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)
1795 struct sock *sk = skb->sk;
1797 if (sk_tx_queue_recorded(sk)) {
1798 queue_index = sk_tx_queue_get(sk);
1800 const struct net_device_ops *ops = dev->netdev_ops;
1802 if (ops->ndo_select_queue) {
1803 queue_index = ops->ndo_select_queue(dev, skb);
1806 if (dev->real_num_tx_queues > 1)
1807 queue_index = skb_tx_hash(dev, skb);
1809 if (sk && sk->sk_dst_cache)
1810 sk_tx_queue_set(sk, queue_index);
1814 skb_set_queue_mapping(skb, queue_index);
1815 return netdev_get_tx_queue(dev, queue_index);
1818 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1819 struct net_device *dev,
1820 struct netdev_queue *txq)
1822 spinlock_t *root_lock = qdisc_lock(q);
1825 spin_lock(root_lock);
1826 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1829 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1830 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1832 * This is a work-conserving queue; there are no old skbs
1833 * waiting to be sent out; and the qdisc is not running -
1834 * xmit the skb directly.
1836 __qdisc_update_bstats(q, skb->len);
1837 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1840 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1842 rc = NET_XMIT_SUCCESS;
1844 rc = qdisc_enqueue_root(skb, q);
1847 spin_unlock(root_lock);
1853 * dev_queue_xmit - transmit a buffer
1854 * @skb: buffer to transmit
1856 * Queue a buffer for transmission to a network device. The caller must
1857 * have set the device and priority and built the buffer before calling
1858 * this function. The function can be called from an interrupt.
1860 * A negative errno code is returned on a failure. A success does not
1861 * guarantee the frame will be transmitted as it may be dropped due
1862 * to congestion or traffic shaping.
1864 * -----------------------------------------------------------------------------------
1865 * I notice this method can also return errors from the queue disciplines,
1866 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1869 * Regardless of the return value, the skb is consumed, so it is currently
1870 * difficult to retry a send to this method. (You can bump the ref count
1871 * before sending to hold a reference for retry if you are careful.)
1873 * When calling this method, interrupts MUST be enabled. This is because
1874 * the BH enable code must have IRQs enabled so that it will not deadlock.
1877 int dev_queue_xmit(struct sk_buff *skb)
1879 struct net_device *dev = skb->dev;
1880 struct netdev_queue *txq;
1884 /* GSO will handle the following emulations directly. */
1885 if (netif_needs_gso(dev, skb))
1888 if (skb_has_frags(skb) &&
1889 !(dev->features & NETIF_F_FRAGLIST) &&
1890 __skb_linearize(skb))
1893 /* Fragmented skb is linearized if device does not support SG,
1894 * or if at least one of fragments is in highmem and device
1895 * does not support DMA from it.
1897 if (skb_shinfo(skb)->nr_frags &&
1898 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1899 __skb_linearize(skb))
1902 /* If packet is not checksummed and device does not support
1903 * checksumming for this protocol, complete checksumming here.
1905 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1906 skb_set_transport_header(skb, skb->csum_start -
1908 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1913 /* Disable soft irqs for various locks below. Also
1914 * stops preemption for RCU.
1918 txq = dev_pick_tx(dev, skb);
1919 q = rcu_dereference(txq->qdisc);
1921 #ifdef CONFIG_NET_CLS_ACT
1922 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
1925 rc = __dev_xmit_skb(skb, q, dev, txq);
1929 /* The device has no queue. Common case for software devices:
1930 loopback, all the sorts of tunnels...
1932 Really, it is unlikely that netif_tx_lock protection is necessary
1933 here. (f.e. loopback and IP tunnels are clean ignoring statistics
1935 However, it is possible, that they rely on protection
1938 Check this and shot the lock. It is not prone from deadlocks.
1939 Either shot noqueue qdisc, it is even simpler 8)
1941 if (dev->flags & IFF_UP) {
1942 int cpu = smp_processor_id(); /* ok because BHs are off */
1944 if (txq->xmit_lock_owner != cpu) {
1946 HARD_TX_LOCK(dev, txq, cpu);
1948 if (!netif_tx_queue_stopped(txq)) {
1949 rc = NET_XMIT_SUCCESS;
1950 if (!dev_hard_start_xmit(skb, dev, txq)) {
1951 HARD_TX_UNLOCK(dev, txq);
1955 HARD_TX_UNLOCK(dev, txq);
1956 if (net_ratelimit())
1957 printk(KERN_CRIT "Virtual device %s asks to "
1958 "queue packet!\n", dev->name);
1960 /* Recursion is detected! It is possible,
1962 if (net_ratelimit())
1963 printk(KERN_CRIT "Dead loop on virtual device "
1964 "%s, fix it urgently!\n", dev->name);
1969 rcu_read_unlock_bh();
1975 rcu_read_unlock_bh();
1978 EXPORT_SYMBOL(dev_queue_xmit);
1981 /*=======================================================================
1983 =======================================================================*/
1985 int netdev_max_backlog __read_mostly = 1000;
1986 int netdev_budget __read_mostly = 300;
1987 int weight_p __read_mostly = 64; /* old backlog weight */
1989 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
1993 * netif_rx - post buffer to the network code
1994 * @skb: buffer to post
1996 * This function receives a packet from a device driver and queues it for
1997 * the upper (protocol) levels to process. It always succeeds. The buffer
1998 * may be dropped during processing for congestion control or by the
2002 * NET_RX_SUCCESS (no congestion)
2003 * NET_RX_DROP (packet was dropped)
2007 int netif_rx(struct sk_buff *skb)
2009 struct softnet_data *queue;
2010 unsigned long flags;
2012 /* if netpoll wants it, pretend we never saw it */
2013 if (netpoll_rx(skb))
2016 if (!skb->tstamp.tv64)
2020 * The code is rearranged so that the path is the most
2021 * short when CPU is congested, but is still operating.
2023 local_irq_save(flags);
2024 queue = &__get_cpu_var(softnet_data);
2026 __get_cpu_var(netdev_rx_stat).total++;
2027 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2028 if (queue->input_pkt_queue.qlen) {
2030 __skb_queue_tail(&queue->input_pkt_queue, skb);
2031 local_irq_restore(flags);
2032 return NET_RX_SUCCESS;
2035 napi_schedule(&queue->backlog);
2039 __get_cpu_var(netdev_rx_stat).dropped++;
2040 local_irq_restore(flags);
2045 EXPORT_SYMBOL(netif_rx);
2047 int netif_rx_ni(struct sk_buff *skb)
2052 err = netif_rx(skb);
2053 if (local_softirq_pending())
2059 EXPORT_SYMBOL(netif_rx_ni);
2061 static void net_tx_action(struct softirq_action *h)
2063 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2065 if (sd->completion_queue) {
2066 struct sk_buff *clist;
2068 local_irq_disable();
2069 clist = sd->completion_queue;
2070 sd->completion_queue = NULL;
2074 struct sk_buff *skb = clist;
2075 clist = clist->next;
2077 WARN_ON(atomic_read(&skb->users));
2082 if (sd->output_queue) {
2085 local_irq_disable();
2086 head = sd->output_queue;
2087 sd->output_queue = NULL;
2091 struct Qdisc *q = head;
2092 spinlock_t *root_lock;
2094 head = head->next_sched;
2096 root_lock = qdisc_lock(q);
2097 if (spin_trylock(root_lock)) {
2098 smp_mb__before_clear_bit();
2099 clear_bit(__QDISC_STATE_SCHED,
2102 spin_unlock(root_lock);
2104 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2106 __netif_reschedule(q);
2108 smp_mb__before_clear_bit();
2109 clear_bit(__QDISC_STATE_SCHED,
2117 static inline int deliver_skb(struct sk_buff *skb,
2118 struct packet_type *pt_prev,
2119 struct net_device *orig_dev)
2121 atomic_inc(&skb->users);
2122 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2125 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2127 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2128 /* This hook is defined here for ATM LANE */
2129 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2130 unsigned char *addr) __read_mostly;
2131 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2135 * If bridge module is loaded call bridging hook.
2136 * returns NULL if packet was consumed.
2138 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2139 struct sk_buff *skb) __read_mostly;
2140 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2142 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2143 struct packet_type **pt_prev, int *ret,
2144 struct net_device *orig_dev)
2146 struct net_bridge_port *port;
2148 if (skb->pkt_type == PACKET_LOOPBACK ||
2149 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2153 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2157 return br_handle_frame_hook(port, skb);
2160 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2163 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2164 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2165 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2167 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2168 struct packet_type **pt_prev,
2170 struct net_device *orig_dev)
2172 if (skb->dev->macvlan_port == NULL)
2176 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2179 return macvlan_handle_frame_hook(skb);
2182 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2185 #ifdef CONFIG_NET_CLS_ACT
2186 /* TODO: Maybe we should just force sch_ingress to be compiled in
2187 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2188 * a compare and 2 stores extra right now if we dont have it on
2189 * but have CONFIG_NET_CLS_ACT
2190 * NOTE: This doesnt stop any functionality; if you dont have
2191 * the ingress scheduler, you just cant add policies on ingress.
2194 static int ing_filter(struct sk_buff *skb)
2196 struct net_device *dev = skb->dev;
2197 u32 ttl = G_TC_RTTL(skb->tc_verd);
2198 struct netdev_queue *rxq;
2199 int result = TC_ACT_OK;
2202 if (MAX_RED_LOOP < ttl++) {
2204 "Redir loop detected Dropping packet (%d->%d)\n",
2205 skb->iif, dev->ifindex);
2209 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2210 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2212 rxq = &dev->rx_queue;
2215 if (q != &noop_qdisc) {
2216 spin_lock(qdisc_lock(q));
2217 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2218 result = qdisc_enqueue_root(skb, q);
2219 spin_unlock(qdisc_lock(q));
2225 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2226 struct packet_type **pt_prev,
2227 int *ret, struct net_device *orig_dev)
2229 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2233 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2236 /* Huh? Why does turning on AF_PACKET affect this? */
2237 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2240 switch (ing_filter(skb)) {
2254 * netif_nit_deliver - deliver received packets to network taps
2257 * This function is used to deliver incoming packets to network
2258 * taps. It should be used when the normal netif_receive_skb path
2259 * is bypassed, for example because of VLAN acceleration.
