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/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <linux/pci.h>
132 #include <linux/inetdevice.h>
134 #include "net-sysfs.h"
136 /* Instead of increasing this, you should create a hash table. */
137 #define MAX_GRO_SKBS 8
139 /* This should be increased if a protocol with a bigger head is added. */
140 #define GRO_MAX_HEAD (MAX_HEADER + 128)
143 * The list of packet types we will receive (as opposed to discard)
144 * and the routines to invoke.
146 * Why 16. Because with 16 the only overlap we get on a hash of the
147 * low nibble of the protocol value is RARP/SNAP/X.25.
149 * NOTE: That is no longer true with the addition of VLAN tags. Not
150 * sure which should go first, but I bet it won't make much
151 * difference if we are running VLANs. The good news is that
152 * this protocol won't be in the list unless compiled in, so
153 * the average user (w/out VLANs) will not be adversely affected.
170 #define PTYPE_HASH_SIZE (16)
171 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
173 static DEFINE_SPINLOCK(ptype_lock);
174 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
175 static struct list_head ptype_all __read_mostly; /* Taps */
178 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
181 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
183 * Writers must hold the rtnl semaphore while they loop through the
184 * dev_base_head list, and hold dev_base_lock for writing when they do the
185 * actual updates. This allows pure readers to access the list even
186 * while a writer is preparing to update it.
188 * To put it another way, dev_base_lock is held for writing only to
189 * protect against pure readers; the rtnl semaphore provides the
190 * protection against other writers.
192 * See, for example usages, register_netdevice() and
193 * unregister_netdevice(), which must be called with the rtnl
196 DEFINE_RWLOCK(dev_base_lock);
197 EXPORT_SYMBOL(dev_base_lock);
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_32(hash, NETDEV_HASHBITS)];
205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
210 static inline void rps_lock(struct softnet_data *sd)
213 spin_lock(&sd->input_pkt_queue.lock);
217 static inline void rps_unlock(struct softnet_data *sd)
220 spin_unlock(&sd->input_pkt_queue.lock);
224 /* Device list insertion */
225 static int list_netdevice(struct net_device *dev)
227 struct net *net = dev_net(dev);
231 write_lock_bh(&dev_base_lock);
232 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head_rcu(&dev->index_hlist,
235 dev_index_hash(net, dev->ifindex));
236 write_unlock_bh(&dev_base_lock);
240 /* Device list removal
241 * caller must respect a RCU grace period before freeing/reusing dev
243 static void unlist_netdevice(struct net_device *dev)
247 /* Unlink dev from the device chain */
248 write_lock_bh(&dev_base_lock);
249 list_del_rcu(&dev->dev_list);
250 hlist_del_rcu(&dev->name_hlist);
251 hlist_del_rcu(&dev->index_hlist);
252 write_unlock_bh(&dev_base_lock);
259 static RAW_NOTIFIER_HEAD(netdev_chain);
262 * Device drivers call our routines to queue packets here. We empty the
263 * queue in the local softnet handler.
266 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
267 EXPORT_PER_CPU_SYMBOL(softnet_data);
269 #ifdef CONFIG_LOCKDEP
271 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
272 * according to dev->type
274 static const unsigned short netdev_lock_type[] =
275 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
276 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
277 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
278 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
279 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
280 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
281 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
282 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
283 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
284 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
285 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
286 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
287 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
288 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
289 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
290 ARPHRD_VOID, ARPHRD_NONE};
292 static const char *const netdev_lock_name[] =
293 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
294 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
295 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
296 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
297 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
298 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
299 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
300 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
301 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
302 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
303 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
304 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
305 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
306 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
307 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
308 "_xmit_VOID", "_xmit_NONE"};
310 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
311 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
313 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
317 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
318 if (netdev_lock_type[i] == dev_type)
320 /* the last key is used by default */
321 return ARRAY_SIZE(netdev_lock_type) - 1;
324 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
325 unsigned short dev_type)
329 i = netdev_lock_pos(dev_type);
330 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
331 netdev_lock_name[i]);
334 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
338 i = netdev_lock_pos(dev->type);
339 lockdep_set_class_and_name(&dev->addr_list_lock,
340 &netdev_addr_lock_key[i],
341 netdev_lock_name[i]);
344 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
345 unsigned short dev_type)
348 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
353 /*******************************************************************************
355 Protocol management and registration routines
357 *******************************************************************************/
360 * Add a protocol ID to the list. Now that the input handler is
361 * smarter we can dispense with all the messy stuff that used to be
364 * BEWARE!!! Protocol handlers, mangling input packets,
365 * MUST BE last in hash buckets and checking protocol handlers
366 * MUST start from promiscuous ptype_all chain in net_bh.
367 * It is true now, do not change it.
368 * Explanation follows: if protocol handler, mangling packet, will
369 * be the first on list, it is not able to sense, that packet
370 * is cloned and should be copied-on-write, so that it will
371 * change it and subsequent readers will get broken packet.
375 static inline struct list_head *ptype_head(const struct packet_type *pt)
377 if (pt->type == htons(ETH_P_ALL))
380 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
384 * dev_add_pack - add packet handler
385 * @pt: packet type declaration
387 * Add a protocol handler to the networking stack. The passed &packet_type
388 * is linked into kernel lists and may not be freed until it has been
389 * removed from the kernel lists.
391 * This call does not sleep therefore it can not
392 * guarantee all CPU's that are in middle of receiving packets
393 * will see the new packet type (until the next received packet).
396 void dev_add_pack(struct packet_type *pt)
398 struct list_head *head = ptype_head(pt);
400 spin_lock(&ptype_lock);
401 list_add_rcu(&pt->list, head);
402 spin_unlock(&ptype_lock);
404 EXPORT_SYMBOL(dev_add_pack);
407 * __dev_remove_pack - remove packet handler
408 * @pt: packet type declaration
410 * Remove a protocol handler that was previously added to the kernel
411 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
412 * from the kernel lists and can be freed or reused once this function
415 * The packet type might still be in use by receivers
416 * and must not be freed until after all the CPU's have gone
417 * through a quiescent state.
419 void __dev_remove_pack(struct packet_type *pt)
421 struct list_head *head = ptype_head(pt);
422 struct packet_type *pt1;
424 spin_lock(&ptype_lock);
426 list_for_each_entry(pt1, head, list) {
428 list_del_rcu(&pt->list);
433 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
435 spin_unlock(&ptype_lock);
437 EXPORT_SYMBOL(__dev_remove_pack);
440 * dev_remove_pack - remove packet handler
441 * @pt: packet type declaration
443 * Remove a protocol handler that was previously added to the kernel
444 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
445 * from the kernel lists and can be freed or reused once this function
448 * This call sleeps to guarantee that no CPU is looking at the packet
451 void dev_remove_pack(struct packet_type *pt)
453 __dev_remove_pack(pt);
457 EXPORT_SYMBOL(dev_remove_pack);
459 /******************************************************************************
461 Device Boot-time Settings Routines
463 *******************************************************************************/
465 /* Boot time configuration table */
466 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
469 * netdev_boot_setup_add - add new setup entry
470 * @name: name of the device
471 * @map: configured settings for the device
473 * Adds new setup entry to the dev_boot_setup list. The function
474 * returns 0 on error and 1 on success. This is a generic routine to
477 static int netdev_boot_setup_add(char *name, struct ifmap *map)
479 struct netdev_boot_setup *s;
483 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
484 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
485 memset(s[i].name, 0, sizeof(s[i].name));
486 strlcpy(s[i].name, name, IFNAMSIZ);
487 memcpy(&s[i].map, map, sizeof(s[i].map));
492 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
496 * netdev_boot_setup_check - check boot time settings
497 * @dev: the netdevice
499 * Check boot time settings for the device.
500 * The found settings are set for the device to be used
501 * later in the device probing.
502 * Returns 0 if no settings found, 1 if they are.
504 int netdev_boot_setup_check(struct net_device *dev)
506 struct netdev_boot_setup *s = dev_boot_setup;
509 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
510 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
511 !strcmp(dev->name, s[i].name)) {
512 dev->irq = s[i].map.irq;
513 dev->base_addr = s[i].map.base_addr;
514 dev->mem_start = s[i].map.mem_start;
515 dev->mem_end = s[i].map.mem_end;
521 EXPORT_SYMBOL(netdev_boot_setup_check);
525 * netdev_boot_base - get address from boot time settings
526 * @prefix: prefix for network device
527 * @unit: id for network device
529 * Check boot time settings for the base address of device.
530 * The found settings are set for the device to be used
531 * later in the device probing.
532 * Returns 0 if no settings found.
534 unsigned long netdev_boot_base(const char *prefix, int unit)
536 const struct netdev_boot_setup *s = dev_boot_setup;
540 sprintf(name, "%s%d", prefix, unit);
543 * If device already registered then return base of 1
544 * to indicate not to probe for this interface
546 if (__dev_get_by_name(&init_net, name))
549 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
550 if (!strcmp(name, s[i].name))
551 return s[i].map.base_addr;
556 * Saves at boot time configured settings for any netdevice.
558 int __init netdev_boot_setup(char *str)
563 str = get_options(str, ARRAY_SIZE(ints), ints);
568 memset(&map, 0, sizeof(map));
572 map.base_addr = ints[2];
574 map.mem_start = ints[3];
576 map.mem_end = ints[4];
578 /* Add new entry to the list */
579 return netdev_boot_setup_add(str, &map);
582 __setup("netdev=", netdev_boot_setup);
584 /*******************************************************************************
586 Device Interface Subroutines
588 *******************************************************************************/
591 * __dev_get_by_name - find a device by its name
592 * @net: the applicable net namespace
593 * @name: name to find
595 * Find an interface by name. Must be called under RTNL semaphore
596 * or @dev_base_lock. If the name is found a pointer to the device
597 * is returned. If the name is not found then %NULL is returned. The
598 * reference counters are not incremented so the caller must be
599 * careful with locks.
602 struct net_device *__dev_get_by_name(struct net *net, const char *name)
604 struct hlist_node *p;
605 struct net_device *dev;
606 struct hlist_head *head = dev_name_hash(net, name);
608 hlist_for_each_entry(dev, p, head, name_hlist)
609 if (!strncmp(dev->name, name, IFNAMSIZ))
614 EXPORT_SYMBOL(__dev_get_by_name);
617 * dev_get_by_name_rcu - find a device by its name
618 * @net: the applicable net namespace
619 * @name: name to find
621 * Find an interface by name.
622 * If the name is found a pointer to the device is returned.
623 * If the name is not found then %NULL is returned.
624 * The reference counters are not incremented so the caller must be
625 * careful with locks. The caller must hold RCU lock.
628 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
630 struct hlist_node *p;
631 struct net_device *dev;
632 struct hlist_head *head = dev_name_hash(net, name);
634 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
635 if (!strncmp(dev->name, name, IFNAMSIZ))
640 EXPORT_SYMBOL(dev_get_by_name_rcu);
643 * dev_get_by_name - find a device by its name
644 * @net: the applicable net namespace
645 * @name: name to find
647 * Find an interface by name. This can be called from any
648 * context and does its own locking. The returned handle has
649 * the usage count incremented and the caller must use dev_put() to
650 * release it when it is no longer needed. %NULL is returned if no
651 * matching device is found.
654 struct net_device *dev_get_by_name(struct net *net, const char *name)
656 struct net_device *dev;
659 dev = dev_get_by_name_rcu(net, name);
665 EXPORT_SYMBOL(dev_get_by_name);
668 * __dev_get_by_index - find a device by its ifindex
669 * @net: the applicable net namespace
670 * @ifindex: index of device
672 * Search for an interface by index. Returns %NULL if the device
673 * is not found or a pointer to the device. The device has not
674 * had its reference counter increased so the caller must be careful
675 * about locking. The caller must hold either the RTNL semaphore
679 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
681 struct hlist_node *p;
682 struct net_device *dev;
683 struct hlist_head *head = dev_index_hash(net, ifindex);
685 hlist_for_each_entry(dev, p, head, index_hlist)
686 if (dev->ifindex == ifindex)
691 EXPORT_SYMBOL(__dev_get_by_index);
694 * dev_get_by_index_rcu - find a device by its ifindex
695 * @net: the applicable net namespace
696 * @ifindex: index of device
698 * Search for an interface by index. Returns %NULL if the device
699 * is not found or a pointer to the device. The device has not
700 * had its reference counter increased so the caller must be careful
701 * about locking. The caller must hold RCU lock.
704 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
706 struct hlist_node *p;
707 struct net_device *dev;
708 struct hlist_head *head = dev_index_hash(net, ifindex);
710 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
711 if (dev->ifindex == ifindex)
716 EXPORT_SYMBOL(dev_get_by_index_rcu);
720 * dev_get_by_index - find a device by its ifindex
721 * @net: the applicable net namespace
722 * @ifindex: index of device
724 * Search for an interface by index. Returns NULL if the device
725 * is not found or a pointer to the device. The device returned has
726 * had a reference added and the pointer is safe until the user calls
727 * dev_put to indicate they have finished with it.
730 struct net_device *dev_get_by_index(struct net *net, int ifindex)
732 struct net_device *dev;
735 dev = dev_get_by_index_rcu(net, ifindex);
741 EXPORT_SYMBOL(dev_get_by_index);
744 * dev_getbyhwaddr - find a device by its hardware address
745 * @net: the applicable net namespace
746 * @type: media type of device
747 * @ha: hardware address
749 * Search for an interface by MAC address. Returns NULL if the device
750 * is not found or a pointer to the device. The caller must hold the
751 * rtnl semaphore. The returned device has not had its ref count increased
752 * and the caller must therefore be careful about locking
755 * If the API was consistent this would be __dev_get_by_hwaddr
758 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
760 struct net_device *dev;
764 for_each_netdev(net, dev)
765 if (dev->type == type &&
766 !memcmp(dev->dev_addr, ha, dev->addr_len))
771 EXPORT_SYMBOL(dev_getbyhwaddr);
773 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
775 struct net_device *dev;
778 for_each_netdev(net, dev)
779 if (dev->type == type)
784 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
786 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
788 struct net_device *dev, *ret = NULL;
791 for_each_netdev_rcu(net, dev)
792 if (dev->type == type) {
800 EXPORT_SYMBOL(dev_getfirstbyhwtype);
803 * dev_get_by_flags_rcu - find any device with given flags
804 * @net: the applicable net namespace
805 * @if_flags: IFF_* values
806 * @mask: bitmask of bits in if_flags to check
808 * Search for any interface with the given flags. Returns NULL if a device
809 * is not found or a pointer to the device. Must be called inside
810 * rcu_read_lock(), and result refcount is unchanged.
813 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
816 struct net_device *dev, *ret;
819 for_each_netdev_rcu(net, dev) {
820 if (((dev->flags ^ if_flags) & mask) == 0) {
827 EXPORT_SYMBOL(dev_get_by_flags_rcu);
830 * dev_valid_name - check if name is okay for network device
833 * Network device names need to be valid file names to
834 * to allow sysfs to work. We also disallow any kind of
837 int dev_valid_name(const char *name)
841 if (strlen(name) >= IFNAMSIZ)
843 if (!strcmp(name, ".") || !strcmp(name, ".."))
847 if (*name == '/' || isspace(*name))
853 EXPORT_SYMBOL(dev_valid_name);
856 * __dev_alloc_name - allocate a name for a device
857 * @net: network namespace to allocate the device name in
858 * @name: name format string
859 * @buf: scratch buffer and result name string
861 * Passed a format string - eg "lt%d" it will try and find a suitable
862 * id. It scans list of devices to build up a free map, then chooses
863 * the first empty slot. The caller must hold the dev_base or rtnl lock
864 * while allocating the name and adding the device in order to avoid
866 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
867 * Returns the number of the unit assigned or a negative errno code.
