Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland...
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / core / dev.c
blob78b5a89b0f40a455e4229fb04b291e87877b491b
1 /*
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
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
14 * Additional Authors:
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>
22 * Changes:
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
34 * drivers
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
44 * call a packet.
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
50 * changes.
51 * Rudi Cilibrasi : Pass the right thing to
52 * set_mac_address()
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
58 * 1 device.
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
66 * the backlog queue.
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>
87 #include <linux/mm.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>
99 #include <net/sock.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
104 #include <net/dst.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>
125 #include <net/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 * The list of packet types we will receive (as opposed to discard)
146 * and the routines to invoke.
148 * Why 16. Because with 16 the only overlap we get on a hash of the
149 * low nibble of the protocol value is RARP/SNAP/X.25.
151 * NOTE: That is no longer true with the addition of VLAN tags. Not
152 * sure which should go first, but I bet it won't make much
153 * difference if we are running VLANs. The good news is that
154 * this protocol won't be in the list unless compiled in, so
155 * the average user (w/out VLANs) will not be adversely affected.
156 * --BLG
158 * 0800 IP
159 * 8100 802.1Q VLAN
160 * 0001 802.3
161 * 0002 AX.25
162 * 0004 802.2
163 * 8035 RARP
164 * 0005 SNAP
165 * 0805 X.25
166 * 0806 ARP
167 * 8137 IPX
168 * 0009 Localtalk
169 * 86DD IPv6
172 #define PTYPE_HASH_SIZE (16)
173 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
175 static DEFINE_SPINLOCK(ptype_lock);
176 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
177 static struct list_head ptype_all __read_mostly; /* Taps */
180 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
181 * semaphore.
183 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
185 * Writers must hold the rtnl semaphore while they loop through the
186 * dev_base_head list, and hold dev_base_lock for writing when they do the
187 * actual updates. This allows pure readers to access the list even
188 * while a writer is preparing to update it.
190 * To put it another way, dev_base_lock is held for writing only to
191 * protect against pure readers; the rtnl semaphore provides the
192 * protection against other writers.
194 * See, for example usages, register_netdevice() and
195 * unregister_netdevice(), which must be called with the rtnl
196 * semaphore held.
198 DEFINE_RWLOCK(dev_base_lock);
199 EXPORT_SYMBOL(dev_base_lock);
201 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
203 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
204 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
207 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
209 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
212 static inline void rps_lock(struct softnet_data *sd)
214 #ifdef CONFIG_RPS
215 spin_lock(&sd->input_pkt_queue.lock);
216 #endif
219 static inline void rps_unlock(struct softnet_data *sd)
221 #ifdef CONFIG_RPS
222 spin_unlock(&sd->input_pkt_queue.lock);
223 #endif
226 /* Device list insertion */
227 static int list_netdevice(struct net_device *dev)
229 struct net *net = dev_net(dev);
231 ASSERT_RTNL();
233 write_lock_bh(&dev_base_lock);
234 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
235 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
236 hlist_add_head_rcu(&dev->index_hlist,
237 dev_index_hash(net, dev->ifindex));
238 write_unlock_bh(&dev_base_lock);
239 return 0;
242 /* Device list removal
243 * caller must respect a RCU grace period before freeing/reusing dev
245 static void unlist_netdevice(struct net_device *dev)
247 ASSERT_RTNL();
249 /* Unlink dev from the device chain */
250 write_lock_bh(&dev_base_lock);
251 list_del_rcu(&dev->dev_list);
252 hlist_del_rcu(&dev->name_hlist);
253 hlist_del_rcu(&dev->index_hlist);
254 write_unlock_bh(&dev_base_lock);
258 * Our notifier list
261 static RAW_NOTIFIER_HEAD(netdev_chain);
264 * Device drivers call our routines to queue packets here. We empty the
265 * queue in the local softnet handler.
268 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
269 EXPORT_PER_CPU_SYMBOL(softnet_data);
271 #ifdef CONFIG_LOCKDEP
273 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
274 * according to dev->type
276 static const unsigned short netdev_lock_type[] =
277 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
278 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
279 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
280 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
281 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
282 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
283 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
284 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
285 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
286 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
287 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
288 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
289 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
290 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
291 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
292 ARPHRD_VOID, ARPHRD_NONE};
294 static const char *const netdev_lock_name[] =
295 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
296 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
297 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
298 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
299 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
300 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
301 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
302 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
303 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
304 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
305 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
306 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
307 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
308 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
309 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
310 "_xmit_VOID", "_xmit_NONE"};
312 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
313 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
315 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
317 int i;
319 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
320 if (netdev_lock_type[i] == dev_type)
321 return i;
322 /* the last key is used by default */
323 return ARRAY_SIZE(netdev_lock_type) - 1;
326 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
327 unsigned short dev_type)
329 int i;
331 i = netdev_lock_pos(dev_type);
332 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
333 netdev_lock_name[i]);
336 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
338 int i;
340 i = netdev_lock_pos(dev->type);
341 lockdep_set_class_and_name(&dev->addr_list_lock,
342 &netdev_addr_lock_key[i],
343 netdev_lock_name[i]);
345 #else
346 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
347 unsigned short dev_type)
350 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
353 #endif
355 /*******************************************************************************
357 Protocol management and registration routines
359 *******************************************************************************/
362 * Add a protocol ID to the list. Now that the input handler is
363 * smarter we can dispense with all the messy stuff that used to be
364 * here.
366 * BEWARE!!! Protocol handlers, mangling input packets,
367 * MUST BE last in hash buckets and checking protocol handlers
368 * MUST start from promiscuous ptype_all chain in net_bh.
369 * It is true now, do not change it.
370 * Explanation follows: if protocol handler, mangling packet, will
371 * be the first on list, it is not able to sense, that packet
372 * is cloned and should be copied-on-write, so that it will
373 * change it and subsequent readers will get broken packet.
374 * --ANK (980803)
377 static inline struct list_head *ptype_head(const struct packet_type *pt)
379 if (pt->type == htons(ETH_P_ALL))
380 return &ptype_all;
381 else
382 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
386 * dev_add_pack - add packet handler
387 * @pt: packet type declaration
389 * Add a protocol handler to the networking stack. The passed &packet_type
390 * is linked into kernel lists and may not be freed until it has been
391 * removed from the kernel lists.
393 * This call does not sleep therefore it can not
394 * guarantee all CPU's that are in middle of receiving packets
395 * will see the new packet type (until the next received packet).
398 void dev_add_pack(struct packet_type *pt)
400 struct list_head *head = ptype_head(pt);
402 spin_lock(&ptype_lock);
403 list_add_rcu(&pt->list, head);
404 spin_unlock(&ptype_lock);
406 EXPORT_SYMBOL(dev_add_pack);
409 * __dev_remove_pack - remove packet handler
410 * @pt: packet type declaration
412 * Remove a protocol handler that was previously added to the kernel
413 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
414 * from the kernel lists and can be freed or reused once this function
415 * returns.
417 * The packet type might still be in use by receivers
418 * and must not be freed until after all the CPU's have gone
419 * through a quiescent state.
421 void __dev_remove_pack(struct packet_type *pt)
423 struct list_head *head = ptype_head(pt);
424 struct packet_type *pt1;
426 spin_lock(&ptype_lock);
428 list_for_each_entry(pt1, head, list) {
429 if (pt == pt1) {
430 list_del_rcu(&pt->list);
431 goto out;
435 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
436 out:
437 spin_unlock(&ptype_lock);
439 EXPORT_SYMBOL(__dev_remove_pack);
442 * dev_remove_pack - remove packet handler
443 * @pt: packet type declaration
445 * Remove a protocol handler that was previously added to the kernel
446 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
447 * from the kernel lists and can be freed or reused once this function
448 * returns.
450 * This call sleeps to guarantee that no CPU is looking at the packet
451 * type after return.
453 void dev_remove_pack(struct packet_type *pt)
455 __dev_remove_pack(pt);
457 synchronize_net();
459 EXPORT_SYMBOL(dev_remove_pack);
461 /******************************************************************************
463 Device Boot-time Settings Routines
465 *******************************************************************************/
467 /* Boot time configuration table */
468 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
471 * netdev_boot_setup_add - add new setup entry
472 * @name: name of the device
473 * @map: configured settings for the device
475 * Adds new setup entry to the dev_boot_setup list. The function
476 * returns 0 on error and 1 on success. This is a generic routine to
477 * all netdevices.
479 static int netdev_boot_setup_add(char *name, struct ifmap *map)
481 struct netdev_boot_setup *s;
482 int i;
484 s = dev_boot_setup;
485 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
486 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
487 memset(s[i].name, 0, sizeof(s[i].name));
488 strlcpy(s[i].name, name, IFNAMSIZ);
489 memcpy(&s[i].map, map, sizeof(s[i].map));
490 break;
494 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
498 * netdev_boot_setup_check - check boot time settings
499 * @dev: the netdevice
501 * Check boot time settings for the device.
502 * The found settings are set for the device to be used
503 * later in the device probing.
504 * Returns 0 if no settings found, 1 if they are.
506 int netdev_boot_setup_check(struct net_device *dev)
508 struct netdev_boot_setup *s = dev_boot_setup;
509 int i;
511 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
512 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
513 !strcmp(dev->name, s[i].name)) {
514 dev->irq = s[i].map.irq;
515 dev->base_addr = s[i].map.base_addr;
516 dev->mem_start = s[i].map.mem_start;
517 dev->mem_end = s[i].map.mem_end;
518 return 1;
521 return 0;
523 EXPORT_SYMBOL(netdev_boot_setup_check);
527 * netdev_boot_base - get address from boot time settings
528 * @prefix: prefix for network device
529 * @unit: id for network device
531 * Check boot time settings for the base address of device.
532 * The found settings are set for the device to be used
533 * later in the device probing.
534 * Returns 0 if no settings found.
536 unsigned long netdev_boot_base(const char *prefix, int unit)
538 const struct netdev_boot_setup *s = dev_boot_setup;
539 char name[IFNAMSIZ];
540 int i;
542 sprintf(name, "%s%d", prefix, unit);
545 * If device already registered then return base of 1
546 * to indicate not to probe for this interface
548 if (__dev_get_by_name(&init_net, name))
549 return 1;
551 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
552 if (!strcmp(name, s[i].name))
553 return s[i].map.base_addr;
554 return 0;
558 * Saves at boot time configured settings for any netdevice.
560 int __init netdev_boot_setup(char *str)
562 int ints[5];
563 struct ifmap map;
565 str = get_options(str, ARRAY_SIZE(ints), ints);
566 if (!str || !*str)
567 return 0;
569 /* Save settings */
570 memset(&map, 0, sizeof(map));
571 if (ints[0] > 0)
572 map.irq = ints[1];
573 if (ints[0] > 1)
574 map.base_addr = ints[2];
575 if (ints[0] > 2)
576 map.mem_start = ints[3];
577 if (ints[0] > 3)
578 map.mem_end = ints[4];
580 /* Add new entry to the list */
581 return netdev_boot_setup_add(str, &map);
584 __setup("netdev=", netdev_boot_setup);
586 /*******************************************************************************
588 Device Interface Subroutines
590 *******************************************************************************/
593 * __dev_get_by_name - find a device by its name
594 * @net: the applicable net namespace
595 * @name: name to find
597 * Find an interface by name. Must be called under RTNL semaphore
598 * or @dev_base_lock. If the name is found a pointer to the device
599 * is returned. If the name is not found then %NULL is returned. The
600 * reference counters are not incremented so the caller must be
601 * careful with locks.
604 struct net_device *__dev_get_by_name(struct net *net, const char *name)
606 struct hlist_node *p;
607 struct net_device *dev;
608 struct hlist_head *head = dev_name_hash(net, name);
610 hlist_for_each_entry(dev, p, head, name_hlist)
611 if (!strncmp(dev->name, name, IFNAMSIZ))
612 return dev;
614 return NULL;
616 EXPORT_SYMBOL(__dev_get_by_name);
619 * dev_get_by_name_rcu - find a device by its name
620 * @net: the applicable net namespace
621 * @name: name to find
623 * Find an interface by name.
624 * If the name is found a pointer to the device is returned.
625 * If the name is not found then %NULL is returned.
626 * The reference counters are not incremented so the caller must be
627 * careful with locks. The caller must hold RCU lock.
630 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
632 struct hlist_node *p;
633 struct net_device *dev;
634 struct hlist_head *head = dev_name_hash(net, name);
636 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
637 if (!strncmp(dev->name, name, IFNAMSIZ))
638 return dev;
640 return NULL;
642 EXPORT_SYMBOL(dev_get_by_name_rcu);
645 * dev_get_by_name - find a device by its name
646 * @net: the applicable net namespace
647 * @name: name to find
649 * Find an interface by name. This can be called from any
650 * context and does its own locking. The returned handle has
651 * the usage count incremented and the caller must use dev_put() to
652 * release it when it is no longer needed. %NULL is returned if no
653 * matching device is found.
656 struct net_device *dev_get_by_name(struct net *net, const char *name)
658 struct net_device *dev;
660 rcu_read_lock();
661 dev = dev_get_by_name_rcu(net, name);
662 if (dev)
663 dev_hold(dev);
664 rcu_read_unlock();
665 return dev;
667 EXPORT_SYMBOL(dev_get_by_name);
670 * __dev_get_by_index - find a device by its ifindex
671 * @net: the applicable net namespace
672 * @ifindex: index of device
674 * Search for an interface by index. Returns %NULL if the device
675 * is not found or a pointer to the device. The device has not
676 * had its reference counter increased so the caller must be careful
677 * about locking. The caller must hold either the RTNL semaphore
678 * or @dev_base_lock.
681 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
683 struct hlist_node *p;
684 struct net_device *dev;
685 struct hlist_head *head = dev_index_hash(net, ifindex);
687 hlist_for_each_entry(dev, p, head, index_hlist)
688 if (dev->ifindex == ifindex)
689 return dev;
691 return NULL;
693 EXPORT_SYMBOL(__dev_get_by_index);
696 * dev_get_by_index_rcu - find a device by its ifindex
697 * @net: the applicable net namespace
698 * @ifindex: index of device
700 * Search for an interface by index. Returns %NULL if the device
701 * is not found or a pointer to the device. The device has not
702 * had its reference counter increased so the caller must be careful
703 * about locking. The caller must hold RCU lock.
706 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
708 struct hlist_node *p;
709 struct net_device *dev;
710 struct hlist_head *head = dev_index_hash(net, ifindex);
712 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
713 if (dev->ifindex == ifindex)
714 return dev;
716 return NULL;
718 EXPORT_SYMBOL(dev_get_by_index_rcu);
722 * dev_get_by_index - find a device by its ifindex
723 * @net: the applicable net namespace
724 * @ifindex: index of device
726 * Search for an interface by index. Returns NULL if the device
727 * is not found or a pointer to the device. The device returned has
728 * had a reference added and the pointer is safe until the user calls
729 * dev_put to indicate they have finished with it.
732 struct net_device *dev_get_by_index(struct net *net, int ifindex)
734 struct net_device *dev;
736 rcu_read_lock();
737 dev = dev_get_by_index_rcu(net, ifindex);
738 if (dev)
739 dev_hold(dev);
740 rcu_read_unlock();
741 return dev;
743 EXPORT_SYMBOL(dev_get_by_index);
746 * dev_getbyhwaddr - find a device by its hardware address
747 * @net: the applicable net namespace
748 * @type: media type of device
749 * @ha: hardware address
751 * Search for an interface by MAC address. Returns NULL if the device
752 * is not found or a pointer to the device. The caller must hold the
753 * rtnl semaphore. The returned device has not had its ref count increased
754 * and the caller must therefore be careful about locking
756 * BUGS:
757 * If the API was consistent this would be __dev_get_by_hwaddr
760 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
762 struct net_device *dev;
764 ASSERT_RTNL();
766 for_each_netdev(net, dev)
767 if (dev->type == type &&
768 !memcmp(dev->dev_addr, ha, dev->addr_len))
769 return dev;
771 return NULL;
773 EXPORT_SYMBOL(dev_getbyhwaddr);
775 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
777 struct net_device *dev;
779 ASSERT_RTNL();
780 for_each_netdev(net, dev)
781 if (dev->type == type)
782 return dev;
784 return NULL;
786 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
788 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
790 struct net_device *dev, *ret = NULL;
792 rcu_read_lock();
793 for_each_netdev_rcu(net, dev)
794 if (dev->type == type) {
795 dev_hold(dev);
796 ret = dev;
797 break;
799 rcu_read_unlock();
800 return ret;
802 EXPORT_SYMBOL(dev_getfirstbyhwtype);
805 * dev_get_by_flags_rcu - find any device with given flags
806 * @net: the applicable net namespace
807 * @if_flags: IFF_* values
808 * @mask: bitmask of bits in if_flags to check
810 * Search for any interface with the given flags. Returns NULL if a device
811 * is not found or a pointer to the device. Must be called inside
812 * rcu_read_lock(), and result refcount is unchanged.
815 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
816 unsigned short mask)
818 struct net_device *dev, *ret;
820 ret = NULL;
821 for_each_netdev_rcu(net, dev) {
822 if (((dev->flags ^ if_flags) & mask) == 0) {
823 ret = dev;
824 break;
827 return ret;
829 EXPORT_SYMBOL(dev_get_by_flags_rcu);
832 * dev_valid_name - check if name is okay for network device
833 * @name: name string
835 * Network device names need to be valid file names to
836 * to allow sysfs to work. We also disallow any kind of
837 * whitespace.
839 int dev_valid_name(const char *name)
841 if (*name == '\0')
842 return 0;
843 if (strlen(name) >= IFNAMSIZ)
844 return 0;
845 if (!strcmp(name, ".") || !strcmp(name, ".."))
846 return 0;
848 while (*name) {
849 if (*name == '/' || isspace(*name))
850 return 0;
851 name++;
853 return 1;
855 EXPORT_SYMBOL(dev_valid_name);
858 * __dev_alloc_name - allocate a name for a device
859 * @net: network namespace to allocate the device name in
860 * @name: name format string
861 * @buf: scratch buffer and result name string
863 * Passed a format string - eg "lt%d" it will try and find a suitable
864 * id. It scans list of devices to build up a free map, then chooses
865 * the first empty slot. The caller must hold the dev_base or rtnl lock
866 * while allocating the name and adding the device in order to avoid
867 * duplicates.
868 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
869 * Returns the number of the unit assigned or a negative errno code.
872 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
874 int i = 0;
875 const char *p;
876 const int max_netdevices = 8*PAGE_SIZE;
877 unsigned long *inuse;
878 struct net_device *d;
880 p = strnchr(name, IFNAMSIZ-1, '%');
881 if (p) {
883 * Verify the string as this thing may have come from
884 * the user. There must be either one "%d" and no other "%"
885 * characters.
887 if (p[1] != 'd' || strchr(p + 2, '%'))
888 return -EINVAL;
890 /* Use one page as a bit array of possible slots */
891 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
892 if (!inuse)
893 return -ENOMEM;
895 for_each_netdev(net, d) {
896 if (!sscanf(d->name, name, &i))
897 continue;
898 if (i < 0 || i >= max_netdevices)
899 continue;
901 /* avoid cases where sscanf is not exact inverse of printf */
902 snprintf(buf, IFNAMSIZ, name, i);
903 if (!strncmp(buf, d->name, IFNAMSIZ))
904 set_bit(i, inuse);
907 i = find_first_zero_bit(inuse, max_netdevices);
908 free_page((unsigned long) inuse);
911 if (buf != name)
912 snprintf(buf, IFNAMSIZ, name, i);
913 if (!__dev_get_by_name(net, buf))
914 return i;
916 /* It is possible to run out of possible slots
917 * when the name is long and there isn't enough space left
918 * for the digits, or if all bits are used.
920 return -ENFILE;
924 * dev_alloc_name - allocate a name for a device
925 * @dev: device
926 * @name: name format string
928 * Passed a format string - eg "lt%d" it will try and find a suitable
929 * id. It scans list of devices to build up a free map, then chooses
930 * the first empty slot. The caller must hold the dev_base or rtnl lock
931 * while allocating the name and adding the device in order to avoid
932 * duplicates.