2261 void netif_nit_deliver(struct sk_buff *skb)
2263 struct packet_type *ptype;
2265 if (list_empty(&ptype_all))
2268 skb_reset_network_header(skb);
2269 skb_reset_transport_header(skb);
2270 skb->mac_len = skb->network_header - skb->mac_header;
2273 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2274 if (!ptype->dev || ptype->dev == skb->dev)
2275 deliver_skb(skb, ptype, skb->dev);
2281 * netif_receive_skb - process receive buffer from network
2282 * @skb: buffer to process
2284 * netif_receive_skb() is the main receive data processing function.
2285 * It always succeeds. The buffer may be dropped during processing
2286 * for congestion control or by the protocol layers.
2288 * This function may only be called from softirq context and interrupts
2289 * should be enabled.
2291 * Return values (usually ignored):
2292 * NET_RX_SUCCESS: no congestion
2293 * NET_RX_DROP: packet was dropped
2295 int netif_receive_skb(struct sk_buff *skb)
2297 struct packet_type *ptype, *pt_prev;
2298 struct net_device *orig_dev;
2299 struct net_device *null_or_orig;
2300 int ret = NET_RX_DROP;
2303 if (!skb->tstamp.tv64)
2306 if (skb->vlan_tci && vlan_hwaccel_do_receive(skb))
2307 return NET_RX_SUCCESS;
2309 /* if we've gotten here through NAPI, check netpoll */
2310 if (netpoll_receive_skb(skb))
2314 skb->iif = skb->dev->ifindex;
2316 null_or_orig = NULL;
2317 orig_dev = skb->dev;
2318 if (orig_dev->master) {
2319 if (skb_bond_should_drop(skb))
2320 null_or_orig = orig_dev; /* deliver only exact match */
2322 skb->dev = orig_dev->master;
2325 __get_cpu_var(netdev_rx_stat).total++;
2327 skb_reset_network_header(skb);
2328 skb_reset_transport_header(skb);
2329 skb->mac_len = skb->network_header - skb->mac_header;
2335 #ifdef CONFIG_NET_CLS_ACT
2336 if (skb->tc_verd & TC_NCLS) {
2337 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2342 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2343 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2344 ptype->dev == orig_dev) {
2346 ret = deliver_skb(skb, pt_prev, orig_dev);
2351 #ifdef CONFIG_NET_CLS_ACT
2352 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2358 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2361 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2365 type = skb->protocol;
2366 list_for_each_entry_rcu(ptype,
2367 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2368 if (ptype->type == type &&
2369 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2370 ptype->dev == orig_dev)) {
2372 ret = deliver_skb(skb, pt_prev, orig_dev);
2378 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2381 /* Jamal, now you will not able to escape explaining
2382 * me how you were going to use this. :-)
2391 EXPORT_SYMBOL(netif_receive_skb);
2393 /* Network device is going away, flush any packets still pending */
2394 static void flush_backlog(void *arg)
2396 struct net_device *dev = arg;
2397 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2398 struct sk_buff *skb, *tmp;
2400 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2401 if (skb->dev == dev) {
2402 __skb_unlink(skb, &queue->input_pkt_queue);
2407 static int napi_gro_complete(struct sk_buff *skb)
2409 struct packet_type *ptype;
2410 __be16 type = skb->protocol;
2411 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2414 if (NAPI_GRO_CB(skb)->count == 1) {
2415 skb_shinfo(skb)->gso_size = 0;
2420 list_for_each_entry_rcu(ptype, head, list) {
2421 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2424 err = ptype->gro_complete(skb);
2430 WARN_ON(&ptype->list == head);
2432 return NET_RX_SUCCESS;
2436 return netif_receive_skb(skb);
2439 void napi_gro_flush(struct napi_struct *napi)
2441 struct sk_buff *skb, *next;
2443 for (skb = napi->gro_list; skb; skb = next) {
2446 napi_gro_complete(skb);
2449 napi->gro_count = 0;
2450 napi->gro_list = NULL;
2452 EXPORT_SYMBOL(napi_gro_flush);
2454 int dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2456 struct sk_buff **pp = NULL;
2457 struct packet_type *ptype;
2458 __be16 type = skb->protocol;
2459 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2464 if (!(skb->dev->features & NETIF_F_GRO))
2467 if (skb_is_gso(skb) || skb_has_frags(skb))
2471 list_for_each_entry_rcu(ptype, head, list) {
2472 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2475 skb_set_network_header(skb, skb_gro_offset(skb));
2476 mac_len = skb->network_header - skb->mac_header;
2477 skb->mac_len = mac_len;
2478 NAPI_GRO_CB(skb)->same_flow = 0;
2479 NAPI_GRO_CB(skb)->flush = 0;
2480 NAPI_GRO_CB(skb)->free = 0;
2482 pp = ptype->gro_receive(&napi->gro_list, skb);
2487 if (&ptype->list == head)
2490 same_flow = NAPI_GRO_CB(skb)->same_flow;
2491 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2494 struct sk_buff *nskb = *pp;
2498 napi_gro_complete(nskb);
2505 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2509 NAPI_GRO_CB(skb)->count = 1;
2510 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2511 skb->next = napi->gro_list;
2512 napi->gro_list = skb;
2516 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2517 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2519 BUG_ON(skb->end - skb->tail < grow);
2521 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2524 skb->data_len -= grow;
2526 skb_shinfo(skb)->frags[0].page_offset += grow;
2527 skb_shinfo(skb)->frags[0].size -= grow;
2529 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2530 put_page(skb_shinfo(skb)->frags[0].page);
2531 memmove(skb_shinfo(skb)->frags,
2532 skb_shinfo(skb)->frags + 1,
2533 --skb_shinfo(skb)->nr_frags);
2544 EXPORT_SYMBOL(dev_gro_receive);
2546 static int __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2550 if (netpoll_rx_on(skb))
2553 for (p = napi->gro_list; p; p = p->next) {
2554 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2555 && !compare_ether_header(skb_mac_header(p),
2556 skb_gro_mac_header(skb));
2557 NAPI_GRO_CB(p)->flush = 0;
2560 return dev_gro_receive(napi, skb);
2563 int napi_skb_finish(int ret, struct sk_buff *skb)
2565 int err = NET_RX_SUCCESS;
2569 return netif_receive_skb(skb);
2575 case GRO_MERGED_FREE:
2582 EXPORT_SYMBOL(napi_skb_finish);
2584 void skb_gro_reset_offset(struct sk_buff *skb)
2586 NAPI_GRO_CB(skb)->data_offset = 0;
2587 NAPI_GRO_CB(skb)->frag0 = NULL;
2588 NAPI_GRO_CB(skb)->frag0_len = 0;
2590 if (skb->mac_header == skb->tail &&
2591 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2592 NAPI_GRO_CB(skb)->frag0 =
2593 page_address(skb_shinfo(skb)->frags[0].page) +
2594 skb_shinfo(skb)->frags[0].page_offset;
2595 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2598 EXPORT_SYMBOL(skb_gro_reset_offset);
2600 int napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2602 skb_gro_reset_offset(skb);
2604 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2606 EXPORT_SYMBOL(napi_gro_receive);
2608 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2610 __skb_pull(skb, skb_headlen(skb));
2611 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2615 EXPORT_SYMBOL(napi_reuse_skb);
2617 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2619 struct sk_buff *skb = napi->skb;
2622 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2628 EXPORT_SYMBOL(napi_get_frags);
2630 int napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb, int ret)
2632 int err = NET_RX_SUCCESS;
2637 skb->protocol = eth_type_trans(skb, napi->dev);
2639 if (ret == GRO_NORMAL)
2640 return netif_receive_skb(skb);
2642 skb_gro_pull(skb, -ETH_HLEN);
2649 case GRO_MERGED_FREE:
2650 napi_reuse_skb(napi, skb);
2656 EXPORT_SYMBOL(napi_frags_finish);
2658 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2660 struct sk_buff *skb = napi->skb;
2667 skb_reset_mac_header(skb);
2668 skb_gro_reset_offset(skb);
2670 off = skb_gro_offset(skb);
2671 hlen = off + sizeof(*eth);
2672 eth = skb_gro_header_fast(skb, off);
2673 if (skb_gro_header_hard(skb, hlen)) {
2674 eth = skb_gro_header_slow(skb, hlen, off);
2675 if (unlikely(!eth)) {
2676 napi_reuse_skb(napi, skb);
2682 skb_gro_pull(skb, sizeof(*eth));
2685 * This works because the only protocols we care about don't require
2686 * special handling. We'll fix it up properly at the end.
2688 skb->protocol = eth->h_proto;
2693 EXPORT_SYMBOL(napi_frags_skb);
2695 int napi_gro_frags(struct napi_struct *napi)
2697 struct sk_buff *skb = napi_frags_skb(napi);
2702 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2704 EXPORT_SYMBOL(napi_gro_frags);
2706 static int process_backlog(struct napi_struct *napi, int quota)
2709 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2710 unsigned long start_time = jiffies;
2712 napi->weight = weight_p;
2714 struct sk_buff *skb;
2716 local_irq_disable();
2717 skb = __skb_dequeue(&queue->input_pkt_queue);
2719 __napi_complete(napi);
2725 netif_receive_skb(skb);
2726 } while (++work < quota && jiffies == start_time);
2732 * __napi_schedule - schedule for receive
2733 * @n: entry to schedule
2735 * The entry's receive function will be scheduled to run
2737 void __napi_schedule(struct napi_struct *n)
2739 unsigned long flags;
2741 local_irq_save(flags);
2742 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2743 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2744 local_irq_restore(flags);
2746 EXPORT_SYMBOL(__napi_schedule);
2748 void __napi_complete(struct napi_struct *n)
2750 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2751 BUG_ON(n->gro_list);
2753 list_del(&n->poll_list);
2754 smp_mb__before_clear_bit();
2755 clear_bit(NAPI_STATE_SCHED, &n->state);
2757 EXPORT_SYMBOL(__napi_complete);
2759 void napi_complete(struct napi_struct *n)
2761 unsigned long flags;
2764 * don't let napi dequeue from the cpu poll list
2765 * just in case its running on a different cpu
2767 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2771 local_irq_save(flags);
2773 local_irq_restore(flags);
2775 EXPORT_SYMBOL(napi_complete);
2777 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2778 int (*poll)(struct napi_struct *, int), int weight)
2780 INIT_LIST_HEAD(&napi->poll_list);
2781 napi->gro_count = 0;
2782 napi->gro_list = NULL;
2785 napi->weight = weight;
2786 list_add(&napi->dev_list, &dev->napi_list);
2788 #ifdef CONFIG_NETPOLL
2789 spin_lock_init(&napi->poll_lock);
2790 napi->poll_owner = -1;
2792 set_bit(NAPI_STATE_SCHED, &napi->state);
2794 EXPORT_SYMBOL(netif_napi_add);
2796 void netif_napi_del(struct napi_struct *napi)
2798 struct sk_buff *skb, *next;
2800 list_del_init(&napi->dev_list);
2801 napi_free_frags(napi);
2803 for (skb = napi->gro_list; skb; skb = next) {
2809 napi->gro_list = NULL;
2810 napi->gro_count = 0;
2812 EXPORT_SYMBOL(netif_napi_del);
2815 static void net_rx_action(struct softirq_action *h)
2817 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2818 unsigned long time_limit = jiffies + 2;
2819 int budget = netdev_budget;
2822 local_irq_disable();
2824 while (!list_empty(list)) {
2825 struct napi_struct *n;
2828 /* If softirq window is exhuasted then punt.
