2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <linux/pci.h>
133 #include "net-sysfs.h"
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
142 * The list of packet types we will receive (as opposed to discard)
143 * and the routines to invoke.
145 * Why 16. Because with 16 the only overlap we get on a hash of the
146 * low nibble of the protocol value is RARP/SNAP/X.25.
148 * NOTE: That is no longer true with the addition of VLAN tags. Not
149 * sure which should go first, but I bet it won't make much
150 * difference if we are running VLANs. The good news is that
151 * this protocol won't be in the list unless compiled in, so
152 * the average user (w/out VLANs) will not be adversely affected.
169 #define PTYPE_HASH_SIZE (16)
170 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
172 static DEFINE_SPINLOCK(ptype_lock);
173 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
174 static struct list_head ptype_all __read_mostly; /* Taps */
177 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
180 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
182 * Writers must hold the rtnl semaphore while they loop through the
183 * dev_base_head list, and hold dev_base_lock for writing when they do the
184 * actual updates. This allows pure readers to access the list even
185 * while a writer is preparing to update it.
187 * To put it another way, dev_base_lock is held for writing only to
188 * protect against pure readers; the rtnl semaphore provides the
189 * protection against other writers.
191 * See, for example usages, register_netdevice() and
192 * unregister_netdevice(), which must be called with the rtnl
195 DEFINE_RWLOCK(dev_base_lock);
196 EXPORT_SYMBOL(dev_base_lock);
198 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
200 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
201 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
204 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
206 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
209 static inline void rps_lock(struct softnet_data *sd)
212 spin_lock(&sd->input_pkt_queue.lock);
216 static inline void rps_unlock(struct softnet_data *sd)
219 spin_unlock(&sd->input_pkt_queue.lock);
223 /* Device list insertion */
224 static int list_netdevice(struct net_device *dev)
226 struct net *net = dev_net(dev);
230 write_lock_bh(&dev_base_lock);
231 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
232 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
233 hlist_add_head_rcu(&dev->index_hlist,
234 dev_index_hash(net, dev->ifindex));
235 write_unlock_bh(&dev_base_lock);
239 /* Device list removal
240 * caller must respect a RCU grace period before freeing/reusing dev
242 static void unlist_netdevice(struct net_device *dev)
246 /* Unlink dev from the device chain */
247 write_lock_bh(&dev_base_lock);
248 list_del_rcu(&dev->dev_list);
249 hlist_del_rcu(&dev->name_hlist);
250 hlist_del_rcu(&dev->index_hlist);
251 write_unlock_bh(&dev_base_lock);
258 static RAW_NOTIFIER_HEAD(netdev_chain);
261 * Device drivers call our routines to queue packets here. We empty the
262 * queue in the local softnet handler.
265 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
266 EXPORT_PER_CPU_SYMBOL(softnet_data);
268 #ifdef CONFIG_LOCKDEP
270 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
271 * according to dev->type
273 static const unsigned short netdev_lock_type[] =
274 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
275 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
276 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
277 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
278 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
279 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
280 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
281 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
282 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
283 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
284 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
285 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
286 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
287 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
288 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
289 ARPHRD_VOID, ARPHRD_NONE};
291 static const char *const netdev_lock_name[] =
292 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
293 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
294 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
295 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
296 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
297 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
298 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
299 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
300 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
301 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
302 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
303 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
304 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
305 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
306 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
307 "_xmit_VOID", "_xmit_NONE"};
309 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
310 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
316 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
317 if (netdev_lock_type[i] == dev_type)
319 /* the last key is used by default */
320 return ARRAY_SIZE(netdev_lock_type) - 1;
323 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
324 unsigned short dev_type)
328 i = netdev_lock_pos(dev_type);
329 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
330 netdev_lock_name[i]);
333 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
337 i = netdev_lock_pos(dev->type);
338 lockdep_set_class_and_name(&dev->addr_list_lock,
339 &netdev_addr_lock_key[i],
340 netdev_lock_name[i]);
343 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
344 unsigned short dev_type)
347 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
352 /*******************************************************************************
354 Protocol management and registration routines
356 *******************************************************************************/
359 * Add a protocol ID to the list. Now that the input handler is
360 * smarter we can dispense with all the messy stuff that used to be
363 * BEWARE!!! Protocol handlers, mangling input packets,
364 * MUST BE last in hash buckets and checking protocol handlers
365 * MUST start from promiscuous ptype_all chain in net_bh.
366 * It is true now, do not change it.
367 * Explanation follows: if protocol handler, mangling packet, will
368 * be the first on list, it is not able to sense, that packet
369 * is cloned and should be copied-on-write, so that it will
370 * change it and subsequent readers will get broken packet.
374 static inline struct list_head *ptype_head(const struct packet_type *pt)
376 if (pt->type == htons(ETH_P_ALL))
379 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
383 * dev_add_pack - add packet handler
384 * @pt: packet type declaration
386 * Add a protocol handler to the networking stack. The passed &packet_type
387 * is linked into kernel lists and may not be freed until it has been
388 * removed from the kernel lists.
390 * This call does not sleep therefore it can not
391 * guarantee all CPU's that are in middle of receiving packets
392 * will see the new packet type (until the next received packet).
395 void dev_add_pack(struct packet_type *pt)
397 struct list_head *head = ptype_head(pt);
399 spin_lock(&ptype_lock);
400 list_add_rcu(&pt->list, head);
401 spin_unlock(&ptype_lock);
403 EXPORT_SYMBOL(dev_add_pack);
406 * __dev_remove_pack - remove packet handler
407 * @pt: packet type declaration
409 * Remove a protocol handler that was previously added to the kernel
410 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
411 * from the kernel lists and can be freed or reused once this function
414 * The packet type might still be in use by receivers
415 * and must not be freed until after all the CPU's have gone
416 * through a quiescent state.
418 void __dev_remove_pack(struct packet_type *pt)
420 struct list_head *head = ptype_head(pt);
421 struct packet_type *pt1;
423 spin_lock(&ptype_lock);
425 list_for_each_entry(pt1, head, list) {
427 list_del_rcu(&pt->list);
432 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
434 spin_unlock(&ptype_lock);
436 EXPORT_SYMBOL(__dev_remove_pack);
439 * dev_remove_pack - remove packet handler
440 * @pt: packet type declaration
442 * Remove a protocol handler that was previously added to the kernel
443 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
444 * from the kernel lists and can be freed or reused once this function
447 * This call sleeps to guarantee that no CPU is looking at the packet
450 void dev_remove_pack(struct packet_type *pt)
452 __dev_remove_pack(pt);
456 EXPORT_SYMBOL(dev_remove_pack);
458 /******************************************************************************
460 Device Boot-time Settings Routines
462 *******************************************************************************/
464 /* Boot time configuration table */
465 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
468 * netdev_boot_setup_add - add new setup entry
469 * @name: name of the device
470 * @map: configured settings for the device
472 * Adds new setup entry to the dev_boot_setup list. The function
473 * returns 0 on error and 1 on success. This is a generic routine to
476 static int netdev_boot_setup_add(char *name, struct ifmap *map)
478 struct netdev_boot_setup *s;
482 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
483 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
484 memset(s[i].name, 0, sizeof(s[i].name));
485 strlcpy(s[i].name, name, IFNAMSIZ);
486 memcpy(&s[i].map, map, sizeof(s[i].map));
491 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
495 * netdev_boot_setup_check - check boot time settings
496 * @dev: the netdevice
498 * Check boot time settings for the device.
499 * The found settings are set for the device to be used
500 * later in the device probing.
501 * Returns 0 if no settings found, 1 if they are.
503 int netdev_boot_setup_check(struct net_device *dev)
505 struct netdev_boot_setup *s = dev_boot_setup;
508 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
509 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
510 !strcmp(dev->name, s[i].name)) {
511 dev->irq = s[i].map.irq;
512 dev->base_addr = s[i].map.base_addr;
513 dev->mem_start = s[i].map.mem_start;
514 dev->mem_end = s[i].map.mem_end;
520 EXPORT_SYMBOL(netdev_boot_setup_check);
524 * netdev_boot_base - get address from boot time settings
525 * @prefix: prefix for network device
526 * @unit: id for network device
528 * Check boot time settings for the base address of device.
529 * The found settings are set for the device to be used
530 * later in the device probing.
531 * Returns 0 if no settings found.
533 unsigned long netdev_boot_base(const char *prefix, int unit)
535 const struct netdev_boot_setup *s = dev_boot_setup;
539 sprintf(name, "%s%d", prefix, unit);
542 * If device already registered then return base of 1
543 * to indicate not to probe for this interface
545 if (__dev_get_by_name(&init_net, name))
548 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
549 if (!strcmp(name, s[i].name))
550 return s[i].map.base_addr;
555 * Saves at boot time configured settings for any netdevice.
557 int __init netdev_boot_setup(char *str)
562 str = get_options(str, ARRAY_SIZE(ints), ints);
567 memset(&map, 0, sizeof(map));
571 map.base_addr = ints[2];
573 map.mem_start = ints[3];
575 map.mem_end = ints[4];
577 /* Add new entry to the list */
578 return netdev_boot_setup_add(str, &map);
581 __setup("netdev=", netdev_boot_setup);
583 /*******************************************************************************
585 Device Interface Subroutines
587 *******************************************************************************/
590 * __dev_get_by_name - find a device by its name
591 * @net: the applicable net namespace
592 * @name: name to find
594 * Find an interface by name. Must be called under RTNL semaphore
595 * or @dev_base_lock. If the name is found a pointer to the device
596 * is returned. If the name is not found then %NULL is returned. The
597 * reference counters are not incremented so the caller must be
598 * careful with locks.
601 struct net_device *__dev_get_by_name(struct net *net, const char *name)
603 struct hlist_node *p;
604 struct net_device *dev;
605 struct hlist_head *head = dev_name_hash(net, name);
607 hlist_for_each_entry(dev, p, head, name_hlist)
608 if (!strncmp(dev->name, name, IFNAMSIZ))
613 EXPORT_SYMBOL(__dev_get_by_name);
616 * dev_get_by_name_rcu - find a device by its name
617 * @net: the applicable net namespace
618 * @name: name to find
620 * Find an interface by name.
621 * If the name is found a pointer to the device is returned.
622 * If the name is not found then %NULL is returned.
623 * The reference counters are not incremented so the caller must be
624 * careful with locks. The caller must hold RCU lock.
627 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
629 struct hlist_node *p;
630 struct net_device *dev;
631 struct hlist_head *head = dev_name_hash(net, name);
633 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
634 if (!strncmp(dev->name, name, IFNAMSIZ))
639 EXPORT_SYMBOL(dev_get_by_name_rcu);
642 * dev_get_by_name - find a device by its name
643 * @net: the applicable net namespace
644 * @name: name to find
646 * Find an interface by name. This can be called from any
647 * context and does its own locking. The returned handle has
648 * the usage count incremented and the caller must use dev_put() to
649 * release it when it is no longer needed. %NULL is returned if no
650 * matching device is found.
653 struct net_device *dev_get_by_name(struct net *net, const char *name)
655 struct net_device *dev;
658 dev = dev_get_by_name_rcu(net, name);
664 EXPORT_SYMBOL(dev_get_by_name);
667 * __dev_get_by_index - find a device by its ifindex
668 * @net: the applicable net namespace
669 * @ifindex: index of device
671 * Search for an interface by index. Returns %NULL if the device
672 * is not found or a pointer to the device. The device has not
673 * had its reference counter increased so the caller must be careful
674 * about locking. The caller must hold either the RTNL semaphore
678 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
680 struct hlist_node *p;
681 struct net_device *dev;
682 struct hlist_head *head = dev_index_hash(net, ifindex);
684 hlist_for_each_entry(dev, p, head, index_hlist)
685 if (dev->ifindex == ifindex)
690 EXPORT_SYMBOL(__dev_get_by_index);
693 * dev_get_by_index_rcu - find a device by its ifindex
694 * @net: the applicable net namespace
695 * @ifindex: index of device
697 * Search for an interface by index. Returns %NULL if the device
698 * is not found or a pointer to the device. The device has not
699 * had its reference counter increased so the caller must be careful
700 * about locking. The caller must hold RCU lock.
703 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
705 struct hlist_node *p;
706 struct net_device *dev;
707 struct hlist_head *head = dev_index_hash(net, ifindex);
709 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
710 if (dev->ifindex == ifindex)
715 EXPORT_SYMBOL(dev_get_by_index_rcu);
719 * dev_get_by_index - find a device by its ifindex
720 * @net: the applicable net namespace
721 * @ifindex: index of device
723 * Search for an interface by index. Returns NULL if the device
724 * is not found or a pointer to the device. The device returned has
725 * had a reference added and the pointer is safe until the user calls
726 * dev_put to indicate they have finished with it.
729 struct net_device *dev_get_by_index(struct net *net, int ifindex)
731 struct net_device *dev;
734 dev = dev_get_by_index_rcu(net, ifindex);
740 EXPORT_SYMBOL(dev_get_by_index);
743 * dev_getbyhwaddr - find a device by its hardware address
744 * @net: the applicable net namespace
745 * @type: media type of device
746 * @ha: hardware address
748 * Search for an interface by MAC address. Returns NULL if the device
749 * is not found or a pointer to the device. The caller must hold the
750 * rtnl semaphore. The returned device has not had its ref count increased
751 * and the caller must therefore be careful about locking
754 * If the API was consistent this would be __dev_get_by_hwaddr
757 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
759 struct net_device *dev;
763 for_each_netdev(net, dev)
764 if (dev->type == type &&
765 !memcmp(dev->dev_addr, ha, dev->addr_len))
770 EXPORT_SYMBOL(dev_getbyhwaddr);
772 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
774 struct net_device *dev;
777 for_each_netdev(net, dev)
778 if (dev->type == type)
783 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
785 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
787 struct net_device *dev, *ret = NULL;
790 for_each_netdev_rcu(net, dev)
791 if (dev->type == type) {
799 EXPORT_SYMBOL(dev_getfirstbyhwtype);
802 * dev_get_by_flags_rcu - find any device with given flags
803 * @net: the applicable net namespace
804 * @if_flags: IFF_* values
805 * @mask: bitmask of bits in if_flags to check
807 * Search for any interface with the given flags. Returns NULL if a device
808 * is not found or a pointer to the device. Must be called inside
809 * rcu_read_lock(), and result refcount is unchanged.
812 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
815 struct net_device *dev, *ret;
818 for_each_netdev_rcu(net, dev) {
819 if (((dev->flags ^ if_flags) & mask) == 0) {
826 EXPORT_SYMBOL(dev_get_by_flags_rcu);
829 * dev_valid_name - check if name is okay for network device
832 * Network device names need to be valid file names to
833 * to allow sysfs to work. We also disallow any kind of
836 int dev_valid_name(const char *name)
840 if (strlen(name) >= IFNAMSIZ)
842 if (!strcmp(name, ".") || !strcmp(name, ".."))
846 if (*name == '/' || isspace(*name))
852 EXPORT_SYMBOL(dev_valid_name);
855 * __dev_alloc_name - allocate a name for a device
856 * @net: network namespace to allocate the device name in
857 * @name: name format string
858 * @buf: scratch buffer and result name string
860 * Passed a format string - eg "lt%d" it will try and find a suitable
861 * id. It scans list of devices to build up a free map, then chooses
862 * the first empty slot. The caller must hold the dev_base or rtnl lock
863 * while allocating the name and adding the device in order to avoid
865 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
866 * Returns the number of the unit assigned or a negative errno code.
