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);
947 "%s: name change rollback failed: %d.\n",
951 memcpy(dev->name, oldname, IFNAMSIZ);
960 * dev_set_alias - change ifalias of a device
962 * @alias: name up to IFALIASZ
963 * @len: limit of bytes to copy from info
965 * Set ifalias for a device,
967 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
982 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
986 strlcpy(dev->ifalias, alias, len+1);
992 * netdev_features_change - device changes features
993 * @dev: device to cause notification
995 * Called to indicate a device has changed features.
997 void netdev_features_change(struct net_device *dev)
999 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1001 EXPORT_SYMBOL(netdev_features_change);
1004 * netdev_state_change - device changes state
1005 * @dev: device to cause notification
1007 * Called to indicate a device has changed state. This function calls
1008 * the notifier chains for netdev_chain and sends a NEWLINK message
1009 * to the routing socket.
1011 void netdev_state_change(struct net_device *dev)
1013 if (dev->flags & IFF_UP) {
1014 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1015 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1018 EXPORT_SYMBOL(netdev_state_change);
1020 void netdev_bonding_change(struct net_device *dev, unsigned long event)
1022 call_netdevice_notifiers(event, dev);
1024 EXPORT_SYMBOL(netdev_bonding_change);
1027 * dev_load - load a network module
1028 * @net: the applicable net namespace
1029 * @name: name of interface
1031 * If a network interface is not present and the process has suitable
1032 * privileges this function loads the module. If module loading is not
1033 * available in this kernel then it becomes a nop.
1036 void dev_load(struct net *net, const char *name)
1038 struct net_device *dev;
1040 read_lock(&dev_base_lock);
1041 dev = __dev_get_by_name(net, name);
1042 read_unlock(&dev_base_lock);
1044 if (!dev && capable(CAP_NET_ADMIN))
1045 request_module("%s", name);
1047 EXPORT_SYMBOL(dev_load);
1050 * dev_open - prepare an interface for use.
1051 * @dev: device to open
1053 * Takes a device from down to up state. The device's private open
1054 * function is invoked and then the multicast lists are loaded. Finally
1055 * the device is moved into the up state and a %NETDEV_UP message is
1056 * sent to the netdev notifier chain.
1058 * Calling this function on an active interface is a nop. On a failure
1059 * a negative errno code is returned.
1061 int dev_open(struct net_device *dev)
1063 const struct net_device_ops *ops = dev->netdev_ops;
1072 if (dev->flags & IFF_UP)
1076 * Is it even present?
1078 if (!netif_device_present(dev))
1081 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1082 ret = notifier_to_errno(ret);
1087 * Call device private open method
1089 set_bit(__LINK_STATE_START, &dev->state);
1091 if (ops->ndo_validate_addr)
1092 ret = ops->ndo_validate_addr(dev);
1094 if (!ret && ops->ndo_open)
1095 ret = ops->ndo_open(dev);
1098 * If it went open OK then:
1102 clear_bit(__LINK_STATE_START, &dev->state);
1107 dev->flags |= IFF_UP;
1112 net_dmaengine_get();
1115 * Initialize multicasting status
1117 dev_set_rx_mode(dev);
1120 * Wakeup transmit queue engine
1125 * ... and announce new interface.
1127 call_netdevice_notifiers(NETDEV_UP, dev);
1132 EXPORT_SYMBOL(dev_open);
1135 * dev_close - shutdown an interface.
1136 * @dev: device to shutdown
1138 * This function moves an active device into down state. A
1139 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1140 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1143 int dev_close(struct net_device *dev)
1145 const struct net_device_ops *ops = dev->netdev_ops;
1150 if (!(dev->flags & IFF_UP))
1154 * Tell people we are going down, so that they can
1155 * prepare to death, when device is still operating.
1157 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1159 clear_bit(__LINK_STATE_START, &dev->state);
1161 /* Synchronize to scheduled poll. We cannot touch poll list,
1162 * it can be even on different cpu. So just clear netif_running().
1164 * dev->stop() will invoke napi_disable() on all of it's
1165 * napi_struct instances on this device.
1167 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1169 dev_deactivate(dev);
1172 * Call the device specific close. This cannot fail.
1173 * Only if device is UP
1175 * We allow it to be called even after a DETACH hot-plug
1182 * Device is now down.
1185 dev->flags &= ~IFF_UP;
1188 * Tell people we are down
1190 call_netdevice_notifiers(NETDEV_DOWN, dev);
1195 net_dmaengine_put();
1199 EXPORT_SYMBOL(dev_close);
1203 * dev_disable_lro - disable Large Receive Offload on a device
1206 * Disable Large Receive Offload (LRO) on a net device. Must be
1207 * called under RTNL. This is needed if received packets may be
1208 * forwarded to another interface.
1210 void dev_disable_lro(struct net_device *dev)
1212 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1213 dev->ethtool_ops->set_flags) {
1214 u32 flags = dev->ethtool_ops->get_flags(dev);
1215 if (flags & ETH_FLAG_LRO) {
1216 flags &= ~ETH_FLAG_LRO;
1217 dev->ethtool_ops->set_flags(dev, flags);
1220 WARN_ON(dev->features & NETIF_F_LRO);
1222 EXPORT_SYMBOL(dev_disable_lro);
1225 static int dev_boot_phase = 1;
1228 * Device change register/unregister. These are not inline or static
1229 * as we export them to the world.
1233 * register_netdevice_notifier - register a network notifier block
1236 * Register a notifier to be called when network device events occur.
1237 * The notifier passed is linked into the kernel structures and must
1238 * not be reused until it has been unregistered. A negative errno code
1239 * is returned on a failure.
1241 * When registered all registration and up events are replayed
1242 * to the new notifier to allow device to have a race free
1243 * view of the network device list.
1246 int register_netdevice_notifier(struct notifier_block *nb)
1248 struct net_device *dev;
1249 struct net_device *last;
1254 err = raw_notifier_chain_register(&netdev_chain, nb);
1260 for_each_netdev(net, dev) {
1261 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1262 err = notifier_to_errno(err);
1266 if (!(dev->flags & IFF_UP))
1269 nb->notifier_call(nb, NETDEV_UP, dev);
1280 for_each_netdev(net, dev) {
1284 if (dev->flags & IFF_UP) {
1285 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1286 nb->notifier_call(nb, NETDEV_DOWN, dev);
1288 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1292 raw_notifier_chain_unregister(&netdev_chain, nb);
1295 EXPORT_SYMBOL(register_netdevice_notifier);
1298 * unregister_netdevice_notifier - unregister a network notifier block
1301 * Unregister a notifier previously registered by
1302 * register_netdevice_notifier(). The notifier is unlinked into the
1303 * kernel structures and may then be reused. A negative errno code
1304 * is returned on a failure.
1307 int unregister_netdevice_notifier(struct notifier_block *nb)
1312 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1316 EXPORT_SYMBOL(unregister_netdevice_notifier);
1319 * call_netdevice_notifiers - call all network notifier blocks
1320 * @val: value passed unmodified to notifier function
1321 * @dev: net_device pointer passed unmodified to notifier function
1323 * Call all network notifier blocks. Parameters and return value
1324 * are as for raw_notifier_call_chain().
1327 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1329 return raw_notifier_call_chain(&netdev_chain, val, dev);
1332 /* When > 0 there are consumers of rx skb time stamps */
1333 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1335 void net_enable_timestamp(void)
1337 atomic_inc(&netstamp_needed);
1339 EXPORT_SYMBOL(net_enable_timestamp);
1341 void net_disable_timestamp(void)
1343 atomic_dec(&netstamp_needed);
1345 EXPORT_SYMBOL(net_disable_timestamp);
1347 static inline void net_timestamp(struct sk_buff *skb)
1349 if (atomic_read(&netstamp_needed))
1350 __net_timestamp(skb);
1352 skb->tstamp.tv64 = 0;
1356 * Support routine. Sends outgoing frames to any network
1357 * taps currently in use.
1360 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1362 struct packet_type *ptype;
1364 #ifdef CONFIG_NET_CLS_ACT
1365 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1372 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1373 /* Never send packets back to the socket
1374 * they originated from - MvS (miquels@drinkel.ow.org)
1376 if ((ptype->dev == dev || !ptype->dev) &&
1377 (ptype->af_packet_priv == NULL ||
1378 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1379 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1383 /* skb->nh should be correctly
1384 set by sender, so that the second statement is
1385 just protection against buggy protocols.
1387 skb_reset_mac_header(skb2);
1389 if (skb_network_header(skb2) < skb2->data ||
1390 skb2->network_header > skb2->tail) {
1391 if (net_ratelimit())
1392 printk(KERN_CRIT "protocol %04x is "
1394 skb2->protocol, dev->name);
1395 skb_reset_network_header(skb2);
1398 skb2->transport_header = skb2->network_header;
1399 skb2->pkt_type = PACKET_OUTGOING;
1400 ptype->func(skb2, skb->dev, ptype, skb->dev);
1407 static inline void __netif_reschedule(struct Qdisc *q)
1409 struct softnet_data *sd;
1410 unsigned long flags;
1412 local_irq_save(flags);
1413 sd = &__get_cpu_var(softnet_data);
1414 q->next_sched = sd->output_queue;
1415 sd->output_queue = q;
1416 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1417 local_irq_restore(flags);
1420 void __netif_schedule(struct Qdisc *q)
1422 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1423 __netif_reschedule(q);
1425 EXPORT_SYMBOL(__netif_schedule);
1427 void dev_kfree_skb_irq(struct sk_buff *skb)
1429 if (atomic_dec_and_test(&skb->users)) {
1430 struct softnet_data *sd;
1431 unsigned long flags;
1433 local_irq_save(flags);
1434 sd = &__get_cpu_var(softnet_data);
1435 skb->next = sd->completion_queue;
1436 sd->completion_queue = skb;
1437 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1438 local_irq_restore(flags);
1441 EXPORT_SYMBOL(dev_kfree_skb_irq);
1443 void dev_kfree_skb_any(struct sk_buff *skb)
1445 if (in_irq() || irqs_disabled())
1446 dev_kfree_skb_irq(skb);
1450 EXPORT_SYMBOL(dev_kfree_skb_any);
1454 * netif_device_detach - mark device as removed
1455 * @dev: network device
1457 * Mark device as removed from system and therefore no longer available.
1459 void netif_device_detach(struct net_device *dev)
1461 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1462 netif_running(dev)) {
1463 netif_tx_stop_all_queues(dev);
1466 EXPORT_SYMBOL(netif_device_detach);
1469 * netif_device_attach - mark device as attached
1470 * @dev: network device
1472 * Mark device as attached from system and restart if needed.
1474 void netif_device_attach(struct net_device *dev)
1476 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1477 netif_running(dev)) {
1478 netif_tx_wake_all_queues(dev);
1479 __netdev_watchdog_up(dev);
1482 EXPORT_SYMBOL(netif_device_attach);
1484 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1486 return ((features & NETIF_F_GEN_CSUM) ||
1487 ((features & NETIF_F_IP_CSUM) &&
1488 protocol == htons(ETH_P_IP)) ||
1489 ((features & NETIF_F_IPV6_CSUM) &&
1490 protocol == htons(ETH_P_IPV6)) ||
1491 ((features & NETIF_F_FCOE_CRC) &&
1492 protocol == htons(ETH_P_FCOE)));
1495 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1497 if (can_checksum_protocol(dev->features, skb->protocol))
1500 if (skb->protocol == htons(ETH_P_8021Q)) {
1501 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1502 if (can_checksum_protocol(dev->features & dev->vlan_features,
1503 veh->h_vlan_encapsulated_proto))
1511 * Invalidate hardware checksum when packet is to be mangled, and
1512 * complete checksum manually on outgoing path.
1514 int skb_checksum_help(struct sk_buff *skb)
1517 int ret = 0, offset;
1519 if (skb->ip_summed == CHECKSUM_COMPLETE)
1520 goto out_set_summed;
1522 if (unlikely(skb_shinfo(skb)->gso_size)) {
1523 /* Let GSO fix up the checksum. */
1524 goto out_set_summed;
1527 offset = skb->csum_start - skb_headroom(skb);
1528 BUG_ON(offset >= skb_headlen(skb));
1529 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1531 offset += skb->csum_offset;
1532 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1534 if (skb_cloned(skb) &&
1535 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1536 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1541 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1543 skb->ip_summed = CHECKSUM_NONE;
1547 EXPORT_SYMBOL(skb_checksum_help);
1550 * skb_gso_segment - Perform segmentation on skb.