2829 * Allow this to run for 2 jiffies since which will allow
2830 * an average latency of 1.5/HZ.
2832 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2837 /* Even though interrupts have been re-enabled, this
2838 * access is safe because interrupts can only add new
2839 * entries to the tail of this list, and only ->poll()
2840 * calls can remove this head entry from the list.
2842 n = list_entry(list->next, struct napi_struct, poll_list);
2844 have = netpoll_poll_lock(n);
2848 /* This NAPI_STATE_SCHED test is for avoiding a race
2849 * with netpoll's poll_napi(). Only the entity which
2850 * obtains the lock and sees NAPI_STATE_SCHED set will
2851 * actually make the ->poll() call. Therefore we avoid
2852 * accidently calling ->poll() when NAPI is not scheduled.
2855 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2856 work = n->poll(n, weight);
2860 WARN_ON_ONCE(work > weight);
2864 local_irq_disable();
2866 /* Drivers must not modify the NAPI state if they
2867 * consume the entire weight. In such cases this code
2868 * still "owns" the NAPI instance and therefore can
2869 * move the instance around on the list at-will.
2871 if (unlikely(work == weight)) {
2872 if (unlikely(napi_disable_pending(n))) {
2875 local_irq_disable();
2877 list_move_tail(&n->poll_list, list);
2880 netpoll_poll_unlock(have);
2885 #ifdef CONFIG_NET_DMA
2887 * There may not be any more sk_buffs coming right now, so push
2888 * any pending DMA copies to hardware
2890 dma_issue_pending_all();
2896 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2897 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2901 static gifconf_func_t *gifconf_list[NPROTO];
2904 * register_gifconf - register a SIOCGIF handler
2905 * @family: Address family
2906 * @gifconf: Function handler
2908 * Register protocol dependent address dumping routines. The handler
2909 * that is passed must not be freed or reused until it has been replaced
2910 * by another handler.
2912 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
2914 if (family >= NPROTO)
2916 gifconf_list[family] = gifconf;
2919 EXPORT_SYMBOL(register_gifconf);
2923 * Map an interface index to its name (SIOCGIFNAME)
2927 * We need this ioctl for efficient implementation of the
2928 * if_indextoname() function required by the IPv6 API. Without
2929 * it, we would have to search all the interfaces to find a
2933 static int dev_ifname(struct net *net, struct ifreq __user *arg)
2935 struct net_device *dev;
2939 * Fetch the caller's info block.
2942 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
2945 read_lock(&dev_base_lock);
2946 dev = __dev_get_by_index(net, ifr.ifr_ifindex);
2948 read_unlock(&dev_base_lock);
2952 strcpy(ifr.ifr_name, dev->name);
2953 read_unlock(&dev_base_lock);
2955 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
2961 * Perform a SIOCGIFCONF call. This structure will change
2962 * size eventually, and there is nothing I can do about it.
2963 * Thus we will need a 'compatibility mode'.
2966 static int dev_ifconf(struct net *net, char __user *arg)
2969 struct net_device *dev;
2976 * Fetch the caller's info block.
2979 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
2986 * Loop over the interfaces, and write an info block for each.
2990 for_each_netdev(net, dev) {
2991 for (i = 0; i < NPROTO; i++) {
2992 if (gifconf_list[i]) {
2995 done = gifconf_list[i](dev, NULL, 0);
2997 done = gifconf_list[i](dev, pos + total,
3007 * All done. Write the updated control block back to the caller.
3009 ifc.ifc_len = total;
3012 * Both BSD and Solaris return 0 here, so we do too.
3014 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3017 #ifdef CONFIG_PROC_FS
3019 * This is invoked by the /proc filesystem handler to display a device
3022 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3023 __acquires(dev_base_lock)
3025 struct net *net = seq_file_net(seq);
3027 struct net_device *dev;
3029 read_lock(&dev_base_lock);
3031 return SEQ_START_TOKEN;
3034 for_each_netdev(net, dev)
3041 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3043 struct net *net = seq_file_net(seq);
3045 return v == SEQ_START_TOKEN ?
3046 first_net_device(net) : next_net_device((struct net_device *)v);
3049 void dev_seq_stop(struct seq_file *seq, void *v)
3050 __releases(dev_base_lock)
3052 read_unlock(&dev_base_lock);
3055 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3057 const struct net_device_stats *stats = dev_get_stats(dev);
3059 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3060 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3061 dev->name, stats->rx_bytes, stats->rx_packets,
3063 stats->rx_dropped + stats->rx_missed_errors,
3064 stats->rx_fifo_errors,
3065 stats->rx_length_errors + stats->rx_over_errors +
3066 stats->rx_crc_errors + stats->rx_frame_errors,
3067 stats->rx_compressed, stats->multicast,
3068 stats->tx_bytes, stats->tx_packets,
3069 stats->tx_errors, stats->tx_dropped,
3070 stats->tx_fifo_errors, stats->collisions,
3071 stats->tx_carrier_errors +
3072 stats->tx_aborted_errors +
3073 stats->tx_window_errors +
3074 stats->tx_heartbeat_errors,
3075 stats->tx_compressed);
3079 * Called from the PROCfs module. This now uses the new arbitrary sized
3080 * /proc/net interface to create /proc/net/dev
3082 static int dev_seq_show(struct seq_file *seq, void *v)
3084 if (v == SEQ_START_TOKEN)
3085 seq_puts(seq, "Inter-| Receive "
3087 " face |bytes packets errs drop fifo frame "
3088 "compressed multicast|bytes packets errs "
3089 "drop fifo colls carrier compressed\n");
3091 dev_seq_printf_stats(seq, v);
3095 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3097 struct netif_rx_stats *rc = NULL;
3099 while (*pos < nr_cpu_ids)
3100 if (cpu_online(*pos)) {
3101 rc = &per_cpu(netdev_rx_stat, *pos);
3108 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3110 return softnet_get_online(pos);
3113 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3116 return softnet_get_online(pos);
3119 static void softnet_seq_stop(struct seq_file *seq, void *v)
3123 static int softnet_seq_show(struct seq_file *seq, void *v)
3125 struct netif_rx_stats *s = v;
3127 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3128 s->total, s->dropped, s->time_squeeze, 0,
3129 0, 0, 0, 0, /* was fastroute */
3134 static const struct seq_operations dev_seq_ops = {
3135 .start = dev_seq_start,
3136 .next = dev_seq_next,
3137 .stop = dev_seq_stop,
3138 .show = dev_seq_show,
3141 static int dev_seq_open(struct inode *inode, struct file *file)
3143 return seq_open_net(inode, file, &dev_seq_ops,
3144 sizeof(struct seq_net_private));
3147 static const struct file_operations dev_seq_fops = {
3148 .owner = THIS_MODULE,
3149 .open = dev_seq_open,
3151 .llseek = seq_lseek,
3152 .release = seq_release_net,
3155 static const struct seq_operations softnet_seq_ops = {
3156 .start = softnet_seq_start,
3157 .next = softnet_seq_next,
3158 .stop = softnet_seq_stop,
3159 .show = softnet_seq_show,
3162 static int softnet_seq_open(struct inode *inode, struct file *file)
3164 return seq_open(file, &softnet_seq_ops);
3167 static const struct file_operations softnet_seq_fops = {
3168 .owner = THIS_MODULE,
3169 .open = softnet_seq_open,
3171 .llseek = seq_lseek,
3172 .release = seq_release,
3175 static void *ptype_get_idx(loff_t pos)
3177 struct packet_type *pt = NULL;
3181 list_for_each_entry_rcu(pt, &ptype_all, list) {
3187 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3188 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3197 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3201 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3204 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3206 struct packet_type *pt;
3207 struct list_head *nxt;
3211 if (v == SEQ_START_TOKEN)
3212 return ptype_get_idx(0);
3215 nxt = pt->list.next;
3216 if (pt->type == htons(ETH_P_ALL)) {
3217 if (nxt != &ptype_all)
3220 nxt = ptype_base[0].next;
3222 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3224 while (nxt == &ptype_base[hash]) {
3225 if (++hash >= PTYPE_HASH_SIZE)
3227 nxt = ptype_base[hash].next;
3230 return list_entry(nxt, struct packet_type, list);
3233 static void ptype_seq_stop(struct seq_file *seq, void *v)
3239 static int ptype_seq_show(struct seq_file *seq, void *v)
3241 struct packet_type *pt = v;
3243 if (v == SEQ_START_TOKEN)
3244 seq_puts(seq, "Type Device Function\n");
3245 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3246 if (pt->type == htons(ETH_P_ALL))
3247 seq_puts(seq, "ALL ");
3249 seq_printf(seq, "%04x", ntohs(pt->type));
3251 seq_printf(seq, " %-8s %pF\n",
3252 pt->dev ? pt->dev->name : "", pt->func);
3258 static const struct seq_operations ptype_seq_ops = {
3259 .start = ptype_seq_start,
3260 .next = ptype_seq_next,
3261 .stop = ptype_seq_stop,
3262 .show = ptype_seq_show,
3265 static int ptype_seq_open(struct inode *inode, struct file *file)
3267 return seq_open_net(inode, file, &ptype_seq_ops,
3268 sizeof(struct seq_net_private));
3271 static const struct file_operations ptype_seq_fops = {
3272 .owner = THIS_MODULE,
3273 .open = ptype_seq_open,
3275 .llseek = seq_lseek,
3276 .release = seq_release_net,
3280 static int __net_init dev_proc_net_init(struct net *net)
3284 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3286 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3288 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3291 if (wext_proc_init(net))
3297 proc_net_remove(net, "ptype");
3299 proc_net_remove(net, "softnet_stat");
3301 proc_net_remove(net, "dev");
3305 static void __net_exit dev_proc_net_exit(struct net *net)
3307 wext_proc_exit(net);
3309 proc_net_remove(net, "ptype");
3310 proc_net_remove(net, "softnet_stat");
3311 proc_net_remove(net, "dev");
3314 static struct pernet_operations __net_initdata dev_proc_ops = {
3315 .init = dev_proc_net_init,
3316 .exit = dev_proc_net_exit,
3319 static int __init dev_proc_init(void)
3321 return register_pernet_subsys(&dev_proc_ops);
3324 #define dev_proc_init() 0
3325 #endif /* CONFIG_PROC_FS */
3329 * netdev_set_master - set up master/slave pair
3330 * @slave: slave device
3331 * @master: new master device
3333 * Changes the master device of the slave. Pass %NULL to break the
3334 * bonding. The caller must hold the RTNL semaphore. On a failure
3335 * a negative errno code is returned. On success the reference counts
3336 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3337 * function returns zero.