869 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
873 const int max_netdevices = 8*PAGE_SIZE;
874 unsigned long *inuse;
875 struct net_device *d;
877 p = strnchr(name, IFNAMSIZ-1, '%');
880 * Verify the string as this thing may have come from
881 * the user. There must be either one "%d" and no other "%"
884 if (p[1] != 'd' || strchr(p + 2, '%'))
887 /* Use one page as a bit array of possible slots */
888 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
892 for_each_netdev(net, d) {
893 if (!sscanf(d->name, name, &i))
895 if (i < 0 || i >= max_netdevices)
898 /* avoid cases where sscanf is not exact inverse of printf */
899 snprintf(buf, IFNAMSIZ, name, i);
900 if (!strncmp(buf, d->name, IFNAMSIZ))
904 i = find_first_zero_bit(inuse, max_netdevices);
905 free_page((unsigned long) inuse);
909 snprintf(buf, IFNAMSIZ, name, i);
910 if (!__dev_get_by_name(net, buf))
913 /* It is possible to run out of possible slots
914 * when the name is long and there isn't enough space left
915 * for the digits, or if all bits are used.
921 * dev_alloc_name - allocate a name for a device
923 * @name: name format string
925 * Passed a format string - eg "lt%d" it will try and find a suitable
926 * id. It scans list of devices to build up a free map, then chooses
927 * the first empty slot. The caller must hold the dev_base or rtnl lock
928 * while allocating the name and adding the device in order to avoid
930 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
931 * Returns the number of the unit assigned or a negative errno code.
934 int dev_alloc_name(struct net_device *dev, const char *name)
940 BUG_ON(!dev_net(dev));
942 ret = __dev_alloc_name(net, name, buf);
944 strlcpy(dev->name, buf, IFNAMSIZ);
947 EXPORT_SYMBOL(dev_alloc_name);
949 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
953 BUG_ON(!dev_net(dev));
956 if (!dev_valid_name(name))
959 if (fmt && strchr(name, '%'))
960 return dev_alloc_name(dev, name);
961 else if (__dev_get_by_name(net, name))
963 else if (dev->name != name)
964 strlcpy(dev->name, name, IFNAMSIZ);
970 * dev_change_name - change name of a device
972 * @newname: name (or format string) must be at least IFNAMSIZ
974 * Change name of a device, can pass format strings "eth%d".
977 int dev_change_name(struct net_device *dev, const char *newname)
979 char oldname[IFNAMSIZ];
985 BUG_ON(!dev_net(dev));
988 if (dev->flags & IFF_UP)
991 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
994 memcpy(oldname, dev->name, IFNAMSIZ);
996 err = dev_get_valid_name(dev, newname, 1);
1001 ret = device_rename(&dev->dev, dev->name);
1003 memcpy(dev->name, oldname, IFNAMSIZ);
1007 write_lock_bh(&dev_base_lock);
1008 hlist_del(&dev->name_hlist);
1009 write_unlock_bh(&dev_base_lock);
1013 write_lock_bh(&dev_base_lock);
1014 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1015 write_unlock_bh(&dev_base_lock);
1017 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1018 ret = notifier_to_errno(ret);
1021 /* err >= 0 after dev_alloc_name() or stores the first errno */
1024 memcpy(dev->name, oldname, IFNAMSIZ);
1028 "%s: name change rollback failed: %d.\n",
1037 * dev_set_alias - change ifalias of a device
1039 * @alias: name up to IFALIASZ
1040 * @len: limit of bytes to copy from info
1042 * Set ifalias for a device,
1044 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1048 if (len >= IFALIASZ)
1053 kfree(dev->ifalias);
1054 dev->ifalias = NULL;
1059 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1063 strlcpy(dev->ifalias, alias, len+1);
1069 * netdev_features_change - device changes features
1070 * @dev: device to cause notification
1072 * Called to indicate a device has changed features.
1074 void netdev_features_change(struct net_device *dev)
1076 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1078 EXPORT_SYMBOL(netdev_features_change);
1081 * netdev_state_change - device changes state
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed state. This function calls
1085 * the notifier chains for netdev_chain and sends a NEWLINK message
1086 * to the routing socket.
1088 void netdev_state_change(struct net_device *dev)
1090 if (dev->flags & IFF_UP) {
1091 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1092 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1095 EXPORT_SYMBOL(netdev_state_change);
1097 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1099 return call_netdevice_notifiers(event, dev);
1101 EXPORT_SYMBOL(netdev_bonding_change);
1104 * dev_load - load a network module
1105 * @net: the applicable net namespace
1106 * @name: name of interface
1108 * If a network interface is not present and the process has suitable
1109 * privileges this function loads the module. If module loading is not
1110 * available in this kernel then it becomes a nop.
1113 void dev_load(struct net *net, const char *name)
1115 struct net_device *dev;
1118 dev = dev_get_by_name_rcu(net, name);
1121 if (!dev && capable(CAP_NET_ADMIN))
1122 request_module("%s", name);
1124 EXPORT_SYMBOL(dev_load);
1126 static int __dev_open(struct net_device *dev)
1128 const struct net_device_ops *ops = dev->netdev_ops;
1134 * Is it even present?
1136 if (!netif_device_present(dev))
1139 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1140 ret = notifier_to_errno(ret);
1145 * Call device private open method
1147 set_bit(__LINK_STATE_START, &dev->state);
1149 if (ops->ndo_validate_addr)
1150 ret = ops->ndo_validate_addr(dev);
1152 if (!ret && ops->ndo_open)
1153 ret = ops->ndo_open(dev);
1156 * If it went open OK then:
1160 clear_bit(__LINK_STATE_START, &dev->state);
1165 dev->flags |= IFF_UP;
1170 net_dmaengine_get();
1173 * Initialize multicasting status
1175 dev_set_rx_mode(dev);
1178 * Wakeup transmit queue engine
1187 * dev_open - prepare an interface for use.
1188 * @dev: device to open
1190 * Takes a device from down to up state. The device's private open
1191 * function is invoked and then the multicast lists are loaded. Finally
1192 * the device is moved into the up state and a %NETDEV_UP message is
1193 * sent to the netdev notifier chain.
1195 * Calling this function on an active interface is a nop. On a failure
1196 * a negative errno code is returned.
1198 int dev_open(struct net_device *dev)
1205 if (dev->flags & IFF_UP)
1211 ret = __dev_open(dev);
1216 * ... and announce new interface.
1218 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1219 call_netdevice_notifiers(NETDEV_UP, dev);
1223 EXPORT_SYMBOL(dev_open);
1225 static int __dev_close(struct net_device *dev)
1227 const struct net_device_ops *ops = dev->netdev_ops;
1233 * Tell people we are going down, so that they can
1234 * prepare to death, when device is still operating.
1236 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1238 clear_bit(__LINK_STATE_START, &dev->state);
1240 /* Synchronize to scheduled poll. We cannot touch poll list,
1241 * it can be even on different cpu. So just clear netif_running().
1243 * dev->stop() will invoke napi_disable() on all of it's
1244 * napi_struct instances on this device.
1246 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1248 dev_deactivate(dev);
1251 * Call the device specific close. This cannot fail.
1252 * Only if device is UP
1254 * We allow it to be called even after a DETACH hot-plug
1261 * Device is now down.
1264 dev->flags &= ~IFF_UP;
1269 net_dmaengine_put();
1275 * dev_close - shutdown an interface.
1276 * @dev: device to shutdown
1278 * This function moves an active device into down state. A
1279 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1280 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1283 int dev_close(struct net_device *dev)
1285 if (!(dev->flags & IFF_UP))
1291 * Tell people we are down
1293 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1294 call_netdevice_notifiers(NETDEV_DOWN, dev);
1298 EXPORT_SYMBOL(dev_close);
1302 * dev_disable_lro - disable Large Receive Offload on a device
1305 * Disable Large Receive Offload (LRO) on a net device. Must be
1306 * called under RTNL. This is needed if received packets may be
1307 * forwarded to another interface.
1309 void dev_disable_lro(struct net_device *dev)
1311 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1312 dev->ethtool_ops->set_flags) {
1313 u32 flags = dev->ethtool_ops->get_flags(dev);
1314 if (flags & ETH_FLAG_LRO) {
1315 flags &= ~ETH_FLAG_LRO;
1316 dev->ethtool_ops->set_flags(dev, flags);
1319 WARN_ON(dev->features & NETIF_F_LRO);
1321 EXPORT_SYMBOL(dev_disable_lro);
1324 static int dev_boot_phase = 1;
1327 * Device change register/unregister. These are not inline or static
1328 * as we export them to the world.
1332 * register_netdevice_notifier - register a network notifier block
1335 * Register a notifier to be called when network device events occur.
1336 * The notifier passed is linked into the kernel structures and must
1337 * not be reused until it has been unregistered. A negative errno code
1338 * is returned on a failure.
1340 * When registered all registration and up events are replayed
1341 * to the new notifier to allow device to have a race free
1342 * view of the network device list.
1345 int register_netdevice_notifier(struct notifier_block *nb)
1347 struct net_device *dev;
1348 struct net_device *last;
1353 err = raw_notifier_chain_register(&netdev_chain, nb);
1359 for_each_netdev(net, dev) {
1360 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1361 err = notifier_to_errno(err);
1365 if (!(dev->flags & IFF_UP))
1368 nb->notifier_call(nb, NETDEV_UP, dev);
1379 for_each_netdev(net, dev) {
1383 if (dev->flags & IFF_UP) {
1384 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1385 nb->notifier_call(nb, NETDEV_DOWN, dev);
1387 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1388 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1392 raw_notifier_chain_unregister(&netdev_chain, nb);
1395 EXPORT_SYMBOL(register_netdevice_notifier);
1398 * unregister_netdevice_notifier - unregister a network notifier block
1401 * Unregister a notifier previously registered by
1402 * register_netdevice_notifier(). The notifier is unlinked into the
1403 * kernel structures and may then be reused. A negative errno code
1404 * is returned on a failure.
1407 int unregister_netdevice_notifier(struct notifier_block *nb)
1412 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1416 EXPORT_SYMBOL(unregister_netdevice_notifier);
1419 * call_netdevice_notifiers - call all network notifier blocks
1420 * @val: value passed unmodified to notifier function
1421 * @dev: net_device pointer passed unmodified to notifier function
1423 * Call all network notifier blocks. Parameters and return value
1424 * are as for raw_notifier_call_chain().
1427 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1430 return raw_notifier_call_chain(&netdev_chain, val, dev);
1433 /* When > 0 there are consumers of rx skb time stamps */
1434 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1436 void net_enable_timestamp(void)
1438 atomic_inc(&netstamp_needed);
1440 EXPORT_SYMBOL(net_enable_timestamp);
1442 void net_disable_timestamp(void)
1444 atomic_dec(&netstamp_needed);
1446 EXPORT_SYMBOL(net_disable_timestamp);
1448 static inline void net_timestamp_set(struct sk_buff *skb)
1450 if (atomic_read(&netstamp_needed))
1451 __net_timestamp(skb);
1453 skb->tstamp.tv64 = 0;
1456 static inline void net_timestamp_check(struct sk_buff *skb)
1458 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1459 __net_timestamp(skb);
1463 * dev_forward_skb - loopback an skb to another netif
1465 * @dev: destination network device
1466 * @skb: buffer to forward
1469 * NET_RX_SUCCESS (no congestion)
1470 * NET_RX_DROP (packet was dropped, but freed)
1472 * dev_forward_skb can be used for injecting an skb from the
1473 * start_xmit function of one device into the receive queue
1474 * of another device.
1476 * The receiving device may be in another namespace, so
1477 * we have to clear all information in the skb that could
1478 * impact namespace isolation.
1480 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1485 if (!(dev->flags & IFF_UP) ||
1486 (skb->len > (dev->mtu + dev->hard_header_len))) {
1490 skb_set_dev(skb, dev);
1491 skb->tstamp.tv64 = 0;
1492 skb->pkt_type = PACKET_HOST;
1493 skb->protocol = eth_type_trans(skb, dev);
1494 return netif_rx(skb);
1496 EXPORT_SYMBOL_GPL(dev_forward_skb);
1499 * Support routine. Sends outgoing frames to any network
1500 * taps currently in use.
1503 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1505 struct packet_type *ptype;
1507 #ifdef CONFIG_NET_CLS_ACT
1508 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1509 net_timestamp_set(skb);
1511 net_timestamp_set(skb);
1515 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1516 /* Never send packets back to the socket
1517 * they originated from - MvS (miquels@drinkel.ow.org)
1519 if ((ptype->dev == dev || !ptype->dev) &&
1520 (ptype->af_packet_priv == NULL ||
1521 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1522 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1526 /* skb->nh should be correctly
1527 set by sender, so that the second statement is
1528 just protection against buggy protocols.
1530 skb_reset_mac_header(skb2);
1532 if (skb_network_header(skb2) < skb2->data ||
1533 skb2->network_header > skb2->tail) {
1534 if (net_ratelimit())
1535 printk(KERN_CRIT "protocol %04x is "
1537 ntohs(skb2->protocol),
1539 skb_reset_network_header(skb2);
1542 skb2->transport_header = skb2->network_header;
1543 skb2->pkt_type = PACKET_OUTGOING;
1544 ptype->func(skb2, skb->dev, ptype, skb->dev);
1551 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1552 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1554 void netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1556 unsigned int real_num = dev->real_num_tx_queues;
1558 if (unlikely(txq > dev->num_tx_queues))
1560 else if (txq > real_num)
1561 dev->real_num_tx_queues = txq;
1562 else if (txq < real_num) {
1563 dev->real_num_tx_queues = txq;
1564 qdisc_reset_all_tx_gt(dev, txq);
1567 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1569 static inline void __netif_reschedule(struct Qdisc *q)
1571 struct softnet_data *sd;
1572 unsigned long flags;
1574 local_irq_save(flags);
1575 sd = &__get_cpu_var(softnet_data);
1576 q->next_sched = NULL;
1577 *sd->output_queue_tailp = q;
1578 sd->output_queue_tailp = &q->next_sched;
1579 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1580 local_irq_restore(flags);
1583 void __netif_schedule(struct Qdisc *q)
1585 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1586 __netif_reschedule(q);
1588 EXPORT_SYMBOL(__netif_schedule);
1590 void dev_kfree_skb_irq(struct sk_buff *skb)
1592 if (atomic_dec_and_test(&skb->users)) {
1593 struct softnet_data *sd;
1594 unsigned long flags;
1596 local_irq_save(flags);
1597 sd = &__get_cpu_var(softnet_data);
1598 skb->next = sd->completion_queue;
1599 sd->completion_queue = skb;
1600 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1601 local_irq_restore(flags);
1604 EXPORT_SYMBOL(dev_kfree_skb_irq);
1606 void dev_kfree_skb_any(struct sk_buff *skb)
1608 if (in_irq() || irqs_disabled())
1609 dev_kfree_skb_irq(skb);
1613 EXPORT_SYMBOL(dev_kfree_skb_any);
1617 * netif_device_detach - mark device as removed
1618 * @dev: network device
1620 * Mark device as removed from system and therefore no longer available.
1622 void netif_device_detach(struct net_device *dev)
1624 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1625 netif_running(dev)) {
1626 netif_tx_stop_all_queues(dev);
1629 EXPORT_SYMBOL(netif_device_detach);
1632 * netif_device_attach - mark device as attached
1633 * @dev: network device
1635 * Mark device as attached from system and restart if needed.
