"};
293 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
294 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
296 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
300 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
301 if (netdev_lock_type[i] == dev_type)
303 /* the last key is used by default */
304 return ARRAY_SIZE(netdev_lock_type) - 1;
307 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
308 unsigned short dev_type)
312 i = netdev_lock_pos(dev_type);
313 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
314 netdev_lock_name[i]);
317 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
321 i = netdev_lock_pos(dev->type);
322 lockdep_set_class_and_name(&dev->addr_list_lock,
323 &netdev_addr_lock_key[i],
324 netdev_lock_name[i]);
327 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
328 unsigned short dev_type)
331 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
336 /*******************************************************************************
338 Protocol management and registration routines
340 *******************************************************************************/
343 * Add a protocol ID to the list. Now that the input handler is
344 * smarter we can dispense with all the messy stuff that used to be
347 * BEWARE!!! Protocol handlers, mangling input packets,
348 * MUST BE last in hash buckets and checking protocol handlers
349 * MUST start from promiscuous ptype_all chain in net_bh.
350 * It is true now, do not change it.
351 * Explanation follows: if protocol handler, mangling packet, will
352 * be the first on list, it is not able to sense, that packet
353 * is cloned and should be copied-on-write, so that it will
354 * change it and subsequent readers will get broken packet.
359 * dev_add_pack - add packet handler
360 * @pt: packet type declaration
362 * Add a protocol handler to the networking stack. The passed &packet_type
363 * is linked into kernel lists and may not be freed until it has been
364 * removed from the kernel lists.
366 * This call does not sleep therefore it can not
367 * guarantee all CPU's that are in middle of receiving packets
368 * will see the new packet type (until the next received packet).
371 void dev_add_pack(struct packet_type *pt)
375 spin_lock_bh(&ptype_lock);
376 if (pt->type == htons(ETH_P_ALL))
377 list_add_rcu(&pt->list, &ptype_all);
379 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
380 list_add_rcu(&pt->list, &ptype_base[hash]);
382 spin_unlock_bh(&ptype_lock);
384 EXPORT_SYMBOL(dev_add_pack);
387 * __dev_remove_pack - remove packet handler
388 * @pt: packet type declaration
390 * Remove a protocol handler that was previously added to the kernel
391 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
392 * from the kernel lists and can be freed or reused once this function
395 * The packet type might still be in use by receivers
396 * and must not be freed until after all the CPU's have gone
397 * through a quiescent state.
399 void __dev_remove_pack(struct packet_type *pt)
401 struct list_head *head;
402 struct packet_type *pt1;
404 spin_lock_bh(&ptype_lock);
406 if (pt->type == htons(ETH_P_ALL))
409 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
411 list_for_each_entry(pt1, head, list) {
413 list_del_rcu(&pt->list);
418 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
420 spin_unlock_bh(&ptype_lock);
422 EXPORT_SYMBOL(__dev_remove_pack);
425 * dev_remove_pack - remove packet handler
426 * @pt: packet type declaration
428 * Remove a protocol handler that was previously added to the kernel
429 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
430 * from the kernel lists and can be freed or reused once this function
433 * This call sleeps to guarantee that no CPU is looking at the packet
436 void dev_remove_pack(struct packet_type *pt)
438 __dev_remove_pack(pt);
442 EXPORT_SYMBOL(dev_remove_pack);
444 /******************************************************************************
446 Device Boot-time Settings Routines
448 *******************************************************************************/
450 /* Boot time configuration table */
451 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
454 * netdev_boot_setup_add - add new setup entry
455 * @name: name of the device
456 * @map: configured settings for the device
458 * Adds new setup entry to the dev_boot_setup list. The function
459 * returns 0 on error and 1 on success. This is a generic routine to
462 static int netdev_boot_setup_add(char *name, struct ifmap *map)
464 struct netdev_boot_setup *s;
468 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
469 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
470 memset(s[i].name, 0, sizeof(s[i].name));
471 strlcpy(s[i].name, name, IFNAMSIZ);
472 memcpy(&s[i].map, map, sizeof(s[i].map));
477 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
481 * netdev_boot_setup_check - check boot time settings
482 * @dev: the netdevice
484 * Check boot time settings for the device.
485 * The found settings are set for the device to be used
486 * later in the device probing.
487 * Returns 0 if no settings found, 1 if they are.
489 int netdev_boot_setup_check(struct net_device *dev)
491 struct netdev_boot_setup *s = dev_boot_setup;
494 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
495 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
496 !strcmp(dev->name, s[i].name)) {
497 dev->irq = s[i].map.irq;
498 dev->base_addr = s[i].map.base_addr;
499 dev->mem_start = s[i].map.mem_start;
500 dev->mem_end = s[i].map.mem_end;
506 EXPORT_SYMBOL(netdev_boot_setup_check);
510 * netdev_boot_base - get address from boot time settings
511 * @prefix: prefix for network device
512 * @unit: id for network device
514 * Check boot time settings for the base address of device.
515 * The found settings are set for the device to be used
516 * later in the device probing.
517 * Returns 0 if no settings found.
519 unsigned long netdev_boot_base(const char *prefix, int unit)
521 const struct netdev_boot_setup *s = dev_boot_setup;
525 sprintf(name, "%s%d", prefix, unit);
528 * If device already registered then return base of 1
529 * to indicate not to probe for this interface
531 if (__dev_get_by_name(&init_net, name))
534 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
535 if (!strcmp(name, s[i].name))
536 return s[i].map.base_addr;
541 * Saves at boot time configured settings for any netdevice.
543 int __init netdev_boot_setup(char *str)
548 str = get_options(str, ARRAY_SIZE(ints), ints);
553 memset(&map, 0, sizeof(map));
557 map.base_addr = ints[2];
559 map.mem_start = ints[3];
561 map.mem_end = ints[4];
563 /* Add new entry to the list */
564 return netdev_boot_setup_add(str, &map);
567 __setup("netdev=", netdev_boot_setup);
569 /*******************************************************************************
571 Device Interface Subroutines
573 *******************************************************************************/
576 * __dev_get_by_name - find a device by its name
577 * @net: the applicable net namespace
578 * @name: name to find
580 * Find an interface by name. Must be called under RTNL semaphore
581 * or @dev_base_lock. If the name is found a pointer to the device
582 * is returned. If the name is not found then %NULL is returned. The
583 * reference counters are not incremented so the caller must be
584 * careful with locks.
587 struct net_device *__dev_get_by_name(struct net *net, const char *name)
589 struct hlist_node *p;
591 hlist_for_each(p, dev_name_hash(net, name)) {
592 struct net_device *dev
593 = hlist_entry(p, struct net_device, name_hlist);
594 if (!strncmp(dev->name, name, IFNAMSIZ))
599 EXPORT_SYMBOL(__dev_get_by_name);
602 * dev_get_by_name - find a device by its name
603 * @net: the applicable net namespace
604 * @name: name to find
606 * Find an interface by name. This can be called from any
607 * context and does its own locking. The returned handle has
608 * the usage count incremented and the caller must use dev_put() to
609 * release it when it is no longer needed. %NULL is returned if no
610 * matching device is found.
613 struct net_device *dev_get_by_name(struct net *net, const char *name)
615 struct net_device *dev;
617 read_lock(&dev_base_lock);
618 dev = __dev_get_by_name(net, name);
621 read_unlock(&dev_base_lock);
624 EXPORT_SYMBOL(dev_get_by_name);
627 * __dev_get_by_index - find a device by its ifindex
628 * @net: the applicable net namespace
629 * @ifindex: index of device
631 * Search for an interface by index. Returns %NULL if the device
632 * is not found or a pointer to the device. The device has not
633 * had its reference counter increased so the caller must be careful
634 * about locking. The caller must hold either the RTNL semaphore
638 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
640 struct hlist_node *p;
642 hlist_for_each(p, dev_index_hash(net, ifindex)) {
643 struct net_device *dev
644 = hlist_entry(p, struct net_device, index_hlist);
645 if (dev->ifindex == ifindex)
650 EXPORT_SYMBOL(__dev_get_by_index);
654 * dev_get_by_index - find a device by its ifindex
655 * @net: the applicable net namespace
656 * @ifindex: index of device
658 * Search for an interface by index. Returns NULL if the device
659 * is not found or a pointer to the device. The device returned has
660 * had a reference added and the pointer is safe until the user calls
661 * dev_put to indicate they have finished with it.
664 struct net_device *dev_get_by_index(struct net *net, int ifindex)
666 struct net_device *dev;
668 read_lock(&dev_base_lock);
669 dev = __dev_get_by_index(net, ifindex);
672 read_unlock(&dev_base_lock);
675 EXPORT_SYMBOL(dev_get_by_index);
678 * dev_getbyhwaddr - find a device by its hardware address
679 * @net: the applicable net namespace
680 * @type: media type of device
681 * @ha: hardware address
683 * Search for an interface by MAC address. Returns NULL if the device
684 * is not found or a pointer to the device. The caller must hold the
685 * rtnl semaphore. The returned device has not had its ref count increased
686 * and the caller must therefore be careful about locking
689 * If the API was consistent this would be __dev_get_by_hwaddr
692 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
694 struct net_device *dev;
698 for_each_netdev(net, dev)
699 if (dev->type == type &&
700 !memcmp(dev->dev_addr, ha, dev->addr_len))
705 EXPORT_SYMBOL(dev_getbyhwaddr);
707 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
709 struct net_device *dev;
712 for_each_netdev(net, dev)
713 if (dev->type == type)
718 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
720 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
722 struct net_device *dev;
725 dev = __dev_getfirstbyhwtype(net, type);
731 EXPORT_SYMBOL(dev_getfirstbyhwtype);
734 * dev_get_by_flags - find any device with given flags
735 * @net: the applicable net namespace
736 * @if_flags: IFF_* values
737 * @mask: bitmask of bits in if_flags to check
739 * Search for any interface with the given flags. Returns NULL if a device
740 * is not found or a pointer to the device. The device returned has
741 * had a reference added and the pointer is safe until the user calls
742 * dev_put to indicate they have finished with it.
745 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
748 struct net_device *dev, *ret;
751 read_lock(&dev_base_lock);
752 for_each_netdev(net, dev) {
753 if (((dev->flags ^ if_flags) & mask) == 0) {
759 read_unlock(&dev_base_lock);
762 EXPORT_SYMBOL(dev_get_by_flags);
765 * dev_valid_name - check if name is okay for network device
768 * Network device names need to be valid file names to
769 * to allow sysfs to work. We also disallow any kind of
772 int dev_valid_name(const char *name)
776 if (strlen(name) >= IFNAMSIZ)
778 if (!strcmp(name, ".") || !strcmp(name, ".."))
782 if (*name == '/' || isspace(*name))
788 EXPORT_SYMBOL(dev_valid_name);
791 * __dev_alloc_name - allocate a name for a device
792 * @net: network namespace to allocate the device name in
793 * @name: name format string
794 * @buf: scratch buffer and result name string
796 * Passed a format string - eg "lt%d" it will try and find a suitable
797 * id. It scans list of devices to build up a free map, then chooses
798 * the first empty slot. The caller must hold the dev_base or rtnl lock
799 * while allocating the name and adding the device in order to avoid
801 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
802 * Returns the number of the unit assigned or a negative errno code.
805 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
809 const int max_netdevices = 8*PAGE_SIZE;
810 unsigned long *inuse;
811 struct net_device *d;
813 p = strnchr(name, IFNAMSIZ-1, '%');
816 * Verify the string as this thing may have come from
817 * the user. There must be either one "%d" and no other "%"
820 if (p[1] != 'd' || strchr(p + 2, '%'))
823 /* Use one page as a bit array of possible slots */
824 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
828 for_each_netdev(net, d) {
829 if (!sscanf(d->name, name, &i))
831 if (i < 0 || i >= max_netdevices)
834 /* avoid cases where sscanf is not exact inverse of printf */
835 snprintf(buf, IFNAMSIZ, name, i);
836 if (!strncmp(buf, d->name, IFNAMSIZ))
840 i = find_first_zero_bit(inuse, max_netdevices);
841 free_page((unsigned long) inuse);
844 snprintf(buf, IFNAMSIZ, name, i);
845 if (!__dev_get_by_name(net, buf))
848 /* It is possible to run out of possible slots
849 * when the name is long and there isn't enough space left
850 * for the digits, or if all bits are used.
856 * dev_alloc_name - allocate a name for a device
858 * @name: name format string
860 * Passed a format string - eg "lt%d" it will try and find a suitable
861 * id. It scans list of devices to build up a free map, then chooses
862 * the first empty slot. The caller must hold the dev_base or rtnl lock
863 * while allocating the name and adding the device in order to avoid
865 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
866 * Returns the number of the unit assigned or a negative errno code.
869 int dev_alloc_name(struct net_device *dev, const char *name)
875 BUG_ON(!dev_net(dev));
877 ret = __dev_alloc_name(net, name, buf);
879 strlcpy(dev->name, buf, IFNAMSIZ);
882 EXPORT_SYMBOL(dev_alloc_name);
886 * dev_change_name - change name of a device
888 * @newname: name (or format string) must be at least IFNAMSIZ
890 * Change name of a device, can pass format strings "eth%d".
893 int dev_change_name(struct net_device *dev, const char *newname)
895 char oldname[IFNAMSIZ];
901 BUG_ON(!dev_net(dev));
904 if (dev->flags & IFF_UP)
907 if (!dev_valid_name(newname))
910 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
913 memcpy(oldname, dev->name, IFNAMSIZ);
915 if (strchr(newname, '%')) {
916 err = dev_alloc_name(dev, newname);
919 } else if (__dev_get_by_name(net, newname))
922 strlcpy(dev->name, newname, IFNAMSIZ);
925 /* For now only devices in the initial network namespace
928 if (net == &init_net) {
929 ret = device_rename(&dev->dev, dev->name);
931 memcpy(dev->name, oldname, IFNAMSIZ);
936 write_lock_bh(&dev_base_lock);
937 hlist_del(&dev->name_hlist);
938 hlist_add_head(&dev->name_hlist, dev_name_hash(net, dev->name));
939 write_unlock_bh(&dev_base_lock);
941 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
942 ret = notifier_to_errno(ret);
945 /* err >= 0 after dev_alloc_name() or stores the first errno */
948 memcpy(dev->name, oldname, IFNAMSIZ);
952 "%s: name change rollback failed: %d.\n",
961 * dev_set_alias - change ifalias of a device
963 * @alias: name up to IFALIASZ
964 * @len: limit of bytes to copy from info
966 * Set ifalias for a device,
968 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
983 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
987 strlcpy(dev->ifalias, alias, len+1);
993 * netdev_features_change - device changes features
994 * @dev: device to cause notification
996 * Called to indicate a device has changed features.