870 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
874 const int max_netdevices = 8*PAGE_SIZE;
875 unsigned long *inuse;
876 struct net_device *d;
878 p = strnchr(name, IFNAMSIZ-1, '%');
881 * Verify the string as this thing may have come from
882 * the user. There must be either one "%d" and no other "%"
885 if (p[1] != 'd' || strchr(p + 2, '%'))
888 /* Use one page as a bit array of possible slots */
889 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
893 for_each_netdev(net, d) {
894 if (!sscanf(d->name, name, &i))
896 if (i < 0 || i >= max_netdevices)
899 /* avoid cases where sscanf is not exact inverse of printf */
900 snprintf(buf, IFNAMSIZ, name, i);
901 if (!strncmp(buf, d->name, IFNAMSIZ))
905 i = find_first_zero_bit(inuse, max_netdevices);
906 free_page((unsigned long) inuse);
910 snprintf(buf, IFNAMSIZ, name, i);
911 if (!__dev_get_by_name(net, buf))
914 /* It is possible to run out of possible slots
915 * when the name is long and there isn't enough space left
916 * for the digits, or if all bits are used.
922 * dev_alloc_name - allocate a name for a device
924 * @name: name format string
926 * Passed a format string - eg "lt%d" it will try and find a suitable
927 * id. It scans list of devices to build up a free map, then chooses
928 * the first empty slot. The caller must hold the dev_base or rtnl lock
929 * while allocating the name and adding the device in order to avoid
931 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
932 * Returns the number of the unit assigned or a negative errno code.
935 int dev_alloc_name(struct net_device *dev, const char *name)
941 BUG_ON(!dev_net(dev));
943 ret = __dev_alloc_name(net, name, buf);
945 strlcpy(dev->name, buf, IFNAMSIZ);
948 EXPORT_SYMBOL(dev_alloc_name);
950 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
954 BUG_ON(!dev_net(dev));
957 if (!dev_valid_name(name))
960 if (fmt && strchr(name, '%'))
961 return dev_alloc_name(dev, name);
962 else if (__dev_get_by_name(net, name))
964 else if (dev->name != name)
965 strlcpy(dev->name, name, IFNAMSIZ);
971 * dev_change_name - change name of a device
973 * @newname: name (or format string) must be at least IFNAMSIZ
975 * Change name of a device, can pass format strings "eth%d".
978 int dev_change_name(struct net_device *dev, const char *newname)
980 char oldname[IFNAMSIZ];
986 BUG_ON(!dev_net(dev));
989 if (dev->flags & IFF_UP)
992 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
995 memcpy(oldname, dev->name, IFNAMSIZ);
997 err = dev_get_valid_name(dev, newname, 1);
1002 ret = device_rename(&dev->dev, dev->name);
1004 memcpy(dev->name, oldname, IFNAMSIZ);
1008 write_lock_bh(&dev_base_lock);
1009 hlist_del(&dev->name_hlist);
1010 write_unlock_bh(&dev_base_lock);
1014 write_lock_bh(&dev_base_lock);
1015 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1016 write_unlock_bh(&dev_base_lock);
1018 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1019 ret = notifier_to_errno(ret);
1022 /* err >= 0 after dev_alloc_name() or stores the first errno */
1025 memcpy(dev->name, oldname, IFNAMSIZ);
1029 "%s: name change rollback failed: %d.\n",
1038 * dev_set_alias - change ifalias of a device
1040 * @alias: name up to IFALIASZ
1041 * @len: limit of bytes to copy from info
1043 * Set ifalias for a device,
1045 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1049 if (len >= IFALIASZ)
1054 kfree(dev->ifalias);
1055 dev->ifalias = NULL;
1060 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1064 strlcpy(dev->ifalias, alias, len+1);
1070 * netdev_features_change - device changes features
1071 * @dev: device to cause notification
1073 * Called to indicate a device has changed features.
1075 void netdev_features_change(struct net_device *dev)
1077 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1079 EXPORT_SYMBOL(netdev_features_change);
1082 * netdev_state_change - device changes state
1083 * @dev: device to cause notification
1085 * Called to indicate a device has changed state. This function calls
1086 * the notifier chains for netdev_chain and sends a NEWLINK message
1087 * to the routing socket.
1089 void netdev_state_change(struct net_device *dev)
1091 if (dev->flags & IFF_UP) {
1092 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1093 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1096 EXPORT_SYMBOL(netdev_state_change);
1098 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1100 return call_netdevice_notifiers(event, dev);
1102 EXPORT_SYMBOL(netdev_bonding_change);
1105 * dev_load - load a network module
1106 * @net: the applicable net namespace
1107 * @name: name of interface
1109 * If a network interface is not present and the process has suitable
1110 * privileges this function loads the module. If module loading is not
1111 * available in this kernel then it becomes a nop.
1114 void dev_load(struct net *net, const char *name)
1116 struct net_device *dev;
1119 dev = dev_get_by_name_rcu(net, name);
1122 if (!dev && capable(CAP_NET_ADMIN))
1123 request_module("%s", name);
1125 EXPORT_SYMBOL(dev_load);
1127 static int __dev_open(struct net_device *dev)
1129 const struct net_device_ops *ops = dev->netdev_ops;
1135 * Is it even present?
1137 if (!netif_device_present(dev))
1140 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1141 ret = notifier_to_errno(ret);
1146 * Call device private open method
1148 set_bit(__LINK_STATE_START, &dev->state);
1150 if (ops->ndo_validate_addr)
1151 ret = ops->ndo_validate_addr(dev);
1153 if (!ret && ops->ndo_open)
1154 ret = ops->ndo_open(dev);
1157 * If it went open OK then:
1161 clear_bit(__LINK_STATE_START, &dev->state);
1166 dev->flags |= IFF_UP;
1171 net_dmaengine_get();
1174 * Initialize multicasting status
1176 dev_set_rx_mode(dev);
1179 * Wakeup transmit queue engine
1188 * dev_open - prepare an interface for use.
1189 * @dev: device to open
1191 * Takes a device from down to up state. The device's private open
1192 * function is invoked and then the multicast lists are loaded. Finally
1193 * the device is moved into the up state and a %NETDEV_UP message is
1194 * sent to the netdev notifier chain.
1196 * Calling this function on an active interface is a nop. On a failure
1197 * a negative errno code is returned.
1199 int dev_open(struct net_device *dev)
1206 if (dev->flags & IFF_UP)
1212 ret = __dev_open(dev);
1217 * ... and announce new interface.
1219 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1220 call_netdevice_notifiers(NETDEV_UP, dev);
1224 EXPORT_SYMBOL(dev_open);
1226 static int __dev_close(struct net_device *dev)
1228 const struct net_device_ops *ops = dev->netdev_ops;
1234 * Tell people we are going down, so that they can
1235 * prepare to death, when device is still operating.
1237 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1239 clear_bit(__LINK_STATE_START, &dev->state);
1241 /* Synchronize to scheduled poll. We cannot touch poll list,
1242 * it can be even on different cpu. So just clear netif_running().
1244 * dev->stop() will invoke napi_disable() on all of it's
1245 * napi_struct instances on this device.
1247 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1249 dev_deactivate(dev);
1252 * Call the device specific close. This cannot fail.
1253 * Only if device is UP
1255 * We allow it to be called even after a DETACH hot-plug
1262 * Device is now down.
1265 dev->flags &= ~IFF_UP;
1270 net_dmaengine_put();
1276 * dev_close - shutdown an interface.
1277 * @dev: device to shutdown
1279 * This function moves an active device into down state. A
1280 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1281 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1284 int dev_close(struct net_device *dev)
1286 if (!(dev->flags & IFF_UP))
1292 * Tell people we are down
1294 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1295 call_netdevice_notifiers(NETDEV_DOWN, dev);
1299 EXPORT_SYMBOL(dev_close);
1303 * dev_disable_lro - disable Large Receive Offload on a device
1306 * Disable Large Receive Offload (LRO) on a net device. Must be
1307 * called under RTNL. This is needed if received packets may be
1308 * forwarded to another interface.
1310 void dev_disable_lro(struct net_device *dev)
1312 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1313 dev->ethtool_ops->set_flags) {
1314 u32 flags = dev->ethtool_ops->get_flags(dev);
1315 if (flags & ETH_FLAG_LRO) {
1316 flags &= ~ETH_FLAG_LRO;
1317 dev->ethtool_ops->set_flags(dev, flags);
1320 WARN_ON(dev->features & NETIF_F_LRO);
1322 EXPORT_SYMBOL(dev_disable_lro);
1325 static int dev_boot_phase = 1;
1328 * Device change register/unregister. These are not inline or static
1329 * as we export them to the world.
1333 * register_netdevice_notifier - register a network notifier block
1336 * Register a notifier to be called when network device events occur.
1337 * The notifier passed is linked into the kernel structures and must
1338 * not be reused until it has been unregistered. A negative errno code
1339 * is returned on a failure.
1341 * When registered all registration and up events are replayed
1342 * to the new notifier to allow device to have a race free
1343 * view of the network device list.
1346 int register_netdevice_notifier(struct notifier_block *nb)
1348 struct net_device *dev;
1349 struct net_device *last;
1354 err = raw_notifier_chain_register(&netdev_chain, nb);
1360 for_each_netdev(net, dev) {
1361 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1362 err = notifier_to_errno(err);
1366 if (!(dev->flags & IFF_UP))
1369 nb->notifier_call(nb, NETDEV_UP, dev);
1380 for_each_netdev(net, dev) {
1384 if (dev->flags & IFF_UP) {
1385 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1386 nb->notifier_call(nb, NETDEV_DOWN, dev);
1388 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1389 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1393 raw_notifier_chain_unregister(&netdev_chain, nb);
1396 EXPORT_SYMBOL(register_netdevice_notifier);
1399 * unregister_netdevice_notifier - unregister a network notifier block
1402 * Unregister a notifier previously registered by
1403 * register_netdevice_notifier(). The notifier is unlinked into the
1404 * kernel structures and may then be reused. A negative errno code
1405 * is returned on a failure.
1408 int unregister_netdevice_notifier(struct notifier_block *nb)
1413 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1417 EXPORT_SYMBOL(unregister_netdevice_notifier);
1420 * call_netdevice_notifiers - call all network notifier blocks
1421 * @val: value passed unmodified to notifier function
1422 * @dev: net_device pointer passed unmodified to notifier function
1424 * Call all network notifier blocks. Parameters and return value
1425 * are as for raw_notifier_call_chain().
1428 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1431 return raw_notifier_call_chain(&netdev_chain, val, dev);
1434 /* When > 0 there are consumers of rx skb time stamps */
1435 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1437 void net_enable_timestamp(void)
1439 atomic_inc(&netstamp_needed);
1441 EXPORT_SYMBOL(net_enable_timestamp);
1443 void net_disable_timestamp(void)
1445 atomic_dec(&netstamp_needed);
1447 EXPORT_SYMBOL(net_disable_timestamp);
1449 static inline void net_timestamp_set(struct sk_buff *skb)
1451 if (atomic_read(&netstamp_needed))
1452 __net_timestamp(skb);
1454 skb->tstamp.tv64 = 0;
1457 static inline void net_timestamp_check(struct sk_buff *skb)
1459 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1460 __net_timestamp(skb);
1464 * dev_forward_skb - loopback an skb to another netif
1466 * @dev: destination network device
1467 * @skb: buffer to forward
1470 * NET_RX_SUCCESS (no congestion)
1471 * NET_RX_DROP (packet was dropped, but freed)
1473 * dev_forward_skb can be used for injecting an skb from the
1474 * start_xmit function of one device into the receive queue
1475 * of another device.
1477 * The receiving device may be in another namespace, so
1478 * we have to clear all information in the skb that could
1479 * impact namespace isolation.
1481 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1486 if (!(dev->flags & IFF_UP) ||
1487 (skb->len > (dev->mtu + dev->hard_header_len))) {
1491 skb_set_dev(skb, dev);
1492 skb->tstamp.tv64 = 0;
1493 skb->pkt_type = PACKET_HOST;
1494 skb->protocol = eth_type_trans(skb, dev);
1495 return netif_rx(skb);
1497 EXPORT_SYMBOL_GPL(dev_forward_skb);
1500 * Support routine. Sends outgoing frames to any network
1501 * taps currently in use.
1504 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1506 struct packet_type *ptype;
1508 #ifdef CONFIG_NET_CLS_ACT
1509 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1510 net_timestamp_set(skb);
1512 net_timestamp_set(skb);
1516 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1517 /* Never send packets back to the socket
1518 * they originated from - MvS (miquels@drinkel.ow.org)
1520 if ((ptype->dev == dev || !ptype->dev) &&
1521 (ptype->af_packet_priv == NULL ||
1522 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1523 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1527 /* skb->nh should be correctly
1528 set by sender, so that the second statement is
1529 just protection against buggy protocols.
1531 skb_reset_mac_header(skb2);
1533 if (skb_network_header(skb2) < skb2->data ||
1534 skb2->network_header > skb2->tail) {
1535 if (net_ratelimit())
1536 printk(KERN_CRIT "protocol %04x is "
1538 ntohs(skb2->protocol),
1540 skb_reset_network_header(skb2);
1543 skb2->transport_header = skb2->network_header;
1544 skb2->pkt_type = PACKET_OUTGOING;
1545 ptype->func(skb2, skb->dev, ptype, skb->dev);
1552 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1553 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1555 void netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1557 unsigned int real_num = dev->real_num_tx_queues;
1559 if (unlikely(txq > dev->num_tx_queues))
1561 else if (txq > real_num)
1562 dev->real_num_tx_queues = txq;
1563 else if (txq < real_num) {
1564 dev->real_num_tx_queues = txq;
1565 qdisc_reset_all_tx_gt(dev, txq);
1568 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1572 * netif_set_real_num_rx_queues - set actual number of RX queues used
1573 * @dev: Network device
1574 * @rxq: Actual number of RX queues
1576 * This must be called either with the rtnl_lock held or before
1577 * registration of the net device. Returns 0 on success, or a
1578 * negative error code. If called before registration, it also
1579 * sets the maximum number of queues, and always succeeds.
1581 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1585 if (dev->reg_state == NETREG_REGISTERED) {
1588 if (rxq > dev->num_rx_queues)
1591 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1596 dev->num_rx_queues = rxq;
1599 dev->real_num_rx_queues = rxq;
1602 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1605 static inline void __netif_reschedule(struct Qdisc *q)
1607 struct softnet_data *sd;
1608 unsigned long flags;
1610 local_irq_save(flags);
1611 sd = &__get_cpu_var(softnet_data);
1612 q->next_sched = NULL;
1613 *sd->output_queue_tailp = q;
1614 sd->output_queue_tailp = &q->next_sched;
1615 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1616 local_irq_restore(flags);
1619 void __netif_schedule(struct Qdisc *q)
1621 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1622 __netif_reschedule(q);
1624 EXPORT_SYMBOL(__netif_schedule);
1626 void dev_kfree_skb_irq(struct sk_buff *skb)
1628 if (atomic_dec_and_test(&skb->users)) {
1629 struct softnet_data *sd;
1630 unsigned long flags;
1632 local_irq_save(flags);
1633 sd = &__get_cpu_var(softnet_data);
1634 skb->next = sd->completion_queue;
1635 sd->completion_queue = skb;
1636 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1637 local_irq_restore(flags);
1640 EXPORT_SYMBOL(dev_kfree_skb_irq);
1642 void dev_kfree_skb_any(struct sk_buff *skb)
1644 if (in_irq() || irqs_disabled())
1645 dev_kfree_skb_irq(skb);
1649 EXPORT_SYMBOL(dev_kfree_skb_any);
1653 * netif_device_detach - mark device as removed
1654 * @dev: network device
1656 * Mark device as removed from system and therefore no longer available.
1658 void netif_device_detach(struct net_device *dev)
1660 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1661 netif_running(dev)) {
1662 netif_tx_stop_all_queues(dev);
1665 EXPORT_SYMBOL(netif_device_detach);
1668 * netif_device_attach - mark device as attached
1669 * @dev: network device
1671 * Mark device as attached from system and restart if needed.
