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 <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/if_tunnel.h>
138 #include "net-sysfs.h"
140 /* Instead of increasing this, you should create a hash table. */
141 #define MAX_GRO_SKBS 8
143 /* This should be increased if a protocol with a bigger head is added. */
144 #define GRO_MAX_HEAD (MAX_HEADER + 128)
147 * The list of packet types we will receive (as opposed to discard)
148 * and the routines to invoke.
150 * Why 16. Because with 16 the only overlap we get on a hash of the
151 * low nibble of the protocol value is RARP/SNAP/X.25.
153 * NOTE: That is no longer true with the addition of VLAN tags. Not
154 * sure which should go first, but I bet it won't make much
155 * difference if we are running VLANs. The good news is that
156 * this protocol won't be in the list unless compiled in, so
157 * the average user (w/out VLANs) will not be adversely affected.
174 #define PTYPE_HASH_SIZE (16)
175 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
177 static DEFINE_SPINLOCK(ptype_lock);
178 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
179 static struct list_head ptype_all __read_mostly; /* Taps */
182 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
185 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
187 * Writers must hold the rtnl semaphore while they loop through the
188 * dev_base_head list, and hold dev_base_lock for writing when they do the
189 * actual updates. This allows pure readers to access the list even
190 * while a writer is preparing to update it.
192 * To put it another way, dev_base_lock is held for writing only to
193 * protect against pure readers; the rtnl semaphore provides the
194 * protection against other writers.
196 * See, for example usages, register_netdevice() and
197 * unregister_netdevice(), which must be called with the rtnl
200 DEFINE_RWLOCK(dev_base_lock);
201 EXPORT_SYMBOL(dev_base_lock);
203 static inline void dev_base_seq_inc(struct net *net)
205 while (++net->dev_base_seq == 0);
208 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
210 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
211 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
214 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
216 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
219 static inline void rps_lock(struct softnet_data *sd)
222 spin_lock(&sd->input_pkt_queue.lock);
226 static inline void rps_unlock(struct softnet_data *sd)
229 spin_unlock(&sd->input_pkt_queue.lock);
233 /* Device list insertion */
234 static int list_netdevice(struct net_device *dev)
236 struct net *net = dev_net(dev);
240 write_lock_bh(&dev_base_lock);
241 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
242 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
243 hlist_add_head_rcu(&dev->index_hlist,
244 dev_index_hash(net, dev->ifindex));
245 write_unlock_bh(&dev_base_lock);
247 dev_base_seq_inc(net);
252 /* Device list removal
253 * caller must respect a RCU grace period before freeing/reusing dev
255 static void unlist_netdevice(struct net_device *dev)
259 /* Unlink dev from the device chain */
260 write_lock_bh(&dev_base_lock);
261 list_del_rcu(&dev->dev_list);
262 hlist_del_rcu(&dev->name_hlist);
263 hlist_del_rcu(&dev->index_hlist);
264 write_unlock_bh(&dev_base_lock);
266 dev_base_seq_inc(dev_net(dev));
273 static RAW_NOTIFIER_HEAD(netdev_chain);
276 * Device drivers call our routines to queue packets here. We empty the
277 * queue in the local softnet handler.
280 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
281 EXPORT_PER_CPU_SYMBOL(softnet_data);
283 #ifdef CONFIG_LOCKDEP
285 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
286 * according to dev->type
288 static const unsigned short netdev_lock_type[] =
289 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
290 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
291 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
292 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
293 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
294 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
295 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
296 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
297 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
298 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
299 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
300 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
301 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
302 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
303 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
304 ARPHRD_VOID, ARPHRD_NONE};
306 static const char *const netdev_lock_name[] =
307 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
308 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
309 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
310 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
311 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
312 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
313 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
314 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
315 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
316 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
317 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
318 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
319 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
320 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
321 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
322 "_xmit_VOID", "_xmit_NONE"};
324 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
325 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
327 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
331 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
332 if (netdev_lock_type[i] == dev_type)
334 /* the last key is used by default */
335 return ARRAY_SIZE(netdev_lock_type) - 1;
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
343 i = netdev_lock_pos(dev_type);
344 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
345 netdev_lock_name[i]);
348 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
352 i = netdev_lock_pos(dev->type);
353 lockdep_set_class_and_name(&dev->addr_list_lock,
354 &netdev_addr_lock_key[i],
355 netdev_lock_name[i]);
358 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
359 unsigned short dev_type)
362 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
367 /*******************************************************************************
369 Protocol management and registration routines
371 *******************************************************************************/
374 * Add a protocol ID to the list. Now that the input handler is
375 * smarter we can dispense with all the messy stuff that used to be
378 * BEWARE!!! Protocol handlers, mangling input packets,
379 * MUST BE last in hash buckets and checking protocol handlers
380 * MUST start from promiscuous ptype_all chain in net_bh.
381 * It is true now, do not change it.
382 * Explanation follows: if protocol handler, mangling packet, will
383 * be the first on list, it is not able to sense, that packet
384 * is cloned and should be copied-on-write, so that it will
385 * change it and subsequent readers will get broken packet.
389 static inline struct list_head *ptype_head(const struct packet_type *pt)
391 if (pt->type == htons(ETH_P_ALL))
394 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
398 * dev_add_pack - add packet handler
399 * @pt: packet type declaration
401 * Add a protocol handler to the networking stack. The passed &packet_type
402 * is linked into kernel lists and may not be freed until it has been
403 * removed from the kernel lists.
405 * This call does not sleep therefore it can not
406 * guarantee all CPU's that are in middle of receiving packets
407 * will see the new packet type (until the next received packet).
410 void dev_add_pack(struct packet_type *pt)
412 struct list_head *head = ptype_head(pt);
414 spin_lock(&ptype_lock);
415 list_add_rcu(&pt->list, head);
416 spin_unlock(&ptype_lock);
418 EXPORT_SYMBOL(dev_add_pack);
421 * __dev_remove_pack - remove packet handler
422 * @pt: packet type declaration
424 * Remove a protocol handler that was previously added to the kernel
425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
426 * from the kernel lists and can be freed or reused once this function
429 * The packet type might still be in use by receivers
430 * and must not be freed until after all the CPU's have gone
431 * through a quiescent state.
433 void __dev_remove_pack(struct packet_type *pt)
435 struct list_head *head = ptype_head(pt);
436 struct packet_type *pt1;
438 spin_lock(&ptype_lock);
440 list_for_each_entry(pt1, head, list) {
442 list_del_rcu(&pt->list);
447 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
449 spin_unlock(&ptype_lock);
451 EXPORT_SYMBOL(__dev_remove_pack);
454 * dev_remove_pack - remove packet handler
455 * @pt: packet type declaration
457 * Remove a protocol handler that was previously added to the kernel
458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
459 * from the kernel lists and can be freed or reused once this function
462 * This call sleeps to guarantee that no CPU is looking at the packet
465 void dev_remove_pack(struct packet_type *pt)
467 __dev_remove_pack(pt);
471 EXPORT_SYMBOL(dev_remove_pack);
473 /******************************************************************************
475 Device Boot-time Settings Routines
477 *******************************************************************************/
479 /* Boot time configuration table */
480 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
483 * netdev_boot_setup_add - add new setup entry
484 * @name: name of the device
485 * @map: configured settings for the device
487 * Adds new setup entry to the dev_boot_setup list. The function
488 * returns 0 on error and 1 on success. This is a generic routine to
491 static int netdev_boot_setup_add(char *name, struct ifmap *map)
493 struct netdev_boot_setup *s;
497 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
498 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
499 memset(s[i].name, 0, sizeof(s[i].name));
500 strlcpy(s[i].name, name, IFNAMSIZ);
501 memcpy(&s[i].map, map, sizeof(s[i].map));
506 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
510 * netdev_boot_setup_check - check boot time settings
511 * @dev: the netdevice
513 * Check boot time settings for the device.
514 * The found settings are set for the device to be used
515 * later in the device probing.
516 * Returns 0 if no settings found, 1 if they are.
518 int netdev_boot_setup_check(struct net_device *dev)
520 struct netdev_boot_setup *s = dev_boot_setup;
523 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
524 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
525 !strcmp(dev->name, s[i].name)) {
526 dev->irq = s[i].map.irq;
527 dev->base_addr = s[i].map.base_addr;
528 dev->mem_start = s[i].map.mem_start;
529 dev->mem_end = s[i].map.mem_end;
535 EXPORT_SYMBOL(netdev_boot_setup_check);
539 * netdev_boot_base - get address from boot time settings
540 * @prefix: prefix for network device
541 * @unit: id for network device
543 * Check boot time settings for the base address of device.
544 * The found settings are set for the device to be used
545 * later in the device probing.
546 * Returns 0 if no settings found.
548 unsigned long netdev_boot_base(const char *prefix, int unit)
550 const struct netdev_boot_setup *s = dev_boot_setup;
554 sprintf(name, "%s%d", prefix, unit);
557 * If device already registered then return base of 1
558 * to indicate not to probe for this interface
560 if (__dev_get_by_name(&init_net, name))
563 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
564 if (!strcmp(name, s[i].name))
565 return s[i].map.base_addr;
570 * Saves at boot time configured settings for any netdevice.
572 int __init netdev_boot_setup(char *str)
577 str = get_options(str, ARRAY_SIZE(ints), ints);
582 memset(&map, 0, sizeof(map));
586 map.base_addr = ints[2];
588 map.mem_start = ints[3];
590 map.mem_end = ints[4];
592 /* Add new entry to the list */
593 return netdev_boot_setup_add(str, &map);
596 __setup("netdev=", netdev_boot_setup);
598 /*******************************************************************************
600 Device Interface Subroutines
602 *******************************************************************************/
605 * __dev_get_by_name - find a device by its name
606 * @net: the applicable net namespace
607 * @name: name to find
609 * Find an interface by name. Must be called under RTNL semaphore
610 * or @dev_base_lock. If the name is found a pointer to the device
611 * is returned. If the name is not found then %NULL is returned. The
612 * reference counters are not incremented so the caller must be
613 * careful with locks.
616 struct net_device *__dev_get_by_name(struct net *net, const char *name)
618 struct hlist_node *p;
619 struct net_device *dev;
620 struct hlist_head *head = dev_name_hash(net, name);
622 hlist_for_each_entry(dev, p, head, name_hlist)
623 if (!strncmp(dev->name, name, IFNAMSIZ))
628 EXPORT_SYMBOL(__dev_get_by_name);
631 * dev_get_by_name_rcu - find a device by its name
632 * @net: the applicable net namespace
633 * @name: name to find
635 * Find an interface by name.
636 * If the name is found a pointer to the device is returned.
637 * If the name is not found then %NULL is returned.
638 * The reference counters are not incremented so the caller must be
639 * careful with locks. The caller must hold RCU lock.
642 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
644 struct hlist_node *p;
645 struct net_device *dev;
646 struct hlist_head *head = dev_name_hash(net, name);
648 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
649 if (!strncmp(dev->name, name, IFNAMSIZ))
654 EXPORT_SYMBOL(dev_get_by_name_rcu);
657 * dev_get_by_name - find a device by its name
658 * @net: the applicable net namespace
659 * @name: name to find
661 * Find an interface by name. This can be called from any
662 * context and does its own locking. The returned handle has
663 * the usage count incremented and the caller must use dev_put() to
664 * release it when it is no longer needed. %NULL is returned if no
665 * matching device is found.
668 struct net_device *dev_get_by_name(struct net *net, const char *name)
670 struct net_device *dev;
673 dev = dev_get_by_name_rcu(net, name);
679 EXPORT_SYMBOL(dev_get_by_name);
682 * __dev_get_by_index - find a device by its ifindex
683 * @net: the applicable net namespace
684 * @ifindex: index of device
686 * Search for an interface by index. Returns %NULL if the device
687 * is not found or a pointer to the device. The device has not
688 * had its reference counter increased so the caller must be careful
689 * about locking. The caller must hold either the RTNL semaphore
693 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
695 struct hlist_node *p;
696 struct net_device *dev;
697 struct hlist_head *head = dev_index_hash(net, ifindex);
699 hlist_for_each_entry(dev, p, head, index_hlist)
700 if (dev->ifindex == ifindex)
705 EXPORT_SYMBOL(__dev_get_by_index);
708 * dev_get_by_index_rcu - find a device by its ifindex
709 * @net: the applicable net namespace
710 * @ifindex: index of device
712 * Search for an interface by index. Returns %NULL if the device
713 * is not found or a pointer to the device. The device has not
714 * had its reference counter increased so the caller must be careful
715 * about locking. The caller must hold RCU lock.
718 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
720 struct hlist_node *p;
721 struct net_device *dev;
722 struct hlist_head *head = dev_index_hash(net, ifindex);
724 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
725 if (dev->ifindex == ifindex)
730 EXPORT_SYMBOL(dev_get_by_index_rcu);
734 * dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns NULL if the device
739 * is not found or a pointer to the device. The device returned has
740 * had a reference added and the pointer is safe until the user calls
741 * dev_put to indicate they have finished with it.
744 struct net_device *dev_get_by_index(struct net *net, int ifindex)
746 struct net_device *dev;
749 dev = dev_get_by_index_rcu(net, ifindex);
755 EXPORT_SYMBOL(dev_get_by_index);
758 * dev_getbyhwaddr_rcu - find a device by its hardware address
759 * @net: the applicable net namespace
760 * @type: media type of device
761 * @ha: hardware address
763 * Search for an interface by MAC address. Returns NULL if the device
764 * is not found or a pointer to the device.
765 * The caller must hold RCU or RTNL.
766 * The returned device has not had its ref count increased
767 * and the caller must therefore be careful about locking
771 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
774 struct net_device *dev;
776 for_each_netdev_rcu(net, dev)
777 if (dev->type == type &&
778 !memcmp(dev->dev_addr, ha, dev->addr_len))
783 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
785 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
787 struct net_device *dev;
790 for_each_netdev(net, dev)
791 if (dev->type == type)
796 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
798 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
800 struct net_device *dev, *ret = NULL;
803 for_each_netdev_rcu(net, dev)
804 if (dev->type == type) {
812 EXPORT_SYMBOL(dev_getfirstbyhwtype);
815 * dev_get_by_flags_rcu - find any device with given flags
816 * @net: the applicable net namespace
817 * @if_flags: IFF_* values
818 * @mask: bitmask of bits in if_flags to check
820 * Search for any interface with the given flags. Returns NULL if a device
821 * is not found or a pointer to the device. Must be called inside
822 * rcu_read_lock(), and result refcount is unchanged.
825 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
828 struct net_device *dev, *ret;
831 for_each_netdev_rcu(net, dev) {
832 if (((dev->flags ^ if_flags) & mask) == 0) {
839 EXPORT_SYMBOL(dev_get_by_flags_rcu);
842 * dev_valid_name - check if name is okay for network device
845 * Network device names need to be valid file names to
846 * to allow sysfs to work. We also disallow any kind of
849 int dev_valid_name(const char *name)
853 if (strlen(name) >= IFNAMSIZ)
855 if (!strcmp(name, ".") || !strcmp(name, ".."))
859 if (*name == '/' || isspace(*name))
865 EXPORT_SYMBOL(dev_valid_name);
868 * __dev_alloc_name - allocate a name for a device
869 * @net: network namespace to allocate the device name in
870 * @name: name format string
871 * @buf: scratch buffer and result name string
873 * Passed a format string - eg "lt%d" it will try and find a suitable
874 * id. It scans list of devices to build up a free map, then chooses
875 * the first empty slot. The caller must hold the dev_base or rtnl lock
876 * while allocating the name and adding the device in order to avoid
878 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
879 * Returns the number of the unit assigned or a negative errno code.
882 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
886 const int max_netdevices = 8*PAGE_SIZE;
887 unsigned long *inuse;
888 struct net_device *d;
890 p = strnchr(name, IFNAMSIZ-1, '%');
893 * Verify the string as this thing may have come from
894 * the user. There must be either one "%d" and no other "%"
897 if (p[1] != 'd' || strchr(p + 2, '%'))
900 /* Use one page as a bit array of possible slots */
901 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
905 for_each_netdev(net, d) {
906 if (!sscanf(d->name, name, &i))
908 if (i < 0 || i >= max_netdevices)
911 /* avoid cases where sscanf is not exact inverse of printf */
912 snprintf(buf, IFNAMSIZ, name, i);
913 if (!strncmp(buf, d->name, IFNAMSIZ))
917 i = find_first_zero_bit(inuse, max_netdevices);
918 free_page((unsigned long) inuse);
922 snprintf(buf, IFNAMSIZ, name, i);
923 if (!__dev_get_by_name(net, buf))
926 /* It is possible to run out of possible slots
927 * when the name is long and there isn't enough space left
928 * for the digits, or if all bits are used.
934 * dev_alloc_name - allocate a name for a device
936 * @name: name format string
938 * Passed a format string - eg "lt%d" it will try and find a suitable
939 * id. It scans list of devices to build up a free map, then chooses
940 * the first empty slot. The caller must hold the dev_base or rtnl lock
941 * while allocating the name and adding the device in order to avoid
943 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
944 * Returns the number of the unit assigned or a negative errno code.