998 void netdev_features_change(struct net_device *dev)
1000 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1002 EXPORT_SYMBOL(netdev_features_change);
1005 * netdev_state_change - device changes state
1006 * @dev: device to cause notification
1008 * Called to indicate a device has changed state. This function calls
1009 * the notifier chains for netdev_chain and sends a NEWLINK message
1010 * to the routing socket.
1012 void netdev_state_change(struct net_device *dev)
1014 if (dev->flags & IFF_UP) {
1015 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1016 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1019 EXPORT_SYMBOL(netdev_state_change);
1021 void netdev_bonding_change(struct net_device *dev, unsigned long event)
1023 call_netdevice_notifiers(event, dev);
1025 EXPORT_SYMBOL(netdev_bonding_change);
1028 * dev_load - load a network module
1029 * @net: the applicable net namespace
1030 * @name: name of interface
1032 * If a network interface is not present and the process has suitable
1033 * privileges this function loads the module. If module loading is not
1034 * available in this kernel then it becomes a nop.
1037 void dev_load(struct net *net, const char *name)
1039 struct net_device *dev;
1042 read_lock(&dev_base_lock);
1043 dev = __dev_get_by_name(net, name);
1044 read_unlock(&dev_base_lock);
1047 if (no_module && capable(CAP_NET_ADMIN))
1048 no_module = request_module("netdev-%s", name);
1049 if (no_module && capable(CAP_SYS_MODULE)) {
1050 if (!request_module("%s", name))
1051 pr_err("Loading kernel module for a network device "
1052 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1056 EXPORT_SYMBOL(dev_load);
1059 * dev_open - prepare an interface for use.
1060 * @dev: device to open
1062 * Takes a device from down to up state. The device's private open
1063 * function is invoked and then the multicast lists are loaded. Finally
1064 * the device is moved into the up state and a %NETDEV_UP message is
1065 * sent to the netdev notifier chain.
1067 * Calling this function on an active interface is a nop. On a failure
1068 * a negative errno code is returned.
1070 int dev_open(struct net_device *dev)
1072 const struct net_device_ops *ops = dev->netdev_ops;
1081 if (dev->flags & IFF_UP)
1085 * Is it even present?
1087 if (!netif_device_present(dev))
1090 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1091 ret = notifier_to_errno(ret);
1096 * Call device private open method
1098 set_bit(__LINK_STATE_START, &dev->state);
1100 if (ops->ndo_validate_addr)
1101 ret = ops->ndo_validate_addr(dev);
1103 if (!ret && ops->ndo_open)
1104 ret = ops->ndo_open(dev);
1107 * If it went open OK then:
1111 clear_bit(__LINK_STATE_START, &dev->state);
1116 dev->flags |= IFF_UP;
1121 net_dmaengine_get();
1124 * Initialize multicasting status
1126 dev_set_rx_mode(dev);
1129 * Wakeup transmit queue engine
1134 * ... and announce new interface.
1136 call_netdevice_notifiers(NETDEV_UP, dev);
1141 EXPORT_SYMBOL(dev_open);
1144 * dev_close - shutdown an interface.
1145 * @dev: device to shutdown
1147 * This function moves an active device into down state. A
1148 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1149 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1152 int dev_close(struct net_device *dev)
1154 const struct net_device_ops *ops = dev->netdev_ops;
1159 if (!(dev->flags & IFF_UP))
1163 * Tell people we are going down, so that they can
1164 * prepare to death, when device is still operating.
1166 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1168 clear_bit(__LINK_STATE_START, &dev->state);
1170 /* Synchronize to scheduled poll. We cannot touch poll list,
1171 * it can be even on different cpu. So just clear netif_running().
1173 * dev->stop() will invoke napi_disable() on all of it's
1174 * napi_struct instances on this device.
1176 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1178 dev_deactivate(dev);
1181 * Call the device specific close. This cannot fail.
1182 * Only if device is UP
1184 * We allow it to be called even after a DETACH hot-plug
1191 * Device is now down.
1194 dev->flags &= ~IFF_UP;
1197 * Tell people we are down
1199 call_netdevice_notifiers(NETDEV_DOWN, dev);
1204 net_dmaengine_put();
1208 EXPORT_SYMBOL(dev_close);
1212 * dev_disable_lro - disable Large Receive Offload on a device
1215 * Disable Large Receive Offload (LRO) on a net device. Must be
1216 * called under RTNL. This is needed if received packets may be
1217 * forwarded to another interface.
1219 void dev_disable_lro(struct net_device *dev)
1221 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1222 dev->ethtool_ops->set_flags) {
1223 u32 flags = dev->ethtool_ops->get_flags(dev);
1224 if (flags & ETH_FLAG_LRO) {
1225 flags &= ~ETH_FLAG_LRO;
1226 dev->ethtool_ops->set_flags(dev, flags);
1229 WARN_ON(dev->features & NETIF_F_LRO);
1231 EXPORT_SYMBOL(dev_disable_lro);
1234 static int dev_boot_phase = 1;
1237 * Device change register/unregister. These are not inline or static
1238 * as we export them to the world.
1242 * register_netdevice_notifier - register a network notifier block
1245 * Register a notifier to be called when network device events occur.
1246 * The notifier passed is linked into the kernel structures and must
1247 * not be reused until it has been unregistered. A negative errno code
1248 * is returned on a failure.
1250 * When registered all registration and up events are replayed
1251 * to the new notifier to allow device to have a race free
1252 * view of the network device list.
1255 int register_netdevice_notifier(struct notifier_block *nb)
1257 struct net_device *dev;
1258 struct net_device *last;
1263 err = raw_notifier_chain_register(&netdev_chain, nb);
1269 for_each_netdev(net, dev) {
1270 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1271 err = notifier_to_errno(err);
1275 if (!(dev->flags & IFF_UP))
1278 nb->notifier_call(nb, NETDEV_UP, dev);
1289 for_each_netdev(net, dev) {
1293 if (dev->flags & IFF_UP) {
1294 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1295 nb->notifier_call(nb, NETDEV_DOWN, dev);
1297 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1301 raw_notifier_chain_unregister(&netdev_chain, nb);
1304 EXPORT_SYMBOL(register_netdevice_notifier);
1307 * unregister_netdevice_notifier - unregister a network notifier block
1310 * Unregister a notifier previously registered by
1311 * register_netdevice_notifier(). The notifier is unlinked into the
1312 * kernel structures and may then be reused. A negative errno code
1313 * is returned on a failure.
1316 int unregister_netdevice_notifier(struct notifier_block *nb)
1321 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1325 EXPORT_SYMBOL(unregister_netdevice_notifier);
1328 * call_netdevice_notifiers - call all network notifier blocks
1329 * @val: value passed unmodified to notifier function
1330 * @dev: net_device pointer passed unmodified to notifier function
1332 * Call all network notifier blocks. Parameters and return value
1333 * are as for raw_notifier_call_chain().
1336 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1338 return raw_notifier_call_chain(&netdev_chain, val, dev);
1341 /* When > 0 there are consumers of rx skb time stamps */
1342 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1344 void net_enable_timestamp(void)
1346 atomic_inc(&netstamp_needed);
1348 EXPORT_SYMBOL(net_enable_timestamp);
1350 void net_disable_timestamp(void)
1352 atomic_dec(&netstamp_needed);
1354 EXPORT_SYMBOL(net_disable_timestamp);
1356 static inline void net_timestamp(struct sk_buff *skb)
1358 if (atomic_read(&netstamp_needed))
1359 __net_timestamp(skb);
1361 skb->tstamp.tv64 = 0;
1365 * Support routine. Sends outgoing frames to any network
1366 * taps currently in use.
1369 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1371 struct packet_type *ptype;
1373 #ifdef CONFIG_NET_CLS_ACT
1374 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1381 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1382 /* Never send packets back to the socket
1383 * they originated from - MvS (miquels@drinkel.ow.org)
1385 if ((ptype->dev == dev || !ptype->dev) &&
1386 (ptype->af_packet_priv == NULL ||
1387 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1388 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1392 /* skb->nh should be correctly
1393 set by sender, so that the second statement is
1394 just protection against buggy protocols.
1396 skb_reset_mac_header(skb2);
1398 if (skb_network_header(skb2) < skb2->data ||
1399 skb2->network_header > skb2->tail) {
1400 if (net_ratelimit())
1401 printk(KERN_CRIT "protocol %04x is "
1403 skb2->protocol, dev->name);
1404 skb_reset_network_header(skb2);
1407 skb2->transport_header = skb2->network_header;
1408 skb2->pkt_type = PACKET_OUTGOING;
1409 ptype->func(skb2, skb->dev, ptype, skb->dev);
1416 static inline void __netif_reschedule(struct Qdisc *q)
1418 struct softnet_data *sd;
1419 unsigned long flags;
1421 local_irq_save(flags);
1422 sd = &__get_cpu_var(softnet_data);
1423 q->next_sched = sd->output_queue;
1424 sd->output_queue = q;
1425 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1426 local_irq_restore(flags);
1429 void __netif_schedule(struct Qdisc *q)
1431 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1432 __netif_reschedule(q);
1434 EXPORT_SYMBOL(__netif_schedule);
1436 void dev_kfree_skb_irq(struct sk_buff *skb)
1438 if (atomic_dec_and_test(&skb->users)) {
1439 struct softnet_data *sd;
1440 unsigned long flags;
1442 local_irq_save(flags);
1443 sd = &__get_cpu_var(softnet_data);
1444 skb->next = sd->completion_queue;
1445 sd->completion_queue = skb;
1446 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1447 local_irq_restore(flags);
1450 EXPORT_SYMBOL(dev_kfree_skb_irq);
1452 void dev_kfree_skb_any(struct sk_buff *skb)
1454 if (in_irq() || irqs_disabled())
1455 dev_kfree_skb_irq(skb);
1459 EXPORT_SYMBOL(dev_kfree_skb_any);
1463 * netif_device_detach - mark device as removed
1464 * @dev: network device
1466 * Mark device as removed from system and therefore no longer available.
1468 void netif_device_detach(struct net_device *dev)
1470 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1471 netif_running(dev)) {
1472 netif_tx_stop_all_queues(dev);
1475 EXPORT_SYMBOL(netif_device_detach);
1478 * netif_device_attach - mark device as attached
1479 * @dev: network device
1481 * Mark device as attached from system and restart if needed.
1483 void netif_device_attach(struct net_device *dev)
1485 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1486 netif_running(dev)) {
1487 netif_tx_wake_all_queues(dev);
1488 __netdev_watchdog_up(dev);
1491 EXPORT_SYMBOL(netif_device_attach);
1493 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1495 return ((features & NETIF_F_NO_CSUM) ||
1496 ((features & NETIF_F_V4_CSUM) &&
1497 protocol == htons(ETH_P_IP)) ||
1498 ((features & NETIF_F_V6_CSUM) &&
1499 protocol == htons(ETH_P_IPV6)) ||
1500 ((features & NETIF_F_FCOE_CRC) &&
1501 protocol == htons(ETH_P_FCOE)));
1504 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1506 if (can_checksum_protocol(dev->features, skb->protocol))
1509 if (skb->protocol == htons(ETH_P_8021Q)) {
1510 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1511 if (can_checksum_protocol(dev->features & dev->vlan_features,
1512 veh->h_vlan_encapsulated_proto))
1520 * Invalidate hardware checksum when packet is to be mangled, and
1521 * complete checksum manually on outgoing path.
1523 int skb_checksum_help(struct sk_buff *skb)
1526 int ret = 0, offset;
1528 if (skb->ip_summed == CHECKSUM_COMPLETE)
1529 goto out_set_summed;
1531 if (unlikely(skb_shinfo(skb)->gso_size)) {
1532 /* Let GSO fix up the checksum. */
1533 goto out_set_summed;
1536 offset = skb->csum_start - skb_headroom(skb);
1537 BUG_ON(offset >= skb_headlen(skb));
1538 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1540 offset += skb->csum_offset;
1541 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1543 if (skb_cloned(skb) &&
1544 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1545 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1550 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1552 skb->ip_summed = CHECKSUM_NONE;
1556 EXPORT_SYMBOL(skb_checksum_help);
1559 * skb_gso_segment - Perform segmentation on skb.
1560 * @skb: buffer to segment
1561 * @features: features for the output path (see dev->features)
1563 * This function segments the given skb and returns a list of segments.
1565 * It may return NULL if the skb requires no segmentation. This is
1566 * only possible when GSO is used for verifying header integrity.
1568 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1570 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1571 struct packet_type *ptype;
1572 __be16 type = skb->protocol;
1575 skb_reset_mac_header(skb);
1576 skb->mac_len = skb->network_header - skb->mac_header;
1577 __skb_pull(skb, skb->mac_len);
1579 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1580 struct net_device *dev = skb->dev;
1581 struct ethtool_drvinfo info = {};
1583 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1584 dev->ethtool_ops->get_drvinfo(dev, &info);
1586 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1588 info.driver, dev ? dev->features : 0L,
1589 skb->sk ? skb->sk->sk_route_caps : 0L,
1590 skb->len, skb->data_len, skb->ip_summed);
1592 if (skb_header_cloned(skb) &&
1593 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1594 return ERR_PTR(err);
1598 list_for_each_entry_rcu(ptype,
1599 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1600 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1601 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1602 err = ptype->gso_send_check(skb);
1603 segs = ERR_PTR(err);
1604 if (err || skb_gso_ok(skb, features))
1606 __skb_push(skb, (skb->data -
1607 skb_network_header(skb)));
1609 segs = ptype->gso_segment(skb, features);
1615 __skb_push(skb, skb->data - skb_mac_header(skb));
1619 EXPORT_SYMBOL(skb_gso_segment);
1621 /* Take action when hardware reception checksum errors are detected. */
1623 void netdev_rx_csum_fault(struct net_device *dev)
1625 if (net_ratelimit()) {
1626 printk(KERN_ERR "%s: hw csum failure.\n",
1627 dev ? dev->name : "<unknown>");
1631 EXPORT_SYMBOL(netdev_rx_csum_fault);
1634 /* Actually, we should eliminate this check as soon as we know, that:
1635 * 1. IOMMU is present and allows to map all the memory.