1637 void netif_device_attach(struct net_device *dev)
1639 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1640 netif_running(dev)) {
1641 netif_tx_wake_all_queues(dev);
1642 __netdev_watchdog_up(dev);
1645 EXPORT_SYMBOL(netif_device_attach);
1647 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1649 return ((features & NETIF_F_GEN_CSUM) ||
1650 ((features & NETIF_F_IP_CSUM) &&
1651 protocol == htons(ETH_P_IP)) ||
1652 ((features & NETIF_F_IPV6_CSUM) &&
1653 protocol == htons(ETH_P_IPV6)) ||
1654 ((features & NETIF_F_FCOE_CRC) &&
1655 protocol == htons(ETH_P_FCOE)));
1658 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1660 if (can_checksum_protocol(dev->features, skb->protocol))
1663 if (skb->protocol == htons(ETH_P_8021Q)) {
1664 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1665 if (can_checksum_protocol(dev->features & dev->vlan_features,
1666 veh->h_vlan_encapsulated_proto))
1674 * skb_dev_set -- assign a new device to a buffer
1675 * @skb: buffer for the new device
1676 * @dev: network device
1678 * If an skb is owned by a device already, we have to reset
1679 * all data private to the namespace a device belongs to
1680 * before assigning it a new device.
1682 #ifdef CONFIG_NET_NS
1683 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1686 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1689 skb_init_secmark(skb);
1693 skb->ipvs_property = 0;
1694 #ifdef CONFIG_NET_SCHED
1700 EXPORT_SYMBOL(skb_set_dev);
1701 #endif /* CONFIG_NET_NS */
1704 * Invalidate hardware checksum when packet is to be mangled, and
1705 * complete checksum manually on outgoing path.
1707 int skb_checksum_help(struct sk_buff *skb)
1710 int ret = 0, offset;
1712 if (skb->ip_summed == CHECKSUM_COMPLETE)
1713 goto out_set_summed;
1715 if (unlikely(skb_shinfo(skb)->gso_size)) {
1716 /* Let GSO fix up the checksum. */
1717 goto out_set_summed;
1720 offset = skb->csum_start - skb_headroom(skb);
1721 BUG_ON(offset >= skb_headlen(skb));
1722 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1724 offset += skb->csum_offset;
1725 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1727 if (skb_cloned(skb) &&
1728 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1729 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1734 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1736 skb->ip_summed = CHECKSUM_NONE;
1740 EXPORT_SYMBOL(skb_checksum_help);
1743 * skb_gso_segment - Perform segmentation on skb.
1744 * @skb: buffer to segment
1745 * @features: features for the output path (see dev->features)
1747 * This function segments the given skb and returns a list of segments.
1749 * It may return NULL if the skb requires no segmentation. This is
1750 * only possible when GSO is used for verifying header integrity.
1752 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1754 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1755 struct packet_type *ptype;
1756 __be16 type = skb->protocol;
1759 skb_reset_mac_header(skb);
1760 skb->mac_len = skb->network_header - skb->mac_header;
1761 __skb_pull(skb, skb->mac_len);
1763 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1764 struct net_device *dev = skb->dev;
1765 struct ethtool_drvinfo info = {};
1767 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1768 dev->ethtool_ops->get_drvinfo(dev, &info);
1770 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1772 info.driver, dev ? dev->features : 0L,
1773 skb->sk ? skb->sk->sk_route_caps : 0L,
1774 skb->len, skb->data_len, skb->ip_summed);
1776 if (skb_header_cloned(skb) &&
1777 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1778 return ERR_PTR(err);
1782 list_for_each_entry_rcu(ptype,
1783 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1784 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1785 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1786 err = ptype->gso_send_check(skb);
1787 segs = ERR_PTR(err);
1788 if (err || skb_gso_ok(skb, features))
1790 __skb_push(skb, (skb->data -
1791 skb_network_header(skb)));
1793 segs = ptype->gso_segment(skb, features);
1799 __skb_push(skb, skb->data - skb_mac_header(skb));
1803 EXPORT_SYMBOL(skb_gso_segment);
1805 /* Take action when hardware reception checksum errors are detected. */
1807 void netdev_rx_csum_fault(struct net_device *dev)
1809 if (net_ratelimit()) {
1810 printk(KERN_ERR "%s: hw csum failure.\n",
1811 dev ? dev->name : "<unknown>");
1815 EXPORT_SYMBOL(netdev_rx_csum_fault);
1818 /* Actually, we should eliminate this check as soon as we know, that:
1819 * 1. IOMMU is present and allows to map all the memory.
1820 * 2. No high memory really exists on this machine.
1823 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1825 #ifdef CONFIG_HIGHMEM
1827 if (!(dev->features & NETIF_F_HIGHDMA)) {
1828 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1829 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1833 if (PCI_DMA_BUS_IS_PHYS) {
1834 struct device *pdev = dev->dev.parent;
1838 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1839 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1840 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1849 void (*destructor)(struct sk_buff *skb);
1852 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1854 static void dev_gso_skb_destructor(struct sk_buff *skb)
1856 struct dev_gso_cb *cb;
1859 struct sk_buff *nskb = skb->next;
1861 skb->next = nskb->next;
1864 } while (skb->next);
1866 cb = DEV_GSO_CB(skb);
1868 cb->destructor(skb);
1872 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1873 * @skb: buffer to segment
1875 * This function segments the given skb and stores the list of segments
1878 static int dev_gso_segment(struct sk_buff *skb)
1880 struct net_device *dev = skb->dev;
1881 struct sk_buff *segs;
1882 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1885 segs = skb_gso_segment(skb, features);
1887 /* Verifying header integrity only. */
1892 return PTR_ERR(segs);
1895 DEV_GSO_CB(skb)->destructor = skb->destructor;
1896 skb->destructor = dev_gso_skb_destructor;
1902 * Try to orphan skb early, right before transmission by the device.
1903 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1904 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1906 static inline void skb_orphan_try(struct sk_buff *skb)
1908 struct sock *sk = skb->sk;
1910 if (sk && !skb_shinfo(skb)->tx_flags) {
1911 /* skb_tx_hash() wont be able to get sk.
1912 * We copy sk_hash into skb->rxhash
1915 skb->rxhash = sk->sk_hash;
1921 * Returns true if either:
1922 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1923 * 2. skb is fragmented and the device does not support SG, or if
1924 * at least one of fragments is in highmem and device does not
1925 * support DMA from it.
1927 static inline int skb_needs_linearize(struct sk_buff *skb,
1928 struct net_device *dev)
1930 return skb_is_nonlinear(skb) &&
1931 ((skb_has_frag_list(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
1932 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
1933 illegal_highdma(dev, skb))));
1936 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1937 struct netdev_queue *txq)
1939 const struct net_device_ops *ops = dev->netdev_ops;
1940 int rc = NETDEV_TX_OK;
1942 if (likely(!skb->next)) {
1943 if (!list_empty(&ptype_all))
1944 dev_queue_xmit_nit(skb, dev);
1947 * If device doesnt need skb->dst, release it right now while
1948 * its hot in this cpu cache
1950 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1953 skb_orphan_try(skb);
1955 if (netif_needs_gso(dev, skb)) {
1956 if (unlikely(dev_gso_segment(skb)))
1961 if (skb_needs_linearize(skb, dev) &&
1962 __skb_linearize(skb))
1965 /* If packet is not checksummed and device does not
1966 * support checksumming for this protocol, complete
1967 * checksumming here.
1969 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1970 skb_set_transport_header(skb, skb->csum_start -
1972 if (!dev_can_checksum(dev, skb) &&
1973 skb_checksum_help(skb))
1978 rc = ops->ndo_start_xmit(skb, dev);
1979 if (rc == NETDEV_TX_OK)
1980 txq_trans_update(txq);
1986 struct sk_buff *nskb = skb->next;
1988 skb->next = nskb->next;
1992 * If device doesnt need nskb->dst, release it right now while
1993 * its hot in this cpu cache
1995 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1998 rc = ops->ndo_start_xmit(nskb, dev);
1999 if (unlikely(rc != NETDEV_TX_OK)) {
2000 if (rc & ~NETDEV_TX_MASK)
2001 goto out_kfree_gso_skb;
2002 nskb->next = skb->next;
2006 txq_trans_update(txq);
2007 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2008 return NETDEV_TX_BUSY;
2009 } while (skb->next);
2012 if (likely(skb->next == NULL))
2013 skb->destructor = DEV_GSO_CB(skb)->destructor;
2019 static u32 hashrnd __read_mostly;
2021 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2025 if (skb_rx_queue_recorded(skb)) {
2026 hash = skb_get_rx_queue(skb);
2027 while (unlikely(hash >= dev->real_num_tx_queues))
2028 hash -= dev->real_num_tx_queues;
2032 if (skb->sk && skb->sk->sk_hash)
2033 hash = skb->sk->sk_hash;
2035 hash = (__force u16) skb->protocol ^ skb->rxhash;
2036 hash = jhash_1word(hash, hashrnd);
2038 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2040 EXPORT_SYMBOL(skb_tx_hash);
2042 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2044 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2045 if (net_ratelimit()) {
2046 pr_warning("%s selects TX queue %d, but "
2047 "real number of TX queues is %d\n",
2048 dev->name, queue_index, dev->real_num_tx_queues);
2055 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2056 struct sk_buff *skb)
2059 struct sock *sk = skb->sk;
2061 queue_index = sk_tx_queue_get(sk);
2062 if (queue_index < 0) {
2063 const struct net_device_ops *ops = dev->netdev_ops;
2065 if (ops->ndo_select_queue) {
2066 queue_index = ops->ndo_select_queue(dev, skb);
2067 queue_index = dev_cap_txqueue(dev, queue_index);
2070 if (dev->real_num_tx_queues > 1)
2071 queue_index = skb_tx_hash(dev, skb);
2074 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2076 if (dst && skb_dst(skb) == dst)
2077 sk_tx_queue_set(sk, queue_index);
2082 skb_set_queue_mapping(skb, queue_index);
2083 return netdev_get_tx_queue(dev, queue_index);
2086 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2087 struct net_device *dev,
2088 struct netdev_queue *txq)
2090 spinlock_t *root_lock = qdisc_lock(q);
2091 bool contended = qdisc_is_running(q);
2095 * Heuristic to force contended enqueues to serialize on a
2096 * separate lock before trying to get qdisc main lock.
2097 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2098 * and dequeue packets faster.
2100 if (unlikely(contended))
2101 spin_lock(&q->busylock);
2103 spin_lock(root_lock);
2104 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2107 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2108 qdisc_run_begin(q)) {
2110 * This is a work-conserving queue; there are no old skbs
2111 * waiting to be sent out; and the qdisc is not running -
2112 * xmit the skb directly.
2114 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2116 __qdisc_update_bstats(q, skb->len);
2117 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2118 if (unlikely(contended)) {
2119 spin_unlock(&q->busylock);
2126 rc = NET_XMIT_SUCCESS;
2129 rc = qdisc_enqueue_root(skb, q);
2130 if (qdisc_run_begin(q)) {
2131 if (unlikely(contended)) {
2132 spin_unlock(&q->busylock);
2138 spin_unlock(root_lock);
2139 if (unlikely(contended))
2140 spin_unlock(&q->busylock);
2145 * dev_queue_xmit - transmit a buffer
2146 * @skb: buffer to transmit
2148 * Queue a buffer for transmission to a network device. The caller must
2149 * have set the device and priority and built the buffer before calling
2150 * this function. The function can be called from an interrupt.
2152 * A negative errno code is returned on a failure. A success does not
2153 * guarantee the frame will be transmitted as it may be dropped due
2154 * to congestion or traffic shaping.
2156 * -----------------------------------------------------------------------------------
2157 * I notice this method can also return errors from the queue disciplines,
2158 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2161 * Regardless of the return value, the skb is consumed, so it is currently
2162 * difficult to retry a send to this method. (You can bump the ref count
2163 * before sending to hold a reference for retry if you are careful.)
2165 * When calling this method, interrupts MUST be enabled. This is because
2166 * the BH enable code must have IRQs enabled so that it will not deadlock.
2169 int dev_queue_xmit(struct sk_buff *skb)
2171 struct net_device *dev = skb->dev;
2172 struct netdev_queue *txq;
2176 /* Disable soft irqs for various locks below. Also
2177 * stops preemption for RCU.
2181 txq = dev_pick_tx(dev, skb);
2182 q = rcu_dereference_bh(txq->qdisc);
2184 #ifdef CONFIG_NET_CLS_ACT
2185 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2188 rc = __dev_xmit_skb(skb, q, dev, txq);
2192 /* The device has no queue. Common case for software devices:
2193 loopback, all the sorts of tunnels...
2195 Really, it is unlikely that netif_tx_lock protection is necessary
2196 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2198 However, it is possible, that they rely on protection
2201 Check this and shot the lock. It is not prone from deadlocks.
2202 Either shot noqueue qdisc, it is even simpler 8)
2204 if (dev->flags & IFF_UP) {
2205 int cpu = smp_processor_id(); /* ok because BHs are off */
2207 if (txq->xmit_lock_owner != cpu) {
2209 HARD_TX_LOCK(dev, txq, cpu);
2211 if (!netif_tx_queue_stopped(txq)) {
2212 rc = dev_hard_start_xmit(skb, dev, txq);
2213 if (dev_xmit_complete(rc)) {
2214 HARD_TX_UNLOCK(dev, txq);
2218 HARD_TX_UNLOCK(dev, txq);
2219 if (net_ratelimit())
2220 printk(KERN_CRIT "Virtual device %s asks to "
2221 "queue packet!\n", dev->name);
2223 /* Recursion is detected! It is possible,
2225 if (net_ratelimit())
2226 printk(KERN_CRIT "Dead loop on virtual device "
2227 "%s, fix it urgently!\n", dev->name);
2232 rcu_read_unlock_bh();
2237 rcu_read_unlock_bh();
2240 EXPORT_SYMBOL(dev_queue_xmit);
2243 /*=======================================================================
2245 =======================================================================*/
2247 int netdev_max_backlog __read_mostly = 1000;
2248 int netdev_tstamp_prequeue __read_mostly = 1;
2249 int netdev_budget __read_mostly = 300;
2250 int weight_p __read_mostly = 64; /* old backlog weight */
2252 /* Called with irq disabled */
2253 static inline void ____napi_schedule(struct softnet_data *sd,
2254 struct napi_struct *napi)
2256 list_add_tail(&napi->poll_list, &sd->poll_list);
2257 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2261 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2262 * and src/dst port numbers. Returns a non-zero hash number on success
2265 __u32 __skb_get_rxhash(struct sk_buff *skb)
2267 int nhoff, hash = 0, poff;
2268 struct ipv6hdr *ip6;
2271 u32 addr1, addr2, ihl;
2277 nhoff = skb_network_offset(skb);
2279 switch (skb->protocol) {
2280 case __constant_htons(ETH_P_IP):
2281 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2284 ip = (struct iphdr *) (skb->data + nhoff);
2285 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2288 ip_proto = ip->protocol;
2289 addr1 = (__force u32) ip->saddr;
2290 addr2 = (__force u32) ip->daddr;
2293 case __constant_htons(ETH_P_IPV6):
2294 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2297 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2298 ip_proto = ip6->nexthdr;
2299 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2300 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2308 poff = proto_ports_offset(ip_proto);
2310 nhoff += ihl * 4 + poff;
2311 if (pskb_may_pull(skb, nhoff + 4)) {
2312 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2313 if (ports.v16[1] < ports.v16[0])
2314 swap(ports.v16[0], ports.v16[1]);
2318 /* get a consistent hash (same value on both flow directions) */
2322 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2329 EXPORT_SYMBOL(__skb_get_rxhash);
2333 /* One global table that all flow-based protocols share. */
2334 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2335 EXPORT_SYMBOL(rps_sock_flow_table);
2338 * get_rps_cpu is called from netif_receive_skb and returns the target
2339 * CPU from the RPS map of the receiving queue for a given skb.
2340 * rcu_read_lock must be held on entry.