3339 int netdev_set_master(struct net_device *slave, struct net_device *master)
3341 struct net_device *old = slave->master;
3351 slave->master = master;
3359 slave->flags |= IFF_SLAVE;
3361 slave->flags &= ~IFF_SLAVE;
3363 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3366 EXPORT_SYMBOL(netdev_set_master);
3368 static void dev_change_rx_flags(struct net_device *dev, int flags)
3370 const struct net_device_ops *ops = dev->netdev_ops;
3372 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3373 ops->ndo_change_rx_flags(dev, flags);
3376 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3378 unsigned short old_flags = dev->flags;
3384 dev->flags |= IFF_PROMISC;
3385 dev->promiscuity += inc;
3386 if (dev->promiscuity == 0) {
3389 * If inc causes overflow, untouch promisc and return error.
3392 dev->flags &= ~IFF_PROMISC;
3394 dev->promiscuity -= inc;
3395 printk(KERN_WARNING "%s: promiscuity touches roof, "
3396 "set promiscuity failed, promiscuity feature "
3397 "of device might be broken.\n", dev->name);
3401 if (dev->flags != old_flags) {
3402 printk(KERN_INFO "device %s %s promiscuous mode\n",
3403 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3405 if (audit_enabled) {
3406 current_uid_gid(&uid, &gid);
3407 audit_log(current->audit_context, GFP_ATOMIC,
3408 AUDIT_ANOM_PROMISCUOUS,
3409 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3410 dev->name, (dev->flags & IFF_PROMISC),
3411 (old_flags & IFF_PROMISC),
3412 audit_get_loginuid(current),
3414 audit_get_sessionid(current));
3417 dev_change_rx_flags(dev, IFF_PROMISC);
3423 * dev_set_promiscuity - update promiscuity count on a device
3427 * Add or remove promiscuity from a device. While the count in the device
3428 * remains above zero the interface remains promiscuous. Once it hits zero
3429 * the device reverts back to normal filtering operation. A negative inc
3430 * value is used to drop promiscuity on the device.
3431 * Return 0 if successful or a negative errno code on error.
3433 int dev_set_promiscuity(struct net_device *dev, int inc)
3435 unsigned short old_flags = dev->flags;
3438 err = __dev_set_promiscuity(dev, inc);
3441 if (dev->flags != old_flags)
3442 dev_set_rx_mode(dev);
3445 EXPORT_SYMBOL(dev_set_promiscuity);
3448 * dev_set_allmulti - update allmulti count on a device
3452 * Add or remove reception of all multicast frames to a device. While the
3453 * count in the device remains above zero the interface remains listening
3454 * to all interfaces. Once it hits zero the device reverts back to normal
3455 * filtering operation. A negative @inc value is used to drop the counter
3456 * when releasing a resource needing all multicasts.
3457 * Return 0 if successful or a negative errno code on error.
3460 int dev_set_allmulti(struct net_device *dev, int inc)
3462 unsigned short old_flags = dev->flags;
3466 dev->flags |= IFF_ALLMULTI;
3467 dev->allmulti += inc;
3468 if (dev->allmulti == 0) {
3471 * If inc causes overflow, untouch allmulti and return error.
3474 dev->flags &= ~IFF_ALLMULTI;
3476 dev->allmulti -= inc;
3477 printk(KERN_WARNING "%s: allmulti touches roof, "
3478 "set allmulti failed, allmulti feature of "
3479 "device might be broken.\n", dev->name);
3483 if (dev->flags ^ old_flags) {
3484 dev_change_rx_flags(dev, IFF_ALLMULTI);
3485 dev_set_rx_mode(dev);
3489 EXPORT_SYMBOL(dev_set_allmulti);
3492 * Upload unicast and multicast address lists to device and
3493 * configure RX filtering. When the device doesn't support unicast
3494 * filtering it is put in promiscuous mode while unicast addresses
3497 void __dev_set_rx_mode(struct net_device *dev)
3499 const struct net_device_ops *ops = dev->netdev_ops;
3501 /* dev_open will call this function so the list will stay sane. */
3502 if (!(dev->flags&IFF_UP))
3505 if (!netif_device_present(dev))
3508 if (ops->ndo_set_rx_mode)
3509 ops->ndo_set_rx_mode(dev);
3511 /* Unicast addresses changes may only happen under the rtnl,
3512 * therefore calling __dev_set_promiscuity here is safe.
3514 if (dev->uc.count > 0 && !dev->uc_promisc) {
3515 __dev_set_promiscuity(dev, 1);
3516 dev->uc_promisc = 1;
3517 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3518 __dev_set_promiscuity(dev, -1);
3519 dev->uc_promisc = 0;
3522 if (ops->ndo_set_multicast_list)
3523 ops->ndo_set_multicast_list(dev);
3527 void dev_set_rx_mode(struct net_device *dev)
3529 netif_addr_lock_bh(dev);
3530 __dev_set_rx_mode(dev);
3531 netif_addr_unlock_bh(dev);
3534 /* hw addresses list handling functions */
3536 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3537 int addr_len, unsigned char addr_type)
3539 struct netdev_hw_addr *ha;
3542 if (addr_len > MAX_ADDR_LEN)
3545 list_for_each_entry(ha, &list->list, list) {
3546 if (!memcmp(ha->addr, addr, addr_len) &&
3547 ha->type == addr_type) {
3554 alloc_size = sizeof(*ha);
3555 if (alloc_size < L1_CACHE_BYTES)
3556 alloc_size = L1_CACHE_BYTES;
3557 ha = kmalloc(alloc_size, GFP_ATOMIC);
3560 memcpy(ha->addr, addr, addr_len);
3561 ha->type = addr_type;
3564 list_add_tail_rcu(&ha->list, &list->list);
3569 static void ha_rcu_free(struct rcu_head *head)
3571 struct netdev_hw_addr *ha;
3573 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3577 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3578 int addr_len, unsigned char addr_type)
3580 struct netdev_hw_addr *ha;
3582 list_for_each_entry(ha, &list->list, list) {
3583 if (!memcmp(ha->addr, addr, addr_len) &&
3584 (ha->type == addr_type || !addr_type)) {
3587 list_del_rcu(&ha->list);
3588 call_rcu(&ha->rcu_head, ha_rcu_free);
3596 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3597 struct netdev_hw_addr_list *from_list,
3599 unsigned char addr_type)
3602 struct netdev_hw_addr *ha, *ha2;
3605 list_for_each_entry(ha, &from_list->list, list) {
3606 type = addr_type ? addr_type : ha->type;
3607 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3614 list_for_each_entry(ha2, &from_list->list, list) {
3617 type = addr_type ? addr_type : ha2->type;
3618 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3623 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3624 struct netdev_hw_addr_list *from_list,
3626 unsigned char addr_type)
3628 struct netdev_hw_addr *ha;
3631 list_for_each_entry(ha, &from_list->list, list) {
3632 type = addr_type ? addr_type : ha->type;
3633 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3637 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3638 struct netdev_hw_addr_list *from_list,
3642 struct netdev_hw_addr *ha, *tmp;
3644 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3646 err = __hw_addr_add(to_list, ha->addr,
3647 addr_len, ha->type);
3652 } else if (ha->refcount == 1) {
3653 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3654 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3660 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3661 struct netdev_hw_addr_list *from_list,
3664 struct netdev_hw_addr *ha, *tmp;
3666 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3668 __hw_addr_del(to_list, ha->addr,
3669 addr_len, ha->type);
3671 __hw_addr_del(from_list, ha->addr,
3672 addr_len, ha->type);
3677 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3679 struct netdev_hw_addr *ha, *tmp;
3681 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3682 list_del_rcu(&ha->list);
3683 call_rcu(&ha->rcu_head, ha_rcu_free);
3688 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3690 INIT_LIST_HEAD(&list->list);
3694 /* Device addresses handling functions */
3696 static void dev_addr_flush(struct net_device *dev)
3698 /* rtnl_mutex must be held here */
3700 __hw_addr_flush(&dev->dev_addrs);
3701 dev->dev_addr = NULL;
3704 static int dev_addr_init(struct net_device *dev)
3706 unsigned char addr[MAX_ADDR_LEN];
3707 struct netdev_hw_addr *ha;
3710 /* rtnl_mutex must be held here */
3712 __hw_addr_init(&dev->dev_addrs);
3713 memset(addr, 0, sizeof(addr));
3714 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3715 NETDEV_HW_ADDR_T_LAN);
3718 * Get the first (previously created) address from the list
3719 * and set dev_addr pointer to this location.
3721 ha = list_first_entry(&dev->dev_addrs.list,
3722 struct netdev_hw_addr, list);
3723 dev->dev_addr = ha->addr;
3729 * dev_addr_add - Add a device address
3731 * @addr: address to add
3732 * @addr_type: address type
3734 * Add a device address to the device or increase the reference count if
3735 * it already exists.