1636 * 2. No high memory really exists on this machine.
1639 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1641 #ifdef CONFIG_HIGHMEM
1644 if (dev->features & NETIF_F_HIGHDMA)
1647 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1648 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1656 void (*destructor)(struct sk_buff *skb);
1659 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1661 static void dev_gso_skb_destructor(struct sk_buff *skb)
1663 struct dev_gso_cb *cb;
1666 struct sk_buff *nskb = skb->next;
1668 skb->next = nskb->next;
1671 } while (skb->next);
1673 cb = DEV_GSO_CB(skb);
1675 cb->destructor(skb);
1679 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1680 * @skb: buffer to segment
1682 * This function segments the given skb and stores the list of segments
1685 static int dev_gso_segment(struct sk_buff *skb)
1687 struct net_device *dev = skb->dev;
1688 struct sk_buff *segs;
1689 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1692 segs = skb_gso_segment(skb, features);
1694 /* Verifying header integrity only. */
1699 return PTR_ERR(segs);
1702 DEV_GSO_CB(skb)->destructor = skb->destructor;
1703 skb->destructor = dev_gso_skb_destructor;
1708 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1709 struct netdev_queue *txq)
1711 const struct net_device_ops *ops = dev->netdev_ops;
1714 if (likely(!skb->next)) {
1715 if (!list_empty(&ptype_all))
1716 dev_queue_xmit_nit(skb, dev);
1718 if (netif_needs_gso(dev, skb)) {
1719 if (unlikely(dev_gso_segment(skb)))
1726 * If device doesnt need skb->dst, release it right now while
1727 * its hot in this cpu cache
1729 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1732 rc = ops->ndo_start_xmit(skb, dev);
1733 if (rc == NETDEV_TX_OK)
1734 txq_trans_update(txq);
1736 * TODO: if skb_orphan() was called by
1737 * dev->hard_start_xmit() (for example, the unmodified
1738 * igb driver does that; bnx2 doesn't), then
1739 * skb_tx_software_timestamp() will be unable to send
1740 * back the time stamp.
1742 * How can this be prevented? Always create another
1743 * reference to the socket before calling
1744 * dev->hard_start_xmit()? Prevent that skb_orphan()
1745 * does anything in dev->hard_start_xmit() by clearing
1746 * the skb destructor before the call and restoring it
1747 * afterwards, then doing the skb_orphan() ourselves?
1754 struct sk_buff *nskb = skb->next;
1756 skb->next = nskb->next;
1760 * If device doesnt need nskb->dst, release it right now while
1761 * its hot in this cpu cache
1763 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1766 rc = ops->ndo_start_xmit(nskb, dev);
1767 if (unlikely(rc != NETDEV_TX_OK)) {
1768 nskb->next = skb->next;
1772 txq_trans_update(txq);
1773 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1774 return NETDEV_TX_BUSY;
1775 } while (skb->next);
1777 skb->destructor = DEV_GSO_CB(skb)->destructor;
1781 return NETDEV_TX_OK;
1784 static u32 skb_tx_hashrnd;
1786 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1790 if (skb_rx_queue_recorded(skb)) {
1791 hash = skb_get_rx_queue(skb);
1792 while (unlikely(hash >= dev->real_num_tx_queues))
1793 hash -= dev->real_num_tx_queues;
1797 if (skb->sk && skb->sk->sk_hash)
1798 hash = skb->sk->sk_hash;
1800 hash = skb->protocol;
1802 hash = jhash_1word(hash, skb_tx_hashrnd);
1804 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1806 EXPORT_SYMBOL(skb_tx_hash);
1808 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1809 struct sk_buff *skb)
1811 const struct net_device_ops *ops = dev->netdev_ops;
1812 u16 queue_index = 0;
1814 if (ops->ndo_select_queue)
1815 queue_index = ops->ndo_select_queue(dev, skb);
1816 else if (dev->real_num_tx_queues > 1)
1817 queue_index = skb_tx_hash(dev, skb);
1819 skb_set_queue_mapping(skb, queue_index);
1820 return netdev_get_tx_queue(dev, queue_index);
1823 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1824 struct net_device *dev,
1825 struct netdev_queue *txq)
1827 spinlock_t *root_lock = qdisc_lock(q);
1830 spin_lock(root_lock);
1831 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1834 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1835 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1837 * This is a work-conserving queue; there are no old skbs
1838 * waiting to be sent out; and the qdisc is not running -
1839 * xmit the skb directly.
1841 __qdisc_update_bstats(q, skb->len);
1842 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1845 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1847 rc = NET_XMIT_SUCCESS;
1849 rc = qdisc_enqueue_root(skb, q);
1852 spin_unlock(root_lock);
1858 * dev_queue_xmit - transmit a buffer
1859 * @skb: buffer to transmit
1861 * Queue a buffer for transmission to a network device. The caller must
1862 * have set the device and priority and built the buffer before calling
1863 * this function. The function can be called from an interrupt.
1865 * A negative errno code is returned on a failure. A success does not
1866 * guarantee the frame will be transmitted as it may be dropped due
1867 * to congestion or traffic shaping.
1869 * -----------------------------------------------------------------------------------
1870 * I notice this method can also return errors from the queue disciplines,
1871 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1874 * Regardless of the return value, the skb is consumed, so it is currently
1875 * difficult to retry a send to this method. (You can bump the ref count
1876 * before sending to hold a reference for retry if you are careful.)
1878 * When calling this method, interrupts MUST be enabled. This is because
1879 * the BH enable code must have IRQs enabled so that it will not deadlock.
1882 int dev_queue_xmit(struct sk_buff *skb)
1884 struct net_device *dev = skb->dev;
1885 struct netdev_queue *txq;
1889 /* GSO will handle the following emulations directly. */
1890 if (netif_needs_gso(dev, skb))
1893 if (skb_has_frags(skb) &&
1894 !(dev->features & NETIF_F_FRAGLIST) &&
1895 __skb_linearize(skb))
1898 /* Fragmented skb is linearized if device does not support SG,
1899 * or if at least one of fragments is in highmem and device
1900 * does not support DMA from it.
1902 if (skb_shinfo(skb)->nr_frags &&
1903 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1904 __skb_linearize(skb))
1907 /* If packet is not checksummed and device does not support
1908 * checksumming for this protocol, complete checksumming here.
1910 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1911 skb_set_transport_header(skb, skb->csum_start -
1913 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1918 /* Disable soft irqs for various locks below. Also
1919 * stops preemption for RCU.
1923 txq = dev_pick_tx(dev, skb);
1924 q = rcu_dereference(txq->qdisc);
1926 #ifdef CONFIG_NET_CLS_ACT
1927 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
1930 rc = __dev_xmit_skb(skb, q, dev, txq);
1934 /* The device has no queue. Common case for software devices:
1935 loopback, all the sorts of tunnels...
1937 Really, it is unlikely that netif_tx_lock protection is necessary
1938 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1940 However, it is possible, that they rely on protection
1943 Check this and shot the lock. It is not prone from deadlocks.
1944 Either shot noqueue qdisc, it is even simpler 8)
1946 if (dev->flags & IFF_UP) {
1947 int cpu = smp_processor_id(); /* ok because BHs are off */
1949 if (txq->xmit_lock_owner != cpu) {
1951 HARD_TX_LOCK(dev, txq, cpu);
1953 if (!netif_tx_queue_stopped(txq)) {
1954 rc = NET_XMIT_SUCCESS;
1955 if (!dev_hard_start_xmit(skb, dev, txq)) {
1956 HARD_TX_UNLOCK(dev, txq);
1960 HARD_TX_UNLOCK(dev, txq);
1961 if (net_ratelimit())
1962 printk(KERN_CRIT "Virtual device %s asks to "
1963 "queue packet!\n", dev->name);
1965 /* Recursion is detected! It is possible,
1967 if (net_ratelimit())
1968 printk(KERN_CRIT "Dead loop on virtual device "
1969 "%s, fix it urgently!\n", dev->name);
1974 rcu_read_unlock_bh();
1980 rcu_read_unlock_bh();
1983 EXPORT_SYMBOL(dev_queue_xmit);
1986 /*=======================================================================
1988 =======================================================================*/
1990 int netdev_max_backlog __read_mostly = 1000;
1991 int netdev_budget __read_mostly = 300;
1992 int weight_p __read_mostly = 64; /* old backlog weight */
1994 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1998 * netif_rx - post buffer to the network code
1999 * @skb: buffer to post
2001 * This function receives a packet from a device driver and queues it for
2002 * the upper (protocol) levels to process. It always succeeds. The buffer
2003 * may be dropped during processing for congestion control or by the
2007 * NET_RX_SUCCESS (no congestion)
2008 * NET_RX_DROP (packet was dropped)
2012 int netif_rx(struct sk_buff *skb)
2014 struct softnet_data *queue;
2015 unsigned long flags;
2017 /* if netpoll wants it, pretend we never saw it */
2018 if (netpoll_rx(skb))
2021 if (!skb->tstamp.tv64)
2025 * The code is rearranged so that the path is the most
2026 * short when CPU is congested, but is still operating.
2028 local_irq_save(flags);
2029 queue = &__get_cpu_var(softnet_data);
2031 __get_cpu_var(netdev_rx_stat).total++;
2032 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2033 if (queue->input_pkt_queue.qlen) {
2035 __skb_queue_tail(&queue->input_pkt_queue, skb);
2036 local_irq_restore(flags);
2037 return NET_RX_SUCCESS;
2040 napi_schedule(&queue->backlog);
2044 __get_cpu_var(netdev_rx_stat).dropped++;
2045 local_irq_restore(flags);
2050 EXPORT_SYMBOL(netif_rx);
2052 int netif_rx_ni(struct sk_buff *skb)
2057 err = netif_rx(skb);
2058 if (local_softirq_pending())
2064 EXPORT_SYMBOL(netif_rx_ni);
2066 static void net_tx_action(struct softirq_action *h)
2068 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2070 if (sd->completion_queue) {
2071 struct sk_buff *clist;
2073 local_irq_disable();
2074 clist = sd->completion_queue;
2075 sd->completion_queue = NULL;
2079 struct sk_buff *skb = clist;
2080 clist = clist->next;
2082 WARN_ON(atomic_read(&skb->users));
2087 if (sd->output_queue) {
2090 local_irq_disable();
2091 head = sd->output_queue;
2092 sd->output_queue = NULL;
2096 struct Qdisc *q = head;
2097 spinlock_t *root_lock;
2099 head = head->next_sched;
2101 root_lock = qdisc_lock(q);
2102 if (spin_trylock(root_lock)) {
2103 smp_mb__before_clear_bit();
2104 clear_bit(__QDISC_STATE_SCHED,
2107 spin_unlock(root_lock);
2109 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2111 __netif_reschedule(q);
2113 smp_mb__before_clear_bit();
2114 clear_bit(__QDISC_STATE_SCHED,
2122 static inline int deliver_skb(struct sk_buff *skb,
2123 struct packet_type *pt_prev,
2124 struct net_device *orig_dev)
2126 atomic_inc(&skb->users);
2127 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2130 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2132 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2133 /* This hook is defined here for ATM LANE */
2134 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2135 unsigned char *addr) __read_mostly;
2136 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2140 * If bridge module is loaded call bridging hook.
2141 * returns NULL if packet was consumed.
2143 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2144 struct sk_buff *skb) __read_mostly;
2145 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2147 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2148 struct packet_type **pt_prev, int *ret,
2149 struct net_device *orig_dev)
2151 struct net_bridge_port *port;
2153 if (skb->pkt_type == PACKET_LOOPBACK ||
2154 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2158 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2162 return br_handle_frame_hook(port, skb);
2165 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2168 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2169 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2170 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2172 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2173 struct packet_type **pt_prev,
2175 struct net_device *orig_dev)
2177 if (skb->dev->macvlan_port == NULL)
2181 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2184 return macvlan_handle_frame_hook(skb);
2187 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2190 #ifdef CONFIG_NET_CLS_ACT
2191 /* TODO: Maybe we should just force sch_ingress to be compiled in
2192 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2193 * a compare and 2 stores extra right now if we dont have it on
2194 * but have CONFIG_NET_CLS_ACT
2195 * NOTE: This doesnt stop any functionality; if you dont have
2196 * the ingress scheduler, you just cant add policies on ingress.
2199 static int ing_filter(struct sk_buff *skb)
2201 struct net_device *dev = skb->dev;
2202 u32 ttl = G_TC_RTTL(skb->tc_verd);
2203 struct netdev_queue *rxq;
2204 int result = TC_ACT_OK;
2207 if (MAX_RED_LOOP < ttl++) {
2209 "Redir loop detected Dropping packet (%d->%d)\n",
2210 skb->iif, dev->ifindex);
2214 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2215 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2217 rxq = &dev->rx_queue;
2220 if (q != &noop_qdisc) {
2221 spin_lock(qdisc_lock(q));
2222 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2223 result = qdisc_enqueue_root(skb, q);
2224 spin_unlock(qdisc_lock(q));
2230 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2231 struct packet_type **pt_prev,
2232 int *ret, struct net_device *orig_dev)
2234 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2238 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2241 /* Huh? Why does turning on AF_PACKET affect this? */
2242 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2245 switch (ing_filter(skb)) {
2259 * netif_nit_deliver - deliver received packets to network taps
2262 * This function is used to deliver incoming packets to network
2263 * taps. It should be used when the normal netif_receive_skb path
2264 * is bypassed, for example because of VLAN acceleration.
2266 void netif_nit_deliver(struct sk_buff *skb)
2268 struct packet_type *ptype;
2270 if (list_empty(&ptype_all))
2273 skb_reset_network_header(skb);
2274 skb_reset_transport_header(skb);
2275 skb->mac_len = skb->network_header - skb->mac_header;
2278 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2279 if (!ptype->dev || ptype->dev == skb->dev)
2280 deliver_skb(skb, ptype, skb->dev);
2286 * netif_receive_skb - process receive buffer from network
2287 * @skb: buffer to process
2289 * netif_receive_skb() is the main receive data processing function.