2342 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2343 struct rps_dev_flow **rflowp)
2345 struct netdev_rx_queue *rxqueue;
2346 struct rps_map *map;
2347 struct rps_dev_flow_table *flow_table;
2348 struct rps_sock_flow_table *sock_flow_table;
2352 if (skb_rx_queue_recorded(skb)) {
2353 u16 index = skb_get_rx_queue(skb);
2354 if (unlikely(index >= dev->num_rx_queues)) {
2355 WARN_ONCE(dev->num_rx_queues > 1, "%s received packet "
2356 "on queue %u, but number of RX queues is %u\n",
2357 dev->name, index, dev->num_rx_queues);
2360 rxqueue = dev->_rx + index;
2364 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2367 skb_reset_network_header(skb);
2368 if (!skb_get_rxhash(skb))
2371 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2372 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2373 if (flow_table && sock_flow_table) {
2375 struct rps_dev_flow *rflow;
2377 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2380 next_cpu = sock_flow_table->ents[skb->rxhash &
2381 sock_flow_table->mask];
2384 * If the desired CPU (where last recvmsg was done) is
2385 * different from current CPU (one in the rx-queue flow
2386 * table entry), switch if one of the following holds:
2387 * - Current CPU is unset (equal to RPS_NO_CPU).
2388 * - Current CPU is offline.
2389 * - The current CPU's queue tail has advanced beyond the
2390 * last packet that was enqueued using this table entry.
2391 * This guarantees that all previous packets for the flow
2392 * have been dequeued, thus preserving in order delivery.
2394 if (unlikely(tcpu != next_cpu) &&
2395 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2396 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2397 rflow->last_qtail)) >= 0)) {
2398 tcpu = rflow->cpu = next_cpu;
2399 if (tcpu != RPS_NO_CPU)
2400 rflow->last_qtail = per_cpu(softnet_data,
2401 tcpu).input_queue_head;
2403 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2410 map = rcu_dereference(rxqueue->rps_map);
2412 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2414 if (cpu_online(tcpu)) {
2424 /* Called from hardirq (IPI) context */
2425 static void rps_trigger_softirq(void *data)
2427 struct softnet_data *sd = data;
2429 ____napi_schedule(sd, &sd->backlog);
2433 #endif /* CONFIG_RPS */
2436 * Check if this softnet_data structure is another cpu one
2437 * If yes, queue it to our IPI list and return 1
2440 static int rps_ipi_queued(struct softnet_data *sd)
2443 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2446 sd->rps_ipi_next = mysd->rps_ipi_list;
2447 mysd->rps_ipi_list = sd;
2449 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2452 #endif /* CONFIG_RPS */
2457 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2458 * queue (may be a remote CPU queue).
2460 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2461 unsigned int *qtail)
2463 struct softnet_data *sd;
2464 unsigned long flags;
2466 sd = &per_cpu(softnet_data, cpu);
2468 local_irq_save(flags);
2471 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2472 if (skb_queue_len(&sd->input_pkt_queue)) {
2474 __skb_queue_tail(&sd->input_pkt_queue, skb);
2475 input_queue_tail_incr_save(sd, qtail);
2477 local_irq_restore(flags);
2478 return NET_RX_SUCCESS;
2481 /* Schedule NAPI for backlog device
2482 * We can use non atomic operation since we own the queue lock
2484 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2485 if (!rps_ipi_queued(sd))
2486 ____napi_schedule(sd, &sd->backlog);
2494 local_irq_restore(flags);
2501 * netif_rx - post buffer to the network code
2502 * @skb: buffer to post
2504 * This function receives a packet from a device driver and queues it for
2505 * the upper (protocol) levels to process. It always succeeds. The buffer
2506 * may be dropped during processing for congestion control or by the
2510 * NET_RX_SUCCESS (no congestion)
2511 * NET_RX_DROP (packet was dropped)
2515 int netif_rx(struct sk_buff *skb)
2519 /* if netpoll wants it, pretend we never saw it */
2520 if (netpoll_rx(skb))
2523 if (netdev_tstamp_prequeue)
2524 net_timestamp_check(skb);
2528 struct rps_dev_flow voidflow, *rflow = &voidflow;
2534 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2536 cpu = smp_processor_id();
2538 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2546 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2552 EXPORT_SYMBOL(netif_rx);
2554 int netif_rx_ni(struct sk_buff *skb)
2559 err = netif_rx(skb);
2560 if (local_softirq_pending())
2566 EXPORT_SYMBOL(netif_rx_ni);
2568 static void net_tx_action(struct softirq_action *h)
2570 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2572 if (sd->completion_queue) {
2573 struct sk_buff *clist;
2575 local_irq_disable();
2576 clist = sd->completion_queue;
2577 sd->completion_queue = NULL;
2581 struct sk_buff *skb = clist;
2582 clist = clist->next;
2584 WARN_ON(atomic_read(&skb->users));
2589 if (sd->output_queue) {
2592 local_irq_disable();
2593 head = sd->output_queue;
2594 sd->output_queue = NULL;
2595 sd->output_queue_tailp = &sd->output_queue;
2599 struct Qdisc *q = head;
2600 spinlock_t *root_lock;
2602 head = head->next_sched;
2604 root_lock = qdisc_lock(q);
2605 if (spin_trylock(root_lock)) {
2606 smp_mb__before_clear_bit();
2607 clear_bit(__QDISC_STATE_SCHED,
2610 spin_unlock(root_lock);
2612 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2614 __netif_reschedule(q);
2616 smp_mb__before_clear_bit();
2617 clear_bit(__QDISC_STATE_SCHED,
2625 static inline int deliver_skb(struct sk_buff *skb,
2626 struct packet_type *pt_prev,
2627 struct net_device *orig_dev)
2629 atomic_inc(&skb->users);
2630 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2633 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2634 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2635 /* This hook is defined here for ATM LANE */
2636 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2637 unsigned char *addr) __read_mostly;
2638 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2641 #ifdef CONFIG_NET_CLS_ACT
2642 /* TODO: Maybe we should just force sch_ingress to be compiled in
2643 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2644 * a compare and 2 stores extra right now if we dont have it on
2645 * but have CONFIG_NET_CLS_ACT
2646 * NOTE: This doesnt stop any functionality; if you dont have
2647 * the ingress scheduler, you just cant add policies on ingress.
2650 static int ing_filter(struct sk_buff *skb)
2652 struct net_device *dev = skb->dev;
2653 u32 ttl = G_TC_RTTL(skb->tc_verd);
2654 struct netdev_queue *rxq;
2655 int result = TC_ACT_OK;
2658 if (unlikely(MAX_RED_LOOP < ttl++)) {
2659 if (net_ratelimit())
2660 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2661 skb->skb_iif, dev->ifindex);
2665 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2666 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2668 rxq = &dev->rx_queue;
2671 if (q != &noop_qdisc) {
2672 spin_lock(qdisc_lock(q));
2673 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2674 result = qdisc_enqueue_root(skb, q);
2675 spin_unlock(qdisc_lock(q));
2681 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2682 struct packet_type **pt_prev,
2683 int *ret, struct net_device *orig_dev)
2685 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2689 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2693 switch (ing_filter(skb)) {
2707 * netif_nit_deliver - deliver received packets to network taps
2710 * This function is used to deliver incoming packets to network
2711 * taps. It should be used when the normal netif_receive_skb path
2712 * is bypassed, for example because of VLAN acceleration.
2714 void netif_nit_deliver(struct sk_buff *skb)
2716 struct packet_type *ptype;
2718 if (list_empty(&ptype_all))
2721 skb_reset_network_header(skb);
2722 skb_reset_transport_header(skb);
2723 skb->mac_len = skb->network_header - skb->mac_header;
2726 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2727 if (!ptype->dev || ptype->dev == skb->dev)
2728 deliver_skb(skb, ptype, skb->dev);
2734 * netdev_rx_handler_register - register receive handler
2735 * @dev: device to register a handler for
2736 * @rx_handler: receive handler to register
2737 * @rx_handler_data: data pointer that is used by rx handler
2739 * Register a receive hander for a device. This handler will then be
2740 * called from __netif_receive_skb. A negative errno code is returned
2743 * The caller must hold the rtnl_mutex.
2745 int netdev_rx_handler_register(struct net_device *dev,
2746 rx_handler_func_t *rx_handler,
2747 void *rx_handler_data)
2751 if (dev->rx_handler)
2754 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2755 rcu_assign_pointer(dev->rx_handler, rx_handler);
2759 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2762 * netdev_rx_handler_unregister - unregister receive handler
2763 * @dev: device to unregister a handler from
2765 * Unregister a receive hander from a device.
2767 * The caller must hold the rtnl_mutex.
2769 void netdev_rx_handler_unregister(struct net_device *dev)
2773 rcu_assign_pointer(dev->rx_handler, NULL);
2774 rcu_assign_pointer(dev->rx_handler_data, NULL);
2776 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2778 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2779 struct net_device *master)
2781 if (skb->pkt_type == PACKET_HOST) {
2782 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2784 memcpy(dest, master->dev_addr, ETH_ALEN);
2788 /* On bonding slaves other than the currently active slave, suppress
2789 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2790 * ARP on active-backup slaves with arp_validate enabled.
2792 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2794 struct net_device *dev = skb->dev;
2796 if (master->priv_flags & IFF_MASTER_ARPMON)
2797 dev->last_rx = jiffies;
2799 if ((master->priv_flags & IFF_MASTER_ALB) &&
2800 (master->priv_flags & IFF_BRIDGE_PORT)) {
2801 /* Do address unmangle. The local destination address
2802 * will be always the one master has. Provides the right
2803 * functionality in a bridge.
2805 skb_bond_set_mac_by_master(skb, master);
2808 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2809 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2810 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2813 if (master->priv_flags & IFF_MASTER_ALB) {
2814 if (skb->pkt_type != PACKET_BROADCAST &&
2815 skb->pkt_type != PACKET_MULTICAST)
2818 if (master->priv_flags & IFF_MASTER_8023AD &&
2819 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2826 EXPORT_SYMBOL(__skb_bond_should_drop);
2828 static int __netif_receive_skb(struct sk_buff *skb)
2830 struct packet_type *ptype, *pt_prev;
2831 rx_handler_func_t *rx_handler;
2832 struct net_device *orig_dev;
2833 struct net_device *master;
2834 struct net_device *null_or_orig;
2835 struct net_device *orig_or_bond;
2836 int ret = NET_RX_DROP;
2839 if (!netdev_tstamp_prequeue)
2840 net_timestamp_check(skb);
2842 if (vlan_tx_tag_present(skb))
2843 vlan_hwaccel_do_receive(skb);
2845 /* if we've gotten here through NAPI, check netpoll */
2846 if (netpoll_receive_skb(skb))
2850 skb->skb_iif = skb->dev->ifindex;
2853 * bonding note: skbs received on inactive slaves should only
2854 * be delivered to pkt handlers that are exact matches. Also
2855 * the deliver_no_wcard flag will be set. If packet handlers
2856 * are sensitive to duplicate packets these skbs will need to
2857 * be dropped at the handler. The vlan accel path may have
2858 * already set the deliver_no_wcard flag.
2860 null_or_orig = NULL;
2861 orig_dev = skb->dev;
2862 master = ACCESS_ONCE(orig_dev->master);
2863 if (skb->deliver_no_wcard)
2864 null_or_orig = orig_dev;
2866 if (skb_bond_should_drop(skb, master)) {
2867 skb->deliver_no_wcard = 1;
2868 null_or_orig = orig_dev; /* deliver only exact match */
2873 __this_cpu_inc(softnet_data.processed);
2874 skb_reset_network_header(skb);
2875 skb_reset_transport_header(skb);
2876 skb->mac_len = skb->network_header - skb->mac_header;
2882 #ifdef CONFIG_NET_CLS_ACT
2883 if (skb->tc_verd & TC_NCLS) {
2884 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2889 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2890 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2891 ptype->dev == orig_dev) {
2893 ret = deliver_skb(skb, pt_prev, orig_dev);
2898 #ifdef CONFIG_NET_CLS_ACT
2899 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2905 /* Handle special case of bridge or macvlan */
2906 rx_handler = rcu_dereference(skb->dev->rx_handler);
2909 ret = deliver_skb(skb, pt_prev, orig_dev);
2912 skb = rx_handler(skb);
2918 * Make sure frames received on VLAN interfaces stacked on
2919 * bonding interfaces still make their way to any base bonding
2920 * device that may have registered for a specific ptype. The
2921 * handler may have to adjust skb->dev and orig_dev.
2923 orig_or_bond = orig_dev;
2924 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2925 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2926 orig_or_bond = vlan_dev_real_dev(skb->dev);
2929 type = skb->protocol;
2930 list_for_each_entry_rcu(ptype,
2931 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2932 if (ptype->type == type && (ptype->dev == null_or_orig ||
2933 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2934 ptype->dev == orig_or_bond)) {
2936 ret = deliver_skb(skb, pt_prev, orig_dev);
2942 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2945 /* Jamal, now you will not able to escape explaining
2946 * me how you were going to use this. :-)
2957 * netif_receive_skb - process receive buffer from network
2958 * @skb: buffer to process
2960 * netif_receive_skb() is the main receive data processing function.
2961 * It always succeeds. The buffer may be dropped during processing
2962 * for congestion control or by the protocol layers.
2964 * This function may only be called from softirq context and interrupts
2965 * should be enabled.
2967 * Return values (usually ignored):
2968 * NET_RX_SUCCESS: no congestion
2969 * NET_RX_DROP: packet was dropped
2971 int netif_receive_skb(struct sk_buff *skb)
2973 if (netdev_tstamp_prequeue)
2974 net_timestamp_check(skb);
2976 if (skb_defer_rx_timestamp(skb))
2977 return NET_RX_SUCCESS;
2981 struct rps_dev_flow voidflow, *rflow = &voidflow;
2986 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2989 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2993 ret = __netif_receive_skb(skb);
2999 return __netif_receive_skb(skb);
3002 EXPORT_SYMBOL(netif_receive_skb);
3004 /* Network device is going away, flush any packets still pending
3005 * Called with irqs disabled.