933 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
934 * Returns the number of the unit assigned or a negative errno code.
937 int dev_alloc_name(struct net_device *dev, const char *name)
939 char buf[IFNAMSIZ];
940 struct net *net;
941 int ret;
943 BUG_ON(!dev_net(dev));
944 net = dev_net(dev);
945 ret = __dev_alloc_name(net, name, buf);
946 if (ret >= 0)
947 strlcpy(dev->name, buf, IFNAMSIZ);
948 return ret;
950 EXPORT_SYMBOL(dev_alloc_name);
952 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
954 struct net *net;
956 BUG_ON(!dev_net(dev));
957 net = dev_net(dev);
959 if (!dev_valid_name(name))
960 return -EINVAL;
962 if (fmt && strchr(name, '%'))
963 return dev_alloc_name(dev, name);
964 else if (__dev_get_by_name(net, name))
965 return -EEXIST;
966 else if (dev->name != name)
967 strlcpy(dev->name, name, IFNAMSIZ);
969 return 0;
973 * dev_change_name - change name of a device
974 * @dev: device
975 * @newname: name (or format string) must be at least IFNAMSIZ
977 * Change name of a device, can pass format strings "eth%d".
978 * for wildcarding.
980 int dev_change_name(struct net_device *dev, const char *newname)
982 char oldname[IFNAMSIZ];
983 int err = 0;
984 int ret;
985 struct net *net;
987 ASSERT_RTNL();
988 BUG_ON(!dev_net(dev));
990 net = dev_net(dev);
991 if (dev->flags & IFF_UP)
992 return -EBUSY;
994 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
995 return 0;
997 memcpy(oldname, dev->name, IFNAMSIZ);
999 err = dev_get_valid_name(dev, newname, 1);
1000 if (err < 0)
1001 return err;
1003 rollback:
1004 ret = device_rename(&dev->dev, dev->name);
1005 if (ret) {
1006 memcpy(dev->name, oldname, IFNAMSIZ);
1007 return ret;
1010 write_lock_bh(&dev_base_lock);
1011 hlist_del(&dev->name_hlist);
1012 write_unlock_bh(&dev_base_lock);
1014 synchronize_rcu();
1016 write_lock_bh(&dev_base_lock);
1017 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1018 write_unlock_bh(&dev_base_lock);
1020 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1021 ret = notifier_to_errno(ret);
1023 if (ret) {
1024 /* err >= 0 after dev_alloc_name() or stores the first errno */
1025 if (err >= 0) {
1026 err = ret;
1027 memcpy(dev->name, oldname, IFNAMSIZ);
1028 goto rollback;
1029 } else {
1030 printk(KERN_ERR
1031 "%s: name change rollback failed: %d.\n",
1032 dev->name, ret);
1036 return err;
1040 * dev_set_alias - change ifalias of a device
1041 * @dev: device
1042 * @alias: name up to IFALIASZ
1043 * @len: limit of bytes to copy from info
1045 * Set ifalias for a device,
1047 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1049 ASSERT_RTNL();
1051 if (len >= IFALIASZ)
1052 return -EINVAL;
1054 if (!len) {
1055 if (dev->ifalias) {
1056 kfree(dev->ifalias);
1057 dev->ifalias = NULL;
1059 return 0;
1062 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1063 if (!dev->ifalias)
1064 return -ENOMEM;
1066 strlcpy(dev->ifalias, alias, len+1);
1067 return len;
1072 * netdev_features_change - device changes features
1073 * @dev: device to cause notification
1075 * Called to indicate a device has changed features.
1077 void netdev_features_change(struct net_device *dev)
1079 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1081 EXPORT_SYMBOL(netdev_features_change);
1084 * netdev_state_change - device changes state
1085 * @dev: device to cause notification
1087 * Called to indicate a device has changed state. This function calls
1088 * the notifier chains for netdev_chain and sends a NEWLINK message
1089 * to the routing socket.
1091 void netdev_state_change(struct net_device *dev)
1093 if (dev->flags & IFF_UP) {
1094 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1095 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1098 EXPORT_SYMBOL(netdev_state_change);
1100 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1102 return call_netdevice_notifiers(event, dev);
1104 EXPORT_SYMBOL(netdev_bonding_change);
1107 * dev_load - load a network module
1108 * @net: the applicable net namespace
1109 * @name: name of interface
1111 * If a network interface is not present and the process has suitable
1112 * privileges this function loads the module. If module loading is not
1113 * available in this kernel then it becomes a nop.
1116 void dev_load(struct net *net, const char *name)
1118 struct net_device *dev;
1120 rcu_read_lock();
1121 dev = dev_get_by_name_rcu(net, name);
1122 rcu_read_unlock();
1124 if (!dev && capable(CAP_NET_ADMIN))
1125 request_module("%s", name);
1127 EXPORT_SYMBOL(dev_load);
1129 static int __dev_open(struct net_device *dev)
1131 const struct net_device_ops *ops = dev->netdev_ops;
1132 int ret;
1134 ASSERT_RTNL();
1137 * Is it even present?
1139 if (!netif_device_present(dev))
1140 return -ENODEV;
1142 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1143 ret = notifier_to_errno(ret);
1144 if (ret)
1145 return ret;
1148 * Call device private open method
1150 set_bit(__LINK_STATE_START, &dev->state);
1152 if (ops->ndo_validate_addr)
1153 ret = ops->ndo_validate_addr(dev);
1155 if (!ret && ops->ndo_open)
1156 ret = ops->ndo_open(dev);
1159 * If it went open OK then:
1162 if (ret)
1163 clear_bit(__LINK_STATE_START, &dev->state);
1164 else {
1166 * Set the flags.
1168 dev->flags |= IFF_UP;
1171 * Enable NET_DMA
1173 net_dmaengine_get();
1176 * Initialize multicasting status
1178 dev_set_rx_mode(dev);
1181 * Wakeup transmit queue engine
1183 dev_activate(dev);
1186 return ret;
1190 * dev_open - prepare an interface for use.
1191 * @dev: device to open
1193 * Takes a device from down to up state. The device's private open
1194 * function is invoked and then the multicast lists are loaded. Finally
1195 * the device is moved into the up state and a %NETDEV_UP message is
1196 * sent to the netdev notifier chain.
1198 * Calling this function on an active interface is a nop. On a failure
1199 * a negative errno code is returned.
1201 int dev_open(struct net_device *dev)
1203 int ret;
1206 * Is it already up?
1208 if (dev->flags & IFF_UP)
1209 return 0;
1212 * Open device
1214 ret = __dev_open(dev);
1215 if (ret < 0)
1216 return ret;
1219 * ... and announce new interface.
1221 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1222 call_netdevice_notifiers(NETDEV_UP, dev);
1224 return ret;
1226 EXPORT_SYMBOL(dev_open);
1228 static int __dev_close(struct net_device *dev)
1230 const struct net_device_ops *ops = dev->netdev_ops;
1232 ASSERT_RTNL();
1233 might_sleep();
1236 * Tell people we are going down, so that they can
1237 * prepare to death, when device is still operating.
1239 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1241 clear_bit(__LINK_STATE_START, &dev->state);
1243 /* Synchronize to scheduled poll. We cannot touch poll list,
1244 * it can be even on different cpu. So just clear netif_running().
1246 * dev->stop() will invoke napi_disable() on all of it's
1247 * napi_struct instances on this device.
1249 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1251 dev_deactivate(dev);
1254 * Call the device specific close. This cannot fail.
1255 * Only if device is UP
1257 * We allow it to be called even after a DETACH hot-plug
1258 * event.
1260 if (ops->ndo_stop)
1261 ops->ndo_stop(dev);
1264 * Device is now down.
1267 dev->flags &= ~IFF_UP;
1270 * Shutdown NET_DMA
1272 net_dmaengine_put();
1274 return 0;
1278 * dev_close - shutdown an interface.
1279 * @dev: device to shutdown
1281 * This function moves an active device into down state. A
1282 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1283 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1284 * chain.
1286 int dev_close(struct net_device *dev)
1288 if (!(dev->flags & IFF_UP))
1289 return 0;
1291 __dev_close(dev);
1294 * Tell people we are down
1296 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1297 call_netdevice_notifiers(NETDEV_DOWN, dev);
1299 return 0;
1301 EXPORT_SYMBOL(dev_close);
1305 * dev_disable_lro - disable Large Receive Offload on a device
1306 * @dev: device
1308 * Disable Large Receive Offload (LRO) on a net device. Must be
1309 * called under RTNL. This is needed if received packets may be
1310 * forwarded to another interface.
1312 void dev_disable_lro(struct net_device *dev)
1314 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1315 dev->ethtool_ops->set_flags) {
1316 u32 flags = dev->ethtool_ops->get_flags(dev);
1317 if (flags & ETH_FLAG_LRO) {
1318 flags &= ~ETH_FLAG_LRO;
1319 dev->ethtool_ops->set_flags(dev, flags);
1322 WARN_ON(dev->features & NETIF_F_LRO);
1324 EXPORT_SYMBOL(dev_disable_lro);
1327 static int dev_boot_phase = 1;
1330 * Device change register/unregister. These are not inline or static
1331 * as we export them to the world.
1335 * register_netdevice_notifier - register a network notifier block
1336 * @nb: notifier
1338 * Register a notifier to be called when network device events occur.
1339 * The notifier passed is linked into the kernel structures and must
1340 * not be reused until it has been unregistered. A negative errno code
1341 * is returned on a failure.
1343 * When registered all registration and up events are replayed
1344 * to the new notifier to allow device to have a race free
1345 * view of the network device list.
1348 int register_netdevice_notifier(struct notifier_block *nb)
1350 struct net_device *dev;
1351 struct net_device *last;
1352 struct net *net;
1353 int err;
1355 rtnl_lock();
1356 err = raw_notifier_chain_register(&netdev_chain, nb);
1357 if (err)
1358 goto unlock;
1359 if (dev_boot_phase)
1360 goto unlock;
1361 for_each_net(net) {
1362 for_each_netdev(net, dev) {
1363 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1364 err = notifier_to_errno(err);
1365 if (err)
1366 goto rollback;
1368 if (!(dev->flags & IFF_UP))
1369 continue;
1371 nb->notifier_call(nb, NETDEV_UP, dev);
1375 unlock:
1376 rtnl_unlock();
1377 return err;
1379 rollback:
1380 last = dev;
1381 for_each_net(net) {
1382 for_each_netdev(net, dev) {
1383 if (dev == last)
1384 break;
1386 if (dev->flags & IFF_UP) {
1387 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1388 nb->notifier_call(nb, NETDEV_DOWN, dev);
1390 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1391 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1395 raw_notifier_chain_unregister(&netdev_chain, nb);
1396 goto unlock;
1398 EXPORT_SYMBOL(register_netdevice_notifier);
1401 * unregister_netdevice_notifier - unregister a network notifier block
1402 * @nb: notifier
1404 * Unregister a notifier previously registered by
1405 * register_netdevice_notifier(). The notifier is unlinked into the
1406 * kernel structures and may then be reused. A negative errno code
1407 * is returned on a failure.
1410 int unregister_netdevice_notifier(struct notifier_block *nb)
1412 int err;
1414 rtnl_lock();
1415 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1416 rtnl_unlock();
1417 return err;
1419 EXPORT_SYMBOL(unregister_netdevice_notifier);
1422 * call_netdevice_notifiers - call all network notifier blocks
1423 * @val: value passed unmodified to notifier function
1424 * @dev: net_device pointer passed unmodified to notifier function
1426 * Call all network notifier blocks. Parameters and return value
1427 * are as for raw_notifier_call_chain().
1430 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1432 ASSERT_RTNL();
1433 return raw_notifier_call_chain(&netdev_chain, val, dev);
1436 /* When > 0 there are consumers of rx skb time stamps */
1437 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1439 void net_enable_timestamp(void)
1441 atomic_inc(&netstamp_needed);
1443 EXPORT_SYMBOL(net_enable_timestamp);
1445 void net_disable_timestamp(void)
1447 atomic_dec(&netstamp_needed);
1449 EXPORT_SYMBOL(net_disable_timestamp);
1451 static inline void net_timestamp_set(struct sk_buff *skb)
1453 if (atomic_read(&netstamp_needed))
1454 __net_timestamp(skb);
1455 else
1456 skb->tstamp.tv64 = 0;
1459 static inline void net_timestamp_check(struct sk_buff *skb)
1461 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1462 __net_timestamp(skb);
1466 * dev_forward_skb - loopback an skb to another netif
1468 * @dev: destination network device
1469 * @skb: buffer to forward
1471 * return values:
1472 * NET_RX_SUCCESS (no congestion)
1473 * NET_RX_DROP (packet was dropped, but freed)
1475 * dev_forward_skb can be used for injecting an skb from the
1476 * start_xmit function of one device into the receive queue
1477 * of another device.
1479 * The receiving device may be in another namespace, so
1480 * we have to clear all information in the skb that could
1481 * impact namespace isolation.
1483 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1485 skb_orphan(skb);
1486 nf_reset(skb);
1488 if (unlikely(!(dev->flags & IFF_UP) ||
1489 (skb->len > (dev->mtu + dev->hard_header_len + VLAN_HLEN)))) {
1490 atomic_long_inc(&dev->rx_dropped);
1491 kfree_skb(skb);
1492 return NET_RX_DROP;
1494 skb_set_dev(skb, dev);
1495 skb->tstamp.tv64 = 0;
1496 skb->pkt_type = PACKET_HOST;
1497 skb->protocol = eth_type_trans(skb, dev);
1498 return netif_rx(skb);
1500 EXPORT_SYMBOL_GPL(dev_forward_skb);
1503 * Support routine. Sends outgoing frames to any network
1504 * taps currently in use.
1507 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1509 struct packet_type *ptype;
1511 #ifdef CONFIG_NET_CLS_ACT
1512 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1513 net_timestamp_set(skb);
1514 #else
1515 net_timestamp_set(skb);
1516 #endif
1518 rcu_read_lock();
1519 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1520 /* Never send packets back to the socket
1521 * they originated from - MvS (miquels@drinkel.ow.org)
1523 if ((ptype->dev == dev || !ptype->dev) &&
1524 (ptype->af_packet_priv == NULL ||
1525 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1526 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1527 if (!skb2)
1528 break;
1530 /* skb->nh should be correctly
1531 set by sender, so that the second statement is
1532 just protection against buggy protocols.
1534 skb_reset_mac_header(skb2);
1536 if (skb_network_header(skb2) < skb2->data ||
1537 skb2->network_header > skb2->tail) {
1538 if (net_ratelimit())
1539 printk(KERN_CRIT "protocol %04x is "
1540 "buggy, dev %s\n",
1541 ntohs(skb2->protocol),
1542 dev->name);
1543 skb_reset_network_header(skb2);
1546 skb2->transport_header = skb2->network_header;
1547 skb2->pkt_type = PACKET_OUTGOING;
1548 ptype->func(skb2, skb->dev, ptype, skb->dev);
1551 rcu_read_unlock();
1555 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1556 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1558 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1560 if (txq < 1 || txq > dev->num_tx_queues)
1561 return -EINVAL;
1563 if (dev->reg_state == NETREG_REGISTERED) {
1564 ASSERT_RTNL();
1566 if (txq < dev->real_num_tx_queues)
1567 qdisc_reset_all_tx_gt(dev, txq);
1570 dev->real_num_tx_queues = txq;
1571 return 0;
1573 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1575 #ifdef CONFIG_RPS
1577 * netif_set_real_num_rx_queues - set actual number of RX queues used
1578 * @dev: Network device
1579 * @rxq: Actual number of RX queues
1581 * This must be called either with the rtnl_lock held or before
1582 * registration of the net device. Returns 0 on success, or a
1583 * negative error code. If called before registration, it always
1584 * succeeds.
1586 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1588 int rc;
1590 if (rxq < 1 || rxq > dev->num_rx_queues)
1591 return -EINVAL;
1593 if (dev->reg_state == NETREG_REGISTERED) {
1594 ASSERT_RTNL();
1596 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1597 rxq);
1598 if (rc)
1599 return rc;
1602 dev->real_num_rx_queues = rxq;
1603 return 0;
1605 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1606 #endif
1608 static inline void __netif_reschedule(struct Qdisc *q)
1610 struct softnet_data *sd;
1611 unsigned long flags;
1613 local_irq_save(flags);
1614 sd = &__get_cpu_var(softnet_data);
1615 q->next_sched = NULL;
1616 *sd->output_queue_tailp = q;
1617 sd->output_queue_tailp = &q->next_sched;
1618 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1619 local_irq_restore(flags);
1622 void __netif_schedule(struct Qdisc *q)
1624 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1625 __netif_reschedule(q);
1627 EXPORT_SYMBOL(__netif_schedule);
1629 void dev_kfree_skb_irq(struct sk_buff *skb)
1631 if (atomic_dec_and_test(&skb->users)) {
1632 struct softnet_data *sd;
1633 unsigned long flags;
1635 local_irq_save(flags);
1636 sd = &__get_cpu_var(softnet_data);
1637 skb->next = sd->completion_queue;
1638 sd->completion_queue = skb;
1639 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1640 local_irq_restore(flags);
1643 EXPORT_SYMBOL(dev_kfree_skb_irq);
1645 void dev_kfree_skb_any(struct sk_buff *skb)
1647 if (in_irq() || irqs_disabled())
1648 dev_kfree_skb_irq(skb);
1649 else
1650 dev_kfree_skb(skb);
1652 EXPORT_SYMBOL(dev_kfree_skb_any);
1656 * netif_device_detach - mark device as removed
1657 * @dev: network device
1659 * Mark device as removed from system and therefore no longer available.
1661 void netif_device_detach(struct net_device *dev)
1663 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1664 netif_running(dev)) {
1665 netif_tx_stop_all_queues(dev);
1668 EXPORT_SYMBOL(netif_device_detach);
1671 * netif_device_attach - mark device as attached
1672 * @dev: network device
1674 * Mark device as attached from system and restart if needed.
1676 void netif_device_attach(struct net_device *dev)
1678 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1679 netif_running(dev)) {
1680 netif_tx_wake_all_queues(dev);
1681 __netdev_watchdog_up(dev);
1684 EXPORT_SYMBOL(netif_device_attach);
1686 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1688 return ((features & NETIF_F_GEN_CSUM) ||
1689 ((features & NETIF_F_IP_CSUM) &&
1690 protocol == htons(ETH_P_IP)) ||
1691 ((features & NETIF_F_IPV6_CSUM) &&
1692 protocol == htons(ETH_P_IPV6)) ||
1693 ((features & NETIF_F_FCOE_CRC) &&
1694 protocol == htons(ETH_P_FCOE)));
1697 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1699 int features = dev->features;
1701 if (vlan_tx_tag_present(skb))
1702 features &= dev->vlan_features;
1704 if (can_checksum_protocol(features, skb->protocol))
1705 return true;
1707 if (skb->protocol == htons(ETH_P_8021Q)) {
1708 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1709 if (can_checksum_protocol(dev->features & dev->vlan_features,
1710 veh->h_vlan_encapsulated_proto))
1711 return true;
1714 return false;
1718 * skb_dev_set -- assign a new device to a buffer
1719 * @skb: buffer for the new device
1720 * @dev: network device
1722 * If an skb is owned by a device already, we have to reset
1723 * all data private to the namespace a device belongs to
1724 * before assigning it a new device.
1726 #ifdef CONFIG_NET_NS
1727 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1729 skb_dst_drop(skb);
1730 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1731 secpath_reset(skb);
1732 nf_reset(skb);
1733 skb_init_secmark(skb);
1734 skb->mark = 0;
1735 skb->priority = 0;
1736 skb->nf_trace = 0;
1737 skb->ipvs_property = 0;
1738 #ifdef CONFIG_NET_SCHED
1739 skb->tc_index = 0;
1740 #endif
1742 skb->dev = dev;
1744 EXPORT_SYMBOL(skb_set_dev);
1745 #endif /* CONFIG_NET_NS */
1748 * Invalidate hardware checksum when packet is to be mangled, and
1749 * complete checksum manually on outgoing path.