2290 * It always succeeds. The buffer may be dropped during processing
2291 * for congestion control or by the protocol layers.
2293 * This function may only be called from softirq context and interrupts
2294 * should be enabled.
2296 * Return values (usually ignored):
2297 * NET_RX_SUCCESS: no congestion
2298 * NET_RX_DROP: packet was dropped
2300 int netif_receive_skb(struct sk_buff *skb)
2302 struct packet_type *ptype, *pt_prev;
2303 struct net_device *orig_dev;
2304 struct net_device *null_or_orig;
2305 int ret = NET_RX_DROP;
2308 if (!skb->tstamp.tv64)
2311 if (skb->vlan_tci && vlan_hwaccel_do_receive(skb))
2312 return NET_RX_SUCCESS;
2314 /* if we've gotten here through NAPI, check netpoll */
2315 if (netpoll_receive_skb(skb))
2319 skb->iif = skb->dev->ifindex;
2321 null_or_orig = NULL;
2322 orig_dev = skb->dev;
2323 if (orig_dev->master) {
2324 if (skb_bond_should_drop(skb))
2325 null_or_orig = orig_dev; /* deliver only exact match */
2327 skb->dev = orig_dev->master;
2330 __get_cpu_var(netdev_rx_stat).total++;
2332 skb_reset_network_header(skb);
2333 skb_reset_transport_header(skb);
2334 skb->mac_len = skb->network_header - skb->mac_header;
2340 #ifdef CONFIG_NET_CLS_ACT
2341 if (skb->tc_verd & TC_NCLS) {
2342 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2347 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2348 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2349 ptype->dev == orig_dev) {
2351 ret = deliver_skb(skb, pt_prev, orig_dev);
2356 #ifdef CONFIG_NET_CLS_ACT
2357 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2363 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2366 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2370 type = skb->protocol;
2371 list_for_each_entry_rcu(ptype,
2372 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2373 if (ptype->type == type &&
2374 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2375 ptype->dev == orig_dev)) {
2377 ret = deliver_skb(skb, pt_prev, orig_dev);
2383 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2386 /* Jamal, now you will not able to escape explaining
2387 * me how you were going to use this. :-)
2396 EXPORT_SYMBOL(netif_receive_skb);
2398 /* Network device is going away, flush any packets still pending */
2399 static void flush_backlog(void *arg)
2401 struct net_device *dev = arg;
2402 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2403 struct sk_buff *skb, *tmp;
2405 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2406 if (skb->dev == dev) {
2407 __skb_unlink(skb, &queue->input_pkt_queue);
2412 static int napi_gro_complete(struct sk_buff *skb)
2414 struct packet_type *ptype;
2415 __be16 type = skb->protocol;
2416 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2419 if (NAPI_GRO_CB(skb)->count == 1) {
2420 skb_shinfo(skb)->gso_size = 0;
2425 list_for_each_entry_rcu(ptype, head, list) {
2426 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2429 err = ptype->gro_complete(skb);
2435 WARN_ON(&ptype->list == head);
2437 return NET_RX_SUCCESS;
2441 return netif_receive_skb(skb);
2444 void napi_gro_flush(struct napi_struct *napi)
2446 struct sk_buff *skb, *next;
2448 for (skb = napi->gro_list; skb; skb = next) {
2451 napi_gro_complete(skb);
2454 napi->gro_count = 0;
2455 napi->gro_list = NULL;
2457 EXPORT_SYMBOL(napi_gro_flush);
2459 int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2461 struct sk_buff **pp = NULL;
2462 struct packet_type *ptype;
2463 __be16 type = skb->protocol;
2464 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2469 if (!(skb->dev->features & NETIF_F_GRO))
2472 if (skb_is_gso(skb) || skb_has_frags(skb))
2476 list_for_each_entry_rcu(ptype, head, list) {
2477 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2480 skb_set_network_header(skb, skb_gro_offset(skb));
2481 mac_len = skb->network_header - skb->mac_header;
2482 skb->mac_len = mac_len;
2483 NAPI_GRO_CB(skb)->same_flow = 0;
2484 NAPI_GRO_CB(skb)->flush = 0;
2485 NAPI_GRO_CB(skb)->free = 0;
2487 pp = ptype->gro_receive(&napi->gro_list, skb);
2492 if (&ptype->list == head)
2495 same_flow = NAPI_GRO_CB(skb)->same_flow;
2496 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2499 struct sk_buff *nskb = *pp;
2503 napi_gro_complete(nskb);
2510 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2514 NAPI_GRO_CB(skb)->count = 1;
2515 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2516 skb->next = napi->gro_list;
2517 napi->gro_list = skb;
2521 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2522 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2524 BUG_ON(skb->end - skb->tail < grow);
2526 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2529 skb->data_len -= grow;
2531 skb_shinfo(skb)->frags[0].page_offset += grow;
2532 skb_shinfo(skb)->frags[0].size -= grow;
2534 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2535 put_page(skb_shinfo(skb)->frags[0].page);
2536 memmove(skb_shinfo(skb)->frags,
2537 skb_shinfo(skb)->frags + 1,
2538 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
2549 EXPORT_SYMBOL(dev_gro_receive);
2551 static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2555 if (netpoll_rx_on(skb))
2558 for (p = napi->gro_list; p; p = p->next) {
2559 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2560 && !compare_ether_header(skb_mac_header(p),
2561 skb_gro_mac_header(skb));
2562 NAPI_GRO_CB(p)->flush = 0;
2565 return dev_gro_receive(napi, skb);
2568 int napi_skb_finish(int ret, struct sk_buff *skb)
2570 int err = NET_RX_SUCCESS;
2574 return netif_receive_skb(skb);
2580 case GRO_MERGED_FREE:
2587 EXPORT_SYMBOL(napi_skb_finish);
2589 void skb_gro_reset_offset(struct sk_buff *skb)
2591 NAPI_GRO_CB(skb)->data_offset = 0;
2592 NAPI_GRO_CB(skb)->frag0 = NULL;
2593 NAPI_GRO_CB(skb)->frag0_len = 0;
2595 if (skb->mac_header == skb->tail &&
2596 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2597 NAPI_GRO_CB(skb)->frag0 =
2598 page_address(skb_shinfo(skb)->frags[0].page) +
2599 skb_shinfo(skb)->frags[0].page_offset;
2600 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2603 EXPORT_SYMBOL(skb_gro_reset_offset);
2605 int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2607 skb_gro_reset_offset(skb);
2609 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2611 EXPORT_SYMBOL(napi_gro_receive);
2613 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2615 __skb_pull(skb, skb_headlen(skb));
2616 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2617 skb->dev = napi->dev;
2622 EXPORT_SYMBOL(napi_reuse_skb);
2624 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2626 struct net_device *dev = napi->dev;
2627 struct sk_buff *skb = napi->skb;
2630 skb = netdev_alloc_skb(dev, GRO_MAX_HEAD + NET_IP_ALIGN);
2634 skb_reserve(skb, NET_IP_ALIGN);
2642 EXPORT_SYMBOL(napi_get_frags);
2644 int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret)
2646 int err = NET_RX_SUCCESS;
2651 skb->protocol = eth_type_trans(skb, skb->dev);
2653 if (ret == GRO_NORMAL)
2654 return netif_receive_skb(skb);
2656 skb_gro_pull(skb, -ETH_HLEN);
2663 case GRO_MERGED_FREE:
2664 napi_reuse_skb(napi, skb);
2670 EXPORT_SYMBOL(napi_frags_finish);
2672 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2674 struct sk_buff *skb = napi->skb;
2681 skb_reset_mac_header(skb);
2682 skb_gro_reset_offset(skb);
2684 off = skb_gro_offset(skb);
2685 hlen = off + sizeof(*eth);
2686 eth = skb_gro_header_fast(skb, off);
2687 if (skb_gro_header_hard(skb, hlen)) {
2688 eth = skb_gro_header_slow(skb, hlen, off);
2689 if (unlikely(!eth)) {
2690 napi_reuse_skb(napi, skb);
2696 skb_gro_pull(skb, sizeof(*eth));
2699 * This works because the only protocols we care about don't require
2700 * special handling. We'll fix it up properly at the end.
2702 skb->protocol = eth->h_proto;
2707 EXPORT_SYMBOL(napi_frags_skb);
2709 int napi_gro_frags(struct napi_struct *napi)
2711 struct sk_buff *skb = napi_frags_skb(napi);
2716 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2718 EXPORT_SYMBOL(napi_gro_frags);
2720 static int process_backlog(struct napi_struct *napi, int quota)
2723 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2724 unsigned long start_time = jiffies;
2726 napi->weight = weight_p;
2728 struct sk_buff *skb;
2730 local_irq_disable();
2731 skb = __skb_dequeue(&queue->input_pkt_queue);
2733 __napi_complete(napi);
2739 netif_receive_skb(skb);
2740 } while (++work < quota && jiffies == start_time);
2746 * __napi_schedule - schedule for receive
2747 * @n: entry to schedule
2749 * The entry's receive function will be scheduled to run
2751 void __napi_schedule(struct napi_struct *n)
2753 unsigned long flags;
2755 local_irq_save(flags);
2756 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2757 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2758 local_irq_restore(flags);
2760 EXPORT_SYMBOL(__napi_schedule);
2762 void __napi_complete(struct napi_struct *n)
2764 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2765 BUG_ON(n->gro_list);
2767 list_del(&n->poll_list);
2768 smp_mb__before_clear_bit();
2769 clear_bit(NAPI_STATE_SCHED, &n->state);
2771 EXPORT_SYMBOL(__napi_complete);
2773 void napi_complete(struct napi_struct *n)
2775 unsigned long flags;
2778 * don't let napi dequeue from the cpu poll list
2779 * just in case its running on a different cpu
2781 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2785 local_irq_save(flags);
2787 local_irq_restore(flags);
2789 EXPORT_SYMBOL(napi_complete);
2791 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2792 int (*poll)(struct napi_struct *, int), int weight)
2794 INIT_LIST_HEAD(&napi->poll_list);
2795 napi->gro_count = 0;
2796 napi->gro_list = NULL;
2799 napi->weight = weight;
2800 list_add(&napi->dev_list, &dev->napi_list);
2802 #ifdef CONFIG_NETPOLL
2803 spin_lock_init(&napi->poll_lock);
2804 napi->poll_owner = -1;
2806 set_bit(NAPI_STATE_SCHED, &napi->state);
2808 EXPORT_SYMBOL(netif_napi_add);
2810 void netif_napi_del(struct napi_struct *napi)
2812 struct sk_buff *skb, *next;
2814 list_del_init(&napi->dev_list);
2815 napi_free_frags(napi);
2817 for (skb = napi->gro_list; skb; skb = next) {
2823 napi->gro_list = NULL;
2824 napi->gro_count = 0;
2826 EXPORT_SYMBOL(netif_napi_del);
2829 static void net_rx_action(struct softirq_action *h)
2831 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2832 unsigned long time_limit = jiffies + 2;
2833 int budget = netdev_budget;
2836 local_irq_disable();
2838 while (!list_empty(list)) {
2839 struct napi_struct *n;
2842 /* If softirq window is exhuasted then punt.
2843 * Allow this to run for 2 jiffies since which will allow
2844 * an average latency of 1.5/HZ.
2846 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2851 /* Even though interrupts have been re-enabled, this
2852 * access is safe because interrupts can only add new
2853 * entries to the tail of this list, and only ->poll()
2854 * calls can remove this head entry from the list.
2856 n = list_entry(list->next, struct napi_struct, poll_list);
2858 have = netpoll_poll_lock(n);
2862 /* This NAPI_STATE_SCHED test is for avoiding a race
2863 * with netpoll's poll_napi(). Only the entity which
2864 * obtains the lock and sees NAPI_STATE_SCHED set will
2865 * actually make the ->poll() call. Therefore we avoid
2866 * accidently calling ->poll() when NAPI is not scheduled.
2869 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2870 work = n->poll(n, weight);
2874 WARN_ON_ONCE(work > weight);
2878 local_irq_disable();
2880 /* Drivers must not modify the NAPI state if they
2881 * consume the entire weight. In such cases this code
2882 * still "owns" the NAPI instance and therefore can
2883 * move the instance around on the list at-will.
2885 if (unlikely(work == weight)) {
2886 if (unlikely(napi_disable_pending(n))) {
2889 local_irq_disable();
2891 list_move_tail(&n->poll_list, list);
2894 netpoll_poll_unlock(have);
2899 #ifdef CONFIG_NET_DMA
2901 * There may not be any more sk_buffs coming right now, so push
2902 * any pending DMA copies to hardware
2904 dma_issue_pending_all();
2910 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2911 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2915 static gifconf_func_t *gifconf_list[NPROTO];
2918 * register_gifconf - register a SIOCGIF handler
2919 * @family: Address family
2920 * @gifconf: Function handler
2922 * Register protocol dependent address dumping routines. The handler
2923 * that is passed must not be freed or reused until it has been replaced
2924 * by another handler.
2926 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
2928 if (family >= NPROTO)
2930 gifconf_list[family] = gifconf;
2933 EXPORT_SYMBOL(register_gifconf);
2937 * Map an interface index to its name (SIOCGIFNAME)
2941 * We need this ioctl for efficient implementation of the
2942 * if_indextoname() function required by the IPv6 API. Without
2943 * it, we would have to search all the interfaces to find a
2947 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2949 struct net_device *dev;
2953 * Fetch the caller's info block.
2956 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2959 read_lock(&dev_base_lock);
2960 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2962 read_unlock(&dev_base_lock);
2966 strcpy(ifr.ifr_name, dev->name);
2967 read_unlock(&dev_base_lock);
2969 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2975 * Perform a SIOCGIFCONF call. This structure will change
2976 * size eventually, and there is nothing I can do about it.
2977 * Thus we will need a 'compatibility mode'.
2980 static int dev_ifconf(struct net *net, char __user *arg)
2983 struct net_device *dev;
2990 * Fetch the caller's info block.
2993 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3000 * Loop over the interfaces, and write an info block for each.
3004 for_each_netdev(net, dev) {
3005 for (i = 0; i < NPROTO; i++) {
3006 if (gifconf_list[i]) {
3009 done = gifconf_list[i](dev, NULL, 0);
3011 done = gifconf_list[i](dev, pos + total,
3021 * All done. Write the updated control block back to the caller.