1551 * @skb: buffer to segment
1552 * @features: features for the output path (see dev->features)
1554 * This function segments the given skb and returns a list of segments.
1556 * It may return NULL if the skb requires no segmentation. This is
1557 * only possible when GSO is used for verifying header integrity.
1559 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1561 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1562 struct packet_type *ptype;
1563 __be16 type = skb->protocol;
1566 skb_reset_mac_header(skb);
1567 skb->mac_len = skb->network_header - skb->mac_header;
1568 __skb_pull(skb, skb->mac_len);
1570 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1571 struct net_device *dev = skb->dev;
1572 struct ethtool_drvinfo info = {};
1574 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1575 dev->ethtool_ops->get_drvinfo(dev, &info);
1577 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1579 info.driver, dev ? dev->features : 0L,
1580 skb->sk ? skb->sk->sk_route_caps : 0L,
1581 skb->len, skb->data_len, skb->ip_summed);
1583 if (skb_header_cloned(skb) &&
1584 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1585 return ERR_PTR(err);
1589 list_for_each_entry_rcu(ptype,
1590 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1591 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1592 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1593 err = ptype->gso_send_check(skb);
1594 segs = ERR_PTR(err);
1595 if (err || skb_gso_ok(skb, features))
1597 __skb_push(skb, (skb->data -
1598 skb_network_header(skb)));
1600 segs = ptype->gso_segment(skb, features);
1606 __skb_push(skb, skb->data - skb_mac_header(skb));
1610 EXPORT_SYMBOL(skb_gso_segment);
1612 /* Take action when hardware reception checksum errors are detected. */
1614 void netdev_rx_csum_fault(struct net_device *dev)
1616 if (net_ratelimit()) {
1617 printk(KERN_ERR "%s: hw csum failure.\n",
1618 dev ? dev->name : "<unknown>");
1622 EXPORT_SYMBOL(netdev_rx_csum_fault);
1625 /* Actually, we should eliminate this check as soon as we know, that:
1626 * 1. IOMMU is present and allows to map all the memory.
1627 * 2. No high memory really exists on this machine.
1630 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1632 #ifdef CONFIG_HIGHMEM
1635 if (dev->features & NETIF_F_HIGHDMA)
1638 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1639 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1647 void (*destructor)(struct sk_buff *skb);
1650 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1652 static void dev_gso_skb_destructor(struct sk_buff *skb)
1654 struct dev_gso_cb *cb;
1657 struct sk_buff *nskb = skb->next;
1659 skb->next = nskb->next;
1662 } while (skb->next);
1664 cb = DEV_GSO_CB(skb);
1666 cb->destructor(skb);
1670 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1671 * @skb: buffer to segment
1673 * This function segments the given skb and stores the list of segments
1676 static int dev_gso_segment(struct sk_buff *skb)
1678 struct net_device *dev = skb->dev;
1679 struct sk_buff *segs;
1680 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1683 segs = skb_gso_segment(skb, features);
1685 /* Verifying header integrity only. */
1690 return PTR_ERR(segs);
1693 DEV_GSO_CB(skb)->destructor = skb->destructor;
1694 skb->destructor = dev_gso_skb_destructor;
1699 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1700 struct netdev_queue *txq)
1702 const struct net_device_ops *ops = dev->netdev_ops;
1705 if (likely(!skb->next)) {
1706 if (!list_empty(&ptype_all))
1707 dev_queue_xmit_nit(skb, dev);
1709 if (netif_needs_gso(dev, skb)) {
1710 if (unlikely(dev_gso_segment(skb)))
1717 * If device doesnt need skb->dst, release it right now while
1718 * its hot in this cpu cache
1720 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1723 rc = ops->ndo_start_xmit(skb, dev);
1724 if (rc == NETDEV_TX_OK)
1725 txq_trans_update(txq);
1727 * TODO: if skb_orphan() was called by
1728 * dev->hard_start_xmit() (for example, the unmodified
1729 * igb driver does that; bnx2 doesn't), then
1730 * skb_tx_software_timestamp() will be unable to send
1731 * back the time stamp.
1733 * How can this be prevented? Always create another
1734 * reference to the socket before calling
1735 * dev->hard_start_xmit()? Prevent that skb_orphan()
1736 * does anything in dev->hard_start_xmit() by clearing
1737 * the skb destructor before the call and restoring it
1738 * afterwards, then doing the skb_orphan() ourselves?
1745 struct sk_buff *nskb = skb->next;
1747 skb->next = nskb->next;
1749 rc = ops->ndo_start_xmit(nskb, dev);
1750 if (unlikely(rc != NETDEV_TX_OK)) {
1751 nskb->next = skb->next;
1755 txq_trans_update(txq);
1756 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1757 return NETDEV_TX_BUSY;
1758 } while (skb->next);
1760 skb->destructor = DEV_GSO_CB(skb)->destructor;
1764 return NETDEV_TX_OK;
1767 static u32 skb_tx_hashrnd;
1769 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1773 if (skb_rx_queue_recorded(skb)) {
1774 hash = skb_get_rx_queue(skb);
1775 while (unlikely(hash >= dev->real_num_tx_queues))
1776 hash -= dev->real_num_tx_queues;
1780 if (skb->sk && skb->sk->sk_hash)
1781 hash = skb->sk->sk_hash;
1783 hash = skb->protocol;
1785 hash = jhash_1word(hash, skb_tx_hashrnd);
1787 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1789 EXPORT_SYMBOL(skb_tx_hash);
1791 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1792 struct sk_buff *skb)
1794 const struct net_device_ops *ops = dev->netdev_ops;
1795 u16 queue_index = 0;
1797 if (ops->ndo_select_queue)
1798 queue_index = ops->ndo_select_queue(dev, skb);
1799 else if (dev->real_num_tx_queues > 1)
1800 queue_index = skb_tx_hash(dev, skb);
1802 skb_set_queue_mapping(skb, queue_index);
1803 return netdev_get_tx_queue(dev, queue_index);
1806 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1807 struct net_device *dev,
1808 struct netdev_queue *txq)
1810 spinlock_t *root_lock = qdisc_lock(q);
1813 spin_lock(root_lock);
1814 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1817 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1818 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1820 * This is a work-conserving queue; there are no old skbs
1821 * waiting to be sent out; and the qdisc is not running -
1822 * xmit the skb directly.
1824 __qdisc_update_bstats(q, skb->len);
1825 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1828 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1830 rc = NET_XMIT_SUCCESS;
1832 rc = qdisc_enqueue_root(skb, q);
1835 spin_unlock(root_lock);
1841 * dev_queue_xmit - transmit a buffer
1842 * @skb: buffer to transmit
1844 * Queue a buffer for transmission to a network device. The caller must
1845 * have set the device and priority and built the buffer before calling
1846 * this function. The function can be called from an interrupt.
1848 * A negative errno code is returned on a failure. A success does not
1849 * guarantee the frame will be transmitted as it may be dropped due
1850 * to congestion or traffic shaping.
1852 * -----------------------------------------------------------------------------------
1853 * I notice this method can also return errors from the queue disciplines,
1854 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1857 * Regardless of the return value, the skb is consumed, so it is currently
1858 * difficult to retry a send to this method. (You can bump the ref count
1859 * before sending to hold a reference for retry if you are careful.)
1861 * When calling this method, interrupts MUST be enabled. This is because
1862 * the BH enable code must have IRQs enabled so that it will not deadlock.
1865 int dev_queue_xmit(struct sk_buff *skb)
1867 struct net_device *dev = skb->dev;
1868 struct netdev_queue *txq;
1872 /* GSO will handle the following emulations directly. */
1873 if (netif_needs_gso(dev, skb))
1876 if (skb_has_frags(skb) &&
1877 !(dev->features & NETIF_F_FRAGLIST) &&
1878 __skb_linearize(skb))
1881 /* Fragmented skb is linearized if device does not support SG,
1882 * or if at least one of fragments is in highmem and device
1883 * does not support DMA from it.
1885 if (skb_shinfo(skb)->nr_frags &&
1886 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1887 __skb_linearize(skb))
1890 /* If packet is not checksummed and device does not support
1891 * checksumming for this protocol, complete checksumming here.
1893 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1894 skb_set_transport_header(skb, skb->csum_start -
1896 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1901 /* Disable soft irqs for various locks below. Also
1902 * stops preemption for RCU.
1906 txq = dev_pick_tx(dev, skb);
1907 q = rcu_dereference(txq->qdisc);
1909 #ifdef CONFIG_NET_CLS_ACT
1910 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
1913 rc = __dev_xmit_skb(skb, q, dev, txq);
1917 /* The device has no queue. Common case for software devices:
1918 loopback, all the sorts of tunnels...
1920 Really, it is unlikely that netif_tx_lock protection is necessary
1921 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1923 However, it is possible, that they rely on protection
1926 Check this and shot the lock. It is not prone from deadlocks.
1927 Either shot noqueue qdisc, it is even simpler 8)
1929 if (dev->flags & IFF_UP) {
1930 int cpu = smp_processor_id(); /* ok because BHs are off */
1932 if (txq->xmit_lock_owner != cpu) {
1934 HARD_TX_LOCK(dev, txq, cpu);
1936 if (!netif_tx_queue_stopped(txq)) {
1937 rc = NET_XMIT_SUCCESS;
1938 if (!dev_hard_start_xmit(skb, dev, txq)) {
1939 HARD_TX_UNLOCK(dev, txq);
1943 HARD_TX_UNLOCK(dev, txq);
1944 if (net_ratelimit())
1945 printk(KERN_CRIT "Virtual device %s asks to "
1946 "queue packet!\n", dev->name);
1948 /* Recursion is detected! It is possible,
1950 if (net_ratelimit())
1951 printk(KERN_CRIT "Dead loop on virtual device "
1952 "%s, fix it urgently!\n", dev->name);
1957 rcu_read_unlock_bh();
1963 rcu_read_unlock_bh();
1966 EXPORT_SYMBOL(dev_queue_xmit);
1969 /*=======================================================================
1971 =======================================================================*/
1973 int netdev_max_backlog __read_mostly = 1000;
1974 int netdev_budget __read_mostly = 300;
1975 int weight_p __read_mostly = 64; /* old backlog weight */
1977 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1981 * netif_rx - post buffer to the network code
1982 * @skb: buffer to post
1984 * This function receives a packet from a device driver and queues it for
1985 * the upper (protocol) levels to process. It always succeeds. The buffer
1986 * may be dropped during processing for congestion control or by the
1990 * NET_RX_SUCCESS (no congestion)
1991 * NET_RX_DROP (packet was dropped)
1995 int netif_rx(struct sk_buff *skb)
1997 struct softnet_data *queue;
1998 unsigned long flags;
2000 /* if netpoll wants it, pretend we never saw it */
2001 if (netpoll_rx(skb))
2004 if (!skb->tstamp.tv64)
2008 * The code is rearranged so that the path is the most
2009 * short when CPU is congested, but is still operating.
2011 local_irq_save(flags);
2012 queue = &__get_cpu_var(softnet_data);
2014 __get_cpu_var(netdev_rx_stat).total++;
2015 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2016 if (queue->input_pkt_queue.qlen) {
2018 __skb_queue_tail(&queue->input_pkt_queue, skb);
2019 local_irq_restore(flags);
2020 return NET_RX_SUCCESS;
2023 napi_schedule(&queue->backlog);
2027 __get_cpu_var(netdev_rx_stat).dropped++;
2028 local_irq_restore(flags);
2033 EXPORT_SYMBOL(netif_rx);
2035 int netif_rx_ni(struct sk_buff *skb)
2040 err = netif_rx(skb);
2041 if (local_softirq_pending())
2047 EXPORT_SYMBOL(netif_rx_ni);
2049 static void net_tx_action(struct softirq_action *h)
2051 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2053 if (sd->completion_queue) {
2054 struct sk_buff *clist;
2056 local_irq_disable();
2057 clist = sd->completion_queue;
2058 sd->completion_queue = NULL;
2062 struct sk_buff *skb = clist;
2063 clist = clist->next;
2065 WARN_ON(atomic_read(&skb->users));
2070 if (sd->output_queue) {
2073 local_irq_disable();
2074 head = sd->output_queue;
2075 sd->output_queue = NULL;
2079 struct Qdisc *q = head;
2080 spinlock_t *root_lock;
2082 head = head->next_sched;
2084 root_lock = qdisc_lock(q);
2085 if (spin_trylock(root_lock)) {
2086 smp_mb__before_clear_bit();
2087 clear_bit(__QDISC_STATE_SCHED,
2090 spin_unlock(root_lock);
2092 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2094 __netif_reschedule(q);
2096 smp_mb__before_clear_bit();
2097 clear_bit(__QDISC_STATE_SCHED,
2105 static inline int deliver_skb(struct sk_buff *skb,
2106 struct packet_type *pt_prev,
2107 struct net_device *orig_dev)
2109 atomic_inc(&skb->users);
2110 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2113 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2115 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2116 /* This hook is defined here for ATM LANE */
2117 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2118 unsigned char *addr) __read_mostly;
2119 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2123 * If bridge module is loaded call bridging hook.