1751 int skb_checksum_help(struct sk_buff *skb)
1753 __wsum csum;
1754 int ret = 0, offset;
1756 if (skb->ip_summed == CHECKSUM_COMPLETE)
1757 goto out_set_summed;
1759 if (unlikely(skb_shinfo(skb)->gso_size)) {
1760 /* Let GSO fix up the checksum. */
1761 goto out_set_summed;
1764 offset = skb->csum_start - skb_headroom(skb);
1765 BUG_ON(offset >= skb_headlen(skb));
1766 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1768 offset += skb->csum_offset;
1769 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1771 if (skb_cloned(skb) &&
1772 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1773 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1774 if (ret)
1775 goto out;
1778 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1779 out_set_summed:
1780 skb->ip_summed = CHECKSUM_NONE;
1781 out:
1782 return ret;
1784 EXPORT_SYMBOL(skb_checksum_help);
1787 * skb_gso_segment - Perform segmentation on skb.
1788 * @skb: buffer to segment
1789 * @features: features for the output path (see dev->features)
1791 * This function segments the given skb and returns a list of segments.
1793 * It may return NULL if the skb requires no segmentation. This is
1794 * only possible when GSO is used for verifying header integrity.
1796 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1798 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1799 struct packet_type *ptype;
1800 __be16 type = skb->protocol;
1801 int err;
1803 if (type == htons(ETH_P_8021Q)) {
1804 struct vlan_ethhdr *veh;
1806 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1807 return ERR_PTR(-EINVAL);
1809 veh = (struct vlan_ethhdr *)skb->data;
1810 type = veh->h_vlan_encapsulated_proto;
1813 skb_reset_mac_header(skb);
1814 skb->mac_len = skb->network_header - skb->mac_header;
1815 __skb_pull(skb, skb->mac_len);
1817 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1818 struct net_device *dev = skb->dev;
1819 struct ethtool_drvinfo info = {};
1821 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1822 dev->ethtool_ops->get_drvinfo(dev, &info);
1824 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1825 "ip_summed=%d",
1826 info.driver, dev ? dev->features : 0L,
1827 skb->sk ? skb->sk->sk_route_caps : 0L,
1828 skb->len, skb->data_len, skb->ip_summed);
1830 if (skb_header_cloned(skb) &&
1831 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1832 return ERR_PTR(err);
1835 rcu_read_lock();
1836 list_for_each_entry_rcu(ptype,
1837 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1838 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1839 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1840 err = ptype->gso_send_check(skb);
1841 segs = ERR_PTR(err);
1842 if (err || skb_gso_ok(skb, features))
1843 break;
1844 __skb_push(skb, (skb->data -
1845 skb_network_header(skb)));
1847 segs = ptype->gso_segment(skb, features);
1848 break;
1851 rcu_read_unlock();
1853 __skb_push(skb, skb->data - skb_mac_header(skb));
1855 return segs;
1857 EXPORT_SYMBOL(skb_gso_segment);
1859 /* Take action when hardware reception checksum errors are detected. */
1860 #ifdef CONFIG_BUG
1861 void netdev_rx_csum_fault(struct net_device *dev)
1863 if (net_ratelimit()) {
1864 printk(KERN_ERR "%s: hw csum failure.\n",
1865 dev ? dev->name : "<unknown>");
1866 dump_stack();
1869 EXPORT_SYMBOL(netdev_rx_csum_fault);
1870 #endif
1872 /* Actually, we should eliminate this check as soon as we know, that:
1873 * 1. IOMMU is present and allows to map all the memory.
1874 * 2. No high memory really exists on this machine.
1877 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1879 #ifdef CONFIG_HIGHMEM
1880 int i;
1881 if (!(dev->features & NETIF_F_HIGHDMA)) {
1882 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1883 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1884 return 1;
1887 if (PCI_DMA_BUS_IS_PHYS) {
1888 struct device *pdev = dev->dev.parent;
1890 if (!pdev)
1891 return 0;
1892 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1893 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1894 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1895 return 1;
1898 #endif
1899 return 0;
1902 struct dev_gso_cb {
1903 void (*destructor)(struct sk_buff *skb);
1906 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1908 static void dev_gso_skb_destructor(struct sk_buff *skb)
1910 struct dev_gso_cb *cb;
1912 do {
1913 struct sk_buff *nskb = skb->next;
1915 skb->next = nskb->next;
1916 nskb->next = NULL;
1917 kfree_skb(nskb);
1918 } while (skb->next);
1920 cb = DEV_GSO_CB(skb);
1921 if (cb->destructor)
1922 cb->destructor(skb);
1926 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1927 * @skb: buffer to segment
1929 * This function segments the given skb and stores the list of segments
1930 * in skb->next.
1932 static int dev_gso_segment(struct sk_buff *skb)
1934 struct net_device *dev = skb->dev;
1935 struct sk_buff *segs;
1936 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1937 NETIF_F_SG : 0);
1939 segs = skb_gso_segment(skb, features);
1941 /* Verifying header integrity only. */
1942 if (!segs)
1943 return 0;
1945 if (IS_ERR(segs))
1946 return PTR_ERR(segs);
1948 skb->next = segs;
1949 DEV_GSO_CB(skb)->destructor = skb->destructor;
1950 skb->destructor = dev_gso_skb_destructor;
1952 return 0;
1956 * Try to orphan skb early, right before transmission by the device.
1957 * We cannot orphan skb if tx timestamp is requested or the sk-reference
1958 * is needed on driver level for other reasons, e.g. see net/can/raw.c
1960 static inline void skb_orphan_try(struct sk_buff *skb)
1962 struct sock *sk = skb->sk;
1964 if (sk && !skb_shinfo(skb)->tx_flags) {
1965 /* skb_tx_hash() wont be able to get sk.
1966 * We copy sk_hash into skb->rxhash
1968 if (!skb->rxhash)
1969 skb->rxhash = sk->sk_hash;
1970 skb_orphan(skb);
1975 * Returns true if either:
1976 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1977 * 2. skb is fragmented and the device does not support SG, or if
1978 * at least one of fragments is in highmem and device does not
1979 * support DMA from it.
1981 static inline int skb_needs_linearize(struct sk_buff *skb,
1982 struct net_device *dev)
1984 int features = dev->features;
1986 if (skb->protocol == htons(ETH_P_8021Q) || vlan_tx_tag_present(skb))
1987 features &= dev->vlan_features;
1989 return skb_is_nonlinear(skb) &&
1990 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
1991 (skb_shinfo(skb)->nr_frags && (!(features & NETIF_F_SG) ||
1992 illegal_highdma(dev, skb))));
1995 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1996 struct netdev_queue *txq)
1998 const struct net_device_ops *ops = dev->netdev_ops;
1999 int rc = NETDEV_TX_OK;
2001 if (likely(!skb->next)) {
2002 if (!list_empty(&ptype_all))
2003 dev_queue_xmit_nit(skb, dev);
2006 * If device doesnt need skb->dst, release it right now while
2007 * its hot in this cpu cache
2009 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2010 skb_dst_drop(skb);
2012 skb_orphan_try(skb);
2014 if (vlan_tx_tag_present(skb) &&
2015 !(dev->features & NETIF_F_HW_VLAN_TX)) {
2016 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2017 if (unlikely(!skb))
2018 goto out;
2020 skb->vlan_tci = 0;
2023 if (netif_needs_gso(dev, skb)) {
2024 if (unlikely(dev_gso_segment(skb)))
2025 goto out_kfree_skb;
2026 if (skb->next)
2027 goto gso;
2028 } else {
2029 if (skb_needs_linearize(skb, dev) &&
2030 __skb_linearize(skb))
2031 goto out_kfree_skb;
2033 /* If packet is not checksummed and device does not
2034 * support checksumming for this protocol, complete
2035 * checksumming here.
2037 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2038 skb_set_transport_header(skb, skb->csum_start -
2039 skb_headroom(skb));
2040 if (!dev_can_checksum(dev, skb) &&
2041 skb_checksum_help(skb))
2042 goto out_kfree_skb;
2046 rc = ops->ndo_start_xmit(skb, dev);
2047 trace_net_dev_xmit(skb, rc);
2048 if (rc == NETDEV_TX_OK)
2049 txq_trans_update(txq);
2050 return rc;
2053 gso:
2054 do {
2055 struct sk_buff *nskb = skb->next;
2057 skb->next = nskb->next;
2058 nskb->next = NULL;
2061 * If device doesnt need nskb->dst, release it right now while
2062 * its hot in this cpu cache
2064 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2065 skb_dst_drop(nskb);
2067 rc = ops->ndo_start_xmit(nskb, dev);
2068 trace_net_dev_xmit(nskb, rc);
2069 if (unlikely(rc != NETDEV_TX_OK)) {
2070 if (rc & ~NETDEV_TX_MASK)
2071 goto out_kfree_gso_skb;
2072 nskb->next = skb->next;
2073 skb->next = nskb;
2074 return rc;
2076 txq_trans_update(txq);
2077 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2078 return NETDEV_TX_BUSY;
2079 } while (skb->next);
2081 out_kfree_gso_skb:
2082 if (likely(skb->next == NULL))
2083 skb->destructor = DEV_GSO_CB(skb)->destructor;
2084 out_kfree_skb:
2085 kfree_skb(skb);
2086 out:
2087 return rc;
2090 static u32 hashrnd __read_mostly;
2092 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
2094 u32 hash;
2096 if (skb_rx_queue_recorded(skb)) {
2097 hash = skb_get_rx_queue(skb);
2098 while (unlikely(hash >= dev->real_num_tx_queues))
2099 hash -= dev->real_num_tx_queues;
2100 return hash;
2103 if (skb->sk && skb->sk->sk_hash)
2104 hash = skb->sk->sk_hash;
2105 else
2106 hash = (__force u16) skb->protocol ^ skb->rxhash;
2107 hash = jhash_1word(hash, hashrnd);
2109 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
2111 EXPORT_SYMBOL(skb_tx_hash);
2113 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2115 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2116 if (net_ratelimit()) {
2117 pr_warning("%s selects TX queue %d, but "
2118 "real number of TX queues is %d\n",
2119 dev->name, queue_index, dev->real_num_tx_queues);
2121 return 0;
2123 return queue_index;
2126 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2127 struct sk_buff *skb)
2129 int queue_index;
2130 const struct net_device_ops *ops = dev->netdev_ops;
2132 if (ops->ndo_select_queue) {
2133 queue_index = ops->ndo_select_queue(dev, skb);
2134 queue_index = dev_cap_txqueue(dev, queue_index);
2135 } else {
2136 struct sock *sk = skb->sk;
2137 queue_index = sk_tx_queue_get(sk);
2138 if (queue_index < 0) {
2140 queue_index = 0;
2141 if (dev->real_num_tx_queues > 1)
2142 queue_index = skb_tx_hash(dev, skb);
2144 if (sk) {
2145 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2147 if (dst && skb_dst(skb) == dst)
2148 sk_tx_queue_set(sk, queue_index);
2153 skb_set_queue_mapping(skb, queue_index);
2154 return netdev_get_tx_queue(dev, queue_index);
2157 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2158 struct net_device *dev,
2159 struct netdev_queue *txq)
2161 spinlock_t *root_lock = qdisc_lock(q);
2162 bool contended = qdisc_is_running(q);
2163 int rc;
2166 * Heuristic to force contended enqueues to serialize on a
2167 * separate lock before trying to get qdisc main lock.
2168 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2169 * and dequeue packets faster.
2171 if (unlikely(contended))
2172 spin_lock(&q->busylock);
2174 spin_lock(root_lock);
2175 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2176 kfree_skb(skb);
2177 rc = NET_XMIT_DROP;
2178 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2179 qdisc_run_begin(q)) {
2181 * This is a work-conserving queue; there are no old skbs
2182 * waiting to be sent out; and the qdisc is not running -
2183 * xmit the skb directly.
2185 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2186 skb_dst_force(skb);
2187 __qdisc_update_bstats(q, skb->len);
2188 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2189 if (unlikely(contended)) {
2190 spin_unlock(&q->busylock);
2191 contended = false;
2193 __qdisc_run(q);
2194 } else
2195 qdisc_run_end(q);
2197 rc = NET_XMIT_SUCCESS;
2198 } else {
2199 skb_dst_force(skb);
2200 rc = qdisc_enqueue_root(skb, q);
2201 if (qdisc_run_begin(q)) {
2202 if (unlikely(contended)) {
2203 spin_unlock(&q->busylock);
2204 contended = false;
2206 __qdisc_run(q);
2209 spin_unlock(root_lock);
2210 if (unlikely(contended))
2211 spin_unlock(&q->busylock);
2212 return rc;
2215 static DEFINE_PER_CPU(int, xmit_recursion);
2216 #define RECURSION_LIMIT 3
2219 * dev_queue_xmit - transmit a buffer
2220 * @skb: buffer to transmit
2222 * Queue a buffer for transmission to a network device. The caller must
2223 * have set the device and priority and built the buffer before calling
2224 * this function. The function can be called from an interrupt.
2226 * A negative errno code is returned on a failure. A success does not
2227 * guarantee the frame will be transmitted as it may be dropped due
2228 * to congestion or traffic shaping.
2230 * -----------------------------------------------------------------------------------
2231 * I notice this method can also return errors from the queue disciplines,
2232 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2233 * be positive.
2235 * Regardless of the return value, the skb is consumed, so it is currently
2236 * difficult to retry a send to this method. (You can bump the ref count
2237 * before sending to hold a reference for retry if you are careful.)
2239 * When calling this method, interrupts MUST be enabled. This is because
2240 * the BH enable code must have IRQs enabled so that it will not deadlock.
2241 * --BLG
2243 int dev_queue_xmit(struct sk_buff *skb)
2245 struct net_device *dev = skb->dev;
2246 struct netdev_queue *txq;
2247 struct Qdisc *q;
2248 int rc = -ENOMEM;
2250 /* Disable soft irqs for various locks below. Also
2251 * stops preemption for RCU.
2253 rcu_read_lock_bh();
2255 txq = dev_pick_tx(dev, skb);
2256 q = rcu_dereference_bh(txq->qdisc);
2258 #ifdef CONFIG_NET_CLS_ACT
2259 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2260 #endif
2261 trace_net_dev_queue(skb);
2262 if (q->enqueue) {
2263 rc = __dev_xmit_skb(skb, q, dev, txq);
2264 goto out;
2267 /* The device has no queue. Common case for software devices:
2268 loopback, all the sorts of tunnels...
2270 Really, it is unlikely that netif_tx_lock protection is necessary
2271 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2272 counters.)
2273 However, it is possible, that they rely on protection
2274 made by us here.
2276 Check this and shot the lock. It is not prone from deadlocks.
2277 Either shot noqueue qdisc, it is even simpler 8)
2279 if (dev->flags & IFF_UP) {
2280 int cpu = smp_processor_id(); /* ok because BHs are off */
2282 if (txq->xmit_lock_owner != cpu) {
2284 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2285 goto recursion_alert;
2287 HARD_TX_LOCK(dev, txq, cpu);
2289 if (!netif_tx_queue_stopped(txq)) {
2290 __this_cpu_inc(xmit_recursion);
2291 rc = dev_hard_start_xmit(skb, dev, txq);
2292 __this_cpu_dec(xmit_recursion);
2293 if (dev_xmit_complete(rc)) {
2294 HARD_TX_UNLOCK(dev, txq);
2295 goto out;
2298 HARD_TX_UNLOCK(dev, txq);
2299 if (net_ratelimit())
2300 printk(KERN_CRIT "Virtual device %s asks to "
2301 "queue packet!\n", dev->name);
2302 } else {
2303 /* Recursion is detected! It is possible,
2304 * unfortunately
2306 recursion_alert:
2307 if (net_ratelimit())
2308 printk(KERN_CRIT "Dead loop on virtual device "
2309 "%s, fix it urgently!\n", dev->name);
2313 rc = -ENETDOWN;
2314 rcu_read_unlock_bh();
2316 kfree_skb(skb);
2317 return rc;
2318 out:
2319 rcu_read_unlock_bh();
2320 return rc;
2322 EXPORT_SYMBOL(dev_queue_xmit);
2325 /*=======================================================================
2326 Receiver routines
2327 =======================================================================*/
2329 int netdev_max_backlog __read_mostly = 1000;
2330 int netdev_tstamp_prequeue __read_mostly = 1;
2331 int netdev_budget __read_mostly = 300;
2332 int weight_p __read_mostly = 64; /* old backlog weight */
2334 /* Called with irq disabled */
2335 static inline void ____napi_schedule(struct softnet_data *sd,
2336 struct napi_struct *napi)
2338 list_add_tail(&napi->poll_list, &sd->poll_list);
2339 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2343 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2344 * and src/dst port numbers. Returns a non-zero hash number on success
2345 * and 0 on failure.
2347 __u32 __skb_get_rxhash(struct sk_buff *skb)
2349 int nhoff, hash = 0, poff;
2350 struct ipv6hdr *ip6;
2351 struct iphdr *ip;
2352 u8 ip_proto;
2353 u32 addr1, addr2, ihl;
2354 union {
2355 u32 v32;
2356 u16 v16[2];
2357 } ports;
2359 nhoff = skb_network_offset(skb);
2361 switch (skb->protocol) {
2362 case __constant_htons(ETH_P_IP):
2363 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2364 goto done;
2366 ip = (struct iphdr *) (skb->data + nhoff);
2367 if (ip->frag_off & htons(IP_MF | IP_OFFSET))
2368 ip_proto = 0;
2369 else
2370 ip_proto = ip->protocol;
2371 addr1 = (__force u32) ip->saddr;
2372 addr2 = (__force u32) ip->daddr;
2373 ihl = ip->ihl;
2374 break;
2375 case __constant_htons(ETH_P_IPV6):
2376 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2377 goto done;
2379 ip6 = (struct ipv6hdr *) (skb->data + nhoff);
2380 ip_proto = ip6->nexthdr;
2381 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2382 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2383 ihl = (40 >> 2);
2384 break;
2385 default:
2386 goto done;
2389 ports.v32 = 0;
2390 poff = proto_ports_offset(ip_proto);
2391 if (poff >= 0) {
2392 nhoff += ihl * 4 + poff;
2393 if (pskb_may_pull(skb, nhoff + 4)) {
2394 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2395 if (ports.v16[1] < ports.v16[0])
2396 swap(ports.v16[0], ports.v16[1]);
2400 /* get a consistent hash (same value on both flow directions) */
2401 if (addr2 < addr1)
2402 swap(addr1, addr2);
2404 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2405 if (!hash)
2406 hash = 1;
2408 done:
2409 return hash;
2411 EXPORT_SYMBOL(__skb_get_rxhash);
2413 #ifdef CONFIG_RPS
2415 /* One global table that all flow-based protocols share. */
2416 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2417 EXPORT_SYMBOL(rps_sock_flow_table);
2420 * get_rps_cpu is called from netif_receive_skb and returns the target
2421 * CPU from the RPS map of the receiving queue for a given skb.
2422 * rcu_read_lock must be held on entry.