3023 ifc.ifc_len = total;
3026 * Both BSD and Solaris return 0 here, so we do too.
3028 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3031 #ifdef CONFIG_PROC_FS
3033 * This is invoked by the /proc filesystem handler to display a device
3036 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3037 __acquires(dev_base_lock)
3039 struct net *net = seq_file_net(seq);
3041 struct net_device *dev;
3043 read_lock(&dev_base_lock);
3045 return SEQ_START_TOKEN;
3048 for_each_netdev(net, dev)
3055 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3057 struct net *net = seq_file_net(seq);
3059 return v == SEQ_START_TOKEN ?
3060 first_net_device(net) : next_net_device((struct net_device *)v);
3063 void dev_seq_stop(struct seq_file *seq, void *v)
3064 __releases(dev_base_lock)
3066 read_unlock(&dev_base_lock);
3069 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3071 const struct net_device_stats *stats = dev_get_stats(dev);
3073 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3074 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3075 dev->name, stats->rx_bytes, stats->rx_packets,
3077 stats->rx_dropped + stats->rx_missed_errors,
3078 stats->rx_fifo_errors,
3079 stats->rx_length_errors + stats->rx_over_errors +
3080 stats->rx_crc_errors + stats->rx_frame_errors,
3081 stats->rx_compressed, stats->multicast,
3082 stats->tx_bytes, stats->tx_packets,
3083 stats->tx_errors, stats->tx_dropped,
3084 stats->tx_fifo_errors, stats->collisions,
3085 stats->tx_carrier_errors +
3086 stats->tx_aborted_errors +
3087 stats->tx_window_errors +
3088 stats->tx_heartbeat_errors,
3089 stats->tx_compressed);
3093 * Called from the PROCfs module. This now uses the new arbitrary sized
3094 * /proc/net interface to create /proc/net/dev
3096 static int dev_seq_show(struct seq_file *seq, void *v)
3098 if (v == SEQ_START_TOKEN)
3099 seq_puts(seq, "Inter-| Receive "
3101 " face |bytes packets errs drop fifo frame "
3102 "compressed multicast|bytes packets errs "
3103 "drop fifo colls carrier compressed\n");
3105 dev_seq_printf_stats(seq, v);
3109 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3111 struct netif_rx_stats *rc = NULL;
3113 while (*pos < nr_cpu_ids)
3114 if (cpu_online(*pos)) {
3115 rc = &per_cpu(netdev_rx_stat, *pos);
3122 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3124 return softnet_get_online(pos);
3127 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3130 return softnet_get_online(pos);
3133 static void softnet_seq_stop(struct seq_file *seq, void *v)
3137 static int softnet_seq_show(struct seq_file *seq, void *v)
3139 struct netif_rx_stats *s = v;
3141 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3142 s->total, s->dropped, s->time_squeeze, 0,
3143 0, 0, 0, 0, /* was fastroute */
3148 static const struct seq_operations dev_seq_ops = {
3149 .start = dev_seq_start,
3150 .next = dev_seq_next,
3151 .stop = dev_seq_stop,
3152 .show = dev_seq_show,
3155 static int dev_seq_open(struct inode *inode, struct file *file)
3157 return seq_open_net(inode, file, &dev_seq_ops,
3158 sizeof(struct seq_net_private));
3161 static const struct file_operations dev_seq_fops = {
3162 .owner = THIS_MODULE,
3163 .open = dev_seq_open,
3165 .llseek = seq_lseek,
3166 .release = seq_release_net,
3169 static const struct seq_operations softnet_seq_ops = {
3170 .start = softnet_seq_start,
3171 .next = softnet_seq_next,
3172 .stop = softnet_seq_stop,
3173 .show = softnet_seq_show,
3176 static int softnet_seq_open(struct inode *inode, struct file *file)
3178 return seq_open(file, &softnet_seq_ops);
3181 static const struct file_operations softnet_seq_fops = {
3182 .owner = THIS_MODULE,
3183 .open = softnet_seq_open,
3185 .llseek = seq_lseek,
3186 .release = seq_release,
3189 static void *ptype_get_idx(loff_t pos)
3191 struct packet_type *pt = NULL;
3195 list_for_each_entry_rcu(pt, &ptype_all, list) {
3201 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3202 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3211 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3215 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3218 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3220 struct packet_type *pt;
3221 struct list_head *nxt;
3225 if (v == SEQ_START_TOKEN)
3226 return ptype_get_idx(0);
3229 nxt = pt->list.next;
3230 if (pt->type == htons(ETH_P_ALL)) {
3231 if (nxt != &ptype_all)
3234 nxt = ptype_base[0].next;
3236 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3238 while (nxt == &ptype_base[hash]) {
3239 if (++hash >= PTYPE_HASH_SIZE)
3241 nxt = ptype_base[hash].next;
3244 return list_entry(nxt, struct packet_type, list);
3247 static void ptype_seq_stop(struct seq_file *seq, void *v)
3253 static int ptype_seq_show(struct seq_file *seq, void *v)
3255 struct packet_type *pt = v;
3257 if (v == SEQ_START_TOKEN)
3258 seq_puts(seq, "Type Device Function\n");
3259 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3260 if (pt->type == htons(ETH_P_ALL))
3261 seq_puts(seq, "ALL ");
3263 seq_printf(seq, "%04x", ntohs(pt->type));
3265 seq_printf(seq, " %-8s %pF\n",
3266 pt->dev ? pt->dev->name : "", pt->func);
3272 static const struct seq_operations ptype_seq_ops = {
3273 .start = ptype_seq_start,
3274 .next = ptype_seq_next,
3275 .stop = ptype_seq_stop,
3276 .show = ptype_seq_show,
3279 static int ptype_seq_open(struct inode *inode, struct file *file)
3281 return seq_open_net(inode, file, &ptype_seq_ops,
3282 sizeof(struct seq_net_private));
3285 static const struct file_operations ptype_seq_fops = {
3286 .owner = THIS_MODULE,
3287 .open = ptype_seq_open,
3289 .llseek = seq_lseek,
3290 .release = seq_release_net,
3294 static int __net_init dev_proc_net_init(struct net *net)
3298 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3300 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3302 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3305 if (wext_proc_init(net))
3311 proc_net_remove(net, "ptype");
3313 proc_net_remove(net, "softnet_stat");
3315 proc_net_remove(net, "dev");
3319 static void __net_exit dev_proc_net_exit(struct net *net)
3321 wext_proc_exit(net);
3323 proc_net_remove(net, "ptype");
3324 proc_net_remove(net, "softnet_stat");
3325 proc_net_remove(net, "dev");
3328 static struct pernet_operations __net_initdata dev_proc_ops = {
3329 .init = dev_proc_net_init,
3330 .exit = dev_proc_net_exit,
3333 static int __init dev_proc_init(void)
3335 return register_pernet_subsys(&dev_proc_ops);
3338 #define dev_proc_init() 0
3339 #endif /* CONFIG_PROC_FS */
3343 * netdev_set_master - set up master/slave pair
3344 * @slave: slave device
3345 * @master: new master device
3347 * Changes the master device of the slave. Pass %NULL to break the
3348 * bonding. The caller must hold the RTNL semaphore. On a failure
3349 * a negative errno code is returned. On success the reference counts
3350 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3351 * function returns zero.
3353 int netdev_set_master(struct net_device *slave, struct net_device *master)
3355 struct net_device *old = slave->master;
3365 slave->master = master;
3373 slave->flags |= IFF_SLAVE;
3375 slave->flags &= ~IFF_SLAVE;
3377 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3380 EXPORT_SYMBOL(netdev_set_master);
3382 static void dev_change_rx_flags(struct net_device *dev, int flags)
3384 const struct net_device_ops *ops = dev->netdev_ops;
3386 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3387 ops->ndo_change_rx_flags(dev, flags);
3390 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3392 unsigned short old_flags = dev->flags;
3398 dev->flags |= IFF_PROMISC;
3399 dev->promiscuity += inc;
3400 if (dev->promiscuity == 0) {
3403 * If inc causes overflow, untouch promisc and return error.
3406 dev->flags &= ~IFF_PROMISC;
3408 dev->promiscuity -= inc;
3409 printk(KERN_WARNING "%s: promiscuity touches roof, "
3410 "set promiscuity failed, promiscuity feature "
3411 "of device might be broken.\n", dev->name);
3415 if (dev->flags != old_flags) {
3416 printk(KERN_INFO "device %s %s promiscuous mode\n",
3417 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3419 if (audit_enabled) {
3420 current_uid_gid(&uid, &gid);
3421 audit_log(current->audit_context, GFP_ATOMIC,
3422 AUDIT_ANOM_PROMISCUOUS,
3423 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3424 dev->name, (dev->flags & IFF_PROMISC),
3425 (old_flags & IFF_PROMISC),
3426 audit_get_loginuid(current),
3428 audit_get_sessionid(current));
3431 dev_change_rx_flags(dev, IFF_PROMISC);
3437 * dev_set_promiscuity - update promiscuity count on a device
3441 * Add or remove promiscuity from a device. While the count in the device
3442 * remains above zero the interface remains promiscuous. Once it hits zero
3443 * the device reverts back to normal filtering operation. A negative inc
3444 * value is used to drop promiscuity on the device.
3445 * Return 0 if successful or a negative errno code on error.
3447 int dev_set_promiscuity(struct net_device *dev, int inc)
3449 unsigned short old_flags = dev->flags;
3452 err = __dev_set_promiscuity(dev, inc);
3455 if (dev->flags != old_flags)
3456 dev_set_rx_mode(dev);
3459 EXPORT_SYMBOL(dev_set_promiscuity);
3462 * dev_set_allmulti - update allmulti count on a device
3466 * Add or remove reception of all multicast frames to a device. While the
3467 * count in the device remains above zero the interface remains listening
3468 * to all interfaces. Once it hits zero the device reverts back to normal
3469 * filtering operation. A negative @inc value is used to drop the counter
3470 * when releasing a resource needing all multicasts.
3471 * Return 0 if successful or a negative errno code on error.
3474 int dev_set_allmulti(struct net_device *dev, int inc)
3476 unsigned short old_flags = dev->flags;
3480 dev->flags |= IFF_ALLMULTI;
3481 dev->allmulti += inc;
3482 if (dev->allmulti == 0) {
3485 * If inc causes overflow, untouch allmulti and return error.
3488 dev->flags &= ~IFF_ALLMULTI;
3490 dev->allmulti -= inc;
3491 printk(KERN_WARNING "%s: allmulti touches roof, "
3492 "set allmulti failed, allmulti feature of "
3493 "device might be broken.\n", dev->name);
3497 if (dev->flags ^ old_flags) {
3498 dev_change_rx_flags(dev, IFF_ALLMULTI);
3499 dev_set_rx_mode(dev);
3503 EXPORT_SYMBOL(dev_set_allmulti);
3506 * Upload unicast and multicast address lists to device and
3507 * configure RX filtering. When the device doesn't support unicast
3508 * filtering it is put in promiscuous mode while unicast addresses
3511 void __dev_set_rx_mode(struct net_device *dev)
3513 const struct net_device_ops *ops = dev->netdev_ops;
3515 /* dev_open will call this function so the list will stay sane. */
3516 if (!(dev->flags&IFF_UP))
3519 if (!netif_device_present(dev))
3522 if (ops->ndo_set_rx_mode)
3523 ops->ndo_set_rx_mode(dev);
3525 /* Unicast addresses changes may only happen under the rtnl,
3526 * therefore calling __dev_set_promiscuity here is safe.
3528 if (dev->uc.count > 0 && !dev->uc_promisc) {
3529 __dev_set_promiscuity(dev, 1);
3530 dev->uc_promisc = 1;
3531 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3532 __dev_set_promiscuity(dev, -1);
3533 dev->uc_promisc = 0;
3536 if (ops->ndo_set_multicast_list)
3537 ops->ndo_set_multicast_list(dev);
3541 void dev_set_rx_mode(struct net_device *dev)
3543 netif_addr_lock_bh(dev);
3544 __dev_set_rx_mode(dev);
3545 netif_addr_unlock_bh(dev);
3548 /* hw addresses list handling functions */
3550 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3551 int addr_len, unsigned char addr_type)
3553 struct netdev_hw_addr *ha;
3556 if (addr_len > MAX_ADDR_LEN)
3559 list_for_each_entry(ha, &list->list, list) {
3560 if (!memcmp(ha->addr, addr, addr_len) &&
3561 ha->type == addr_type) {
3568 alloc_size = sizeof(*ha);
3569 if (alloc_size < L1_CACHE_BYTES)
3570 alloc_size = L1_CACHE_BYTES;
3571 ha = kmalloc(alloc_size, GFP_ATOMIC);
3574 memcpy(ha->addr, addr, addr_len);
3575 ha->type = addr_type;
3578 list_add_tail_rcu(&ha->list, &list->list);
3583 static void ha_rcu_free(struct rcu_head *head)
3585 struct netdev_hw_addr *ha;
3587 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3591 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3592 int addr_len, unsigned char addr_type)
3594 struct netdev_hw_addr *ha;
3596 list_for_each_entry(ha, &list->list, list) {
3597 if (!memcmp(ha->addr, addr, addr_len) &&
3598 (ha->type == addr_type || !addr_type)) {
3601 list_del_rcu(&ha->list);
3602 call_rcu(&ha->rcu_head, ha_rcu_free);
3610 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3611 struct netdev_hw_addr_list *from_list,
3613 unsigned char addr_type)
3616 struct netdev_hw_addr *ha, *ha2;
3619 list_for_each_entry(ha, &from_list->list, list) {
3620 type = addr_type ? addr_type : ha->type;
3621 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3628 list_for_each_entry(ha2, &from_list->list, list) {
3631 type = addr_type ? addr_type : ha2->type;
3632 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3637 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3638 struct netdev_hw_addr_list *from_list,
3640 unsigned char addr_type)
3642 struct netdev_hw_addr *ha;
3645 list_for_each_entry(ha, &from_list->list, list) {
3646 type = addr_type ? addr_type : ha->type;
3647 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3651 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3652 struct netdev_hw_addr_list *from_list,
3656 struct netdev_hw_addr *ha, *tmp;
3658 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3660 err = __hw_addr_add(to_list, ha->addr,
3661 addr_len, ha->type);
3666 } else if (ha->refcount == 1) {
3667 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3668 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3674 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3675 struct netdev_hw_addr_list *from_list,
3678 struct netdev_hw_addr *ha, *tmp;
3680 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3682 __hw_addr_del(to_list, ha->addr,
3683 addr_len, ha->type);
3685 __hw_addr_del(from_list, ha->addr,
3686 addr_len, ha->type);
3691 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3693 struct netdev_hw_addr *ha, *tmp;
3695 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3696 list_del_rcu(&ha->list);
3697 call_rcu(&ha->rcu_head, ha_rcu_free);
3702 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3704 INIT_LIST_HEAD(&list->list);
3708 /* Device addresses handling functions */
3710 static void dev_addr_flush(struct net_device *dev)
3712 /* rtnl_mutex must be held here */
3714 __hw_addr_flush(&dev->dev_addrs);
3715 dev->dev_addr = NULL;
3718 static int dev_addr_init(struct net_device *dev)
3720 unsigned char addr[MAX_ADDR_LEN];
3721 struct netdev_hw_addr *ha;
3724 /* rtnl_mutex must be held here */
3726 __hw_addr_init(&dev->dev_addrs);
3727 memset(addr, 0, sizeof(addr));
3728 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3729 NETDEV_HW_ADDR_T_LAN);
3732 * Get the first (previously created) address from the list
3733 * and set dev_addr pointer to this location.