947 int dev_alloc_name(struct net_device *dev, const char *name)
953 BUG_ON(!dev_net(dev));
955 ret = __dev_alloc_name(net, name, buf);
957 strlcpy(dev->name, buf, IFNAMSIZ);
960 EXPORT_SYMBOL(dev_alloc_name);
962 static int dev_get_valid_name(struct net_device *dev, const char *name)
966 BUG_ON(!dev_net(dev));
969 if (!dev_valid_name(name))
972 if (strchr(name, '%'))
973 return dev_alloc_name(dev, name);
974 else if (__dev_get_by_name(net, name))
976 else if (dev->name != name)
977 strlcpy(dev->name, name, IFNAMSIZ);
983 * dev_change_name - change name of a device
985 * @newname: name (or format string) must be at least IFNAMSIZ
987 * Change name of a device, can pass format strings "eth%d".
990 int dev_change_name(struct net_device *dev, const char *newname)
992 char oldname[IFNAMSIZ];
998 BUG_ON(!dev_net(dev));
1001 if (dev->flags & IFF_UP)
1004 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1007 memcpy(oldname, dev->name, IFNAMSIZ);
1009 err = dev_get_valid_name(dev, newname);
1014 ret = device_rename(&dev->dev, dev->name);
1016 memcpy(dev->name, oldname, IFNAMSIZ);
1020 write_lock_bh(&dev_base_lock);
1021 hlist_del_rcu(&dev->name_hlist);
1022 write_unlock_bh(&dev_base_lock);
1026 write_lock_bh(&dev_base_lock);
1027 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1028 write_unlock_bh(&dev_base_lock);
1030 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1031 ret = notifier_to_errno(ret);
1034 /* err >= 0 after dev_alloc_name() or stores the first errno */
1037 memcpy(dev->name, oldname, IFNAMSIZ);
1041 "%s: name change rollback failed: %d.\n",
1050 * dev_set_alias - change ifalias of a device
1052 * @alias: name up to IFALIASZ
1053 * @len: limit of bytes to copy from info
1055 * Set ifalias for a device,
1057 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1061 if (len >= IFALIASZ)
1066 kfree(dev->ifalias);
1067 dev->ifalias = NULL;
1072 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1076 strlcpy(dev->ifalias, alias, len+1);
1082 * netdev_features_change - device changes features
1083 * @dev: device to cause notification
1085 * Called to indicate a device has changed features.
1087 void netdev_features_change(struct net_device *dev)
1089 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1091 EXPORT_SYMBOL(netdev_features_change);
1094 * netdev_state_change - device changes state
1095 * @dev: device to cause notification
1097 * Called to indicate a device has changed state. This function calls
1098 * the notifier chains for netdev_chain and sends a NEWLINK message
1099 * to the routing socket.
1101 void netdev_state_change(struct net_device *dev)
1103 if (dev->flags & IFF_UP) {
1104 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1105 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1108 EXPORT_SYMBOL(netdev_state_change);
1110 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1112 return call_netdevice_notifiers(event, dev);
1114 EXPORT_SYMBOL(netdev_bonding_change);
1117 * dev_load - load a network module
1118 * @net: the applicable net namespace
1119 * @name: name of interface
1121 * If a network interface is not present and the process has suitable
1122 * privileges this function loads the module. If module loading is not
1123 * available in this kernel then it becomes a nop.
1126 void dev_load(struct net *net, const char *name)
1128 struct net_device *dev;
1132 dev = dev_get_by_name_rcu(net, name);
1136 if (no_module && capable(CAP_NET_ADMIN))
1137 no_module = request_module("netdev-%s", name);
1138 if (no_module && capable(CAP_SYS_MODULE)) {
1139 if (!request_module("%s", name))
1140 pr_err("Loading kernel module for a network device "
1141 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1145 EXPORT_SYMBOL(dev_load);
1147 static int __dev_open(struct net_device *dev)
1149 const struct net_device_ops *ops = dev->netdev_ops;
1154 if (!netif_device_present(dev))
1157 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1158 ret = notifier_to_errno(ret);
1162 set_bit(__LINK_STATE_START, &dev->state);
1164 if (ops->ndo_validate_addr)
1165 ret = ops->ndo_validate_addr(dev);
1167 if (!ret && ops->ndo_open)
1168 ret = ops->ndo_open(dev);
1171 clear_bit(__LINK_STATE_START, &dev->state);
1173 dev->flags |= IFF_UP;
1174 net_dmaengine_get();
1175 dev_set_rx_mode(dev);
1183 * dev_open - prepare an interface for use.
1184 * @dev: device to open
1186 * Takes a device from down to up state. The device's private open
1187 * function is invoked and then the multicast lists are loaded. Finally
1188 * the device is moved into the up state and a %NETDEV_UP message is
1189 * sent to the netdev notifier chain.
1191 * Calling this function on an active interface is a nop. On a failure
1192 * a negative errno code is returned.
1194 int dev_open(struct net_device *dev)
1198 if (dev->flags & IFF_UP)
1201 ret = __dev_open(dev);
1205 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1206 call_netdevice_notifiers(NETDEV_UP, dev);
1210 EXPORT_SYMBOL(dev_open);
1212 static int __dev_close_many(struct list_head *head)
1214 struct net_device *dev;
1219 list_for_each_entry(dev, head, unreg_list) {
1220 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1222 clear_bit(__LINK_STATE_START, &dev->state);
1224 /* Synchronize to scheduled poll. We cannot touch poll list, it
1225 * can be even on different cpu. So just clear netif_running().
1227 * dev->stop() will invoke napi_disable() on all of it's
1228 * napi_struct instances on this device.
1230 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1233 dev_deactivate_many(head);
1235 list_for_each_entry(dev, head, unreg_list) {
1236 const struct net_device_ops *ops = dev->netdev_ops;
1239 * Call the device specific close. This cannot fail.
1240 * Only if device is UP
1242 * We allow it to be called even after a DETACH hot-plug
1248 dev->flags &= ~IFF_UP;
1249 net_dmaengine_put();
1255 static int __dev_close(struct net_device *dev)
1260 list_add(&dev->unreg_list, &single);
1261 retval = __dev_close_many(&single);
1266 static int dev_close_many(struct list_head *head)
1268 struct net_device *dev, *tmp;
1269 LIST_HEAD(tmp_list);
1271 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1272 if (!(dev->flags & IFF_UP))
1273 list_move(&dev->unreg_list, &tmp_list);
1275 __dev_close_many(head);
1277 list_for_each_entry(dev, head, unreg_list) {
1278 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1279 call_netdevice_notifiers(NETDEV_DOWN, dev);
1282 /* rollback_registered_many needs the complete original list */
1283 list_splice(&tmp_list, head);
1288 * dev_close - shutdown an interface.
1289 * @dev: device to shutdown
1291 * This function moves an active device into down state. A
1292 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1293 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1296 int dev_close(struct net_device *dev)
1298 if (dev->flags & IFF_UP) {
1301 list_add(&dev->unreg_list, &single);
1302 dev_close_many(&single);
1307 EXPORT_SYMBOL(dev_close);
1311 * dev_disable_lro - disable Large Receive Offload on a device
1314 * Disable Large Receive Offload (LRO) on a net device. Must be
1315 * called under RTNL. This is needed if received packets may be
1316 * forwarded to another interface.
1318 void dev_disable_lro(struct net_device *dev)
1323 * If we're trying to disable lro on a vlan device
1324 * use the underlying physical device instead
1326 if (is_vlan_dev(dev))
1327 dev = vlan_dev_real_dev(dev);
1329 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1330 flags = dev->ethtool_ops->get_flags(dev);
1332 flags = ethtool_op_get_flags(dev);
1334 if (!(flags & ETH_FLAG_LRO))
1337 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1338 if (unlikely(dev->features & NETIF_F_LRO))
1339 netdev_WARN(dev, "failed to disable LRO!\n");
1341 EXPORT_SYMBOL(dev_disable_lro);
1344 static int dev_boot_phase = 1;
1347 * register_netdevice_notifier - register a network notifier block
1350 * Register a notifier to be called when network device events occur.
1351 * The notifier passed is linked into the kernel structures and must
1352 * not be reused until it has been unregistered. A negative errno code
1353 * is returned on a failure.
1355 * When registered all registration and up events are replayed
1356 * to the new notifier to allow device to have a race free
1357 * view of the network device list.
1360 int register_netdevice_notifier(struct notifier_block *nb)
1362 struct net_device *dev;
1363 struct net_device *last;
1368 err = raw_notifier_chain_register(&netdev_chain, nb);
1374 for_each_netdev(net, dev) {
1375 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1376 err = notifier_to_errno(err);
1380 if (!(dev->flags & IFF_UP))
1383 nb->notifier_call(nb, NETDEV_UP, dev);
1394 for_each_netdev(net, dev) {
1398 if (dev->flags & IFF_UP) {
1399 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1400 nb->notifier_call(nb, NETDEV_DOWN, dev);
1402 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1403 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1407 raw_notifier_chain_unregister(&netdev_chain, nb);
1410 EXPORT_SYMBOL(register_netdevice_notifier);
1413 * unregister_netdevice_notifier - unregister a network notifier block
1416 * Unregister a notifier previously registered by
1417 * register_netdevice_notifier(). The notifier is unlinked into the
1418 * kernel structures and may then be reused. A negative errno code
1419 * is returned on a failure.
1422 int unregister_netdevice_notifier(struct notifier_block *nb)
1427 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1431 EXPORT_SYMBOL(unregister_netdevice_notifier);
1434 * call_netdevice_notifiers - call all network notifier blocks
1435 * @val: value passed unmodified to notifier function
1436 * @dev: net_device pointer passed unmodified to notifier function
1438 * Call all network notifier blocks. Parameters and return value
1439 * are as for raw_notifier_call_chain().
1442 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1445 return raw_notifier_call_chain(&netdev_chain, val, dev);
1447 EXPORT_SYMBOL(call_netdevice_notifiers);
1449 /* When > 0 there are consumers of rx skb time stamps */
1450 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1452 void net_enable_timestamp(void)
1454 atomic_inc(&netstamp_needed);
1456 EXPORT_SYMBOL(net_enable_timestamp);
1458 void net_disable_timestamp(void)
1460 atomic_dec(&netstamp_needed);
1462 EXPORT_SYMBOL(net_disable_timestamp);
1464 static inline void net_timestamp_set(struct sk_buff *skb)
1466 if (atomic_read(&netstamp_needed))
1467 __net_timestamp(skb);
1469 skb->tstamp.tv64 = 0;
1472 static inline void net_timestamp_check(struct sk_buff *skb)
1474 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1475 __net_timestamp(skb);
1478 static inline bool is_skb_forwardable(struct net_device *dev,
1479 struct sk_buff *skb)
1483 if (!(dev->flags & IFF_UP))
1486 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1487 if (skb->len <= len)
1490 /* if TSO is enabled, we don't care about the length as the packet
1491 * could be forwarded without being segmented before
1493 if (skb_is_gso(skb))
1500 * dev_forward_skb - loopback an skb to another netif
1502 * @dev: destination network device
1503 * @skb: buffer to forward
1506 * NET_RX_SUCCESS (no congestion)
1507 * NET_RX_DROP (packet was dropped, but freed)
1509 * dev_forward_skb can be used for injecting an skb from the
1510 * start_xmit function of one device into the receive queue
1511 * of another device.
1513 * The receiving device may be in another namespace, so
1514 * we have to clear all information in the skb that could
1515 * impact namespace isolation.
1517 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1522 if (unlikely(!is_skb_forwardable(dev, skb))) {
1523 atomic_long_inc(&dev->rx_dropped);
1527 skb_set_dev(skb, dev);
1528 skb->tstamp.tv64 = 0;
1529 skb->pkt_type = PACKET_HOST;
1530 skb->protocol = eth_type_trans(skb, dev);
1531 return netif_rx(skb);
1533 EXPORT_SYMBOL_GPL(dev_forward_skb);
1535 static inline int deliver_skb(struct sk_buff *skb,
1536 struct packet_type *pt_prev,
1537 struct net_device *orig_dev)
1539 atomic_inc(&skb->users);
1540 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1544 * Support routine. Sends outgoing frames to any network
1545 * taps currently in use.
1548 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1550 struct packet_type *ptype;
1551 struct sk_buff *skb2 = NULL;
1552 struct packet_type *pt_prev = NULL;
1555 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1556 /* Never send packets back to the socket
1557 * they originated from - MvS (miquels@drinkel.ow.org)
1559 if ((ptype->dev == dev || !ptype->dev) &&
1560 (ptype->af_packet_priv == NULL ||
1561 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1563 deliver_skb(skb2, pt_prev, skb->dev);
1568 skb2 = skb_clone(skb, GFP_ATOMIC);
1572 net_timestamp_set(skb2);
1574 /* skb->nh should be correctly
1575 set by sender, so that the second statement is
1576 just protection against buggy protocols.
1578 skb_reset_mac_header(skb2);
1580 if (skb_network_header(skb2) < skb2->data ||
1581 skb2->network_header > skb2->tail) {
1582 if (net_ratelimit())
1583 printk(KERN_CRIT "protocol %04x is "
1585 ntohs(skb2->protocol),
1587 skb_reset_network_header(skb2);
1590 skb2->transport_header = skb2->network_header;
1591 skb2->pkt_type = PACKET_OUTGOING;
1596 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1600 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1601 * @dev: Network device
1602 * @txq: number of queues available
1604 * If real_num_tx_queues is changed the tc mappings may no longer be
1605 * valid. To resolve this verify the tc mapping remains valid and if
1606 * not NULL the mapping. With no priorities mapping to this
1607 * offset/count pair it will no longer be used. In the worst case TC0
1608 * is invalid nothing can be done so disable priority mappings. If is
1609 * expected that drivers will fix this mapping if they can before
1610 * calling netif_set_real_num_tx_queues.
1612 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1615 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1617 /* If TC0 is invalidated disable TC mapping */
1618 if (tc->offset + tc->count > txq) {
1619 pr_warning("Number of in use tx queues changed "
1620 "invalidating tc mappings. Priority "
1621 "traffic classification disabled!\n");
1626 /* Invalidated prio to tc mappings set to TC0 */
1627 for (i = 1; i < TC_BITMASK + 1; i++) {
1628 int q = netdev_get_prio_tc_map(dev, i);
1630 tc = &dev->tc_to_txq[q];
1631 if (tc->offset + tc->count > txq) {
1632 pr_warning("Number of in use tx queues "
1633 "changed. Priority %i to tc "
1634 "mapping %i is no longer valid "
1635 "setting map to 0\n",
1637 netdev_set_prio_tc_map(dev, i, 0);
1643 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1644 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1646 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1650 if (txq < 1 || txq > dev->num_tx_queues)
1653 if (dev->reg_state == NETREG_REGISTERED ||
1654 dev->reg_state == NETREG_UNREGISTERING) {
1657 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1663 netif_setup_tc(dev, txq);
1665 if (txq < dev->real_num_tx_queues)
1666 qdisc_reset_all_tx_gt(dev, txq);
1669 dev->real_num_tx_queues = txq;
1672 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1676 * netif_set_real_num_rx_queues - set actual number of RX queues used
1677 * @dev: Network device
1678 * @rxq: Actual number of RX queues
1680 * This must be called either with the rtnl_lock held or before
1681 * registration of the net device. Returns 0 on success, or a
1682 * negative error code. If called before registration, it always
1685 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1689 if (rxq < 1 || rxq > dev->num_rx_queues)
1692 if (dev->reg_state == NETREG_REGISTERED) {
1695 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1701 dev->real_num_rx_queues = rxq;
1704 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1707 static inline void __netif_reschedule(struct Qdisc *q)
1709 struct softnet_data *sd;
1710 unsigned long flags;
1712 local_irq_save(flags);
1713 sd = &__get_cpu_var(softnet_data);
1714 q->next_sched = NULL;
1715 *sd->output_queue_tailp = q;
1716 sd->output_queue_tailp = &q->next_sched;
1717 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1718 local_irq_restore(flags);
1721 void __netif_schedule(struct Qdisc *q)
1723 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1724 __netif_reschedule(q);
1726 EXPORT_SYMBOL(__netif_schedule);
1728 void dev_kfree_skb_irq(struct sk_buff *skb)
1730 if (atomic_dec_and_test(&skb->users)) {
1731 struct softnet_data *sd;
1732 unsigned long flags;
1734 local_irq_save(flags);
1735 sd = &__get_cpu_var(softnet_data);
1736 skb->next = sd->completion_queue;
1737 sd->completion_queue = skb;
1738 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1739 local_irq_restore(flags);
1742 EXPORT_SYMBOL(dev_kfree_skb_irq);
1744 void dev_kfree_skb_any(struct sk_buff *skb)
1746 if (in_irq() || irqs_disabled())
1747 dev_kfree_skb_irq(skb);
1751 EXPORT_SYMBOL(dev_kfree_skb_any);
1755 * netif_device_detach - mark device as removed
1756 * @dev: network device
1758 * Mark device as removed from system and therefore no longer available.