2124 * returns NULL if packet was consumed.
2126 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2127 struct sk_buff *skb) __read_mostly;
2128 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2130 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2131 struct packet_type **pt_prev, int *ret,
2132 struct net_device *orig_dev)
2134 struct net_bridge_port *port;
2136 if (skb->pkt_type == PACKET_LOOPBACK ||
2137 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2141 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2145 return br_handle_frame_hook(port, skb);
2148 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2151 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2152 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2153 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2155 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2156 struct packet_type **pt_prev,
2158 struct net_device *orig_dev)
2160 if (skb->dev->macvlan_port == NULL)
2164 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2167 return macvlan_handle_frame_hook(skb);
2170 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2173 #ifdef CONFIG_NET_CLS_ACT
2174 /* TODO: Maybe we should just force sch_ingress to be compiled in
2175 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2176 * a compare and 2 stores extra right now if we dont have it on
2177 * but have CONFIG_NET_CLS_ACT
2178 * NOTE: This doesnt stop any functionality; if you dont have
2179 * the ingress scheduler, you just cant add policies on ingress.
2182 static int ing_filter(struct sk_buff *skb)
2184 struct net_device *dev = skb->dev;
2185 u32 ttl = G_TC_RTTL(skb->tc_verd);
2186 struct netdev_queue *rxq;
2187 int result = TC_ACT_OK;
2190 if (MAX_RED_LOOP < ttl++) {
2192 "Redir loop detected Dropping packet (%d->%d)\n",
2193 skb->iif, dev->ifindex);
2197 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2198 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2200 rxq = &dev->rx_queue;
2203 if (q != &noop_qdisc) {
2204 spin_lock(qdisc_lock(q));
2205 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2206 result = qdisc_enqueue_root(skb, q);
2207 spin_unlock(qdisc_lock(q));
2213 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2214 struct packet_type **pt_prev,
2215 int *ret, struct net_device *orig_dev)
2217 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2221 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2224 /* Huh? Why does turning on AF_PACKET affect this? */
2225 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2228 switch (ing_filter(skb)) {
2242 * netif_nit_deliver - deliver received packets to network taps
2245 * This function is used to deliver incoming packets to network
2246 * taps. It should be used when the normal netif_receive_skb path
2247 * is bypassed, for example because of VLAN acceleration.
2249 void netif_nit_deliver(struct sk_buff *skb)
2251 struct packet_type *ptype;
2253 if (list_empty(&ptype_all))
2256 skb_reset_network_header(skb);
2257 skb_reset_transport_header(skb);
2258 skb->mac_len = skb->network_header - skb->mac_header;
2261 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2262 if (!ptype->dev || ptype->dev == skb->dev)
2263 deliver_skb(skb, ptype, skb->dev);
2269 * netif_receive_skb - process receive buffer from network
2270 * @skb: buffer to process
2272 * netif_receive_skb() is the main receive data processing function.
2273 * It always succeeds. The buffer may be dropped during processing
2274 * for congestion control or by the protocol layers.
2276 * This function may only be called from softirq context and interrupts
2277 * should be enabled.
2279 * Return values (usually ignored):
2280 * NET_RX_SUCCESS: no congestion
2281 * NET_RX_DROP: packet was dropped
2283 int netif_receive_skb(struct sk_buff *skb)
2285 struct packet_type *ptype, *pt_prev;
2286 struct net_device *orig_dev;
2287 struct net_device *null_or_orig;
2288 int ret = NET_RX_DROP;
2291 if (!skb->tstamp.tv64)
2294 if (skb->vlan_tci && vlan_hwaccel_do_receive(skb))
2295 return NET_RX_SUCCESS;
2297 /* if we've gotten here through NAPI, check netpoll */
2298 if (netpoll_receive_skb(skb))
2302 skb->iif = skb->dev->ifindex;
2304 null_or_orig = NULL;
2305 orig_dev = skb->dev;
2306 if (orig_dev->master) {
2307 if (skb_bond_should_drop(skb))
2308 null_or_orig = orig_dev; /* deliver only exact match */
2310 skb->dev = orig_dev->master;
2313 __get_cpu_var(netdev_rx_stat).total++;
2315 skb_reset_network_header(skb);
2316 skb_reset_transport_header(skb);
2317 skb->mac_len = skb->network_header - skb->mac_header;
2323 #ifdef CONFIG_NET_CLS_ACT
2324 if (skb->tc_verd & TC_NCLS) {
2325 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2330 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2331 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2332 ptype->dev == orig_dev) {
2334 ret = deliver_skb(skb, pt_prev, orig_dev);
2339 #ifdef CONFIG_NET_CLS_ACT
2340 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2346 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2349 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2353 type = skb->protocol;
2354 list_for_each_entry_rcu(ptype,
2355 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2356 if (ptype->type == type &&
2357 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2358 ptype->dev == orig_dev)) {
2360 ret = deliver_skb(skb, pt_prev, orig_dev);
2366 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2369 /* Jamal, now you will not able to escape explaining
2370 * me how you were going to use this. :-)
2379 EXPORT_SYMBOL(netif_receive_skb);
2381 /* Network device is going away, flush any packets still pending */
2382 static void flush_backlog(void *arg)
2384 struct net_device *dev = arg;
2385 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2386 struct sk_buff *skb, *tmp;
2388 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2389 if (skb->dev == dev) {
2390 __skb_unlink(skb, &queue->input_pkt_queue);
2395 static int napi_gro_complete(struct sk_buff *skb)
2397 struct packet_type *ptype;
2398 __be16 type = skb->protocol;
2399 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2402 if (NAPI_GRO_CB(skb)->count == 1) {
2403 skb_shinfo(skb)->gso_size = 0;
2408 list_for_each_entry_rcu(ptype, head, list) {
2409 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2412 err = ptype->gro_complete(skb);
2418 WARN_ON(&ptype->list == head);
2420 return NET_RX_SUCCESS;
2424 return netif_receive_skb(skb);
2427 void napi_gro_flush(struct napi_struct *napi)
2429 struct sk_buff *skb, *next;
2431 for (skb = napi->gro_list; skb; skb = next) {
2434 napi_gro_complete(skb);
2437 napi->gro_count = 0;
2438 napi->gro_list = NULL;
2440 EXPORT_SYMBOL(napi_gro_flush);
2442 int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2444 struct sk_buff **pp = NULL;
2445 struct packet_type *ptype;
2446 __be16 type = skb->protocol;
2447 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2452 if (!(skb->dev->features & NETIF_F_GRO))
2455 if (skb_is_gso(skb) || skb_has_frags(skb))
2459 list_for_each_entry_rcu(ptype, head, list) {
2460 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2463 skb_set_network_header(skb, skb_gro_offset(skb));
2464 mac_len = skb->network_header - skb->mac_header;
2465 skb->mac_len = mac_len;
2466 NAPI_GRO_CB(skb)->same_flow = 0;
2467 NAPI_GRO_CB(skb)->flush = 0;
2468 NAPI_GRO_CB(skb)->free = 0;
2470 pp = ptype->gro_receive(&napi->gro_list, skb);
2475 if (&ptype->list == head)
2478 same_flow = NAPI_GRO_CB(skb)->same_flow;
2479 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2482 struct sk_buff *nskb = *pp;
2486 napi_gro_complete(nskb);
2493 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2497 NAPI_GRO_CB(skb)->count = 1;
2498 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2499 skb->next = napi->gro_list;
2500 napi->gro_list = skb;
2504 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2505 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2507 BUG_ON(skb->end - skb->tail < grow);
2509 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2512 skb->data_len -= grow;
2514 skb_shinfo(skb)->frags[0].page_offset += grow;
2515 skb_shinfo(skb)->frags[0].size -= grow;
2517 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2518 put_page(skb_shinfo(skb)->frags[0].page);
2519 memmove(skb_shinfo(skb)->frags,
2520 skb_shinfo(skb)->frags + 1,
2521 --skb_shinfo(skb)->nr_frags);
2532 EXPORT_SYMBOL(dev_gro_receive);
2534 static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2538 if (netpoll_rx_on(skb))
2541 for (p = napi->gro_list; p; p = p->next) {
2542 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2543 && !compare_ether_header(skb_mac_header(p),
2544 skb_gro_mac_header(skb));
2545 NAPI_GRO_CB(p)->flush = 0;
2548 return dev_gro_receive(napi, skb);
2551 int napi_skb_finish(int ret, struct sk_buff *skb)
2553 int err = NET_RX_SUCCESS;
2557 return netif_receive_skb(skb);
2563 case GRO_MERGED_FREE:
2570 EXPORT_SYMBOL(napi_skb_finish);
2572 void skb_gro_reset_offset(struct sk_buff *skb)
2574 NAPI_GRO_CB(skb)->data_offset = 0;
2575 NAPI_GRO_CB(skb)->frag0 = NULL;
2576 NAPI_GRO_CB(skb)->frag0_len = 0;
2578 if (skb->mac_header == skb->tail &&
2579 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2580 NAPI_GRO_CB(skb)->frag0 =
2581 page_address(skb_shinfo(skb)->frags[0].page) +
2582 skb_shinfo(skb)->frags[0].page_offset;
2583 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2586 EXPORT_SYMBOL(skb_gro_reset_offset);
2588 int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2590 skb_gro_reset_offset(skb);
2592 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2594 EXPORT_SYMBOL(napi_gro_receive);
2596 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2598 __skb_pull(skb, skb_headlen(skb));
2599 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2603 EXPORT_SYMBOL(napi_reuse_skb);
2605 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2607 struct sk_buff *skb = napi->skb;
2610 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2616 EXPORT_SYMBOL(napi_get_frags);
2618 int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret)
2620 int err = NET_RX_SUCCESS;
2625 skb->protocol = eth_type_trans(skb, napi->dev);
2627 if (ret == GRO_NORMAL)
2628 return netif_receive_skb(skb);
2630 skb_gro_pull(skb, -ETH_HLEN);
2637 case GRO_MERGED_FREE:
2638 napi_reuse_skb(napi, skb);
2644 EXPORT_SYMBOL(napi_frags_finish);
2646 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2648 struct sk_buff *skb = napi->skb;
2655 skb_reset_mac_header(skb);
2656 skb_gro_reset_offset(skb);
2658 off = skb_gro_offset(skb);
2659 hlen = off + sizeof(*eth);
2660 eth = skb_gro_header_fast(skb, off);
2661 if (skb_gro_header_hard(skb, hlen)) {
2662 eth = skb_gro_header_slow(skb, hlen, off);
2663 if (unlikely(!eth)) {
2664 napi_reuse_skb(napi, skb);
2670 skb_gro_pull(skb, sizeof(*eth));
2673 * This works because the only protocols we care about don't require
2674 * special handling. We'll fix it up properly at the end.