2424 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2425 struct rps_dev_flow **rflowp)
2427 struct netdev_rx_queue *rxqueue;
2428 struct rps_map *map = NULL;
2429 struct rps_dev_flow_table *flow_table;
2430 struct rps_sock_flow_table *sock_flow_table;
2431 int cpu = -1;
2432 u16 tcpu;
2434 if (skb_rx_queue_recorded(skb)) {
2435 u16 index = skb_get_rx_queue(skb);
2436 if (unlikely(index >= dev->real_num_rx_queues)) {
2437 WARN_ONCE(dev->real_num_rx_queues > 1,
2438 "%s received packet on queue %u, but number "
2439 "of RX queues is %u\n",
2440 dev->name, index, dev->real_num_rx_queues);
2441 goto done;
2443 rxqueue = dev->_rx + index;
2444 } else
2445 rxqueue = dev->_rx;
2447 if (rxqueue->rps_map) {
2448 map = rcu_dereference(rxqueue->rps_map);
2449 if (map && map->len == 1) {
2450 tcpu = map->cpus[0];
2451 if (cpu_online(tcpu))
2452 cpu = tcpu;
2453 goto done;
2455 } else if (!rxqueue->rps_flow_table) {
2456 goto done;
2459 skb_reset_network_header(skb);
2460 if (!skb_get_rxhash(skb))
2461 goto done;
2463 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2464 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2465 if (flow_table && sock_flow_table) {
2466 u16 next_cpu;
2467 struct rps_dev_flow *rflow;
2469 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2470 tcpu = rflow->cpu;
2472 next_cpu = sock_flow_table->ents[skb->rxhash &
2473 sock_flow_table->mask];
2476 * If the desired CPU (where last recvmsg was done) is
2477 * different from current CPU (one in the rx-queue flow
2478 * table entry), switch if one of the following holds:
2479 * - Current CPU is unset (equal to RPS_NO_CPU).
2480 * - Current CPU is offline.
2481 * - The current CPU's queue tail has advanced beyond the
2482 * last packet that was enqueued using this table entry.
2483 * This guarantees that all previous packets for the flow
2484 * have been dequeued, thus preserving in order delivery.
2486 if (unlikely(tcpu != next_cpu) &&
2487 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2488 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2489 rflow->last_qtail)) >= 0)) {
2490 tcpu = rflow->cpu = next_cpu;
2491 if (tcpu != RPS_NO_CPU)
2492 rflow->last_qtail = per_cpu(softnet_data,
2493 tcpu).input_queue_head;
2495 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2496 *rflowp = rflow;
2497 cpu = tcpu;
2498 goto done;
2502 if (map) {
2503 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2505 if (cpu_online(tcpu)) {
2506 cpu = tcpu;
2507 goto done;
2511 done:
2512 return cpu;
2515 /* Called from hardirq (IPI) context */
2516 static void rps_trigger_softirq(void *data)
2518 struct softnet_data *sd = data;
2520 ____napi_schedule(sd, &sd->backlog);
2521 sd->received_rps++;
2524 #endif /* CONFIG_RPS */
2527 * Check if this softnet_data structure is another cpu one
2528 * If yes, queue it to our IPI list and return 1
2529 * If no, return 0
2531 static int rps_ipi_queued(struct softnet_data *sd)
2533 #ifdef CONFIG_RPS
2534 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2536 if (sd != mysd) {
2537 sd->rps_ipi_next = mysd->rps_ipi_list;
2538 mysd->rps_ipi_list = sd;
2540 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2541 return 1;
2543 #endif /* CONFIG_RPS */
2544 return 0;
2548 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2549 * queue (may be a remote CPU queue).
2551 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2552 unsigned int *qtail)
2554 struct softnet_data *sd;
2555 unsigned long flags;
2557 sd = &per_cpu(softnet_data, cpu);
2559 local_irq_save(flags);
2561 rps_lock(sd);
2562 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2563 if (skb_queue_len(&sd->input_pkt_queue)) {
2564 enqueue:
2565 __skb_queue_tail(&sd->input_pkt_queue, skb);
2566 input_queue_tail_incr_save(sd, qtail);
2567 rps_unlock(sd);
2568 local_irq_restore(flags);
2569 return NET_RX_SUCCESS;
2572 /* Schedule NAPI for backlog device
2573 * We can use non atomic operation since we own the queue lock
2575 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2576 if (!rps_ipi_queued(sd))
2577 ____napi_schedule(sd, &sd->backlog);
2579 goto enqueue;
2582 sd->dropped++;
2583 rps_unlock(sd);
2585 local_irq_restore(flags);
2587 atomic_long_inc(&skb->dev->rx_dropped);
2588 kfree_skb(skb);
2589 return NET_RX_DROP;
2593 * netif_rx - post buffer to the network code
2594 * @skb: buffer to post
2596 * This function receives a packet from a device driver and queues it for
2597 * the upper (protocol) levels to process. It always succeeds. The buffer
2598 * may be dropped during processing for congestion control or by the
2599 * protocol layers.
2601 * return values:
2602 * NET_RX_SUCCESS (no congestion)
2603 * NET_RX_DROP (packet was dropped)
2607 int netif_rx(struct sk_buff *skb)
2609 int ret;
2611 /* if netpoll wants it, pretend we never saw it */
2612 if (netpoll_rx(skb))
2613 return NET_RX_DROP;
2615 if (netdev_tstamp_prequeue)
2616 net_timestamp_check(skb);
2618 trace_netif_rx(skb);
2619 #ifdef CONFIG_RPS
2621 struct rps_dev_flow voidflow, *rflow = &voidflow;
2622 int cpu;
2624 preempt_disable();
2625 rcu_read_lock();
2627 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2628 if (cpu < 0)
2629 cpu = smp_processor_id();
2631 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2633 rcu_read_unlock();
2634 preempt_enable();
2636 #else
2638 unsigned int qtail;
2639 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2640 put_cpu();
2642 #endif
2643 return ret;
2645 EXPORT_SYMBOL(netif_rx);
2647 int netif_rx_ni(struct sk_buff *skb)
2649 int err;
2651 preempt_disable();
2652 err = netif_rx(skb);
2653 if (local_softirq_pending())
2654 do_softirq();
2655 preempt_enable();
2657 return err;
2659 EXPORT_SYMBOL(netif_rx_ni);
2661 static void net_tx_action(struct softirq_action *h)
2663 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2665 if (sd->completion_queue) {
2666 struct sk_buff *clist;
2668 local_irq_disable();
2669 clist = sd->completion_queue;
2670 sd->completion_queue = NULL;
2671 local_irq_enable();
2673 while (clist) {
2674 struct sk_buff *skb = clist;
2675 clist = clist->next;
2677 WARN_ON(atomic_read(&skb->users));
2678 trace_kfree_skb(skb, net_tx_action);
2679 __kfree_skb(skb);
2683 if (sd->output_queue) {
2684 struct Qdisc *head;
2686 local_irq_disable();
2687 head = sd->output_queue;
2688 sd->output_queue = NULL;
2689 sd->output_queue_tailp = &sd->output_queue;
2690 local_irq_enable();
2692 while (head) {
2693 struct Qdisc *q = head;
2694 spinlock_t *root_lock;
2696 head = head->next_sched;
2698 root_lock = qdisc_lock(q);
2699 if (spin_trylock(root_lock)) {
2700 smp_mb__before_clear_bit();
2701 clear_bit(__QDISC_STATE_SCHED,
2702 &q->state);
2703 qdisc_run(q);
2704 spin_unlock(root_lock);
2705 } else {
2706 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2707 &q->state)) {
2708 __netif_reschedule(q);
2709 } else {
2710 smp_mb__before_clear_bit();
2711 clear_bit(__QDISC_STATE_SCHED,
2712 &q->state);
2719 static inline int deliver_skb(struct sk_buff *skb,
2720 struct packet_type *pt_prev,
2721 struct net_device *orig_dev)
2723 atomic_inc(&skb->users);
2724 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2727 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2728 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2729 /* This hook is defined here for ATM LANE */
2730 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2731 unsigned char *addr) __read_mostly;
2732 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2733 #endif
2735 #ifdef CONFIG_NET_CLS_ACT
2736 /* TODO: Maybe we should just force sch_ingress to be compiled in
2737 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2738 * a compare and 2 stores extra right now if we dont have it on
2739 * but have CONFIG_NET_CLS_ACT
2740 * NOTE: This doesnt stop any functionality; if you dont have
2741 * the ingress scheduler, you just cant add policies on ingress.
2744 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
2746 struct net_device *dev = skb->dev;
2747 u32 ttl = G_TC_RTTL(skb->tc_verd);
2748 int result = TC_ACT_OK;
2749 struct Qdisc *q;
2751 if (unlikely(MAX_RED_LOOP < ttl++)) {
2752 if (net_ratelimit())
2753 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
2754 skb->skb_iif, dev->ifindex);
2755 return TC_ACT_SHOT;
2758 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2759 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2761 q = rxq->qdisc;
2762 if (q != &noop_qdisc) {
2763 spin_lock(qdisc_lock(q));
2764 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2765 result = qdisc_enqueue_root(skb, q);
2766 spin_unlock(qdisc_lock(q));
2769 return result;
2772 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2773 struct packet_type **pt_prev,
2774 int *ret, struct net_device *orig_dev)
2776 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
2778 if (!rxq || rxq->qdisc == &noop_qdisc)
2779 goto out;
2781 if (*pt_prev) {
2782 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2783 *pt_prev = NULL;
2786 switch (ing_filter(skb, rxq)) {
2787 case TC_ACT_SHOT:
2788 case TC_ACT_STOLEN:
2789 kfree_skb(skb);
2790 return NULL;
2793 out:
2794 skb->tc_verd = 0;
2795 return skb;
2797 #endif
2800 * netdev_rx_handler_register - register receive handler
2801 * @dev: device to register a handler for
2802 * @rx_handler: receive handler to register
2803 * @rx_handler_data: data pointer that is used by rx handler
2805 * Register a receive hander for a device. This handler will then be
2806 * called from __netif_receive_skb. A negative errno code is returned
2807 * on a failure.
2809 * The caller must hold the rtnl_mutex.
2811 int netdev_rx_handler_register(struct net_device *dev,
2812 rx_handler_func_t *rx_handler,
2813 void *rx_handler_data)
2815 ASSERT_RTNL();
2817 if (dev->rx_handler)
2818 return -EBUSY;
2820 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
2821 rcu_assign_pointer(dev->rx_handler, rx_handler);
2823 return 0;
2825 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
2828 * netdev_rx_handler_unregister - unregister receive handler
2829 * @dev: device to unregister a handler from
2831 * Unregister a receive hander from a device.
2833 * The caller must hold the rtnl_mutex.
2835 void netdev_rx_handler_unregister(struct net_device *dev)
2838 ASSERT_RTNL();
2839 rcu_assign_pointer(dev->rx_handler, NULL);
2840 rcu_assign_pointer(dev->rx_handler_data, NULL);
2842 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
2844 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2845 struct net_device *master)
2847 if (skb->pkt_type == PACKET_HOST) {
2848 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2850 memcpy(dest, master->dev_addr, ETH_ALEN);
2854 /* On bonding slaves other than the currently active slave, suppress
2855 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2856 * ARP on active-backup slaves with arp_validate enabled.
2858 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2860 struct net_device *dev = skb->dev;
2862 if (master->priv_flags & IFF_MASTER_ARPMON)
2863 dev->last_rx = jiffies;
2865 if ((master->priv_flags & IFF_MASTER_ALB) &&
2866 (master->priv_flags & IFF_BRIDGE_PORT)) {
2867 /* Do address unmangle. The local destination address
2868 * will be always the one master has. Provides the right
2869 * functionality in a bridge.
2871 skb_bond_set_mac_by_master(skb, master);
2874 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2875 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2876 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2877 return 0;
2879 if (master->priv_flags & IFF_MASTER_ALB) {
2880 if (skb->pkt_type != PACKET_BROADCAST &&
2881 skb->pkt_type != PACKET_MULTICAST)
2882 return 0;
2884 if (master->priv_flags & IFF_MASTER_8023AD &&
2885 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2886 return 0;
2888 return 1;
2890 return 0;
2892 EXPORT_SYMBOL(__skb_bond_should_drop);
2894 static int __netif_receive_skb(struct sk_buff *skb)
2896 struct packet_type *ptype, *pt_prev;
2897 rx_handler_func_t *rx_handler;
2898 struct net_device *orig_dev;
2899 struct net_device *master;
2900 struct net_device *null_or_orig;
2901 struct net_device *orig_or_bond;
2902 int ret = NET_RX_DROP;
2903 __be16 type;
2905 if (!netdev_tstamp_prequeue)
2906 net_timestamp_check(skb);
2908 trace_netif_receive_skb(skb);
2910 /* if we've gotten here through NAPI, check netpoll */
2911 if (netpoll_receive_skb(skb))
2912 return NET_RX_DROP;
2914 if (!skb->skb_iif)
2915 skb->skb_iif = skb->dev->ifindex;
2918 * bonding note: skbs received on inactive slaves should only
2919 * be delivered to pkt handlers that are exact matches. Also
2920 * the deliver_no_wcard flag will be set. If packet handlers
2921 * are sensitive to duplicate packets these skbs will need to
2922 * be dropped at the handler.
2924 null_or_orig = NULL;
2925 orig_dev = skb->dev;
2926 master = ACCESS_ONCE(orig_dev->master);
2927 if (skb->deliver_no_wcard)
2928 null_or_orig = orig_dev;
2929 else if (master) {
2930 if (skb_bond_should_drop(skb, master)) {
2931 skb->deliver_no_wcard = 1;
2932 null_or_orig = orig_dev; /* deliver only exact match */
2933 } else
2934 skb->dev = master;
2937 __this_cpu_inc(softnet_data.processed);
2938 skb_reset_network_header(skb);
2939 skb_reset_transport_header(skb);
2940 skb->mac_len = skb->network_header - skb->mac_header;
2942 pt_prev = NULL;
2944 rcu_read_lock();
2946 #ifdef CONFIG_NET_CLS_ACT
2947 if (skb->tc_verd & TC_NCLS) {
2948 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2949 goto ncls;
2951 #endif
2953 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2954 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2955 ptype->dev == orig_dev) {
2956 if (pt_prev)
2957 ret = deliver_skb(skb, pt_prev, orig_dev);
2958 pt_prev = ptype;
2962 #ifdef CONFIG_NET_CLS_ACT
2963 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2964 if (!skb)
2965 goto out;
2966 ncls:
2967 #endif
2969 /* Handle special case of bridge or macvlan */
2970 rx_handler = rcu_dereference(skb->dev->rx_handler);
2971 if (rx_handler) {
2972 if (pt_prev) {
2973 ret = deliver_skb(skb, pt_prev, orig_dev);
2974 pt_prev = NULL;
2976 skb = rx_handler(skb);
2977 if (!skb)
2978 goto out;
2981 if (vlan_tx_tag_present(skb)) {
2982 if (pt_prev) {
2983 ret = deliver_skb(skb, pt_prev, orig_dev);
2984 pt_prev = NULL;
2986 if (vlan_hwaccel_do_receive(&skb)) {
2987 ret = __netif_receive_skb(skb);
2988 goto out;
2989 } else if (unlikely(!skb))
2990 goto out;
2994 * Make sure frames received on VLAN interfaces stacked on
2995 * bonding interfaces still make their way to any base bonding
2996 * device that may have registered for a specific ptype. The
2997 * handler may have to adjust skb->dev and orig_dev.
2999 orig_or_bond = orig_dev;
3000 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
3001 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
3002 orig_or_bond = vlan_dev_real_dev(skb->dev);
3005 type = skb->protocol;
3006 list_for_each_entry_rcu(ptype,
3007 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3008 if (ptype->type == type && (ptype->dev == null_or_orig ||
3009 ptype->dev == skb->dev || ptype->dev == orig_dev ||
3010 ptype->dev == orig_or_bond)) {
3011 if (pt_prev)
3012 ret = deliver_skb(skb, pt_prev, orig_dev);
3013 pt_prev = ptype;
3017 if (pt_prev) {
3018 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3019 } else {
3020 atomic_long_inc(&skb->dev->rx_dropped);
3021 kfree_skb(skb);
3022 /* Jamal, now you will not able to escape explaining
3023 * me how you were going to use this. :-)
3025 ret = NET_RX_DROP;
3028 out:
3029 rcu_read_unlock();
3030 return ret;
3034 * netif_receive_skb - process receive buffer from network
3035 * @skb: buffer to process
3037 * netif_receive_skb() is the main receive data processing function.
3038 * It always succeeds. The buffer may be dropped during processing
3039 * for congestion control or by the protocol layers.
3041 * This function may only be called from softirq context and interrupts
3042 * should be enabled.
3044 * Return values (usually ignored):
3045 * NET_RX_SUCCESS: no congestion
3046 * NET_RX_DROP: packet was dropped
3048 int netif_receive_skb(struct sk_buff *skb)
3050 if (netdev_tstamp_prequeue)
3051 net_timestamp_check(skb);
3053 if (skb_defer_rx_timestamp(skb))
3054 return NET_RX_SUCCESS;
3056 #ifdef CONFIG_RPS
3058 struct rps_dev_flow voidflow, *rflow = &voidflow;
3059 int cpu, ret;
3061 rcu_read_lock();
3063 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3065 if (cpu >= 0) {
3066 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3067 rcu_read_unlock();
3068 } else {
3069 rcu_read_unlock();
3070 ret = __netif_receive_skb(skb);
3073 return ret;
3075 #else
3076 return __netif_receive_skb(skb);
3077 #endif
3079 EXPORT_SYMBOL(netif_receive_skb);
3081 /* Network device is going away, flush any packets still pending
3082 * Called with irqs disabled.