1760 void netif_device_detach(struct net_device *dev)
1762 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1763 netif_running(dev)) {
1764 netif_tx_stop_all_queues(dev);
1767 EXPORT_SYMBOL(netif_device_detach);
1770 * netif_device_attach - mark device as attached
1771 * @dev: network device
1773 * Mark device as attached from system and restart if needed.
1775 void netif_device_attach(struct net_device *dev)
1777 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1778 netif_running(dev)) {
1779 netif_tx_wake_all_queues(dev);
1780 __netdev_watchdog_up(dev);
1783 EXPORT_SYMBOL(netif_device_attach);
1786 * skb_dev_set -- assign a new device to a buffer
1787 * @skb: buffer for the new device
1788 * @dev: network device
1790 * If an skb is owned by a device already, we have to reset
1791 * all data private to the namespace a device belongs to
1792 * before assigning it a new device.
1794 #ifdef CONFIG_NET_NS
1795 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1798 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1801 skb_init_secmark(skb);
1805 skb->ipvs_property = 0;
1806 #ifdef CONFIG_NET_SCHED
1812 EXPORT_SYMBOL(skb_set_dev);
1813 #endif /* CONFIG_NET_NS */
1816 * Invalidate hardware checksum when packet is to be mangled, and
1817 * complete checksum manually on outgoing path.
1819 int skb_checksum_help(struct sk_buff *skb)
1822 int ret = 0, offset;
1824 if (skb->ip_summed == CHECKSUM_COMPLETE)
1825 goto out_set_summed;
1827 if (unlikely(skb_shinfo(skb)->gso_size)) {
1828 /* Let GSO fix up the checksum. */
1829 goto out_set_summed;
1832 offset = skb_checksum_start_offset(skb);
1833 BUG_ON(offset >= skb_headlen(skb));
1834 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1836 offset += skb->csum_offset;
1837 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1839 if (skb_cloned(skb) &&
1840 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1841 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1846 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1848 skb->ip_summed = CHECKSUM_NONE;
1852 EXPORT_SYMBOL(skb_checksum_help);
1855 * skb_gso_segment - Perform segmentation on skb.
1856 * @skb: buffer to segment
1857 * @features: features for the output path (see dev->features)
1859 * This function segments the given skb and returns a list of segments.
1861 * It may return NULL if the skb requires no segmentation. This is
1862 * only possible when GSO is used for verifying header integrity.
1864 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1866 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1867 struct packet_type *ptype;
1868 __be16 type = skb->protocol;
1869 int vlan_depth = ETH_HLEN;
1872 while (type == htons(ETH_P_8021Q)) {
1873 struct vlan_hdr *vh;
1875 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1876 return ERR_PTR(-EINVAL);
1878 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1879 type = vh->h_vlan_encapsulated_proto;
1880 vlan_depth += VLAN_HLEN;
1883 skb_reset_mac_header(skb);
1884 skb->mac_len = skb->network_header - skb->mac_header;
1885 __skb_pull(skb, skb->mac_len);
1887 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1888 struct net_device *dev = skb->dev;
1889 struct ethtool_drvinfo info = {};
1891 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1892 dev->ethtool_ops->get_drvinfo(dev, &info);
1894 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1895 info.driver, dev ? dev->features : 0L,
1896 skb->sk ? skb->sk->sk_route_caps : 0L,
1897 skb->len, skb->data_len, skb->ip_summed);
1899 if (skb_header_cloned(skb) &&
1900 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1901 return ERR_PTR(err);
1905 list_for_each_entry_rcu(ptype,
1906 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1907 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1908 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1909 err = ptype->gso_send_check(skb);
1910 segs = ERR_PTR(err);
1911 if (err || skb_gso_ok(skb, features))
1913 __skb_push(skb, (skb->data -
1914 skb_network_header(skb)));
1916 segs = ptype->gso_segment(skb, features);
1922 __skb_push(skb, skb->data - skb_mac_header(skb));
1926 EXPORT_SYMBOL(skb_gso_segment);
1928 /* Take action when hardware reception checksum errors are detected. */
1930 void netdev_rx_csum_fault(struct net_device *dev)
1932 if (net_ratelimit()) {
1933 printk(KERN_ERR "%s: hw csum failure.\n",
1934 dev ? dev->name : "<unknown>");
1938 EXPORT_SYMBOL(netdev_rx_csum_fault);
1941 /* Actually, we should eliminate this check as soon as we know, that:
1942 * 1. IOMMU is present and allows to map all the memory.
1943 * 2. No high memory really exists on this machine.
1946 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1948 #ifdef CONFIG_HIGHMEM
1950 if (!(dev->features & NETIF_F_HIGHDMA)) {
1951 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1952 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1956 if (PCI_DMA_BUS_IS_PHYS) {
1957 struct device *pdev = dev->dev.parent;
1961 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1962 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1963 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1972 void (*destructor)(struct sk_buff *skb);
1975 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1977 static void dev_gso_skb_destructor(struct sk_buff *skb)
1979 struct dev_gso_cb *cb;
1982 struct sk_buff *nskb = skb->next;
1984 skb->next = nskb->next;
1987 } while (skb->next);
1989 cb = DEV_GSO_CB(skb);
1991 cb->destructor(skb);
1995 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1996 * @skb: buffer to segment
1997 * @features: device features as applicable to this skb
1999 * This function segments the given skb and stores the list of segments
2002 static int dev_gso_segment(struct sk_buff *skb, int features)
2004 struct sk_buff *segs;
2006 segs = skb_gso_segment(skb, features);
2008 /* Verifying header integrity only. */
2013 return PTR_ERR(segs);
2016 DEV_GSO_CB(skb)->destructor = skb->destructor;
2017 skb->destructor = dev_gso_skb_destructor;
2023 * Try to orphan skb early, right before transmission by the device.
2024 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2025 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2027 static inline void skb_orphan_try(struct sk_buff *skb)
2029 struct sock *sk = skb->sk;
2031 if (sk && !skb_shinfo(skb)->tx_flags) {
2032 /* skb_tx_hash() wont be able to get sk.
2033 * We copy sk_hash into skb->rxhash
2036 skb->rxhash = sk->sk_hash;
2041 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2043 return ((features & NETIF_F_GEN_CSUM) ||
2044 ((features & NETIF_F_V4_CSUM) &&
2045 protocol == htons(ETH_P_IP)) ||
2046 ((features & NETIF_F_V6_CSUM) &&
2047 protocol == htons(ETH_P_IPV6)) ||
2048 ((features & NETIF_F_FCOE_CRC) &&
2049 protocol == htons(ETH_P_FCOE)));
2052 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2054 if (!can_checksum_protocol(features, protocol)) {
2055 features &= ~NETIF_F_ALL_CSUM;
2056 features &= ~NETIF_F_SG;
2057 } else if (illegal_highdma(skb->dev, skb)) {
2058 features &= ~NETIF_F_SG;
2064 u32 netif_skb_features(struct sk_buff *skb)
2066 __be16 protocol = skb->protocol;
2067 u32 features = skb->dev->features;
2069 if (protocol == htons(ETH_P_8021Q)) {
2070 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2071 protocol = veh->h_vlan_encapsulated_proto;
2072 } else if (!vlan_tx_tag_present(skb)) {
2073 return harmonize_features(skb, protocol, features);
2076 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2078 if (protocol != htons(ETH_P_8021Q)) {
2079 return harmonize_features(skb, protocol, features);
2081 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2082 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2083 return harmonize_features(skb, protocol, features);
2086 EXPORT_SYMBOL(netif_skb_features);
2089 * Returns true if either:
2090 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2091 * 2. skb is fragmented and the device does not support SG, or if
2092 * at least one of fragments is in highmem and device does not
2093 * support DMA from it.
2095 static inline int skb_needs_linearize(struct sk_buff *skb,
2098 return skb_is_nonlinear(skb) &&
2099 ((skb_has_frag_list(skb) &&
2100 !(features & NETIF_F_FRAGLIST)) ||
2101 (skb_shinfo(skb)->nr_frags &&
2102 !(features & NETIF_F_SG)));
2105 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2106 struct netdev_queue *txq)
2108 const struct net_device_ops *ops = dev->netdev_ops;
2109 int rc = NETDEV_TX_OK;
2110 unsigned int skb_len;
2112 if (likely(!skb->next)) {
2116 * If device doesn't need skb->dst, release it right now while
2117 * its hot in this cpu cache
2119 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2122 if (!list_empty(&ptype_all))
2123 dev_queue_xmit_nit(skb, dev);
2125 skb_orphan_try(skb);
2127 features = netif_skb_features(skb);
2129 if (vlan_tx_tag_present(skb) &&
2130 !(features & NETIF_F_HW_VLAN_TX)) {
2131 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2138 if (netif_needs_gso(skb, features)) {
2139 if (unlikely(dev_gso_segment(skb, features)))
2144 if (skb_needs_linearize(skb, features) &&
2145 __skb_linearize(skb))
2148 /* If packet is not checksummed and device does not
2149 * support checksumming for this protocol, complete
2150 * checksumming here.
2152 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2153 skb_set_transport_header(skb,
2154 skb_checksum_start_offset(skb));
2155 if (!(features & NETIF_F_ALL_CSUM) &&
2156 skb_checksum_help(skb))
2162 rc = ops->ndo_start_xmit(skb, dev);
2163 trace_net_dev_xmit(skb, rc, dev, skb_len);
2164 if (rc == NETDEV_TX_OK)
2165 txq_trans_update(txq);
2171 struct sk_buff *nskb = skb->next;
2173 skb->next = nskb->next;
2177 * If device doesn't need nskb->dst, release it right now while
2178 * its hot in this cpu cache
2180 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2183 skb_len = nskb->len;
2184 rc = ops->ndo_start_xmit(nskb, dev);
2185 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2186 if (unlikely(rc != NETDEV_TX_OK)) {
2187 if (rc & ~NETDEV_TX_MASK)
2188 goto out_kfree_gso_skb;
2189 nskb->next = skb->next;
2193 txq_trans_update(txq);
2194 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2195 return NETDEV_TX_BUSY;
2196 } while (skb->next);
2199 if (likely(skb->next == NULL))
2200 skb->destructor = DEV_GSO_CB(skb)->destructor;
2207 static u32 hashrnd __read_mostly;
2210 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2211 * to be used as a distribution range.
2213 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2214 unsigned int num_tx_queues)
2218 u16 qcount = num_tx_queues;
2220 if (skb_rx_queue_recorded(skb)) {
2221 hash = skb_get_rx_queue(skb);
2222 while (unlikely(hash >= num_tx_queues))
2223 hash -= num_tx_queues;
2228 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2229 qoffset = dev->tc_to_txq[tc].offset;
2230 qcount = dev->tc_to_txq[tc].count;
2233 if (skb->sk && skb->sk->sk_hash)
2234 hash = skb->sk->sk_hash;
2236 hash = (__force u16) skb->protocol ^ skb->rxhash;
2237 hash = jhash_1word(hash, hashrnd);
2239 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2241 EXPORT_SYMBOL(__skb_tx_hash);
2243 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2245 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2246 if (net_ratelimit()) {
2247 pr_warning("%s selects TX queue %d, but "
2248 "real number of TX queues is %d\n",
2249 dev->name, queue_index, dev->real_num_tx_queues);
2256 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2259 struct xps_dev_maps *dev_maps;
2260 struct xps_map *map;
2261 int queue_index = -1;
2264 dev_maps = rcu_dereference(dev->xps_maps);
2266 map = rcu_dereference(
2267 dev_maps->cpu_map[raw_smp_processor_id()]);
2270 queue_index = map->queues[0];
2273 if (skb->sk && skb->sk->sk_hash)
2274 hash = skb->sk->sk_hash;
2276 hash = (__force u16) skb->protocol ^
2278 hash = jhash_1word(hash, hashrnd);
2279 queue_index = map->queues[
2280 ((u64)hash * map->len) >> 32];
2282 if (unlikely(queue_index >= dev->real_num_tx_queues))
2294 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2295 struct sk_buff *skb)
2298 const struct net_device_ops *ops = dev->netdev_ops;
2300 if (dev->real_num_tx_queues == 1)
2302 else if (ops->ndo_select_queue) {
2303 queue_index = ops->ndo_select_queue(dev, skb);
2304 queue_index = dev_cap_txqueue(dev, queue_index);
2306 struct sock *sk = skb->sk;
2307 queue_index = sk_tx_queue_get(sk);
2309 if (queue_index < 0 || skb->ooo_okay ||
2310 queue_index >= dev->real_num_tx_queues) {
2311 int old_index = queue_index;
2313 queue_index = get_xps_queue(dev, skb);
2314 if (queue_index < 0)
2315 queue_index = skb_tx_hash(dev, skb);
2317 if (queue_index != old_index && sk) {
2318 struct dst_entry *dst =
2319 rcu_dereference_check(sk->sk_dst_cache, 1);
2321 if (dst && skb_dst(skb) == dst)
2322 sk_tx_queue_set(sk, queue_index);
2327 skb_set_queue_mapping(skb, queue_index);
2328 return netdev_get_tx_queue(dev, queue_index);
2331 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2332 struct net_device *dev,
2333 struct netdev_queue *txq)
2335 spinlock_t *root_lock = qdisc_lock(q);
2339 qdisc_skb_cb(skb)->pkt_len = skb->len;
2340 qdisc_calculate_pkt_len(skb, q);
2342 * Heuristic to force contended enqueues to serialize on a
2343 * separate lock before trying to get qdisc main lock.
2344 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2345 * and dequeue packets faster.
2347 contended = qdisc_is_running(q);
2348 if (unlikely(contended))
2349 spin_lock(&q->busylock);
2351 spin_lock(root_lock);
2352 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2355 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2356 qdisc_run_begin(q)) {
2358 * This is a work-conserving queue; there are no old skbs
2359 * waiting to be sent out; and the qdisc is not running -
2360 * xmit the skb directly.
2362 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2365 qdisc_bstats_update(q, skb);
2367 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2368 if (unlikely(contended)) {
2369 spin_unlock(&q->busylock);
2376 rc = NET_XMIT_SUCCESS;
2379 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2380 if (qdisc_run_begin(q)) {
2381 if (unlikely(contended)) {
2382 spin_unlock(&q->busylock);
2388 spin_unlock(root_lock);
2389 if (unlikely(contended))
2390 spin_unlock(&q->busylock);
2394 static DEFINE_PER_CPU(int, xmit_recursion);
2395 #define RECURSION_LIMIT 10
2398 * dev_queue_xmit - transmit a buffer
2399 * @skb: buffer to transmit
2401 * Queue a buffer for transmission to a network device. The caller must
2402 * have set the device and priority and built the buffer before calling
2403 * this function. The function can be called from an interrupt.
2405 * A negative errno code is returned on a failure. A success does not
2406 * guarantee the frame will be transmitted as it may be dropped due
2407 * to congestion or traffic shaping.
2409 * -----------------------------------------------------------------------------------
2410 * I notice this method can also return errors from the queue disciplines,
2411 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2414 * Regardless of the return value, the skb is consumed, so it is currently
2415 * difficult to retry a send to this method. (You can bump the ref count
2416 * before sending to hold a reference for retry if you are careful.)
2418 * When calling this method, interrupts MUST be enabled. This is because
2419 * the BH enable code must have IRQs enabled so that it will not deadlock.
2422 int dev_queue_xmit(struct sk_buff *skb)
2424 struct net_device *dev = skb->dev;
2425 struct netdev_queue *txq;
2429 /* Disable soft irqs for various locks below. Also
2430 * stops preemption for RCU.
2434 txq = dev_pick_tx(dev, skb);
2435 q = rcu_dereference_bh(txq->qdisc);
2437 #ifdef CONFIG_NET_CLS_ACT
2438 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2440 trace_net_dev_queue(skb);
2442 rc = __dev_xmit_skb(skb, q, dev, txq);
2446 /* The device has no queue. Common case for software devices:
2447 loopback, all the sorts of tunnels...