2676 skb->protocol = eth->h_proto;
2681 EXPORT_SYMBOL(napi_frags_skb);
2683 int napi_gro_frags(struct napi_struct *napi)
2685 struct sk_buff *skb = napi_frags_skb(napi);
2690 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2692 EXPORT_SYMBOL(napi_gro_frags);
2694 static int process_backlog(struct napi_struct *napi, int quota)
2697 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2698 unsigned long start_time = jiffies;
2700 napi->weight = weight_p;
2702 struct sk_buff *skb;
2704 local_irq_disable();
2705 skb = __skb_dequeue(&queue->input_pkt_queue);
2707 __napi_complete(napi);
2713 netif_receive_skb(skb);
2714 } while (++work < quota && jiffies == start_time);
2720 * __napi_schedule - schedule for receive
2721 * @n: entry to schedule
2723 * The entry's receive function will be scheduled to run
2725 void __napi_schedule(struct napi_struct *n)
2727 unsigned long flags;
2729 local_irq_save(flags);
2730 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2731 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2732 local_irq_restore(flags);
2734 EXPORT_SYMBOL(__napi_schedule);
2736 void __napi_complete(struct napi_struct *n)
2738 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2739 BUG_ON(n->gro_list);
2741 list_del(&n->poll_list);
2742 smp_mb__before_clear_bit();
2743 clear_bit(NAPI_STATE_SCHED, &n->state);
2745 EXPORT_SYMBOL(__napi_complete);
2747 void napi_complete(struct napi_struct *n)
2749 unsigned long flags;
2752 * don't let napi dequeue from the cpu poll list
2753 * just in case its running on a different cpu
2755 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2759 local_irq_save(flags);
2761 local_irq_restore(flags);
2763 EXPORT_SYMBOL(napi_complete);
2765 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2766 int (*poll)(struct napi_struct *, int), int weight)
2768 INIT_LIST_HEAD(&napi->poll_list);
2769 napi->gro_count = 0;
2770 napi->gro_list = NULL;
2773 napi->weight = weight;
2774 list_add(&napi->dev_list, &dev->napi_list);
2776 #ifdef CONFIG_NETPOLL
2777 spin_lock_init(&napi->poll_lock);
2778 napi->poll_owner = -1;
2780 set_bit(NAPI_STATE_SCHED, &napi->state);
2782 EXPORT_SYMBOL(netif_napi_add);
2784 void netif_napi_del(struct napi_struct *napi)
2786 struct sk_buff *skb, *next;
2788 list_del_init(&napi->dev_list);
2789 napi_free_frags(napi);
2791 for (skb = napi->gro_list; skb; skb = next) {
2797 napi->gro_list = NULL;
2798 napi->gro_count = 0;
2800 EXPORT_SYMBOL(netif_napi_del);
2803 static void net_rx_action(struct softirq_action *h)
2805 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2806 unsigned long time_limit = jiffies + 2;
2807 int budget = netdev_budget;
2810 local_irq_disable();
2812 while (!list_empty(list)) {
2813 struct napi_struct *n;
2816 /* If softirq window is exhuasted then punt.
2817 * Allow this to run for 2 jiffies since which will allow
2818 * an average latency of 1.5/HZ.
2820 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2825 /* Even though interrupts have been re-enabled, this
2826 * access is safe because interrupts can only add new
2827 * entries to the tail of this list, and only ->poll()
2828 * calls can remove this head entry from the list.
2830 n = list_entry(list->next, struct napi_struct, poll_list);
2832 have = netpoll_poll_lock(n);
2836 /* This NAPI_STATE_SCHED test is for avoiding a race
2837 * with netpoll's poll_napi(). Only the entity which
2838 * obtains the lock and sees NAPI_STATE_SCHED set will
2839 * actually make the ->poll() call. Therefore we avoid
2840 * accidently calling ->poll() when NAPI is not scheduled.
2843 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2844 work = n->poll(n, weight);
2848 WARN_ON_ONCE(work > weight);
2852 local_irq_disable();
2854 /* Drivers must not modify the NAPI state if they
2855 * consume the entire weight. In such cases this code
2856 * still "owns" the NAPI instance and therefore can
2857 * move the instance around on the list at-will.
2859 if (unlikely(work == weight)) {
2860 if (unlikely(napi_disable_pending(n))) {
2863 local_irq_disable();
2865 list_move_tail(&n->poll_list, list);
2868 netpoll_poll_unlock(have);
2873 #ifdef CONFIG_NET_DMA
2875 * There may not be any more sk_buffs coming right now, so push
2876 * any pending DMA copies to hardware
2878 dma_issue_pending_all();
2884 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2885 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2889 static gifconf_func_t *gifconf_list[NPROTO];
2892 * register_gifconf - register a SIOCGIF handler
2893 * @family: Address family
2894 * @gifconf: Function handler
2896 * Register protocol dependent address dumping routines. The handler
2897 * that is passed must not be freed or reused until it has been replaced
2898 * by another handler.
2900 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
2902 if (family >= NPROTO)
2904 gifconf_list[family] = gifconf;
2907 EXPORT_SYMBOL(register_gifconf);
2911 * Map an interface index to its name (SIOCGIFNAME)
2915 * We need this ioctl for efficient implementation of the
2916 * if_indextoname() function required by the IPv6 API. Without
2917 * it, we would have to search all the interfaces to find a
2921 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2923 struct net_device *dev;
2927 * Fetch the caller's info block.
2930 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2933 read_lock(&dev_base_lock);
2934 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2936 read_unlock(&dev_base_lock);
2940 strcpy(ifr.ifr_name, dev->name);
2941 read_unlock(&dev_base_lock);
2943 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2949 * Perform a SIOCGIFCONF call. This structure will change
2950 * size eventually, and there is nothing I can do about it.
2951 * Thus we will need a 'compatibility mode'.
2954 static int dev_ifconf(struct net *net, char __user *arg)
2957 struct net_device *dev;
2964 * Fetch the caller's info block.
2967 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2974 * Loop over the interfaces, and write an info block for each.
2978 for_each_netdev(net, dev) {
2979 for (i = 0; i < NPROTO; i++) {
2980 if (gifconf_list[i]) {
2983 done = gifconf_list[i](dev, NULL, 0);
2985 done = gifconf_list[i](dev, pos + total,
2995 * All done. Write the updated control block back to the caller.
2997 ifc.ifc_len = total;
3000 * Both BSD and Solaris return 0 here, so we do too.
3002 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3005 #ifdef CONFIG_PROC_FS
3007 * This is invoked by the /proc filesystem handler to display a device
3010 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3011 __acquires(dev_base_lock)
3013 struct net *net = seq_file_net(seq);
3015 struct net_device *dev;
3017 read_lock(&dev_base_lock);
3019 return SEQ_START_TOKEN;
3022 for_each_netdev(net, dev)
3029 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3031 struct net *net = seq_file_net(seq);
3033 return v == SEQ_START_TOKEN ?
3034 first_net_device(net) : next_net_device((struct net_device *)v);
3037 void dev_seq_stop(struct seq_file *seq, void *v)
3038 __releases(dev_base_lock)
3040 read_unlock(&dev_base_lock);
3043 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3045 const struct net_device_stats *stats = dev_get_stats(dev);
3047 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3048 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3049 dev->name, stats->rx_bytes, stats->rx_packets,
3051 stats->rx_dropped + stats->rx_missed_errors,
3052 stats->rx_fifo_errors,
3053 stats->rx_length_errors + stats->rx_over_errors +
3054 stats->rx_crc_errors + stats->rx_frame_errors,
3055 stats->rx_compressed, stats->multicast,
3056 stats->tx_bytes, stats->tx_packets,
3057 stats->tx_errors, stats->tx_dropped,
3058 stats->tx_fifo_errors, stats->collisions,
3059 stats->tx_carrier_errors +
3060 stats->tx_aborted_errors +
3061 stats->tx_window_errors +
3062 stats->tx_heartbeat_errors,
3063 stats->tx_compressed);
3067 * Called from the PROCfs module. This now uses the new arbitrary sized
3068 * /proc/net interface to create /proc/net/dev
3070 static int dev_seq_show(struct seq_file *seq, void *v)
3072 if (v == SEQ_START_TOKEN)
3073 seq_puts(seq, "Inter-| Receive "
3075 " face |bytes packets errs drop fifo frame "
3076 "compressed multicast|bytes packets errs "
3077 "drop fifo colls carrier compressed\n");
3079 dev_seq_printf_stats(seq, v);
3083 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3085 struct netif_rx_stats *rc = NULL;
3087 while (*pos < nr_cpu_ids)
3088 if (cpu_online(*pos)) {
3089 rc = &per_cpu(netdev_rx_stat, *pos);
3096 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3098 return softnet_get_online(pos);
3101 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3104 return softnet_get_online(pos);
3107 static void softnet_seq_stop(struct seq_file *seq, void *v)
3111 static int softnet_seq_show(struct seq_file *seq, void *v)
3113 struct netif_rx_stats *s = v;
3115 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3116 s->total, s->dropped, s->time_squeeze, 0,
3117 0, 0, 0, 0, /* was fastroute */
3122 static const struct seq_operations dev_seq_ops = {
3123 .start = dev_seq_start,
3124 .next = dev_seq_next,
3125 .stop = dev_seq_stop,
3126 .show = dev_seq_show,
3129 static int dev_seq_open(struct inode *inode, struct file *file)
3131 return seq_open_net(inode, file, &dev_seq_ops,
3132 sizeof(struct seq_net_private));
3135 static const struct file_operations dev_seq_fops = {
3136 .owner = THIS_MODULE,
3137 .open = dev_seq_open,
3139 .llseek = seq_lseek,
3140 .release = seq_release_net,
3143 static const struct seq_operations softnet_seq_ops = {
3144 .start = softnet_seq_start,
3145 .next = softnet_seq_next,
3146 .stop = softnet_seq_stop,
3147 .show = softnet_seq_show,
3150 static int softnet_seq_open(struct inode *inode, struct file *file)
3152 return seq_open(file, &softnet_seq_ops);
3155 static const struct file_operations softnet_seq_fops = {
3156 .owner = THIS_MODULE,
3157 .open = softnet_seq_open,
3159 .llseek = seq_lseek,
3160 .release = seq_release,
3163 static void *ptype_get_idx(loff_t pos)
3165 struct packet_type *pt = NULL;
3169 list_for_each_entry_rcu(pt, &ptype_all, list) {
3175 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3176 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3185 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3189 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3192 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3194 struct packet_type *pt;
3195 struct list_head *nxt;
3199 if (v == SEQ_START_TOKEN)
3200 return ptype_get_idx(0);
3203 nxt = pt->list.next;
3204 if (pt->type == htons(ETH_P_ALL)) {
3205 if (nxt != &ptype_all)
3208 nxt = ptype_base[0].next;
3210 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3212 while (nxt == &ptype_base[hash]) {
3213 if (++hash >= PTYPE_HASH_SIZE)
3215 nxt = ptype_base[hash].next;
3218 return list_entry(nxt, struct packet_type, list);
3221 static void ptype_seq_stop(struct seq_file *seq, void *v)
3227 static int ptype_seq_show(struct seq_file *seq, void *v)
3229 struct packet_type *pt = v;
3231 if (v == SEQ_START_TOKEN)
3232 seq_puts(seq, "Type Device Function\n");
3233 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3234 if (pt->type == htons(ETH_P_ALL))
3235 seq_puts(seq, "ALL ");
3237 seq_printf(seq, "%04x", ntohs(pt->type));
3239 seq_printf(seq, " %-8s %pF\n",
3240 pt->dev ? pt->dev->name : "", pt->func);
3246 static const struct seq_operations ptype_seq_ops = {
3247 .start = ptype_seq_start,
3248 .next = ptype_seq_next,
3249 .stop = ptype_seq_stop,
3250 .show = ptype_seq_show,
3253 static int ptype_seq_open(struct inode *inode, struct file *file)
3255 return seq_open_net(inode, file, &ptype_seq_ops,
3256 sizeof(struct seq_net_private));
3259 static const struct file_operations ptype_seq_fops = {
3260 .owner = THIS_MODULE,
3261 .open = ptype_seq_open,
3263 .llseek = seq_lseek,
3264 .release = seq_release_net,
3268 static int __net_init dev_proc_net_init(struct net *net)
3272 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3274 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3276 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3279 if (wext_proc_init(net))
3285 proc_net_remove(net, "ptype");
3287 proc_net_remove(net, "softnet_stat");
3289 proc_net_remove(net, "dev");
3293 static void __net_exit dev_proc_net_exit(struct net *net)
3295 wext_proc_exit(net);
3297 proc_net_remove(net, "ptype");
3298 proc_net_remove(net, "softnet_stat");
3299 proc_net_remove(net, "dev");
3302 static struct pernet_operations __net_initdata dev_proc_ops = {
3303 .init = dev_proc_net_init,
3304 .exit = dev_proc_net_exit,
3307 static int __init dev_proc_init(void)
3309 return register_pernet_subsys(&dev_proc_ops);
3312 #define dev_proc_init() 0
3313 #endif /* CONFIG_PROC_FS */
3317 * netdev_set_master - set up master/slave pair
3318 * @slave: slave device
3319 * @master: new master device
3321 * Changes the master device of the slave. Pass %NULL to break the
3322 * bonding. The caller must hold the RTNL semaphore. On a failure
3323 * a negative errno code is returned. On success the reference counts
3324 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3325 * function returns zero.