1673 void netif_device_attach(struct net_device *dev)
1675 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1676 netif_running(dev)) {
1677 netif_tx_wake_all_queues(dev);
1678 __netdev_watchdog_up(dev);
1681 EXPORT_SYMBOL(netif_device_attach);
1683 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1685 return ((features & NETIF_F_GEN_CSUM) ||
1686 ((features & NETIF_F_IP_CSUM) &&
1687 protocol == htons(ETH_P_IP)) ||
1688 ((features & NETIF_F_IPV6_CSUM) &&
1689 protocol == htons(ETH_P_IPV6)) ||
1690 ((features & NETIF_F_FCOE_CRC) &&
1691 protocol == htons(ETH_P_FCOE)));
1694 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1696 if (can_checksum_protocol(dev->features, skb->protocol))
1699 if (skb->protocol == htons(ETH_P_8021Q)) {
1700 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1701 if (can_checksum_protocol(dev->features & dev->vlan_features,
1702 veh->h_vlan_encapsulated_proto))
1710 * skb_dev_set -- assign a new device to a buffer
1711 * @skb: buffer for the new device
1712 * @dev: network device
1714 * If an skb is owned by a device already, we have to reset
1715 * all data private to the namespace a device belongs to
1716 * before assigning it a new device.
1718 #ifdef CONFIG_NET_NS
1719 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1722 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1725 skb_init_secmark(skb);
1729 skb->ipvs_property = 0;
1730 #ifdef CONFIG_NET_SCHED
1736 EXPORT_SYMBOL(skb_set_dev);
1737 #endif /* CONFIG_NET_NS */
1740 * Invalidate hardware checksum when packet is to be mangled, and
1741 * complete checksum manually on outgoing path.
1743 int skb_checksum_help(struct sk_buff *skb)
1746 int ret = 0, offset;
1748 if (skb->ip_summed == CHECKSUM_COMPLETE)
1749 goto out_set_summed;
1751 if (unlikely(skb_shinfo(skb)->gso_size)) {
1752 /* Let GSO fix up the checksum. */
1753 goto out_set_summed;
1756 offset = skb->csum_start - skb_headroom(skb);
1757 BUG_ON(offset >= skb_headlen(skb));
1758 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1760 offset += skb->csum_offset;
1761 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1763 if (skb_cloned(skb) &&
1764 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1765 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1770 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1772 skb->ip_summed = CHECKSUM_NONE;
1776 EXPORT_SYMBOL(skb_checksum_help);
1779 * skb_gso_segment - Perform segmentation on skb.
1780 * @skb: buffer to segment
1781 * @features: features for the output path (see dev->features)
1783 * This function segments the given skb and returns a list of segments.
1785 * It may return NULL if the skb requires no segmentation. This is
1786 * only possible when GSO is used for verifying header integrity.
1788 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1790 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1791 struct packet_type *ptype;
1792 __be16 type = skb->protocol;
1795 skb_reset_mac_header(skb);
1796 skb->mac_len = skb->network_header - skb->mac_header;
1797 __skb_pull(skb, skb->mac_len);
1799 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1800 struct net_device *dev = skb->dev;
1801 struct ethtool_drvinfo info = {};
1803 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1804 dev->ethtool_ops->get_drvinfo(dev, &info);
1806 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1808 info.driver, dev ? dev->features : 0L,
1809 skb->sk ? skb->sk->sk_route_caps : 0L,
1810 skb->len, skb->data_len, skb->ip_summed);
1812 if (skb_header_cloned(skb) &&
1813 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1814 return ERR_PTR(err);
1818 list_for_each_entry_rcu(ptype,
1819 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1820 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1821 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1822 err = ptype->gso_send_check(skb);
1823 segs = ERR_PTR(err);
1824 if (err || skb_gso_ok(skb, features))
1826 __skb_push(skb, (skb->data -
1827 skb_network_header(skb)));
1829 segs = ptype->gso_segment(skb, features);
1835 __skb_push(skb, skb->data - skb_mac_header(skb));
1839 EXPORT_SYMBOL(skb_gso_segment);
1841 /* Take action when hardware reception checksum errors are detected. */
1843 void netdev_rx_csum_fault(struct net_device *dev)
1845 if (net_ratelimit()) {
1846 printk(KERN_ERR "%s: hw csum failure.\n",
1847 dev ? dev->name : "<unknown>");
1851 EXPORT_SYMBOL(netdev_rx_csum_fault);
1854 /* Actually, we should eliminate this check as soon as we know, that:
1855 * 1. IOMMU is present and allows to map all the memory.
1856 * 2. No high memory really exists on this machine.
1859 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1861 #ifdef CONFIG_HIGHMEM
1863 if (!(dev->features & NETIF_F_HIGHDMA)) {
1864 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1865 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1869 if (PCI_DMA_BUS_IS_PHYS) {
1870 struct device *pdev = dev->dev.parent;
1874 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1875 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1876 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1885 void (*destructor)(struct sk_buff *skb);
1888 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1890 static void dev_gso_skb_destructor(struct sk_buff *skb)
1892 struct dev_gso_cb *cb;
1895 struct sk_buff *nskb = skb->next;
1897 skb->next = nskb->next;
1900 } while (skb->next);
1902 cb = DEV_GSO_CB(skb);
1904 cb->destructor(skb);
1908 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1909 * @skb: buffer to segment
1911 * This function segments the given skb and stores the list of segments
1914 static int dev_gso_segment(struct sk_buff *skb)
1916 struct net_device *dev = skb->dev;
1917 struct sk_buff *segs;
1918 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1921 segs = skb_gso_segment(skb, features);
1923 /* Verifying header integrity only. */
1928 return PTR_ERR(segs);
1931 DEV_GSO_CB(skb)->destructor = skb->destructor;
1932 skb->destructor = dev_gso_skb_destructor;
1938 * Try to orphan skb early, right before transmission by the device.
1939 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1940 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1942 static inline void skb_orphan_try(struct sk_buff *skb)
1944 struct sock *sk = skb->sk;
1946 if (sk && !skb_shinfo(skb)->tx_flags) {
1947 /* skb_tx_hash() wont be able to get sk.
1948 * We copy sk_hash into skb->rxhash
1951 skb->rxhash = sk->sk_hash;
1957 * Returns true if either:
1958 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1959 * 2. skb is fragmented and the device does not support SG, or if
1960 * at least one of fragments is in highmem and device does not
1961 * support DMA from it.
1963 static inline int skb_needs_linearize(struct sk_buff *skb,
1964 struct net_device *dev)
1966 return skb_is_nonlinear(skb) &&
1967 ((skb_has_frag_list(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
1968 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
1969 illegal_highdma(dev, skb))));
1972 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1973 struct netdev_queue *txq)
1975 const struct net_device_ops *ops = dev->netdev_ops;
1976 int rc = NETDEV_TX_OK;
1978 if (likely(!skb->next)) {
1979 if (!list_empty(&ptype_all))
1980 dev_queue_xmit_nit(skb, dev);
1983 * If device doesnt need skb->dst, release it right now while
1984 * its hot in this cpu cache
1986 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1989 skb_orphan_try(skb);
1991 if (netif_needs_gso(dev, skb)) {
1992 if (unlikely(dev_gso_segment(skb)))
1997 if (skb_needs_linearize(skb, dev) &&
1998 __skb_linearize(skb))
2001 /* If packet is not checksummed and device does not
2002 * support checksumming for this protocol, complete
2003 * checksumming here.
2005 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2006 skb_set_transport_header(skb, skb->csum_start -
2008 if (!dev_can_checksum(dev, skb) &&
2009 skb_checksum_help(skb))
2014 rc = ops->ndo_start_xmit(skb, dev);
2015 if (rc == NETDEV_TX_OK)
2016 txq_trans_update(txq);
2022 struct sk_buff *nskb = skb->next;
2024 skb->next = nskb->next;
2028 * If device doesnt need nskb->dst, release it right now while
2029 * its hot in this cpu cache
2031 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2034 rc = ops->ndo_start_xmit(nskb, dev);
2035 if (unlikely(rc != NETDEV_TX_OK)) {
2036 if (rc & ~NETDEV_TX_MASK)
2037 goto out_kfree_gso_skb;
2038 nskb->next = skb->next;
2042 txq_trans_update(txq);
2043 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2044 return NETDEV_TX_BUSY;
2045 } while (skb->next);
2048 if (likely(skb->next == NULL))
2049 skb->destructor = DEV_GSO_CB(skb)->destructor;
2055 static u32 hashrnd __read_mostly;
2057 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2061 if (skb_rx_queue_recorded(skb)) {
2062 hash = skb_get_rx_queue(skb);
2063 while (unlikely(hash >= dev->real_num_tx_queues))
2064 hash -= dev->real_num_tx_queues;
2068 if (skb->sk && skb->sk->sk_hash)
2069 hash = skb->sk->sk_hash;
2071 hash = (__force u16) skb->protocol ^ skb->rxhash;
2072 hash = jhash_1word(hash, hashrnd);
2074 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2076 EXPORT_SYMBOL(skb_tx_hash);
2078 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2080 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2081 if (net_ratelimit()) {
2082 pr_warning("%s selects TX queue %d, but "
2083 "real number of TX queues is %d\n",
2084 dev->name, queue_index, dev->real_num_tx_queues);
2091 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2092 struct sk_buff *skb)
2095 const struct net_device_ops *ops = dev->netdev_ops;
2097 if (ops->ndo_select_queue) {
2098 queue_index = ops->ndo_select_queue(dev, skb);
2099 queue_index = dev_cap_txqueue(dev, queue_index);
2101 struct sock *sk = skb->sk;
2102 queue_index = sk_tx_queue_get(sk);
2103 if (queue_index < 0) {
2106 if (dev->real_num_tx_queues > 1)
2107 queue_index = skb_tx_hash(dev, skb);
2110 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2112 if (dst && skb_dst(skb) == dst)
2113 sk_tx_queue_set(sk, queue_index);
2118 skb_set_queue_mapping(skb, queue_index);
2119 return netdev_get_tx_queue(dev, queue_index);
2122 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2123 struct net_device *dev,
2124 struct netdev_queue *txq)
2126 spinlock_t *root_lock = qdisc_lock(q);
2127 bool contended = qdisc_is_running(q);
2131 * Heuristic to force contended enqueues to serialize on a
2132 * separate lock before trying to get qdisc main lock.
2133 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2134 * and dequeue packets faster.
2136 if (unlikely(contended))
2137 spin_lock(&q->busylock);
2139 spin_lock(root_lock);
2140 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2143 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2144 qdisc_run_begin(q)) {
2146 * This is a work-conserving queue; there are no old skbs
2147 * waiting to be sent out; and the qdisc is not running -
2148 * xmit the skb directly.
2150 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2152 __qdisc_update_bstats(q, skb->len);
2153 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2154 if (unlikely(contended)) {
2155 spin_unlock(&q->busylock);
2162 rc = NET_XMIT_SUCCESS;
2165 rc = qdisc_enqueue_root(skb, q);
2166 if (qdisc_run_begin(q)) {
2167 if (unlikely(contended)) {
2168 spin_unlock(&q->busylock);
2174 spin_unlock(root_lock);
2175 if (unlikely(contended))
2176 spin_unlock(&q->busylock);
2181 * dev_queue_xmit - transmit a buffer
2182 * @skb: buffer to transmit
2184 * Queue a buffer for transmission to a network device. The caller must
2185 * have set the device and priority and built the buffer before calling
2186 * this function. The function can be called from an interrupt.
2188 * A negative errno code is returned on a failure. A success does not
2189 * guarantee the frame will be transmitted as it may be dropped due
2190 * to congestion or traffic shaping.
2192 * -----------------------------------------------------------------------------------
2193 * I notice this method can also return errors from the queue disciplines,
2194 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2197 * Regardless of the return value, the skb is consumed, so it is currently
2198 * difficult to retry a send to this method. (You can bump the ref count
2199 * before sending to hold a reference for retry if you are careful.)
2201 * When calling this method, interrupts MUST be enabled. This is because
2202 * the BH enable code must have IRQs enabled so that it will not deadlock.
2205 int dev_queue_xmit(struct sk_buff *skb)
2207 struct net_device *dev = skb->dev;
2208 struct netdev_queue *txq;
2212 /* Disable soft irqs for various locks below. Also
2213 * stops preemption for RCU.
2217 txq = dev_pick_tx(dev, skb);
2218 q = rcu_dereference_bh(txq->qdisc);
2220 #ifdef CONFIG_NET_CLS_ACT
2221 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2224 rc = __dev_xmit_skb(skb, q, dev, txq);
2228 /* The device has no queue. Common case for software devices:
2229 loopback, all the sorts of tunnels...
2231 Really, it is unlikely that netif_tx_lock protection is necessary
2232 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2234 However, it is possible, that they rely on protection
2237 Check this and shot the lock. It is not prone from deadlocks.
2238 Either shot noqueue qdisc, it is even simpler 8)
2240 if (dev->flags & IFF_UP) {
2241 int cpu = smp_processor_id(); /* ok because BHs are off */
2243 if (txq->xmit_lock_owner != cpu) {
2245 HARD_TX_LOCK(dev, txq, cpu);
2247 if (!netif_tx_queue_stopped(txq)) {
2248 rc = dev_hard_start_xmit(skb, dev, txq);
2249 if (dev_xmit_complete(rc)) {
2250 HARD_TX_UNLOCK(dev, txq);
2254 HARD_TX_UNLOCK(dev, txq);
2255 if (net_ratelimit())
2256 printk(KERN_CRIT "Virtual device %s asks to "
2257 "queue packet!\n", dev->name);
2259 /* Recursion is detected! It is possible,
2261 if (net_ratelimit())
2262 printk(KERN_CRIT "Dead loop on virtual device "
2263 "%s, fix it urgently!\n", dev->name);
2268 rcu_read_unlock_bh();
2273 rcu_read_unlock_bh();
2276 EXPORT_SYMBOL(dev_queue_xmit);
2279 /*=======================================================================
2281 =======================================================================*/
2283 int netdev_max_backlog __read_mostly = 1000;
2284 int netdev_tstamp_prequeue __read_mostly = 1;
2285 int netdev_budget __read_mostly = 300;
2286 int weight_p __read_mostly = 64; /* old backlog weight */
2288 /* Called with irq disabled */
2289 static inline void ____napi_schedule(struct softnet_data *sd,
2290 struct napi_struct *napi)
2292 list_add_tail(&napi->poll_list, &sd->poll_list);
2293 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2297 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2298 * and src/dst port numbers. Returns a non-zero hash number on success
2301 __u32 __skb_get_rxhash(struct sk_buff *skb)
2303 int nhoff, hash = 0, poff;
2304 struct ipv6hdr *ip6;
2307 u32 addr1, addr2, ihl;
2313 nhoff = skb_network_offset(skb);
2315 switch (skb->protocol) {
2316 case __constant_htons(ETH_P_IP):
2317 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2320 ip = (struct iphdr *) (skb->data + nhoff);
2321 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2324 ip_proto = ip->protocol;
2325 addr1 = (__force u32) ip->saddr;
2326 addr2 = (__force u32) ip->daddr;
2329 case __constant_htons(ETH_P_IPV6):
2330 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2333 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2334 ip_proto = ip6->nexthdr;
2335 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2336 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2344 poff = proto_ports_offset(ip_proto);
2346 nhoff += ihl * 4 + poff;
2347 if (pskb_may_pull(skb, nhoff + 4)) {
2348 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2349 if (ports.v16[1] < ports.v16[0])
2350 swap(ports.v16[0], ports.v16[1]);
2354 /* get a consistent hash (same value on both flow directions) */
2358 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2365 EXPORT_SYMBOL(__skb_get_rxhash);
2369 /* One global table that all flow-based protocols share. */
2370 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2371 EXPORT_SYMBOL(rps_sock_flow_table);
2374 * get_rps_cpu is called from netif_receive_skb and returns the target
2375 * CPU from the RPS map of the receiving queue for a given skb.