3007 static void flush_backlog(void *arg)
3009 struct net_device *dev = arg;
3010 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3011 struct sk_buff *skb, *tmp;
3014 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3015 if (skb->dev == dev) {
3016 __skb_unlink(skb, &sd->input_pkt_queue);
3018 input_queue_head_incr(sd);
3023 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3024 if (skb->dev == dev) {
3025 __skb_unlink(skb, &sd->process_queue);
3027 input_queue_head_incr(sd);
3032 static int napi_gro_complete(struct sk_buff *skb)
3034 struct packet_type *ptype;
3035 __be16 type = skb->protocol;
3036 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3039 if (NAPI_GRO_CB(skb)->count == 1) {
3040 skb_shinfo(skb)->gso_size = 0;
3045 list_for_each_entry_rcu(ptype, head, list) {
3046 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3049 err = ptype->gro_complete(skb);
3055 WARN_ON(&ptype->list == head);
3057 return NET_RX_SUCCESS;
3061 return netif_receive_skb(skb);
3064 inline void napi_gro_flush(struct napi_struct *napi)
3066 struct sk_buff *skb, *next;
3068 for (skb = napi->gro_list; skb; skb = next) {
3071 napi_gro_complete(skb);
3074 napi->gro_count = 0;
3075 napi->gro_list = NULL;
3077 EXPORT_SYMBOL(napi_gro_flush);
3079 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3081 struct sk_buff **pp = NULL;
3082 struct packet_type *ptype;
3083 __be16 type = skb->protocol;
3084 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3087 enum gro_result ret;
3089 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3092 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3096 list_for_each_entry_rcu(ptype, head, list) {
3097 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3100 skb_set_network_header(skb, skb_gro_offset(skb));
3101 mac_len = skb->network_header - skb->mac_header;
3102 skb->mac_len = mac_len;
3103 NAPI_GRO_CB(skb)->same_flow = 0;
3104 NAPI_GRO_CB(skb)->flush = 0;
3105 NAPI_GRO_CB(skb)->free = 0;
3107 pp = ptype->gro_receive(&napi->gro_list, skb);
3112 if (&ptype->list == head)
3115 same_flow = NAPI_GRO_CB(skb)->same_flow;
3116 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3119 struct sk_buff *nskb = *pp;
3123 napi_gro_complete(nskb);
3130 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3134 NAPI_GRO_CB(skb)->count = 1;
3135 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3136 skb->next = napi->gro_list;
3137 napi->gro_list = skb;
3141 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3142 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3144 BUG_ON(skb->end - skb->tail < grow);
3146 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3149 skb->data_len -= grow;
3151 skb_shinfo(skb)->frags[0].page_offset += grow;
3152 skb_shinfo(skb)->frags[0].size -= grow;
3154 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3155 put_page(skb_shinfo(skb)->frags[0].page);
3156 memmove(skb_shinfo(skb)->frags,
3157 skb_shinfo(skb)->frags + 1,
3158 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3169 EXPORT_SYMBOL(dev_gro_receive);
3171 static inline gro_result_t
3172 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3176 for (p = napi->gro_list; p; p = p->next) {
3177 unsigned long diffs;
3179 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3180 diffs |= compare_ether_header(skb_mac_header(p),
3181 skb_gro_mac_header(skb));
3182 NAPI_GRO_CB(p)->same_flow = !diffs;
3183 NAPI_GRO_CB(p)->flush = 0;
3186 return dev_gro_receive(napi, skb);
3189 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3193 if (netif_receive_skb(skb))
3198 case GRO_MERGED_FREE:
3209 EXPORT_SYMBOL(napi_skb_finish);
3211 void skb_gro_reset_offset(struct sk_buff *skb)
3213 NAPI_GRO_CB(skb)->data_offset = 0;
3214 NAPI_GRO_CB(skb)->frag0 = NULL;
3215 NAPI_GRO_CB(skb)->frag0_len = 0;
3217 if (skb->mac_header == skb->tail &&
3218 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3219 NAPI_GRO_CB(skb)->frag0 =
3220 page_address(skb_shinfo(skb)->frags[0].page) +
3221 skb_shinfo(skb)->frags[0].page_offset;
3222 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3225 EXPORT_SYMBOL(skb_gro_reset_offset);
3227 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3229 skb_gro_reset_offset(skb);
3231 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3233 EXPORT_SYMBOL(napi_gro_receive);
3235 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3237 __skb_pull(skb, skb_headlen(skb));
3238 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3242 EXPORT_SYMBOL(napi_reuse_skb);
3244 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3246 struct sk_buff *skb = napi->skb;
3249 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3255 EXPORT_SYMBOL(napi_get_frags);
3257 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3263 skb->protocol = eth_type_trans(skb, skb->dev);
3265 if (ret == GRO_HELD)
3266 skb_gro_pull(skb, -ETH_HLEN);
3267 else if (netif_receive_skb(skb))
3272 case GRO_MERGED_FREE:
3273 napi_reuse_skb(napi, skb);
3282 EXPORT_SYMBOL(napi_frags_finish);
3284 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3286 struct sk_buff *skb = napi->skb;
3293 skb_reset_mac_header(skb);
3294 skb_gro_reset_offset(skb);
3296 off = skb_gro_offset(skb);
3297 hlen = off + sizeof(*eth);
3298 eth = skb_gro_header_fast(skb, off);
3299 if (skb_gro_header_hard(skb, hlen)) {
3300 eth = skb_gro_header_slow(skb, hlen, off);
3301 if (unlikely(!eth)) {
3302 napi_reuse_skb(napi, skb);
3308 skb_gro_pull(skb, sizeof(*eth));
3311 * This works because the only protocols we care about don't require
3312 * special handling. We'll fix it up properly at the end.
3314 skb->protocol = eth->h_proto;
3319 EXPORT_SYMBOL(napi_frags_skb);
3321 gro_result_t napi_gro_frags(struct napi_struct *napi)
3323 struct sk_buff *skb = napi_frags_skb(napi);
3328 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3330 EXPORT_SYMBOL(napi_gro_frags);
3333 * net_rps_action sends any pending IPI's for rps.
3334 * Note: called with local irq disabled, but exits with local irq enabled.
3336 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3339 struct softnet_data *remsd = sd->rps_ipi_list;
3342 sd->rps_ipi_list = NULL;
3346 /* Send pending IPI's to kick RPS processing on remote cpus. */
3348 struct softnet_data *next = remsd->rps_ipi_next;
3350 if (cpu_online(remsd->cpu))
3351 __smp_call_function_single(remsd->cpu,
3360 static int process_backlog(struct napi_struct *napi, int quota)
3363 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3366 /* Check if we have pending ipi, its better to send them now,
3367 * not waiting net_rx_action() end.
3369 if (sd->rps_ipi_list) {
3370 local_irq_disable();
3371 net_rps_action_and_irq_enable(sd);
3374 napi->weight = weight_p;
3375 local_irq_disable();
3376 while (work < quota) {
3377 struct sk_buff *skb;
3380 while ((skb = __skb_dequeue(&sd->process_queue))) {
3382 __netif_receive_skb(skb);
3383 local_irq_disable();
3384 input_queue_head_incr(sd);
3385 if (++work >= quota) {
3392 qlen = skb_queue_len(&sd->input_pkt_queue);
3394 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3395 &sd->process_queue);
3397 if (qlen < quota - work) {
3399 * Inline a custom version of __napi_complete().
3400 * only current cpu owns and manipulates this napi,
3401 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3402 * we can use a plain write instead of clear_bit(),
3403 * and we dont need an smp_mb() memory barrier.
3405 list_del(&napi->poll_list);
3408 quota = work + qlen;
3418 * __napi_schedule - schedule for receive
3419 * @n: entry to schedule
3421 * The entry's receive function will be scheduled to run
3423 void __napi_schedule(struct napi_struct *n)
3425 unsigned long flags;
3427 local_irq_save(flags);
3428 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3429 local_irq_restore(flags);
3431 EXPORT_SYMBOL(__napi_schedule);
3433 void __napi_complete(struct napi_struct *n)
3435 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3436 BUG_ON(n->gro_list);
3438 list_del(&n->poll_list);
3439 smp_mb__before_clear_bit();
3440 clear_bit(NAPI_STATE_SCHED, &n->state);
3442 EXPORT_SYMBOL(__napi_complete);
3444 void napi_complete(struct napi_struct *n)
3446 unsigned long flags;
3449 * don't let napi dequeue from the cpu poll list
3450 * just in case its running on a different cpu
3452 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3456 local_irq_save(flags);
3458 local_irq_restore(flags);
3460 EXPORT_SYMBOL(napi_complete);
3462 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3463 int (*poll)(struct napi_struct *, int), int weight)
3465 INIT_LIST_HEAD(&napi->poll_list);
3466 napi->gro_count = 0;
3467 napi->gro_list = NULL;
3470 napi->weight = weight;
3471 list_add(&napi->dev_list, &dev->napi_list);
3473 #ifdef CONFIG_NETPOLL
3474 spin_lock_init(&napi->poll_lock);
3475 napi->poll_owner = -1;
3477 set_bit(NAPI_STATE_SCHED, &napi->state);
3479 EXPORT_SYMBOL(netif_napi_add);
3481 void netif_napi_del(struct napi_struct *napi)
3483 struct sk_buff *skb, *next;
3485 list_del_init(&napi->dev_list);
3486 napi_free_frags(napi);
3488 for (skb = napi->gro_list; skb; skb = next) {
3494 napi->gro_list = NULL;
3495 napi->gro_count = 0;
3497 EXPORT_SYMBOL(netif_napi_del);
3499 static void net_rx_action(struct softirq_action *h)
3501 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3502 unsigned long time_limit = jiffies + 2;
3503 int budget = netdev_budget;
3506 local_irq_disable();
3508 while (!list_empty(&sd->poll_list)) {
3509 struct napi_struct *n;
3512 /* If softirq window is exhuasted then punt.
3513 * Allow this to run for 2 jiffies since which will allow
3514 * an average latency of 1.5/HZ.
3516 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3521 /* Even though interrupts have been re-enabled, this
3522 * access is safe because interrupts can only add new
3523 * entries to the tail of this list, and only ->poll()
3524 * calls can remove this head entry from the list.
3526 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3528 have = netpoll_poll_lock(n);
3532 /* This NAPI_STATE_SCHED test is for avoiding a race
3533 * with netpoll's poll_napi(). Only the entity which
3534 * obtains the lock and sees NAPI_STATE_SCHED set will
3535 * actually make the ->poll() call. Therefore we avoid
3536 * accidently calling ->poll() when NAPI is not scheduled.
3539 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3540 work = n->poll(n, weight);
3544 WARN_ON_ONCE(work > weight);
3548 local_irq_disable();
3550 /* Drivers must not modify the NAPI state if they
3551 * consume the entire weight. In such cases this code
3552 * still "owns" the NAPI instance and therefore can
3553 * move the instance around on the list at-will.
3555 if (unlikely(work == weight)) {
3556 if (unlikely(napi_disable_pending(n))) {
3559 local_irq_disable();
3561 list_move_tail(&n->poll_list, &sd->poll_list);
3564 netpoll_poll_unlock(have);
3567 net_rps_action_and_irq_enable(sd);
3569 #ifdef CONFIG_NET_DMA
3571 * There may not be any more sk_buffs coming right now, so push
3572 * any pending DMA copies to hardware
3574 dma_issue_pending_all();
3581 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3585 static gifconf_func_t *gifconf_list[NPROTO];
3588 * register_gifconf - register a SIOCGIF handler
3589 * @family: Address family
3590 * @gifconf: Function handler
3592 * Register protocol dependent address dumping routines. The handler
3593 * that is passed must not be freed or reused until it has been replaced
3594 * by another handler.
3596 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3598 if (family >= NPROTO)
3600 gifconf_list[family] = gifconf;
3603 EXPORT_SYMBOL(register_gifconf);
3607 * Map an interface index to its name (SIOCGIFNAME)
3611 * We need this ioctl for efficient implementation of the
3612 * if_indextoname() function required by the IPv6 API. Without
3613 * it, we would have to search all the interfaces to find a
3617 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3619 struct net_device *dev;
3623 * Fetch the caller's info block.
3626 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3630 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3636 strcpy(ifr.ifr_name, dev->name);
3639 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3645 * Perform a SIOCGIFCONF call. This structure will change
3646 * size eventually, and there is nothing I can do about it.
3647 * Thus we will need a 'compatibility mode'.
3650 static int dev_ifconf(struct net *net, char __user *arg)
3653 struct net_device *dev;
3660 * Fetch the caller's info block.
3663 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3670 * Loop over the interfaces, and write an info block for each.
3674 for_each_netdev(net, dev) {
3675 for (i = 0; i < NPROTO; i++) {
3676 if (gifconf_list[i]) {
3679 done = gifconf_list[i](dev, NULL, 0);
3681 done = gifconf_list[i](dev, pos + total,
3691 * All done. Write the updated control block back to the caller.
3693 ifc.ifc_len = total;
3696 * Both BSD and Solaris return 0 here, so we do too.
3698 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3701 #ifdef CONFIG_PROC_FS
3703 * This is invoked by the /proc filesystem handler to display a device
3706 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3709 struct net *net = seq_file_net(seq);
3711 struct net_device *dev;
3715 return SEQ_START_TOKEN;
3718 for_each_netdev_rcu(net, dev)
3725 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3727 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3728 first_net_device(seq_file_net(seq)) :
3729 next_net_device((struct net_device *)v);
3732 return rcu_dereference(dev);
3735 void dev_seq_stop(struct seq_file *seq, void *v)
3741 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3743 struct rtnl_link_stats64 temp;
3744 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3746 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3747 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3748 dev->name, stats->rx_bytes, stats->rx_packets,
3750 stats->rx_dropped + stats->rx_missed_errors,
3751 stats->rx_fifo_errors,
3752 stats->rx_length_errors + stats->rx_over_errors +
3753 stats->rx_crc_errors + stats->rx_frame_errors,
3754 stats->rx_compressed, stats->multicast,
3755 stats->tx_bytes, stats->tx_packets,
3756 stats->tx_errors, stats->tx_dropped,
3757 stats->tx_fifo_errors, stats->collisions,
3758 stats->tx_carrier_errors +
3759 stats->tx_aborted_errors +
3760 stats->tx_window_errors +
3761 stats->tx_heartbeat_errors,
3762 stats->tx_compressed);
3766 * Called from the PROCfs module. This now uses the new arbitrary sized
3767 * /proc/net interface to create /proc/net/dev
3769 static int dev_seq_show(struct seq_file *seq, void *v)
3771 if (v == SEQ_START_TOKEN)
3772 seq_puts(seq, "Inter-| Receive "
3774 " face |bytes packets errs drop fifo frame "
3775 "compressed multicast|bytes packets errs "
3776 "drop fifo colls carrier compressed\n");
3778 dev_seq_printf_stats(seq, v);
3782 static struct softnet_data *softnet_get_online(loff_t *pos)
3784 struct softnet_data *sd = NULL;
3786 while (*pos < nr_cpu_ids)
3787 if (cpu_online(*pos)) {
3788 sd = &per_cpu(softnet_data, *pos);
3795 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3797 return softnet_get_online(pos);
3800 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3803 return softnet_get_online(pos);
3806 static void softnet_seq_stop(struct seq_file *seq, void *v)
3810 static int softnet_seq_show(struct seq_file *seq, void *v)
3812 struct softnet_data *sd = v;
3814 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3815 sd->processed, sd->dropped, sd->time_squeeze, 0,
3816 0, 0, 0, 0, /* was fastroute */
3817 sd->cpu_collision, sd->received_rps);
3821 static const struct seq_operations dev_seq_ops = {
3822 .start = dev_seq_start,
3823 .next = dev_seq_next,
3824 .stop = dev_seq_stop,
3825 .show = dev_seq_show,
3828 static int dev_seq_open(struct inode *inode, struct file *file)
3830 return seq_open_net(inode, file, &dev_seq_ops,
3831 sizeof(struct seq_net_private));
3834 static const struct file_operations dev_seq_fops = {
3835 .owner = THIS_MODULE,
3836 .open = dev_seq_open,
3838 .llseek = seq_lseek,
3839 .release = seq_release_net,
3842 static const struct seq_operations softnet_seq_ops = {
3843 .start = softnet_seq_start,
3844 .next = softnet_seq_next,
3845 .stop = softnet_seq_stop,
3846 .show = softnet_seq_show,
3849 static int softnet_seq_open(struct inode *inode, struct file *file)
3851 return seq_open(file, &softnet_seq_ops);
3854 static const struct file_operations softnet_seq_fops = {
3855 .owner = THIS_MODULE,
3856 .open = softnet_seq_open,
3858 .llseek = seq_lseek,
3859 .release = seq_release,
3862 static void *ptype_get_idx(loff_t pos)
3864 struct packet_type *pt = NULL;
3868 list_for_each_entry_rcu(pt, &ptype_all, list) {
3874 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3875 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3884 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3888 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3891 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3893 struct packet_type *pt;
3894 struct list_head *nxt;
3898 if (v == SEQ_START_TOKEN)
3899 return ptype_get_idx(0);
3902 nxt = pt->list.next;
3903 if (pt->type == htons(ETH_P_ALL)) {
3904 if (nxt != &ptype_all)
3907 nxt = ptype_base[0].next;
3909 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3911 while (nxt == &ptype_base[hash]) {
3912 if (++hash >= PTYPE_HASH_SIZE)
3914 nxt = ptype_base[hash].next;
3917 return list_entry(nxt, struct packet_type, list);
3920 static void ptype_seq_stop(struct seq_file *seq, void *v)
3926 static int ptype_seq_show(struct seq_file *seq, void *v)
3928 struct packet_type *pt = v;
3930 if (v == SEQ_START_TOKEN)
3931 seq_puts(seq, "Type Device Function\n");
3932 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3933 if (pt->type == htons(ETH_P_ALL))
3934 seq_puts(seq, "ALL ");
3936 seq_printf(seq, "%04x", ntohs(pt->type));
3938 seq_printf(seq, " %-8s %pF\n",
3939 pt->dev ? pt->dev->name : "", pt->func);
3945 static const struct seq_operations ptype_seq_ops = {
3946 .start = ptype_seq_start,
3947 .next = ptype_seq_next,
3948 .stop = ptype_seq_stop,
3949 .show = ptype_seq_show,
3952 static int ptype_seq_open(struct inode *inode, struct file *file)
3954 return seq_open_net(inode, file, &ptype_seq_ops,
3955 sizeof(struct seq_net_private));
3958 static const struct file_operations ptype_seq_fops = {
3959 .owner = THIS_MODULE,
3960 .open = ptype_seq_open,
3962 .llseek = seq_lseek,
3963 .release = seq_release_net,
3967 static int __net_init dev_proc_net_init(struct net *net)
3971 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3973 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3975 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3978 if (wext_proc_init(net))
3984 proc_net_remove(net, "ptype");
3986 proc_net_remove(net, "softnet_stat");
3988 proc_net_remove(net, "dev");
3992 static void __net_exit dev_proc_net_exit(struct net *net)
3994 wext_proc_exit(net);
3996 proc_net_remove(net, "ptype");
3997 proc_net_remove(net, "softnet_stat");
3998 proc_net_remove(net, "dev");
4001 static struct pernet_operations __net_initdata dev_proc_ops = {
4002 .init = dev_proc_net_init,
4003 .exit = dev_proc_net_exit,
4006 static int __init dev_proc_init(void)
4008 return register_pernet_subsys(&dev_proc_ops);
4011 #define dev_proc_init() 0
4012 #endif /* CONFIG_PROC_FS */
4016 * netdev_set_master - set up master/slave pair
4017 * @slave: slave device
4018 * @master: new master device
4020 * Changes the master device of the slave. Pass %NULL to break the
4021 * bonding. The caller must hold the RTNL semaphore. On a failure
4022 * a negative errno code is returned. On success the reference counts
4023 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4024 * function returns zero.