3084 static void flush_backlog(void *arg)
3086 struct net_device *dev = arg;
3087 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3088 struct sk_buff *skb, *tmp;
3090 rps_lock(sd);
3091 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3092 if (skb->dev == dev) {
3093 __skb_unlink(skb, &sd->input_pkt_queue);
3094 kfree_skb(skb);
3095 input_queue_head_incr(sd);
3098 rps_unlock(sd);
3100 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3101 if (skb->dev == dev) {
3102 __skb_unlink(skb, &sd->process_queue);
3103 kfree_skb(skb);
3104 input_queue_head_incr(sd);
3109 static int napi_gro_complete(struct sk_buff *skb)
3111 struct packet_type *ptype;
3112 __be16 type = skb->protocol;
3113 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3114 int err = -ENOENT;
3116 if (NAPI_GRO_CB(skb)->count == 1) {
3117 skb_shinfo(skb)->gso_size = 0;
3118 goto out;
3121 rcu_read_lock();
3122 list_for_each_entry_rcu(ptype, head, list) {
3123 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3124 continue;
3126 err = ptype->gro_complete(skb);
3127 break;
3129 rcu_read_unlock();
3131 if (err) {
3132 WARN_ON(&ptype->list == head);
3133 kfree_skb(skb);
3134 return NET_RX_SUCCESS;
3137 out:
3138 return netif_receive_skb(skb);
3141 inline void napi_gro_flush(struct napi_struct *napi)
3143 struct sk_buff *skb, *next;
3145 for (skb = napi->gro_list; skb; skb = next) {
3146 next = skb->next;
3147 skb->next = NULL;
3148 napi_gro_complete(skb);
3151 napi->gro_count = 0;
3152 napi->gro_list = NULL;
3154 EXPORT_SYMBOL(napi_gro_flush);
3156 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3158 struct sk_buff **pp = NULL;
3159 struct packet_type *ptype;
3160 __be16 type = skb->protocol;
3161 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3162 int same_flow;
3163 int mac_len;
3164 enum gro_result ret;
3166 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3167 goto normal;
3169 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3170 goto normal;
3172 rcu_read_lock();
3173 list_for_each_entry_rcu(ptype, head, list) {
3174 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3175 continue;
3177 skb_set_network_header(skb, skb_gro_offset(skb));
3178 mac_len = skb->network_header - skb->mac_header;
3179 skb->mac_len = mac_len;
3180 NAPI_GRO_CB(skb)->same_flow = 0;
3181 NAPI_GRO_CB(skb)->flush = 0;
3182 NAPI_GRO_CB(skb)->free = 0;
3184 pp = ptype->gro_receive(&napi->gro_list, skb);
3185 break;
3187 rcu_read_unlock();
3189 if (&ptype->list == head)
3190 goto normal;
3192 same_flow = NAPI_GRO_CB(skb)->same_flow;
3193 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3195 if (pp) {
3196 struct sk_buff *nskb = *pp;
3198 *pp = nskb->next;
3199 nskb->next = NULL;
3200 napi_gro_complete(nskb);
3201 napi->gro_count--;
3204 if (same_flow)
3205 goto ok;
3207 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3208 goto normal;
3210 napi->gro_count++;
3211 NAPI_GRO_CB(skb)->count = 1;
3212 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3213 skb->next = napi->gro_list;
3214 napi->gro_list = skb;
3215 ret = GRO_HELD;
3217 pull:
3218 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3219 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3221 BUG_ON(skb->end - skb->tail < grow);
3223 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3225 skb->tail += grow;
3226 skb->data_len -= grow;
3228 skb_shinfo(skb)->frags[0].page_offset += grow;
3229 skb_shinfo(skb)->frags[0].size -= grow;
3231 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3232 put_page(skb_shinfo(skb)->frags[0].page);
3233 memmove(skb_shinfo(skb)->frags,
3234 skb_shinfo(skb)->frags + 1,
3235 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3240 return ret;
3242 normal:
3243 ret = GRO_NORMAL;
3244 goto pull;
3246 EXPORT_SYMBOL(dev_gro_receive);
3248 static inline gro_result_t
3249 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3251 struct sk_buff *p;
3253 for (p = napi->gro_list; p; p = p->next) {
3254 unsigned long diffs;
3256 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3257 diffs |= p->vlan_tci ^ skb->vlan_tci;
3258 diffs |= compare_ether_header(skb_mac_header(p),
3259 skb_gro_mac_header(skb));
3260 NAPI_GRO_CB(p)->same_flow = !diffs;
3261 NAPI_GRO_CB(p)->flush = 0;
3264 return dev_gro_receive(napi, skb);
3267 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3269 switch (ret) {
3270 case GRO_NORMAL:
3271 if (netif_receive_skb(skb))
3272 ret = GRO_DROP;
3273 break;
3275 case GRO_DROP:
3276 case GRO_MERGED_FREE:
3277 kfree_skb(skb);
3278 break;
3280 case GRO_HELD:
3281 case GRO_MERGED:
3282 break;
3285 return ret;
3287 EXPORT_SYMBOL(napi_skb_finish);
3289 void skb_gro_reset_offset(struct sk_buff *skb)
3291 NAPI_GRO_CB(skb)->data_offset = 0;
3292 NAPI_GRO_CB(skb)->frag0 = NULL;
3293 NAPI_GRO_CB(skb)->frag0_len = 0;
3295 if (skb->mac_header == skb->tail &&
3296 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3297 NAPI_GRO_CB(skb)->frag0 =
3298 page_address(skb_shinfo(skb)->frags[0].page) +
3299 skb_shinfo(skb)->frags[0].page_offset;
3300 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3303 EXPORT_SYMBOL(skb_gro_reset_offset);
3305 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3307 skb_gro_reset_offset(skb);
3309 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3311 EXPORT_SYMBOL(napi_gro_receive);
3313 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3315 __skb_pull(skb, skb_headlen(skb));
3316 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3317 skb->vlan_tci = 0;
3319 napi->skb = skb;
3322 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3324 struct sk_buff *skb = napi->skb;
3326 if (!skb) {
3327 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3328 if (skb)
3329 napi->skb = skb;
3331 return skb;
3333 EXPORT_SYMBOL(napi_get_frags);
3335 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3336 gro_result_t ret)
3338 switch (ret) {
3339 case GRO_NORMAL:
3340 case GRO_HELD:
3341 skb->protocol = eth_type_trans(skb, skb->dev);
3343 if (ret == GRO_HELD)
3344 skb_gro_pull(skb, -ETH_HLEN);
3345 else if (netif_receive_skb(skb))
3346 ret = GRO_DROP;
3347 break;
3349 case GRO_DROP:
3350 case GRO_MERGED_FREE:
3351 napi_reuse_skb(napi, skb);
3352 break;
3354 case GRO_MERGED:
3355 break;
3358 return ret;
3360 EXPORT_SYMBOL(napi_frags_finish);
3362 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3364 struct sk_buff *skb = napi->skb;
3365 struct ethhdr *eth;
3366 unsigned int hlen;
3367 unsigned int off;
3369 napi->skb = NULL;
3371 skb_reset_mac_header(skb);
3372 skb_gro_reset_offset(skb);
3374 off = skb_gro_offset(skb);
3375 hlen = off + sizeof(*eth);
3376 eth = skb_gro_header_fast(skb, off);
3377 if (skb_gro_header_hard(skb, hlen)) {
3378 eth = skb_gro_header_slow(skb, hlen, off);
3379 if (unlikely(!eth)) {
3380 napi_reuse_skb(napi, skb);
3381 skb = NULL;
3382 goto out;
3386 skb_gro_pull(skb, sizeof(*eth));
3389 * This works because the only protocols we care about don't require
3390 * special handling. We'll fix it up properly at the end.
3392 skb->protocol = eth->h_proto;
3394 out:
3395 return skb;
3397 EXPORT_SYMBOL(napi_frags_skb);
3399 gro_result_t napi_gro_frags(struct napi_struct *napi)
3401 struct sk_buff *skb = napi_frags_skb(napi);
3403 if (!skb)
3404 return GRO_DROP;
3406 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3408 EXPORT_SYMBOL(napi_gro_frags);
3411 * net_rps_action sends any pending IPI's for rps.
3412 * Note: called with local irq disabled, but exits with local irq enabled.
3414 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3416 #ifdef CONFIG_RPS
3417 struct softnet_data *remsd = sd->rps_ipi_list;
3419 if (remsd) {
3420 sd->rps_ipi_list = NULL;
3422 local_irq_enable();
3424 /* Send pending IPI's to kick RPS processing on remote cpus. */
3425 while (remsd) {
3426 struct softnet_data *next = remsd->rps_ipi_next;
3428 if (cpu_online(remsd->cpu))
3429 __smp_call_function_single(remsd->cpu,
3430 &remsd->csd, 0);
3431 remsd = next;
3433 } else
3434 #endif
3435 local_irq_enable();
3438 static int process_backlog(struct napi_struct *napi, int quota)
3440 int work = 0;
3441 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3443 #ifdef CONFIG_RPS
3444 /* Check if we have pending ipi, its better to send them now,
3445 * not waiting net_rx_action() end.
3447 if (sd->rps_ipi_list) {
3448 local_irq_disable();
3449 net_rps_action_and_irq_enable(sd);
3451 #endif
3452 napi->weight = weight_p;
3453 local_irq_disable();
3454 while (work < quota) {
3455 struct sk_buff *skb;
3456 unsigned int qlen;
3458 while ((skb = __skb_dequeue(&sd->process_queue))) {
3459 local_irq_enable();
3460 __netif_receive_skb(skb);
3461 local_irq_disable();
3462 input_queue_head_incr(sd);
3463 if (++work >= quota) {
3464 local_irq_enable();
3465 return work;
3469 rps_lock(sd);
3470 qlen = skb_queue_len(&sd->input_pkt_queue);
3471 if (qlen)
3472 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3473 &sd->process_queue);
3475 if (qlen < quota - work) {
3477 * Inline a custom version of __napi_complete().
3478 * only current cpu owns and manipulates this napi,
3479 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3480 * we can use a plain write instead of clear_bit(),
3481 * and we dont need an smp_mb() memory barrier.
3483 list_del(&napi->poll_list);
3484 napi->state = 0;
3486 quota = work + qlen;
3488 rps_unlock(sd);
3490 local_irq_enable();
3492 return work;
3496 * __napi_schedule - schedule for receive
3497 * @n: entry to schedule
3499 * The entry's receive function will be scheduled to run
3501 void __napi_schedule(struct napi_struct *n)
3503 unsigned long flags;
3505 local_irq_save(flags);
3506 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3507 local_irq_restore(flags);
3509 EXPORT_SYMBOL(__napi_schedule);
3511 void __napi_complete(struct napi_struct *n)
3513 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3514 BUG_ON(n->gro_list);
3516 list_del(&n->poll_list);
3517 smp_mb__before_clear_bit();
3518 clear_bit(NAPI_STATE_SCHED, &n->state);
3520 EXPORT_SYMBOL(__napi_complete);
3522 void napi_complete(struct napi_struct *n)
3524 unsigned long flags;
3527 * don't let napi dequeue from the cpu poll list
3528 * just in case its running on a different cpu
3530 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3531 return;
3533 napi_gro_flush(n);
3534 local_irq_save(flags);
3535 __napi_complete(n);
3536 local_irq_restore(flags);
3538 EXPORT_SYMBOL(napi_complete);
3540 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3541 int (*poll)(struct napi_struct *, int), int weight)
3543 INIT_LIST_HEAD(&napi->poll_list);
3544 napi->gro_count = 0;
3545 napi->gro_list = NULL;
3546 napi->skb = NULL;
3547 napi->poll = poll;
3548 napi->weight = weight;
3549 list_add(&napi->dev_list, &dev->napi_list);
3550 napi->dev = dev;
3551 #ifdef CONFIG_NETPOLL
3552 spin_lock_init(&napi->poll_lock);
3553 napi->poll_owner = -1;
3554 #endif
3555 set_bit(NAPI_STATE_SCHED, &napi->state);
3557 EXPORT_SYMBOL(netif_napi_add);
3559 void netif_napi_del(struct napi_struct *napi)
3561 struct sk_buff *skb, *next;
3563 list_del_init(&napi->dev_list);
3564 napi_free_frags(napi);
3566 for (skb = napi->gro_list; skb; skb = next) {
3567 next = skb->next;
3568 skb->next = NULL;
3569 kfree_skb(skb);
3572 napi->gro_list = NULL;
3573 napi->gro_count = 0;
3575 EXPORT_SYMBOL(netif_napi_del);
3577 static void net_rx_action(struct softirq_action *h)
3579 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3580 unsigned long time_limit = jiffies + 2;
3581 int budget = netdev_budget;
3582 void *have;
3584 local_irq_disable();
3586 while (!list_empty(&sd->poll_list)) {
3587 struct napi_struct *n;
3588 int work, weight;
3590 /* If softirq window is exhuasted then punt.
3591 * Allow this to run for 2 jiffies since which will allow
3592 * an average latency of 1.5/HZ.
3594 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3595 goto softnet_break;
3597 local_irq_enable();
3599 /* Even though interrupts have been re-enabled, this
3600 * access is safe because interrupts can only add new
3601 * entries to the tail of this list, and only ->poll()
3602 * calls can remove this head entry from the list.
3604 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3606 have = netpoll_poll_lock(n);
3608 weight = n->weight;
3610 /* This NAPI_STATE_SCHED test is for avoiding a race
3611 * with netpoll's poll_napi(). Only the entity which
3612 * obtains the lock and sees NAPI_STATE_SCHED set will
3613 * actually make the ->poll() call. Therefore we avoid
3614 * accidently calling ->poll() when NAPI is not scheduled.
3616 work = 0;
3617 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3618 work = n->poll(n, weight);
3619 trace_napi_poll(n);
3622 WARN_ON_ONCE(work > weight);
3624 budget -= work;
3626 local_irq_disable();
3628 /* Drivers must not modify the NAPI state if they
3629 * consume the entire weight. In such cases this code
3630 * still "owns" the NAPI instance and therefore can
3631 * move the instance around on the list at-will.
3633 if (unlikely(work == weight)) {
3634 if (unlikely(napi_disable_pending(n))) {
3635 local_irq_enable();
3636 napi_complete(n);
3637 local_irq_disable();
3638 } else
3639 list_move_tail(&n->poll_list, &sd->poll_list);
3642 netpoll_poll_unlock(have);
3644 out:
3645 net_rps_action_and_irq_enable(sd);
3647 #ifdef CONFIG_NET_DMA
3649 * There may not be any more sk_buffs coming right now, so push
3650 * any pending DMA copies to hardware
3652 dma_issue_pending_all();
3653 #endif
3655 return;
3657 softnet_break:
3658 sd->time_squeeze++;
3659 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3660 goto out;
3663 static gifconf_func_t *gifconf_list[NPROTO];
3666 * register_gifconf - register a SIOCGIF handler
3667 * @family: Address family
3668 * @gifconf: Function handler
3670 * Register protocol dependent address dumping routines. The handler
3671 * that is passed must not be freed or reused until it has been replaced
3672 * by another handler.
3674 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3676 if (family >= NPROTO)
3677 return -EINVAL;
3678 gifconf_list[family] = gifconf;
3679 return 0;
3681 EXPORT_SYMBOL(register_gifconf);
3685 * Map an interface index to its name (SIOCGIFNAME)
3689 * We need this ioctl for efficient implementation of the
3690 * if_indextoname() function required by the IPv6 API. Without
3691 * it, we would have to search all the interfaces to find a
3692 * match. --pb
3695 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3697 struct net_device *dev;
3698 struct ifreq ifr;
3701 * Fetch the caller's info block.
3704 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3705 return -EFAULT;
3707 rcu_read_lock();
3708 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3709 if (!dev) {
3710 rcu_read_unlock();
3711 return -ENODEV;
3714 strcpy(ifr.ifr_name, dev->name);
3715 rcu_read_unlock();
3717 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3718 return -EFAULT;
3719 return 0;
3723 * Perform a SIOCGIFCONF call. This structure will change
3724 * size eventually, and there is nothing I can do about it.
3725 * Thus we will need a 'compatibility mode'.
3728 static int dev_ifconf(struct net *net, char __user *arg)
3730 struct ifconf ifc;
3731 struct net_device *dev;
3732 char __user *pos;
3733 int len;
3734 int total;
3735 int i;
3738 * Fetch the caller's info block.
3741 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3742 return -EFAULT;
3744 pos = ifc.ifc_buf;
3745 len = ifc.ifc_len;
3748 * Loop over the interfaces, and write an info block for each.
3751 total = 0;
3752 for_each_netdev(net, dev) {
3753 for (i = 0; i < NPROTO; i++) {
3754 if (gifconf_list[i]) {
3755 int done;
3756 if (!pos)
3757 done = gifconf_list[i](dev, NULL, 0);
3758 else
3759 done = gifconf_list[i](dev, pos + total,
3760 len - total);
3761 if (done < 0)
3762 return -EFAULT;
3763 total += done;
3769 * All done. Write the updated control block back to the caller.
3771 ifc.ifc_len = total;
3774 * Both BSD and Solaris return 0 here, so we do too.
3776 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3779 #ifdef CONFIG_PROC_FS
3781 * This is invoked by the /proc filesystem handler to display a device
3782 * in detail.
3784 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3785 __acquires(RCU)
3787 struct net *net = seq_file_net(seq);
3788 loff_t off;
3789 struct net_device *dev;
3791 rcu_read_lock();
3792 if (!*pos)
3793 return SEQ_START_TOKEN;
3795 off = 1;
3796 for_each_netdev_rcu(net, dev)
3797 if (off++ == *pos)
3798 return dev;
3800 return NULL;
3803 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3805 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3806 first_net_device(seq_file_net(seq)) :
3807 next_net_device((struct net_device *)v);
3809 ++*pos;
3810 return rcu_dereference(dev);
3813 void dev_seq_stop(struct seq_file *seq, void *v)
3814 __releases(RCU)
3816 rcu_read_unlock();
3819 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3821 struct rtnl_link_stats64 temp;
3822 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
3824 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
3825 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
3826 dev->name, stats->rx_bytes, stats->rx_packets,
3827 stats->rx_errors,
3828 stats->rx_dropped + stats->rx_missed_errors,
3829 stats->rx_fifo_errors,
3830 stats->rx_length_errors + stats->rx_over_errors +
3831 stats->rx_crc_errors + stats->rx_frame_errors,
3832 stats->rx_compressed, stats->multicast,
3833 stats->tx_bytes, stats->tx_packets,
3834 stats->tx_errors, stats->tx_dropped,
3835 stats->tx_fifo_errors, stats->collisions,
3836 stats->tx_carrier_errors +
3837 stats->tx_aborted_errors +
3838 stats->tx_window_errors +
3839 stats->tx_heartbeat_errors,
3840 stats->tx_compressed);
3844 * Called from the PROCfs module. This now uses the new arbitrary sized
3845 * /proc/net interface to create /proc/net/dev
3847 static int dev_seq_show(struct seq_file *seq, void *v)
3849 if (v == SEQ_START_TOKEN)
3850 seq_puts(seq, "Inter-| Receive "
3851 " | Transmit\n"
3852 " face |bytes packets errs drop fifo frame "
3853 "compressed multicast|bytes packets errs "
3854 "drop fifo colls carrier compressed\n");
3855 else
3856 dev_seq_printf_stats(seq, v);
3857 return 0;
3860 static struct softnet_data *softnet_get_online(loff_t *pos)
3862 struct softnet_data *sd = NULL;
3864 while (*pos < nr_cpu_ids)
3865 if (cpu_online(*pos)) {
3866 sd = &per_cpu(softnet_data, *pos);
3867 break;
3868 } else
3869 ++*pos;
3870 return sd;
3873 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3875 return softnet_get_online(pos);
3878 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3880 ++*pos;
3881 return softnet_get_online(pos);
3884 static void softnet_seq_stop(struct seq_file *seq, void *v)
3888 static int softnet_seq_show(struct seq_file *seq, void *v)
3890 struct softnet_data *sd = v;
3892 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3893 sd->processed, sd->dropped, sd->time_squeeze, 0,
3894 0, 0, 0, 0, /* was fastroute */
3895 sd->cpu_collision, sd->received_rps);
3896 return 0;
3899 static const struct seq_operations dev_seq_ops = {
3900 .start = dev_seq_start,
3901 .next = dev_seq_next,
3902 .stop = dev_seq_stop,
3903 .show = dev_seq_show,
3906 static int dev_seq_open(struct inode *inode, struct file *file)
3908 return seq_open_net(inode, file, &dev_seq_ops,
3909 sizeof(struct seq_net_private));
3912 static const struct file_operations dev_seq_fops = {
3913 .owner = THIS_MODULE,
3914 .open = dev_seq_open,
3915 .read = seq_read,
3916 .llseek = seq_lseek,
3917 .release = seq_release_net,
3920 static const struct seq_operations softnet_seq_ops = {
3921 .start = softnet_seq_start,
3922 .next = softnet_seq_next,
3923 .stop = softnet_seq_stop,
3924 .show = softnet_seq_show,
3927 static int softnet_seq_open(struct inode *inode, struct file *file)
3929 return seq_open(file, &softnet_seq_ops);
3932 static const struct file_operations softnet_seq_fops = {
3933 .owner = THIS_MODULE,
3934 .open = softnet_seq_open,
3935 .read = seq_read,
3936 .llseek = seq_lseek,
3937 .release = seq_release,
3940 static void *ptype_get_idx(loff_t pos)
3942 struct packet_type *pt = NULL;
3943 loff_t i = 0;
3944 int t;
3946 list_for_each_entry_rcu(pt, &ptype_all, list) {
3947 if (i == pos)
3948 return pt;
3949 ++i;
3952 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3953 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3954 if (i == pos)
3955 return pt;
3956 ++i;
3959 return NULL;
3962 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3963 __acquires(RCU)
3965 rcu_read_lock();
3966 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3969 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3971 struct packet_type *pt;
3972 struct list_head *nxt;
3973 int hash;
3975 ++*pos;
3976 if (v == SEQ_START_TOKEN)
3977 return ptype_get_idx(0);
3979 pt = v;
3980 nxt = pt->list.next;
3981 if (pt->type == htons(ETH_P_ALL)) {
3982 if (nxt != &ptype_all)
3983 goto found;
3984 hash = 0;
3985 nxt = ptype_base[0].next;
3986 } else
3987 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3989 while (nxt == &ptype_base[hash]) {
3990 if (++hash >= PTYPE_HASH_SIZE)
3991 return NULL;
3992 nxt = ptype_base[hash].next;
3994 found:
3995 return list_entry(nxt, struct packet_type, list);
3998 static void ptype_seq_stop(struct seq_file *seq, void *v)
3999 __releases(RCU)
4001 rcu_read_unlock();
4004 static int ptype_seq_show(struct seq_file *seq, void *v)
4006 struct packet_type *pt = v;
4008 if (v == SEQ_START_TOKEN)
4009 seq_puts(seq, "Type Device Function\n");
4010 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4011 if (pt->type == htons(ETH_P_ALL))
4012 seq_puts(seq, "ALL ");
4013 else
4014 seq_printf(seq, "%04x", ntohs(pt->type));
4016 seq_printf(seq, " %-8s %pF\n",
4017 pt->dev ? pt->dev->name : "", pt->func);
4020 return 0;
4023 static const struct seq_operations ptype_seq_ops = {
4024 .start = ptype_seq_start,
4025 .next = ptype_seq_next,
4026 .stop = ptype_seq_stop,
4027 .show = ptype_seq_show,
4030 static int ptype_seq_open(struct inode *inode, struct file *file)
4032 return seq_open_net(inode, file, &ptype_seq_ops,
4033 sizeof(struct seq_net_private));
4036 static const struct file_operations ptype_seq_fops = {
4037 .owner = THIS_MODULE,
4038 .open = ptype_seq_open,
4039 .read = seq_read,
4040 .llseek = seq_lseek,
4041 .release = seq_release_net,
4045 static int __net_init dev_proc_net_init(struct net *net)
4047 int rc = -ENOMEM;
4049 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4050 goto out;
4051 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4052 goto out_dev;
4053 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4054 goto out_softnet;
4056 if (wext_proc_init(net))
4057 goto out_ptype;
4058 rc = 0;
4059 out:
4060 return rc;
4061 out_ptype:
4062 proc_net_remove(net, "ptype");
4063 out_softnet:
4064 proc_net_remove(net, "softnet_stat");
4065 out_dev:
4066 proc_net_remove(net, "dev");
4067 goto out;
4070 static void __net_exit dev_proc_net_exit(struct net *net)
4072 wext_proc_exit(net);
4074 proc_net_remove(net, "ptype");
4075 proc_net_remove(net, "softnet_stat");
4076 proc_net_remove(net, "dev");
4079 static struct pernet_operations __net_initdata dev_proc_ops = {
4080 .init = dev_proc_net_init,
4081 .exit = dev_proc_net_exit,
4084 static int __init dev_proc_init(void)
4086 return register_pernet_subsys(&dev_proc_ops);
4088 #else
4089 #define dev_proc_init() 0
4090 #endif /* CONFIG_PROC_FS */
4094 * netdev_set_master - set up master/slave pair
4095 * @slave: slave device
4096 * @master: new master device
4098 * Changes the master device of the slave. Pass %NULL to break the
4099 * bonding. The caller must hold the RTNL semaphore. On a failure
4100 * a negative errno code is returned. On success the reference counts
4101 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4102 * function returns zero.