2376 * rcu_read_lock must be held on entry.
2378 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2379 struct rps_dev_flow **rflowp)
2381 struct netdev_rx_queue *rxqueue;
2382 struct rps_map *map = NULL;
2383 struct rps_dev_flow_table *flow_table;
2384 struct rps_sock_flow_table *sock_flow_table;
2388 if (skb_rx_queue_recorded(skb)) {
2389 u16 index = skb_get_rx_queue(skb);
2390 if (unlikely(index >= dev->real_num_rx_queues)) {
2391 WARN_ONCE(dev->real_num_rx_queues > 1,
2392 "%s received packet on queue %u, but number "
2393 "of RX queues is %u\n",
2394 dev->name, index, dev->real_num_rx_queues);
2397 rxqueue = dev->_rx + index;
2401 if (rxqueue->rps_map) {
2402 map = rcu_dereference(rxqueue->rps_map);
2403 if (map && map->len == 1) {
2404 tcpu = map->cpus[0];
2405 if (cpu_online(tcpu))
2409 } else if (!rxqueue->rps_flow_table) {
2413 skb_reset_network_header(skb);
2414 if (!skb_get_rxhash(skb))
2417 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2418 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2419 if (flow_table && sock_flow_table) {
2421 struct rps_dev_flow *rflow;
2423 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2426 next_cpu = sock_flow_table->ents[skb->rxhash &
2427 sock_flow_table->mask];
2430 * If the desired CPU (where last recvmsg was done) is
2431 * different from current CPU (one in the rx-queue flow
2432 * table entry), switch if one of the following holds:
2433 * - Current CPU is unset (equal to RPS_NO_CPU).
2434 * - Current CPU is offline.
2435 * - The current CPU's queue tail has advanced beyond the
2436 * last packet that was enqueued using this table entry.
2437 * This guarantees that all previous packets for the flow
2438 * have been dequeued, thus preserving in order delivery.
2440 if (unlikely(tcpu != next_cpu) &&
2441 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2442 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2443 rflow->last_qtail)) >= 0)) {
2444 tcpu = rflow->cpu = next_cpu;
2445 if (tcpu != RPS_NO_CPU)
2446 rflow->last_qtail = per_cpu(softnet_data,
2447 tcpu).input_queue_head;
2449 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2457 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2459 if (cpu_online(tcpu)) {
2469 /* Called from hardirq (IPI) context */
2470 static void rps_trigger_softirq(void *data)
2472 struct softnet_data *sd = data;
2474 ____napi_schedule(sd, &sd->backlog);
2478 #endif /* CONFIG_RPS */
2481 * Check if this softnet_data structure is another cpu one
2482 * If yes, queue it to our IPI list and return 1
2485 static int rps_ipi_queued(struct softnet_data *sd)
2488 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2491 sd->rps_ipi_next = mysd->rps_ipi_list;
2492 mysd->rps_ipi_list = sd;
2494 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2497 #endif /* CONFIG_RPS */
2502 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2503 * queue (may be a remote CPU queue).
2505 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2506 unsigned int *qtail)
2508 struct softnet_data *sd;
2509 unsigned long flags;
2511 sd = &per_cpu(softnet_data, cpu);
2513 local_irq_save(flags);
2516 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2517 if (skb_queue_len(&sd->input_pkt_queue)) {
2519 __skb_queue_tail(&sd->input_pkt_queue, skb);
2520 input_queue_tail_incr_save(sd, qtail);
2522 local_irq_restore(flags);
2523 return NET_RX_SUCCESS;
2526 /* Schedule NAPI for backlog device
2527 * We can use non atomic operation since we own the queue lock
2529 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2530 if (!rps_ipi_queued(sd))
2531 ____napi_schedule(sd, &sd->backlog);
2539 local_irq_restore(flags);
2546 * netif_rx - post buffer to the network code
2547 * @skb: buffer to post
2549 * This function receives a packet from a device driver and queues it for
2550 * the upper (protocol) levels to process. It always succeeds. The buffer
2551 * may be dropped during processing for congestion control or by the
2555 * NET_RX_SUCCESS (no congestion)
2556 * NET_RX_DROP (packet was dropped)
2560 int netif_rx(struct sk_buff *skb)
2564 /* if netpoll wants it, pretend we never saw it */
2565 if (netpoll_rx(skb))
2568 if (netdev_tstamp_prequeue)
2569 net_timestamp_check(skb);
2573 struct rps_dev_flow voidflow, *rflow = &voidflow;
2579 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2581 cpu = smp_processor_id();
2583 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2591 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2597 EXPORT_SYMBOL(netif_rx);
2599 int netif_rx_ni(struct sk_buff *skb)
2604 err = netif_rx(skb);
2605 if (local_softirq_pending())
2611 EXPORT_SYMBOL(netif_rx_ni);
2613 static void net_tx_action(struct softirq_action *h)
2615 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2617 if (sd->completion_queue) {
2618 struct sk_buff *clist;
2620 local_irq_disable();
2621 clist = sd->completion_queue;
2622 sd->completion_queue = NULL;
2626 struct sk_buff *skb = clist;
2627 clist = clist->next;
2629 WARN_ON(atomic_read(&skb->users));
2634 if (sd->output_queue) {
2637 local_irq_disable();
2638 head = sd->output_queue;
2639 sd->output_queue = NULL;
2640 sd->output_queue_tailp = &sd->output_queue;
2644 struct Qdisc *q = head;
2645 spinlock_t *root_lock;
2647 head = head->next_sched;
2649 root_lock = qdisc_lock(q);
2650 if (spin_trylock(root_lock)) {
2651 smp_mb__before_clear_bit();
2652 clear_bit(__QDISC_STATE_SCHED,
2655 spin_unlock(root_lock);
2657 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2659 __netif_reschedule(q);
2661 smp_mb__before_clear_bit();
2662 clear_bit(__QDISC_STATE_SCHED,
2670 static inline int deliver_skb(struct sk_buff *skb,
2671 struct packet_type *pt_prev,
2672 struct net_device *orig_dev)
2674 atomic_inc(&skb->users);
2675 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2678 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2679 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2680 /* This hook is defined here for ATM LANE */
2681 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2682 unsigned char *addr) __read_mostly;
2683 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2686 #ifdef CONFIG_NET_CLS_ACT
2687 /* TODO: Maybe we should just force sch_ingress to be compiled in
2688 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2689 * a compare and 2 stores extra right now if we dont have it on
2690 * but have CONFIG_NET_CLS_ACT
2691 * NOTE: This doesnt stop any functionality; if you dont have
2692 * the ingress scheduler, you just cant add policies on ingress.
2695 static int ing_filter(struct sk_buff *skb)
2697 struct net_device *dev = skb->dev;
2698 u32 ttl = G_TC_RTTL(skb->tc_verd);
2699 struct netdev_queue *rxq;
2700 int result = TC_ACT_OK;
2703 if (unlikely(MAX_RED_LOOP < ttl++)) {
2704 if (net_ratelimit())
2705 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2706 skb->skb_iif, dev->ifindex);
2710 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2711 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2713 rxq = &dev->rx_queue;
2716 if (q != &noop_qdisc) {
2717 spin_lock(qdisc_lock(q));
2718 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2719 result = qdisc_enqueue_root(skb, q);
2720 spin_unlock(qdisc_lock(q));
2726 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2727 struct packet_type **pt_prev,
2728 int *ret, struct net_device *orig_dev)
2730 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2734 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2738 switch (ing_filter(skb)) {
2752 * netif_nit_deliver - deliver received packets to network taps
2755 * This function is used to deliver incoming packets to network
2756 * taps. It should be used when the normal netif_receive_skb path
2757 * is bypassed, for example because of VLAN acceleration.
2759 void netif_nit_deliver(struct sk_buff *skb)
2761 struct packet_type *ptype;
2763 if (list_empty(&ptype_all))
2766 skb_reset_network_header(skb);
2767 skb_reset_transport_header(skb);
2768 skb->mac_len = skb->network_header - skb->mac_header;
2771 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2772 if (!ptype->dev || ptype->dev == skb->dev)
2773 deliver_skb(skb, ptype, skb->dev);
2779 * netdev_rx_handler_register - register receive handler
2780 * @dev: device to register a handler for
2781 * @rx_handler: receive handler to register
2782 * @rx_handler_data: data pointer that is used by rx handler
2784 * Register a receive hander for a device. This handler will then be
2785 * called from __netif_receive_skb. A negative errno code is returned
2788 * The caller must hold the rtnl_mutex.
2790 int netdev_rx_handler_register(struct net_device *dev,
2791 rx_handler_func_t *rx_handler,
2792 void *rx_handler_data)
2796 if (dev->rx_handler)
2799 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2800 rcu_assign_pointer(dev->rx_handler, rx_handler);
2804 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2807 * netdev_rx_handler_unregister - unregister receive handler
2808 * @dev: device to unregister a handler from
2810 * Unregister a receive hander from a device.
2812 * The caller must hold the rtnl_mutex.
2814 void netdev_rx_handler_unregister(struct net_device *dev)
2818 rcu_assign_pointer(dev->rx_handler, NULL);
2819 rcu_assign_pointer(dev->rx_handler_data, NULL);
2821 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2823 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2824 struct net_device *master)
2826 if (skb->pkt_type == PACKET_HOST) {
2827 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2829 memcpy(dest, master->dev_addr, ETH_ALEN);
2833 /* On bonding slaves other than the currently active slave, suppress
2834 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2835 * ARP on active-backup slaves with arp_validate enabled.
2837 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2839 struct net_device *dev = skb->dev;
2841 if (master->priv_flags & IFF_MASTER_ARPMON)
2842 dev->last_rx = jiffies;
2844 if ((master->priv_flags & IFF_MASTER_ALB) &&
2845 (master->priv_flags & IFF_BRIDGE_PORT)) {
2846 /* Do address unmangle. The local destination address
2847 * will be always the one master has. Provides the right
2848 * functionality in a bridge.
2850 skb_bond_set_mac_by_master(skb, master);
2853 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2854 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2855 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2858 if (master->priv_flags & IFF_MASTER_ALB) {
2859 if (skb->pkt_type != PACKET_BROADCAST &&
2860 skb->pkt_type != PACKET_MULTICAST)
2863 if (master->priv_flags & IFF_MASTER_8023AD &&
2864 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2871 EXPORT_SYMBOL(__skb_bond_should_drop);
2873 static int __netif_receive_skb(struct sk_buff *skb)
2875 struct packet_type *ptype, *pt_prev;
2876 rx_handler_func_t *rx_handler;
2877 struct net_device *orig_dev;
2878 struct net_device *master;
2879 struct net_device *null_or_orig;
2880 struct net_device *orig_or_bond;
2881 int ret = NET_RX_DROP;
2884 if (!netdev_tstamp_prequeue)
2885 net_timestamp_check(skb);
2887 if (vlan_tx_tag_present(skb))
2888 vlan_hwaccel_do_receive(skb);
2890 /* if we've gotten here through NAPI, check netpoll */
2891 if (netpoll_receive_skb(skb))
2895 skb->skb_iif = skb->dev->ifindex;
2898 * bonding note: skbs received on inactive slaves should only
2899 * be delivered to pkt handlers that are exact matches. Also
2900 * the deliver_no_wcard flag will be set. If packet handlers
2901 * are sensitive to duplicate packets these skbs will need to
2902 * be dropped at the handler. The vlan accel path may have
2903 * already set the deliver_no_wcard flag.
2905 null_or_orig = NULL;
2906 orig_dev = skb->dev;
2907 master = ACCESS_ONCE(orig_dev->master);
2908 if (skb->deliver_no_wcard)
2909 null_or_orig = orig_dev;
2911 if (skb_bond_should_drop(skb, master)) {
2912 skb->deliver_no_wcard = 1;
2913 null_or_orig = orig_dev; /* deliver only exact match */
2918 __this_cpu_inc(softnet_data.processed);
2919 skb_reset_network_header(skb);
2920 skb_reset_transport_header(skb);
2921 skb->mac_len = skb->network_header - skb->mac_header;
2927 #ifdef CONFIG_NET_CLS_ACT
2928 if (skb->tc_verd & TC_NCLS) {
2929 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2934 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2935 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2936 ptype->dev == orig_dev) {
2938 ret = deliver_skb(skb, pt_prev, orig_dev);
2943 #ifdef CONFIG_NET_CLS_ACT
2944 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2950 /* Handle special case of bridge or macvlan */
2951 rx_handler = rcu_dereference(skb->dev->rx_handler);
2954 ret = deliver_skb(skb, pt_prev, orig_dev);
2957 skb = rx_handler(skb);
2963 * Make sure frames received on VLAN interfaces stacked on
2964 * bonding interfaces still make their way to any base bonding
2965 * device that may have registered for a specific ptype. The
2966 * handler may have to adjust skb->dev and orig_dev.
2968 orig_or_bond = orig_dev;
2969 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2970 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2971 orig_or_bond = vlan_dev_real_dev(skb->dev);
2974 type = skb->protocol;
2975 list_for_each_entry_rcu(ptype,
2976 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2977 if (ptype->type == type && (ptype->dev == null_or_orig ||
2978 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2979 ptype->dev == orig_or_bond)) {
2981 ret = deliver_skb(skb, pt_prev, orig_dev);
2987 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2990 /* Jamal, now you will not able to escape explaining
2991 * me how you were going to use this. :-)
3002 * netif_receive_skb - process receive buffer from network
3003 * @skb: buffer to process
3005 * netif_receive_skb() is the main receive data processing function.
3006 * It always succeeds. The buffer may be dropped during processing
3007 * for congestion control or by the protocol layers.
3009 * This function may only be called from softirq context and interrupts
3010 * should be enabled.
3012 * Return values (usually ignored):
3013 * NET_RX_SUCCESS: no congestion
3014 * NET_RX_DROP: packet was dropped
3016 int netif_receive_skb(struct sk_buff *skb)
3018 if (netdev_tstamp_prequeue)
3019 net_timestamp_check(skb);
3021 if (skb_defer_rx_timestamp(skb))
3022 return NET_RX_SUCCESS;
3026 struct rps_dev_flow voidflow, *rflow = &voidflow;
3031 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3034 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3038 ret = __netif_receive_skb(skb);
3044 return __netif_receive_skb(skb);
3047 EXPORT_SYMBOL(netif_receive_skb);
3049 /* Network device is going away, flush any packets still pending
3050 * Called with irqs disabled.