2449 Really, it is unlikely that netif_tx_lock protection is necessary
2450 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2452 However, it is possible, that they rely on protection
2455 Check this and shot the lock. It is not prone from deadlocks.
2456 Either shot noqueue qdisc, it is even simpler 8)
2458 if (dev->flags & IFF_UP) {
2459 int cpu = smp_processor_id(); /* ok because BHs are off */
2461 if (txq->xmit_lock_owner != cpu) {
2463 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2464 goto recursion_alert;
2466 HARD_TX_LOCK(dev, txq, cpu);
2468 if (!netif_tx_queue_stopped(txq)) {
2469 __this_cpu_inc(xmit_recursion);
2470 rc = dev_hard_start_xmit(skb, dev, txq);
2471 __this_cpu_dec(xmit_recursion);
2472 if (dev_xmit_complete(rc)) {
2473 HARD_TX_UNLOCK(dev, txq);
2477 HARD_TX_UNLOCK(dev, txq);
2478 if (net_ratelimit())
2479 printk(KERN_CRIT "Virtual device %s asks to "
2480 "queue packet!\n", dev->name);
2482 /* Recursion is detected! It is possible,
2486 if (net_ratelimit())
2487 printk(KERN_CRIT "Dead loop on virtual device "
2488 "%s, fix it urgently!\n", dev->name);
2493 rcu_read_unlock_bh();
2498 rcu_read_unlock_bh();
2501 EXPORT_SYMBOL(dev_queue_xmit);
2504 /*=======================================================================
2506 =======================================================================*/
2508 int netdev_max_backlog __read_mostly = 1000;
2509 int netdev_tstamp_prequeue __read_mostly = 1;
2510 int netdev_budget __read_mostly = 300;
2511 int weight_p __read_mostly = 64; /* old backlog weight */
2513 /* Called with irq disabled */
2514 static inline void ____napi_schedule(struct softnet_data *sd,
2515 struct napi_struct *napi)
2517 list_add_tail(&napi->poll_list, &sd->poll_list);
2518 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2522 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2523 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2524 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2525 * if hash is a canonical 4-tuple hash over transport ports.
2527 void __skb_get_rxhash(struct sk_buff *skb)
2529 int nhoff, hash = 0, poff;
2530 const struct ipv6hdr *ip6;
2531 const struct iphdr *ip;
2532 const struct vlan_hdr *vlan;
2541 nhoff = skb_network_offset(skb);
2542 proto = skb->protocol;
2546 case __constant_htons(ETH_P_IP):
2547 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2550 ip = (const struct iphdr *) (skb->data + nhoff);
2551 if (ip_is_fragment(ip))
2554 ip_proto = ip->protocol;
2555 addr1 = (__force u32) ip->saddr;
2556 addr2 = (__force u32) ip->daddr;
2557 nhoff += ip->ihl * 4;
2559 case __constant_htons(ETH_P_IPV6):
2560 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2563 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2564 ip_proto = ip6->nexthdr;
2565 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2566 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2569 case __constant_htons(ETH_P_8021Q):
2570 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2572 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2573 proto = vlan->h_vlan_encapsulated_proto;
2574 nhoff += sizeof(*vlan);
2582 if (pskb_may_pull(skb, nhoff + 16)) {
2583 u8 *h = skb->data + nhoff;
2584 __be16 flags = *(__be16 *)h;
2587 * Only look inside GRE if version zero and no
2590 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2591 proto = *(__be16 *)(h + 2);
2593 if (flags & GRE_CSUM)
2595 if (flags & GRE_KEY)
2597 if (flags & GRE_SEQ)
2608 poff = proto_ports_offset(ip_proto);
2611 if (pskb_may_pull(skb, nhoff + 4)) {
2612 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2613 if (ports.v16[1] < ports.v16[0])
2614 swap(ports.v16[0], ports.v16[1]);
2619 /* get a consistent hash (same value on both flow directions) */
2623 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2630 EXPORT_SYMBOL(__skb_get_rxhash);
2634 /* One global table that all flow-based protocols share. */
2635 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2636 EXPORT_SYMBOL(rps_sock_flow_table);
2638 static struct rps_dev_flow *
2639 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2640 struct rps_dev_flow *rflow, u16 next_cpu)
2644 tcpu = rflow->cpu = next_cpu;
2645 if (tcpu != RPS_NO_CPU) {
2646 #ifdef CONFIG_RFS_ACCEL
2647 struct netdev_rx_queue *rxqueue;
2648 struct rps_dev_flow_table *flow_table;
2649 struct rps_dev_flow *old_rflow;
2654 /* Should we steer this flow to a different hardware queue? */
2655 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2656 !(dev->features & NETIF_F_NTUPLE))
2658 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2659 if (rxq_index == skb_get_rx_queue(skb))
2662 rxqueue = dev->_rx + rxq_index;
2663 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2666 flow_id = skb->rxhash & flow_table->mask;
2667 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2668 rxq_index, flow_id);
2672 rflow = &flow_table->flows[flow_id];
2673 rflow->cpu = next_cpu;
2675 if (old_rflow->filter == rflow->filter)
2676 old_rflow->filter = RPS_NO_FILTER;
2680 per_cpu(softnet_data, tcpu).input_queue_head;
2687 * get_rps_cpu is called from netif_receive_skb and returns the target
2688 * CPU from the RPS map of the receiving queue for a given skb.
2689 * rcu_read_lock must be held on entry.
2691 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2692 struct rps_dev_flow **rflowp)
2694 struct netdev_rx_queue *rxqueue;
2695 struct rps_map *map;
2696 struct rps_dev_flow_table *flow_table;
2697 struct rps_sock_flow_table *sock_flow_table;
2701 if (skb_rx_queue_recorded(skb)) {
2702 u16 index = skb_get_rx_queue(skb);
2703 if (unlikely(index >= dev->real_num_rx_queues)) {
2704 WARN_ONCE(dev->real_num_rx_queues > 1,
2705 "%s received packet on queue %u, but number "
2706 "of RX queues is %u\n",
2707 dev->name, index, dev->real_num_rx_queues);
2710 rxqueue = dev->_rx + index;
2714 map = rcu_dereference(rxqueue->rps_map);
2716 if (map->len == 1 &&
2717 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2718 tcpu = map->cpus[0];
2719 if (cpu_online(tcpu))
2723 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2727 skb_reset_network_header(skb);
2728 if (!skb_get_rxhash(skb))
2731 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2732 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2733 if (flow_table && sock_flow_table) {
2735 struct rps_dev_flow *rflow;
2737 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2740 next_cpu = sock_flow_table->ents[skb->rxhash &
2741 sock_flow_table->mask];
2744 * If the desired CPU (where last recvmsg was done) is
2745 * different from current CPU (one in the rx-queue flow
2746 * table entry), switch if one of the following holds:
2747 * - Current CPU is unset (equal to RPS_NO_CPU).
2748 * - Current CPU is offline.
2749 * - The current CPU's queue tail has advanced beyond the
2750 * last packet that was enqueued using this table entry.
2751 * This guarantees that all previous packets for the flow
2752 * have been dequeued, thus preserving in order delivery.
2754 if (unlikely(tcpu != next_cpu) &&
2755 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2756 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2757 rflow->last_qtail)) >= 0))
2758 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2760 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2768 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2770 if (cpu_online(tcpu)) {
2780 #ifdef CONFIG_RFS_ACCEL
2783 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2784 * @dev: Device on which the filter was set
2785 * @rxq_index: RX queue index
2786 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2787 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2789 * Drivers that implement ndo_rx_flow_steer() should periodically call
2790 * this function for each installed filter and remove the filters for
2791 * which it returns %true.
2793 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2794 u32 flow_id, u16 filter_id)
2796 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2797 struct rps_dev_flow_table *flow_table;
2798 struct rps_dev_flow *rflow;
2803 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2804 if (flow_table && flow_id <= flow_table->mask) {
2805 rflow = &flow_table->flows[flow_id];
2806 cpu = ACCESS_ONCE(rflow->cpu);
2807 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2808 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2809 rflow->last_qtail) <
2810 (int)(10 * flow_table->mask)))
2816 EXPORT_SYMBOL(rps_may_expire_flow);
2818 #endif /* CONFIG_RFS_ACCEL */
2820 /* Called from hardirq (IPI) context */
2821 static void rps_trigger_softirq(void *data)
2823 struct softnet_data *sd = data;
2825 ____napi_schedule(sd, &sd->backlog);
2829 #endif /* CONFIG_RPS */
2832 * Check if this softnet_data structure is another cpu one
2833 * If yes, queue it to our IPI list and return 1
2836 static int rps_ipi_queued(struct softnet_data *sd)
2839 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2842 sd->rps_ipi_next = mysd->rps_ipi_list;
2843 mysd->rps_ipi_list = sd;
2845 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2848 #endif /* CONFIG_RPS */
2853 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2854 * queue (may be a remote CPU queue).
2856 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2857 unsigned int *qtail)
2859 struct softnet_data *sd;
2860 unsigned long flags;
2862 sd = &per_cpu(softnet_data, cpu);
2864 local_irq_save(flags);
2867 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2868 if (skb_queue_len(&sd->input_pkt_queue)) {
2870 __skb_queue_tail(&sd->input_pkt_queue, skb);
2871 input_queue_tail_incr_save(sd, qtail);
2873 local_irq_restore(flags);
2874 return NET_RX_SUCCESS;
2877 /* Schedule NAPI for backlog device
2878 * We can use non atomic operation since we own the queue lock
2880 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2881 if (!rps_ipi_queued(sd))
2882 ____napi_schedule(sd, &sd->backlog);
2890 local_irq_restore(flags);
2892 atomic_long_inc(&skb->dev->rx_dropped);
2898 * netif_rx - post buffer to the network code
2899 * @skb: buffer to post
2901 * This function receives a packet from a device driver and queues it for
2902 * the upper (protocol) levels to process. It always succeeds. The buffer
2903 * may be dropped during processing for congestion control or by the
2907 * NET_RX_SUCCESS (no congestion)
2908 * NET_RX_DROP (packet was dropped)
2912 int netif_rx(struct sk_buff *skb)
2916 /* if netpoll wants it, pretend we never saw it */
2917 if (netpoll_rx(skb))
2920 if (netdev_tstamp_prequeue)
2921 net_timestamp_check(skb);
2923 trace_netif_rx(skb);
2926 struct rps_dev_flow voidflow, *rflow = &voidflow;
2932 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2934 cpu = smp_processor_id();
2936 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2944 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2950 EXPORT_SYMBOL(netif_rx);
2952 int netif_rx_ni(struct sk_buff *skb)
2957 err = netif_rx(skb);
2958 if (local_softirq_pending())
2964 EXPORT_SYMBOL(netif_rx_ni);
2966 static void net_tx_action(struct softirq_action *h)
2968 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2970 if (sd->completion_queue) {
2971 struct sk_buff *clist;
2973 local_irq_disable();
2974 clist = sd->completion_queue;
2975 sd->completion_queue = NULL;
2979 struct sk_buff *skb = clist;
2980 clist = clist->next;
2982 WARN_ON(atomic_read(&skb->users));
2983 trace_kfree_skb(skb, net_tx_action);
2988 if (sd->output_queue) {
2991 local_irq_disable();
2992 head = sd->output_queue;
2993 sd->output_queue = NULL;
2994 sd->output_queue_tailp = &sd->output_queue;
2998 struct Qdisc *q = head;
2999 spinlock_t *root_lock;
3001 head = head->next_sched;
3003 root_lock = qdisc_lock(q);
3004 if (spin_trylock(root_lock)) {
3005 smp_mb__before_clear_bit();
3006 clear_bit(__QDISC_STATE_SCHED,
3009 spin_unlock(root_lock);
3011 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3013 __netif_reschedule(q);
3015 smp_mb__before_clear_bit();
3016 clear_bit(__QDISC_STATE_SCHED,
3024 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3025 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3026 /* This hook is defined here for ATM LANE */
3027 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3028 unsigned char *addr) __read_mostly;
3029 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3032 #ifdef CONFIG_NET_CLS_ACT
3033 /* TODO: Maybe we should just force sch_ingress to be compiled in
3034 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3035 * a compare and 2 stores extra right now if we dont have it on
3036 * but have CONFIG_NET_CLS_ACT
3037 * NOTE: This doesn't stop any functionality; if you dont have
3038 * the ingress scheduler, you just can't add policies on ingress.
3041 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3043 struct net_device *dev = skb->dev;
3044 u32 ttl = G_TC_RTTL(skb->tc_verd);
3045 int result = TC_ACT_OK;
3048 if (unlikely(MAX_RED_LOOP < ttl++)) {
3049 if (net_ratelimit())
3050 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3051 skb->skb_iif, dev->ifindex);
3055 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3056 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3059 if (q != &noop_qdisc) {
3060 spin_lock(qdisc_lock(q));
3061 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3062 result = qdisc_enqueue_root(skb, q);
3063 spin_unlock(qdisc_lock(q));
3069 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3070 struct packet_type **pt_prev,
3071 int *ret, struct net_device *orig_dev)
3073 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3075 if (!rxq || rxq->qdisc == &noop_qdisc)
3079 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3083 switch (ing_filter(skb, rxq)) {
3097 * netdev_rx_handler_register - register receive handler
3098 * @dev: device to register a handler for
3099 * @rx_handler: receive handler to register
3100 * @rx_handler_data: data pointer that is used by rx handler
3102 * Register a receive hander for a device. This handler will then be
3103 * called from __netif_receive_skb. A negative errno code is returned
3106 * The caller must hold the rtnl_mutex.
3108 * For a general description of rx_handler, see enum rx_handler_result.
3110 int netdev_rx_handler_register(struct net_device *dev,
3111 rx_handler_func_t *rx_handler,
3112 void *rx_handler_data)
3116 if (dev->rx_handler)
3119 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3120 rcu_assign_pointer(dev->rx_handler, rx_handler);
3124 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3127 * netdev_rx_handler_unregister - unregister receive handler
3128 * @dev: device to unregister a handler from
3130 * Unregister a receive hander from a device.
3132 * The caller must hold the rtnl_mutex.
3134 void netdev_rx_handler_unregister(struct net_device *dev)
3138 RCU_INIT_POINTER(dev->rx_handler, NULL);
3139 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3141 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3143 static int __netif_receive_skb(struct sk_buff *skb)
3145 struct packet_type *ptype, *pt_prev;
3146 rx_handler_func_t *rx_handler;
3147 struct net_device *orig_dev;
3148 struct net_device *null_or_dev;
3149 bool deliver_exact = false;
3150 int ret = NET_RX_DROP;
3153 if (!netdev_tstamp_prequeue)
3154 net_timestamp_check(skb);
3156 trace_netif_receive_skb(skb);
3158 /* if we've gotten here through NAPI, check netpoll */
3159 if (netpoll_receive_skb(skb))
3163 skb->skb_iif = skb->dev->ifindex;
3164 orig_dev = skb->dev;
3166 skb_reset_network_header(skb);
3167 skb_reset_transport_header(skb);
3168 skb_reset_mac_len(skb);
3176 __this_cpu_inc(softnet_data.processed);
3178 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3179 skb = vlan_untag(skb);
3184 #ifdef CONFIG_NET_CLS_ACT
3185 if (skb->tc_verd & TC_NCLS) {
3186 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3191 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3192 if (!ptype->dev || ptype->dev == skb->dev) {
3194 ret = deliver_skb(skb, pt_prev, orig_dev);
3199 #ifdef CONFIG_NET_CLS_ACT
3200 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3206 rx_handler = rcu_dereference(skb->dev->rx_handler);
3209 ret = deliver_skb(skb, pt_prev, orig_dev);
3212 switch (rx_handler(&skb)) {
3213 case RX_HANDLER_CONSUMED:
3215 case RX_HANDLER_ANOTHER:
3217 case RX_HANDLER_EXACT:
3218 deliver_exact = true;
3219 case RX_HANDLER_PASS:
3226 if (vlan_tx_tag_present(skb)) {
3228 ret = deliver_skb(skb, pt_prev, orig_dev);
3231 if (vlan_do_receive(&skb)) {
3232 ret = __netif_receive_skb(skb);
3234 } else if (unlikely(!skb))
3238 /* deliver only exact match when indicated */
3239 null_or_dev = deliver_exact ? skb->dev : NULL;
3241 type = skb->protocol;
3242 list_for_each_entry_rcu(ptype,
3243 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3244 if (ptype->type == type &&
3245 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3246 ptype->dev == orig_dev)) {
3248 ret = deliver_skb(skb, pt_prev, orig_dev);
3254 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3256 atomic_long_inc(&skb->dev->rx_dropped);
3258 /* Jamal, now you will not able to escape explaining
3259 * me how you were going to use this. :-)
3270 * netif_receive_skb - process receive buffer from network
3271 * @skb: buffer to process
3273 * netif_receive_skb() is the main receive data processing function.
3274 * It always succeeds. The buffer may be dropped during processing
3275 * for congestion control or by the protocol layers.
3277 * This function may only be called from softirq context and interrupts
3278 * should be enabled.
3280 * Return values (usually ignored):
3281 * NET_RX_SUCCESS: no congestion
3282 * NET_RX_DROP: packet was dropped
3284 int netif_receive_skb(struct sk_buff *skb)
3286 if (netdev_tstamp_prequeue)
3287 net_timestamp_check(skb);
3289 if (skb_defer_rx_timestamp(skb))
3290 return NET_RX_SUCCESS;
3294 struct rps_dev_flow voidflow, *rflow = &voidflow;
3299 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3302 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3306 ret = __netif_receive_skb(skb);
3312 return __netif_receive_skb(skb);
3315 EXPORT_SYMBOL(netif_receive_skb);
3317 /* Network device is going away, flush any packets still pending
3318 * Called with irqs disabled.