4104 int netdev_set_master(struct net_device *slave, struct net_device *master)
4106 struct net_device *old = slave->master;
4108 ASSERT_RTNL();
4110 if (master) {
4111 if (old)
4112 return -EBUSY;
4113 dev_hold(master);
4116 slave->master = master;
4118 if (old) {
4119 synchronize_net();
4120 dev_put(old);
4122 if (master)
4123 slave->flags |= IFF_SLAVE;
4124 else
4125 slave->flags &= ~IFF_SLAVE;
4127 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4128 return 0;
4130 EXPORT_SYMBOL(netdev_set_master);
4132 static void dev_change_rx_flags(struct net_device *dev, int flags)
4134 const struct net_device_ops *ops = dev->netdev_ops;
4136 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4137 ops->ndo_change_rx_flags(dev, flags);
4140 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4142 unsigned short old_flags = dev->flags;
4143 uid_t uid;
4144 gid_t gid;
4146 ASSERT_RTNL();
4148 dev->flags |= IFF_PROMISC;
4149 dev->promiscuity += inc;
4150 if (dev->promiscuity == 0) {
4152 * Avoid overflow.
4153 * If inc causes overflow, untouch promisc and return error.
4155 if (inc < 0)
4156 dev->flags &= ~IFF_PROMISC;
4157 else {
4158 dev->promiscuity -= inc;
4159 printk(KERN_WARNING "%s: promiscuity touches roof, "
4160 "set promiscuity failed, promiscuity feature "
4161 "of device might be broken.\n", dev->name);
4162 return -EOVERFLOW;
4165 if (dev->flags != old_flags) {
4166 printk(KERN_INFO "device %s %s promiscuous mode\n",
4167 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4168 "left");
4169 if (audit_enabled) {
4170 current_uid_gid(&uid, &gid);
4171 audit_log(current->audit_context, GFP_ATOMIC,
4172 AUDIT_ANOM_PROMISCUOUS,
4173 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4174 dev->name, (dev->flags & IFF_PROMISC),
4175 (old_flags & IFF_PROMISC),
4176 audit_get_loginuid(current),
4177 uid, gid,
4178 audit_get_sessionid(current));
4181 dev_change_rx_flags(dev, IFF_PROMISC);
4183 return 0;
4187 * dev_set_promiscuity - update promiscuity count on a device
4188 * @dev: device
4189 * @inc: modifier
4191 * Add or remove promiscuity from a device. While the count in the device
4192 * remains above zero the interface remains promiscuous. Once it hits zero
4193 * the device reverts back to normal filtering operation. A negative inc
4194 * value is used to drop promiscuity on the device.
4195 * Return 0 if successful or a negative errno code on error.
4197 int dev_set_promiscuity(struct net_device *dev, int inc)
4199 unsigned short old_flags = dev->flags;
4200 int err;
4202 err = __dev_set_promiscuity(dev, inc);
4203 if (err < 0)
4204 return err;
4205 if (dev->flags != old_flags)
4206 dev_set_rx_mode(dev);
4207 return err;
4209 EXPORT_SYMBOL(dev_set_promiscuity);
4212 * dev_set_allmulti - update allmulti count on a device
4213 * @dev: device
4214 * @inc: modifier
4216 * Add or remove reception of all multicast frames to a device. While the
4217 * count in the device remains above zero the interface remains listening
4218 * to all interfaces. Once it hits zero the device reverts back to normal
4219 * filtering operation. A negative @inc value is used to drop the counter
4220 * when releasing a resource needing all multicasts.
4221 * Return 0 if successful or a negative errno code on error.
4224 int dev_set_allmulti(struct net_device *dev, int inc)
4226 unsigned short old_flags = dev->flags;
4228 ASSERT_RTNL();
4230 dev->flags |= IFF_ALLMULTI;
4231 dev->allmulti += inc;
4232 if (dev->allmulti == 0) {
4234 * Avoid overflow.
4235 * If inc causes overflow, untouch allmulti and return error.
4237 if (inc < 0)
4238 dev->flags &= ~IFF_ALLMULTI;
4239 else {
4240 dev->allmulti -= inc;
4241 printk(KERN_WARNING "%s: allmulti touches roof, "
4242 "set allmulti failed, allmulti feature of "
4243 "device might be broken.\n", dev->name);
4244 return -EOVERFLOW;
4247 if (dev->flags ^ old_flags) {
4248 dev_change_rx_flags(dev, IFF_ALLMULTI);
4249 dev_set_rx_mode(dev);
4251 return 0;
4253 EXPORT_SYMBOL(dev_set_allmulti);
4256 * Upload unicast and multicast address lists to device and
4257 * configure RX filtering. When the device doesn't support unicast
4258 * filtering it is put in promiscuous mode while unicast addresses
4259 * are present.
4261 void __dev_set_rx_mode(struct net_device *dev)
4263 const struct net_device_ops *ops = dev->netdev_ops;
4265 /* dev_open will call this function so the list will stay sane. */
4266 if (!(dev->flags&IFF_UP))
4267 return;
4269 if (!netif_device_present(dev))
4270 return;
4272 if (ops->ndo_set_rx_mode)
4273 ops->ndo_set_rx_mode(dev);
4274 else {
4275 /* Unicast addresses changes may only happen under the rtnl,
4276 * therefore calling __dev_set_promiscuity here is safe.
4278 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4279 __dev_set_promiscuity(dev, 1);
4280 dev->uc_promisc = 1;
4281 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4282 __dev_set_promiscuity(dev, -1);
4283 dev->uc_promisc = 0;
4286 if (ops->ndo_set_multicast_list)
4287 ops->ndo_set_multicast_list(dev);
4291 void dev_set_rx_mode(struct net_device *dev)
4293 netif_addr_lock_bh(dev);
4294 __dev_set_rx_mode(dev);
4295 netif_addr_unlock_bh(dev);
4299 * dev_get_flags - get flags reported to userspace
4300 * @dev: device
4302 * Get the combination of flag bits exported through APIs to userspace.
4304 unsigned dev_get_flags(const struct net_device *dev)
4306 unsigned flags;
4308 flags = (dev->flags & ~(IFF_PROMISC |
4309 IFF_ALLMULTI |
4310 IFF_RUNNING |
4311 IFF_LOWER_UP |
4312 IFF_DORMANT)) |
4313 (dev->gflags & (IFF_PROMISC |
4314 IFF_ALLMULTI));
4316 if (netif_running(dev)) {
4317 if (netif_oper_up(dev))
4318 flags |= IFF_RUNNING;
4319 if (netif_carrier_ok(dev))
4320 flags |= IFF_LOWER_UP;
4321 if (netif_dormant(dev))
4322 flags |= IFF_DORMANT;
4325 return flags;
4327 EXPORT_SYMBOL(dev_get_flags);
4329 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4331 int old_flags = dev->flags;
4332 int ret;
4334 ASSERT_RTNL();
4337 * Set the flags on our device.
4340 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4341 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4342 IFF_AUTOMEDIA)) |
4343 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4344 IFF_ALLMULTI));
4347 * Load in the correct multicast list now the flags have changed.
4350 if ((old_flags ^ flags) & IFF_MULTICAST)
4351 dev_change_rx_flags(dev, IFF_MULTICAST);
4353 dev_set_rx_mode(dev);
4356 * Have we downed the interface. We handle IFF_UP ourselves
4357 * according to user attempts to set it, rather than blindly
4358 * setting it.
4361 ret = 0;
4362 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4363 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4365 if (!ret)
4366 dev_set_rx_mode(dev);
4369 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4370 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4372 dev->gflags ^= IFF_PROMISC;
4373 dev_set_promiscuity(dev, inc);
4376 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4377 is important. Some (broken) drivers set IFF_PROMISC, when
4378 IFF_ALLMULTI is requested not asking us and not reporting.
4380 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4381 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4383 dev->gflags ^= IFF_ALLMULTI;
4384 dev_set_allmulti(dev, inc);
4387 return ret;
4390 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4392 unsigned int changes = dev->flags ^ old_flags;
4394 if (changes & IFF_UP) {
4395 if (dev->flags & IFF_UP)
4396 call_netdevice_notifiers(NETDEV_UP, dev);
4397 else
4398 call_netdevice_notifiers(NETDEV_DOWN, dev);
4401 if (dev->flags & IFF_UP &&
4402 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4403 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4407 * dev_change_flags - change device settings
4408 * @dev: device
4409 * @flags: device state flags
4411 * Change settings on device based state flags. The flags are
4412 * in the userspace exported format.
4414 int dev_change_flags(struct net_device *dev, unsigned flags)
4416 int ret, changes;
4417 int old_flags = dev->flags;
4419 ret = __dev_change_flags(dev, flags);
4420 if (ret < 0)
4421 return ret;
4423 changes = old_flags ^ dev->flags;
4424 if (changes)
4425 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4427 __dev_notify_flags(dev, old_flags);
4428 return ret;
4430 EXPORT_SYMBOL(dev_change_flags);
4433 * dev_set_mtu - Change maximum transfer unit
4434 * @dev: device
4435 * @new_mtu: new transfer unit
4437 * Change the maximum transfer size of the network device.
4439 int dev_set_mtu(struct net_device *dev, int new_mtu)
4441 const struct net_device_ops *ops = dev->netdev_ops;
4442 int err;
4444 if (new_mtu == dev->mtu)
4445 return 0;
4447 /* MTU must be positive. */
4448 if (new_mtu < 0)
4449 return -EINVAL;
4451 if (!netif_device_present(dev))
4452 return -ENODEV;
4454 err = 0;
4455 if (ops->ndo_change_mtu)
4456 err = ops->ndo_change_mtu(dev, new_mtu);
4457 else
4458 dev->mtu = new_mtu;
4460 if (!err && dev->flags & IFF_UP)
4461 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4462 return err;
4464 EXPORT_SYMBOL(dev_set_mtu);
4467 * dev_set_mac_address - Change Media Access Control Address
4468 * @dev: device
4469 * @sa: new address
4471 * Change the hardware (MAC) address of the device
4473 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4475 const struct net_device_ops *ops = dev->netdev_ops;
4476 int err;
4478 if (!ops->ndo_set_mac_address)
4479 return -EOPNOTSUPP;
4480 if (sa->sa_family != dev->type)
4481 return -EINVAL;
4482 if (!netif_device_present(dev))
4483 return -ENODEV;
4484 err = ops->ndo_set_mac_address(dev, sa);
4485 if (!err)
4486 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4487 return err;
4489 EXPORT_SYMBOL(dev_set_mac_address);
4492 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4494 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4496 int err;
4497 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4499 if (!dev)
4500 return -ENODEV;
4502 switch (cmd) {
4503 case SIOCGIFFLAGS: /* Get interface flags */
4504 ifr->ifr_flags = (short) dev_get_flags(dev);
4505 return 0;
4507 case SIOCGIFMETRIC: /* Get the metric on the interface
4508 (currently unused) */
4509 ifr->ifr_metric = 0;
4510 return 0;
4512 case SIOCGIFMTU: /* Get the MTU of a device */
4513 ifr->ifr_mtu = dev->mtu;
4514 return 0;
4516 case SIOCGIFHWADDR:
4517 if (!dev->addr_len)
4518 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4519 else
4520 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4521 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4522 ifr->ifr_hwaddr.sa_family = dev->type;
4523 return 0;
4525 case SIOCGIFSLAVE:
4526 err = -EINVAL;
4527 break;
4529 case SIOCGIFMAP:
4530 ifr->ifr_map.mem_start = dev->mem_start;
4531 ifr->ifr_map.mem_end = dev->mem_end;
4532 ifr->ifr_map.base_addr = dev->base_addr;
4533 ifr->ifr_map.irq = dev->irq;
4534 ifr->ifr_map.dma = dev->dma;
4535 ifr->ifr_map.port = dev->if_port;
4536 return 0;
4538 case SIOCGIFINDEX:
4539 ifr->ifr_ifindex = dev->ifindex;
4540 return 0;
4542 case SIOCGIFTXQLEN:
4543 ifr->ifr_qlen = dev->tx_queue_len;
4544 return 0;
4546 default:
4547 /* dev_ioctl() should ensure this case
4548 * is never reached
4550 WARN_ON(1);
4551 err = -EINVAL;
4552 break;
4555 return err;
4559 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4561 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4563 int err;
4564 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4565 const struct net_device_ops *ops;
4567 if (!dev)
4568 return -ENODEV;
4570 ops = dev->netdev_ops;
4572 switch (cmd) {
4573 case SIOCSIFFLAGS: /* Set interface flags */
4574 return dev_change_flags(dev, ifr->ifr_flags);
4576 case SIOCSIFMETRIC: /* Set the metric on the interface
4577 (currently unused) */
4578 return -EOPNOTSUPP;
4580 case SIOCSIFMTU: /* Set the MTU of a device */
4581 return dev_set_mtu(dev, ifr->ifr_mtu);
4583 case SIOCSIFHWADDR:
4584 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4586 case SIOCSIFHWBROADCAST:
4587 if (ifr->ifr_hwaddr.sa_family != dev->type)
4588 return -EINVAL;
4589 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4590 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4591 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4592 return 0;
4594 case SIOCSIFMAP:
4595 if (ops->ndo_set_config) {
4596 if (!netif_device_present(dev))
4597 return -ENODEV;
4598 return ops->ndo_set_config(dev, &ifr->ifr_map);
4600 return -EOPNOTSUPP;
4602 case SIOCADDMULTI:
4603 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4604 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4605 return -EINVAL;
4606 if (!netif_device_present(dev))
4607 return -ENODEV;
4608 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4610 case SIOCDELMULTI:
4611 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4612 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4613 return -EINVAL;
4614 if (!netif_device_present(dev))
4615 return -ENODEV;
4616 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4618 case SIOCSIFTXQLEN:
4619 if (ifr->ifr_qlen < 0)
4620 return -EINVAL;
4621 dev->tx_queue_len = ifr->ifr_qlen;
4622 return 0;
4624 case SIOCSIFNAME:
4625 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4626 return dev_change_name(dev, ifr->ifr_newname);
4629 * Unknown or private ioctl
4631 default:
4632 if ((cmd >= SIOCDEVPRIVATE &&
4633 cmd <= SIOCDEVPRIVATE + 15) ||
4634 cmd == SIOCBONDENSLAVE ||
4635 cmd == SIOCBONDRELEASE ||
4636 cmd == SIOCBONDSETHWADDR ||
4637 cmd == SIOCBONDSLAVEINFOQUERY ||
4638 cmd == SIOCBONDINFOQUERY ||
4639 cmd == SIOCBONDCHANGEACTIVE ||
4640 cmd == SIOCGMIIPHY ||
4641 cmd == SIOCGMIIREG ||
4642 cmd == SIOCSMIIREG ||
4643 cmd == SIOCBRADDIF ||
4644 cmd == SIOCBRDELIF ||
4645 cmd == SIOCSHWTSTAMP ||
4646 cmd == SIOCWANDEV) {
4647 err = -EOPNOTSUPP;
4648 if (ops->ndo_do_ioctl) {
4649 if (netif_device_present(dev))
4650 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4651 else
4652 err = -ENODEV;
4654 } else
4655 err = -EINVAL;
4658 return err;
4662 * This function handles all "interface"-type I/O control requests. The actual
4663 * 'doing' part of this is dev_ifsioc above.
4667 * dev_ioctl - network device ioctl
4668 * @net: the applicable net namespace
4669 * @cmd: command to issue
4670 * @arg: pointer to a struct ifreq in user space
4672 * Issue ioctl functions to devices. This is normally called by the
4673 * user space syscall interfaces but can sometimes be useful for
4674 * other purposes. The return value is the return from the syscall if
4675 * positive or a negative errno code on error.
4678 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4680 struct ifreq ifr;
4681 int ret;
4682 char *colon;
4684 /* One special case: SIOCGIFCONF takes ifconf argument
4685 and requires shared lock, because it sleeps writing
4686 to user space.
4689 if (cmd == SIOCGIFCONF) {
4690 rtnl_lock();
4691 ret = dev_ifconf(net, (char __user *) arg);
4692 rtnl_unlock();
4693 return ret;
4695 if (cmd == SIOCGIFNAME)
4696 return dev_ifname(net, (struct ifreq __user *)arg);
4698 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4699 return -EFAULT;
4701 ifr.ifr_name[IFNAMSIZ-1] = 0;
4703 colon = strchr(ifr.ifr_name, ':');
4704 if (colon)
4705 *colon = 0;
4708 * See which interface the caller is talking about.
4711 switch (cmd) {
4713 * These ioctl calls:
4714 * - can be done by all.
4715 * - atomic and do not require locking.
4716 * - return a value
4718 case SIOCGIFFLAGS:
4719 case SIOCGIFMETRIC:
4720 case SIOCGIFMTU:
4721 case SIOCGIFHWADDR:
4722 case SIOCGIFSLAVE:
4723 case SIOCGIFMAP:
4724 case SIOCGIFINDEX:
4725 case SIOCGIFTXQLEN:
4726 dev_load(net, ifr.ifr_name);
4727 rcu_read_lock();
4728 ret = dev_ifsioc_locked(net, &ifr, cmd);
4729 rcu_read_unlock();
4730 if (!ret) {
4731 if (colon)
4732 *colon = ':';
4733 if (copy_to_user(arg, &ifr,
4734 sizeof(struct ifreq)))
4735 ret = -EFAULT;
4737 return ret;
4739 case SIOCETHTOOL:
4740 dev_load(net, ifr.ifr_name);
4741 rtnl_lock();
4742 ret = dev_ethtool(net, &ifr);
4743 rtnl_unlock();
4744 if (!ret) {
4745 if (colon)
4746 *colon = ':';
4747 if (copy_to_user(arg, &ifr,
4748 sizeof(struct ifreq)))
4749 ret = -EFAULT;
4751 return ret;
4754 * These ioctl calls:
4755 * - require superuser power.