3052 static void flush_backlog(void *arg)
3054 struct net_device *dev = arg;
3055 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3056 struct sk_buff *skb, *tmp;
3059 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3060 if (skb->dev == dev) {
3061 __skb_unlink(skb, &sd->input_pkt_queue);
3063 input_queue_head_incr(sd);
3068 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3069 if (skb->dev == dev) {
3070 __skb_unlink(skb, &sd->process_queue);
3072 input_queue_head_incr(sd);
3077 static int napi_gro_complete(struct sk_buff *skb)
3079 struct packet_type *ptype;
3080 __be16 type = skb->protocol;
3081 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3084 if (NAPI_GRO_CB(skb)->count == 1) {
3085 skb_shinfo(skb)->gso_size = 0;
3090 list_for_each_entry_rcu(ptype, head, list) {
3091 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3094 err = ptype->gro_complete(skb);
3100 WARN_ON(&ptype->list == head);
3102 return NET_RX_SUCCESS;
3106 return netif_receive_skb(skb);
3109 inline void napi_gro_flush(struct napi_struct *napi)
3111 struct sk_buff *skb, *next;
3113 for (skb = napi->gro_list; skb; skb = next) {
3116 napi_gro_complete(skb);
3119 napi->gro_count = 0;
3120 napi->gro_list = NULL;
3122 EXPORT_SYMBOL(napi_gro_flush);
3124 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3126 struct sk_buff **pp = NULL;
3127 struct packet_type *ptype;
3128 __be16 type = skb->protocol;
3129 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3132 enum gro_result ret;
3134 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3137 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3141 list_for_each_entry_rcu(ptype, head, list) {
3142 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3145 skb_set_network_header(skb, skb_gro_offset(skb));
3146 mac_len = skb->network_header - skb->mac_header;
3147 skb->mac_len = mac_len;
3148 NAPI_GRO_CB(skb)->same_flow = 0;
3149 NAPI_GRO_CB(skb)->flush = 0;
3150 NAPI_GRO_CB(skb)->free = 0;
3152 pp = ptype->gro_receive(&napi->gro_list, skb);
3157 if (&ptype->list == head)
3160 same_flow = NAPI_GRO_CB(skb)->same_flow;
3161 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3164 struct sk_buff *nskb = *pp;
3168 napi_gro_complete(nskb);
3175 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3179 NAPI_GRO_CB(skb)->count = 1;
3180 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3181 skb->next = napi->gro_list;
3182 napi->gro_list = skb;
3186 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3187 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3189 BUG_ON(skb->end - skb->tail < grow);
3191 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3194 skb->data_len -= grow;
3196 skb_shinfo(skb)->frags[0].page_offset += grow;
3197 skb_shinfo(skb)->frags[0].size -= grow;
3199 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3200 put_page(skb_shinfo(skb)->frags[0].page);
3201 memmove(skb_shinfo(skb)->frags,
3202 skb_shinfo(skb)->frags + 1,
3203 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3214 EXPORT_SYMBOL(dev_gro_receive);
3216 static inline gro_result_t
3217 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3221 for (p = napi->gro_list; p; p = p->next) {
3222 unsigned long diffs;
3224 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3225 diffs |= compare_ether_header(skb_mac_header(p),
3226 skb_gro_mac_header(skb));
3227 NAPI_GRO_CB(p)->same_flow = !diffs;
3228 NAPI_GRO_CB(p)->flush = 0;
3231 return dev_gro_receive(napi, skb);
3234 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3238 if (netif_receive_skb(skb))
3243 case GRO_MERGED_FREE:
3254 EXPORT_SYMBOL(napi_skb_finish);
3256 void skb_gro_reset_offset(struct sk_buff *skb)
3258 NAPI_GRO_CB(skb)->data_offset = 0;
3259 NAPI_GRO_CB(skb)->frag0 = NULL;
3260 NAPI_GRO_CB(skb)->frag0_len = 0;
3262 if (skb->mac_header == skb->tail &&
3263 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3264 NAPI_GRO_CB(skb)->frag0 =
3265 page_address(skb_shinfo(skb)->frags[0].page) +
3266 skb_shinfo(skb)->frags[0].page_offset;
3267 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3270 EXPORT_SYMBOL(skb_gro_reset_offset);
3272 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3274 skb_gro_reset_offset(skb);
3276 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3278 EXPORT_SYMBOL(napi_gro_receive);
3280 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3282 __skb_pull(skb, skb_headlen(skb));
3283 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3287 EXPORT_SYMBOL(napi_reuse_skb);
3289 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3291 struct sk_buff *skb = napi->skb;
3294 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3300 EXPORT_SYMBOL(napi_get_frags);
3302 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3308 skb->protocol = eth_type_trans(skb, skb->dev);
3310 if (ret == GRO_HELD)
3311 skb_gro_pull(skb, -ETH_HLEN);
3312 else if (netif_receive_skb(skb))
3317 case GRO_MERGED_FREE:
3318 napi_reuse_skb(napi, skb);
3327 EXPORT_SYMBOL(napi_frags_finish);
3329 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3331 struct sk_buff *skb = napi->skb;
3338 skb_reset_mac_header(skb);
3339 skb_gro_reset_offset(skb);
3341 off = skb_gro_offset(skb);
3342 hlen = off + sizeof(*eth);
3343 eth = skb_gro_header_fast(skb, off);
3344 if (skb_gro_header_hard(skb, hlen)) {
3345 eth = skb_gro_header_slow(skb, hlen, off);
3346 if (unlikely(!eth)) {
3347 napi_reuse_skb(napi, skb);
3353 skb_gro_pull(skb, sizeof(*eth));
3356 * This works because the only protocols we care about don't require
3357 * special handling. We'll fix it up properly at the end.
3359 skb->protocol = eth->h_proto;
3364 EXPORT_SYMBOL(napi_frags_skb);
3366 gro_result_t napi_gro_frags(struct napi_struct *napi)
3368 struct sk_buff *skb = napi_frags_skb(napi);
3373 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3375 EXPORT_SYMBOL(napi_gro_frags);
3378 * net_rps_action sends any pending IPI's for rps.
3379 * Note: called with local irq disabled, but exits with local irq enabled.
3381 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3384 struct softnet_data *remsd = sd->rps_ipi_list;
3387 sd->rps_ipi_list = NULL;
3391 /* Send pending IPI's to kick RPS processing on remote cpus. */
3393 struct softnet_data *next = remsd->rps_ipi_next;
3395 if (cpu_online(remsd->cpu))
3396 __smp_call_function_single(remsd->cpu,
3405 static int process_backlog(struct napi_struct *napi, int quota)
3408 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3411 /* Check if we have pending ipi, its better to send them now,
3412 * not waiting net_rx_action() end.
3414 if (sd->rps_ipi_list) {
3415 local_irq_disable();
3416 net_rps_action_and_irq_enable(sd);
3419 napi->weight = weight_p;
3420 local_irq_disable();
3421 while (work < quota) {
3422 struct sk_buff *skb;
3425 while ((skb = __skb_dequeue(&sd->process_queue))) {
3427 __netif_receive_skb(skb);
3428 local_irq_disable();
3429 input_queue_head_incr(sd);
3430 if (++work >= quota) {
3437 qlen = skb_queue_len(&sd->input_pkt_queue);
3439 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3440 &sd->process_queue);
3442 if (qlen < quota - work) {
3444 * Inline a custom version of __napi_complete().
3445 * only current cpu owns and manipulates this napi,
3446 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3447 * we can use a plain write instead of clear_bit(),
3448 * and we dont need an smp_mb() memory barrier.
3450 list_del(&napi->poll_list);
3453 quota = work + qlen;
3463 * __napi_schedule - schedule for receive
3464 * @n: entry to schedule
3466 * The entry's receive function will be scheduled to run
3468 void __napi_schedule(struct napi_struct *n)
3470 unsigned long flags;
3472 local_irq_save(flags);
3473 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3474 local_irq_restore(flags);
3476 EXPORT_SYMBOL(__napi_schedule);
3478 void __napi_complete(struct napi_struct *n)
3480 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3481 BUG_ON(n->gro_list);
3483 list_del(&n->poll_list);
3484 smp_mb__before_clear_bit();
3485 clear_bit(NAPI_STATE_SCHED, &n->state);
3487 EXPORT_SYMBOL(__napi_complete);
3489 void napi_complete(struct napi_struct *n)
3491 unsigned long flags;
3494 * don't let napi dequeue from the cpu poll list
3495 * just in case its running on a different cpu
3497 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3501 local_irq_save(flags);
3503 local_irq_restore(flags);
3505 EXPORT_SYMBOL(napi_complete);
3507 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3508 int (*poll)(struct napi_struct *, int), int weight)
3510 INIT_LIST_HEAD(&napi->poll_list);
3511 napi->gro_count = 0;
3512 napi->gro_list = NULL;
3515 napi->weight = weight;
3516 list_add(&napi->dev_list, &dev->napi_list);
3518 #ifdef CONFIG_NETPOLL
3519 spin_lock_init(&napi->poll_lock);
3520 napi->poll_owner = -1;
3522 set_bit(NAPI_STATE_SCHED, &napi->state);
3524 EXPORT_SYMBOL(netif_napi_add);
3526 void netif_napi_del(struct napi_struct *napi)
3528 struct sk_buff *skb, *next;
3530 list_del_init(&napi->dev_list);
3531 napi_free_frags(napi);
3533 for (skb = napi->gro_list; skb; skb = next) {
3539 napi->gro_list = NULL;
3540 napi->gro_count = 0;
3542 EXPORT_SYMBOL(netif_napi_del);
3544 static void net_rx_action(struct softirq_action *h)
3546 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3547 unsigned long time_limit = jiffies + 2;
3548 int budget = netdev_budget;
3551 local_irq_disable();
3553 while (!list_empty(&sd->poll_list)) {
3554 struct napi_struct *n;
3557 /* If softirq window is exhuasted then punt.
3558 * Allow this to run for 2 jiffies since which will allow
3559 * an average latency of 1.5/HZ.
3561 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3566 /* Even though interrupts have been re-enabled, this
3567 * access is safe because interrupts can only add new
3568 * entries to the tail of this list, and only ->poll()
3569 * calls can remove this head entry from the list.
3571 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3573 have = netpoll_poll_lock(n);
3577 /* This NAPI_STATE_SCHED test is for avoiding a race
3578 * with netpoll's poll_napi(). Only the entity which
3579 * obtains the lock and sees NAPI_STATE_SCHED set will
3580 * actually make the ->poll() call. Therefore we avoid
3581 * accidently calling ->poll() when NAPI is not scheduled.
3584 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3585 work = n->poll(n, weight);
3589 WARN_ON_ONCE(work > weight);
3593 local_irq_disable();
3595 /* Drivers must not modify the NAPI state if they
3596 * consume the entire weight. In such cases this code
3597 * still "owns" the NAPI instance and therefore can
3598 * move the instance around on the list at-will.
3600 if (unlikely(work == weight)) {
3601 if (unlikely(napi_disable_pending(n))) {
3604 local_irq_disable();
3606 list_move_tail(&n->poll_list, &sd->poll_list);
3609 netpoll_poll_unlock(have);
3612 net_rps_action_and_irq_enable(sd);
3614 #ifdef CONFIG_NET_DMA
3616 * There may not be any more sk_buffs coming right now, so push
3617 * any pending DMA copies to hardware
3619 dma_issue_pending_all();
3626 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3630 static gifconf_func_t *gifconf_list[NPROTO];
3633 * register_gifconf - register a SIOCGIF handler
3634 * @family: Address family
3635 * @gifconf: Function handler
3637 * Register protocol dependent address dumping routines. The handler
3638 * that is passed must not be freed or reused until it has been replaced
3639 * by another handler.
3641 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3643 if (family >= NPROTO)
3645 gifconf_list[family] = gifconf;
3648 EXPORT_SYMBOL(register_gifconf);
3652 * Map an interface index to its name (SIOCGIFNAME)
3656 * We need this ioctl for efficient implementation of the
3657 * if_indextoname() function required by the IPv6 API. Without
3658 * it, we would have to search all the interfaces to find a
3662 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3664 struct net_device *dev;
3668 * Fetch the caller's info block.
3671 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3675 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3681 strcpy(ifr.ifr_name, dev->name);
3684 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3690 * Perform a SIOCGIFCONF call. This structure will change
3691 * size eventually, and there is nothing I can do about it.
3692 * Thus we will need a 'compatibility mode'.
3695 static int dev_ifconf(struct net *net, char __user *arg)
3698 struct net_device *dev;
3705 * Fetch the caller's info block.
3708 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3715 * Loop over the interfaces, and write an info block for each.
3719 for_each_netdev(net, dev) {
3720 for (i = 0; i < NPROTO; i++) {
3721 if (gifconf_list[i]) {
3724 done = gifconf_list[i](dev, NULL, 0);
3726 done = gifconf_list[i](dev, pos + total,
3736 * All done. Write the updated control block back to the caller.
3738 ifc.ifc_len = total;
3741 * Both BSD and Solaris return 0 here, so we do too.
3743 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3746 #ifdef CONFIG_PROC_FS
3748 * This is invoked by the /proc filesystem handler to display a device
3751 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3754 struct net *net = seq_file_net(seq);
3756 struct net_device *dev;
3760 return SEQ_START_TOKEN;
3763 for_each_netdev_rcu(net, dev)
3770 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3772 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3773 first_net_device(seq_file_net(seq)) :
3774 next_net_device((struct net_device *)v);
3777 return rcu_dereference(dev);
3780 void dev_seq_stop(struct seq_file *seq, void *v)
3786 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3788 struct rtnl_link_stats64 temp;
3789 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3791 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3792 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3793 dev->name, stats->rx_bytes, stats->rx_packets,
3795 stats->rx_dropped + stats->rx_missed_errors,
3796 stats->rx_fifo_errors,
3797 stats->rx_length_errors + stats->rx_over_errors +
3798 stats->rx_crc_errors + stats->rx_frame_errors,
3799 stats->rx_compressed, stats->multicast,
3800 stats->tx_bytes, stats->tx_packets,
3801 stats->tx_errors, stats->tx_dropped,
3802 stats->tx_fifo_errors, stats->collisions,
3803 stats->tx_carrier_errors +
3804 stats->tx_aborted_errors +
3805 stats->tx_window_errors +
3806 stats->tx_heartbeat_errors,
3807 stats->tx_compressed);
3811 * Called from the PROCfs module. This now uses the new arbitrary sized
3812 * /proc/net interface to create /proc/net/dev
3814 static int dev_seq_show(struct seq_file *seq, void *v)
3816 if (v == SEQ_START_TOKEN)
3817 seq_puts(seq, "Inter-| Receive "
3819 " face |bytes packets errs drop fifo frame "
3820 "compressed multicast|bytes packets errs "
3821 "drop fifo colls carrier compressed\n");
3823 dev_seq_printf_stats(seq, v);
3827 static struct softnet_data *softnet_get_online(loff_t *pos)
3829 struct softnet_data *sd = NULL;
3831 while (*pos < nr_cpu_ids)
3832 if (cpu_online(*pos)) {
3833 sd = &per_cpu(softnet_data, *pos);
3840 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3842 return softnet_get_online(pos);
3845 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3848 return softnet_get_online(pos);
3851 static void softnet_seq_stop(struct seq_file *seq, void *v)
3855 static int softnet_seq_show(struct seq_file *seq, void *v)
3857 struct softnet_data *sd = v;
3859 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3860 sd->processed, sd->dropped, sd->time_squeeze, 0,
3861 0, 0, 0, 0, /* was fastroute */
3862 sd->cpu_collision, sd->received_rps);
3866 static const struct seq_operations dev_seq_ops = {
3867 .start = dev_seq_start,
3868 .next = dev_seq_next,
3869 .stop = dev_seq_stop,
3870 .show = dev_seq_show,
3873 static int dev_seq_open(struct inode *inode, struct file *file)
3875 return seq_open_net(inode, file, &dev_seq_ops,
3876 sizeof(struct seq_net_private));
3879 static const struct file_operations dev_seq_fops = {
3880 .owner = THIS_MODULE,
3881 .open = dev_seq_open,
3883 .llseek = seq_lseek,
3884 .release = seq_release_net,
3887 static const struct seq_operations softnet_seq_ops = {
3888 .start = softnet_seq_start,
3889 .next = softnet_seq_next,
3890 .stop = softnet_seq_stop,
3891 .show = softnet_seq_show,
3894 static int softnet_seq_open(struct inode *inode, struct file *file)
3896 return seq_open(file, &softnet_seq_ops);
3899 static const struct file_operations softnet_seq_fops = {
3900 .owner = THIS_MODULE,
3901 .open = softnet_seq_open,
3903 .llseek = seq_lseek,
3904 .release = seq_release,
3907 static void *ptype_get_idx(loff_t pos)
3909 struct packet_type *pt = NULL;
3913 list_for_each_entry_rcu(pt, &ptype_all, list) {
3919 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3920 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3929 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3933 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3936 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3938 struct packet_type *pt;
3939 struct list_head *nxt;
3943 if (v == SEQ_START_TOKEN)
3944 return ptype_get_idx(0);
3947 nxt = pt->list.next;
3948 if (pt->type == htons(ETH_P_ALL)) {
3949 if (nxt != &ptype_all)
3952 nxt = ptype_base[0].next;
3954 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3956 while (nxt == &ptype_base[hash]) {
3957 if (++hash >= PTYPE_HASH_SIZE)
3959 nxt = ptype_base[hash].next;
3962 return list_entry(nxt, struct packet_type, list);
3965 static void ptype_seq_stop(struct seq_file *seq, void *v)
3971 static int ptype_seq_show(struct seq_file *seq, void *v)
3973 struct packet_type *pt = v;
3975 if (v == SEQ_START_TOKEN)
3976 seq_puts(seq, "Type Device Function\n");
3977 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3978 if (pt->type == htons(ETH_P_ALL))
3979 seq_puts(seq, "ALL ");
3981 seq_printf(seq, "%04x", ntohs(pt->type));
3983 seq_printf(seq, " %-8s %pF\n",
3984 pt->dev ? pt->dev->name : "", pt->func);
3990 static const struct seq_operations ptype_seq_ops = {
3991 .start = ptype_seq_start,
3992 .next = ptype_seq_next,
3993 .stop = ptype_seq_stop,
3994 .show = ptype_seq_show,
3997 static int ptype_seq_open(struct inode *inode, struct file *file)
3999 return seq_open_net(inode, file, &ptype_seq_ops,
4000 sizeof(struct seq_net_private));
4003 static const struct file_operations ptype_seq_fops = {
4004 .owner = THIS_MODULE,
4005 .open = ptype_seq_open,
4007 .llseek = seq_lseek,
4008 .release = seq_release_net,
4012 static int __net_init dev_proc_net_init(struct net *net)
4016 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4018 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4020 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4023 if (wext_proc_init(net))
4029 proc_net_remove(net, "ptype");
4031 proc_net_remove(net, "softnet_stat");
4033 proc_net_remove(net, "dev");
4037 static void __net_exit dev_proc_net_exit(struct net *net)
4039 wext_proc_exit(net);
4041 proc_net_remove(net, "ptype");
4042 proc_net_remove(net, "softnet_stat");
4043 proc_net_remove(net, "dev");
4046 static struct pernet_operations __net_initdata dev_proc_ops = {
4047 .init = dev_proc_net_init,
4048 .exit = dev_proc_net_exit,
4051 static int __init dev_proc_init(void)
4053 return register_pernet_subsys(&dev_proc_ops);
4056 #define dev_proc_init() 0
4057 #endif /* CONFIG_PROC_FS */
4061 * netdev_set_master - set up master/slave pair
4062 * @slave: slave device
4063 * @master: new master device
4065 * Changes the master device of the slave. Pass %NULL to break the
4066 * bonding. The caller must hold the RTNL semaphore. On a failure
4067 * a negative errno code is returned. On success the reference counts
4068 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4069 * function returns zero.