3320 static void flush_backlog(void *arg)
3322 struct net_device *dev = arg;
3323 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3324 struct sk_buff *skb, *tmp;
3327 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3328 if (skb->dev == dev) {
3329 __skb_unlink(skb, &sd->input_pkt_queue);
3331 input_queue_head_incr(sd);
3336 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3337 if (skb->dev == dev) {
3338 __skb_unlink(skb, &sd->process_queue);
3340 input_queue_head_incr(sd);
3345 static int napi_gro_complete(struct sk_buff *skb)
3347 struct packet_type *ptype;
3348 __be16 type = skb->protocol;
3349 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3352 if (NAPI_GRO_CB(skb)->count == 1) {
3353 skb_shinfo(skb)->gso_size = 0;
3358 list_for_each_entry_rcu(ptype, head, list) {
3359 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3362 err = ptype->gro_complete(skb);
3368 WARN_ON(&ptype->list == head);
3370 return NET_RX_SUCCESS;
3374 return netif_receive_skb(skb);
3377 inline void napi_gro_flush(struct napi_struct *napi)
3379 struct sk_buff *skb, *next;
3381 for (skb = napi->gro_list; skb; skb = next) {
3384 napi_gro_complete(skb);
3387 napi->gro_count = 0;
3388 napi->gro_list = NULL;
3390 EXPORT_SYMBOL(napi_gro_flush);
3392 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3394 struct sk_buff **pp = NULL;
3395 struct packet_type *ptype;
3396 __be16 type = skb->protocol;
3397 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3400 enum gro_result ret;
3402 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3405 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3409 list_for_each_entry_rcu(ptype, head, list) {
3410 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3413 skb_set_network_header(skb, skb_gro_offset(skb));
3414 mac_len = skb->network_header - skb->mac_header;
3415 skb->mac_len = mac_len;
3416 NAPI_GRO_CB(skb)->same_flow = 0;
3417 NAPI_GRO_CB(skb)->flush = 0;
3418 NAPI_GRO_CB(skb)->free = 0;
3420 pp = ptype->gro_receive(&napi->gro_list, skb);
3425 if (&ptype->list == head)
3428 same_flow = NAPI_GRO_CB(skb)->same_flow;
3429 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3432 struct sk_buff *nskb = *pp;
3436 napi_gro_complete(nskb);
3443 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3447 NAPI_GRO_CB(skb)->count = 1;
3448 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3449 skb->next = napi->gro_list;
3450 napi->gro_list = skb;
3454 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3455 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3457 BUG_ON(skb->end - skb->tail < grow);
3459 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3462 skb->data_len -= grow;
3464 skb_shinfo(skb)->frags[0].page_offset += grow;
3465 skb_shinfo(skb)->frags[0].size -= grow;
3467 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3468 put_page(skb_shinfo(skb)->frags[0].page);
3469 memmove(skb_shinfo(skb)->frags,
3470 skb_shinfo(skb)->frags + 1,
3471 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3482 EXPORT_SYMBOL(dev_gro_receive);
3484 static inline gro_result_t
3485 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3489 for (p = napi->gro_list; p; p = p->next) {
3490 unsigned long diffs;
3492 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3493 diffs |= p->vlan_tci ^ skb->vlan_tci;
3494 diffs |= compare_ether_header(skb_mac_header(p),
3495 skb_gro_mac_header(skb));
3496 NAPI_GRO_CB(p)->same_flow = !diffs;
3497 NAPI_GRO_CB(p)->flush = 0;
3500 return dev_gro_receive(napi, skb);
3503 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3507 if (netif_receive_skb(skb))
3512 case GRO_MERGED_FREE:
3523 EXPORT_SYMBOL(napi_skb_finish);
3525 void skb_gro_reset_offset(struct sk_buff *skb)
3527 NAPI_GRO_CB(skb)->data_offset = 0;
3528 NAPI_GRO_CB(skb)->frag0 = NULL;
3529 NAPI_GRO_CB(skb)->frag0_len = 0;
3531 if (skb->mac_header == skb->tail &&
3532 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3533 NAPI_GRO_CB(skb)->frag0 =
3534 page_address(skb_shinfo(skb)->frags[0].page) +
3535 skb_shinfo(skb)->frags[0].page_offset;
3536 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3539 EXPORT_SYMBOL(skb_gro_reset_offset);
3541 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3543 skb_gro_reset_offset(skb);
3545 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3547 EXPORT_SYMBOL(napi_gro_receive);
3549 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3551 __skb_pull(skb, skb_headlen(skb));
3552 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3554 skb->dev = napi->dev;
3560 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3562 struct sk_buff *skb = napi->skb;
3565 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3571 EXPORT_SYMBOL(napi_get_frags);
3573 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3579 skb->protocol = eth_type_trans(skb, skb->dev);
3581 if (ret == GRO_HELD)
3582 skb_gro_pull(skb, -ETH_HLEN);
3583 else if (netif_receive_skb(skb))
3588 case GRO_MERGED_FREE:
3589 napi_reuse_skb(napi, skb);
3598 EXPORT_SYMBOL(napi_frags_finish);
3600 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3602 struct sk_buff *skb = napi->skb;
3609 skb_reset_mac_header(skb);
3610 skb_gro_reset_offset(skb);
3612 off = skb_gro_offset(skb);
3613 hlen = off + sizeof(*eth);
3614 eth = skb_gro_header_fast(skb, off);
3615 if (skb_gro_header_hard(skb, hlen)) {
3616 eth = skb_gro_header_slow(skb, hlen, off);
3617 if (unlikely(!eth)) {
3618 napi_reuse_skb(napi, skb);
3624 skb_gro_pull(skb, sizeof(*eth));
3627 * This works because the only protocols we care about don't require
3628 * special handling. We'll fix it up properly at the end.
3630 skb->protocol = eth->h_proto;
3635 EXPORT_SYMBOL(napi_frags_skb);
3637 gro_result_t napi_gro_frags(struct napi_struct *napi)
3639 struct sk_buff *skb = napi_frags_skb(napi);
3644 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3646 EXPORT_SYMBOL(napi_gro_frags);
3649 * net_rps_action sends any pending IPI's for rps.
3650 * Note: called with local irq disabled, but exits with local irq enabled.
3652 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3655 struct softnet_data *remsd = sd->rps_ipi_list;
3658 sd->rps_ipi_list = NULL;
3662 /* Send pending IPI's to kick RPS processing on remote cpus. */
3664 struct softnet_data *next = remsd->rps_ipi_next;
3666 if (cpu_online(remsd->cpu))
3667 __smp_call_function_single(remsd->cpu,
3676 static int process_backlog(struct napi_struct *napi, int quota)
3679 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3682 /* Check if we have pending ipi, its better to send them now,
3683 * not waiting net_rx_action() end.
3685 if (sd->rps_ipi_list) {
3686 local_irq_disable();
3687 net_rps_action_and_irq_enable(sd);
3690 napi->weight = weight_p;
3691 local_irq_disable();
3692 while (work < quota) {
3693 struct sk_buff *skb;
3696 while ((skb = __skb_dequeue(&sd->process_queue))) {
3698 __netif_receive_skb(skb);
3699 local_irq_disable();
3700 input_queue_head_incr(sd);
3701 if (++work >= quota) {
3708 qlen = skb_queue_len(&sd->input_pkt_queue);
3710 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3711 &sd->process_queue);
3713 if (qlen < quota - work) {
3715 * Inline a custom version of __napi_complete().
3716 * only current cpu owns and manipulates this napi,
3717 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3718 * we can use a plain write instead of clear_bit(),
3719 * and we dont need an smp_mb() memory barrier.
3721 list_del(&napi->poll_list);
3724 quota = work + qlen;
3734 * __napi_schedule - schedule for receive
3735 * @n: entry to schedule
3737 * The entry's receive function will be scheduled to run
3739 void __napi_schedule(struct napi_struct *n)
3741 unsigned long flags;
3743 local_irq_save(flags);
3744 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3745 local_irq_restore(flags);
3747 EXPORT_SYMBOL(__napi_schedule);
3749 void __napi_complete(struct napi_struct *n)
3751 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3752 BUG_ON(n->gro_list);
3754 list_del(&n->poll_list);
3755 smp_mb__before_clear_bit();
3756 clear_bit(NAPI_STATE_SCHED, &n->state);
3758 EXPORT_SYMBOL(__napi_complete);
3760 void napi_complete(struct napi_struct *n)
3762 unsigned long flags;
3765 * don't let napi dequeue from the cpu poll list
3766 * just in case its running on a different cpu
3768 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3772 local_irq_save(flags);
3774 local_irq_restore(flags);
3776 EXPORT_SYMBOL(napi_complete);
3778 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3779 int (*poll)(struct napi_struct *, int), int weight)
3781 INIT_LIST_HEAD(&napi->poll_list);
3782 napi->gro_count = 0;
3783 napi->gro_list = NULL;
3786 napi->weight = weight;
3787 list_add(&napi->dev_list, &dev->napi_list);
3789 #ifdef CONFIG_NETPOLL
3790 spin_lock_init(&napi->poll_lock);
3791 napi->poll_owner = -1;
3793 set_bit(NAPI_STATE_SCHED, &napi->state);
3795 EXPORT_SYMBOL(netif_napi_add);
3797 void netif_napi_del(struct napi_struct *napi)
3799 struct sk_buff *skb, *next;
3801 list_del_init(&napi->dev_list);
3802 napi_free_frags(napi);
3804 for (skb = napi->gro_list; skb; skb = next) {
3810 napi->gro_list = NULL;
3811 napi->gro_count = 0;
3813 EXPORT_SYMBOL(netif_napi_del);
3815 static void net_rx_action(struct softirq_action *h)
3817 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3818 unsigned long time_limit = jiffies + 2;
3819 int budget = netdev_budget;
3822 local_irq_disable();
3824 while (!list_empty(&sd->poll_list)) {
3825 struct napi_struct *n;
3828 /* If softirq window is exhuasted then punt.
3829 * Allow this to run for 2 jiffies since which will allow
3830 * an average latency of 1.5/HZ.
3832 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3837 /* Even though interrupts have been re-enabled, this
3838 * access is safe because interrupts can only add new
3839 * entries to the tail of this list, and only ->poll()
3840 * calls can remove this head entry from the list.
3842 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3844 have = netpoll_poll_lock(n);
3848 /* This NAPI_STATE_SCHED test is for avoiding a race
3849 * with netpoll's poll_napi(). Only the entity which
3850 * obtains the lock and sees NAPI_STATE_SCHED set will
3851 * actually make the ->poll() call. Therefore we avoid
3852 * accidentally calling ->poll() when NAPI is not scheduled.
3855 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3856 work = n->poll(n, weight);
3860 WARN_ON_ONCE(work > weight);
3864 local_irq_disable();
3866 /* Drivers must not modify the NAPI state if they
3867 * consume the entire weight. In such cases this code
3868 * still "owns" the NAPI instance and therefore can
3869 * move the instance around on the list at-will.
3871 if (unlikely(work == weight)) {
3872 if (unlikely(napi_disable_pending(n))) {
3875 local_irq_disable();
3877 list_move_tail(&n->poll_list, &sd->poll_list);
3880 netpoll_poll_unlock(have);
3883 net_rps_action_and_irq_enable(sd);
3885 #ifdef CONFIG_NET_DMA
3887 * There may not be any more sk_buffs coming right now, so push
3888 * any pending DMA copies to hardware
3890 dma_issue_pending_all();
3897 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3901 static gifconf_func_t *gifconf_list[NPROTO];
3904 * register_gifconf - register a SIOCGIF handler
3905 * @family: Address family
3906 * @gifconf: Function handler
3908 * Register protocol dependent address dumping routines. The handler
3909 * that is passed must not be freed or reused until it has been replaced
3910 * by another handler.
3912 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3914 if (family >= NPROTO)
3916 gifconf_list[family] = gifconf;
3919 EXPORT_SYMBOL(register_gifconf);
3923 * Map an interface index to its name (SIOCGIFNAME)
3927 * We need this ioctl for efficient implementation of the
3928 * if_indextoname() function required by the IPv6 API. Without
3929 * it, we would have to search all the interfaces to find a
3933 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3935 struct net_device *dev;
3939 * Fetch the caller's info block.
3942 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3946 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3952 strcpy(ifr.ifr_name, dev->name);
3955 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3961 * Perform a SIOCGIFCONF call. This structure will change
3962 * size eventually, and there is nothing I can do about it.
3963 * Thus we will need a 'compatibility mode'.
3966 static int dev_ifconf(struct net *net, char __user *arg)
3969 struct net_device *dev;
3976 * Fetch the caller's info block.
3979 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3986 * Loop over the interfaces, and write an info block for each.
3990 for_each_netdev(net, dev) {
3991 for (i = 0; i < NPROTO; i++) {
3992 if (gifconf_list[i]) {
3995 done = gifconf_list[i](dev, NULL, 0);
3997 done = gifconf_list[i](dev, pos + total,
4007 * All done. Write the updated control block back to the caller.
4009 ifc.ifc_len = total;
4012 * Both BSD and Solaris return 0 here, so we do too.
4014 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4017 #ifdef CONFIG_PROC_FS
4019 * This is invoked by the /proc filesystem handler to display a device
4022 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4025 struct net *net = seq_file_net(seq);
4027 struct net_device *dev;
4031 return SEQ_START_TOKEN;
4034 for_each_netdev_rcu(net, dev)
4041 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4043 struct net_device *dev = v;
4045 if (v == SEQ_START_TOKEN)
4046 dev = first_net_device_rcu(seq_file_net(seq));
4048 dev = next_net_device_rcu(dev);
4054 void dev_seq_stop(struct seq_file *seq, void *v)
4060 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4062 struct rtnl_link_stats64 temp;
4063 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4065 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4066 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4067 dev->name, stats->rx_bytes, stats->rx_packets,
4069 stats->rx_dropped + stats->rx_missed_errors,
4070 stats->rx_fifo_errors,
4071 stats->rx_length_errors + stats->rx_over_errors +
4072 stats->rx_crc_errors + stats->rx_frame_errors,
4073 stats->rx_compressed, stats->multicast,
4074 stats->tx_bytes, stats->tx_packets,
4075 stats->tx_errors, stats->tx_dropped,
4076 stats->tx_fifo_errors, stats->collisions,
4077 stats->tx_carrier_errors +
4078 stats->tx_aborted_errors +
4079 stats->tx_window_errors +
4080 stats->tx_heartbeat_errors,
4081 stats->tx_compressed);
4085 * Called from the PROCfs module. This now uses the new arbitrary sized
4086 * /proc/net interface to create /proc/net/dev
4088 static int dev_seq_show(struct seq_file *seq, void *v)
4090 if (v == SEQ_START_TOKEN)
4091 seq_puts(seq, "Inter-| Receive "
4093 " face |bytes packets errs drop fifo frame "
4094 "compressed multicast|bytes packets errs "
4095 "drop fifo colls carrier compressed\n");
4097 dev_seq_printf_stats(seq, v);
4101 static struct softnet_data *softnet_get_online(loff_t *pos)
4103 struct softnet_data *sd = NULL;
4105 while (*pos < nr_cpu_ids)
4106 if (cpu_online(*pos)) {
4107 sd = &per_cpu(softnet_data, *pos);
4114 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4116 return softnet_get_online(pos);
4119 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4122 return softnet_get_online(pos);
4125 static void softnet_seq_stop(struct seq_file *seq, void *v)
4129 static int softnet_seq_show(struct seq_file *seq, void *v)
4131 struct softnet_data *sd = v;
4133 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4134 sd->processed, sd->dropped, sd->time_squeeze, 0,
4135 0, 0, 0, 0, /* was fastroute */
4136 sd->cpu_collision, sd->received_rps);
4140 static const struct seq_operations dev_seq_ops = {
4141 .start = dev_seq_start,
4142 .next = dev_seq_next,
4143 .stop = dev_seq_stop,
4144 .show = dev_seq_show,
4147 static int dev_seq_open(struct inode *inode, struct file *file)
4149 return seq_open_net(inode, file, &dev_seq_ops,
4150 sizeof(struct seq_net_private));
4153 static const struct file_operations dev_seq_fops = {
4154 .owner = THIS_MODULE,
4155 .open = dev_seq_open,
4157 .llseek = seq_lseek,
4158 .release = seq_release_net,
4161 static const struct seq_operations softnet_seq_ops = {
4162 .start = softnet_seq_start,
4163 .next = softnet_seq_next,
4164 .stop = softnet_seq_stop,
4165 .show = softnet_seq_show,
4168 static int softnet_seq_open(struct inode *inode, struct file *file)
4170 return seq_open(file, &softnet_seq_ops);
4173 static const struct file_operations softnet_seq_fops = {
4174 .owner = THIS_MODULE,
4175 .open = softnet_seq_open,
4177 .llseek = seq_lseek,
4178 .release = seq_release,
4181 static void *ptype_get_idx(loff_t pos)
4183 struct packet_type *pt = NULL;
4187 list_for_each_entry_rcu(pt, &ptype_all, list) {
4193 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4194 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4203 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4207 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4210 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4212 struct packet_type *pt;
4213 struct list_head *nxt;
4217 if (v == SEQ_START_TOKEN)
4218 return ptype_get_idx(0);
4221 nxt = pt->list.next;
4222 if (pt->type == htons(ETH_P_ALL)) {
4223 if (nxt != &ptype_all)
4226 nxt = ptype_base[0].next;
4228 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4230 while (nxt == &ptype_base[hash]) {
4231 if (++hash >= PTYPE_HASH_SIZE)
4233 nxt = ptype_base[hash].next;
4236 return list_entry(nxt, struct packet_type, list);
4239 static void ptype_seq_stop(struct seq_file *seq, void *v)
4245 static int ptype_seq_show(struct seq_file *seq, void *v)
4247 struct packet_type *pt = v;
4249 if (v == SEQ_START_TOKEN)
4250 seq_puts(seq, "Type Device Function\n");
4251 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4252 if (pt->type == htons(ETH_P_ALL))
4253 seq_puts(seq, "ALL ");
4255 seq_printf(seq, "%04x", ntohs(pt->type));
4257 seq_printf(seq, " %-8s %pF\n",
4258 pt->dev ? pt->dev->name : "", pt->func);
4264 static const struct seq_operations ptype_seq_ops = {
4265 .start = ptype_seq_start,
4266 .next = ptype_seq_next,
4267 .stop = ptype_seq_stop,
4268 .show = ptype_seq_show,
4271 static int ptype_seq_open(struct inode *inode, struct file *file)
4273 return seq_open_net(inode, file, &ptype_seq_ops,
4274 sizeof(struct seq_net_private));
4277 static const struct file_operations ptype_seq_fops = {
4278 .owner = THIS_MODULE,
4279 .open = ptype_seq_open,
4281 .llseek = seq_lseek,
4282 .release = seq_release_net,
4286 static int __net_init dev_proc_net_init(struct net *net)
4290 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4292 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4294 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4297 if (wext_proc_init(net))
4303 proc_net_remove(net, "ptype");
4305 proc_net_remove(net, "softnet_stat");
4307 proc_net_remove(net, "dev");
4311 static void __net_exit dev_proc_net_exit(struct net *net)
4313 wext_proc_exit(net);
4315 proc_net_remove(net, "ptype");
4316 proc_net_remove(net, "softnet_stat");
4317 proc_net_remove(net, "dev");
4320 static struct pernet_operations __net_initdata dev_proc_ops = {
4321 .init = dev_proc_net_init,
4322 .exit = dev_proc_net_exit,
4325 static int __init dev_proc_init(void)
4327 return register_pernet_subsys(&dev_proc_ops);
4330 #define dev_proc_init() 0
4331 #endif /* CONFIG_PROC_FS */
4335 * netdev_set_master - set up master pointer
4336 * @slave: slave device
4337 * @master: new master device
4339 * Changes the master device of the slave. Pass %NULL to break the
4340 * bonding. The caller must hold the RTNL semaphore. On a failure
4341 * a negative errno code is returned. On success the reference counts
4342 * are adjusted and the function returns zero.