3327 int netdev_set_master(struct net_device *slave, struct net_device *master)
3329 struct net_device *old = slave->master;
3339 slave->master = master;
3347 slave->flags |= IFF_SLAVE;
3349 slave->flags &= ~IFF_SLAVE;
3351 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3354 EXPORT_SYMBOL(netdev_set_master);
3356 static void dev_change_rx_flags(struct net_device *dev, int flags)
3358 const struct net_device_ops *ops = dev->netdev_ops;
3360 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3361 ops->ndo_change_rx_flags(dev, flags);
3364 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3366 unsigned short old_flags = dev->flags;
3372 dev->flags |= IFF_PROMISC;
3373 dev->promiscuity += inc;
3374 if (dev->promiscuity == 0) {
3377 * If inc causes overflow, untouch promisc and return error.
3380 dev->flags &= ~IFF_PROMISC;
3382 dev->promiscuity -= inc;
3383 printk(KERN_WARNING "%s: promiscuity touches roof, "
3384 "set promiscuity failed, promiscuity feature "
3385 "of device might be broken.\n", dev->name);
3389 if (dev->flags != old_flags) {
3390 printk(KERN_INFO "device %s %s promiscuous mode\n",
3391 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3393 if (audit_enabled) {
3394 current_uid_gid(&uid, &gid);
3395 audit_log(current->audit_context, GFP_ATOMIC,
3396 AUDIT_ANOM_PROMISCUOUS,
3397 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3398 dev->name, (dev->flags & IFF_PROMISC),
3399 (old_flags & IFF_PROMISC),
3400 audit_get_loginuid(current),
3402 audit_get_sessionid(current));
3405 dev_change_rx_flags(dev, IFF_PROMISC);
3411 * dev_set_promiscuity - update promiscuity count on a device
3415 * Add or remove promiscuity from a device. While the count in the device
3416 * remains above zero the interface remains promiscuous. Once it hits zero
3417 * the device reverts back to normal filtering operation. A negative inc
3418 * value is used to drop promiscuity on the device.
3419 * Return 0 if successful or a negative errno code on error.
3421 int dev_set_promiscuity(struct net_device *dev, int inc)
3423 unsigned short old_flags = dev->flags;
3426 err = __dev_set_promiscuity(dev, inc);
3429 if (dev->flags != old_flags)
3430 dev_set_rx_mode(dev);
3433 EXPORT_SYMBOL(dev_set_promiscuity);
3436 * dev_set_allmulti - update allmulti count on a device
3440 * Add or remove reception of all multicast frames to a device. While the
3441 * count in the device remains above zero the interface remains listening
3442 * to all interfaces. Once it hits zero the device reverts back to normal
3443 * filtering operation. A negative @inc value is used to drop the counter
3444 * when releasing a resource needing all multicasts.
3445 * Return 0 if successful or a negative errno code on error.
3448 int dev_set_allmulti(struct net_device *dev, int inc)
3450 unsigned short old_flags = dev->flags;
3454 dev->flags |= IFF_ALLMULTI;
3455 dev->allmulti += inc;
3456 if (dev->allmulti == 0) {
3459 * If inc causes overflow, untouch allmulti and return error.
3462 dev->flags &= ~IFF_ALLMULTI;
3464 dev->allmulti -= inc;
3465 printk(KERN_WARNING "%s: allmulti touches roof, "
3466 "set allmulti failed, allmulti feature of "
3467 "device might be broken.\n", dev->name);
3471 if (dev->flags ^ old_flags) {
3472 dev_change_rx_flags(dev, IFF_ALLMULTI);
3473 dev_set_rx_mode(dev);
3477 EXPORT_SYMBOL(dev_set_allmulti);
3480 * Upload unicast and multicast address lists to device and
3481 * configure RX filtering. When the device doesn't support unicast
3482 * filtering it is put in promiscuous mode while unicast addresses
3485 void __dev_set_rx_mode(struct net_device *dev)
3487 const struct net_device_ops *ops = dev->netdev_ops;
3489 /* dev_open will call this function so the list will stay sane. */
3490 if (!(dev->flags&IFF_UP))
3493 if (!netif_device_present(dev))
3496 if (ops->ndo_set_rx_mode)
3497 ops->ndo_set_rx_mode(dev);
3499 /* Unicast addresses changes may only happen under the rtnl,
3500 * therefore calling __dev_set_promiscuity here is safe.
3502 if (dev->uc.count > 0 && !dev->uc_promisc) {
3503 __dev_set_promiscuity(dev, 1);
3504 dev->uc_promisc = 1;
3505 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3506 __dev_set_promiscuity(dev, -1);
3507 dev->uc_promisc = 0;
3510 if (ops->ndo_set_multicast_list)
3511 ops->ndo_set_multicast_list(dev);
3515 void dev_set_rx_mode(struct net_device *dev)
3517 netif_addr_lock_bh(dev);
3518 __dev_set_rx_mode(dev);
3519 netif_addr_unlock_bh(dev);
3522 /* hw addresses list handling functions */
3524 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3525 int addr_len, unsigned char addr_type)
3527 struct netdev_hw_addr *ha;
3530 if (addr_len > MAX_ADDR_LEN)
3533 list_for_each_entry(ha, &list->list, list) {
3534 if (!memcmp(ha->addr, addr, addr_len) &&
3535 ha->type == addr_type) {
3542 alloc_size = sizeof(*ha);
3543 if (alloc_size < L1_CACHE_BYTES)
3544 alloc_size = L1_CACHE_BYTES;
3545 ha = kmalloc(alloc_size, GFP_ATOMIC);
3548 memcpy(ha->addr, addr, addr_len);
3549 ha->type = addr_type;
3552 list_add_tail_rcu(&ha->list, &list->list);
3557 static void ha_rcu_free(struct rcu_head *head)
3559 struct netdev_hw_addr *ha;
3561 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3565 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3566 int addr_len, unsigned char addr_type)
3568 struct netdev_hw_addr *ha;
3570 list_for_each_entry(ha, &list->list, list) {
3571 if (!memcmp(ha->addr, addr, addr_len) &&
3572 (ha->type == addr_type || !addr_type)) {
3575 list_del_rcu(&ha->list);
3576 call_rcu(&ha->rcu_head, ha_rcu_free);
3584 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3585 struct netdev_hw_addr_list *from_list,
3587 unsigned char addr_type)
3590 struct netdev_hw_addr *ha, *ha2;
3593 list_for_each_entry(ha, &from_list->list, list) {
3594 type = addr_type ? addr_type : ha->type;
3595 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3602 list_for_each_entry(ha2, &from_list->list, list) {
3605 type = addr_type ? addr_type : ha2->type;
3606 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3611 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3612 struct netdev_hw_addr_list *from_list,
3614 unsigned char addr_type)
3616 struct netdev_hw_addr *ha;
3619 list_for_each_entry(ha, &from_list->list, list) {
3620 type = addr_type ? addr_type : ha->type;
3621 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3625 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3626 struct netdev_hw_addr_list *from_list,
3630 struct netdev_hw_addr *ha, *tmp;
3632 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3634 err = __hw_addr_add(to_list, ha->addr,
3635 addr_len, ha->type);
3640 } else if (ha->refcount == 1) {
3641 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3642 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3648 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3649 struct netdev_hw_addr_list *from_list,
3652 struct netdev_hw_addr *ha, *tmp;
3654 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3656 __hw_addr_del(to_list, ha->addr,
3657 addr_len, ha->type);
3659 __hw_addr_del(from_list, ha->addr,
3660 addr_len, ha->type);
3665 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3667 struct netdev_hw_addr *ha, *tmp;
3669 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3670 list_del_rcu(&ha->list);
3671 call_rcu(&ha->rcu_head, ha_rcu_free);
3676 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3678 INIT_LIST_HEAD(&list->list);
3682 /* Device addresses handling functions */
3684 static void dev_addr_flush(struct net_device *dev)
3686 /* rtnl_mutex must be held here */
3688 __hw_addr_flush(&dev->dev_addrs);
3689 dev->dev_addr = NULL;
3692 static int dev_addr_init(struct net_device *dev)
3694 unsigned char addr[MAX_ADDR_LEN];
3695 struct netdev_hw_addr *ha;
3698 /* rtnl_mutex must be held here */
3700 __hw_addr_init(&dev->dev_addrs);
3701 memset(addr, 0, sizeof(addr));
3702 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3703 NETDEV_HW_ADDR_T_LAN);
3706 * Get the first (previously created) address from the list
3707 * and set dev_addr pointer to this location.
3709 ha = list_first_entry(&dev->dev_addrs.list,
3710 struct netdev_hw_addr, list);
3711 dev->dev_addr = ha->addr;
3717 * dev_addr_add - Add a device address
3719 * @addr: address to add
3720 * @addr_type: address type
3722 * Add a device address to the device or increase the reference count if
3723 * it already exists.
3725 * The caller must hold the rtnl_mutex.
3727 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3728 unsigned char addr_type)
3734 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3736 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3739 EXPORT_SYMBOL(dev_addr_add);
3742 * dev_addr_del - Release a device address.
3744 * @addr: address to delete
3745 * @addr_type: address type
3747 * Release reference to a device address and remove it from the device
3748 * if the reference count drops to zero.
3750 * The caller must hold the rtnl_mutex.
3752 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3753 unsigned char addr_type)
3756 struct netdev_hw_addr *ha;
3761 * We can not remove the first address from the list because
3762 * dev->dev_addr points to that.
3764 ha = list_first_entry(&dev->dev_addrs.list,
3765 struct netdev_hw_addr, list);
3766 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3769 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3772 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3775 EXPORT_SYMBOL(dev_addr_del);
3778 * dev_addr_add_multiple - Add device addresses from another device
3779 * @to_dev: device to which addresses will be added
3780 * @from_dev: device from which addresses will be added
3781 * @addr_type: address type - 0 means type will be used from from_dev
3783 * Add device addresses of the one device to another.
3785 * The caller must hold the rtnl_mutex.
3787 int dev_addr_add_multiple(struct net_device *to_dev,
3788 struct net_device *from_dev,
3789 unsigned char addr_type)
3795 if (from_dev->addr_len != to_dev->addr_len)
3797 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3798 to_dev->addr_len, addr_type);
3800 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3803 EXPORT_SYMBOL(dev_addr_add_multiple);
3806 * dev_addr_del_multiple - Delete device addresses by another device
3807 * @to_dev: device where the addresses will be deleted
3808 * @from_dev: device by which addresses the addresses will be deleted
3809 * @addr_type: address type - 0 means type will used from from_dev
3811 * Deletes addresses in to device by the list of addresses in from device.
3813 * The caller must hold the rtnl_mutex.
3815 int dev_addr_del_multiple(struct net_device *to_dev,
3816 struct net_device *from_dev,
3817 unsigned char addr_type)
3821 if (from_dev->addr_len != to_dev->addr_len)
3823 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3824 to_dev->addr_len, addr_type);
3825 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3828 EXPORT_SYMBOL(dev_addr_del_multiple);
3830 /* multicast addresses handling functions */
3832 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3833 void *addr, int alen, int glbl)
3835 struct dev_addr_list *da;
3837 for (; (da = *list) != NULL; list = &da->next) {
3838 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3839 alen == da->da_addrlen) {
3841 int old_glbl = da->da_gusers;
3858 int __dev_addr_add(struct dev_addr_list **list, int *count,
3859 void *addr, int alen, int glbl)
3861 struct dev_addr_list *da;
3863 for (da = *list; da != NULL; da = da->next) {
3864 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3865 da->da_addrlen == alen) {
3867 int old_glbl = da->da_gusers;
3877 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3880 memcpy(da->da_addr, addr, alen);
3881 da->da_addrlen = alen;
3883 da->da_gusers = glbl ? 1 : 0;
3891 * dev_unicast_delete - Release secondary unicast address.
3893 * @addr: address to delete
3895 * Release reference to a secondary unicast address and remove it
3896 * from the device if the reference count drops to zero.
3898 * The caller must hold the rtnl_mutex.