4071 int netdev_set_master(struct net_device *slave, struct net_device *master)
4073 struct net_device *old = slave->master;
4083 slave->master = master;
4090 slave->flags |= IFF_SLAVE;
4092 slave->flags &= ~IFF_SLAVE;
4094 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4097 EXPORT_SYMBOL(netdev_set_master);
4099 static void dev_change_rx_flags(struct net_device *dev, int flags)
4101 const struct net_device_ops *ops = dev->netdev_ops;
4103 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4104 ops->ndo_change_rx_flags(dev, flags);
4107 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4109 unsigned short old_flags = dev->flags;
4115 dev->flags |= IFF_PROMISC;
4116 dev->promiscuity += inc;
4117 if (dev->promiscuity == 0) {
4120 * If inc causes overflow, untouch promisc and return error.
4123 dev->flags &= ~IFF_PROMISC;
4125 dev->promiscuity -= inc;
4126 printk(KERN_WARNING "%s: promiscuity touches roof, "
4127 "set promiscuity failed, promiscuity feature "
4128 "of device might be broken.\n", dev->name);
4132 if (dev->flags != old_flags) {
4133 printk(KERN_INFO "device %s %s promiscuous mode\n",
4134 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4136 if (audit_enabled) {
4137 current_uid_gid(&uid, &gid);
4138 audit_log(current->audit_context, GFP_ATOMIC,
4139 AUDIT_ANOM_PROMISCUOUS,
4140 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4141 dev->name, (dev->flags & IFF_PROMISC),
4142 (old_flags & IFF_PROMISC),
4143 audit_get_loginuid(current),
4145 audit_get_sessionid(current));
4148 dev_change_rx_flags(dev, IFF_PROMISC);
4154 * dev_set_promiscuity - update promiscuity count on a device
4158 * Add or remove promiscuity from a device. While the count in the device
4159 * remains above zero the interface remains promiscuous. Once it hits zero
4160 * the device reverts back to normal filtering operation. A negative inc
4161 * value is used to drop promiscuity on the device.
4162 * Return 0 if successful or a negative errno code on error.
4164 int dev_set_promiscuity(struct net_device *dev, int inc)
4166 unsigned short old_flags = dev->flags;
4169 err = __dev_set_promiscuity(dev, inc);
4172 if (dev->flags != old_flags)
4173 dev_set_rx_mode(dev);
4176 EXPORT_SYMBOL(dev_set_promiscuity);
4179 * dev_set_allmulti - update allmulti count on a device
4183 * Add or remove reception of all multicast frames to a device. While the
4184 * count in the device remains above zero the interface remains listening
4185 * to all interfaces. Once it hits zero the device reverts back to normal
4186 * filtering operation. A negative @inc value is used to drop the counter
4187 * when releasing a resource needing all multicasts.
4188 * Return 0 if successful or a negative errno code on error.
4191 int dev_set_allmulti(struct net_device *dev, int inc)
4193 unsigned short old_flags = dev->flags;
4197 dev->flags |= IFF_ALLMULTI;
4198 dev->allmulti += inc;
4199 if (dev->allmulti == 0) {
4202 * If inc causes overflow, untouch allmulti and return error.
4205 dev->flags &= ~IFF_ALLMULTI;
4207 dev->allmulti -= inc;
4208 printk(KERN_WARNING "%s: allmulti touches roof, "
4209 "set allmulti failed, allmulti feature of "
4210 "device might be broken.\n", dev->name);
4214 if (dev->flags ^ old_flags) {
4215 dev_change_rx_flags(dev, IFF_ALLMULTI);
4216 dev_set_rx_mode(dev);
4220 EXPORT_SYMBOL(dev_set_allmulti);
4223 * Upload unicast and multicast address lists to device and
4224 * configure RX filtering. When the device doesn't support unicast
4225 * filtering it is put in promiscuous mode while unicast addresses
4228 void __dev_set_rx_mode(struct net_device *dev)
4230 const struct net_device_ops *ops = dev->netdev_ops;
4232 /* dev_open will call this function so the list will stay sane. */
4233 if (!(dev->flags&IFF_UP))
4236 if (!netif_device_present(dev))
4239 if (ops->ndo_set_rx_mode)
4240 ops->ndo_set_rx_mode(dev);
4242 /* Unicast addresses changes may only happen under the rtnl,
4243 * therefore calling __dev_set_promiscuity here is safe.
4245 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4246 __dev_set_promiscuity(dev, 1);
4247 dev->uc_promisc = 1;
4248 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4249 __dev_set_promiscuity(dev, -1);
4250 dev->uc_promisc = 0;
4253 if (ops->ndo_set_multicast_list)
4254 ops->ndo_set_multicast_list(dev);
4258 void dev_set_rx_mode(struct net_device *dev)
4260 netif_addr_lock_bh(dev);
4261 __dev_set_rx_mode(dev);
4262 netif_addr_unlock_bh(dev);
4266 * dev_get_flags - get flags reported to userspace
4269 * Get the combination of flag bits exported through APIs to userspace.
4271 unsigned dev_get_flags(const struct net_device *dev)
4275 flags = (dev->flags & ~(IFF_PROMISC |
4280 (dev->gflags & (IFF_PROMISC |
4283 if (netif_running(dev)) {
4284 if (netif_oper_up(dev))
4285 flags |= IFF_RUNNING;
4286 if (netif_carrier_ok(dev))
4287 flags |= IFF_LOWER_UP;
4288 if (netif_dormant(dev))
4289 flags |= IFF_DORMANT;
4294 EXPORT_SYMBOL(dev_get_flags);
4296 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4298 int old_flags = dev->flags;
4304 * Set the flags on our device.
4307 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4308 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4310 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4314 * Load in the correct multicast list now the flags have changed.
4317 if ((old_flags ^ flags) & IFF_MULTICAST)
4318 dev_change_rx_flags(dev, IFF_MULTICAST);
4320 dev_set_rx_mode(dev);
4323 * Have we downed the interface. We handle IFF_UP ourselves
4324 * according to user attempts to set it, rather than blindly
4329 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4330 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4333 dev_set_rx_mode(dev);
4336 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4337 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4339 dev->gflags ^= IFF_PROMISC;
4340 dev_set_promiscuity(dev, inc);
4343 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4344 is important. Some (broken) drivers set IFF_PROMISC, when
4345 IFF_ALLMULTI is requested not asking us and not reporting.
4347 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4348 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4350 dev->gflags ^= IFF_ALLMULTI;
4351 dev_set_allmulti(dev, inc);
4357 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4359 unsigned int changes = dev->flags ^ old_flags;
4361 if (changes & IFF_UP) {
4362 if (dev->flags & IFF_UP)
4363 call_netdevice_notifiers(NETDEV_UP, dev);
4365 call_netdevice_notifiers(NETDEV_DOWN, dev);
4368 if (dev->flags & IFF_UP &&
4369 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4370 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4374 * dev_change_flags - change device settings
4376 * @flags: device state flags
4378 * Change settings on device based state flags. The flags are
4379 * in the userspace exported format.
4381 int dev_change_flags(struct net_device *dev, unsigned flags)
4384 int old_flags = dev->flags;
4386 ret = __dev_change_flags(dev, flags);
4390 changes = old_flags ^ dev->flags;
4392 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4394 __dev_notify_flags(dev, old_flags);
4397 EXPORT_SYMBOL(dev_change_flags);
4400 * dev_set_mtu - Change maximum transfer unit
4402 * @new_mtu: new transfer unit
4404 * Change the maximum transfer size of the network device.
4406 int dev_set_mtu(struct net_device *dev, int new_mtu)
4408 const struct net_device_ops *ops = dev->netdev_ops;
4411 if (new_mtu == dev->mtu)
4414 /* MTU must be positive. */
4418 if (!netif_device_present(dev))
4422 if (ops->ndo_change_mtu)
4423 err = ops->ndo_change_mtu(dev, new_mtu);
4427 if (!err && dev->flags & IFF_UP)
4428 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4431 EXPORT_SYMBOL(dev_set_mtu);
4434 * dev_set_mac_address - Change Media Access Control Address
4438 * Change the hardware (MAC) address of the device
4440 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4442 const struct net_device_ops *ops = dev->netdev_ops;
4445 if (!ops->ndo_set_mac_address)
4447 if (sa->sa_family != dev->type)
4449 if (!netif_device_present(dev))
4451 err = ops->ndo_set_mac_address(dev, sa);
4453 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4456 EXPORT_SYMBOL(dev_set_mac_address);
4459 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4461 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4464 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4470 case SIOCGIFFLAGS: /* Get interface flags */
4471 ifr->ifr_flags = (short) dev_get_flags(dev);
4474 case SIOCGIFMETRIC: /* Get the metric on the interface
4475 (currently unused) */
4476 ifr->ifr_metric = 0;
4479 case SIOCGIFMTU: /* Get the MTU of a device */
4480 ifr->ifr_mtu = dev->mtu;
4485 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4487 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4488 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4489 ifr->ifr_hwaddr.sa_family = dev->type;
4497 ifr->ifr_map.mem_start = dev->mem_start;
4498 ifr->ifr_map.mem_end = dev->mem_end;
4499 ifr->ifr_map.base_addr = dev->base_addr;
4500 ifr->ifr_map.irq = dev->irq;
4501 ifr->ifr_map.dma = dev->dma;
4502 ifr->ifr_map.port = dev->if_port;
4506 ifr->ifr_ifindex = dev->ifindex;
4510 ifr->ifr_qlen = dev->tx_queue_len;
4514 /* dev_ioctl() should ensure this case
4526 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4528 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4531 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4532 const struct net_device_ops *ops;
4537 ops = dev->netdev_ops;
4540 case SIOCSIFFLAGS: /* Set interface flags */
4541 return dev_change_flags(dev, ifr->ifr_flags);
4543 case SIOCSIFMETRIC: /* Set the metric on the interface
4544 (currently unused) */
4547 case SIOCSIFMTU: /* Set the MTU of a device */
4548 return dev_set_mtu(dev, ifr->ifr_mtu);
4551 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4553 case SIOCSIFHWBROADCAST:
4554 if (ifr->ifr_hwaddr.sa_family != dev->type)
4556 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4557 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4558 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4562 if (ops->ndo_set_config) {
4563 if (!netif_device_present(dev))
4565 return ops->ndo_set_config(dev, &ifr->ifr_map);
4570 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4571 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4573 if (!netif_device_present(dev))
4575 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4578 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4579 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4581 if (!netif_device_present(dev))
4583 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4586 if (ifr->ifr_qlen < 0)
4588 dev->tx_queue_len = ifr->ifr_qlen;
4592 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4593 return dev_change_name(dev, ifr->ifr_newname);
4596 * Unknown or private ioctl
4599 if ((cmd >= SIOCDEVPRIVATE &&
4600 cmd <= SIOCDEVPRIVATE + 15) ||
4601 cmd == SIOCBONDENSLAVE ||
4602 cmd == SIOCBONDRELEASE ||
4603 cmd == SIOCBONDSETHWADDR ||
4604 cmd == SIOCBONDSLAVEINFOQUERY ||
4605 cmd == SIOCBONDINFOQUERY ||
4606 cmd == SIOCBONDCHANGEACTIVE ||
4607 cmd == SIOCGMIIPHY ||
4608 cmd == SIOCGMIIREG ||
4609 cmd == SIOCSMIIREG ||
4610 cmd == SIOCBRADDIF ||
4611 cmd == SIOCBRDELIF ||
4612 cmd == SIOCSHWTSTAMP ||
4613 cmd == SIOCWANDEV) {
4615 if (ops->ndo_do_ioctl) {
4616 if (netif_device_present(dev))
4617 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4629 * This function handles all "interface"-type I/O control requests. The actual
4630 * 'doing' part of this is dev_ifsioc above.
4634 * dev_ioctl - network device ioctl
4635 * @net: the applicable net namespace
4636 * @cmd: command to issue
4637 * @arg: pointer to a struct ifreq in user space
4639 * Issue ioctl functions to devices. This is normally called by the
4640 * user space syscall interfaces but can sometimes be useful for
4641 * other purposes. The return value is the return from the syscall if
4642 * positive or a negative errno code on error.
4645 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4651 /* One special case: SIOCGIFCONF takes ifconf argument
4652 and requires shared lock, because it sleeps writing
4656 if (cmd == SIOCGIFCONF) {
4658 ret = dev_ifconf(net, (char __user *) arg);
4662 if (cmd == SIOCGIFNAME)
4663 return dev_ifname(net, (struct ifreq __user *)arg);
4665 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4668 ifr.ifr_name[IFNAMSIZ-1] = 0;
4670 colon = strchr(ifr.ifr_name, ':');
4675 * See which interface the caller is talking about.
4680 * These ioctl calls:
4681 * - can be done by all.
4682 * - atomic and do not require locking.
4693 dev_load(net, ifr.ifr_name);
4695 ret = dev_ifsioc_locked(net, &ifr, cmd);
4700 if (copy_to_user(arg, &ifr,
4701 sizeof(struct ifreq)))
4707 dev_load(net, ifr.ifr_name);
4709 ret = dev_ethtool(net, &ifr);
4714 if (copy_to_user(arg, &ifr,
4715 sizeof(struct ifreq)))
4721 * These ioctl calls:
4722 * - require superuser power.