4344 int netdev_set_master(struct net_device *slave, struct net_device *master)
4346 struct net_device *old = slave->master;
4356 slave->master = master;
4362 EXPORT_SYMBOL(netdev_set_master);
4365 * netdev_set_bond_master - set up bonding master/slave pair
4366 * @slave: slave device
4367 * @master: new master device
4369 * Changes the master device of the slave. Pass %NULL to break the
4370 * bonding. The caller must hold the RTNL semaphore. On a failure
4371 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4372 * to the routing socket and the function returns zero.
4374 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4380 err = netdev_set_master(slave, master);
4384 slave->flags |= IFF_SLAVE;
4386 slave->flags &= ~IFF_SLAVE;
4388 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4391 EXPORT_SYMBOL(netdev_set_bond_master);
4393 static void dev_change_rx_flags(struct net_device *dev, int flags)
4395 const struct net_device_ops *ops = dev->netdev_ops;
4397 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4398 ops->ndo_change_rx_flags(dev, flags);
4401 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4403 unsigned short old_flags = dev->flags;
4409 dev->flags |= IFF_PROMISC;
4410 dev->promiscuity += inc;
4411 if (dev->promiscuity == 0) {
4414 * If inc causes overflow, untouch promisc and return error.
4417 dev->flags &= ~IFF_PROMISC;
4419 dev->promiscuity -= inc;
4420 printk(KERN_WARNING "%s: promiscuity touches roof, "
4421 "set promiscuity failed, promiscuity feature "
4422 "of device might be broken.\n", dev->name);
4426 if (dev->flags != old_flags) {
4427 printk(KERN_INFO "device %s %s promiscuous mode\n",
4428 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4430 if (audit_enabled) {
4431 current_uid_gid(&uid, &gid);
4432 audit_log(current->audit_context, GFP_ATOMIC,
4433 AUDIT_ANOM_PROMISCUOUS,
4434 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4435 dev->name, (dev->flags & IFF_PROMISC),
4436 (old_flags & IFF_PROMISC),
4437 audit_get_loginuid(current),
4439 audit_get_sessionid(current));
4442 dev_change_rx_flags(dev, IFF_PROMISC);
4448 * dev_set_promiscuity - update promiscuity count on a device
4452 * Add or remove promiscuity from a device. While the count in the device
4453 * remains above zero the interface remains promiscuous. Once it hits zero
4454 * the device reverts back to normal filtering operation. A negative inc
4455 * value is used to drop promiscuity on the device.
4456 * Return 0 if successful or a negative errno code on error.
4458 int dev_set_promiscuity(struct net_device *dev, int inc)
4460 unsigned short old_flags = dev->flags;
4463 err = __dev_set_promiscuity(dev, inc);
4466 if (dev->flags != old_flags)
4467 dev_set_rx_mode(dev);
4470 EXPORT_SYMBOL(dev_set_promiscuity);
4473 * dev_set_allmulti - update allmulti count on a device
4477 * Add or remove reception of all multicast frames to a device. While the
4478 * count in the device remains above zero the interface remains listening
4479 * to all interfaces. Once it hits zero the device reverts back to normal
4480 * filtering operation. A negative @inc value is used to drop the counter
4481 * when releasing a resource needing all multicasts.
4482 * Return 0 if successful or a negative errno code on error.
4485 int dev_set_allmulti(struct net_device *dev, int inc)
4487 unsigned short old_flags = dev->flags;
4491 dev->flags |= IFF_ALLMULTI;
4492 dev->allmulti += inc;
4493 if (dev->allmulti == 0) {
4496 * If inc causes overflow, untouch allmulti and return error.
4499 dev->flags &= ~IFF_ALLMULTI;
4501 dev->allmulti -= inc;
4502 printk(KERN_WARNING "%s: allmulti touches roof, "
4503 "set allmulti failed, allmulti feature of "
4504 "device might be broken.\n", dev->name);
4508 if (dev->flags ^ old_flags) {
4509 dev_change_rx_flags(dev, IFF_ALLMULTI);
4510 dev_set_rx_mode(dev);
4514 EXPORT_SYMBOL(dev_set_allmulti);
4517 * Upload unicast and multicast address lists to device and
4518 * configure RX filtering. When the device doesn't support unicast
4519 * filtering it is put in promiscuous mode while unicast addresses
4522 void __dev_set_rx_mode(struct net_device *dev)
4524 const struct net_device_ops *ops = dev->netdev_ops;
4526 /* dev_open will call this function so the list will stay sane. */
4527 if (!(dev->flags&IFF_UP))
4530 if (!netif_device_present(dev))
4533 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4534 /* Unicast addresses changes may only happen under the rtnl,
4535 * therefore calling __dev_set_promiscuity here is safe.
4537 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4538 __dev_set_promiscuity(dev, 1);
4539 dev->uc_promisc = true;
4540 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4541 __dev_set_promiscuity(dev, -1);
4542 dev->uc_promisc = false;
4546 if (ops->ndo_set_rx_mode)
4547 ops->ndo_set_rx_mode(dev);
4550 void dev_set_rx_mode(struct net_device *dev)
4552 netif_addr_lock_bh(dev);
4553 __dev_set_rx_mode(dev);
4554 netif_addr_unlock_bh(dev);
4558 * dev_ethtool_get_settings - call device's ethtool_ops::get_settings()
4560 * @cmd: memory area for ethtool_ops::get_settings() result
4562 * The cmd arg is initialized properly (cleared and
4563 * ethtool_cmd::cmd field set to ETHTOOL_GSET).
4565 * Return device's ethtool_ops::get_settings() result value or
4566 * -EOPNOTSUPP when device doesn't expose
4567 * ethtool_ops::get_settings() operation.
4569 int dev_ethtool_get_settings(struct net_device *dev,
4570 struct ethtool_cmd *cmd)
4572 if (!dev->ethtool_ops || !dev->ethtool_ops->get_settings)
4575 memset(cmd, 0, sizeof(struct ethtool_cmd));
4576 cmd->cmd = ETHTOOL_GSET;
4577 return dev->ethtool_ops->get_settings(dev, cmd);
4579 EXPORT_SYMBOL(dev_ethtool_get_settings);
4582 * dev_get_flags - get flags reported to userspace
4585 * Get the combination of flag bits exported through APIs to userspace.
4587 unsigned dev_get_flags(const struct net_device *dev)
4591 flags = (dev->flags & ~(IFF_PROMISC |
4596 (dev->gflags & (IFF_PROMISC |
4599 if (netif_running(dev)) {
4600 if (netif_oper_up(dev))
4601 flags |= IFF_RUNNING;
4602 if (netif_carrier_ok(dev))
4603 flags |= IFF_LOWER_UP;
4604 if (netif_dormant(dev))
4605 flags |= IFF_DORMANT;
4610 EXPORT_SYMBOL(dev_get_flags);
4612 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4614 int old_flags = dev->flags;
4620 * Set the flags on our device.
4623 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4624 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4626 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4630 * Load in the correct multicast list now the flags have changed.
4633 if ((old_flags ^ flags) & IFF_MULTICAST)
4634 dev_change_rx_flags(dev, IFF_MULTICAST);
4636 dev_set_rx_mode(dev);
4639 * Have we downed the interface. We handle IFF_UP ourselves
4640 * according to user attempts to set it, rather than blindly
4645 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4646 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4649 dev_set_rx_mode(dev);
4652 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4653 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4655 dev->gflags ^= IFF_PROMISC;
4656 dev_set_promiscuity(dev, inc);
4659 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4660 is important. Some (broken) drivers set IFF_PROMISC, when
4661 IFF_ALLMULTI is requested not asking us and not reporting.
4663 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4664 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4666 dev->gflags ^= IFF_ALLMULTI;
4667 dev_set_allmulti(dev, inc);
4673 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4675 unsigned int changes = dev->flags ^ old_flags;
4677 if (changes & IFF_UP) {
4678 if (dev->flags & IFF_UP)
4679 call_netdevice_notifiers(NETDEV_UP, dev);
4681 call_netdevice_notifiers(NETDEV_DOWN, dev);
4684 if (dev->flags & IFF_UP &&
4685 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4686 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4690 * dev_change_flags - change device settings
4692 * @flags: device state flags
4694 * Change settings on device based state flags. The flags are
4695 * in the userspace exported format.
4697 int dev_change_flags(struct net_device *dev, unsigned flags)
4700 int old_flags = dev->flags;
4702 ret = __dev_change_flags(dev, flags);
4706 changes = old_flags ^ dev->flags;
4708 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4710 __dev_notify_flags(dev, old_flags);
4713 EXPORT_SYMBOL(dev_change_flags);
4716 * dev_set_mtu - Change maximum transfer unit
4718 * @new_mtu: new transfer unit
4720 * Change the maximum transfer size of the network device.
4722 int dev_set_mtu(struct net_device *dev, int new_mtu)
4724 const struct net_device_ops *ops = dev->netdev_ops;
4727 if (new_mtu == dev->mtu)
4730 /* MTU must be positive. */
4734 if (!netif_device_present(dev))
4738 if (ops->ndo_change_mtu)
4739 err = ops->ndo_change_mtu(dev, new_mtu);
4743 if (!err && dev->flags & IFF_UP)
4744 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4747 EXPORT_SYMBOL(dev_set_mtu);
4750 * dev_set_group - Change group this device belongs to
4752 * @new_group: group this device should belong to
4754 void dev_set_group(struct net_device *dev, int new_group)
4756 dev->group = new_group;
4758 EXPORT_SYMBOL(dev_set_group);
4761 * dev_set_mac_address - Change Media Access Control Address
4765 * Change the hardware (MAC) address of the device
4767 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4769 const struct net_device_ops *ops = dev->netdev_ops;
4772 if (!ops->ndo_set_mac_address)
4774 if (sa->sa_family != dev->type)
4776 if (!netif_device_present(dev))
4778 err = ops->ndo_set_mac_address(dev, sa);
4780 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4783 EXPORT_SYMBOL(dev_set_mac_address);
4786 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4788 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4791 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4797 case SIOCGIFFLAGS: /* Get interface flags */
4798 ifr->ifr_flags = (short) dev_get_flags(dev);
4801 case SIOCGIFMETRIC: /* Get the metric on the interface
4802 (currently unused) */
4803 ifr->ifr_metric = 0;
4806 case SIOCGIFMTU: /* Get the MTU of a device */
4807 ifr->ifr_mtu = dev->mtu;
4812 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4814 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4815 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4816 ifr->ifr_hwaddr.sa_family = dev->type;
4824 ifr->ifr_map.mem_start = dev->mem_start;
4825 ifr->ifr_map.mem_end = dev->mem_end;
4826 ifr->ifr_map.base_addr = dev->base_addr;
4827 ifr->ifr_map.irq = dev->irq;
4828 ifr->ifr_map.dma = dev->dma;
4829 ifr->ifr_map.port = dev->if_port;
4833 ifr->ifr_ifindex = dev->ifindex;
4837 ifr->ifr_qlen = dev->tx_queue_len;
4841 /* dev_ioctl() should ensure this case
4853 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4855 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4858 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4859 const struct net_device_ops *ops;
4864 ops = dev->netdev_ops;
4867 case SIOCSIFFLAGS: /* Set interface flags */
4868 return dev_change_flags(dev, ifr->ifr_flags);
4870 case SIOCSIFMETRIC: /* Set the metric on the interface
4871 (currently unused) */
4874 case SIOCSIFMTU: /* Set the MTU of a device */
4875 return dev_set_mtu(dev, ifr->ifr_mtu);
4878 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4880 case SIOCSIFHWBROADCAST:
4881 if (ifr->ifr_hwaddr.sa_family != dev->type)
4883 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4884 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4885 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4889 if (ops->ndo_set_config) {
4890 if (!netif_device_present(dev))
4892 return ops->ndo_set_config(dev, &ifr->ifr_map);
4897 if (!ops->ndo_set_rx_mode ||
4898 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4900 if (!netif_device_present(dev))
4902 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4905 if (!ops->ndo_set_rx_mode ||
4906 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4908 if (!netif_device_present(dev))
4910 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4913 if (ifr->ifr_qlen < 0)
4915 dev->tx_queue_len = ifr->ifr_qlen;
4919 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4920 return dev_change_name(dev, ifr->ifr_newname);
4923 * Unknown or private ioctl
4926 if ((cmd >= SIOCDEVPRIVATE &&
4927 cmd <= SIOCDEVPRIVATE + 15) ||
4928 cmd == SIOCBONDENSLAVE ||
4929 cmd == SIOCBONDRELEASE ||
4930 cmd == SIOCBONDSETHWADDR ||
4931 cmd == SIOCBONDSLAVEINFOQUERY ||
4932 cmd == SIOCBONDINFOQUERY ||
4933 cmd == SIOCBONDCHANGEACTIVE ||
4934 cmd == SIOCGMIIPHY ||
4935 cmd == SIOCGMIIREG ||
4936 cmd == SIOCSMIIREG ||
4937 cmd == SIOCBRADDIF ||
4938 cmd == SIOCBRDELIF ||
4939 cmd == SIOCSHWTSTAMP ||
4940 cmd == SIOCWANDEV) {
4942 if (ops->ndo_do_ioctl) {
4943 if (netif_device_present(dev))
4944 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4956 * This function handles all "interface"-type I/O control requests. The actual
4957 * 'doing' part of this is dev_ifsioc above.
4961 * dev_ioctl - network device ioctl
4962 * @net: the applicable net namespace
4963 * @cmd: command to issue
4964 * @arg: pointer to a struct ifreq in user space
4966 * Issue ioctl functions to devices. This is normally called by the
4967 * user space syscall interfaces but can sometimes be useful for
4968 * other purposes. The return value is the return from the syscall if
4969 * positive or a negative errno code on error.