4026 int netdev_set_master(struct net_device *slave, struct net_device *master)
4028 struct net_device *old = slave->master;
4038 slave->master = master;
4045 slave->flags |= IFF_SLAVE;
4047 slave->flags &= ~IFF_SLAVE;
4049 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4052 EXPORT_SYMBOL(netdev_set_master);
4054 static void dev_change_rx_flags(struct net_device *dev, int flags)
4056 const struct net_device_ops *ops = dev->netdev_ops;
4058 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4059 ops->ndo_change_rx_flags(dev, flags);
4062 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4064 unsigned short old_flags = dev->flags;
4070 dev->flags |= IFF_PROMISC;
4071 dev->promiscuity += inc;
4072 if (dev->promiscuity == 0) {
4075 * If inc causes overflow, untouch promisc and return error.
4078 dev->flags &= ~IFF_PROMISC;
4080 dev->promiscuity -= inc;
4081 printk(KERN_WARNING "%s: promiscuity touches roof, "
4082 "set promiscuity failed, promiscuity feature "
4083 "of device might be broken.\n", dev->name);
4087 if (dev->flags != old_flags) {
4088 printk(KERN_INFO "device %s %s promiscuous mode\n",
4089 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4091 if (audit_enabled) {
4092 current_uid_gid(&uid, &gid);
4093 audit_log(current->audit_context, GFP_ATOMIC,
4094 AUDIT_ANOM_PROMISCUOUS,
4095 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4096 dev->name, (dev->flags & IFF_PROMISC),
4097 (old_flags & IFF_PROMISC),
4098 audit_get_loginuid(current),
4100 audit_get_sessionid(current));
4103 dev_change_rx_flags(dev, IFF_PROMISC);
4109 * dev_set_promiscuity - update promiscuity count on a device
4113 * Add or remove promiscuity from a device. While the count in the device
4114 * remains above zero the interface remains promiscuous. Once it hits zero
4115 * the device reverts back to normal filtering operation. A negative inc
4116 * value is used to drop promiscuity on the device.
4117 * Return 0 if successful or a negative errno code on error.
4119 int dev_set_promiscuity(struct net_device *dev, int inc)
4121 unsigned short old_flags = dev->flags;
4124 err = __dev_set_promiscuity(dev, inc);
4127 if (dev->flags != old_flags)
4128 dev_set_rx_mode(dev);
4131 EXPORT_SYMBOL(dev_set_promiscuity);
4134 * dev_set_allmulti - update allmulti count on a device
4138 * Add or remove reception of all multicast frames to a device. While the
4139 * count in the device remains above zero the interface remains listening
4140 * to all interfaces. Once it hits zero the device reverts back to normal
4141 * filtering operation. A negative @inc value is used to drop the counter
4142 * when releasing a resource needing all multicasts.
4143 * Return 0 if successful or a negative errno code on error.
4146 int dev_set_allmulti(struct net_device *dev, int inc)
4148 unsigned short old_flags = dev->flags;
4152 dev->flags |= IFF_ALLMULTI;
4153 dev->allmulti += inc;
4154 if (dev->allmulti == 0) {
4157 * If inc causes overflow, untouch allmulti and return error.
4160 dev->flags &= ~IFF_ALLMULTI;
4162 dev->allmulti -= inc;
4163 printk(KERN_WARNING "%s: allmulti touches roof, "
4164 "set allmulti failed, allmulti feature of "
4165 "device might be broken.\n", dev->name);
4169 if (dev->flags ^ old_flags) {
4170 dev_change_rx_flags(dev, IFF_ALLMULTI);
4171 dev_set_rx_mode(dev);
4175 EXPORT_SYMBOL(dev_set_allmulti);
4178 * Upload unicast and multicast address lists to device and
4179 * configure RX filtering. When the device doesn't support unicast
4180 * filtering it is put in promiscuous mode while unicast addresses
4183 void __dev_set_rx_mode(struct net_device *dev)
4185 const struct net_device_ops *ops = dev->netdev_ops;
4187 /* dev_open will call this function so the list will stay sane. */
4188 if (!(dev->flags&IFF_UP))
4191 if (!netif_device_present(dev))
4194 if (ops->ndo_set_rx_mode)
4195 ops->ndo_set_rx_mode(dev);
4197 /* Unicast addresses changes may only happen under the rtnl,
4198 * therefore calling __dev_set_promiscuity here is safe.
4200 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4201 __dev_set_promiscuity(dev, 1);
4202 dev->uc_promisc = 1;
4203 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4204 __dev_set_promiscuity(dev, -1);
4205 dev->uc_promisc = 0;
4208 if (ops->ndo_set_multicast_list)
4209 ops->ndo_set_multicast_list(dev);
4213 void dev_set_rx_mode(struct net_device *dev)
4215 netif_addr_lock_bh(dev);
4216 __dev_set_rx_mode(dev);
4217 netif_addr_unlock_bh(dev);
4221 * dev_get_flags - get flags reported to userspace
4224 * Get the combination of flag bits exported through APIs to userspace.
4226 unsigned dev_get_flags(const struct net_device *dev)
4230 flags = (dev->flags & ~(IFF_PROMISC |
4235 (dev->gflags & (IFF_PROMISC |
4238 if (netif_running(dev)) {
4239 if (netif_oper_up(dev))
4240 flags |= IFF_RUNNING;
4241 if (netif_carrier_ok(dev))
4242 flags |= IFF_LOWER_UP;
4243 if (netif_dormant(dev))
4244 flags |= IFF_DORMANT;
4249 EXPORT_SYMBOL(dev_get_flags);
4251 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4253 int old_flags = dev->flags;
4259 * Set the flags on our device.
4262 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4263 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4265 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4269 * Load in the correct multicast list now the flags have changed.
4272 if ((old_flags ^ flags) & IFF_MULTICAST)
4273 dev_change_rx_flags(dev, IFF_MULTICAST);
4275 dev_set_rx_mode(dev);
4278 * Have we downed the interface. We handle IFF_UP ourselves
4279 * according to user attempts to set it, rather than blindly
4284 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4285 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4288 dev_set_rx_mode(dev);
4291 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4292 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4294 dev->gflags ^= IFF_PROMISC;
4295 dev_set_promiscuity(dev, inc);
4298 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4299 is important. Some (broken) drivers set IFF_PROMISC, when
4300 IFF_ALLMULTI is requested not asking us and not reporting.
4302 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4303 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4305 dev->gflags ^= IFF_ALLMULTI;
4306 dev_set_allmulti(dev, inc);
4312 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4314 unsigned int changes = dev->flags ^ old_flags;
4316 if (changes & IFF_UP) {
4317 if (dev->flags & IFF_UP)
4318 call_netdevice_notifiers(NETDEV_UP, dev);
4320 call_netdevice_notifiers(NETDEV_DOWN, dev);
4323 if (dev->flags & IFF_UP &&
4324 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4325 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4329 * dev_change_flags - change device settings
4331 * @flags: device state flags
4333 * Change settings on device based state flags. The flags are
4334 * in the userspace exported format.
4336 int dev_change_flags(struct net_device *dev, unsigned flags)
4339 int old_flags = dev->flags;
4341 ret = __dev_change_flags(dev, flags);
4345 changes = old_flags ^ dev->flags;
4347 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4349 __dev_notify_flags(dev, old_flags);
4352 EXPORT_SYMBOL(dev_change_flags);
4355 * dev_set_mtu - Change maximum transfer unit
4357 * @new_mtu: new transfer unit
4359 * Change the maximum transfer size of the network device.
4361 int dev_set_mtu(struct net_device *dev, int new_mtu)
4363 const struct net_device_ops *ops = dev->netdev_ops;
4366 if (new_mtu == dev->mtu)
4369 /* MTU must be positive. */
4373 if (!netif_device_present(dev))
4377 if (ops->ndo_change_mtu)
4378 err = ops->ndo_change_mtu(dev, new_mtu);
4382 if (!err && dev->flags & IFF_UP)
4383 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4386 EXPORT_SYMBOL(dev_set_mtu);
4389 * dev_set_mac_address - Change Media Access Control Address
4393 * Change the hardware (MAC) address of the device
4395 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4397 const struct net_device_ops *ops = dev->netdev_ops;
4400 if (!ops->ndo_set_mac_address)
4402 if (sa->sa_family != dev->type)
4404 if (!netif_device_present(dev))
4406 err = ops->ndo_set_mac_address(dev, sa);
4408 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4411 EXPORT_SYMBOL(dev_set_mac_address);
4414 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4416 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4419 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4425 case SIOCGIFFLAGS: /* Get interface flags */
4426 ifr->ifr_flags = (short) dev_get_flags(dev);
4429 case SIOCGIFMETRIC: /* Get the metric on the interface
4430 (currently unused) */
4431 ifr->ifr_metric = 0;
4434 case SIOCGIFMTU: /* Get the MTU of a device */
4435 ifr->ifr_mtu = dev->mtu;
4440 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4442 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4443 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4444 ifr->ifr_hwaddr.sa_family = dev->type;
4452 ifr->ifr_map.mem_start = dev->mem_start;
4453 ifr->ifr_map.mem_end = dev->mem_end;
4454 ifr->ifr_map.base_addr = dev->base_addr;
4455 ifr->ifr_map.irq = dev->irq;
4456 ifr->ifr_map.dma = dev->dma;
4457 ifr->ifr_map.port = dev->if_port;
4461 ifr->ifr_ifindex = dev->ifindex;
4465 ifr->ifr_qlen = dev->tx_queue_len;
4469 /* dev_ioctl() should ensure this case
4481 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4483 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4486 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4487 const struct net_device_ops *ops;
4492 ops = dev->netdev_ops;
4495 case SIOCSIFFLAGS: /* Set interface flags */
4496 return dev_change_flags(dev, ifr->ifr_flags);
4498 case SIOCSIFMETRIC: /* Set the metric on the interface
4499 (currently unused) */
4502 case SIOCSIFMTU: /* Set the MTU of a device */
4503 return dev_set_mtu(dev, ifr->ifr_mtu);
4506 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4508 case SIOCSIFHWBROADCAST:
4509 if (ifr->ifr_hwaddr.sa_family != dev->type)
4511 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4512 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4513 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4517 if (ops->ndo_set_config) {
4518 if (!netif_device_present(dev))
4520 return ops->ndo_set_config(dev, &ifr->ifr_map);
4525 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4526 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4528 if (!netif_device_present(dev))
4530 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4533 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4534 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4536 if (!netif_device_present(dev))
4538 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4541 if (ifr->ifr_qlen < 0)
4543 dev->tx_queue_len = ifr->ifr_qlen;
4547 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4548 return dev_change_name(dev, ifr->ifr_newname);
4551 * Unknown or private ioctl
4554 if ((cmd >= SIOCDEVPRIVATE &&
4555 cmd <= SIOCDEVPRIVATE + 15) ||
4556 cmd == SIOCBONDENSLAVE ||
4557 cmd == SIOCBONDRELEASE ||
4558 cmd == SIOCBONDSETHWADDR ||
4559 cmd == SIOCBONDSLAVEINFOQUERY ||
4560 cmd == SIOCBONDINFOQUERY ||
4561 cmd == SIOCBONDCHANGEACTIVE ||
4562 cmd == SIOCGMIIPHY ||
4563 cmd == SIOCGMIIREG ||
4564 cmd == SIOCSMIIREG ||
4565 cmd == SIOCBRADDIF ||
4566 cmd == SIOCBRDELIF ||
4567 cmd == SIOCSHWTSTAMP ||
4568 cmd == SIOCWANDEV) {
4570 if (ops->ndo_do_ioctl) {
4571 if (netif_device_present(dev))
4572 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4584 * This function handles all "interface"-type I/O control requests. The actual
4585 * 'doing' part of this is dev_ifsioc above.
4589 * dev_ioctl - network device ioctl
4590 * @net: the applicable net namespace
4591 * @cmd: command to issue
4592 * @arg: pointer to a struct ifreq in user space
4594 * Issue ioctl functions to devices. This is normally called by the
4595 * user space syscall interfaces but can sometimes be useful for
4596 * other purposes. The return value is the return from the syscall if
4597 * positive or a negative errno code on error.
4600 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4606 /* One special case: SIOCGIFCONF takes ifconf argument
4607 and requires shared lock, because it sleeps writing
4611 if (cmd == SIOCGIFCONF) {
4613 ret = dev_ifconf(net, (char __user *) arg);
4617 if (cmd == SIOCGIFNAME)
4618 return dev_ifname(net, (struct ifreq __user *)arg);
4620 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4623 ifr.ifr_name[IFNAMSIZ-1] = 0;
4625 colon = strchr(ifr.ifr_name, ':');
4630 * See which interface the caller is talking about.
4635 * These ioctl calls:
4636 * - can be done by all.
4637 * - atomic and do not require locking.
4648 dev_load(net, ifr.ifr_name);
4650 ret = dev_ifsioc_locked(net, &ifr, cmd);
4655 if (copy_to_user(arg, &ifr,
4656 sizeof(struct ifreq)))
4662 dev_load(net, ifr.ifr_name);
4664 ret = dev_ethtool(net, &ifr);
4669 if (copy_to_user(arg, &ifr,
4670 sizeof(struct ifreq)))
4676 * These ioctl calls:
4677 * - require superuser power.