4756 * - require strict serialization.
4757 * - return a value
4759 case SIOCGMIIPHY:
4760 case SIOCGMIIREG:
4761 case SIOCSIFNAME:
4762 if (!capable(CAP_NET_ADMIN))
4763 return -EPERM;
4764 dev_load(net, ifr.ifr_name);
4765 rtnl_lock();
4766 ret = dev_ifsioc(net, &ifr, cmd);
4767 rtnl_unlock();
4768 if (!ret) {
4769 if (colon)
4770 *colon = ':';
4771 if (copy_to_user(arg, &ifr,
4772 sizeof(struct ifreq)))
4773 ret = -EFAULT;
4775 return ret;
4778 * These ioctl calls:
4779 * - require superuser power.
4780 * - require strict serialization.
4781 * - do not return a value
4783 case SIOCSIFFLAGS:
4784 case SIOCSIFMETRIC:
4785 case SIOCSIFMTU:
4786 case SIOCSIFMAP:
4787 case SIOCSIFHWADDR:
4788 case SIOCSIFSLAVE:
4789 case SIOCADDMULTI:
4790 case SIOCDELMULTI:
4791 case SIOCSIFHWBROADCAST:
4792 case SIOCSIFTXQLEN:
4793 case SIOCSMIIREG:
4794 case SIOCBONDENSLAVE:
4795 case SIOCBONDRELEASE:
4796 case SIOCBONDSETHWADDR:
4797 case SIOCBONDCHANGEACTIVE:
4798 case SIOCBRADDIF:
4799 case SIOCBRDELIF:
4800 case SIOCSHWTSTAMP:
4801 if (!capable(CAP_NET_ADMIN))
4802 return -EPERM;
4803 /* fall through */
4804 case SIOCBONDSLAVEINFOQUERY:
4805 case SIOCBONDINFOQUERY:
4806 dev_load(net, ifr.ifr_name);
4807 rtnl_lock();
4808 ret = dev_ifsioc(net, &ifr, cmd);
4809 rtnl_unlock();
4810 return ret;
4812 case SIOCGIFMEM:
4813 /* Get the per device memory space. We can add this but
4814 * currently do not support it */
4815 case SIOCSIFMEM:
4816 /* Set the per device memory buffer space.
4817 * Not applicable in our case */
4818 case SIOCSIFLINK:
4819 return -EINVAL;
4822 * Unknown or private ioctl.
4824 default:
4825 if (cmd == SIOCWANDEV ||
4826 (cmd >= SIOCDEVPRIVATE &&
4827 cmd <= SIOCDEVPRIVATE + 15)) {
4828 dev_load(net, ifr.ifr_name);
4829 rtnl_lock();
4830 ret = dev_ifsioc(net, &ifr, cmd);
4831 rtnl_unlock();
4832 if (!ret && copy_to_user(arg, &ifr,
4833 sizeof(struct ifreq)))
4834 ret = -EFAULT;
4835 return ret;
4837 /* Take care of Wireless Extensions */
4838 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4839 return wext_handle_ioctl(net, &ifr, cmd, arg);
4840 return -EINVAL;
4846 * dev_new_index - allocate an ifindex
4847 * @net: the applicable net namespace
4849 * Returns a suitable unique value for a new device interface
4850 * number. The caller must hold the rtnl semaphore or the
4851 * dev_base_lock to be sure it remains unique.
4853 static int dev_new_index(struct net *net)
4855 static int ifindex;
4856 for (;;) {
4857 if (++ifindex <= 0)
4858 ifindex = 1;
4859 if (!__dev_get_by_index(net, ifindex))
4860 return ifindex;
4864 /* Delayed registration/unregisteration */
4865 static LIST_HEAD(net_todo_list);
4867 static void net_set_todo(struct net_device *dev)
4869 list_add_tail(&dev->todo_list, &net_todo_list);
4872 static void rollback_registered_many(struct list_head *head)
4874 struct net_device *dev, *tmp;
4876 BUG_ON(dev_boot_phase);
4877 ASSERT_RTNL();
4879 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4880 /* Some devices call without registering
4881 * for initialization unwind. Remove those
4882 * devices and proceed with the remaining.
4884 if (dev->reg_state == NETREG_UNINITIALIZED) {
4885 pr_debug("unregister_netdevice: device %s/%p never "
4886 "was registered\n", dev->name, dev);
4888 WARN_ON(1);
4889 list_del(&dev->unreg_list);
4890 continue;
4893 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4895 /* If device is running, close it first. */
4896 dev_close(dev);
4898 /* And unlink it from device chain. */
4899 unlist_netdevice(dev);
4901 dev->reg_state = NETREG_UNREGISTERING;
4904 synchronize_net();
4906 list_for_each_entry(dev, head, unreg_list) {
4907 /* Shutdown queueing discipline. */
4908 dev_shutdown(dev);
4911 /* Notify protocols, that we are about to destroy
4912 this device. They should clean all the things.
4914 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4916 if (!dev->rtnl_link_ops ||
4917 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4918 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4921 * Flush the unicast and multicast chains
4923 dev_uc_flush(dev);
4924 dev_mc_flush(dev);
4926 if (dev->netdev_ops->ndo_uninit)
4927 dev->netdev_ops->ndo_uninit(dev);
4929 /* Notifier chain MUST detach us from master device. */
4930 WARN_ON(dev->master);
4932 /* Remove entries from kobject tree */
4933 netdev_unregister_kobject(dev);
4936 /* Process any work delayed until the end of the batch */
4937 dev = list_first_entry(head, struct net_device, unreg_list);
4938 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4940 rcu_barrier();
4942 list_for_each_entry(dev, head, unreg_list)
4943 dev_put(dev);
4946 static void rollback_registered(struct net_device *dev)
4948 LIST_HEAD(single);
4950 list_add(&dev->unreg_list, &single);
4951 rollback_registered_many(&single);
4954 unsigned long netdev_fix_features(unsigned long features, const char *name)
4956 /* Fix illegal SG+CSUM combinations. */
4957 if ((features & NETIF_F_SG) &&
4958 !(features & NETIF_F_ALL_CSUM)) {
4959 if (name)
4960 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4961 "checksum feature.\n", name);
4962 features &= ~NETIF_F_SG;
4965 /* TSO requires that SG is present as well. */
4966 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4967 if (name)
4968 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4969 "SG feature.\n", name);
4970 features &= ~NETIF_F_TSO;
4973 if (features & NETIF_F_UFO) {
4974 if (!(features & NETIF_F_GEN_CSUM)) {
4975 if (name)
4976 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4977 "since no NETIF_F_HW_CSUM feature.\n",
4978 name);
4979 features &= ~NETIF_F_UFO;
4982 if (!(features & NETIF_F_SG)) {
4983 if (name)
4984 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4985 "since no NETIF_F_SG feature.\n", name);
4986 features &= ~NETIF_F_UFO;
4990 return features;
4992 EXPORT_SYMBOL(netdev_fix_features);
4995 * netif_stacked_transfer_operstate - transfer operstate
4996 * @rootdev: the root or lower level device to transfer state from
4997 * @dev: the device to transfer operstate to
4999 * Transfer operational state from root to device. This is normally
5000 * called when a stacking relationship exists between the root
5001 * device and the device(a leaf device).
5003 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5004 struct net_device *dev)
5006 if (rootdev->operstate == IF_OPER_DORMANT)
5007 netif_dormant_on(dev);
5008 else
5009 netif_dormant_off(dev);
5011 if (netif_carrier_ok(rootdev)) {
5012 if (!netif_carrier_ok(dev))
5013 netif_carrier_on(dev);
5014 } else {
5015 if (netif_carrier_ok(dev))
5016 netif_carrier_off(dev);
5019 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5021 static int netif_alloc_rx_queues(struct net_device *dev)
5023 #ifdef CONFIG_RPS
5024 unsigned int i, count = dev->num_rx_queues;
5025 struct netdev_rx_queue *rx;
5027 BUG_ON(count < 1);
5029 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5030 if (!rx) {
5031 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5032 return -ENOMEM;
5034 dev->_rx = rx;
5037 * Set a pointer to first element in the array which holds the
5038 * reference count.
5040 for (i = 0; i < count; i++)
5041 rx[i].first = rx;
5042 #endif
5043 return 0;
5046 static int netif_alloc_netdev_queues(struct net_device *dev)
5048 unsigned int count = dev->num_tx_queues;
5049 struct netdev_queue *tx;
5051 BUG_ON(count < 1);
5053 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5054 if (!tx) {
5055 pr_err("netdev: Unable to allocate %u tx queues.\n",
5056 count);
5057 return -ENOMEM;
5059 dev->_tx = tx;
5060 return 0;
5063 static void netdev_init_one_queue(struct net_device *dev,
5064 struct netdev_queue *queue,
5065 void *_unused)
5067 queue->dev = dev;
5069 /* Initialize queue lock */
5070 spin_lock_init(&queue->_xmit_lock);
5071 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5072 queue->xmit_lock_owner = -1;
5075 static void netdev_init_queues(struct net_device *dev)
5077 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5078 spin_lock_init(&dev->tx_global_lock);
5082 * register_netdevice - register a network device
5083 * @dev: device to register
5085 * Take a completed network device structure and add it to the kernel
5086 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5087 * chain. 0 is returned on success. A negative errno code is returned
5088 * on a failure to set up the device, or if the name is a duplicate.
5090 * Callers must hold the rtnl semaphore. You may want
5091 * register_netdev() instead of this.
5093 * BUGS:
5094 * The locking appears insufficient to guarantee two parallel registers
5095 * will not get the same name.
5098 int register_netdevice(struct net_device *dev)
5100 int ret;
5101 struct net *net = dev_net(dev);
5103 BUG_ON(dev_boot_phase);
5104 ASSERT_RTNL();
5106 might_sleep();
5108 /* When net_device's are persistent, this will be fatal. */
5109 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5110 BUG_ON(!net);
5112 spin_lock_init(&dev->addr_list_lock);
5113 netdev_set_addr_lockdep_class(dev);
5115 dev->iflink = -1;
5117 ret = netif_alloc_rx_queues(dev);
5118 if (ret)
5119 goto out;
5121 ret = netif_alloc_netdev_queues(dev);
5122 if (ret)
5123 goto out;
5125 netdev_init_queues(dev);
5127 /* Init, if this function is available */
5128 if (dev->netdev_ops->ndo_init) {
5129 ret = dev->netdev_ops->ndo_init(dev);
5130 if (ret) {
5131 if (ret > 0)
5132 ret = -EIO;
5133 goto out;
5137 ret = dev_get_valid_name(dev, dev->name, 0);
5138 if (ret)
5139 goto err_uninit;
5141 dev->ifindex = dev_new_index(net);
5142 if (dev->iflink == -1)
5143 dev->iflink = dev->ifindex;
5145 /* Fix illegal checksum combinations */
5146 if ((dev->features & NETIF_F_HW_CSUM) &&
5147 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5148 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5149 dev->name);
5150 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5153 if ((dev->features & NETIF_F_NO_CSUM) &&
5154 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5155 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5156 dev->name);
5157 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5160 dev->features = netdev_fix_features(dev->features, dev->name);
5162 /* Enable software GSO if SG is supported. */
5163 if (dev->features & NETIF_F_SG)
5164 dev->features |= NETIF_F_GSO;
5166 /* Enable GRO and NETIF_F_HIGHDMA for vlans by default,
5167 * vlan_dev_init() will do the dev->features check, so these features
5168 * are enabled only if supported by underlying device.
5170 dev->vlan_features |= (NETIF_F_GRO | NETIF_F_HIGHDMA);
5172 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5173 ret = notifier_to_errno(ret);
5174 if (ret)
5175 goto err_uninit;
5177 ret = netdev_register_kobject(dev);
5178 if (ret)
5179 goto err_uninit;
5180 dev->reg_state = NETREG_REGISTERED;
5183 * Default initial state at registry is that the
5184 * device is present.
5187 set_bit(__LINK_STATE_PRESENT, &dev->state);
5189 dev_init_scheduler(dev);
5190 dev_hold(dev);
5191 list_netdevice(dev);
5193 /* Notify protocols, that a new device appeared. */
5194 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5195 ret = notifier_to_errno(ret);
5196 if (ret) {
5197 rollback_registered(dev);
5198 dev->reg_state = NETREG_UNREGISTERED;
5201 * Prevent userspace races by waiting until the network
5202 * device is fully setup before sending notifications.
5204 if (!dev->rtnl_link_ops ||
5205 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5206 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5208 out:
5209 return ret;
5211 err_uninit:
5212 if (dev->netdev_ops->ndo_uninit)
5213 dev->netdev_ops->ndo_uninit(dev);
5214 goto out;
5216 EXPORT_SYMBOL(register_netdevice);
5219 * init_dummy_netdev - init a dummy network device for NAPI
5220 * @dev: device to init
5222 * This takes a network device structure and initialize the minimum
5223 * amount of fields so it can be used to schedule NAPI polls without
5224 * registering a full blown interface. This is to be used by drivers
5225 * that need to tie several hardware interfaces to a single NAPI
5226 * poll scheduler due to HW limitations.
5228 int init_dummy_netdev(struct net_device *dev)
5230 /* Clear everything. Note we don't initialize spinlocks
5231 * are they aren't supposed to be taken by any of the
5232 * NAPI code and this dummy netdev is supposed to be
5233 * only ever used for NAPI polls
5235 memset(dev, 0, sizeof(struct net_device));
5237 /* make sure we BUG if trying to hit standard
5238 * register/unregister code path
5240 dev->reg_state = NETREG_DUMMY;
5242 /* NAPI wants this */
5243 INIT_LIST_HEAD(&dev->napi_list);
5245 /* a dummy interface is started by default */
5246 set_bit(__LINK_STATE_PRESENT, &dev->state);
5247 set_bit(__LINK_STATE_START, &dev->state);
5249 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5250 * because users of this 'device' dont need to change
5251 * its refcount.
5254 return 0;
5256 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5260 * register_netdev - register a network device
5261 * @dev: device to register
5263 * Take a completed network device structure and add it to the kernel
5264 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5265 * chain. 0 is returned on success. A negative errno code is returned
5266 * on a failure to set up the device, or if the name is a duplicate.
5268 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5269 * and expands the device name if you passed a format string to
5270 * alloc_netdev.
5272 int register_netdev(struct net_device *dev)
5274 int err;
5276 rtnl_lock();
5279 * If the name is a format string the caller wants us to do a
5280 * name allocation.
5282 if (strchr(dev->name, '%')) {
5283 err = dev_alloc_name(dev, dev->name);
5284 if (err < 0)
5285 goto out;
5288 err = register_netdevice(dev);
5289 out:
5290 rtnl_unlock();
5291 return err;
5293 EXPORT_SYMBOL(register_netdev);
5295 int netdev_refcnt_read(const struct net_device *dev)
5297 int i, refcnt = 0;
5299 for_each_possible_cpu(i)
5300 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5301 return refcnt;
5303 EXPORT_SYMBOL(netdev_refcnt_read);
5306 * netdev_wait_allrefs - wait until all references are gone.
5308 * This is called when unregistering network devices.
5310 * Any protocol or device that holds a reference should register
5311 * for netdevice notification, and cleanup and put back the
5312 * reference if they receive an UNREGISTER event.
5313 * We can get stuck here if buggy protocols don't correctly
5314 * call dev_put.
5316 static void netdev_wait_allrefs(struct net_device *dev)
5318 unsigned long rebroadcast_time, warning_time;
5319 int refcnt;
5321 linkwatch_forget_dev(dev);
5323 rebroadcast_time = warning_time = jiffies;
5324 refcnt = netdev_refcnt_read(dev);
5326 while (refcnt != 0) {
5327 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5328 rtnl_lock();
5330 /* Rebroadcast unregister notification */
5331 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5332 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5333 * should have already handle it the first time */
5335 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5336 &dev->state)) {
5337 /* We must not have linkwatch events
5338 * pending on unregister. If this
5339 * happens, we simply run the queue
5340 * unscheduled, resulting in a noop
5341 * for this device.
5343 linkwatch_run_queue();
5346 __rtnl_unlock();
5348 rebroadcast_time = jiffies;
5351 msleep(250);
5353 refcnt = netdev_refcnt_read(dev);
5355 if (time_after(jiffies, warning_time + 10 * HZ)) {
5356 printk(KERN_EMERG "unregister_netdevice: "
5357 "waiting for %s to become free. Usage "
5358 "count = %d\n",
5359 dev->name, refcnt);
5360 warning_time = jiffies;
5365 /* The sequence is:
5367 * rtnl_lock();
5368 * ...
5369 * register_netdevice(x1);
5370 * register_netdevice(x2);
5371 * ...
5372 * unregister_netdevice(y1);
5373 * unregister_netdevice(y2);
5374 * ...
5375 * rtnl_unlock();
5376 * free_netdev(y1);
5377 * free_netdev(y2);
5379 * We are invoked by rtnl_unlock().
5380 * This allows us to deal with problems:
5381 * 1) We can delete sysfs objects which invoke hotplug
5382 * without deadlocking with linkwatch via keventd.
5383 * 2) Since we run with the RTNL semaphore not held, we can sleep
5384 * safely in order to wait for the netdev refcnt to drop to zero.
5386 * We must not return until all unregister events added during
5387 * the interval the lock was held have been completed.
5389 void netdev_run_todo(void)
5391 struct list_head list;
5393 /* Snapshot list, allow later requests */
5394 list_replace_init(&net_todo_list, &list);
5396 __rtnl_unlock();
5398 while (!list_empty(&list)) {
5399 struct net_device *dev
5400 = list_first_entry(&list, struct net_device, todo_list);
5401 list_del(&dev->todo_list);
5403 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5404 printk(KERN_ERR "network todo '%s' but state %d\n",
5405 dev->name, dev->reg_state);
5406 dump_stack();
5407 continue;
5410 dev->reg_state = NETREG_UNREGISTERED;
5412 on_each_cpu(flush_backlog, dev, 1);
5414 netdev_wait_allrefs(dev);
5416 /* paranoia */
5417 BUG_ON(netdev_refcnt_read(dev));
5418 WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5419 WARN_ON(dev->ip6_ptr);
5420 WARN_ON(dev->dn_ptr);
5422 if (dev->destructor)
5423 dev->destructor(dev);
5425 /* Free network device */
5426 kobject_put(&dev->dev.kobj);
5431 * dev_txq_stats_fold - fold tx_queues stats
5432 * @dev: device to get statistics from
5433 * @stats: struct rtnl_link_stats64 to hold results
5435 void dev_txq_stats_fold(const struct net_device *dev,
5436 struct rtnl_link_stats64 *stats)
5438 u64 tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5439 unsigned int i;
5440 struct netdev_queue *txq;
5442 for (i = 0; i < dev->num_tx_queues; i++) {
5443 txq = netdev_get_tx_queue(dev, i);
5444 spin_lock_bh(&txq->_xmit_lock);
5445 tx_bytes += txq->tx_bytes;
5446 tx_packets += txq->tx_packets;
5447 tx_dropped += txq->tx_dropped;
5448 spin_unlock_bh(&txq->_xmit_lock);
5450 if (tx_bytes || tx_packets || tx_dropped) {
5451 stats->tx_bytes = tx_bytes;
5452 stats->tx_packets = tx_packets;
5453 stats->tx_dropped = tx_dropped;
5456 EXPORT_SYMBOL(dev_txq_stats_fold);
5458 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5459 * fields in the same order, with only the type differing.
5461 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5462 const struct net_device_stats *netdev_stats)
5464 #if BITS_PER_LONG == 64
5465 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5466 memcpy(stats64, netdev_stats, sizeof(*stats64));
5467 #else
5468 size_t i, n = sizeof(*stats64) / sizeof(u64);
5469 const unsigned long *src = (const unsigned long *)netdev_stats;
5470 u64 *dst = (u64 *)stats64;
5472 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5473 sizeof(*stats64) / sizeof(u64));
5474 for (i = 0; i < n; i++)
5475 dst[i] = src[i];
5476 #endif
5480 * dev_get_stats - get network device statistics
5481 * @dev: device to get statistics from
5482 * @storage: place to store stats
5484 * Get network statistics from device. Return @storage.