4972 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4978 /* One special case: SIOCGIFCONF takes ifconf argument
4979 and requires shared lock, because it sleeps writing
4983 if (cmd == SIOCGIFCONF) {
4985 ret = dev_ifconf(net, (char __user *) arg);
4989 if (cmd == SIOCGIFNAME)
4990 return dev_ifname(net, (struct ifreq __user *)arg);
4992 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4995 ifr.ifr_name[IFNAMSIZ-1] = 0;
4997 colon = strchr(ifr.ifr_name, ':');
5002 * See which interface the caller is talking about.
5007 * These ioctl calls:
5008 * - can be done by all.
5009 * - atomic and do not require locking.
5020 dev_load(net, ifr.ifr_name);
5022 ret = dev_ifsioc_locked(net, &ifr, cmd);
5027 if (copy_to_user(arg, &ifr,
5028 sizeof(struct ifreq)))
5034 dev_load(net, ifr.ifr_name);
5036 ret = dev_ethtool(net, &ifr);
5041 if (copy_to_user(arg, &ifr,
5042 sizeof(struct ifreq)))
5048 * These ioctl calls:
5049 * - require superuser power.
5050 * - require strict serialization.
5056 if (!capable(CAP_NET_ADMIN))
5058 dev_load(net, ifr.ifr_name);
5060 ret = dev_ifsioc(net, &ifr, cmd);
5065 if (copy_to_user(arg, &ifr,
5066 sizeof(struct ifreq)))
5072 * These ioctl calls:
5073 * - require superuser power.
5074 * - require strict serialization.
5075 * - do not return a value
5085 case SIOCSIFHWBROADCAST:
5088 case SIOCBONDENSLAVE:
5089 case SIOCBONDRELEASE:
5090 case SIOCBONDSETHWADDR:
5091 case SIOCBONDCHANGEACTIVE:
5095 if (!capable(CAP_NET_ADMIN))
5098 case SIOCBONDSLAVEINFOQUERY:
5099 case SIOCBONDINFOQUERY:
5100 dev_load(net, ifr.ifr_name);
5102 ret = dev_ifsioc(net, &ifr, cmd);
5107 /* Get the per device memory space. We can add this but
5108 * currently do not support it */
5110 /* Set the per device memory buffer space.
5111 * Not applicable in our case */
5116 * Unknown or private ioctl.
5119 if (cmd == SIOCWANDEV ||
5120 (cmd >= SIOCDEVPRIVATE &&
5121 cmd <= SIOCDEVPRIVATE + 15)) {
5122 dev_load(net, ifr.ifr_name);
5124 ret = dev_ifsioc(net, &ifr, cmd);
5126 if (!ret && copy_to_user(arg, &ifr,
5127 sizeof(struct ifreq)))
5131 /* Take care of Wireless Extensions */
5132 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5133 return wext_handle_ioctl(net, &ifr, cmd, arg);
5140 * dev_new_index - allocate an ifindex
5141 * @net: the applicable net namespace
5143 * Returns a suitable unique value for a new device interface
5144 * number. The caller must hold the rtnl semaphore or the
5145 * dev_base_lock to be sure it remains unique.
5147 static int dev_new_index(struct net *net)
5153 if (!__dev_get_by_index(net, ifindex))
5158 /* Delayed registration/unregisteration */
5159 static LIST_HEAD(net_todo_list);
5161 static void net_set_todo(struct net_device *dev)
5163 list_add_tail(&dev->todo_list, &net_todo_list);
5166 static void rollback_registered_many(struct list_head *head)
5168 struct net_device *dev, *tmp;
5170 BUG_ON(dev_boot_phase);
5173 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5174 /* Some devices call without registering
5175 * for initialization unwind. Remove those
5176 * devices and proceed with the remaining.
5178 if (dev->reg_state == NETREG_UNINITIALIZED) {
5179 pr_debug("unregister_netdevice: device %s/%p never "
5180 "was registered\n", dev->name, dev);
5183 list_del(&dev->unreg_list);
5186 dev->dismantle = true;
5187 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5190 /* If device is running, close it first. */
5191 dev_close_many(head);
5193 list_for_each_entry(dev, head, unreg_list) {
5194 /* And unlink it from device chain. */
5195 unlist_netdevice(dev);
5197 dev->reg_state = NETREG_UNREGISTERING;
5202 list_for_each_entry(dev, head, unreg_list) {
5203 /* Shutdown queueing discipline. */
5207 /* Notify protocols, that we are about to destroy
5208 this device. They should clean all the things.
5210 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5212 if (!dev->rtnl_link_ops ||
5213 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5214 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5217 * Flush the unicast and multicast chains
5222 if (dev->netdev_ops->ndo_uninit)
5223 dev->netdev_ops->ndo_uninit(dev);
5225 /* Notifier chain MUST detach us from master device. */
5226 WARN_ON(dev->master);
5228 /* Remove entries from kobject tree */
5229 netdev_unregister_kobject(dev);
5232 /* Process any work delayed until the end of the batch */
5233 dev = list_first_entry(head, struct net_device, unreg_list);
5234 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5238 list_for_each_entry(dev, head, unreg_list)
5242 static void rollback_registered(struct net_device *dev)
5246 list_add(&dev->unreg_list, &single);
5247 rollback_registered_many(&single);
5251 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5253 /* Fix illegal checksum combinations */
5254 if ((features & NETIF_F_HW_CSUM) &&
5255 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5256 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5257 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5260 if ((features & NETIF_F_NO_CSUM) &&
5261 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5262 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5263 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5266 /* Fix illegal SG+CSUM combinations. */
5267 if ((features & NETIF_F_SG) &&
5268 !(features & NETIF_F_ALL_CSUM)) {
5270 "Dropping NETIF_F_SG since no checksum feature.\n");
5271 features &= ~NETIF_F_SG;
5274 /* TSO requires that SG is present as well. */
5275 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5276 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5277 features &= ~NETIF_F_ALL_TSO;
5280 /* TSO ECN requires that TSO is present as well. */
5281 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5282 features &= ~NETIF_F_TSO_ECN;
5284 /* Software GSO depends on SG. */
5285 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5286 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5287 features &= ~NETIF_F_GSO;
5290 /* UFO needs SG and checksumming */
5291 if (features & NETIF_F_UFO) {
5292 /* maybe split UFO into V4 and V6? */
5293 if (!((features & NETIF_F_GEN_CSUM) ||
5294 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5295 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5297 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5298 features &= ~NETIF_F_UFO;
5301 if (!(features & NETIF_F_SG)) {
5303 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5304 features &= ~NETIF_F_UFO;
5311 int __netdev_update_features(struct net_device *dev)
5318 features = netdev_get_wanted_features(dev);
5320 if (dev->netdev_ops->ndo_fix_features)
5321 features = dev->netdev_ops->ndo_fix_features(dev, features);
5323 /* driver might be less strict about feature dependencies */
5324 features = netdev_fix_features(dev, features);
5326 if (dev->features == features)
5329 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5330 dev->features, features);
5332 if (dev->netdev_ops->ndo_set_features)
5333 err = dev->netdev_ops->ndo_set_features(dev, features);
5335 if (unlikely(err < 0)) {
5337 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5338 err, features, dev->features);
5343 dev->features = features;
5349 * netdev_update_features - recalculate device features
5350 * @dev: the device to check
5352 * Recalculate dev->features set and send notifications if it
5353 * has changed. Should be called after driver or hardware dependent
5354 * conditions might have changed that influence the features.
5356 void netdev_update_features(struct net_device *dev)
5358 if (__netdev_update_features(dev))
5359 netdev_features_change(dev);
5361 EXPORT_SYMBOL(netdev_update_features);
5364 * netdev_change_features - recalculate device features
5365 * @dev: the device to check
5367 * Recalculate dev->features set and send notifications even
5368 * if they have not changed. Should be called instead of
5369 * netdev_update_features() if also dev->vlan_features might
5370 * have changed to allow the changes to be propagated to stacked
5373 void netdev_change_features(struct net_device *dev)
5375 __netdev_update_features(dev);
5376 netdev_features_change(dev);
5378 EXPORT_SYMBOL(netdev_change_features);
5381 * netif_stacked_transfer_operstate - transfer operstate
5382 * @rootdev: the root or lower level device to transfer state from
5383 * @dev: the device to transfer operstate to
5385 * Transfer operational state from root to device. This is normally
5386 * called when a stacking relationship exists between the root
5387 * device and the device(a leaf device).
5389 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5390 struct net_device *dev)
5392 if (rootdev->operstate == IF_OPER_DORMANT)
5393 netif_dormant_on(dev);
5395 netif_dormant_off(dev);
5397 if (netif_carrier_ok(rootdev)) {
5398 if (!netif_carrier_ok(dev))
5399 netif_carrier_on(dev);
5401 if (netif_carrier_ok(dev))
5402 netif_carrier_off(dev);
5405 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5408 static int netif_alloc_rx_queues(struct net_device *dev)
5410 unsigned int i, count = dev->num_rx_queues;
5411 struct netdev_rx_queue *rx;
5415 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5417 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5422 for (i = 0; i < count; i++)
5428 static void netdev_init_one_queue(struct net_device *dev,
5429 struct netdev_queue *queue, void *_unused)
5431 /* Initialize queue lock */
5432 spin_lock_init(&queue->_xmit_lock);
5433 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5434 queue->xmit_lock_owner = -1;
5435 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5439 static int netif_alloc_netdev_queues(struct net_device *dev)
5441 unsigned int count = dev->num_tx_queues;
5442 struct netdev_queue *tx;
5446 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5448 pr_err("netdev: Unable to allocate %u tx queues.\n",
5454 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5455 spin_lock_init(&dev->tx_global_lock);
5461 * register_netdevice - register a network device
5462 * @dev: device to register
5464 * Take a completed network device structure and add it to the kernel
5465 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5466 * chain. 0 is returned on success. A negative errno code is returned
5467 * on a failure to set up the device, or if the name is a duplicate.
5469 * Callers must hold the rtnl semaphore. You may want
5470 * register_netdev() instead of this.
5473 * The locking appears insufficient to guarantee two parallel registers
5474 * will not get the same name.
5477 int register_netdevice(struct net_device *dev)
5480 struct net *net = dev_net(dev);
5482 BUG_ON(dev_boot_phase);
5487 /* When net_device's are persistent, this will be fatal. */
5488 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5491 spin_lock_init(&dev->addr_list_lock);
5492 netdev_set_addr_lockdep_class(dev);
5496 ret = dev_get_valid_name(dev, dev->name);
5500 /* Init, if this function is available */
5501 if (dev->netdev_ops->ndo_init) {
5502 ret = dev->netdev_ops->ndo_init(dev);
5510 dev->ifindex = dev_new_index(net);
5511 if (dev->iflink == -1)
5512 dev->iflink = dev->ifindex;
5514 /* Transfer changeable features to wanted_features and enable
5515 * software offloads (GSO and GRO).
5517 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5518 dev->features |= NETIF_F_SOFT_FEATURES;
5519 dev->wanted_features = dev->features & dev->hw_features;
5521 /* Turn on no cache copy if HW is doing checksum */
5522 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5523 if ((dev->features & NETIF_F_ALL_CSUM) &&
5524 !(dev->features & NETIF_F_NO_CSUM)) {
5525 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5526 dev->features |= NETIF_F_NOCACHE_COPY;
5529 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5531 dev->vlan_features |= NETIF_F_HIGHDMA;
5533 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5534 ret = notifier_to_errno(ret);
5538 ret = netdev_register_kobject(dev);
5541 dev->reg_state = NETREG_REGISTERED;
5543 __netdev_update_features(dev);
5546 * Default initial state at registry is that the
5547 * device is present.
5550 set_bit(__LINK_STATE_PRESENT, &dev->state);
5552 dev_init_scheduler(dev);
5554 list_netdevice(dev);
5556 /* Notify protocols, that a new device appeared. */
5557 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5558 ret = notifier_to_errno(ret);
5560 rollback_registered(dev);
5561 dev->reg_state = NETREG_UNREGISTERED;
5564 * Prevent userspace races by waiting until the network
5565 * device is fully setup before sending notifications.
5567 if (!dev->rtnl_link_ops ||
5568 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5569 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5575 if (dev->netdev_ops->ndo_uninit)
5576 dev->netdev_ops->ndo_uninit(dev);
5579 EXPORT_SYMBOL(register_netdevice);
5582 * init_dummy_netdev - init a dummy network device for NAPI
5583 * @dev: device to init
5585 * This takes a network device structure and initialize the minimum
5586 * amount of fields so it can be used to schedule NAPI polls without
5587 * registering a full blown interface. This is to be used by drivers
5588 * that need to tie several hardware interfaces to a single NAPI
5589 * poll scheduler due to HW limitations.
5591 int init_dummy_netdev(struct net_device *dev)
5593 /* Clear everything. Note we don't initialize spinlocks
5594 * are they aren't supposed to be taken by any of the
5595 * NAPI code and this dummy netdev is supposed to be
5596 * only ever used for NAPI polls
5598 memset(dev, 0, sizeof(struct net_device));
5600 /* make sure we BUG if trying to hit standard
5601 * register/unregister code path
5603 dev->reg_state = NETREG_DUMMY;
5605 /* NAPI wants this */
5606 INIT_LIST_HEAD(&dev->napi_list);
5608 /* a dummy interface is started by default */
5609 set_bit(__LINK_STATE_PRESENT, &dev->state);
5610 set_bit(__LINK_STATE_START, &dev->state);
5612 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5613 * because users of this 'device' dont need to change
5619 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5623 * register_netdev - register a network device
5624 * @dev: device to register
5626 * Take a completed network device structure and add it to the kernel
5627 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5628 * chain. 0 is returned on success. A negative errno code is returned
5629 * on a failure to set up the device, or if the name is a duplicate.
5631 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5632 * and expands the device name if you passed a format string to
5635 int register_netdev(struct net_device *dev)
5640 err = register_netdevice(dev);
5644 EXPORT_SYMBOL(register_netdev);
5646 int netdev_refcnt_read(const struct net_device *dev)
5650 for_each_possible_cpu(i)
5651 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5654 EXPORT_SYMBOL(netdev_refcnt_read);
5657 * netdev_wait_allrefs - wait until all references are gone.
5659 * This is called when unregistering network devices.
5661 * Any protocol or device that holds a reference should register
5662 * for netdevice notification, and cleanup and put back the
5663 * reference if they receive an UNREGISTER event.
5664 * We can get stuck here if buggy protocols don't correctly
5667 static void netdev_wait_allrefs(struct net_device *dev)
5669 unsigned long rebroadcast_time, warning_time;
5672 linkwatch_forget_dev(dev);
5674 rebroadcast_time = warning_time = jiffies;
5675 refcnt = netdev_refcnt_read(dev);
5677 while (refcnt != 0) {
5678 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5681 /* Rebroadcast unregister notification */
5682 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5683 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5684 * should have already handle it the first time */
5686 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5688 /* We must not have linkwatch events
5689 * pending on unregister. If this
5690 * happens, we simply run the queue
5691 * unscheduled, resulting in a noop
5694 linkwatch_run_queue();
5699 rebroadcast_time = jiffies;
5704 refcnt = netdev_refcnt_read(dev);
5706 if (time_after(jiffies, warning_time + 10 * HZ)) {
5707 printk(KERN_EMERG "unregister_netdevice: "
5708 "waiting for %s to become free. Usage "
5711 warning_time = jiffies;
5720 * register_netdevice(x1);
5721 * register_netdevice(x2);
5723 * unregister_netdevice(y1);
5724 * unregister_netdevice(y2);
5730 * We are invoked by rtnl_unlock().
5731 * This allows us to deal with problems:
5732 * 1) We can delete sysfs objects which invoke hotplug
5733 * without deadlocking with linkwatch via keventd.
5734 * 2) Since we run with the RTNL semaphore not held, we can sleep
5735 * safely in order to wait for the netdev refcnt to drop to zero.
5737 * We must not return until all unregister events added during
5738 * the interval the lock was held have been completed.
5740 void netdev_run_todo(void)
5742 struct list_head list;
5744 /* Snapshot list, allow later requests */
5745 list_replace_init(&net_todo_list, &list);
5749 while (!list_empty(&list)) {
5750 struct net_device *dev
5751 = list_first_entry(&list, struct net_device, todo_list);
5752 list_del(&dev->todo_list);
5754 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5755 printk(KERN_ERR "network todo '%s' but state %d\n",
5756 dev->name, dev->reg_state);
5761 dev->reg_state = NETREG_UNREGISTERED;
5763 on_each_cpu(flush_backlog, dev, 1);
5765 netdev_wait_allrefs(dev);
5768 BUG_ON(netdev_refcnt_read(dev));
5769 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5770 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5771 WARN_ON(dev->dn_ptr);
5773 if (dev->destructor)
5774 dev->destructor(dev);
5776 /* Free network device */
5777 kobject_put(&dev->dev.kobj);
5781 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5782 * fields in the same order, with only the type differing.