3735 ha = list_first_entry(&dev->dev_addrs.list,
3736 struct netdev_hw_addr, list);
3737 dev->dev_addr = ha->addr;
3743 * dev_addr_add - Add a device address
3745 * @addr: address to add
3746 * @addr_type: address type
3748 * Add a device address to the device or increase the reference count if
3749 * it already exists.
3751 * The caller must hold the rtnl_mutex.
3753 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3754 unsigned char addr_type)
3760 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3762 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3765 EXPORT_SYMBOL(dev_addr_add);
3768 * dev_addr_del - Release a device address.
3770 * @addr: address to delete
3771 * @addr_type: address type
3773 * Release reference to a device address and remove it from the device
3774 * if the reference count drops to zero.
3776 * The caller must hold the rtnl_mutex.
3778 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3779 unsigned char addr_type)
3782 struct netdev_hw_addr *ha;
3787 * We can not remove the first address from the list because
3788 * dev->dev_addr points to that.
3790 ha = list_first_entry(&dev->dev_addrs.list,
3791 struct netdev_hw_addr, list);
3792 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3795 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3798 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3801 EXPORT_SYMBOL(dev_addr_del);
3804 * dev_addr_add_multiple - Add device addresses from another device
3805 * @to_dev: device to which addresses will be added
3806 * @from_dev: device from which addresses will be added
3807 * @addr_type: address type - 0 means type will be used from from_dev
3809 * Add device addresses of the one device to another.
3811 * The caller must hold the rtnl_mutex.
3813 int dev_addr_add_multiple(struct net_device *to_dev,
3814 struct net_device *from_dev,
3815 unsigned char addr_type)
3821 if (from_dev->addr_len != to_dev->addr_len)
3823 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3824 to_dev->addr_len, addr_type);
3826 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3829 EXPORT_SYMBOL(dev_addr_add_multiple);
3832 * dev_addr_del_multiple - Delete device addresses by another device
3833 * @to_dev: device where the addresses will be deleted
3834 * @from_dev: device by which addresses the addresses will be deleted
3835 * @addr_type: address type - 0 means type will used from from_dev
3837 * Deletes addresses in to device by the list of addresses in from device.
3839 * The caller must hold the rtnl_mutex.
3841 int dev_addr_del_multiple(struct net_device *to_dev,
3842 struct net_device *from_dev,
3843 unsigned char addr_type)
3847 if (from_dev->addr_len != to_dev->addr_len)
3849 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3850 to_dev->addr_len, addr_type);
3851 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3854 EXPORT_SYMBOL(dev_addr_del_multiple);
3856 /* multicast addresses handling functions */
3858 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3859 void *addr, int alen, int glbl)
3861 struct dev_addr_list *da;
3863 for (; (da = *list) != NULL; list = &da->next) {
3864 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3865 alen == da->da_addrlen) {
3867 int old_glbl = da->da_gusers;
3884 int __dev_addr_add(struct dev_addr_list **list, int *count,
3885 void *addr, int alen, int glbl)
3887 struct dev_addr_list *da;
3889 for (da = *list; da != NULL; da = da->next) {
3890 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3891 da->da_addrlen == alen) {
3893 int old_glbl = da->da_gusers;
3903 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3906 memcpy(da->da_addr, addr, alen);
3907 da->da_addrlen = alen;
3909 da->da_gusers = glbl ? 1 : 0;
3917 * dev_unicast_delete - Release secondary unicast address.
3919 * @addr: address to delete
3921 * Release reference to a secondary unicast address and remove it
3922 * from the device if the reference count drops to zero.
3924 * The caller must hold the rtnl_mutex.
3926 int dev_unicast_delete(struct net_device *dev, void *addr)
3932 netif_addr_lock_bh(dev);
3933 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
3934 NETDEV_HW_ADDR_T_UNICAST);
3936 __dev_set_rx_mode(dev);
3937 netif_addr_unlock_bh(dev);
3940 EXPORT_SYMBOL(dev_unicast_delete);
3943 * dev_unicast_add - add a secondary unicast address
3945 * @addr: address to add
3947 * Add a secondary unicast address to the device or increase
3948 * the reference count if it already exists.
3950 * The caller must hold the rtnl_mutex.
3952 int dev_unicast_add(struct net_device *dev, void *addr)
3958 netif_addr_lock_bh(dev);
3959 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
3960 NETDEV_HW_ADDR_T_UNICAST);
3962 __dev_set_rx_mode(dev);
3963 netif_addr_unlock_bh(dev);
3966 EXPORT_SYMBOL(dev_unicast_add);
3968 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3969 struct dev_addr_list **from, int *from_count)
3971 struct dev_addr_list *da, *next;
3975 while (da != NULL) {
3977 if (!da->da_synced) {
3978 err = __dev_addr_add(to, to_count,
3979 da->da_addr, da->da_addrlen, 0);
3984 } else if (da->da_users == 1) {
3985 __dev_addr_delete(to, to_count,
3986 da->da_addr, da->da_addrlen, 0);
3987 __dev_addr_delete(from, from_count,
3988 da->da_addr, da->da_addrlen, 0);
3994 EXPORT_SYMBOL_GPL(__dev_addr_sync);
3996 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3997 struct dev_addr_list **from, int *from_count)
3999 struct dev_addr_list *da, *next;
4002 while (da != NULL) {
4004 if (da->da_synced) {
4005 __dev_addr_delete(to, to_count,
4006 da->da_addr, da->da_addrlen, 0);
4008 __dev_addr_delete(from, from_count,
4009 da->da_addr, da->da_addrlen, 0);
4014 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4017 * dev_unicast_sync - Synchronize device's unicast list to another device
4018 * @to: destination device
4019 * @from: source device
4021 * Add newly added addresses to the destination device and release
4022 * addresses that have no users left. The source device must be
4023 * locked by netif_tx_lock_bh.
4025 * This function is intended to be called from the dev->set_rx_mode
4026 * function of layered software devices.
4028 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4032 if (to->addr_len != from->addr_len)
4035 netif_addr_lock_bh(to);
4036 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4038 __dev_set_rx_mode(to);
4039 netif_addr_unlock_bh(to);
4042 EXPORT_SYMBOL(dev_unicast_sync);
4045 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4046 * @to: destination device
4047 * @from: source device
4049 * Remove all addresses that were added to the destination device by
4050 * dev_unicast_sync(). This function is intended to be called from the
4051 * dev->stop function of layered software devices.
4053 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4055 if (to->addr_len != from->addr_len)
4058 netif_addr_lock_bh(from);
4059 netif_addr_lock(to);
4060 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4061 __dev_set_rx_mode(to);
4062 netif_addr_unlock(to);
4063 netif_addr_unlock_bh(from);
4065 EXPORT_SYMBOL(dev_unicast_unsync);
4067 static void dev_unicast_flush(struct net_device *dev)
4069 netif_addr_lock_bh(dev);
4070 __hw_addr_flush(&dev->uc);
4071 netif_addr_unlock_bh(dev);
4074 static void dev_unicast_init(struct net_device *dev)
4076 __hw_addr_init(&dev->uc);
4080 static void __dev_addr_discard(struct dev_addr_list **list)
4082 struct dev_addr_list *tmp;
4084 while (*list != NULL) {
4087 if (tmp->da_users > tmp->da_gusers)
4088 printk("__dev_addr_discard: address leakage! "
4089 "da_users=%d\n", tmp->da_users);
4094 static void dev_addr_discard(struct net_device *dev)
4096 netif_addr_lock_bh(dev);
4098 __dev_addr_discard(&dev->mc_list);
4101 netif_addr_unlock_bh(dev);
4105 * dev_get_flags - get flags reported to userspace
4108 * Get the combination of flag bits exported through APIs to userspace.
4110 unsigned dev_get_flags(const struct net_device *dev)
4114 flags = (dev->flags & ~(IFF_PROMISC |
4119 (dev->gflags & (IFF_PROMISC |
4122 if (netif_running(dev)) {
4123 if (netif_oper_up(dev))
4124 flags |= IFF_RUNNING;
4125 if (netif_carrier_ok(dev))
4126 flags |= IFF_LOWER_UP;
4127 if (netif_dormant(dev))
4128 flags |= IFF_DORMANT;
4133 EXPORT_SYMBOL(dev_get_flags);
4136 * dev_change_flags - change device settings
4138 * @flags: device state flags
4140 * Change settings on device based state flags. The flags are
4141 * in the userspace exported format.
4143 int dev_change_flags(struct net_device *dev, unsigned flags)
4146 int old_flags = dev->flags;
4151 * Set the flags on our device.
4154 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4155 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4157 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4161 * Load in the correct multicast list now the flags have changed.
4164 if ((old_flags ^ flags) & IFF_MULTICAST)
4165 dev_change_rx_flags(dev, IFF_MULTICAST);
4167 dev_set_rx_mode(dev);
4170 * Have we downed the interface. We handle IFF_UP ourselves
4171 * according to user attempts to set it, rather than blindly
4176 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4177 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4180 dev_set_rx_mode(dev);
4183 if (dev->flags & IFF_UP &&
4184 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4186 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4188 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4189 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4191 dev->gflags ^= IFF_PROMISC;
4192 dev_set_promiscuity(dev, inc);
4195 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4196 is important. Some (broken) drivers set IFF_PROMISC, when
4197 IFF_ALLMULTI is requested not asking us and not reporting.
4199 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4200 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4202 dev->gflags ^= IFF_ALLMULTI;
4203 dev_set_allmulti(dev, inc);
4206 /* Exclude state transition flags, already notified */
4207 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4209 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4213 EXPORT_SYMBOL(dev_change_flags);
4216 * dev_set_mtu - Change maximum transfer unit
4218 * @new_mtu: new transfer unit
4220 * Change the maximum transfer size of the network device.
4222 int dev_set_mtu(struct net_device *dev, int new_mtu)
4224 const struct net_device_ops *ops = dev->netdev_ops;
4227 if (new_mtu == dev->mtu)
4230 /* MTU must be positive. */
4234 if (!netif_device_present(dev))
4238 if (ops->ndo_change_mtu)
4239 err = ops->ndo_change_mtu(dev, new_mtu);
4243 if (!err && dev->flags & IFF_UP)
4244 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4247 EXPORT_SYMBOL(dev_set_mtu);
4250 * dev_set_mac_address - Change Media Access Control Address
4254 * Change the hardware (MAC) address of the device
4256 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4258 const struct net_device_ops *ops = dev->netdev_ops;
4261 if (!ops->ndo_set_mac_address)
4263 if (sa->sa_family != dev->type)
4265 if (!netif_device_present(dev))
4267 err = ops->ndo_set_mac_address(dev, sa);
4269 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4272 EXPORT_SYMBOL(dev_set_mac_address);
4275 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
4277 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4280 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4286 case SIOCGIFFLAGS: /* Get interface flags */
4287 ifr->ifr_flags = (short) dev_get_flags(dev);
4290 case SIOCGIFMETRIC: /* Get the metric on the interface
4291 (currently unused) */
4292 ifr->ifr_metric = 0;
4295 case SIOCGIFMTU: /* Get the MTU of a device */
4296 ifr->ifr_mtu = dev->mtu;
4301 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4303 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4304 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4305 ifr->ifr_hwaddr.sa_family = dev->type;
4313 ifr->ifr_map.mem_start = dev->mem_start;
4314 ifr->ifr_map.mem_end = dev->mem_end;
4315 ifr->ifr_map.base_addr = dev->base_addr;
4316 ifr->ifr_map.irq = dev->irq;
4317 ifr->ifr_map.dma = dev->dma;
4318 ifr->ifr_map.port = dev->if_port;
4322 ifr->ifr_ifindex = dev->ifindex;
4326 ifr->ifr_qlen = dev->tx_queue_len;
4330 /* dev_ioctl() should ensure this case
4342 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4344 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4347 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4348 const struct net_device_ops *ops;
4353 ops = dev->netdev_ops;
4356 case SIOCSIFFLAGS: /* Set interface flags */
4357 return dev_change_flags(dev, ifr->ifr_flags);
4359 case SIOCSIFMETRIC: /* Set the metric on the interface
4360 (currently unused) */
4363 case SIOCSIFMTU: /* Set the MTU of a device */
4364 return dev_set_mtu(dev, ifr->ifr_mtu);
4367 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4369 case SIOCSIFHWBROADCAST:
4370 if (ifr->ifr_hwaddr.sa_family != dev->type)
4372 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4373 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4374 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4378 if (ops->ndo_set_config) {
4379 if (!netif_device_present(dev))
4381 return ops->ndo_set_config(dev, &ifr->ifr_map);
4386 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4387 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4389 if (!netif_device_present(dev))
4391 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4395 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4396 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4398 if (!netif_device_present(dev))
4400 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4404 if (ifr->ifr_qlen < 0)
4406 dev->tx_queue_len = ifr->ifr_qlen;
4410 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4411 return dev_change_name(dev, ifr->ifr_newname);
4414 * Unknown or private ioctl
4417 if ((cmd >= SIOCDEVPRIVATE &&
4418 cmd <= SIOCDEVPRIVATE + 15) ||
4419 cmd == SIOCBONDENSLAVE ||
4420 cmd == SIOCBONDRELEASE ||
4421 cmd == SIOCBONDSETHWADDR ||
4422 cmd == SIOCBONDSLAVEINFOQUERY ||
4423 cmd == SIOCBONDINFOQUERY ||
4424 cmd == SIOCBONDCHANGEACTIVE ||
4425 cmd == SIOCGMIIPHY ||
4426 cmd == SIOCGMIIREG ||
4427 cmd == SIOCSMIIREG ||
4428 cmd == SIOCBRADDIF ||
4429 cmd == SIOCBRDELIF ||
4430 cmd == SIOCSHWTSTAMP ||
4431 cmd == SIOCWANDEV) {
4433 if (ops->ndo_do_ioctl) {
4434 if (netif_device_present(dev))
4435 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4447 * This function handles all "interface"-type I/O control requests. The actual
4448 * 'doing' part of this is dev_ifsioc above.