3737 * The caller must hold the rtnl_mutex.
3739 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3740 unsigned char addr_type)
3746 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3748 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3751 EXPORT_SYMBOL(dev_addr_add);
3754 * dev_addr_del - Release a device address.
3756 * @addr: address to delete
3757 * @addr_type: address type
3759 * Release reference to a device address and remove it from the device
3760 * if the reference count drops to zero.
3762 * The caller must hold the rtnl_mutex.
3764 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3765 unsigned char addr_type)
3768 struct netdev_hw_addr *ha;
3773 * We can not remove the first address from the list because
3774 * dev->dev_addr points to that.
3776 ha = list_first_entry(&dev->dev_addrs.list,
3777 struct netdev_hw_addr, list);
3778 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3781 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3784 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3787 EXPORT_SYMBOL(dev_addr_del);
3790 * dev_addr_add_multiple - Add device addresses from another device
3791 * @to_dev: device to which addresses will be added
3792 * @from_dev: device from which addresses will be added
3793 * @addr_type: address type - 0 means type will be used from from_dev
3795 * Add device addresses of the one device to another.
3797 * The caller must hold the rtnl_mutex.
3799 int dev_addr_add_multiple(struct net_device *to_dev,
3800 struct net_device *from_dev,
3801 unsigned char addr_type)
3807 if (from_dev->addr_len != to_dev->addr_len)
3809 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3810 to_dev->addr_len, addr_type);
3812 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3815 EXPORT_SYMBOL(dev_addr_add_multiple);
3818 * dev_addr_del_multiple - Delete device addresses by another device
3819 * @to_dev: device where the addresses will be deleted
3820 * @from_dev: device by which addresses the addresses will be deleted
3821 * @addr_type: address type - 0 means type will used from from_dev
3823 * Deletes addresses in to device by the list of addresses in from device.
3825 * The caller must hold the rtnl_mutex.
3827 int dev_addr_del_multiple(struct net_device *to_dev,
3828 struct net_device *from_dev,
3829 unsigned char addr_type)
3833 if (from_dev->addr_len != to_dev->addr_len)
3835 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3836 to_dev->addr_len, addr_type);
3837 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3840 EXPORT_SYMBOL(dev_addr_del_multiple);
3842 /* multicast addresses handling functions */
3844 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3845 void *addr, int alen, int glbl)
3847 struct dev_addr_list *da;
3849 for (; (da = *list) != NULL; list = &da->next) {
3850 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3851 alen == da->da_addrlen) {
3853 int old_glbl = da->da_gusers;
3870 int __dev_addr_add(struct dev_addr_list **list, int *count,
3871 void *addr, int alen, int glbl)
3873 struct dev_addr_list *da;
3875 for (da = *list; da != NULL; da = da->next) {
3876 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3877 da->da_addrlen == alen) {
3879 int old_glbl = da->da_gusers;
3889 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3892 memcpy(da->da_addr, addr, alen);
3893 da->da_addrlen = alen;
3895 da->da_gusers = glbl ? 1 : 0;
3903 * dev_unicast_delete - Release secondary unicast address.
3905 * @addr: address to delete
3907 * Release reference to a secondary unicast address and remove it
3908 * from the device if the reference count drops to zero.
3910 * The caller must hold the rtnl_mutex.
3912 int dev_unicast_delete(struct net_device *dev, void *addr)
3918 netif_addr_lock_bh(dev);
3919 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
3920 NETDEV_HW_ADDR_T_UNICAST);
3922 __dev_set_rx_mode(dev);
3923 netif_addr_unlock_bh(dev);
3926 EXPORT_SYMBOL(dev_unicast_delete);
3929 * dev_unicast_add - add a secondary unicast address
3931 * @addr: address to add
3933 * Add a secondary unicast address to the device or increase
3934 * the reference count if it already exists.
3936 * The caller must hold the rtnl_mutex.
3938 int dev_unicast_add(struct net_device *dev, void *addr)
3944 netif_addr_lock_bh(dev);
3945 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
3946 NETDEV_HW_ADDR_T_UNICAST);
3948 __dev_set_rx_mode(dev);
3949 netif_addr_unlock_bh(dev);
3952 EXPORT_SYMBOL(dev_unicast_add);
3954 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
3955 struct dev_addr_list **from, int *from_count)
3957 struct dev_addr_list *da, *next;
3961 while (da != NULL) {
3963 if (!da->da_synced) {
3964 err = __dev_addr_add(to, to_count,
3965 da->da_addr, da->da_addrlen, 0);
3970 } else if (da->da_users == 1) {
3971 __dev_addr_delete(to, to_count,
3972 da->da_addr, da->da_addrlen, 0);
3973 __dev_addr_delete(from, from_count,
3974 da->da_addr, da->da_addrlen, 0);
3980 EXPORT_SYMBOL_GPL(__dev_addr_sync);
3982 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
3983 struct dev_addr_list **from, int *from_count)
3985 struct dev_addr_list *da, *next;
3988 while (da != NULL) {
3990 if (da->da_synced) {
3991 __dev_addr_delete(to, to_count,
3992 da->da_addr, da->da_addrlen, 0);
3994 __dev_addr_delete(from, from_count,
3995 da->da_addr, da->da_addrlen, 0);
4000 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4003 * dev_unicast_sync - Synchronize device's unicast list to another device
4004 * @to: destination device
4005 * @from: source device
4007 * Add newly added addresses to the destination device and release
4008 * addresses that have no users left. The source device must be
4009 * locked by netif_tx_lock_bh.
4011 * This function is intended to be called from the dev->set_rx_mode
4012 * function of layered software devices.
4014 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4018 if (to->addr_len != from->addr_len)
4021 netif_addr_lock_bh(to);
4022 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4024 __dev_set_rx_mode(to);
4025 netif_addr_unlock_bh(to);
4028 EXPORT_SYMBOL(dev_unicast_sync);
4031 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4032 * @to: destination device
4033 * @from: source device
4035 * Remove all addresses that were added to the destination device by
4036 * dev_unicast_sync(). This function is intended to be called from the
4037 * dev->stop function of layered software devices.
4039 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4041 if (to->addr_len != from->addr_len)
4044 netif_addr_lock_bh(from);
4045 netif_addr_lock(to);
4046 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4047 __dev_set_rx_mode(to);
4048 netif_addr_unlock(to);
4049 netif_addr_unlock_bh(from);
4051 EXPORT_SYMBOL(dev_unicast_unsync);
4053 static void dev_unicast_flush(struct net_device *dev)
4055 netif_addr_lock_bh(dev);
4056 __hw_addr_flush(&dev->uc);
4057 netif_addr_unlock_bh(dev);
4060 static void dev_unicast_init(struct net_device *dev)
4062 __hw_addr_init(&dev->uc);
4066 static void __dev_addr_discard(struct dev_addr_list **list)
4068 struct dev_addr_list *tmp;
4070 while (*list != NULL) {
4073 if (tmp->da_users > tmp->da_gusers)
4074 printk("__dev_addr_discard: address leakage! "
4075 "da_users=%d\n", tmp->da_users);
4080 static void dev_addr_discard(struct net_device *dev)
4082 netif_addr_lock_bh(dev);
4084 __dev_addr_discard(&dev->mc_list);
4087 netif_addr_unlock_bh(dev);
4091 * dev_get_flags - get flags reported to userspace
4094 * Get the combination of flag bits exported through APIs to userspace.
4096 unsigned dev_get_flags(const struct net_device *dev)
4100 flags = (dev->flags & ~(IFF_PROMISC |
4105 (dev->gflags & (IFF_PROMISC |
4108 if (netif_running(dev)) {
4109 if (netif_oper_up(dev))
4110 flags |= IFF_RUNNING;
4111 if (netif_carrier_ok(dev))
4112 flags |= IFF_LOWER_UP;
4113 if (netif_dormant(dev))
4114 flags |= IFF_DORMANT;
4119 EXPORT_SYMBOL(dev_get_flags);
4122 * dev_change_flags - change device settings
4124 * @flags: device state flags
4126 * Change settings on device based state flags. The flags are
4127 * in the userspace exported format.
4129 int dev_change_flags(struct net_device *dev, unsigned flags)
4132 int old_flags = dev->flags;
4137 * Set the flags on our device.
4140 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4141 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4143 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4147 * Load in the correct multicast list now the flags have changed.
4150 if ((old_flags ^ flags) & IFF_MULTICAST)
4151 dev_change_rx_flags(dev, IFF_MULTICAST);
4153 dev_set_rx_mode(dev);
4156 * Have we downed the interface. We handle IFF_UP ourselves
4157 * according to user attempts to set it, rather than blindly
4162 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4163 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4166 dev_set_rx_mode(dev);
4169 if (dev->flags & IFF_UP &&
4170 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4172 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4174 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4175 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4177 dev->gflags ^= IFF_PROMISC;
4178 dev_set_promiscuity(dev, inc);
4181 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4182 is important. Some (broken) drivers set IFF_PROMISC, when
4183 IFF_ALLMULTI is requested not asking us and not reporting.
4185 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4186 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4188 dev->gflags ^= IFF_ALLMULTI;
4189 dev_set_allmulti(dev, inc);
4192 /* Exclude state transition flags, already notified */
4193 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4195 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4199 EXPORT_SYMBOL(dev_change_flags);
4202 * dev_set_mtu - Change maximum transfer unit
4204 * @new_mtu: new transfer unit
4206 * Change the maximum transfer size of the network device.