4723 * - require strict serialization.
4729 if (!capable(CAP_NET_ADMIN))
4731 dev_load(net, ifr.ifr_name);
4733 ret = dev_ifsioc(net, &ifr, cmd);
4738 if (copy_to_user(arg, &ifr,
4739 sizeof(struct ifreq)))
4745 * These ioctl calls:
4746 * - require superuser power.
4747 * - require strict serialization.
4748 * - do not return a value
4758 case SIOCSIFHWBROADCAST:
4761 case SIOCBONDENSLAVE:
4762 case SIOCBONDRELEASE:
4763 case SIOCBONDSETHWADDR:
4764 case SIOCBONDCHANGEACTIVE:
4768 if (!capable(CAP_NET_ADMIN))
4771 case SIOCBONDSLAVEINFOQUERY:
4772 case SIOCBONDINFOQUERY:
4773 dev_load(net, ifr.ifr_name);
4775 ret = dev_ifsioc(net, &ifr, cmd);
4780 /* Get the per device memory space. We can add this but
4781 * currently do not support it */
4783 /* Set the per device memory buffer space.
4784 * Not applicable in our case */
4789 * Unknown or private ioctl.
4792 if (cmd == SIOCWANDEV ||
4793 (cmd >= SIOCDEVPRIVATE &&
4794 cmd <= SIOCDEVPRIVATE + 15)) {
4795 dev_load(net, ifr.ifr_name);
4797 ret = dev_ifsioc(net, &ifr, cmd);
4799 if (!ret && copy_to_user(arg, &ifr,
4800 sizeof(struct ifreq)))
4804 /* Take care of Wireless Extensions */
4805 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4806 return wext_handle_ioctl(net, &ifr, cmd, arg);
4813 * dev_new_index - allocate an ifindex
4814 * @net: the applicable net namespace
4816 * Returns a suitable unique value for a new device interface
4817 * number. The caller must hold the rtnl semaphore or the
4818 * dev_base_lock to be sure it remains unique.
4820 static int dev_new_index(struct net *net)
4826 if (!__dev_get_by_index(net, ifindex))
4831 /* Delayed registration/unregisteration */
4832 static LIST_HEAD(net_todo_list);
4834 static void net_set_todo(struct net_device *dev)
4836 list_add_tail(&dev->todo_list, &net_todo_list);
4839 static void rollback_registered_many(struct list_head *head)
4841 struct net_device *dev, *tmp;
4843 BUG_ON(dev_boot_phase);
4846 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4847 /* Some devices call without registering
4848 * for initialization unwind. Remove those
4849 * devices and proceed with the remaining.
4851 if (dev->reg_state == NETREG_UNINITIALIZED) {
4852 pr_debug("unregister_netdevice: device %s/%p never "
4853 "was registered\n", dev->name, dev);
4856 list_del(&dev->unreg_list);
4860 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4862 /* If device is running, close it first. */
4865 /* And unlink it from device chain. */
4866 unlist_netdevice(dev);
4868 dev->reg_state = NETREG_UNREGISTERING;
4873 list_for_each_entry(dev, head, unreg_list) {
4874 /* Shutdown queueing discipline. */
4878 /* Notify protocols, that we are about to destroy
4879 this device. They should clean all the things.
4881 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4883 if (!dev->rtnl_link_ops ||
4884 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4885 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4888 * Flush the unicast and multicast chains
4893 if (dev->netdev_ops->ndo_uninit)
4894 dev->netdev_ops->ndo_uninit(dev);
4896 /* Notifier chain MUST detach us from master device. */
4897 WARN_ON(dev->master);
4899 /* Remove entries from kobject tree */
4900 netdev_unregister_kobject(dev);
4903 /* Process any work delayed until the end of the batch */
4904 dev = list_first_entry(head, struct net_device, unreg_list);
4905 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4909 list_for_each_entry(dev, head, unreg_list)
4913 static void rollback_registered(struct net_device *dev)
4917 list_add(&dev->unreg_list, &single);
4918 rollback_registered_many(&single);
4921 static void __netdev_init_queue_locks_one(struct net_device *dev,
4922 struct netdev_queue *dev_queue,
4925 spin_lock_init(&dev_queue->_xmit_lock);
4926 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4927 dev_queue->xmit_lock_owner = -1;
4930 static void netdev_init_queue_locks(struct net_device *dev)
4932 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4933 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4936 unsigned long netdev_fix_features(unsigned long features, const char *name)
4938 /* Fix illegal SG+CSUM combinations. */
4939 if ((features & NETIF_F_SG) &&
4940 !(features & NETIF_F_ALL_CSUM)) {
4942 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4943 "checksum feature.\n", name);
4944 features &= ~NETIF_F_SG;
4947 /* TSO requires that SG is present as well. */
4948 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4950 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4951 "SG feature.\n", name);
4952 features &= ~NETIF_F_TSO;
4955 if (features & NETIF_F_UFO) {
4956 if (!(features & NETIF_F_GEN_CSUM)) {
4958 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4959 "since no NETIF_F_HW_CSUM feature.\n",
4961 features &= ~NETIF_F_UFO;
4964 if (!(features & NETIF_F_SG)) {
4966 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4967 "since no NETIF_F_SG feature.\n", name);
4968 features &= ~NETIF_F_UFO;
4974 EXPORT_SYMBOL(netdev_fix_features);
4977 * netif_stacked_transfer_operstate - transfer operstate
4978 * @rootdev: the root or lower level device to transfer state from
4979 * @dev: the device to transfer operstate to
4981 * Transfer operational state from root to device. This is normally
4982 * called when a stacking relationship exists between the root
4983 * device and the device(a leaf device).
4985 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4986 struct net_device *dev)
4988 if (rootdev->operstate == IF_OPER_DORMANT)
4989 netif_dormant_on(dev);
4991 netif_dormant_off(dev);
4993 if (netif_carrier_ok(rootdev)) {
4994 if (!netif_carrier_ok(dev))
4995 netif_carrier_on(dev);
4997 if (netif_carrier_ok(dev))
4998 netif_carrier_off(dev);
5001 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5003 static int netif_alloc_rx_queues(struct net_device *dev)
5006 unsigned int i, count = dev->num_rx_queues;
5009 struct netdev_rx_queue *rx;
5011 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5013 pr_err("netdev: Unable to allocate %u rx queues.\n",
5018 atomic_set(&rx->count, count);
5021 * Set a pointer to first element in the array which holds the
5024 for (i = 0; i < count; i++)
5032 * register_netdevice - register a network device
5033 * @dev: device to register
5035 * Take a completed network device structure and add it to the kernel
5036 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5037 * chain. 0 is returned on success. A negative errno code is returned
5038 * on a failure to set up the device, or if the name is a duplicate.
5040 * Callers must hold the rtnl semaphore. You may want
5041 * register_netdev() instead of this.
5044 * The locking appears insufficient to guarantee two parallel registers
5045 * will not get the same name.
5048 int register_netdevice(struct net_device *dev)
5051 struct net *net = dev_net(dev);
5053 BUG_ON(dev_boot_phase);
5058 /* When net_device's are persistent, this will be fatal. */
5059 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5062 spin_lock_init(&dev->addr_list_lock);
5063 netdev_set_addr_lockdep_class(dev);
5064 netdev_init_queue_locks(dev);
5068 ret = netif_alloc_rx_queues(dev);
5072 /* Init, if this function is available */
5073 if (dev->netdev_ops->ndo_init) {
5074 ret = dev->netdev_ops->ndo_init(dev);
5082 ret = dev_get_valid_name(dev, dev->name, 0);
5086 dev->ifindex = dev_new_index(net);
5087 if (dev->iflink == -1)
5088 dev->iflink = dev->ifindex;
5090 /* Fix illegal checksum combinations */
5091 if ((dev->features & NETIF_F_HW_CSUM) &&
5092 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5093 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5095 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5098 if ((dev->features & NETIF_F_NO_CSUM) &&
5099 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5100 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5102 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5105 dev->features = netdev_fix_features(dev->features, dev->name);
5107 /* Enable software GSO if SG is supported. */
5108 if (dev->features & NETIF_F_SG)
5109 dev->features |= NETIF_F_GSO;
5111 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5112 * vlan_dev_init() will do the dev->features check, so these features
5113 * are enabled only if supported by underlying device.
5115 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5117 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5118 ret = notifier_to_errno(ret);
5122 ret = netdev_register_kobject(dev);
5125 dev->reg_state = NETREG_REGISTERED;
5128 * Default initial state at registry is that the
5129 * device is present.
5132 set_bit(__LINK_STATE_PRESENT, &dev->state);
5134 dev_init_scheduler(dev);
5136 list_netdevice(dev);
5138 /* Notify protocols, that a new device appeared. */
5139 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5140 ret = notifier_to_errno(ret);
5142 rollback_registered(dev);
5143 dev->reg_state = NETREG_UNREGISTERED;
5146 * Prevent userspace races by waiting until the network
5147 * device is fully setup before sending notifications.
5149 if (!dev->rtnl_link_ops ||
5150 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5151 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5157 if (dev->netdev_ops->ndo_uninit)
5158 dev->netdev_ops->ndo_uninit(dev);
5161 EXPORT_SYMBOL(register_netdevice);
5164 * init_dummy_netdev - init a dummy network device for NAPI
5165 * @dev: device to init
5167 * This takes a network device structure and initialize the minimum
5168 * amount of fields so it can be used to schedule NAPI polls without
5169 * registering a full blown interface. This is to be used by drivers
5170 * that need to tie several hardware interfaces to a single NAPI
5171 * poll scheduler due to HW limitations.
5173 int init_dummy_netdev(struct net_device *dev)
5175 /* Clear everything. Note we don't initialize spinlocks
5176 * are they aren't supposed to be taken by any of the
5177 * NAPI code and this dummy netdev is supposed to be
5178 * only ever used for NAPI polls
5180 memset(dev, 0, sizeof(struct net_device));
5182 /* make sure we BUG if trying to hit standard
5183 * register/unregister code path
5185 dev->reg_state = NETREG_DUMMY;
5187 /* initialize the ref count */
5188 atomic_set(&dev->refcnt, 1);
5190 /* NAPI wants this */
5191 INIT_LIST_HEAD(&dev->napi_list);
5193 /* a dummy interface is started by default */
5194 set_bit(__LINK_STATE_PRESENT, &dev->state);
5195 set_bit(__LINK_STATE_START, &dev->state);
5199 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5203 * register_netdev - register a network device
5204 * @dev: device to register
5206 * Take a completed network device structure and add it to the kernel
5207 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5208 * chain. 0 is returned on success. A negative errno code is returned
5209 * on a failure to set up the device, or if the name is a duplicate.
5211 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5212 * and expands the device name if you passed a format string to
5215 int register_netdev(struct net_device *dev)
5222 * If the name is a format string the caller wants us to do a
5225 if (strchr(dev->name, '%')) {
5226 err = dev_alloc_name(dev, dev->name);
5231 err = register_netdevice(dev);
5236 EXPORT_SYMBOL(register_netdev);
5239 * netdev_wait_allrefs - wait until all references are gone.
5241 * This is called when unregistering network devices.
5243 * Any protocol or device that holds a reference should register
5244 * for netdevice notification, and cleanup and put back the
5245 * reference if they receive an UNREGISTER event.
5246 * We can get stuck here if buggy protocols don't correctly
5249 static void netdev_wait_allrefs(struct net_device *dev)
5251 unsigned long rebroadcast_time, warning_time;
5253 linkwatch_forget_dev(dev);
5255 rebroadcast_time = warning_time = jiffies;
5256 while (atomic_read(&dev->refcnt) != 0) {
5257 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5260 /* Rebroadcast unregister notification */
5261 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5262 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5263 * should have already handle it the first time */
5265 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5267 /* We must not have linkwatch events
5268 * pending on unregister. If this
5269 * happens, we simply run the queue
5270 * unscheduled, resulting in a noop
5273 linkwatch_run_queue();
5278 rebroadcast_time = jiffies;
5283 if (time_after(jiffies, warning_time + 10 * HZ)) {
5284 printk(KERN_EMERG "unregister_netdevice: "
5285 "waiting for %s to become free. Usage "
5287 dev->name, atomic_read(&dev->refcnt));
5288 warning_time = jiffies;
5297 * register_netdevice(x1);
5298 * register_netdevice(x2);
5300 * unregister_netdevice(y1);
5301 * unregister_netdevice(y2);
5307 * We are invoked by rtnl_unlock().
5308 * This allows us to deal with problems:
5309 * 1) We can delete sysfs objects which invoke hotplug
5310 * without deadlocking with linkwatch via keventd.
5311 * 2) Since we run with the RTNL semaphore not held, we can sleep
5312 * safely in order to wait for the netdev refcnt to drop to zero.
5314 * We must not return until all unregister events added during
5315 * the interval the lock was held have been completed.
5317 void netdev_run_todo(void)
5319 struct list_head list;
5321 /* Snapshot list, allow later requests */
5322 list_replace_init(&net_todo_list, &list);
5326 while (!list_empty(&list)) {
5327 struct net_device *dev
5328 = list_first_entry(&list, struct net_device, todo_list);
5329 list_del(&dev->todo_list);
5331 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5332 printk(KERN_ERR "network todo '%s' but state %d\n",
5333 dev->name, dev->reg_state);
5338 dev->reg_state = NETREG_UNREGISTERED;
5340 on_each_cpu(flush_backlog, dev, 1);
5342 netdev_wait_allrefs(dev);
5345 BUG_ON(atomic_read(&dev->refcnt));
5346 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5347 WARN_ON(dev->ip6_ptr);
5348 WARN_ON(dev->dn_ptr);
5350 if (dev->destructor)
5351 dev->destructor(dev);
5353 /* Free network device */
5354 kobject_put(&dev->dev.kobj);
5359 * dev_txq_stats_fold - fold tx_queues stats
5360 * @dev: device to get statistics from
5361 * @stats: struct rtnl_link_stats64 to hold results
5363 void dev_txq_stats_fold(const struct net_device *dev,
5364 struct rtnl_link_stats64 *stats)
5366 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5368 struct netdev_queue *txq;
5370 for (i = 0; i < dev->num_tx_queues; i++) {
5371 txq = netdev_get_tx_queue(dev, i);
5372 spin_lock_bh(&txq->_xmit_lock);
5373 tx_bytes += txq->tx_bytes;
5374 tx_packets += txq->tx_packets;
5375 tx_dropped += txq->tx_dropped;
5376 spin_unlock_bh(&txq->_xmit_lock);
5378 if (tx_bytes || tx_packets || tx_dropped) {
5379 stats->tx_bytes = tx_bytes;
5380 stats->tx_packets = tx_packets;
5381 stats->tx_dropped = tx_dropped;
5384 EXPORT_SYMBOL(dev_txq_stats_fold);
5386 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5387 * fields in the same order, with only the type differing.
5389 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5390 const struct net_device_stats *netdev_stats)
5392 #if BITS_PER_LONG == 64
5393 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5394 memcpy(stats64, netdev_stats, sizeof(*stats64));
5396 size_t i, n = sizeof(*stats64) / sizeof(u64);
5397 const unsigned long *src = (const unsigned long *)netdev_stats;
5398 u64 *dst = (u64 *)stats64;
5400 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5401 sizeof(*stats64) / sizeof(u64));
5402 for (i = 0; i < n; i++)
5408 * dev_get_stats - get network device statistics
5409 * @dev: device to get statistics from
5410 * @storage: place to store stats
5412 * Get network statistics from device. Return @storage.
5413 * The device driver may provide its own method by setting
5414 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5415 * otherwise the internal statistics structure is used.