4452 * dev_ioctl - network device ioctl
4453 * @net: the applicable net namespace
4454 * @cmd: command to issue
4455 * @arg: pointer to a struct ifreq in user space
4457 * Issue ioctl functions to devices. This is normally called by the
4458 * user space syscall interfaces but can sometimes be useful for
4459 * other purposes. The return value is the return from the syscall if
4460 * positive or a negative errno code on error.
4463 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4469 /* One special case: SIOCGIFCONF takes ifconf argument
4470 and requires shared lock, because it sleeps writing
4474 if (cmd == SIOCGIFCONF) {
4476 ret = dev_ifconf(net, (char __user *) arg);
4480 if (cmd == SIOCGIFNAME)
4481 return dev_ifname(net, (struct ifreq __user *)arg);
4483 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4486 ifr.ifr_name[IFNAMSIZ-1] = 0;
4488 colon = strchr(ifr.ifr_name, ':');
4493 * See which interface the caller is talking about.
4498 * These ioctl calls:
4499 * - can be done by all.
4500 * - atomic and do not require locking.
4511 dev_load(net, ifr.ifr_name);
4512 read_lock(&dev_base_lock);
4513 ret = dev_ifsioc_locked(net, &ifr, cmd);
4514 read_unlock(&dev_base_lock);
4518 if (copy_to_user(arg, &ifr,
4519 sizeof(struct ifreq)))
4525 dev_load(net, ifr.ifr_name);
4527 ret = dev_ethtool(net, &ifr);
4532 if (copy_to_user(arg, &ifr,
4533 sizeof(struct ifreq)))
4539 * These ioctl calls:
4540 * - require superuser power.
4541 * - require strict serialization.
4547 if (!capable(CAP_NET_ADMIN))
4549 dev_load(net, ifr.ifr_name);
4551 ret = dev_ifsioc(net, &ifr, cmd);
4556 if (copy_to_user(arg, &ifr,
4557 sizeof(struct ifreq)))
4563 * These ioctl calls:
4564 * - require superuser power.
4565 * - require strict serialization.
4566 * - do not return a value
4576 case SIOCSIFHWBROADCAST:
4579 case SIOCBONDENSLAVE:
4580 case SIOCBONDRELEASE:
4581 case SIOCBONDSETHWADDR:
4582 case SIOCBONDCHANGEACTIVE:
4586 if (!capable(CAP_NET_ADMIN))
4589 case SIOCBONDSLAVEINFOQUERY:
4590 case SIOCBONDINFOQUERY:
4591 dev_load(net, ifr.ifr_name);
4593 ret = dev_ifsioc(net, &ifr, cmd);
4598 /* Get the per device memory space. We can add this but
4599 * currently do not support it */
4601 /* Set the per device memory buffer space.
4602 * Not applicable in our case */
4607 * Unknown or private ioctl.
4610 if (cmd == SIOCWANDEV ||
4611 (cmd >= SIOCDEVPRIVATE &&
4612 cmd <= SIOCDEVPRIVATE + 15)) {
4613 dev_load(net, ifr.ifr_name);
4615 ret = dev_ifsioc(net, &ifr, cmd);
4617 if (!ret && copy_to_user(arg, &ifr,
4618 sizeof(struct ifreq)))
4622 /* Take care of Wireless Extensions */
4623 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4624 return wext_handle_ioctl(net, &ifr, cmd, arg);
4631 * dev_new_index - allocate an ifindex
4632 * @net: the applicable net namespace
4634 * Returns a suitable unique value for a new device interface
4635 * number. The caller must hold the rtnl semaphore or the
4636 * dev_base_lock to be sure it remains unique.
4638 static int dev_new_index(struct net *net)
4644 if (!__dev_get_by_index(net, ifindex))
4649 /* Delayed registration/unregisteration */
4650 static LIST_HEAD(net_todo_list);
4652 static void net_set_todo(struct net_device *dev)
4654 list_add_tail(&dev->todo_list, &net_todo_list);
4657 static void rollback_registered(struct net_device *dev)
4659 BUG_ON(dev_boot_phase);
4662 /* Some devices call without registering for initialization unwind. */
4663 if (dev->reg_state == NETREG_UNINITIALIZED) {
4664 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
4665 "was registered\n", dev->name, dev);
4671 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4673 /* If device is running, close it first. */
4676 /* And unlink it from device chain. */
4677 unlist_netdevice(dev);
4679 dev->reg_state = NETREG_UNREGISTERING;
4683 /* Shutdown queueing discipline. */
4687 /* Notify protocols, that we are about to destroy
4688 this device. They should clean all the things.
4690 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4693 * Flush the unicast and multicast chains
4695 dev_unicast_flush(dev);
4696 dev_addr_discard(dev);
4698 if (dev->netdev_ops->ndo_uninit)
4699 dev->netdev_ops->ndo_uninit(dev);
4701 /* Notifier chain MUST detach us from master device. */
4702 WARN_ON(dev->master);
4704 /* Remove entries from kobject tree */
4705 netdev_unregister_kobject(dev);
4712 static void __netdev_init_queue_locks_one(struct net_device *dev,
4713 struct netdev_queue *dev_queue,
4716 spin_lock_init(&dev_queue->_xmit_lock);
4717 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4718 dev_queue->xmit_lock_owner = -1;
4721 static void netdev_init_queue_locks(struct net_device *dev)
4723 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4724 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4727 unsigned long netdev_fix_features(unsigned long features, const char *name)
4729 /* Fix illegal SG+CSUM combinations. */
4730 if ((features & NETIF_F_SG) &&
4731 !(features & NETIF_F_ALL_CSUM)) {
4733 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4734 "checksum feature.\n", name);
4735 features &= ~NETIF_F_SG;
4738 /* TSO requires that SG is present as well. */
4739 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4741 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4742 "SG feature.\n", name);
4743 features &= ~NETIF_F_TSO;
4746 if (features & NETIF_F_UFO) {
4747 if (!(features & NETIF_F_GEN_CSUM)) {
4749 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4750 "since no NETIF_F_HW_CSUM feature.\n",
4752 features &= ~NETIF_F_UFO;
4755 if (!(features & NETIF_F_SG)) {
4757 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4758 "since no NETIF_F_SG feature.\n", name);
4759 features &= ~NETIF_F_UFO;
4765 EXPORT_SYMBOL(netdev_fix_features);
4768 * register_netdevice - register a network device
4769 * @dev: device to register
4771 * Take a completed network device structure and add it to the kernel
4772 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4773 * chain. 0 is returned on success. A negative errno code is returned
4774 * on a failure to set up the device, or if the name is a duplicate.
4776 * Callers must hold the rtnl semaphore. You may want
4777 * register_netdev() instead of this.
4780 * The locking appears insufficient to guarantee two parallel registers
4781 * will not get the same name.
4784 int register_netdevice(struct net_device *dev)
4786 struct hlist_head *head;
4787 struct hlist_node *p;
4789 struct net *net = dev_net(dev);
4791 BUG_ON(dev_boot_phase);
4796 /* When net_device's are persistent, this will be fatal. */
4797 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4800 spin_lock_init(&dev->addr_list_lock);
4801 netdev_set_addr_lockdep_class(dev);
4802 netdev_init_queue_locks(dev);
4806 /* Init, if this function is available */
4807 if (dev->netdev_ops->ndo_init) {
4808 ret = dev->netdev_ops->ndo_init(dev);
4816 if (!dev_valid_name(dev->name)) {
4821 dev->ifindex = dev_new_index(net);
4822 if (dev->iflink == -1)
4823 dev->iflink = dev->ifindex;
4825 /* Check for existence of name */
4826 head = dev_name_hash(net, dev->name);
4827 hlist_for_each(p, head) {
4828 struct net_device *d
4829 = hlist_entry(p, struct net_device, name_hlist);
4830 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4836 /* Fix illegal checksum combinations */
4837 if ((dev->features & NETIF_F_HW_CSUM) &&
4838 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4839 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4841 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4844 if ((dev->features & NETIF_F_NO_CSUM) &&
4845 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4846 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4848 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4851 dev->features = netdev_fix_features(dev->features, dev->name);
4853 /* Enable software GSO if SG is supported. */
4854 if (dev->features & NETIF_F_SG)
4855 dev->features |= NETIF_F_GSO;
4857 netdev_initialize_kobject(dev);
4858 ret = netdev_register_kobject(dev);
4861 dev->reg_state = NETREG_REGISTERED;
4864 * Default initial state at registry is that the
4865 * device is present.
4868 set_bit(__LINK_STATE_PRESENT, &dev->state);
4870 dev_init_scheduler(dev);
4872 list_netdevice(dev);
4874 /* Notify protocols, that a new device appeared. */
4875 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4876 ret = notifier_to_errno(ret);
4878 rollback_registered(dev);
4879 dev->reg_state = NETREG_UNREGISTERED;
4882 * Prevent userspace races by waiting until the network
4883 * device is fully setup before sending notifications.
4885 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
4891 if (dev->netdev_ops->ndo_uninit)
4892 dev->netdev_ops->ndo_uninit(dev);
4895 EXPORT_SYMBOL(register_netdevice);
4898 * init_dummy_netdev - init a dummy network device for NAPI
4899 * @dev: device to init
4901 * This takes a network device structure and initialize the minimum
4902 * amount of fields so it can be used to schedule NAPI polls without
4903 * registering a full blown interface. This is to be used by drivers
4904 * that need to tie several hardware interfaces to a single NAPI
4905 * poll scheduler due to HW limitations.
4907 int init_dummy_netdev(struct net_device *dev)
4909 /* Clear everything. Note we don't initialize spinlocks
4910 * are they aren't supposed to be taken by any of the
4911 * NAPI code and this dummy netdev is supposed to be
4912 * only ever used for NAPI polls
4914 memset(dev, 0, sizeof(struct net_device));
4916 /* make sure we BUG if trying to hit standard
4917 * register/unregister code path
4919 dev->reg_state = NETREG_DUMMY;
4921 /* initialize the ref count */
4922 atomic_set(&dev->refcnt, 1);
4924 /* NAPI wants this */
4925 INIT_LIST_HEAD(&dev->napi_list);
4927 /* a dummy interface is started by default */
4928 set_bit(__LINK_STATE_PRESENT, &dev->state);
4929 set_bit(__LINK_STATE_START, &dev->state);
4933 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4937 * register_netdev - register a network device
4938 * @dev: device to register
4940 * Take a completed network device structure and add it to the kernel
4941 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4942 * chain. 0 is returned on success. A negative errno code is returned
4943 * on a failure to set up the device, or if the name is a duplicate.
4945 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4946 * and expands the device name if you passed a format string to
4949 int register_netdev(struct net_device *dev)
4956 * If the name is a format string the caller wants us to do a
4959 if (strchr(dev->name, '%')) {
4960 err = dev_alloc_name(dev, dev->name);
4965 err = register_netdevice(dev);
4970 EXPORT_SYMBOL(register_netdev);
4973 * netdev_wait_allrefs - wait until all references are gone.
4975 * This is called when unregistering network devices.
4977 * Any protocol or device that holds a reference should register
4978 * for netdevice notification, and cleanup and put back the
4979 * reference if they receive an UNREGISTER event.
4980 * We can get stuck here if buggy protocols don't correctly
4983 static void netdev_wait_allrefs(struct net_device *dev)
4985 unsigned long rebroadcast_time, warning_time;
4987 rebroadcast_time = warning_time = jiffies;
4988 while (atomic_read(&dev->refcnt) != 0) {
4989 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4992 /* Rebroadcast unregister notification */
4993 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4995 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4997 /* We must not have linkwatch events
4998 * pending on unregister. If this
4999 * happens, we simply run the queue
5000 * unscheduled, resulting in a noop
5003 linkwatch_run_queue();
5008 rebroadcast_time = jiffies;
5013 if (time_after(jiffies, warning_time + 10 * HZ)) {
5014 printk(KERN_EMERG "unregister_netdevice: "
5015 "waiting for %s to become free. Usage "
5017 dev->name, atomic_read(&dev->refcnt));
5018 warning_time = jiffies;
5027 * register_netdevice(x1);
5028 * register_netdevice(x2);
5030 * unregister_netdevice(y1);
5031 * unregister_netdevice(y2);
5037 * We are invoked by rtnl_unlock().
5038 * This allows us to deal with problems:
5039 * 1) We can delete sysfs objects which invoke hotplug
5040 * without deadlocking with linkwatch via keventd.
5041 * 2) Since we run with the RTNL semaphore not held, we can sleep
5042 * safely in order to wait for the netdev refcnt to drop to zero.
5044 * We must not return until all unregister events added during
5045 * the interval the lock was held have been completed.