4208 int dev_set_mtu(struct net_device *dev, int new_mtu)
4210 const struct net_device_ops *ops = dev->netdev_ops;
4213 if (new_mtu == dev->mtu)
4216 /* MTU must be positive. */
4220 if (!netif_device_present(dev))
4224 if (ops->ndo_change_mtu)
4225 err = ops->ndo_change_mtu(dev, new_mtu);
4229 if (!err && dev->flags & IFF_UP)
4230 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4233 EXPORT_SYMBOL(dev_set_mtu);
4236 * dev_set_mac_address - Change Media Access Control Address
4240 * Change the hardware (MAC) address of the device
4242 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4244 const struct net_device_ops *ops = dev->netdev_ops;
4247 if (!ops->ndo_set_mac_address)
4249 if (sa->sa_family != dev->type)
4251 if (!netif_device_present(dev))
4253 err = ops->ndo_set_mac_address(dev, sa);
4255 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4258 EXPORT_SYMBOL(dev_set_mac_address);
4261 * Perform the SIOCxIFxxx calls, inside read_lock(dev_base_lock)
4263 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4266 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4272 case SIOCGIFFLAGS: /* Get interface flags */
4273 ifr->ifr_flags = (short) dev_get_flags(dev);
4276 case SIOCGIFMETRIC: /* Get the metric on the interface
4277 (currently unused) */
4278 ifr->ifr_metric = 0;
4281 case SIOCGIFMTU: /* Get the MTU of a device */
4282 ifr->ifr_mtu = dev->mtu;
4287 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4289 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4290 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4291 ifr->ifr_hwaddr.sa_family = dev->type;
4299 ifr->ifr_map.mem_start = dev->mem_start;
4300 ifr->ifr_map.mem_end = dev->mem_end;
4301 ifr->ifr_map.base_addr = dev->base_addr;
4302 ifr->ifr_map.irq = dev->irq;
4303 ifr->ifr_map.dma = dev->dma;
4304 ifr->ifr_map.port = dev->if_port;
4308 ifr->ifr_ifindex = dev->ifindex;
4312 ifr->ifr_qlen = dev->tx_queue_len;
4316 /* dev_ioctl() should ensure this case
4328 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4330 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4333 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4334 const struct net_device_ops *ops;
4339 ops = dev->netdev_ops;
4342 case SIOCSIFFLAGS: /* Set interface flags */
4343 return dev_change_flags(dev, ifr->ifr_flags);
4345 case SIOCSIFMETRIC: /* Set the metric on the interface
4346 (currently unused) */
4349 case SIOCSIFMTU: /* Set the MTU of a device */
4350 return dev_set_mtu(dev, ifr->ifr_mtu);
4353 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4355 case SIOCSIFHWBROADCAST:
4356 if (ifr->ifr_hwaddr.sa_family != dev->type)
4358 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4359 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4360 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4364 if (ops->ndo_set_config) {
4365 if (!netif_device_present(dev))
4367 return ops->ndo_set_config(dev, &ifr->ifr_map);
4372 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4373 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4375 if (!netif_device_present(dev))
4377 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4381 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4382 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4384 if (!netif_device_present(dev))
4386 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4390 if (ifr->ifr_qlen < 0)
4392 dev->tx_queue_len = ifr->ifr_qlen;
4396 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4397 return dev_change_name(dev, ifr->ifr_newname);
4400 * Unknown or private ioctl
4403 if ((cmd >= SIOCDEVPRIVATE &&
4404 cmd <= SIOCDEVPRIVATE + 15) ||
4405 cmd == SIOCBONDENSLAVE ||
4406 cmd == SIOCBONDRELEASE ||
4407 cmd == SIOCBONDSETHWADDR ||
4408 cmd == SIOCBONDSLAVEINFOQUERY ||
4409 cmd == SIOCBONDINFOQUERY ||
4410 cmd == SIOCBONDCHANGEACTIVE ||
4411 cmd == SIOCGMIIPHY ||
4412 cmd == SIOCGMIIREG ||
4413 cmd == SIOCSMIIREG ||
4414 cmd == SIOCBRADDIF ||
4415 cmd == SIOCBRDELIF ||
4416 cmd == SIOCSHWTSTAMP ||
4417 cmd == SIOCWANDEV) {
4419 if (ops->ndo_do_ioctl) {
4420 if (netif_device_present(dev))
4421 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4433 * This function handles all "interface"-type I/O control requests. The actual
4434 * 'doing' part of this is dev_ifsioc above.
4438 * dev_ioctl - network device ioctl
4439 * @net: the applicable net namespace
4440 * @cmd: command to issue
4441 * @arg: pointer to a struct ifreq in user space
4443 * Issue ioctl functions to devices. This is normally called by the
4444 * user space syscall interfaces but can sometimes be useful for
4445 * other purposes. The return value is the return from the syscall if
4446 * positive or a negative errno code on error.
4449 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4455 /* One special case: SIOCGIFCONF takes ifconf argument
4456 and requires shared lock, because it sleeps writing
4460 if (cmd == SIOCGIFCONF) {
4462 ret = dev_ifconf(net, (char __user *) arg);
4466 if (cmd == SIOCGIFNAME)
4467 return dev_ifname(net, (struct ifreq __user *)arg);
4469 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4472 ifr.ifr_name[IFNAMSIZ-1] = 0;
4474 colon = strchr(ifr.ifr_name, ':');
4479 * See which interface the caller is talking about.
4484 * These ioctl calls:
4485 * - can be done by all.
4486 * - atomic and do not require locking.
4497 dev_load(net, ifr.ifr_name);
4498 read_lock(&dev_base_lock);
4499 ret = dev_ifsioc_locked(net, &ifr, cmd);
4500 read_unlock(&dev_base_lock);
4504 if (copy_to_user(arg, &ifr,
4505 sizeof(struct ifreq)))
4511 dev_load(net, ifr.ifr_name);
4513 ret = dev_ethtool(net, &ifr);
4518 if (copy_to_user(arg, &ifr,
4519 sizeof(struct ifreq)))
4525 * These ioctl calls:
4526 * - require superuser power.
4527 * - require strict serialization.
4533 if (!capable(CAP_NET_ADMIN))
4535 dev_load(net, ifr.ifr_name);
4537 ret = dev_ifsioc(net, &ifr, cmd);
4542 if (copy_to_user(arg, &ifr,
4543 sizeof(struct ifreq)))
4549 * These ioctl calls:
4550 * - require superuser power.
4551 * - require strict serialization.
4552 * - do not return a value
4562 case SIOCSIFHWBROADCAST:
4565 case SIOCBONDENSLAVE:
4566 case SIOCBONDRELEASE:
4567 case SIOCBONDSETHWADDR:
4568 case SIOCBONDCHANGEACTIVE:
4572 if (!capable(CAP_NET_ADMIN))
4575 case SIOCBONDSLAVEINFOQUERY:
4576 case SIOCBONDINFOQUERY:
4577 dev_load(net, ifr.ifr_name);
4579 ret = dev_ifsioc(net, &ifr, cmd);
4584 /* Get the per device memory space. We can add this but
4585 * currently do not support it */
4587 /* Set the per device memory buffer space.
4588 * Not applicable in our case */
4593 * Unknown or private ioctl.
4596 if (cmd == SIOCWANDEV ||
4597 (cmd >= SIOCDEVPRIVATE &&
4598 cmd <= SIOCDEVPRIVATE + 15)) {
4599 dev_load(net, ifr.ifr_name);
4601 ret = dev_ifsioc(net, &ifr, cmd);
4603 if (!ret && copy_to_user(arg, &ifr,
4604 sizeof(struct ifreq)))
4608 /* Take care of Wireless Extensions */
4609 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4610 return wext_handle_ioctl(net, &ifr, cmd, arg);
4617 * dev_new_index - allocate an ifindex
4618 * @net: the applicable net namespace
4620 * Returns a suitable unique value for a new device interface
4621 * number. The caller must hold the rtnl semaphore or the
4622 * dev_base_lock to be sure it remains unique.
4624 static int dev_new_index(struct net *net)
4630 if (!__dev_get_by_index(net, ifindex))
4635 /* Delayed registration/unregisteration */
4636 static LIST_HEAD(net_todo_list);
4638 static void net_set_todo(struct net_device *dev)
4640 list_add_tail(&dev->todo_list, &net_todo_list);
4643 static void rollback_registered(struct net_device *dev)
4645 BUG_ON(dev_boot_phase);
4648 /* Some devices call without registering for initialization unwind. */
4649 if (dev->reg_state == NETREG_UNINITIALIZED) {
4650 printk(KERN_DEBUG "unregister_netdevice: device %s/%p never "
4651 "was registered\n", dev->name, dev);
4657 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4659 /* If device is running, close it first. */
4662 /* And unlink it from device chain. */
4663 unlist_netdevice(dev);
4665 dev->reg_state = NETREG_UNREGISTERING;
4669 /* Shutdown queueing discipline. */
4673 /* Notify protocols, that we are about to destroy
4674 this device. They should clean all the things.
4676 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4679 * Flush the unicast and multicast chains
4681 dev_unicast_flush(dev);
4682 dev_addr_discard(dev);
4684 if (dev->netdev_ops->ndo_uninit)
4685 dev->netdev_ops->ndo_uninit(dev);
4687 /* Notifier chain MUST detach us from master device. */
4688 WARN_ON(dev->master);
4690 /* Remove entries from kobject tree */
4691 netdev_unregister_kobject(dev);
4698 static void __netdev_init_queue_locks_one(struct net_device *dev,
4699 struct netdev_queue *dev_queue,
4702 spin_lock_init(&dev_queue->_xmit_lock);
4703 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4704 dev_queue->xmit_lock_owner = -1;
4707 static void netdev_init_queue_locks(struct net_device *dev)
4709 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4710 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4713 unsigned long netdev_fix_features(unsigned long features, const char *name)
4715 /* Fix illegal SG+CSUM combinations. */
4716 if ((features & NETIF_F_SG) &&
4717 !(features & NETIF_F_ALL_CSUM)) {
4719 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4720 "checksum feature.\n", name);
4721 features &= ~NETIF_F_SG;
4724 /* TSO requires that SG is present as well. */
4725 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4727 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4728 "SG feature.\n", name);
4729 features &= ~NETIF_F_TSO;
4732 if (features & NETIF_F_UFO) {
4733 if (!(features & NETIF_F_GEN_CSUM)) {
4735 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4736 "since no NETIF_F_HW_CSUM feature.\n",
4738 features &= ~NETIF_F_UFO;
4741 if (!(features & NETIF_F_SG)) {
4743 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4744 "since no NETIF_F_SG feature.\n", name);
4745 features &= ~NETIF_F_UFO;
4751 EXPORT_SYMBOL(netdev_fix_features);
4754 * register_netdevice - register a network device
4755 * @dev: device to register
4757 * Take a completed network device structure and add it to the kernel
4758 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4759 * chain. 0 is returned on success. A negative errno code is returned
4760 * on a failure to set up the device, or if the name is a duplicate.
4762 * Callers must hold the rtnl semaphore. You may want
4763 * register_netdev() instead of this.
4766 * The locking appears insufficient to guarantee two parallel registers
4767 * will not get the same name.