3900 int dev_unicast_delete(struct net_device *dev, void *addr)
3906 netif_addr_lock_bh(dev);
3907 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
3908 NETDEV_HW_ADDR_T_UNICAST);
3910 __dev_set_rx_mode(dev);
3911 netif_addr_unlock_bh(dev);
3914 EXPORT_SYMBOL(dev_unicast_delete);
3917 * dev_unicast_add - add a secondary unicast address
3919 * @addr: address to add
3921 * Add a secondary unicast address to the device or increase
3922 * the reference count if it already exists.
3924 * The caller must hold the rtnl_mutex.
3926 int dev_unicast_add(struct net_device *dev, void *addr)
3932 netif_addr_lock_bh(dev);
3933 err = __hw_addr_add(&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_add);
3942 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3943 struct dev_addr_list **from, int *from_count)
3945 struct dev_addr_list *da, *next;
3949 while (da != NULL) {
3951 if (!da->da_synced) {
3952 err = __dev_addr_add(to, to_count,
3953 da->da_addr, da->da_addrlen, 0);
3958 } else if (da->da_users == 1) {
3959 __dev_addr_delete(to, to_count,
3960 da->da_addr, da->da_addrlen, 0);
3961 __dev_addr_delete(from, from_count,
3962 da->da_addr, da->da_addrlen, 0);
3968 EXPORT_SYMBOL_GPL(__dev_addr_sync);
3970 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3971 struct dev_addr_list **from, int *from_count)
3973 struct dev_addr_list *da, *next;
3976 while (da != NULL) {
3978 if (da->da_synced) {
3979 __dev_addr_delete(to, to_count,
3980 da->da_addr, da->da_addrlen, 0);
3982 __dev_addr_delete(from, from_count,
3983 da->da_addr, da->da_addrlen, 0);
3988 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
3991 * dev_unicast_sync - Synchronize device's unicast list to another device
3992 * @to: destination device
3993 * @from: source device
3995 * Add newly added addresses to the destination device and release
3996 * addresses that have no users left. The source device must be
3997 * locked by netif_tx_lock_bh.
3999 * This function is intended to be called from the dev->set_rx_mode
4000 * function of layered software devices.
4002 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4006 if (to->addr_len != from->addr_len)
4009 netif_addr_lock_bh(to);
4010 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4012 __dev_set_rx_mode(to);
4013 netif_addr_unlock_bh(to);
4016 EXPORT_SYMBOL(dev_unicast_sync);
4019 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4020 * @to: destination device
4021 * @from: source device
4023 * Remove all addresses that were added to the destination device by
4024 * dev_unicast_sync(). This function is intended to be called from the
4025 * dev->stop function of layered software devices.
4027 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4029 if (to->addr_len != from->addr_len)
4032 netif_addr_lock_bh(from);
4033 netif_addr_lock(to);
4034 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4035 __dev_set_rx_mode(to);
4036 netif_addr_unlock(to);
4037 netif_addr_unlock_bh(from);
4039 EXPORT_SYMBOL(dev_unicast_unsync);
4041 static void dev_unicast_flush(struct net_device *dev)
4043 netif_addr_lock_bh(dev);
4044 __hw_addr_flush(&dev->uc);
4045 netif_addr_unlock_bh(dev);
4048 static void dev_unicast_init(struct net_device *dev)
4050 __hw_addr_init(&dev->uc);
4054 static void __dev_addr_discard(struct dev_addr_list **list)
4056 struct dev_addr_list *tmp;
4058 while (*list != NULL) {
4061 if (tmp->da_users > tmp->da_gusers)
4062 printk("__dev_addr_discard: address leakage! "
4063 "da_users=%d\n", tmp->da_users);
4068 static void dev_addr_discard(struct net_device *dev)
4070 netif_addr_lock_bh(dev);
4072 __dev_addr_discard(&dev->mc_list);
4075 netif_addr_unlock_bh(dev);
4079 * dev_get_flags - get flags reported to userspace
4082 * Get the combination of flag bits exported through APIs to userspace.
4084 unsigned dev_get_flags(const struct net_device *dev)
4088 flags = (dev->flags & ~(IFF_PROMISC |
4093 (dev->gflags & (IFF_PROMISC |
4096 if (netif_running(dev)) {
4097 if (netif_oper_up(dev))
4098 flags |= IFF_RUNNING;
4099 if (netif_carrier_ok(dev))
4100 flags |= IFF_LOWER_UP;
4101 if (netif_dormant(dev))
4102 flags |= IFF_DORMANT;
4107 EXPORT_SYMBOL(dev_get_flags);
4110 * dev_change_flags - change device settings
4112 * @flags: device state flags
4114 * Change settings on device based state flags. The flags are
4115 * in the userspace exported format.
4117 int dev_change_flags(struct net_device *dev, unsigned flags)
4120 int old_flags = dev->flags;
4125 * Set the flags on our device.
4128 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4129 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4131 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4135 * Load in the correct multicast list now the flags have changed.
4138 if ((old_flags ^ flags) & IFF_MULTICAST)
4139 dev_change_rx_flags(dev, IFF_MULTICAST);
4141 dev_set_rx_mode(dev);
4144 * Have we downed the interface. We handle IFF_UP ourselves
4145 * according to user attempts to set it, rather than blindly
4150 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4151 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4154 dev_set_rx_mode(dev);
4157 if (dev->flags & IFF_UP &&
4158 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4160 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4162 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4163 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4165 dev->gflags ^= IFF_PROMISC;
4166 dev_set_promiscuity(dev, inc);
4169 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4170 is important. Some (broken) drivers set IFF_PROMISC, when
4171 IFF_ALLMULTI is requested not asking us and not reporting.
4173 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4174 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4176 dev->gflags ^= IFF_ALLMULTI;
4177 dev_set_allmulti(dev, inc);
4180 /* Exclude state transition flags, already notified */
4181 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4183 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4187 EXPORT_SYMBOL(dev_change_flags);
4190 * dev_set_mtu - Change maximum transfer unit
4192 * @new_mtu: new transfer unit
4194 * Change the maximum transfer size of the network device.
4196 int dev_set_mtu(struct net_device *dev, int new_mtu)
4198 const struct net_device_ops *ops = dev->netdev_ops;
4201 if (new_mtu == dev->mtu)
4204 /* MTU must be positive. */
4208 if (!netif_device_present(dev))
4212 if (ops->ndo_change_mtu)
4213 err = ops->ndo_change_mtu(dev, new_mtu);
4217 if (!err && dev->flags & IFF_UP)
4218 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4221 EXPORT_SYMBOL(dev_set_mtu);
4224 * dev_set_mac_address - Change Media Access Control Address
4228 * Change the hardware (MAC) address of the device
4230 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4232 const struct net_device_ops *ops = dev->netdev_ops;
4235 if (!ops->ndo_set_mac_address)
4237 if (sa->sa_family != dev->type)
4239 if (!netif_device_present(dev))
4241 err = ops->ndo_set_mac_address(dev, sa);
4243 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4246 EXPORT_SYMBOL(dev_set_mac_address);
4249 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
4251 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4254 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4260 case SIOCGIFFLAGS: /* Get interface flags */
4261 ifr->ifr_flags = (short) dev_get_flags(dev);
4264 case SIOCGIFMETRIC: /* Get the metric on the interface
4265 (currently unused) */
4266 ifr->ifr_metric = 0;
4269 case SIOCGIFMTU: /* Get the MTU of a device */
4270 ifr->ifr_mtu = dev->mtu;
4275 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4277 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4278 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4279 ifr->ifr_hwaddr.sa_family = dev->type;
4287 ifr->ifr_map.mem_start = dev->mem_start;
4288 ifr->ifr_map.mem_end = dev->mem_end;
4289 ifr->ifr_map.base_addr = dev->base_addr;
4290 ifr->ifr_map.irq = dev->irq;
4291 ifr->ifr_map.dma = dev->dma;
4292 ifr->ifr_map.port = dev->if_port;
4296 ifr->ifr_ifindex = dev->ifindex;
4300 ifr->ifr_qlen = dev->tx_queue_len;
4304 /* dev_ioctl() should ensure this case
4316 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4318 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4321 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4322 const struct net_device_ops *ops;
4327 ops = dev->netdev_ops;
4330 case SIOCSIFFLAGS: /* Set interface flags */
4331 return dev_change_flags(dev, ifr->ifr_flags);
4333 case SIOCSIFMETRIC: /* Set the metric on the interface
4334 (currently unused) */
4337 case SIOCSIFMTU: /* Set the MTU of a device */
4338 return dev_set_mtu(dev, ifr->ifr_mtu);
4341 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4343 case SIOCSIFHWBROADCAST:
4344 if (ifr->ifr_hwaddr.sa_family != dev->type)
4346 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4347 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4348 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4352 if (ops->ndo_set_config) {
4353 if (!netif_device_present(dev))
4355 return ops->ndo_set_config(dev, &ifr->ifr_map);
4360 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4361 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4363 if (!netif_device_present(dev))
4365 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4369 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4370 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4372 if (!netif_device_present(dev))
4374 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4378 if (ifr->ifr_qlen < 0)
4380 dev->tx_queue_len = ifr->ifr_qlen;
4384 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4385 return dev_change_name(dev, ifr->ifr_newname);
4388 * Unknown or private ioctl
4391 if ((cmd >= SIOCDEVPRIVATE &&
4392 cmd <= SIOCDEVPRIVATE + 15) ||
4393 cmd == SIOCBONDENSLAVE ||
4394 cmd == SIOCBONDRELEASE ||
4395 cmd == SIOCBONDSETHWADDR ||
4396 cmd == SIOCBONDSLAVEINFOQUERY ||
4397 cmd == SIOCBONDINFOQUERY ||
4398 cmd == SIOCBONDCHANGEACTIVE ||
4399 cmd == SIOCGMIIPHY ||
4400 cmd == SIOCGMIIREG ||
4401 cmd == SIOCSMIIREG ||
4402 cmd == SIOCBRADDIF ||
4403 cmd == SIOCBRDELIF ||
4404 cmd == SIOCSHWTSTAMP ||
4405 cmd == SIOCWANDEV) {
4407 if (ops->ndo_do_ioctl) {
4408 if (netif_device_present(dev))
4409 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4421 * This function handles all "interface"-type I/O control requests. The actual
4422 * 'doing' part of this is dev_ifsioc above.
4426 * dev_ioctl - network device ioctl
4427 * @net: the applicable net namespace
4428 * @cmd: command to issue
4429 * @arg: pointer to a struct ifreq in user space
4431 * Issue ioctl functions to devices. This is normally called by the
4432 * user space syscall interfaces but can sometimes be useful for
4433 * other purposes. The return value is the return from the syscall if
4434 * positive or a negative errno code on error.
4437 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4443 /* One special case: SIOCGIFCONF takes ifconf argument
4444 and requires shared lock, because it sleeps writing
4448 if (cmd == SIOCGIFCONF) {
4450 ret = dev_ifconf(net, (char __user *) arg);
4454 if (cmd == SIOCGIFNAME)
4455 return dev_ifname(net, (struct ifreq __user *)arg);
4457 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4460 ifr.ifr_name[IFNAMSIZ-1] = 0;
4462 colon = strchr(ifr.ifr_name, ':');
4467 * See which interface the caller is talking about.
4472 * These ioctl calls:
4473 * - can be done by all.
4474 * - atomic and do not require locking.
4485 dev_load(net, ifr.ifr_name);
4486 read_lock(&dev_base_lock);
4487 ret = dev_ifsioc_locked(net, &ifr, cmd);
4488 read_unlock(&dev_base_lock);
4492 if (copy_to_user(arg, &ifr,
4493 sizeof(struct ifreq)))
4499 dev_load(net, ifr.ifr_name);
4501 ret = dev_ethtool(net, &ifr);
4506 if (copy_to_user(arg, &ifr,
4507 sizeof(struct ifreq)))
4513 * These ioctl calls:
4514 * - require superuser power.
4515 * - require strict serialization.
4521 if (!capable(CAP_NET_ADMIN))
4523 dev_load(net, ifr.ifr_name);
4525 ret = dev_ifsioc(net, &ifr, cmd);
4530 if (copy_to_user(arg, &ifr,
4531 sizeof(struct ifreq)))
4537 * These ioctl calls:
4538 * - require superuser power.
4539 * - require strict serialization.