5485 * The device driver may provide its own method by setting
5486 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5487 * otherwise the internal statistics structure is used.
5489 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5490 struct rtnl_link_stats64 *storage)
5492 const struct net_device_ops *ops = dev->netdev_ops;
5494 if (ops->ndo_get_stats64) {
5495 memset(storage, 0, sizeof(*storage));
5496 ops->ndo_get_stats64(dev, storage);
5497 } else if (ops->ndo_get_stats) {
5498 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5499 } else {
5500 netdev_stats_to_stats64(storage, &dev->stats);
5501 dev_txq_stats_fold(dev, storage);
5503 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5504 return storage;
5506 EXPORT_SYMBOL(dev_get_stats);
5508 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5510 struct netdev_queue *queue = dev_ingress_queue(dev);
5512 #ifdef CONFIG_NET_CLS_ACT
5513 if (queue)
5514 return queue;
5515 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5516 if (!queue)
5517 return NULL;
5518 netdev_init_one_queue(dev, queue, NULL);
5519 queue->qdisc = &noop_qdisc;
5520 queue->qdisc_sleeping = &noop_qdisc;
5521 rcu_assign_pointer(dev->ingress_queue, queue);
5522 #endif
5523 return queue;
5527 * alloc_netdev_mq - allocate network device
5528 * @sizeof_priv: size of private data to allocate space for
5529 * @name: device name format string
5530 * @setup: callback to initialize device
5531 * @queue_count: the number of subqueues to allocate
5533 * Allocates a struct net_device with private data area for driver use
5534 * and performs basic initialization. Also allocates subquue structs
5535 * for each queue on the device at the end of the netdevice.
5537 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5538 void (*setup)(struct net_device *), unsigned int queue_count)
5540 struct net_device *dev;
5541 size_t alloc_size;
5542 struct net_device *p;
5544 BUG_ON(strlen(name) >= sizeof(dev->name));
5546 if (queue_count < 1) {
5547 pr_err("alloc_netdev: Unable to allocate device "
5548 "with zero queues.\n");
5549 return NULL;
5552 alloc_size = sizeof(struct net_device);
5553 if (sizeof_priv) {
5554 /* ensure 32-byte alignment of private area */
5555 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5556 alloc_size += sizeof_priv;
5558 /* ensure 32-byte alignment of whole construct */
5559 alloc_size += NETDEV_ALIGN - 1;
5561 p = kzalloc(alloc_size, GFP_KERNEL);
5562 if (!p) {
5563 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5564 return NULL;
5567 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5568 dev->padded = (char *)dev - (char *)p;
5570 dev->pcpu_refcnt = alloc_percpu(int);
5571 if (!dev->pcpu_refcnt)
5572 goto free_p;
5574 if (dev_addr_init(dev))
5575 goto free_pcpu;
5577 dev_mc_init(dev);
5578 dev_uc_init(dev);
5580 dev_net_set(dev, &init_net);
5582 dev->num_tx_queues = queue_count;
5583 dev->real_num_tx_queues = queue_count;
5585 #ifdef CONFIG_RPS
5586 dev->num_rx_queues = queue_count;
5587 dev->real_num_rx_queues = queue_count;
5588 #endif
5590 dev->gso_max_size = GSO_MAX_SIZE;
5592 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5593 dev->ethtool_ntuple_list.count = 0;
5594 INIT_LIST_HEAD(&dev->napi_list);
5595 INIT_LIST_HEAD(&dev->unreg_list);
5596 INIT_LIST_HEAD(&dev->link_watch_list);
5597 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5598 setup(dev);
5599 strcpy(dev->name, name);
5600 return dev;
5602 free_pcpu:
5603 free_percpu(dev->pcpu_refcnt);
5604 free_p:
5605 kfree(p);
5606 return NULL;
5608 EXPORT_SYMBOL(alloc_netdev_mq);
5611 * free_netdev - free network device
5612 * @dev: device
5614 * This function does the last stage of destroying an allocated device
5615 * interface. The reference to the device object is released.
5616 * If this is the last reference then it will be freed.
5618 void free_netdev(struct net_device *dev)
5620 struct napi_struct *p, *n;
5622 release_net(dev_net(dev));
5624 kfree(dev->_tx);
5626 kfree(rcu_dereference_raw(dev->ingress_queue));
5628 /* Flush device addresses */
5629 dev_addr_flush(dev);
5631 /* Clear ethtool n-tuple list */
5632 ethtool_ntuple_flush(dev);
5634 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5635 netif_napi_del(p);
5637 free_percpu(dev->pcpu_refcnt);
5638 dev->pcpu_refcnt = NULL;
5640 /* Compatibility with error handling in drivers */
5641 if (dev->reg_state == NETREG_UNINITIALIZED) {
5642 kfree((char *)dev - dev->padded);
5643 return;
5646 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5647 dev->reg_state = NETREG_RELEASED;
5649 /* will free via device release */
5650 put_device(&dev->dev);
5652 EXPORT_SYMBOL(free_netdev);
5655 * synchronize_net - Synchronize with packet receive processing
5657 * Wait for packets currently being received to be done.
5658 * Does not block later packets from starting.
5660 void synchronize_net(void)
5662 might_sleep();
5663 synchronize_rcu();
5665 EXPORT_SYMBOL(synchronize_net);
5668 * unregister_netdevice_queue - remove device from the kernel
5669 * @dev: device
5670 * @head: list
5672 * This function shuts down a device interface and removes it
5673 * from the kernel tables.
5674 * If head not NULL, device is queued to be unregistered later.
5676 * Callers must hold the rtnl semaphore. You may want
5677 * unregister_netdev() instead of this.
5680 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5682 ASSERT_RTNL();
5684 if (head) {
5685 list_move_tail(&dev->unreg_list, head);
5686 } else {
5687 rollback_registered(dev);
5688 /* Finish processing unregister after unlock */
5689 net_set_todo(dev);
5692 EXPORT_SYMBOL(unregister_netdevice_queue);
5695 * unregister_netdevice_many - unregister many devices
5696 * @head: list of devices
5698 void unregister_netdevice_many(struct list_head *head)
5700 struct net_device *dev;
5702 if (!list_empty(head)) {
5703 rollback_registered_many(head);
5704 list_for_each_entry(dev, head, unreg_list)
5705 net_set_todo(dev);
5708 EXPORT_SYMBOL(unregister_netdevice_many);
5711 * unregister_netdev - remove device from the kernel
5712 * @dev: device
5714 * This function shuts down a device interface and removes it
5715 * from the kernel tables.
5717 * This is just a wrapper for unregister_netdevice that takes
5718 * the rtnl semaphore. In general you want to use this and not
5719 * unregister_netdevice.
5721 void unregister_netdev(struct net_device *dev)
5723 rtnl_lock();
5724 unregister_netdevice(dev);
5725 rtnl_unlock();
5727 EXPORT_SYMBOL(unregister_netdev);
5730 * dev_change_net_namespace - move device to different nethost namespace
5731 * @dev: device
5732 * @net: network namespace
5733 * @pat: If not NULL name pattern to try if the current device name
5734 * is already taken in the destination network namespace.
5736 * This function shuts down a device interface and moves it
5737 * to a new network namespace. On success 0 is returned, on
5738 * a failure a netagive errno code is returned.
5740 * Callers must hold the rtnl semaphore.
5743 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5745 int err;
5747 ASSERT_RTNL();
5749 /* Don't allow namespace local devices to be moved. */
5750 err = -EINVAL;
5751 if (dev->features & NETIF_F_NETNS_LOCAL)
5752 goto out;
5754 /* Ensure the device has been registrered */
5755 err = -EINVAL;
5756 if (dev->reg_state != NETREG_REGISTERED)
5757 goto out;
5759 /* Get out if there is nothing todo */
5760 err = 0;
5761 if (net_eq(dev_net(dev), net))
5762 goto out;
5764 /* Pick the destination device name, and ensure
5765 * we can use it in the destination network namespace.
5767 err = -EEXIST;
5768 if (__dev_get_by_name(net, dev->name)) {
5769 /* We get here if we can't use the current device name */
5770 if (!pat)
5771 goto out;
5772 if (dev_get_valid_name(dev, pat, 1))
5773 goto out;
5777 * And now a mini version of register_netdevice unregister_netdevice.
5780 /* If device is running close it first. */
5781 dev_close(dev);
5783 /* And unlink it from device chain */
5784 err = -ENODEV;
5785 unlist_netdevice(dev);
5787 synchronize_net();
5789 /* Shutdown queueing discipline. */
5790 dev_shutdown(dev);
5792 /* Notify protocols, that we are about to destroy
5793 this device. They should clean all the things.
5795 Note that dev->reg_state stays at NETREG_REGISTERED.
5796 This is wanted because this way 8021q and macvlan know
5797 the device is just moving and can keep their slaves up.
5799 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5800 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5803 * Flush the unicast and multicast chains
5805 dev_uc_flush(dev);
5806 dev_mc_flush(dev);
5808 /* Actually switch the network namespace */
5809 dev_net_set(dev, net);
5811 /* If there is an ifindex conflict assign a new one */
5812 if (__dev_get_by_index(net, dev->ifindex)) {
5813 int iflink = (dev->iflink == dev->ifindex);
5814 dev->ifindex = dev_new_index(net);
5815 if (iflink)
5816 dev->iflink = dev->ifindex;
5819 /* Fixup kobjects */
5820 err = device_rename(&dev->dev, dev->name);
5821 WARN_ON(err);
5823 /* Add the device back in the hashes */
5824 list_netdevice(dev);
5826 /* Notify protocols, that a new device appeared. */
5827 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5830 * Prevent userspace races by waiting until the network
5831 * device is fully setup before sending notifications.
5833 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5835 synchronize_net();
5836 err = 0;
5837 out:
5838 return err;
5840 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5842 static int dev_cpu_callback(struct notifier_block *nfb,
5843 unsigned long action,
5844 void *ocpu)
5846 struct sk_buff **list_skb;
5847 struct sk_buff *skb;
5848 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5849 struct softnet_data *sd, *oldsd;
5851 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5852 return NOTIFY_OK;
5854 local_irq_disable();
5855 cpu = smp_processor_id();
5856 sd = &per_cpu(softnet_data, cpu);
5857 oldsd = &per_cpu(softnet_data, oldcpu);
5859 /* Find end of our completion_queue. */
5860 list_skb = &sd->completion_queue;
5861 while (*list_skb)
5862 list_skb = &(*list_skb)->next;
5863 /* Append completion queue from offline CPU. */
5864 *list_skb = oldsd->completion_queue;
5865 oldsd->completion_queue = NULL;
5867 /* Append output queue from offline CPU. */
5868 if (oldsd->output_queue) {
5869 *sd->output_queue_tailp = oldsd->output_queue;
5870 sd->output_queue_tailp = oldsd->output_queue_tailp;
5871 oldsd->output_queue = NULL;
5872 oldsd->output_queue_tailp = &oldsd->output_queue;
5875 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5876 local_irq_enable();
5878 /* Process offline CPU's input_pkt_queue */
5879 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5880 netif_rx(skb);
5881 input_queue_head_incr(oldsd);
5883 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5884 netif_rx(skb);
5885 input_queue_head_incr(oldsd);
5888 return NOTIFY_OK;
5893 * netdev_increment_features - increment feature set by one
5894 * @all: current feature set
5895 * @one: new feature set
5896 * @mask: mask feature set
5898 * Computes a new feature set after adding a device with feature set
5899 * @one to the master device with current feature set @all. Will not
5900 * enable anything that is off in @mask. Returns the new feature set.
5902 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5903 unsigned long mask)
5905 /* If device needs checksumming, downgrade to it. */
5906 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5907 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5908 else if (mask & NETIF_F_ALL_CSUM) {
5909 /* If one device supports v4/v6 checksumming, set for all. */
5910 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5911 !(all & NETIF_F_GEN_CSUM)) {
5912 all &= ~NETIF_F_ALL_CSUM;
5913 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5916 /* If one device supports hw checksumming, set for all. */
5917 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5918 all &= ~NETIF_F_ALL_CSUM;
5919 all |= NETIF_F_HW_CSUM;
5923 one |= NETIF_F_ALL_CSUM;
5925 one |= all & NETIF_F_ONE_FOR_ALL;
5926 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5927 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5929 return all;
5931 EXPORT_SYMBOL(netdev_increment_features);
5933 static struct hlist_head *netdev_create_hash(void)
5935 int i;
5936 struct hlist_head *hash;
5938 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5939 if (hash != NULL)
5940 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5941 INIT_HLIST_HEAD(&hash[i]);
5943 return hash;
5946 /* Initialize per network namespace state */
5947 static int __net_init netdev_init(struct net *net)
5949 INIT_LIST_HEAD(&net->dev_base_head);
5951 net->dev_name_head = netdev_create_hash();
5952 if (net->dev_name_head == NULL)
5953 goto err_name;
5955 net->dev_index_head = netdev_create_hash();
5956 if (net->dev_index_head == NULL)
5957 goto err_idx;
5959 return 0;
5961 err_idx:
5962 kfree(net->dev_name_head);
5963 err_name:
5964 return -ENOMEM;
5968 * netdev_drivername - network driver for the device
5969 * @dev: network device
5970 * @buffer: buffer for resulting name
5971 * @len: size of buffer
5973 * Determine network driver for device.
5975 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5977 const struct device_driver *driver;
5978 const struct device *parent;
5980 if (len <= 0 || !buffer)
5981 return buffer;
5982 buffer[0] = 0;
5984 parent = dev->dev.parent;
5986 if (!parent)
5987 return buffer;
5989 driver = parent->driver;
5990 if (driver && driver->name)
5991 strlcpy(buffer, driver->name, len);
5992 return buffer;
5995 static int __netdev_printk(const char *level, const struct net_device *dev,
5996 struct va_format *vaf)
5998 int r;
6000 if (dev && dev->dev.parent)
6001 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6002 netdev_name(dev), vaf);
6003 else if (dev)
6004 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6005 else
6006 r = printk("%s(NULL net_device): %pV", level, vaf);
6008 return r;
6011 int netdev_printk(const char *level, const struct net_device *dev,
6012 const char *format, ...)
6014 struct va_format vaf;
6015 va_list args;
6016 int r;
6018 va_start(args, format);
6020 vaf.fmt = format;
6021 vaf.va = &args;
6023 r = __netdev_printk(level, dev, &vaf);
6024 va_end(args);
6026 return r;
6028 EXPORT_SYMBOL(netdev_printk);
6030 #define define_netdev_printk_level(func, level) \
6031 int func(const struct net_device *dev, const char *fmt, ...) \
6033 int r; \
6034 struct va_format vaf; \
6035 va_list args; \
6037 va_start(args, fmt); \
6039 vaf.fmt = fmt; \
6040 vaf.va = &args; \
6042 r = __netdev_printk(level, dev, &vaf); \
6043 va_end(args); \
6045 return r; \
6047 EXPORT_SYMBOL(func);
6049 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6050 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6051 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6052 define_netdev_printk_level(netdev_err, KERN_ERR);
6053 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6054 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6055 define_netdev_printk_level(netdev_info, KERN_INFO);
6057 static void __net_exit netdev_exit(struct net *net)
6059 kfree(net->dev_name_head);
6060 kfree(net->dev_index_head);
6063 static struct pernet_operations __net_initdata netdev_net_ops = {
6064 .init = netdev_init,
6065 .exit = netdev_exit,
6068 static void __net_exit default_device_exit(struct net *net)
6070 struct net_device *dev, *aux;
6072 * Push all migratable network devices back to the
6073 * initial network namespace
6075 rtnl_lock();
6076 for_each_netdev_safe(net, dev, aux) {
6077 int err;
6078 char fb_name[IFNAMSIZ];
6080 /* Ignore unmoveable devices (i.e. loopback) */
6081 if (dev->features & NETIF_F_NETNS_LOCAL)
6082 continue;
6084 /* Leave virtual devices for the generic cleanup */
6085 if (dev->rtnl_link_ops)
6086 continue;
6088 /* Push remaing network devices to init_net */
6089 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6090 err = dev_change_net_namespace(dev, &init_net, fb_name);
6091 if (err) {
6092 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6093 __func__, dev->name, err);
6094 BUG();
6097 rtnl_unlock();
6100 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6102 /* At exit all network devices most be removed from a network
6103 * namespace. Do this in the reverse order of registeration.
6104 * Do this across as many network namespaces as possible to
6105 * improve batching efficiency.
6107 struct net_device *dev;
6108 struct net *net;
6109 LIST_HEAD(dev_kill_list);
6111 rtnl_lock();
6112 list_for_each_entry(net, net_list, exit_list) {
6113 for_each_netdev_reverse(net, dev) {
6114 if (dev->rtnl_link_ops)
6115 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6116 else
6117 unregister_netdevice_queue(dev, &dev_kill_list);
6120 unregister_netdevice_many(&dev_kill_list);
6121 rtnl_unlock();
6124 static struct pernet_operations __net_initdata default_device_ops = {
6125 .exit = default_device_exit,
6126 .exit_batch = default_device_exit_batch,
6130 * Initialize the DEV module. At boot time this walks the device list and
6131 * unhooks any devices that fail to initialise (normally hardware not
6132 * present) and leaves us with a valid list of present and active devices.
6137 * This is called single threaded during boot, so no need
6138 * to take the rtnl semaphore.
6140 static int __init net_dev_init(void)
6142 int i, rc = -ENOMEM;
6144 BUG_ON(!dev_boot_phase);
6146 if (dev_proc_init())
6147 goto out;
6149 if (netdev_kobject_init())
6150 goto out;
6152 INIT_LIST_HEAD(&ptype_all);
6153 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6154 INIT_LIST_HEAD(&ptype_base[i]);
6156 if (register_pernet_subsys(&netdev_net_ops))
6157 goto out;
6160 * Initialise the packet receive queues.
6163 for_each_possible_cpu(i) {
6164 struct softnet_data *sd = &per_cpu(softnet_data, i);
6166 memset(sd, 0, sizeof(*sd));
6167 skb_queue_head_init(&sd->input_pkt_queue);
6168 skb_queue_head_init(&sd->process_queue);
6169 sd->completion_queue = NULL;
6170 INIT_LIST_HEAD(&sd->poll_list);
6171 sd->output_queue = NULL;
6172 sd->output_queue_tailp = &sd->output_queue;
6173 #ifdef CONFIG_RPS
6174 sd->csd.func = rps_trigger_softirq;
6175 sd->csd.info = sd;
6176 sd->csd.flags = 0;
6177 sd->cpu = i;
6178 #endif
6180 sd->backlog.poll = process_backlog;
6181 sd->backlog.weight = weight_p;
6182 sd->backlog.gro_list = NULL;
6183 sd->backlog.gro_count = 0;
6186 dev_boot_phase = 0;
6188 /* The loopback device is special if any other network devices
6189 * is present in a network namespace the loopback device must
6190 * be present. Since we now dynamically allocate and free the
6191 * loopback device ensure this invariant is maintained by
6192 * keeping the loopback device as the first device on the
6193 * list of network devices. Ensuring the loopback devices
6194 * is the first device that appears and the last network device
6195 * that disappears.
6197 if (register_pernet_device(&loopback_net_ops))
6198 goto out;
6200 if (register_pernet_device(&default_device_ops))
6201 goto out;
6203 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6204 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6206 hotcpu_notifier(dev_cpu_callback, 0);
6207 dst_init();
6208 dev_mcast_init();
6209 rc = 0;
6210 out:
6211 return rc;
6214 subsys_initcall(net_dev_init);
6216 static int __init initialize_hashrnd(void)
6218 get_random_bytes(&hashrnd, sizeof(hashrnd));
6219 return 0;
6222 late_initcall_sync(initialize_hashrnd);