4770 int register_netdevice(struct net_device *dev)
4772 struct hlist_head *head;
4773 struct hlist_node *p;
4775 struct net *net = dev_net(dev);
4777 BUG_ON(dev_boot_phase);
4782 /* When net_device's are persistent, this will be fatal. */
4783 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4786 spin_lock_init(&dev->addr_list_lock);
4787 netdev_set_addr_lockdep_class(dev);
4788 netdev_init_queue_locks(dev);
4792 /* Init, if this function is available */
4793 if (dev->netdev_ops->ndo_init) {
4794 ret = dev->netdev_ops->ndo_init(dev);
4802 if (!dev_valid_name(dev->name)) {
4807 dev->ifindex = dev_new_index(net);
4808 if (dev->iflink == -1)
4809 dev->iflink = dev->ifindex;
4811 /* Check for existence of name */
4812 head = dev_name_hash(net, dev->name);
4813 hlist_for_each(p, head) {
4814 struct net_device *d
4815 = hlist_entry(p, struct net_device, name_hlist);
4816 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4822 /* Fix illegal checksum combinations */
4823 if ((dev->features & NETIF_F_HW_CSUM) &&
4824 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4825 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4827 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4830 if ((dev->features & NETIF_F_NO_CSUM) &&
4831 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4832 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4834 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4837 dev->features = netdev_fix_features(dev->features, dev->name);
4839 /* Enable software GSO if SG is supported. */
4840 if (dev->features & NETIF_F_SG)
4841 dev->features |= NETIF_F_GSO;
4843 netdev_initialize_kobject(dev);
4845 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4846 ret = notifier_to_errno(ret);
4850 ret = netdev_register_kobject(dev);
4853 dev->reg_state = NETREG_REGISTERED;
4856 * Default initial state at registry is that the
4857 * device is present.
4860 set_bit(__LINK_STATE_PRESENT, &dev->state);
4862 dev_init_scheduler(dev);
4864 list_netdevice(dev);
4866 /* Notify protocols, that a new device appeared. */
4867 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4868 ret = notifier_to_errno(ret);
4870 rollback_registered(dev);
4871 dev->reg_state = NETREG_UNREGISTERED;
4878 if (dev->netdev_ops->ndo_uninit)
4879 dev->netdev_ops->ndo_uninit(dev);
4882 EXPORT_SYMBOL(register_netdevice);
4885 * init_dummy_netdev - init a dummy network device for NAPI
4886 * @dev: device to init
4888 * This takes a network device structure and initialize the minimum
4889 * amount of fields so it can be used to schedule NAPI polls without
4890 * registering a full blown interface. This is to be used by drivers
4891 * that need to tie several hardware interfaces to a single NAPI
4892 * poll scheduler due to HW limitations.
4894 int init_dummy_netdev(struct net_device *dev)
4896 /* Clear everything. Note we don't initialize spinlocks
4897 * are they aren't supposed to be taken by any of the
4898 * NAPI code and this dummy netdev is supposed to be
4899 * only ever used for NAPI polls
4901 memset(dev, 0, sizeof(struct net_device));
4903 /* make sure we BUG if trying to hit standard
4904 * register/unregister code path
4906 dev->reg_state = NETREG_DUMMY;
4908 /* initialize the ref count */
4909 atomic_set(&dev->refcnt, 1);
4911 /* NAPI wants this */
4912 INIT_LIST_HEAD(&dev->napi_list);
4914 /* a dummy interface is started by default */
4915 set_bit(__LINK_STATE_PRESENT, &dev->state);
4916 set_bit(__LINK_STATE_START, &dev->state);
4920 EXPORT_SYMBOL_GPL(init_dummy_netdev);
4924 * register_netdev - register a network device
4925 * @dev: device to register
4927 * Take a completed network device structure and add it to the kernel
4928 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4929 * chain. 0 is returned on success. A negative errno code is returned
4930 * on a failure to set up the device, or if the name is a duplicate.
4932 * This is a wrapper around register_netdevice that takes the rtnl semaphore
4933 * and expands the device name if you passed a format string to
4936 int register_netdev(struct net_device *dev)
4943 * If the name is a format string the caller wants us to do a
4946 if (strchr(dev->name, '%')) {
4947 err = dev_alloc_name(dev, dev->name);
4952 err = register_netdevice(dev);
4957 EXPORT_SYMBOL(register_netdev);
4960 * netdev_wait_allrefs - wait until all references are gone.
4962 * This is called when unregistering network devices.
4964 * Any protocol or device that holds a reference should register
4965 * for netdevice notification, and cleanup and put back the
4966 * reference if they receive an UNREGISTER event.
4967 * We can get stuck here if buggy protocols don't correctly
4970 static void netdev_wait_allrefs(struct net_device *dev)
4972 unsigned long rebroadcast_time, warning_time;
4974 rebroadcast_time = warning_time = jiffies;
4975 while (atomic_read(&dev->refcnt) != 0) {
4976 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
4979 /* Rebroadcast unregister notification */
4980 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4982 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
4984 /* We must not have linkwatch events
4985 * pending on unregister. If this
4986 * happens, we simply run the queue
4987 * unscheduled, resulting in a noop
4990 linkwatch_run_queue();
4995 rebroadcast_time = jiffies;
5000 if (time_after(jiffies, warning_time + 10 * HZ)) {
5001 printk(KERN_EMERG "unregister_netdevice: "
5002 "waiting for %s to become free. Usage "
5004 dev->name, atomic_read(&dev->refcnt));
5005 warning_time = jiffies;
5014 * register_netdevice(x1);
5015 * register_netdevice(x2);
5017 * unregister_netdevice(y1);
5018 * unregister_netdevice(y2);
5024 * We are invoked by rtnl_unlock().
5025 * This allows us to deal with problems:
5026 * 1) We can delete sysfs objects which invoke hotplug
5027 * without deadlocking with linkwatch via keventd.
5028 * 2) Since we run with the RTNL semaphore not held, we can sleep
5029 * safely in order to wait for the netdev refcnt to drop to zero.
5031 * We must not return until all unregister events added during
5032 * the interval the lock was held have been completed.
5034 void netdev_run_todo(void)
5036 struct list_head list;
5038 /* Snapshot list, allow later requests */
5039 list_replace_init(&net_todo_list, &list);
5043 while (!list_empty(&list)) {
5044 struct net_device *dev
5045 = list_entry(list.next, struct net_device, todo_list);
5046 list_del(&dev->todo_list);
5048 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5049 printk(KERN_ERR "network todo '%s' but state %d\n",
5050 dev->name, dev->reg_state);
5055 dev->reg_state = NETREG_UNREGISTERED;
5057 on_each_cpu(flush_backlog, dev, 1);
5059 netdev_wait_allrefs(dev);
5062 BUG_ON(atomic_read(&dev->refcnt));
5063 WARN_ON(dev->ip_ptr);
5064 WARN_ON(dev->ip6_ptr);
5065 WARN_ON(dev->dn_ptr);
5067 if (dev->destructor)
5068 dev->destructor(dev);
5070 /* Free network device */
5071 kobject_put(&dev->dev.kobj);
5076 * dev_get_stats - get network device statistics
5077 * @dev: device to get statistics from
5079 * Get network statistics from device. The device driver may provide
5080 * its own method by setting dev->netdev_ops->get_stats; otherwise
5081 * the internal statistics structure is used.
5083 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5085 const struct net_device_ops *ops = dev->netdev_ops;
5087 if (ops->ndo_get_stats)
5088 return ops->ndo_get_stats(dev);
5090 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5091 struct net_device_stats *stats = &dev->stats;
5093 struct netdev_queue *txq;
5095 for (i = 0; i < dev->num_tx_queues; i++) {
5096 txq = netdev_get_tx_queue(dev, i);
5097 tx_bytes += txq->tx_bytes;
5098 tx_packets += txq->tx_packets;
5099 tx_dropped += txq->tx_dropped;
5101 if (tx_bytes || tx_packets || tx_dropped) {
5102 stats->tx_bytes = tx_bytes;
5103 stats->tx_packets = tx_packets;
5104 stats->tx_dropped = tx_dropped;
5109 EXPORT_SYMBOL(dev_get_stats);
5111 static void netdev_init_one_queue(struct net_device *dev,
5112 struct netdev_queue *queue,
5118 static void netdev_init_queues(struct net_device *dev)
5120 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5121 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5122 spin_lock_init(&dev->tx_global_lock);
5126 * alloc_netdev_mq - allocate network device
5127 * @sizeof_priv: size of private data to allocate space for
5128 * @name: device name format string
5129 * @setup: callback to initialize device
5130 * @queue_count: the number of subqueues to allocate
5132 * Allocates a struct net_device with private data area for driver use
5133 * and performs basic initialization. Also allocates subquue structs
5134 * for each queue on the device at the end of the netdevice.
5136 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5137 void (*setup)(struct net_device *), unsigned int queue_count)
5139 struct netdev_queue *tx;
5140 struct net_device *dev;
5142 struct net_device *p;
5144 BUG_ON(strlen(name) >= sizeof(dev->name));
5146 alloc_size = sizeof(struct net_device);
5148 /* ensure 32-byte alignment of private area */
5149 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5150 alloc_size += sizeof_priv;
5152 /* ensure 32-byte alignment of whole construct */
5153 alloc_size += NETDEV_ALIGN - 1;
5155 p = kzalloc(alloc_size, GFP_KERNEL);
5157 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5161 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5163 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5168 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5169 dev->padded = (char *)dev - (char *)p;
5171 if (dev_addr_init(dev))
5174 dev_unicast_init(dev);
5176 dev_net_set(dev, &init_net);
5179 dev->num_tx_queues = queue_count;
5180 dev->real_num_tx_queues = queue_count;
5182 dev->gso_max_size = GSO_MAX_SIZE;
5184 netdev_init_queues(dev);
5186 INIT_LIST_HEAD(&dev->napi_list);
5187 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5189 strcpy(dev->name, name);
5199 EXPORT_SYMBOL(alloc_netdev_mq);
5202 * free_netdev - free network device
5205 * This function does the last stage of destroying an allocated device
5206 * interface. The reference to the device object is released.
5207 * If this is the last reference then it will be freed.