4540 * - do not return a value
4550 case SIOCSIFHWBROADCAST:
4553 case SIOCBONDENSLAVE:
4554 case SIOCBONDRELEASE:
4555 case SIOCBONDSETHWADDR:
4556 case SIOCBONDCHANGEACTIVE:
4560 if (!capable(CAP_NET_ADMIN))
4563 case SIOCBONDSLAVEINFOQUERY:
4564 case SIOCBONDINFOQUERY:
4565 dev_load(net, ifr.ifr_name);
4567 ret = dev_ifsioc(net, &ifr, cmd);
4572 /* Get the per device memory space. We can add this but
4573 * currently do not support it */
4575 /* Set the per device memory buffer space.
4576 * Not applicable in our case */
4581 * Unknown or private ioctl.
4584 if (cmd == SIOCWANDEV ||
4585 (cmd >= SIOCDEVPRIVATE &&
4586 cmd <= SIOCDEVPRIVATE + 15)) {
4587 dev_load(net, ifr.ifr_name);
4589 ret = dev_ifsioc(net, &ifr, cmd);
4591 if (!ret && copy_to_user(arg, &ifr,
4592 sizeof(struct ifreq)))
4596 /* Take care of Wireless Extensions */
4597 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4598 return wext_handle_ioctl(net, &ifr, cmd, arg);
4605 * dev_new_index - allocate an ifindex
4606 * @net: the applicable net namespace
4608 * Returns a suitable unique value for a new device interface
4609 * number. The caller must hold the rtnl semaphore or the
4610 * dev_base_lock to be sure it remains unique.
4612 static int dev_new_index(struct net *net)
4618 if (!__dev_get_by_index(net, ifindex))
4623 /* Delayed registration/unregisteration */
4624 static LIST_HEAD(net_todo_list);
4626 static void net_set_todo(struct net_device *dev)
4628 list_add_tail(&dev->todo_list, &net_todo_list);
4631 static void rollback_registered(struct net_device *dev)
4633 BUG_ON(dev_boot_phase);
4636 /* Some devices call without registering for initialization unwind. */
4637 if (dev->reg_state == NETREG_UNINITIALIZED) {
4638 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
4639 "was registered\n", dev->name, dev);
4645 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4647 /* If device is running, close it first. */
4650 /* And unlink it from device chain. */
4651 unlist_netdevice(dev);
4653 dev->reg_state = NETREG_UNREGISTERING;
4657 /* Shutdown queueing discipline. */
4661 /* Notify protocols, that we are about to destroy
4662 this device. They should clean all the things.
4664 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4667 * Flush the unicast and multicast chains
4669 dev_unicast_flush(dev);
4670 dev_addr_discard(dev);
4672 if (dev->netdev_ops->ndo_uninit)
4673 dev->netdev_ops->ndo_uninit(dev);
4675 /* Notifier chain MUST detach us from master device. */
4676 WARN_ON(dev->master);
4678 /* Remove entries from kobject tree */
4679 netdev_unregister_kobject(dev);
4686 static void __netdev_init_queue_locks_one(struct net_device *dev,
4687 struct netdev_queue *dev_queue,
4690 spin_lock_init(&dev_queue->_xmit_lock);
4691 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4692 dev_queue->xmit_lock_owner = -1;
4695 static void netdev_init_queue_locks(struct net_device *dev)
4697 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4698 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4701 unsigned long netdev_fix_features(unsigned long features, const char *name)
4703 /* Fix illegal SG+CSUM combinations. */
4704 if ((features & NETIF_F_SG) &&
4705 !(features & NETIF_F_ALL_CSUM)) {
4707 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4708 "checksum feature.\n", name);
4709 features &= ~NETIF_F_SG;
4712 /* TSO requires that SG is present as well. */
4713 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4715 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4716 "SG feature.\n", name);
4717 features &= ~NETIF_F_TSO;
4720 if (features & NETIF_F_UFO) {
4721 if (!(features & NETIF_F_GEN_CSUM)) {
4723 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4724 "since no NETIF_F_HW_CSUM feature.\n",
4726 features &= ~NETIF_F_UFO;
4729 if (!(features & NETIF_F_SG)) {
4731 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4732 "since no NETIF_F_SG feature.\n", name);
4733 features &= ~NETIF_F_UFO;
4739 EXPORT_SYMBOL(netdev_fix_features);
4742 * register_netdevice - register a network device
4743 * @dev: device to register
4745 * Take a completed network device structure and add it to the kernel
4746 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4747 * chain. 0 is returned on success. A negative errno code is returned
4748 * on a failure to set up the device, or if the name is a duplicate.
4750 * Callers must hold the rtnl semaphore. You may want
4751 * register_netdev() instead of this.
4754 * The locking appears insufficient to guarantee two parallel registers
4755 * will not get the same name.
4758 int register_netdevice(struct net_device *dev)
4760 struct hlist_head *head;
4761 struct hlist_node *p;
4763 struct net *net = dev_net(dev);
4765 BUG_ON(dev_boot_phase);
4770 /* When net_device's are persistent, this will be fatal. */
4771 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4774 spin_lock_init(&dev->addr_list_lock);
4775 netdev_set_addr_lockdep_class(dev);
4776 netdev_init_queue_locks(dev);
4780 /* Init, if this function is available */
4781 if (dev->netdev_ops->ndo_init) {
4782 ret = dev->netdev_ops->ndo_init(dev);
4790 if (!dev_valid_name(dev->name)) {
4795 dev->ifindex = dev_new_index(net);
4796 if (dev->iflink == -1)
4797 dev->iflink = dev->ifindex;
4799 /* Check for existence of name */
4800 head = dev_name_hash(net, dev->name);
4801 hlist_for_each(p, head) {
4802 struct net_device *d
4803 = hlist_entry(p, struct net_device, name_hlist);
4804 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4810 /* Fix illegal checksum combinations */
4811 if ((dev->features & NETIF_F_HW_CSUM) &&
4812 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4813 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4815 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4818 if ((dev->features & NETIF_F_NO_CSUM) &&
4819 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4820 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4822 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4825 dev->features = netdev_fix_features(dev->features, dev->name);
4827 /* Enable software GSO if SG is supported. */
4828 if (dev->features & NETIF_F_SG)
4829 dev->features |= NETIF_F_GSO;
4831 netdev_initialize_kobject(dev);
4833 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4834 ret = notifier_to_errno(ret);
4838 ret = netdev_register_kobject(dev);
4841 dev->reg_state = NETREG_REGISTERED;
4844 * Default initial state at registry is that the
4845 * device is present.
4848 set_bit(__LINK_STATE_PRESENT, &dev->state);
4850 dev_init_scheduler(dev);
4852 list_netdevice(dev);
4854 /* Notify protocols, that a new device appeared. */
4855 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4856 ret = notifier_to_errno(ret);
4858 rollback_registered(dev);
4859 dev->reg_state = NETREG_UNREGISTERED;
4866 if (dev->netdev_ops->ndo_uninit)
4867 dev->netdev_ops->ndo_uninit(dev);
4870 EXPORT_SYMBOL(register_netdevice);
4873 * init_dummy_netdev - init a dummy network device for NAPI
4874 * @dev: device to init
4876 * This takes a network device structure and initialize the minimum
4877 * amount of fields so it can be used to schedule NAPI polls without
4878 * registering a full blown interface. This is to be used by drivers
4879 * that need to tie several hardware interfaces to a single NAPI
4880 * poll scheduler due to HW limitations.
4882 int init_dummy_netdev(struct net_device *dev)
4884 /* Clear everything. Note we don't initialize spinlocks
4885 * are they aren't supposed to be taken by any of the
4886 * NAPI code and this dummy netdev is supposed to be
4887 * only ever used for NAPI polls
4889 memset(dev, 0, sizeof(struct net_device));
4891 /* make sure we BUG if trying to hit standard
4892 * register/unregister code path
4894 dev->reg_state = NETREG_DUMMY;
4896 /* initialize the ref count */
4897 atomic_set(&dev->refcnt, 1);
4899 /* NAPI wants this */
4900 INIT_LIST_HEAD(&dev->napi_list);
4902 /* a dummy interface is started by default */
4903 set_bit(__LINK_STATE_PRESENT, &dev->state);
4904 set_bit(__LINK_STATE_START, &dev->state);
4908 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4912 * register_netdev - register a network device
4913 * @dev: device to register
4915 * Take a completed network device structure and add it to the kernel
4916 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4917 * chain. 0 is returned on success. A negative errno code is returned
4918 * on a failure to set up the device, or if the name is a duplicate.
4920 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4921 * and expands the device name if you passed a format string to
4924 int register_netdev(struct net_device *dev)
4931 * If the name is a format string the caller wants us to do a
4934 if (strchr(dev->name, '%')) {
4935 err = dev_alloc_name(dev, dev->name);
4940 err = register_netdevice(dev);
4945 EXPORT_SYMBOL(register_netdev);
4948 * netdev_wait_allrefs - wait until all references are gone.
4950 * This is called when unregistering network devices.
4952 * Any protocol or device that holds a reference should register
4953 * for netdevice notification, and cleanup and put back the
4954 * reference if they receive an UNREGISTER event.
4955 * We can get stuck here if buggy protocols don't correctly
4958 static void netdev_wait_allrefs(struct net_device *dev)
4960 unsigned long rebroadcast_time, warning_time;
4962 rebroadcast_time = warning_time = jiffies;
4963 while (atomic_read(&dev->refcnt) != 0) {
4964 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4967 /* Rebroadcast unregister notification */
4968 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4970 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4972 /* We must not have linkwatch events
4973 * pending on unregister. If this
4974 * happens, we simply run the queue
4975 * unscheduled, resulting in a noop
4978 linkwatch_run_queue();
4983 rebroadcast_time = jiffies;
4988 if (time_after(jiffies, warning_time + 10 * HZ)) {
4989 printk(KERN_EMERG "unregister_netdevice: "
4990 "waiting for %s to become free. Usage "
4992 dev->name, atomic_read(&dev->refcnt));
4993 warning_time = jiffies;
5002 * register_netdevice(x1);
5003 * register_netdevice(x2);
5005 * unregister_netdevice(y1);
5006 * unregister_netdevice(y2);
5012 * We are invoked by rtnl_unlock().
5013 * This allows us to deal with problems:
5014 * 1) We can delete sysfs objects which invoke hotplug
5015 * without deadlocking with linkwatch via keventd.
5016 * 2) Since we run with the RTNL semaphore not held, we can sleep
5017 * safely in order to wait for the netdev refcnt to drop to zero.
5019 * We must not return until all unregister events added during
5020 * the interval the lock was held have been completed.
5022 void netdev_run_todo(void)
5024 struct list_head list;
5026 /* Snapshot list, allow later requests */
5027 list_replace_init(&net_todo_list, &list);
5031 while (!list_empty(&list)) {
5032 struct net_device *dev
5033 = list_entry(list.next, struct net_device, todo_list);
5034 list_del(&dev->todo_list);
5036 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5037 printk(KERN_ERR "network todo '%s' but state %d\n",
5038 dev->name, dev->reg_state);
5043 dev->reg_state = NETREG_UNREGISTERED;
5045 on_each_cpu(flush_backlog, dev, 1);
5047 netdev_wait_allrefs(dev);
5050 BUG_ON(atomic_read(&dev->refcnt));
5051 WARN_ON(dev->ip_ptr);
5052 WARN_ON(dev->ip6_ptr);
5053 WARN_ON(dev->dn_ptr);
5055 if (dev->destructor)
5056 dev->destructor(dev);
5058 /* Free network device */
5059 kobject_put(&dev->dev.kobj);
5064 * dev_get_stats - get network device statistics
5065 * @dev: device to get statistics from
5067 * Get network statistics from device. The device driver may provide
5068 * its own method by setting dev->netdev_ops->get_stats; otherwise
5069 * the internal statistics structure is used.
5071 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5073 const struct net_device_ops *ops = dev->netdev_ops;
5075 if (ops->ndo_get_stats)
5076 return ops->ndo_get_stats(dev);
5078 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5079 struct net_device_stats *stats = &dev->stats;
5081 struct netdev_queue *txq;
5083 for (i = 0; i < dev->num_tx_queues; i++) {
5084 txq = netdev_get_tx_queue(dev, i);
5085 tx_bytes += txq->tx_bytes;
5086 tx_packets += txq->tx_packets;
5087 tx_dropped += txq->tx_dropped;
5089 if (tx_bytes || tx_packets || tx_dropped) {
5090 stats->tx_bytes = tx_bytes;
5091 stats->tx_packets = tx_packets;
5092 stats->tx_dropped = tx_dropped;
5097 EXPORT_SYMBOL(dev_get_stats);
5099 static void netdev_init_one_queue(struct net_device *dev,
5100 struct netdev_queue *queue,
5106 static void netdev_init_queues(struct net_device *dev)
5108 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5109 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5110 spin_lock_init(&dev->tx_global_lock);
5114 * alloc_netdev_mq - allocate network device
5115 * @sizeof_priv: size of private data to allocate space for
5116 * @name: device name format string
5117 * @setup: callback to initialize device
5118 * @queue_count: the number of subqueues to allocate
5120 * Allocates a struct net_device with private data area for driver use
5121 * and performs basic initialization. Also allocates subquue structs
5122 * for each queue on the device at the end of the netdevice.