4678 * - require strict serialization.
4684 if (!capable(CAP_NET_ADMIN))
4686 dev_load(net, ifr.ifr_name);
4688 ret = dev_ifsioc(net, &ifr, cmd);
4693 if (copy_to_user(arg, &ifr,
4694 sizeof(struct ifreq)))
4700 * These ioctl calls:
4701 * - require superuser power.
4702 * - require strict serialization.
4703 * - do not return a value
4713 case SIOCSIFHWBROADCAST:
4716 case SIOCBONDENSLAVE:
4717 case SIOCBONDRELEASE:
4718 case SIOCBONDSETHWADDR:
4719 case SIOCBONDCHANGEACTIVE:
4723 if (!capable(CAP_NET_ADMIN))
4726 case SIOCBONDSLAVEINFOQUERY:
4727 case SIOCBONDINFOQUERY:
4728 dev_load(net, ifr.ifr_name);
4730 ret = dev_ifsioc(net, &ifr, cmd);
4735 /* Get the per device memory space. We can add this but
4736 * currently do not support it */
4738 /* Set the per device memory buffer space.
4739 * Not applicable in our case */
4744 * Unknown or private ioctl.
4747 if (cmd == SIOCWANDEV ||
4748 (cmd >= SIOCDEVPRIVATE &&
4749 cmd <= SIOCDEVPRIVATE + 15)) {
4750 dev_load(net, ifr.ifr_name);
4752 ret = dev_ifsioc(net, &ifr, cmd);
4754 if (!ret && copy_to_user(arg, &ifr,
4755 sizeof(struct ifreq)))
4759 /* Take care of Wireless Extensions */
4760 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4761 return wext_handle_ioctl(net, &ifr, cmd, arg);
4768 * dev_new_index - allocate an ifindex
4769 * @net: the applicable net namespace
4771 * Returns a suitable unique value for a new device interface
4772 * number. The caller must hold the rtnl semaphore or the
4773 * dev_base_lock to be sure it remains unique.
4775 static int dev_new_index(struct net *net)
4781 if (!__dev_get_by_index(net, ifindex))
4786 /* Delayed registration/unregisteration */
4787 static LIST_HEAD(net_todo_list);
4789 static void net_set_todo(struct net_device *dev)
4791 list_add_tail(&dev->todo_list, &net_todo_list);
4794 static void rollback_registered_many(struct list_head *head)
4796 struct net_device *dev, *tmp;
4798 BUG_ON(dev_boot_phase);
4801 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4802 /* Some devices call without registering
4803 * for initialization unwind. Remove those
4804 * devices and proceed with the remaining.
4806 if (dev->reg_state == NETREG_UNINITIALIZED) {
4807 pr_debug("unregister_netdevice: device %s/%p never "
4808 "was registered\n", dev->name, dev);
4811 list_del(&dev->unreg_list);
4815 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4817 /* If device is running, close it first. */
4820 /* And unlink it from device chain. */
4821 unlist_netdevice(dev);
4823 dev->reg_state = NETREG_UNREGISTERING;
4828 list_for_each_entry(dev, head, unreg_list) {
4829 /* Shutdown queueing discipline. */
4833 /* Notify protocols, that we are about to destroy
4834 this device. They should clean all the things.
4836 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4838 if (!dev->rtnl_link_ops ||
4839 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4840 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4843 * Flush the unicast and multicast chains
4848 if (dev->netdev_ops->ndo_uninit)
4849 dev->netdev_ops->ndo_uninit(dev);
4851 /* Notifier chain MUST detach us from master device. */
4852 WARN_ON(dev->master);
4854 /* Remove entries from kobject tree */
4855 netdev_unregister_kobject(dev);
4858 /* Process any work delayed until the end of the batch */
4859 dev = list_first_entry(head, struct net_device, unreg_list);
4860 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4864 list_for_each_entry(dev, head, unreg_list)
4868 static void rollback_registered(struct net_device *dev)
4872 list_add(&dev->unreg_list, &single);
4873 rollback_registered_many(&single);
4876 static void __netdev_init_queue_locks_one(struct net_device *dev,
4877 struct netdev_queue *dev_queue,
4880 spin_lock_init(&dev_queue->_xmit_lock);
4881 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4882 dev_queue->xmit_lock_owner = -1;
4885 static void netdev_init_queue_locks(struct net_device *dev)
4887 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4888 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4891 unsigned long netdev_fix_features(unsigned long features, const char *name)
4893 /* Fix illegal SG+CSUM combinations. */
4894 if ((features & NETIF_F_SG) &&
4895 !(features & NETIF_F_ALL_CSUM)) {
4897 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4898 "checksum feature.\n", name);
4899 features &= ~NETIF_F_SG;
4902 /* TSO requires that SG is present as well. */
4903 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4905 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4906 "SG feature.\n", name);
4907 features &= ~NETIF_F_TSO;
4910 if (features & NETIF_F_UFO) {
4911 if (!(features & NETIF_F_GEN_CSUM)) {
4913 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4914 "since no NETIF_F_HW_CSUM feature.\n",
4916 features &= ~NETIF_F_UFO;
4919 if (!(features & NETIF_F_SG)) {
4921 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4922 "since no NETIF_F_SG feature.\n", name);
4923 features &= ~NETIF_F_UFO;
4929 EXPORT_SYMBOL(netdev_fix_features);
4932 * netif_stacked_transfer_operstate - transfer operstate
4933 * @rootdev: the root or lower level device to transfer state from
4934 * @dev: the device to transfer operstate to
4936 * Transfer operational state from root to device. This is normally
4937 * called when a stacking relationship exists between the root
4938 * device and the device(a leaf device).
4940 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4941 struct net_device *dev)
4943 if (rootdev->operstate == IF_OPER_DORMANT)
4944 netif_dormant_on(dev);
4946 netif_dormant_off(dev);
4948 if (netif_carrier_ok(rootdev)) {
4949 if (!netif_carrier_ok(dev))
4950 netif_carrier_on(dev);
4952 if (netif_carrier_ok(dev))
4953 netif_carrier_off(dev);
4956 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4959 * register_netdevice - register a network device
4960 * @dev: device to register
4962 * Take a completed network device structure and add it to the kernel
4963 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4964 * chain. 0 is returned on success. A negative errno code is returned
4965 * on a failure to set up the device, or if the name is a duplicate.
4967 * Callers must hold the rtnl semaphore. You may want
4968 * register_netdev() instead of this.
4971 * The locking appears insufficient to guarantee two parallel registers
4972 * will not get the same name.
4975 int register_netdevice(struct net_device *dev)
4978 struct net *net = dev_net(dev);
4980 BUG_ON(dev_boot_phase);
4985 /* When net_device's are persistent, this will be fatal. */
4986 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4989 spin_lock_init(&dev->addr_list_lock);
4990 netdev_set_addr_lockdep_class(dev);
4991 netdev_init_queue_locks(dev);
4996 if (!dev->num_rx_queues) {
4998 * Allocate a single RX queue if driver never called
5002 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
5008 dev->_rx->first = dev->_rx;
5009 atomic_set(&dev->_rx->count, 1);
5010 dev->num_rx_queues = 1;
5013 /* Init, if this function is available */
5014 if (dev->netdev_ops->ndo_init) {
5015 ret = dev->netdev_ops->ndo_init(dev);
5023 ret = dev_get_valid_name(dev, dev->name, 0);
5027 dev->ifindex = dev_new_index(net);
5028 if (dev->iflink == -1)
5029 dev->iflink = dev->ifindex;
5031 /* Fix illegal checksum combinations */
5032 if ((dev->features & NETIF_F_HW_CSUM) &&
5033 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5034 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5036 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5039 if ((dev->features & NETIF_F_NO_CSUM) &&
5040 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5041 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5043 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5046 dev->features = netdev_fix_features(dev->features, dev->name);
5048 /* Enable software GSO if SG is supported. */
5049 if (dev->features & NETIF_F_SG)
5050 dev->features |= NETIF_F_GSO;
5052 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5053 ret = notifier_to_errno(ret);
5057 ret = netdev_register_kobject(dev);
5060 dev->reg_state = NETREG_REGISTERED;
5063 * Default initial state at registry is that the
5064 * device is present.
5067 set_bit(__LINK_STATE_PRESENT, &dev->state);
5069 dev_init_scheduler(dev);
5071 list_netdevice(dev);
5073 /* Notify protocols, that a new device appeared. */
5074 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5075 ret = notifier_to_errno(ret);
5077 rollback_registered(dev);
5078 dev->reg_state = NETREG_UNREGISTERED;
5081 * Prevent userspace races by waiting until the network
5082 * device is fully setup before sending notifications.
5084 if (!dev->rtnl_link_ops ||
5085 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5086 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5092 if (dev->netdev_ops->ndo_uninit)
5093 dev->netdev_ops->ndo_uninit(dev);
5096 EXPORT_SYMBOL(register_netdevice);
5099 * init_dummy_netdev - init a dummy network device for NAPI
5100 * @dev: device to init
5102 * This takes a network device structure and initialize the minimum
5103 * amount of fields so it can be used to schedule NAPI polls without
5104 * registering a full blown interface. This is to be used by drivers
5105 * that need to tie several hardware interfaces to a single NAPI
5106 * poll scheduler due to HW limitations.
5108 int init_dummy_netdev(struct net_device *dev)
5110 /* Clear everything. Note we don't initialize spinlocks
5111 * are they aren't supposed to be taken by any of the
5112 * NAPI code and this dummy netdev is supposed to be
5113 * only ever used for NAPI polls
5115 memset(dev, 0, sizeof(struct net_device));
5117 /* make sure we BUG if trying to hit standard
5118 * register/unregister code path
5120 dev->reg_state = NETREG_DUMMY;
5122 /* initialize the ref count */
5123 atomic_set(&dev->refcnt, 1);
5125 /* NAPI wants this */
5126 INIT_LIST_HEAD(&dev->napi_list);
5128 /* a dummy interface is started by default */
5129 set_bit(__LINK_STATE_PRESENT, &dev->state);
5130 set_bit(__LINK_STATE_START, &dev->state);
5134 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5138 * register_netdev - register a network device
5139 * @dev: device to register
5141 * Take a completed network device structure and add it to the kernel
5142 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5143 * chain. 0 is returned on success. A negative errno code is returned
5144 * on a failure to set up the device, or if the name is a duplicate.
5146 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5147 * and expands the device name if you passed a format string to
5150 int register_netdev(struct net_device *dev)
5157 * If the name is a format string the caller wants us to do a
5160 if (strchr(dev->name, '%')) {
5161 err = dev_alloc_name(dev, dev->name);
5166 err = register_netdevice(dev);
5171 EXPORT_SYMBOL(register_netdev);
5174 * netdev_wait_allrefs - wait until all references are gone.
5176 * This is called when unregistering network devices.
5178 * Any protocol or device that holds a reference should register
5179 * for netdevice notification, and cleanup and put back the
5180 * reference if they receive an UNREGISTER event.
5181 * We can get stuck here if buggy protocols don't correctly
5184 static void netdev_wait_allrefs(struct net_device *dev)
5186 unsigned long rebroadcast_time, warning_time;
5188 linkwatch_forget_dev(dev);
5190 rebroadcast_time = warning_time = jiffies;
5191 while (atomic_read(&dev->refcnt) != 0) {
5192 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5195 /* Rebroadcast unregister notification */
5196 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5197 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5198 * should have already handle it the first time */
5200 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5202 /* We must not have linkwatch events
5203 * pending on unregister. If this
5204 * happens, we simply run the queue
5205 * unscheduled, resulting in a noop
5208 linkwatch_run_queue();
5213 rebroadcast_time = jiffies;
5218 if (time_after(jiffies, warning_time + 10 * HZ)) {
5219 printk(KERN_EMERG "unregister_netdevice: "
5220 "waiting for %s to become free. Usage "
5222 dev->name, atomic_read(&dev->refcnt));
5223 warning_time = jiffies;
5232 * register_netdevice(x1);
5233 * register_netdevice(x2);
5235 * unregister_netdevice(y1);
5236 * unregister_netdevice(y2);
5242 * We are invoked by rtnl_unlock().
5243 * This allows us to deal with problems:
5244 * 1) We can delete sysfs objects which invoke hotplug
5245 * without deadlocking with linkwatch via keventd.
5246 * 2) Since we run with the RTNL semaphore not held, we can sleep
5247 * safely in order to wait for the netdev refcnt to drop to zero.
5249 * We must not return until all unregister events added during
5250 * the interval the lock was held have been completed.
5252 void netdev_run_todo(void)
5254 struct list_head list;
5256 /* Snapshot list, allow later requests */
5257 list_replace_init(&net_todo_list, &list);
5261 while (!list_empty(&list)) {
5262 struct net_device *dev
5263 = list_first_entry(&list, struct net_device, todo_list);
5264 list_del(&dev->todo_list);
5266 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5267 printk(KERN_ERR "network todo '%s' but state %d\n",
5268 dev->name, dev->reg_state);
5273 dev->reg_state = NETREG_UNREGISTERED;
5275 on_each_cpu(flush_backlog, dev, 1);
5277 netdev_wait_allrefs(dev);
5280 BUG_ON(atomic_read(&dev->refcnt));
5281 WARN_ON(dev->ip_ptr);
5282 WARN_ON(dev->ip6_ptr);
5283 WARN_ON(dev->dn_ptr);
5285 if (dev->destructor)
5286 dev->destructor(dev);
5288 /* Free network device */
5289 kobject_put(&dev->dev.kobj);
5294 * dev_txq_stats_fold - fold tx_queues stats
5295 * @dev: device to get statistics from
5296 * @stats: struct rtnl_link_stats64 to hold results
5298 void dev_txq_stats_fold(const struct net_device *dev,
5299 struct rtnl_link_stats64 *stats)
5301 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5303 struct netdev_queue *txq;
5305 for (i = 0; i < dev->num_tx_queues; i++) {
5306 txq = netdev_get_tx_queue(dev, i);
5307 spin_lock_bh(&txq->_xmit_lock);
5308 tx_bytes += txq->tx_bytes;
5309 tx_packets += txq->tx_packets;
5310 tx_dropped += txq->tx_dropped;
5311 spin_unlock_bh(&txq->_xmit_lock);
5313 if (tx_bytes || tx_packets || tx_dropped) {
5314 stats->tx_bytes = tx_bytes;
5315 stats->tx_packets = tx_packets;
5316 stats->tx_dropped = tx_dropped;
5319 EXPORT_SYMBOL(dev_txq_stats_fold);
5321 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5322 * fields in the same order, with only the type differing.
5324 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5325 const struct net_device_stats *netdev_stats)
5327 #if BITS_PER_LONG == 64
5328 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5329 memcpy(stats64, netdev_stats, sizeof(*stats64));
5331 size_t i, n = sizeof(*stats64) / sizeof(u64);
5332 const unsigned long *src = (const unsigned long *)netdev_stats;
5333 u64 *dst = (u64 *)stats64;
5335 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5336 sizeof(*stats64) / sizeof(u64));
5337 for (i = 0; i < n; i++)
5343 * dev_get_stats - get network device statistics
5344 * @dev: device to get statistics from
5345 * @storage: place to store stats
5347 * Get network statistics from device. Return @storage.
5348 * The device driver may provide its own method by setting
5349 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5350 * otherwise the internal statistics structure is used.