5209 void free_netdev(struct net_device *dev)
5211 struct napi_struct *p, *n;
5213 release_net(dev_net(dev));
5217 /* Flush device addresses */
5218 dev_addr_flush(dev);
5220 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5223 /* Compatibility with error handling in drivers */
5224 if (dev->reg_state == NETREG_UNINITIALIZED) {
5225 kfree((char *)dev - dev->padded);
5229 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5230 dev->reg_state = NETREG_RELEASED;
5232 /* will free via device release */
5233 put_device(&dev->dev);
5235 EXPORT_SYMBOL(free_netdev);
5238 * synchronize_net - Synchronize with packet receive processing
5240 * Wait for packets currently being received to be done.
5241 * Does not block later packets from starting.
5243 void synchronize_net(void)
5248 EXPORT_SYMBOL(synchronize_net);
5251 * unregister_netdevice - remove device from the kernel
5254 * This function shuts down a device interface and removes it
5255 * from the kernel tables.
5257 * Callers must hold the rtnl semaphore. You may want
5258 * unregister_netdev() instead of this.
5261 void unregister_netdevice(struct net_device *dev)
5265 rollback_registered(dev);
5266 /* Finish processing unregister after unlock */
5269 EXPORT_SYMBOL(unregister_netdevice);
5272 * unregister_netdev - remove device from the kernel
5275 * This function shuts down a device interface and removes it
5276 * from the kernel tables.
5278 * This is just a wrapper for unregister_netdevice that takes
5279 * the rtnl semaphore. In general you want to use this and not
5280 * unregister_netdevice.
5282 void unregister_netdev(struct net_device *dev)
5285 unregister_netdevice(dev);
5288 EXPORT_SYMBOL(unregister_netdev);
5291 * dev_change_net_namespace - move device to different nethost namespace
5293 * @net: network namespace
5294 * @pat: If not NULL name pattern to try if the current device name
5295 * is already taken in the destination network namespace.
5297 * This function shuts down a device interface and moves it
5298 * to a new network namespace. On success 0 is returned, on
5299 * a failure a netagive errno code is returned.
5301 * Callers must hold the rtnl semaphore.
5304 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5307 const char *destname;
5312 /* Don't allow namespace local devices to be moved. */
5314 if (dev->features & NETIF_F_NETNS_LOCAL)
5318 /* Don't allow real devices to be moved when sysfs
5322 if (dev->dev.parent)
5326 /* Ensure the device has been registrered */
5328 if (dev->reg_state != NETREG_REGISTERED)
5331 /* Get out if there is nothing todo */
5333 if (net_eq(dev_net(dev), net))
5336 /* Pick the destination device name, and ensure
5337 * we can use it in the destination network namespace.
5340 destname = dev->name;
5341 if (__dev_get_by_name(net, destname)) {
5342 /* We get here if we can't use the current device name */
5345 if (!dev_valid_name(pat))
5347 if (strchr(pat, '%')) {
5348 if (__dev_alloc_name(net, pat, buf) < 0)
5353 if (__dev_get_by_name(net, destname))
5358 * And now a mini version of register_netdevice unregister_netdevice.
5361 /* If device is running close it first. */
5364 /* And unlink it from device chain */
5366 unlist_netdevice(dev);
5370 /* Shutdown queueing discipline. */
5373 /* Notify protocols, that we are about to destroy
5374 this device. They should clean all the things.
5376 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5379 * Flush the unicast and multicast chains
5381 dev_unicast_flush(dev);
5382 dev_addr_discard(dev);
5384 netdev_unregister_kobject(dev);
5386 /* Actually switch the network namespace */
5387 dev_net_set(dev, net);
5389 /* Assign the new device name */
5390 if (destname != dev->name)
5391 strcpy(dev->name, destname);
5393 /* If there is an ifindex conflict assign a new one */
5394 if (__dev_get_by_index(net, dev->ifindex)) {
5395 int iflink = (dev->iflink == dev->ifindex);
5396 dev->ifindex = dev_new_index(net);
5398 dev->iflink = dev->ifindex;
5401 /* Fixup kobjects */
5402 err = netdev_register_kobject(dev);
5405 /* Add the device back in the hashes */
5406 list_netdevice(dev);
5408 /* Notify protocols, that a new device appeared. */
5409 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5416 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5418 static int dev_cpu_callback(struct notifier_block *nfb,
5419 unsigned long action,
5422 struct sk_buff **list_skb;
5423 struct Qdisc **list_net;
5424 struct sk_buff *skb;
5425 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5426 struct softnet_data *sd, *oldsd;
5428 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5431 local_irq_disable();
5432 cpu = smp_processor_id();
5433 sd = &per_cpu(softnet_data, cpu);
5434 oldsd = &per_cpu(softnet_data, oldcpu);
5436 /* Find end of our completion_queue. */
5437 list_skb = &sd->completion_queue;
5439 list_skb = &(*list_skb)->next;
5440 /* Append completion queue from offline CPU. */
5441 *list_skb = oldsd->completion_queue;
5442 oldsd->completion_queue = NULL;
5444 /* Find end of our output_queue. */
5445 list_net = &sd->output_queue;
5447 list_net = &(*list_net)->next_sched;
5448 /* Append output queue from offline CPU. */
5449 *list_net = oldsd->output_queue;
5450 oldsd->output_queue = NULL;
5452 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5455 /* Process offline CPU's input_pkt_queue */
5456 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5464 * netdev_increment_features - increment feature set by one
5465 * @all: current feature set
5466 * @one: new feature set
5467 * @mask: mask feature set
5469 * Computes a new feature set after adding a device with feature set
5470 * @one to the master device with current feature set @all. Will not
5471 * enable anything that is off in @mask. Returns the new feature set.
5473 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5476 /* If device needs checksumming, downgrade to it. */
5477 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5478 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5479 else if (mask & NETIF_F_ALL_CSUM) {
5480 /* If one device supports v4/v6 checksumming, set for all. */
5481 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5482 !(all & NETIF_F_GEN_CSUM)) {
5483 all &= ~NETIF_F_ALL_CSUM;
5484 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5487 /* If one device supports hw checksumming, set for all. */
5488 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5489 all &= ~NETIF_F_ALL_CSUM;
5490 all |= NETIF_F_HW_CSUM;
5494 one |= NETIF_F_ALL_CSUM;
5496 one |= all & NETIF_F_ONE_FOR_ALL;
5497 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5498 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5502 EXPORT_SYMBOL(netdev_increment_features);
5504 static struct hlist_head *netdev_create_hash(void)
5507 struct hlist_head *hash;
5509 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5511 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5512 INIT_HLIST_HEAD(&hash[i]);
5517 /* Initialize per network namespace state */
5518 static int __net_init netdev_init(struct net *net)
5520 INIT_LIST_HEAD(&net->dev_base_head);
5522 net->dev_name_head = netdev_create_hash();
5523 if (net->dev_name_head == NULL)
5526 net->dev_index_head = netdev_create_hash();
5527 if (net->dev_index_head == NULL)
5533 kfree(net->dev_name_head);
5539 * netdev_drivername - network driver for the device
5540 * @dev: network device
5541 * @buffer: buffer for resulting name
5542 * @len: size of buffer
5544 * Determine network driver for device.
5546 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5548 const struct device_driver *driver;
5549 const struct device *parent;
5551 if (len <= 0 || !buffer)
5555 parent = dev->dev.parent;
5560 driver = parent->driver;
5561 if (driver && driver->name)
5562 strlcpy(buffer, driver->name, len);
5566 static void __net_exit netdev_exit(struct net *net)
5568 kfree(net->dev_name_head);
5569 kfree(net->dev_index_head);
5572 static struct pernet_operations __net_initdata netdev_net_ops = {
5573 .init = netdev_init,
5574 .exit = netdev_exit,
5577 static void __net_exit default_device_exit(struct net *net)
5579 struct net_device *dev;
5581 * Push all migratable of the network devices back to the
5582 * initial network namespace
5586 for_each_netdev(net, dev) {
5588 char fb_name[IFNAMSIZ];
5590 /* Ignore unmoveable devices (i.e. loopback) */
5591 if (dev->features & NETIF_F_NETNS_LOCAL)
5594 /* Delete virtual devices */
5595 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5596 dev->rtnl_link_ops->dellink(dev);
5600 /* Push remaing network devices to init_net */
5601 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5602 err = dev_change_net_namespace(dev, &init_net, fb_name);
5604 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5605 __func__, dev->name, err);
5613 static struct pernet_operations __net_initdata default_device_ops = {
5614 .exit = default_device_exit,
5618 * Initialize the DEV module. At boot time this walks the device list and
5619 * unhooks any devices that fail to initialise (normally hardware not
5620 * present) and leaves us with a valid list of present and active devices.
5625 * This is called single threaded during boot, so no need
5626 * to take the rtnl semaphore.
5628 static int __init net_dev_init(void)
5630 int i, rc = -ENOMEM;
5632 BUG_ON(!dev_boot_phase);
5634 if (dev_proc_init())
5637 if (netdev_kobject_init())
5640 INIT_LIST_HEAD(&ptype_all);
5641 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5642 INIT_LIST_HEAD(&ptype_base[i]);
5644 if (register_pernet_subsys(&netdev_net_ops))
5648 * Initialise the packet receive queues.
5651 for_each_possible_cpu(i) {
5652 struct softnet_data *queue;
5654 queue = &per_cpu(softnet_data, i);
5655 skb_queue_head_init(&queue->input_pkt_queue);
5656 queue->completion_queue = NULL;
5657 INIT_LIST_HEAD(&queue->poll_list);
5659 queue->backlog.poll = process_backlog;
5660 queue->backlog.weight = weight_p;
5661 queue->backlog.gro_list = NULL;
5662 queue->backlog.gro_count = 0;
5667 /* The loopback device is special if any other network devices
5668 * is present in a network namespace the loopback device must
5669 * be present. Since we now dynamically allocate and free the
5670 * loopback device ensure this invariant is maintained by
5671 * keeping the loopback device as the first device on the
5672 * list of network devices. Ensuring the loopback devices
5673 * is the first device that appears and the last network device
5676 if (register_pernet_device(&loopback_net_ops))
5679 if (register_pernet_device(&default_device_ops))
5682 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5683 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5685 hotcpu_notifier(dev_cpu_callback, 0);
5693 subsys_initcall(net_dev_init);
5695 static int __init initialize_hashrnd(void)
5697 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5701 late_initcall_sync(initialize_hashrnd);