5124 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5125 void (*setup)(struct net_device *), unsigned int queue_count)
5127 struct netdev_queue *tx;
5128 struct net_device *dev;
5130 struct net_device *p;
5132 BUG_ON(strlen(name) >= sizeof(dev->name));
5134 alloc_size = sizeof(struct net_device);
5136 /* ensure 32-byte alignment of private area */
5137 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5138 alloc_size += sizeof_priv;
5140 /* ensure 32-byte alignment of whole construct */
5141 alloc_size += NETDEV_ALIGN - 1;
5143 p = kzalloc(alloc_size, GFP_KERNEL);
5145 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5149 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5151 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5156 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5157 dev->padded = (char *)dev - (char *)p;
5159 if (dev_addr_init(dev))
5162 dev_unicast_init(dev);
5164 dev_net_set(dev, &init_net);
5167 dev->num_tx_queues = queue_count;
5168 dev->real_num_tx_queues = queue_count;
5170 dev->gso_max_size = GSO_MAX_SIZE;
5172 netdev_init_queues(dev);
5174 INIT_LIST_HEAD(&dev->napi_list);
5175 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5177 strcpy(dev->name, name);
5187 EXPORT_SYMBOL(alloc_netdev_mq);
5190 * free_netdev - free network device
5193 * This function does the last stage of destroying an allocated device
5194 * interface. The reference to the device object is released.
5195 * If this is the last reference then it will be freed.
5197 void free_netdev(struct net_device *dev)
5199 struct napi_struct *p, *n;
5201 release_net(dev_net(dev));
5205 /* Flush device addresses */
5206 dev_addr_flush(dev);
5208 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5211 /* Compatibility with error handling in drivers */
5212 if (dev->reg_state == NETREG_UNINITIALIZED) {
5213 kfree((char *)dev - dev->padded);
5217 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5218 dev->reg_state = NETREG_RELEASED;
5220 /* will free via device release */
5221 put_device(&dev->dev);
5223 EXPORT_SYMBOL(free_netdev);
5226 * synchronize_net - Synchronize with packet receive processing
5228 * Wait for packets currently being received to be done.
5229 * Does not block later packets from starting.
5231 void synchronize_net(void)
5236 EXPORT_SYMBOL(synchronize_net);
5239 * unregister_netdevice - remove device from the kernel
5242 * This function shuts down a device interface and removes it
5243 * from the kernel tables.
5245 * Callers must hold the rtnl semaphore. You may want
5246 * unregister_netdev() instead of this.
5249 void unregister_netdevice(struct net_device *dev)
5253 rollback_registered(dev);
5254 /* Finish processing unregister after unlock */
5257 EXPORT_SYMBOL(unregister_netdevice);
5260 * unregister_netdev - remove device from the kernel
5263 * This function shuts down a device interface and removes it
5264 * from the kernel tables.
5266 * This is just a wrapper for unregister_netdevice that takes
5267 * the rtnl semaphore. In general you want to use this and not
5268 * unregister_netdevice.
5270 void unregister_netdev(struct net_device *dev)
5273 unregister_netdevice(dev);
5276 EXPORT_SYMBOL(unregister_netdev);
5279 * dev_change_net_namespace - move device to different nethost namespace
5281 * @net: network namespace
5282 * @pat: If not NULL name pattern to try if the current device name
5283 * is already taken in the destination network namespace.
5285 * This function shuts down a device interface and moves it
5286 * to a new network namespace. On success 0 is returned, on
5287 * a failure a netagive errno code is returned.
5289 * Callers must hold the rtnl semaphore.
5292 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5295 const char *destname;
5300 /* Don't allow namespace local devices to be moved. */
5302 if (dev->features & NETIF_F_NETNS_LOCAL)
5306 /* Don't allow real devices to be moved when sysfs
5310 if (dev->dev.parent)
5314 /* Ensure the device has been registrered */
5316 if (dev->reg_state != NETREG_REGISTERED)
5319 /* Get out if there is nothing todo */
5321 if (net_eq(dev_net(dev), net))
5324 /* Pick the destination device name, and ensure
5325 * we can use it in the destination network namespace.
5328 destname = dev->name;
5329 if (__dev_get_by_name(net, destname)) {
5330 /* We get here if we can't use the current device name */
5333 if (!dev_valid_name(pat))
5335 if (strchr(pat, '%')) {
5336 if (__dev_alloc_name(net, pat, buf) < 0)
5341 if (__dev_get_by_name(net, destname))
5346 * And now a mini version of register_netdevice unregister_netdevice.
5349 /* If device is running close it first. */
5352 /* And unlink it from device chain */
5354 unlist_netdevice(dev);
5358 /* Shutdown queueing discipline. */
5361 /* Notify protocols, that we are about to destroy
5362 this device. They should clean all the things.
5364 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5367 * Flush the unicast and multicast chains
5369 dev_unicast_flush(dev);
5370 dev_addr_discard(dev);
5372 netdev_unregister_kobject(dev);
5374 /* Actually switch the network namespace */
5375 dev_net_set(dev, net);
5377 /* Assign the new device name */
5378 if (destname != dev->name)
5379 strcpy(dev->name, destname);
5381 /* If there is an ifindex conflict assign a new one */
5382 if (__dev_get_by_index(net, dev->ifindex)) {
5383 int iflink = (dev->iflink == dev->ifindex);
5384 dev->ifindex = dev_new_index(net);
5386 dev->iflink = dev->ifindex;
5389 /* Fixup kobjects */
5390 err = netdev_register_kobject(dev);
5393 /* Add the device back in the hashes */
5394 list_netdevice(dev);
5396 /* Notify protocols, that a new device appeared. */
5397 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5404 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5406 static int dev_cpu_callback(struct notifier_block *nfb,
5407 unsigned long action,
5410 struct sk_buff **list_skb;
5411 struct Qdisc **list_net;
5412 struct sk_buff *skb;
5413 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5414 struct softnet_data *sd, *oldsd;
5416 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5419 local_irq_disable();
5420 cpu = smp_processor_id();
5421 sd = &per_cpu(softnet_data, cpu);
5422 oldsd = &per_cpu(softnet_data, oldcpu);
5424 /* Find end of our completion_queue. */
5425 list_skb = &sd->completion_queue;
5427 list_skb = &(*list_skb)->next;
5428 /* Append completion queue from offline CPU. */
5429 *list_skb = oldsd->completion_queue;
5430 oldsd->completion_queue = NULL;
5432 /* Find end of our output_queue. */
5433 list_net = &sd->output_queue;
5435 list_net = &(*list_net)->next_sched;
5436 /* Append output queue from offline CPU. */
5437 *list_net = oldsd->output_queue;
5438 oldsd->output_queue = NULL;
5440 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5443 /* Process offline CPU's input_pkt_queue */
5444 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5452 * netdev_increment_features - increment feature set by one
5453 * @all: current feature set
5454 * @one: new feature set
5455 * @mask: mask feature set
5457 * Computes a new feature set after adding a device with feature set
5458 * @one to the master device with current feature set @all. Will not
5459 * enable anything that is off in @mask. Returns the new feature set.
5461 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5464 /* If device needs checksumming, downgrade to it. */
5465 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5466 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5467 else if (mask & NETIF_F_ALL_CSUM) {
5468 /* If one device supports v4/v6 checksumming, set for all. */
5469 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5470 !(all & NETIF_F_GEN_CSUM)) {
5471 all &= ~NETIF_F_ALL_CSUM;
5472 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5475 /* If one device supports hw checksumming, set for all. */
5476 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5477 all &= ~NETIF_F_ALL_CSUM;
5478 all |= NETIF_F_HW_CSUM;
5482 one |= NETIF_F_ALL_CSUM;
5484 one |= all & NETIF_F_ONE_FOR_ALL;
5485 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5486 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5490 EXPORT_SYMBOL(netdev_increment_features);
5492 static struct hlist_head *netdev_create_hash(void)
5495 struct hlist_head *hash;
5497 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5499 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5500 INIT_HLIST_HEAD(&hash[i]);
5505 /* Initialize per network namespace state */
5506 static int __net_init netdev_init(struct net *net)
5508 INIT_LIST_HEAD(&net->dev_base_head);
5510 net->dev_name_head = netdev_create_hash();
5511 if (net->dev_name_head == NULL)
5514 net->dev_index_head = netdev_create_hash();
5515 if (net->dev_index_head == NULL)
5521 kfree(net->dev_name_head);
5527 * netdev_drivername - network driver for the device
5528 * @dev: network device
5529 * @buffer: buffer for resulting name
5530 * @len: size of buffer
5532 * Determine network driver for device.
5534 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5536 const struct device_driver *driver;
5537 const struct device *parent;
5539 if (len <= 0 || !buffer)
5543 parent = dev->dev.parent;
5548 driver = parent->driver;
5549 if (driver && driver->name)
5550 strlcpy(buffer, driver->name, len);
5554 static void __net_exit netdev_exit(struct net *net)
5556 kfree(net->dev_name_head);
5557 kfree(net->dev_index_head);
5560 static struct pernet_operations __net_initdata netdev_net_ops = {
5561 .init = netdev_init,
5562 .exit = netdev_exit,
5565 static void __net_exit default_device_exit(struct net *net)
5567 struct net_device *dev;
5569 * Push all migratable of the network devices back to the
5570 * initial network namespace
5574 for_each_netdev(net, dev) {
5576 char fb_name[IFNAMSIZ];
5578 /* Ignore unmoveable devices (i.e. loopback) */
5579 if (dev->features & NETIF_F_NETNS_LOCAL)
5582 /* Delete virtual devices */
5583 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5584 dev->rtnl_link_ops->dellink(dev);
5588 /* Push remaing network devices to init_net */
5589 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5590 err = dev_change_net_namespace(dev, &init_net, fb_name);
5592 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5593 __func__, dev->name, err);
5601 static struct pernet_operations __net_initdata default_device_ops = {
5602 .exit = default_device_exit,
5606 * Initialize the DEV module. At boot time this walks the device list and
5607 * unhooks any devices that fail to initialise (normally hardware not
5608 * present) and leaves us with a valid list of present and active devices.
5613 * This is called single threaded during boot, so no need
5614 * to take the rtnl semaphore.
5616 static int __init net_dev_init(void)
5618 int i, rc = -ENOMEM;
5620 BUG_ON(!dev_boot_phase);
5622 if (dev_proc_init())
5625 if (netdev_kobject_init())
5628 INIT_LIST_HEAD(&ptype_all);
5629 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5630 INIT_LIST_HEAD(&ptype_base[i]);
5632 if (register_pernet_subsys(&netdev_net_ops))
5636 * Initialise the packet receive queues.
5639 for_each_possible_cpu(i) {
5640 struct softnet_data *queue;
5642 queue = &per_cpu(softnet_data, i);
5643 skb_queue_head_init(&queue->input_pkt_queue);
5644 queue->completion_queue = NULL;
5645 INIT_LIST_HEAD(&queue->poll_list);
5647 queue->backlog.poll = process_backlog;
5648 queue->backlog.weight = weight_p;
5649 queue->backlog.gro_list = NULL;
5650 queue->backlog.gro_count = 0;
5655 /* The loopback device is special if any other network devices
5656 * is present in a network namespace the loopback device must
5657 * be present. Since we now dynamically allocate and free the
5658 * loopback device ensure this invariant is maintained by
5659 * keeping the loopback device as the first device on the
5660 * list of network devices. Ensuring the loopback devices
5661 * is the first device that appears and the last network device
5664 if (register_pernet_device(&loopback_net_ops))
5667 if (register_pernet_device(&default_device_ops))
5670 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5671 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5673 hotcpu_notifier(dev_cpu_callback, 0);
5681 subsys_initcall(net_dev_init);
5683 static int __init initialize_hashrnd(void)
5685 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5689 late_initcall_sync(initialize_hashrnd);