5047 void netdev_run_todo(void)
5049 struct list_head list;
5051 /* Snapshot list, allow later requests */
5052 list_replace_init(&net_todo_list, &list);
5056 while (!list_empty(&list)) {
5057 struct net_device *dev
5058 = list_entry(list.next, struct net_device, todo_list);
5059 list_del(&dev->todo_list);
5061 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5062 printk(KERN_ERR "network todo '%s' but state %d\n",
5063 dev->name, dev->reg_state);
5068 dev->reg_state = NETREG_UNREGISTERED;
5070 on_each_cpu(flush_backlog, dev, 1);
5072 netdev_wait_allrefs(dev);
5075 BUG_ON(atomic_read(&dev->refcnt));
5076 WARN_ON(dev->ip_ptr);
5077 WARN_ON(dev->ip6_ptr);
5078 WARN_ON(dev->dn_ptr);
5080 if (dev->destructor)
5081 dev->destructor(dev);
5083 /* Free network device */
5084 kobject_put(&dev->dev.kobj);
5089 * dev_get_stats - get network device statistics
5090 * @dev: device to get statistics from
5092 * Get network statistics from device. The device driver may provide
5093 * its own method by setting dev->netdev_ops->get_stats; otherwise
5094 * the internal statistics structure is used.
5096 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5098 const struct net_device_ops *ops = dev->netdev_ops;
5100 if (ops->ndo_get_stats)
5101 return ops->ndo_get_stats(dev);
5103 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5104 struct net_device_stats *stats = &dev->stats;
5106 struct netdev_queue *txq;
5108 for (i = 0; i < dev->num_tx_queues; i++) {
5109 txq = netdev_get_tx_queue(dev, i);
5110 tx_bytes += txq->tx_bytes;
5111 tx_packets += txq->tx_packets;
5112 tx_dropped += txq->tx_dropped;
5114 if (tx_bytes || tx_packets || tx_dropped) {
5115 stats->tx_bytes = tx_bytes;
5116 stats->tx_packets = tx_packets;
5117 stats->tx_dropped = tx_dropped;
5122 EXPORT_SYMBOL(dev_get_stats);
5124 static void netdev_init_one_queue(struct net_device *dev,
5125 struct netdev_queue *queue,
5131 static void netdev_init_queues(struct net_device *dev)
5133 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5134 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5135 spin_lock_init(&dev->tx_global_lock);
5139 * alloc_netdev_mq - allocate network device
5140 * @sizeof_priv: size of private data to allocate space for
5141 * @name: device name format string
5142 * @setup: callback to initialize device
5143 * @queue_count: the number of subqueues to allocate
5145 * Allocates a struct net_device with private data area for driver use
5146 * and performs basic initialization. Also allocates subquue structs
5147 * for each queue on the device at the end of the netdevice.
5149 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5150 void (*setup)(struct net_device *), unsigned int queue_count)
5152 struct netdev_queue *tx;
5153 struct net_device *dev;
5155 struct net_device *p;
5157 BUG_ON(strlen(name) >= sizeof(dev->name));
5159 alloc_size = sizeof(struct net_device);
5161 /* ensure 32-byte alignment of private area */
5162 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5163 alloc_size += sizeof_priv;
5165 /* ensure 32-byte alignment of whole construct */
5166 alloc_size += NETDEV_ALIGN - 1;
5168 p = kzalloc(alloc_size, GFP_KERNEL);
5170 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5174 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5176 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5181 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5182 dev->padded = (char *)dev - (char *)p;
5184 if (dev_addr_init(dev))
5187 dev_unicast_init(dev);
5189 dev_net_set(dev, &init_net);
5192 dev->num_tx_queues = queue_count;
5193 dev->real_num_tx_queues = queue_count;
5195 dev->gso_max_size = GSO_MAX_SIZE;
5197 netdev_init_queues(dev);
5199 INIT_LIST_HEAD(&dev->napi_list);
5200 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5202 strcpy(dev->name, name);
5212 EXPORT_SYMBOL(alloc_netdev_mq);
5215 * free_netdev - free network device
5218 * This function does the last stage of destroying an allocated device
5219 * interface. The reference to the device object is released.
5220 * If this is the last reference then it will be freed.
5222 void free_netdev(struct net_device *dev)
5224 struct napi_struct *p, *n;
5226 release_net(dev_net(dev));
5230 /* Flush device addresses */
5231 dev_addr_flush(dev);
5233 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5236 /* Compatibility with error handling in drivers */
5237 if (dev->reg_state == NETREG_UNINITIALIZED) {
5238 kfree((char *)dev - dev->padded);
5242 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5243 dev->reg_state = NETREG_RELEASED;
5245 /* will free via device release */
5246 put_device(&dev->dev);
5248 EXPORT_SYMBOL(free_netdev);
5251 * synchronize_net - Synchronize with packet receive processing
5253 * Wait for packets currently being received to be done.
5254 * Does not block later packets from starting.
5256 void synchronize_net(void)
5261 EXPORT_SYMBOL(synchronize_net);
5264 * unregister_netdevice - remove device from the kernel
5267 * This function shuts down a device interface and removes it
5268 * from the kernel tables.
5270 * Callers must hold the rtnl semaphore. You may want
5271 * unregister_netdev() instead of this.
5274 void unregister_netdevice(struct net_device *dev)
5278 rollback_registered(dev);
5279 /* Finish processing unregister after unlock */
5282 EXPORT_SYMBOL(unregister_netdevice);
5285 * unregister_netdev - remove device from the kernel
5288 * This function shuts down a device interface and removes it
5289 * from the kernel tables.
5291 * This is just a wrapper for unregister_netdevice that takes
5292 * the rtnl semaphore. In general you want to use this and not
5293 * unregister_netdevice.
5295 void unregister_netdev(struct net_device *dev)
5298 unregister_netdevice(dev);
5301 EXPORT_SYMBOL(unregister_netdev);
5304 * dev_change_net_namespace - move device to different nethost namespace
5306 * @net: network namespace
5307 * @pat: If not NULL name pattern to try if the current device name
5308 * is already taken in the destination network namespace.
5310 * This function shuts down a device interface and moves it
5311 * to a new network namespace. On success 0 is returned, on
5312 * a failure a netagive errno code is returned.
5314 * Callers must hold the rtnl semaphore.
5317 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5320 const char *destname;
5325 /* Don't allow namespace local devices to be moved. */
5327 if (dev->features & NETIF_F_NETNS_LOCAL)
5331 /* Don't allow real devices to be moved when sysfs
5335 if (dev->dev.parent)
5339 /* Ensure the device has been registrered */
5341 if (dev->reg_state != NETREG_REGISTERED)
5344 /* Get out if there is nothing todo */
5346 if (net_eq(dev_net(dev), net))
5349 /* Pick the destination device name, and ensure
5350 * we can use it in the destination network namespace.
5353 destname = dev->name;
5354 if (__dev_get_by_name(net, destname)) {
5355 /* We get here if we can't use the current device name */
5358 if (!dev_valid_name(pat))
5360 if (strchr(pat, '%')) {
5361 if (__dev_alloc_name(net, pat, buf) < 0)
5366 if (__dev_get_by_name(net, destname))
5371 * And now a mini version of register_netdevice unregister_netdevice.
5374 /* If device is running close it first. */
5377 /* And unlink it from device chain */
5379 unlist_netdevice(dev);
5383 /* Shutdown queueing discipline. */
5386 /* Notify protocols, that we are about to destroy
5387 this device. They should clean all the things.
5389 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5392 * Flush the unicast and multicast chains
5394 dev_unicast_flush(dev);
5395 dev_addr_discard(dev);
5397 netdev_unregister_kobject(dev);
5399 /* Actually switch the network namespace */
5400 dev_net_set(dev, net);
5402 /* Assign the new device name */
5403 if (destname != dev->name)
5404 strcpy(dev->name, destname);
5406 /* If there is an ifindex conflict assign a new one */
5407 if (__dev_get_by_index(net, dev->ifindex)) {
5408 int iflink = (dev->iflink == dev->ifindex);
5409 dev->ifindex = dev_new_index(net);
5411 dev->iflink = dev->ifindex;
5414 /* Fixup kobjects */
5415 err = netdev_register_kobject(dev);
5418 /* Add the device back in the hashes */
5419 list_netdevice(dev);
5421 /* Notify protocols, that a new device appeared. */
5422 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5425 * Prevent userspace races by waiting until the network
5426 * device is fully setup before sending notifications.
5428 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5435 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5437 static int dev_cpu_callback(struct notifier_block *nfb,
5438 unsigned long action,
5441 struct sk_buff **list_skb;
5442 struct Qdisc **list_net;
5443 struct sk_buff *skb;
5444 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5445 struct softnet_data *sd, *oldsd;
5447 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5450 local_irq_disable();
5451 cpu = smp_processor_id();
5452 sd = &per_cpu(softnet_data, cpu);
5453 oldsd = &per_cpu(softnet_data, oldcpu);
5455 /* Find end of our completion_queue. */
5456 list_skb = &sd->completion_queue;
5458 list_skb = &(*list_skb)->next;
5459 /* Append completion queue from offline CPU. */
5460 *list_skb = oldsd->completion_queue;
5461 oldsd->completion_queue = NULL;
5463 /* Find end of our output_queue. */
5464 list_net = &sd->output_queue;
5466 list_net = &(*list_net)->next_sched;
5467 /* Append output queue from offline CPU. */
5468 *list_net = oldsd->output_queue;
5469 oldsd->output_queue = NULL;
5471 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5474 /* Process offline CPU's input_pkt_queue */
5475 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5483 * netdev_increment_features - increment feature set by one
5484 * @all: current feature set
5485 * @one: new feature set
5486 * @mask: mask feature set
5488 * Computes a new feature set after adding a device with feature set
5489 * @one to the master device with current feature set @all. Will not
5490 * enable anything that is off in @mask. Returns the new feature set.
5492 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5495 /* If device needs checksumming, downgrade to it. */
5496 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5497 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5498 else if (mask & NETIF_F_ALL_CSUM) {
5499 /* If one device supports v4/v6 checksumming, set for all. */
5500 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5501 !(all & NETIF_F_GEN_CSUM)) {
5502 all &= ~NETIF_F_ALL_CSUM;
5503 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5506 /* If one device supports hw checksumming, set for all. */
5507 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5508 all &= ~NETIF_F_ALL_CSUM;
5509 all |= NETIF_F_HW_CSUM;
5513 one |= NETIF_F_ALL_CSUM;
5515 one |= all & NETIF_F_ONE_FOR_ALL;
5516 all &= one | NETIF_F_LLTX | NETIF_F_GSO;
5517 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5521 EXPORT_SYMBOL(netdev_increment_features);
5523 static struct hlist_head *netdev_create_hash(void)
5526 struct hlist_head *hash;
5528 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5530 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5531 INIT_HLIST_HEAD(&hash[i]);
5536 /* Initialize per network namespace state */
5537 static int __net_init netdev_init(struct net *net)
5539 INIT_LIST_HEAD(&net->dev_base_head);
5541 net->dev_name_head = netdev_create_hash();
5542 if (net->dev_name_head == NULL)
5545 net->dev_index_head = netdev_create_hash();
5546 if (net->dev_index_head == NULL)
5552 kfree(net->dev_name_head);
5558 * netdev_drivername - network driver for the device
5559 * @dev: network device
5560 * @buffer: buffer for resulting name
5561 * @len: size of buffer
5563 * Determine network driver for device.
5565 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5567 const struct device_driver *driver;
5568 const struct device *parent;
5570 if (len <= 0 || !buffer)
5574 parent = dev->dev.parent;
5579 driver = parent->driver;
5580 if (driver && driver->name)
5581 strlcpy(buffer, driver->name, len);
5585 static void __net_exit netdev_exit(struct net *net)
5587 kfree(net->dev_name_head);
5588 kfree(net->dev_index_head);
5591 static struct pernet_operations __net_initdata netdev_net_ops = {
5592 .init = netdev_init,
5593 .exit = netdev_exit,
5596 static void __net_exit default_device_exit(struct net *net)
5598 struct net_device *dev;
5600 * Push all migratable of the network devices back to the
5601 * initial network namespace
5605 for_each_netdev(net, dev) {
5607 char fb_name[IFNAMSIZ];
5609 /* Ignore unmoveable devices (i.e. loopback) */
5610 if (dev->features & NETIF_F_NETNS_LOCAL)
5613 /* Delete virtual devices */
5614 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5615 dev->rtnl_link_ops->dellink(dev);
5619 /* Push remaing network devices to init_net */
5620 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5621 err = dev_change_net_namespace(dev, &init_net, fb_name);
5623 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5624 __func__, dev->name, err);
5632 static struct pernet_operations __net_initdata default_device_ops = {
5633 .exit = default_device_exit,
5637 * Initialize the DEV module. At boot time this walks the device list and
5638 * unhooks any devices that fail to initialise (normally hardware not
5639 * present) and leaves us with a valid list of present and active devices.
5644 * This is called single threaded during boot, so no need
5645 * to take the rtnl semaphore.
5647 static int __init net_dev_init(void)
5649 int i, rc = -ENOMEM;
5651 BUG_ON(!dev_boot_phase);
5653 if (dev_proc_init())
5656 if (netdev_kobject_init())
5659 INIT_LIST_HEAD(&ptype_all);
5660 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5661 INIT_LIST_HEAD(&ptype_base[i]);
5663 if (register_pernet_subsys(&netdev_net_ops))
5667 * Initialise the packet receive queues.
5670 for_each_possible_cpu(i) {
5671 struct softnet_data *queue;
5673 queue = &per_cpu(softnet_data, i);
5674 skb_queue_head_init(&queue->input_pkt_queue);
5675 queue->completion_queue = NULL;
5676 INIT_LIST_HEAD(&queue->poll_list);
5678 queue->backlog.poll = process_backlog;
5679 queue->backlog.weight = weight_p;
5680 queue->backlog.gro_list = NULL;
5681 queue->backlog.gro_count = 0;
5686 /* The loopback device is special if any other network devices
5687 * is present in a network namespace the loopback device must
5688 * be present. Since we now dynamically allocate and free the
5689 * loopback device ensure this invariant is maintained by
5690 * keeping the loopback device as the first device on the
5691 * list of network devices. Ensuring the loopback devices
5692 * is the first device that appears and the last network device
5695 if (register_pernet_device(&loopback_net_ops))
5698 if (register_pernet_device(&default_device_ops))
5701 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5702 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5704 hotcpu_notifier(dev_cpu_callback, 0);
5712 subsys_initcall(net_dev_init);
5714 static int __init initialize_hashrnd(void)
5716 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5720 late_initcall_sync(initialize_hashrnd);