5352 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5353 struct rtnl_link_stats64 *storage)
5355 const struct net_device_ops *ops = dev->netdev_ops;
5357 if (ops->ndo_get_stats64) {
5358 memset(storage, 0, sizeof(*storage));
5359 return ops->ndo_get_stats64(dev, storage);
5361 if (ops->ndo_get_stats) {
5362 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5365 netdev_stats_to_stats64(storage, &dev->stats);
5366 dev_txq_stats_fold(dev, storage);
5369 EXPORT_SYMBOL(dev_get_stats);
5371 static void netdev_init_one_queue(struct net_device *dev,
5372 struct netdev_queue *queue,
5378 static void netdev_init_queues(struct net_device *dev)
5380 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5381 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5382 spin_lock_init(&dev->tx_global_lock);
5386 * alloc_netdev_mq - allocate network device
5387 * @sizeof_priv: size of private data to allocate space for
5388 * @name: device name format string
5389 * @setup: callback to initialize device
5390 * @queue_count: the number of subqueues to allocate
5392 * Allocates a struct net_device with private data area for driver use
5393 * and performs basic initialization. Also allocates subquue structs
5394 * for each queue on the device at the end of the netdevice.
5396 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5397 void (*setup)(struct net_device *), unsigned int queue_count)
5399 struct netdev_queue *tx;
5400 struct net_device *dev;
5402 struct net_device *p;
5404 struct netdev_rx_queue *rx;
5408 BUG_ON(strlen(name) >= sizeof(dev->name));
5410 alloc_size = sizeof(struct net_device);
5412 /* ensure 32-byte alignment of private area */
5413 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5414 alloc_size += sizeof_priv;
5416 /* ensure 32-byte alignment of whole construct */
5417 alloc_size += NETDEV_ALIGN - 1;
5419 p = kzalloc(alloc_size, GFP_KERNEL);
5421 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5425 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5427 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5433 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5435 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5440 atomic_set(&rx->count, queue_count);
5443 * Set a pointer to first element in the array which holds the
5446 for (i = 0; i < queue_count; i++)
5450 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5451 dev->padded = (char *)dev - (char *)p;
5453 if (dev_addr_init(dev))
5459 dev_net_set(dev, &init_net);
5462 dev->num_tx_queues = queue_count;
5463 dev->real_num_tx_queues = queue_count;
5467 dev->num_rx_queues = queue_count;
5470 dev->gso_max_size = GSO_MAX_SIZE;
5472 netdev_init_queues(dev);
5474 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5475 dev->ethtool_ntuple_list.count = 0;
5476 INIT_LIST_HEAD(&dev->napi_list);
5477 INIT_LIST_HEAD(&dev->unreg_list);
5478 INIT_LIST_HEAD(&dev->link_watch_list);
5479 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5481 strcpy(dev->name, name);
5494 EXPORT_SYMBOL(alloc_netdev_mq);
5497 * free_netdev - free network device
5500 * This function does the last stage of destroying an allocated device
5501 * interface. The reference to the device object is released.
5502 * If this is the last reference then it will be freed.
5504 void free_netdev(struct net_device *dev)
5506 struct napi_struct *p, *n;
5508 release_net(dev_net(dev));
5512 /* Flush device addresses */
5513 dev_addr_flush(dev);
5515 /* Clear ethtool n-tuple list */
5516 ethtool_ntuple_flush(dev);
5518 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5521 /* Compatibility with error handling in drivers */
5522 if (dev->reg_state == NETREG_UNINITIALIZED) {
5523 kfree((char *)dev - dev->padded);
5527 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5528 dev->reg_state = NETREG_RELEASED;
5530 /* will free via device release */
5531 put_device(&dev->dev);
5533 EXPORT_SYMBOL(free_netdev);
5536 * synchronize_net - Synchronize with packet receive processing
5538 * Wait for packets currently being received to be done.
5539 * Does not block later packets from starting.
5541 void synchronize_net(void)
5546 EXPORT_SYMBOL(synchronize_net);
5549 * unregister_netdevice_queue - remove device from the kernel
5553 * This function shuts down a device interface and removes it
5554 * from the kernel tables.
5555 * If head not NULL, device is queued to be unregistered later.
5557 * Callers must hold the rtnl semaphore. You may want
5558 * unregister_netdev() instead of this.
5561 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5566 list_move_tail(&dev->unreg_list, head);
5568 rollback_registered(dev);
5569 /* Finish processing unregister after unlock */
5573 EXPORT_SYMBOL(unregister_netdevice_queue);
5576 * unregister_netdevice_many - unregister many devices
5577 * @head: list of devices
5579 void unregister_netdevice_many(struct list_head *head)
5581 struct net_device *dev;
5583 if (!list_empty(head)) {
5584 rollback_registered_many(head);
5585 list_for_each_entry(dev, head, unreg_list)
5589 EXPORT_SYMBOL(unregister_netdevice_many);
5592 * unregister_netdev - remove device from the kernel
5595 * This function shuts down a device interface and removes it
5596 * from the kernel tables.
5598 * This is just a wrapper for unregister_netdevice that takes
5599 * the rtnl semaphore. In general you want to use this and not
5600 * unregister_netdevice.
5602 void unregister_netdev(struct net_device *dev)
5605 unregister_netdevice(dev);
5608 EXPORT_SYMBOL(unregister_netdev);
5611 * dev_change_net_namespace - move device to different nethost namespace
5613 * @net: network namespace
5614 * @pat: If not NULL name pattern to try if the current device name
5615 * is already taken in the destination network namespace.
5617 * This function shuts down a device interface and moves it
5618 * to a new network namespace. On success 0 is returned, on
5619 * a failure a netagive errno code is returned.
5621 * Callers must hold the rtnl semaphore.
5624 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5630 /* Don't allow namespace local devices to be moved. */
5632 if (dev->features & NETIF_F_NETNS_LOCAL)
5635 /* Ensure the device has been registrered */
5637 if (dev->reg_state != NETREG_REGISTERED)
5640 /* Get out if there is nothing todo */
5642 if (net_eq(dev_net(dev), net))
5645 /* Pick the destination device name, and ensure
5646 * we can use it in the destination network namespace.
5649 if (__dev_get_by_name(net, dev->name)) {
5650 /* We get here if we can't use the current device name */
5653 if (dev_get_valid_name(dev, pat, 1))
5658 * And now a mini version of register_netdevice unregister_netdevice.
5661 /* If device is running close it first. */
5664 /* And unlink it from device chain */
5666 unlist_netdevice(dev);
5670 /* Shutdown queueing discipline. */
5673 /* Notify protocols, that we are about to destroy
5674 this device. They should clean all the things.
5676 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5677 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5680 * Flush the unicast and multicast chains
5685 /* Actually switch the network namespace */
5686 dev_net_set(dev, net);
5688 /* If there is an ifindex conflict assign a new one */
5689 if (__dev_get_by_index(net, dev->ifindex)) {
5690 int iflink = (dev->iflink == dev->ifindex);
5691 dev->ifindex = dev_new_index(net);
5693 dev->iflink = dev->ifindex;
5696 /* Fixup kobjects */
5697 err = device_rename(&dev->dev, dev->name);
5700 /* Add the device back in the hashes */
5701 list_netdevice(dev);
5703 /* Notify protocols, that a new device appeared. */
5704 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5707 * Prevent userspace races by waiting until the network
5708 * device is fully setup before sending notifications.
5710 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5717 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5719 static int dev_cpu_callback(struct notifier_block *nfb,
5720 unsigned long action,
5723 struct sk_buff **list_skb;
5724 struct sk_buff *skb;
5725 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5726 struct softnet_data *sd, *oldsd;
5728 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5731 local_irq_disable();
5732 cpu = smp_processor_id();
5733 sd = &per_cpu(softnet_data, cpu);
5734 oldsd = &per_cpu(softnet_data, oldcpu);
5736 /* Find end of our completion_queue. */
5737 list_skb = &sd->completion_queue;
5739 list_skb = &(*list_skb)->next;
5740 /* Append completion queue from offline CPU. */
5741 *list_skb = oldsd->completion_queue;
5742 oldsd->completion_queue = NULL;
5744 /* Append output queue from offline CPU. */
5745 if (oldsd->output_queue) {
5746 *sd->output_queue_tailp = oldsd->output_queue;
5747 sd->output_queue_tailp = oldsd->output_queue_tailp;
5748 oldsd->output_queue = NULL;
5749 oldsd->output_queue_tailp = &oldsd->output_queue;
5752 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5755 /* Process offline CPU's input_pkt_queue */
5756 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5758 input_queue_head_incr(oldsd);
5760 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5762 input_queue_head_incr(oldsd);
5770 * netdev_increment_features - increment feature set by one
5771 * @all: current feature set
5772 * @one: new feature set
5773 * @mask: mask feature set
5775 * Computes a new feature set after adding a device with feature set
5776 * @one to the master device with current feature set @all. Will not
5777 * enable anything that is off in @mask. Returns the new feature set.
5779 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5782 /* If device needs checksumming, downgrade to it. */
5783 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5784 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5785 else if (mask & NETIF_F_ALL_CSUM) {
5786 /* If one device supports v4/v6 checksumming, set for all. */
5787 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5788 !(all & NETIF_F_GEN_CSUM)) {
5789 all &= ~NETIF_F_ALL_CSUM;
5790 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5793 /* If one device supports hw checksumming, set for all. */
5794 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5795 all &= ~NETIF_F_ALL_CSUM;
5796 all |= NETIF_F_HW_CSUM;
5800 one |= NETIF_F_ALL_CSUM;
5802 one |= all & NETIF_F_ONE_FOR_ALL;
5803 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5804 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5808 EXPORT_SYMBOL(netdev_increment_features);
5810 static struct hlist_head *netdev_create_hash(void)
5813 struct hlist_head *hash;
5815 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5817 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5818 INIT_HLIST_HEAD(&hash[i]);
5823 /* Initialize per network namespace state */
5824 static int __net_init netdev_init(struct net *net)
5826 INIT_LIST_HEAD(&net->dev_base_head);
5828 net->dev_name_head = netdev_create_hash();
5829 if (net->dev_name_head == NULL)
5832 net->dev_index_head = netdev_create_hash();
5833 if (net->dev_index_head == NULL)
5839 kfree(net->dev_name_head);
5845 * netdev_drivername - network driver for the device
5846 * @dev: network device
5847 * @buffer: buffer for resulting name
5848 * @len: size of buffer
5850 * Determine network driver for device.
5852 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5854 const struct device_driver *driver;
5855 const struct device *parent;
5857 if (len <= 0 || !buffer)
5861 parent = dev->dev.parent;
5866 driver = parent->driver;
5867 if (driver && driver->name)
5868 strlcpy(buffer, driver->name, len);
5872 static int __netdev_printk(const char *level, const struct net_device *dev,
5873 struct va_format *vaf)
5877 if (dev && dev->dev.parent)
5878 r = dev_printk(level, dev->dev.parent, "%s: %pV",
5879 netdev_name(dev), vaf);
5881 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
5883 r = printk("%s(NULL net_device): %pV", level, vaf);
5888 int netdev_printk(const char *level, const struct net_device *dev,
5889 const char *format, ...)
5891 struct va_format vaf;
5895 va_start(args, format);
5900 r = __netdev_printk(level, dev, &vaf);
5905 EXPORT_SYMBOL(netdev_printk);
5907 #define define_netdev_printk_level(func, level) \
5908 int func(const struct net_device *dev, const char *fmt, ...) \
5911 struct va_format vaf; \
5914 va_start(args, fmt); \
5919 r = __netdev_printk(level, dev, &vaf); \
5924 EXPORT_SYMBOL(func);
5926 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
5927 define_netdev_printk_level(netdev_alert, KERN_ALERT);
5928 define_netdev_printk_level(netdev_crit, KERN_CRIT);
5929 define_netdev_printk_level(netdev_err, KERN_ERR);
5930 define_netdev_printk_level(netdev_warn, KERN_WARNING);
5931 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
5932 define_netdev_printk_level(netdev_info, KERN_INFO);
5934 static void __net_exit netdev_exit(struct net *net)
5936 kfree(net->dev_name_head);
5937 kfree(net->dev_index_head);
5940 static struct pernet_operations __net_initdata netdev_net_ops = {
5941 .init = netdev_init,
5942 .exit = netdev_exit,
5945 static void __net_exit default_device_exit(struct net *net)
5947 struct net_device *dev, *aux;
5949 * Push all migratable network devices back to the
5950 * initial network namespace
5953 for_each_netdev_safe(net, dev, aux) {
5955 char fb_name[IFNAMSIZ];
5957 /* Ignore unmoveable devices (i.e. loopback) */
5958 if (dev->features & NETIF_F_NETNS_LOCAL)
5961 /* Leave virtual devices for the generic cleanup */
5962 if (dev->rtnl_link_ops)
5965 /* Push remaing network devices to init_net */
5966 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5967 err = dev_change_net_namespace(dev, &init_net, fb_name);
5969 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5970 __func__, dev->name, err);
5977 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5979 /* At exit all network devices most be removed from a network
5980 * namespace. Do this in the reverse order of registeration.
5981 * Do this across as many network namespaces as possible to
5982 * improve batching efficiency.
5984 struct net_device *dev;
5986 LIST_HEAD(dev_kill_list);
5989 list_for_each_entry(net, net_list, exit_list) {
5990 for_each_netdev_reverse(net, dev) {
5991 if (dev->rtnl_link_ops)
5992 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5994 unregister_netdevice_queue(dev, &dev_kill_list);
5997 unregister_netdevice_many(&dev_kill_list);
6001 static struct pernet_operations __net_initdata default_device_ops = {
6002 .exit = default_device_exit,
6003 .exit_batch = default_device_exit_batch,
6007 * Initialize the DEV module. At boot time this walks the device list and
6008 * unhooks any devices that fail to initialise (normally hardware not
6009 * present) and leaves us with a valid list of present and active devices.
6014 * This is called single threaded during boot, so no need
6015 * to take the rtnl semaphore.
6017 static int __init net_dev_init(void)
6019 int i, rc = -ENOMEM;
6021 BUG_ON(!dev_boot_phase);
6023 if (dev_proc_init())
6026 if (netdev_kobject_init())
6029 INIT_LIST_HEAD(&ptype_all);
6030 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6031 INIT_LIST_HEAD(&ptype_base[i]);
6033 if (register_pernet_subsys(&netdev_net_ops))
6037 * Initialise the packet receive queues.
6040 for_each_possible_cpu(i) {
6041 struct softnet_data *sd = &per_cpu(softnet_data, i);
6043 memset(sd, 0, sizeof(*sd));
6044 skb_queue_head_init(&sd->input_pkt_queue);
6045 skb_queue_head_init(&sd->process_queue);
6046 sd->completion_queue = NULL;
6047 INIT_LIST_HEAD(&sd->poll_list);
6048 sd->output_queue = NULL;
6049 sd->output_queue_tailp = &sd->output_queue;
6051 sd->csd.func = rps_trigger_softirq;
6057 sd->backlog.poll = process_backlog;
6058 sd->backlog.weight = weight_p;
6059 sd->backlog.gro_list = NULL;
6060 sd->backlog.gro_count = 0;
6065 /* The loopback device is special if any other network devices
6066 * is present in a network namespace the loopback device must
6067 * be present. Since we now dynamically allocate and free the
6068 * loopback device ensure this invariant is maintained by
6069 * keeping the loopback device as the first device on the
6070 * list of network devices. Ensuring the loopback devices
6071 * is the first device that appears and the last network device
6074 if (register_pernet_device(&loopback_net_ops))
6077 if (register_pernet_device(&default_device_ops))
6080 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6081 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6083 hotcpu_notifier(dev_cpu_callback, 0);
6091 subsys_initcall(net_dev_init);
6093 static int __init initialize_hashrnd(void)
6095 get_random_bytes(&hashrnd, sizeof(hashrnd));
6099 late_initcall_sync(initialize_hashrnd);