5784 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5785 const struct net_device_stats *netdev_stats)
5787 #if BITS_PER_LONG == 64
5788 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5789 memcpy(stats64, netdev_stats, sizeof(*stats64));
5791 size_t i, n = sizeof(*stats64) / sizeof(u64);
5792 const unsigned long *src = (const unsigned long *)netdev_stats;
5793 u64 *dst = (u64 *)stats64;
5795 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5796 sizeof(*stats64) / sizeof(u64));
5797 for (i = 0; i < n; i++)
5803 * dev_get_stats - get network device statistics
5804 * @dev: device to get statistics from
5805 * @storage: place to store stats
5807 * Get network statistics from device. Return @storage.
5808 * The device driver may provide its own method by setting
5809 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5810 * otherwise the internal statistics structure is used.
5812 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5813 struct rtnl_link_stats64 *storage)
5815 const struct net_device_ops *ops = dev->netdev_ops;
5817 if (ops->ndo_get_stats64) {
5818 memset(storage, 0, sizeof(*storage));
5819 ops->ndo_get_stats64(dev, storage);
5820 } else if (ops->ndo_get_stats) {
5821 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5823 netdev_stats_to_stats64(storage, &dev->stats);
5825 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5828 EXPORT_SYMBOL(dev_get_stats);
5830 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5832 struct netdev_queue *queue = dev_ingress_queue(dev);
5834 #ifdef CONFIG_NET_CLS_ACT
5837 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5840 netdev_init_one_queue(dev, queue, NULL);
5841 queue->qdisc = &noop_qdisc;
5842 queue->qdisc_sleeping = &noop_qdisc;
5843 rcu_assign_pointer(dev->ingress_queue, queue);
5849 * alloc_netdev_mqs - allocate network device
5850 * @sizeof_priv: size of private data to allocate space for
5851 * @name: device name format string
5852 * @setup: callback to initialize device
5853 * @txqs: the number of TX subqueues to allocate
5854 * @rxqs: the number of RX subqueues to allocate
5856 * Allocates a struct net_device with private data area for driver use
5857 * and performs basic initialization. Also allocates subquue structs
5858 * for each queue on the device.
5860 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5861 void (*setup)(struct net_device *),
5862 unsigned int txqs, unsigned int rxqs)
5864 struct net_device *dev;
5866 struct net_device *p;
5868 BUG_ON(strlen(name) >= sizeof(dev->name));
5871 pr_err("alloc_netdev: Unable to allocate device "
5872 "with zero queues.\n");
5878 pr_err("alloc_netdev: Unable to allocate device "
5879 "with zero RX queues.\n");
5884 alloc_size = sizeof(struct net_device);
5886 /* ensure 32-byte alignment of private area */
5887 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5888 alloc_size += sizeof_priv;
5890 /* ensure 32-byte alignment of whole construct */
5891 alloc_size += NETDEV_ALIGN - 1;
5893 p = kzalloc(alloc_size, GFP_KERNEL);
5895 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5899 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5900 dev->padded = (char *)dev - (char *)p;
5902 dev->pcpu_refcnt = alloc_percpu(int);
5903 if (!dev->pcpu_refcnt)
5906 if (dev_addr_init(dev))
5912 dev_net_set(dev, &init_net);
5914 dev->gso_max_size = GSO_MAX_SIZE;
5916 INIT_LIST_HEAD(&dev->napi_list);
5917 INIT_LIST_HEAD(&dev->unreg_list);
5918 INIT_LIST_HEAD(&dev->link_watch_list);
5919 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5922 dev->num_tx_queues = txqs;
5923 dev->real_num_tx_queues = txqs;
5924 if (netif_alloc_netdev_queues(dev))
5928 dev->num_rx_queues = rxqs;
5929 dev->real_num_rx_queues = rxqs;
5930 if (netif_alloc_rx_queues(dev))
5934 strcpy(dev->name, name);
5935 dev->group = INIT_NETDEV_GROUP;
5943 free_percpu(dev->pcpu_refcnt);
5953 EXPORT_SYMBOL(alloc_netdev_mqs);
5956 * free_netdev - free network device
5959 * This function does the last stage of destroying an allocated device
5960 * interface. The reference to the device object is released.
5961 * If this is the last reference then it will be freed.
5963 void free_netdev(struct net_device *dev)
5965 struct napi_struct *p, *n;
5967 release_net(dev_net(dev));
5974 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5976 /* Flush device addresses */
5977 dev_addr_flush(dev);
5979 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5982 free_percpu(dev->pcpu_refcnt);
5983 dev->pcpu_refcnt = NULL;
5985 /* Compatibility with error handling in drivers */
5986 if (dev->reg_state == NETREG_UNINITIALIZED) {
5987 kfree((char *)dev - dev->padded);
5991 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5992 dev->reg_state = NETREG_RELEASED;
5994 /* will free via device release */
5995 put_device(&dev->dev);
5997 EXPORT_SYMBOL(free_netdev);
6000 * synchronize_net - Synchronize with packet receive processing
6002 * Wait for packets currently being received to be done.
6003 * Does not block later packets from starting.
6005 void synchronize_net(void)
6008 if (rtnl_is_locked())
6009 synchronize_rcu_expedited();
6013 EXPORT_SYMBOL(synchronize_net);
6016 * unregister_netdevice_queue - remove device from the kernel
6020 * This function shuts down a device interface and removes it
6021 * from the kernel tables.
6022 * If head not NULL, device is queued to be unregistered later.
6024 * Callers must hold the rtnl semaphore. You may want
6025 * unregister_netdev() instead of this.
6028 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6033 list_move_tail(&dev->unreg_list, head);
6035 rollback_registered(dev);
6036 /* Finish processing unregister after unlock */
6040 EXPORT_SYMBOL(unregister_netdevice_queue);
6043 * unregister_netdevice_many - unregister many devices
6044 * @head: list of devices
6046 void unregister_netdevice_many(struct list_head *head)
6048 struct net_device *dev;
6050 if (!list_empty(head)) {
6051 rollback_registered_many(head);
6052 list_for_each_entry(dev, head, unreg_list)
6056 EXPORT_SYMBOL(unregister_netdevice_many);
6059 * unregister_netdev - remove device from the kernel
6062 * This function shuts down a device interface and removes it
6063 * from the kernel tables.
6065 * This is just a wrapper for unregister_netdevice that takes
6066 * the rtnl semaphore. In general you want to use this and not
6067 * unregister_netdevice.
6069 void unregister_netdev(struct net_device *dev)
6072 unregister_netdevice(dev);
6075 EXPORT_SYMBOL(unregister_netdev);
6078 * dev_change_net_namespace - move device to different nethost namespace
6080 * @net: network namespace
6081 * @pat: If not NULL name pattern to try if the current device name
6082 * is already taken in the destination network namespace.
6084 * This function shuts down a device interface and moves it
6085 * to a new network namespace. On success 0 is returned, on
6086 * a failure a netagive errno code is returned.
6088 * Callers must hold the rtnl semaphore.
6091 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6097 /* Don't allow namespace local devices to be moved. */
6099 if (dev->features & NETIF_F_NETNS_LOCAL)
6102 /* Ensure the device has been registrered */
6104 if (dev->reg_state != NETREG_REGISTERED)
6107 /* Get out if there is nothing todo */
6109 if (net_eq(dev_net(dev), net))
6112 /* Pick the destination device name, and ensure
6113 * we can use it in the destination network namespace.
6116 if (__dev_get_by_name(net, dev->name)) {
6117 /* We get here if we can't use the current device name */
6120 if (dev_get_valid_name(dev, pat) < 0)
6125 * And now a mini version of register_netdevice unregister_netdevice.
6128 /* If device is running close it first. */
6131 /* And unlink it from device chain */
6133 unlist_netdevice(dev);
6137 /* Shutdown queueing discipline. */
6140 /* Notify protocols, that we are about to destroy
6141 this device. They should clean all the things.
6143 Note that dev->reg_state stays at NETREG_REGISTERED.
6144 This is wanted because this way 8021q and macvlan know
6145 the device is just moving and can keep their slaves up.
6147 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6148 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6151 * Flush the unicast and multicast chains
6156 /* Actually switch the network namespace */
6157 dev_net_set(dev, net);
6159 /* If there is an ifindex conflict assign a new one */
6160 if (__dev_get_by_index(net, dev->ifindex)) {
6161 int iflink = (dev->iflink == dev->ifindex);
6162 dev->ifindex = dev_new_index(net);
6164 dev->iflink = dev->ifindex;
6167 /* Fixup kobjects */
6168 err = device_rename(&dev->dev, dev->name);
6171 /* Add the device back in the hashes */
6172 list_netdevice(dev);
6174 /* Notify protocols, that a new device appeared. */
6175 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6178 * Prevent userspace races by waiting until the network
6179 * device is fully setup before sending notifications.
6181 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6188 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6190 static int dev_cpu_callback(struct notifier_block *nfb,
6191 unsigned long action,
6194 struct sk_buff **list_skb;
6195 struct sk_buff *skb;
6196 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6197 struct softnet_data *sd, *oldsd;
6199 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6202 local_irq_disable();
6203 cpu = smp_processor_id();
6204 sd = &per_cpu(softnet_data, cpu);
6205 oldsd = &per_cpu(softnet_data, oldcpu);
6207 /* Find end of our completion_queue. */
6208 list_skb = &sd->completion_queue;
6210 list_skb = &(*list_skb)->next;
6211 /* Append completion queue from offline CPU. */
6212 *list_skb = oldsd->completion_queue;
6213 oldsd->completion_queue = NULL;
6215 /* Append output queue from offline CPU. */
6216 if (oldsd->output_queue) {
6217 *sd->output_queue_tailp = oldsd->output_queue;
6218 sd->output_queue_tailp = oldsd->output_queue_tailp;
6219 oldsd->output_queue = NULL;
6220 oldsd->output_queue_tailp = &oldsd->output_queue;
6222 /* Append NAPI poll list from offline CPU. */
6223 if (!list_empty(&oldsd->poll_list)) {
6224 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6225 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6228 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6231 /* Process offline CPU's input_pkt_queue */
6232 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6234 input_queue_head_incr(oldsd);
6236 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6238 input_queue_head_incr(oldsd);
6246 * netdev_increment_features - increment feature set by one
6247 * @all: current feature set
6248 * @one: new feature set
6249 * @mask: mask feature set
6251 * Computes a new feature set after adding a device with feature set
6252 * @one to the master device with current feature set @all. Will not
6253 * enable anything that is off in @mask. Returns the new feature set.
6255 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6257 if (mask & NETIF_F_GEN_CSUM)
6258 mask |= NETIF_F_ALL_CSUM;
6259 mask |= NETIF_F_VLAN_CHALLENGED;
6261 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6262 all &= one | ~NETIF_F_ALL_FOR_ALL;
6264 /* If device needs checksumming, downgrade to it. */
6265 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6266 all &= ~NETIF_F_NO_CSUM;
6268 /* If one device supports hw checksumming, set for all. */
6269 if (all & NETIF_F_GEN_CSUM)
6270 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6274 EXPORT_SYMBOL(netdev_increment_features);
6276 static struct hlist_head *netdev_create_hash(void)
6279 struct hlist_head *hash;
6281 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6283 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6284 INIT_HLIST_HEAD(&hash[i]);
6289 /* Initialize per network namespace state */
6290 static int __net_init netdev_init(struct net *net)
6292 INIT_LIST_HEAD(&net->dev_base_head);
6294 net->dev_name_head = netdev_create_hash();
6295 if (net->dev_name_head == NULL)
6298 net->dev_index_head = netdev_create_hash();
6299 if (net->dev_index_head == NULL)
6305 kfree(net->dev_name_head);
6311 * netdev_drivername - network driver for the device
6312 * @dev: network device
6314 * Determine network driver for device.
6316 const char *netdev_drivername(const struct net_device *dev)
6318 const struct device_driver *driver;
6319 const struct device *parent;
6320 const char *empty = "";
6322 parent = dev->dev.parent;
6326 driver = parent->driver;
6327 if (driver && driver->name)
6328 return driver->name;
6332 static int __netdev_printk(const char *level, const struct net_device *dev,
6333 struct va_format *vaf)
6337 if (dev && dev->dev.parent)
6338 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6339 netdev_name(dev), vaf);
6341 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6343 r = printk("%s(NULL net_device): %pV", level, vaf);
6348 int netdev_printk(const char *level, const struct net_device *dev,
6349 const char *format, ...)
6351 struct va_format vaf;
6355 va_start(args, format);
6360 r = __netdev_printk(level, dev, &vaf);
6365 EXPORT_SYMBOL(netdev_printk);
6367 #define define_netdev_printk_level(func, level) \
6368 int func(const struct net_device *dev, const char *fmt, ...) \
6371 struct va_format vaf; \
6374 va_start(args, fmt); \
6379 r = __netdev_printk(level, dev, &vaf); \
6384 EXPORT_SYMBOL(func);
6386 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6387 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6388 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6389 define_netdev_printk_level(netdev_err, KERN_ERR);
6390 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6391 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6392 define_netdev_printk_level(netdev_info, KERN_INFO);
6394 static void __net_exit netdev_exit(struct net *net)
6396 kfree(net->dev_name_head);
6397 kfree(net->dev_index_head);
6400 static struct pernet_operations __net_initdata netdev_net_ops = {
6401 .init = netdev_init,
6402 .exit = netdev_exit,
6405 static void __net_exit default_device_exit(struct net *net)
6407 struct net_device *dev, *aux;
6409 * Push all migratable network devices back to the
6410 * initial network namespace
6413 for_each_netdev_safe(net, dev, aux) {
6415 char fb_name[IFNAMSIZ];
6417 /* Ignore unmoveable devices (i.e. loopback) */
6418 if (dev->features & NETIF_F_NETNS_LOCAL)
6421 /* Leave virtual devices for the generic cleanup */
6422 if (dev->rtnl_link_ops)
6425 /* Push remaining network devices to init_net */
6426 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6427 err = dev_change_net_namespace(dev, &init_net, fb_name);
6429 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6430 __func__, dev->name, err);
6437 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6439 /* At exit all network devices most be removed from a network
6440 * namespace. Do this in the reverse order of registration.
6441 * Do this across as many network namespaces as possible to
6442 * improve batching efficiency.
6444 struct net_device *dev;
6446 LIST_HEAD(dev_kill_list);
6449 list_for_each_entry(net, net_list, exit_list) {
6450 for_each_netdev_reverse(net, dev) {
6451 if (dev->rtnl_link_ops)
6452 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6454 unregister_netdevice_queue(dev, &dev_kill_list);
6457 unregister_netdevice_many(&dev_kill_list);
6458 list_del(&dev_kill_list);
6462 static struct pernet_operations __net_initdata default_device_ops = {
6463 .exit = default_device_exit,
6464 .exit_batch = default_device_exit_batch,
6468 * Initialize the DEV module. At boot time this walks the device list and
6469 * unhooks any devices that fail to initialise (normally hardware not
6470 * present) and leaves us with a valid list of present and active devices.
6475 * This is called single threaded during boot, so no need
6476 * to take the rtnl semaphore.
6478 static int __init net_dev_init(void)
6480 int i, rc = -ENOMEM;
6482 BUG_ON(!dev_boot_phase);
6484 if (dev_proc_init())
6487 if (netdev_kobject_init())
6490 INIT_LIST_HEAD(&ptype_all);
6491 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6492 INIT_LIST_HEAD(&ptype_base[i]);
6494 if (register_pernet_subsys(&netdev_net_ops))
6498 * Initialise the packet receive queues.
6501 for_each_possible_cpu(i) {
6502 struct softnet_data *sd = &per_cpu(softnet_data, i);
6504 memset(sd, 0, sizeof(*sd));
6505 skb_queue_head_init(&sd->input_pkt_queue);
6506 skb_queue_head_init(&sd->process_queue);
6507 sd->completion_queue = NULL;
6508 INIT_LIST_HEAD(&sd->poll_list);
6509 sd->output_queue = NULL;
6510 sd->output_queue_tailp = &sd->output_queue;
6512 sd->csd.func = rps_trigger_softirq;
6518 sd->backlog.poll = process_backlog;
6519 sd->backlog.weight = weight_p;
6520 sd->backlog.gro_list = NULL;
6521 sd->backlog.gro_count = 0;
6526 /* The loopback device is special if any other network devices
6527 * is present in a network namespace the loopback device must
6528 * be present. Since we now dynamically allocate and free the
6529 * loopback device ensure this invariant is maintained by
6530 * keeping the loopback device as the first device on the
6531 * list of network devices. Ensuring the loopback devices
6532 * is the first device that appears and the last network device
6535 if (register_pernet_device(&loopback_net_ops))
6538 if (register_pernet_device(&default_device_ops))
6541 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6542 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6544 hotcpu_notifier(dev_cpu_callback, 0);
6552 subsys_initcall(net_dev_init);
6554 static int __init initialize_hashrnd(void)
6556 get_random_bytes(&hashrnd, sizeof(hashrnd));
6560 late_initcall_sync(initialize_hashrnd);