5417 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5418 struct rtnl_link_stats64 *storage)
5420 const struct net_device_ops *ops = dev->netdev_ops;
5422 if (ops->ndo_get_stats64) {
5423 memset(storage, 0, sizeof(*storage));
5424 return ops->ndo_get_stats64(dev, storage);
5426 if (ops->ndo_get_stats) {
5427 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5430 netdev_stats_to_stats64(storage, &dev->stats);
5431 dev_txq_stats_fold(dev, storage);
5434 EXPORT_SYMBOL(dev_get_stats);
5436 static void netdev_init_one_queue(struct net_device *dev,
5437 struct netdev_queue *queue,
5443 static void netdev_init_queues(struct net_device *dev)
5445 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5446 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5447 spin_lock_init(&dev->tx_global_lock);
5451 * alloc_netdev_mq - allocate network device
5452 * @sizeof_priv: size of private data to allocate space for
5453 * @name: device name format string
5454 * @setup: callback to initialize device
5455 * @queue_count: the number of subqueues to allocate
5457 * Allocates a struct net_device with private data area for driver use
5458 * and performs basic initialization. Also allocates subquue structs
5459 * for each queue on the device at the end of the netdevice.
5461 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5462 void (*setup)(struct net_device *), unsigned int queue_count)
5464 struct netdev_queue *tx;
5465 struct net_device *dev;
5467 struct net_device *p;
5469 BUG_ON(strlen(name) >= sizeof(dev->name));
5471 alloc_size = sizeof(struct net_device);
5473 /* ensure 32-byte alignment of private area */
5474 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5475 alloc_size += sizeof_priv;
5477 /* ensure 32-byte alignment of whole construct */
5478 alloc_size += NETDEV_ALIGN - 1;
5480 p = kzalloc(alloc_size, GFP_KERNEL);
5482 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5486 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5488 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5494 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5495 dev->padded = (char *)dev - (char *)p;
5497 if (dev_addr_init(dev))
5503 dev_net_set(dev, &init_net);
5506 dev->num_tx_queues = queue_count;
5507 dev->real_num_tx_queues = queue_count;
5510 dev->num_rx_queues = queue_count;
5511 dev->real_num_rx_queues = queue_count;
5514 dev->gso_max_size = GSO_MAX_SIZE;
5516 netdev_init_queues(dev);
5518 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5519 dev->ethtool_ntuple_list.count = 0;
5520 INIT_LIST_HEAD(&dev->napi_list);
5521 INIT_LIST_HEAD(&dev->unreg_list);
5522 INIT_LIST_HEAD(&dev->link_watch_list);
5523 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5525 strcpy(dev->name, name);
5534 EXPORT_SYMBOL(alloc_netdev_mq);
5537 * free_netdev - free network device
5540 * This function does the last stage of destroying an allocated device
5541 * interface. The reference to the device object is released.
5542 * If this is the last reference then it will be freed.
5544 void free_netdev(struct net_device *dev)
5546 struct napi_struct *p, *n;
5548 release_net(dev_net(dev));
5552 /* Flush device addresses */
5553 dev_addr_flush(dev);
5555 /* Clear ethtool n-tuple list */
5556 ethtool_ntuple_flush(dev);
5558 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5561 /* Compatibility with error handling in drivers */
5562 if (dev->reg_state == NETREG_UNINITIALIZED) {
5563 kfree((char *)dev - dev->padded);
5567 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5568 dev->reg_state = NETREG_RELEASED;
5570 /* will free via device release */
5571 put_device(&dev->dev);
5573 EXPORT_SYMBOL(free_netdev);
5576 * synchronize_net - Synchronize with packet receive processing
5578 * Wait for packets currently being received to be done.
5579 * Does not block later packets from starting.
5581 void synchronize_net(void)
5586 EXPORT_SYMBOL(synchronize_net);
5589 * unregister_netdevice_queue - remove device from the kernel
5593 * This function shuts down a device interface and removes it
5594 * from the kernel tables.
5595 * If head not NULL, device is queued to be unregistered later.
5597 * Callers must hold the rtnl semaphore. You may want
5598 * unregister_netdev() instead of this.
5601 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5606 list_move_tail(&dev->unreg_list, head);
5608 rollback_registered(dev);
5609 /* Finish processing unregister after unlock */
5613 EXPORT_SYMBOL(unregister_netdevice_queue);
5616 * unregister_netdevice_many - unregister many devices
5617 * @head: list of devices
5619 void unregister_netdevice_many(struct list_head *head)
5621 struct net_device *dev;
5623 if (!list_empty(head)) {
5624 rollback_registered_many(head);
5625 list_for_each_entry(dev, head, unreg_list)
5629 EXPORT_SYMBOL(unregister_netdevice_many);
5632 * unregister_netdev - remove device from the kernel
5635 * This function shuts down a device interface and removes it
5636 * from the kernel tables.
5638 * This is just a wrapper for unregister_netdevice that takes
5639 * the rtnl semaphore. In general you want to use this and not
5640 * unregister_netdevice.
5642 void unregister_netdev(struct net_device *dev)
5645 unregister_netdevice(dev);
5648 EXPORT_SYMBOL(unregister_netdev);
5651 * dev_change_net_namespace - move device to different nethost namespace
5653 * @net: network namespace
5654 * @pat: If not NULL name pattern to try if the current device name
5655 * is already taken in the destination network namespace.
5657 * This function shuts down a device interface and moves it
5658 * to a new network namespace. On success 0 is returned, on
5659 * a failure a netagive errno code is returned.
5661 * Callers must hold the rtnl semaphore.
5664 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5670 /* Don't allow namespace local devices to be moved. */
5672 if (dev->features & NETIF_F_NETNS_LOCAL)
5675 /* Ensure the device has been registrered */
5677 if (dev->reg_state != NETREG_REGISTERED)
5680 /* Get out if there is nothing todo */
5682 if (net_eq(dev_net(dev), net))
5685 /* Pick the destination device name, and ensure
5686 * we can use it in the destination network namespace.
5689 if (__dev_get_by_name(net, dev->name)) {
5690 /* We get here if we can't use the current device name */
5693 if (dev_get_valid_name(dev, pat, 1))
5698 * And now a mini version of register_netdevice unregister_netdevice.
5701 /* If device is running close it first. */
5704 /* And unlink it from device chain */
5706 unlist_netdevice(dev);
5710 /* Shutdown queueing discipline. */
5713 /* Notify protocols, that we are about to destroy
5714 this device. They should clean all the things.
5716 Note that dev->reg_state stays at NETREG_REGISTERED.
5717 This is wanted because this way 8021q and macvlan know
5718 the device is just moving and can keep their slaves up.
5720 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5721 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5724 * Flush the unicast and multicast chains
5729 /* Actually switch the network namespace */
5730 dev_net_set(dev, net);
5732 /* If there is an ifindex conflict assign a new one */
5733 if (__dev_get_by_index(net, dev->ifindex)) {
5734 int iflink = (dev->iflink == dev->ifindex);
5735 dev->ifindex = dev_new_index(net);
5737 dev->iflink = dev->ifindex;
5740 /* Fixup kobjects */
5741 err = device_rename(&dev->dev, dev->name);
5744 /* Add the device back in the hashes */
5745 list_netdevice(dev);
5747 /* Notify protocols, that a new device appeared. */
5748 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5751 * Prevent userspace races by waiting until the network
5752 * device is fully setup before sending notifications.
5754 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5761 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5763 static int dev_cpu_callback(struct notifier_block *nfb,
5764 unsigned long action,
5767 struct sk_buff **list_skb;
5768 struct sk_buff *skb;
5769 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5770 struct softnet_data *sd, *oldsd;
5772 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5775 local_irq_disable();
5776 cpu = smp_processor_id();
5777 sd = &per_cpu(softnet_data, cpu);
5778 oldsd = &per_cpu(softnet_data, oldcpu);
5780 /* Find end of our completion_queue. */
5781 list_skb = &sd->completion_queue;
5783 list_skb = &(*list_skb)->next;
5784 /* Append completion queue from offline CPU. */
5785 *list_skb = oldsd->completion_queue;
5786 oldsd->completion_queue = NULL;
5788 /* Append output queue from offline CPU. */
5789 if (oldsd->output_queue) {
5790 *sd->output_queue_tailp = oldsd->output_queue;
5791 sd->output_queue_tailp = oldsd->output_queue_tailp;
5792 oldsd->output_queue = NULL;
5793 oldsd->output_queue_tailp = &oldsd->output_queue;
5796 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5799 /* Process offline CPU's input_pkt_queue */
5800 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5802 input_queue_head_incr(oldsd);
5804 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5806 input_queue_head_incr(oldsd);
5814 * netdev_increment_features - increment feature set by one
5815 * @all: current feature set
5816 * @one: new feature set
5817 * @mask: mask feature set
5819 * Computes a new feature set after adding a device with feature set
5820 * @one to the master device with current feature set @all. Will not
5821 * enable anything that is off in @mask. Returns the new feature set.
5823 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5826 /* If device needs checksumming, downgrade to it. */
5827 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5828 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5829 else if (mask & NETIF_F_ALL_CSUM) {
5830 /* If one device supports v4/v6 checksumming, set for all. */
5831 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5832 !(all & NETIF_F_GEN_CSUM)) {
5833 all &= ~NETIF_F_ALL_CSUM;
5834 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5837 /* If one device supports hw checksumming, set for all. */
5838 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5839 all &= ~NETIF_F_ALL_CSUM;
5840 all |= NETIF_F_HW_CSUM;
5844 one |= NETIF_F_ALL_CSUM;
5846 one |= all & NETIF_F_ONE_FOR_ALL;
5847 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5848 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5852 EXPORT_SYMBOL(netdev_increment_features);
5854 static struct hlist_head *netdev_create_hash(void)
5857 struct hlist_head *hash;
5859 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5861 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5862 INIT_HLIST_HEAD(&hash[i]);
5867 /* Initialize per network namespace state */
5868 static int __net_init netdev_init(struct net *net)
5870 INIT_LIST_HEAD(&net->dev_base_head);
5872 net->dev_name_head = netdev_create_hash();
5873 if (net->dev_name_head == NULL)
5876 net->dev_index_head = netdev_create_hash();
5877 if (net->dev_index_head == NULL)
5883 kfree(net->dev_name_head);
5889 * netdev_drivername - network driver for the device
5890 * @dev: network device
5891 * @buffer: buffer for resulting name
5892 * @len: size of buffer
5894 * Determine network driver for device.
5896 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5898 const struct device_driver *driver;
5899 const struct device *parent;
5901 if (len <= 0 || !buffer)
5905 parent = dev->dev.parent;
5910 driver = parent->driver;
5911 if (driver && driver->name)
5912 strlcpy(buffer, driver->name, len);
5916 static int __netdev_printk(const char *level, const struct net_device *dev,
5917 struct va_format *vaf)
5921 if (dev && dev->dev.parent)
5922 r = dev_printk(level, dev->dev.parent, "%s: %pV",
5923 netdev_name(dev), vaf);
5925 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
5927 r = printk("%s(NULL net_device): %pV", level, vaf);
5932 int netdev_printk(const char *level, const struct net_device *dev,
5933 const char *format, ...)
5935 struct va_format vaf;
5939 va_start(args, format);
5944 r = __netdev_printk(level, dev, &vaf);
5949 EXPORT_SYMBOL(netdev_printk);
5951 #define define_netdev_printk_level(func, level) \
5952 int func(const struct net_device *dev, const char *fmt, ...) \
5955 struct va_format vaf; \
5958 va_start(args, fmt); \
5963 r = __netdev_printk(level, dev, &vaf); \
5968 EXPORT_SYMBOL(func);
5970 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
5971 define_netdev_printk_level(netdev_alert, KERN_ALERT);
5972 define_netdev_printk_level(netdev_crit, KERN_CRIT);
5973 define_netdev_printk_level(netdev_err, KERN_ERR);
5974 define_netdev_printk_level(netdev_warn, KERN_WARNING);
5975 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
5976 define_netdev_printk_level(netdev_info, KERN_INFO);
5978 static void __net_exit netdev_exit(struct net *net)
5980 kfree(net->dev_name_head);
5981 kfree(net->dev_index_head);
5984 static struct pernet_operations __net_initdata netdev_net_ops = {
5985 .init = netdev_init,
5986 .exit = netdev_exit,
5989 static void __net_exit default_device_exit(struct net *net)
5991 struct net_device *dev, *aux;
5993 * Push all migratable network devices back to the
5994 * initial network namespace
5997 for_each_netdev_safe(net, dev, aux) {
5999 char fb_name[IFNAMSIZ];
6001 /* Ignore unmoveable devices (i.e. loopback) */
6002 if (dev->features & NETIF_F_NETNS_LOCAL)
6005 /* Leave virtual devices for the generic cleanup */
6006 if (dev->rtnl_link_ops)
6009 /* Push remaing network devices to init_net */
6010 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6011 err = dev_change_net_namespace(dev, &init_net, fb_name);
6013 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6014 __func__, dev->name, err);
6021 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6023 /* At exit all network devices most be removed from a network
6024 * namespace. Do this in the reverse order of registeration.
6025 * Do this across as many network namespaces as possible to
6026 * improve batching efficiency.
6028 struct net_device *dev;
6030 LIST_HEAD(dev_kill_list);
6033 list_for_each_entry(net, net_list, exit_list) {
6034 for_each_netdev_reverse(net, dev) {
6035 if (dev->rtnl_link_ops)
6036 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6038 unregister_netdevice_queue(dev, &dev_kill_list);
6041 unregister_netdevice_many(&dev_kill_list);
6045 static struct pernet_operations __net_initdata default_device_ops = {
6046 .exit = default_device_exit,
6047 .exit_batch = default_device_exit_batch,
6051 * Initialize the DEV module. At boot time this walks the device list and
6052 * unhooks any devices that fail to initialise (normally hardware not
6053 * present) and leaves us with a valid list of present and active devices.
6058 * This is called single threaded during boot, so no need
6059 * to take the rtnl semaphore.
6061 static int __init net_dev_init(void)
6063 int i, rc = -ENOMEM;
6065 BUG_ON(!dev_boot_phase);
6067 if (dev_proc_init())
6070 if (netdev_kobject_init())
6073 INIT_LIST_HEAD(&ptype_all);
6074 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6075 INIT_LIST_HEAD(&ptype_base[i]);
6077 if (register_pernet_subsys(&netdev_net_ops))
6081 * Initialise the packet receive queues.
6084 for_each_possible_cpu(i) {
6085 struct softnet_data *sd = &per_cpu(softnet_data, i);
6087 memset(sd, 0, sizeof(*sd));
6088 skb_queue_head_init(&sd->input_pkt_queue);
6089 skb_queue_head_init(&sd->process_queue);
6090 sd->completion_queue = NULL;
6091 INIT_LIST_HEAD(&sd->poll_list);
6092 sd->output_queue = NULL;
6093 sd->output_queue_tailp = &sd->output_queue;
6095 sd->csd.func = rps_trigger_softirq;
6101 sd->backlog.poll = process_backlog;
6102 sd->backlog.weight = weight_p;
6103 sd->backlog.gro_list = NULL;
6104 sd->backlog.gro_count = 0;
6109 /* The loopback device is special if any other network devices
6110 * is present in a network namespace the loopback device must
6111 * be present. Since we now dynamically allocate and free the
6112 * loopback device ensure this invariant is maintained by
6113 * keeping the loopback device as the first device on the
6114 * list of network devices. Ensuring the loopback devices
6115 * is the first device that appears and the last network device
6118 if (register_pernet_device(&loopback_net_ops))
6121 if (register_pernet_device(&default_device_ops))
6124 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6125 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6127 hotcpu_notifier(dev_cpu_callback, 0);
6135 subsys_initcall(net_dev_init);
6137 static int __init initialize_hashrnd(void)
6139 get_random_bytes(&hashrnd, sizeof(hashrnd));
6143 late_initcall_sync(initialize_hashrnd);