2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/if_tunnel.h>
137 #include <linux/if_pppox.h>
138 #include <linux/ppp_defs.h>
139 #include <linux/net_tstamp.h>
140 #include <linux/jump_label.h>
142 #include "net-sysfs.h"
144 /* Instead of increasing this, you should create a hash table. */
145 #define MAX_GRO_SKBS 8
147 /* This should be increased if a protocol with a bigger head is added. */
148 #define GRO_MAX_HEAD (MAX_HEADER + 128)
151 * The list of packet types we will receive (as opposed to discard)
152 * and the routines to invoke.
154 * Why 16. Because with 16 the only overlap we get on a hash of the
155 * low nibble of the protocol value is RARP/SNAP/X.25.
157 * NOTE: That is no longer true with the addition of VLAN tags. Not
158 * sure which should go first, but I bet it won't make much
159 * difference if we are running VLANs. The good news is that
160 * this protocol won't be in the list unless compiled in, so
161 * the average user (w/out VLANs) will not be adversely affected.
178 #define PTYPE_HASH_SIZE (16)
179 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
181 static DEFINE_SPINLOCK(ptype_lock);
182 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
183 static struct list_head ptype_all __read_mostly; /* Taps */
186 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
189 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
191 * Writers must hold the rtnl semaphore while they loop through the
192 * dev_base_head list, and hold dev_base_lock for writing when they do the
193 * actual updates. This allows pure readers to access the list even
194 * while a writer is preparing to update it.
196 * To put it another way, dev_base_lock is held for writing only to
197 * protect against pure readers; the rtnl semaphore provides the
198 * protection against other writers.
200 * See, for example usages, register_netdevice() and
201 * unregister_netdevice(), which must be called with the rtnl
204 DEFINE_RWLOCK(dev_base_lock);
205 EXPORT_SYMBOL(dev_base_lock);
207 static inline void dev_base_seq_inc(struct net *net)
209 while (++net->dev_base_seq == 0);
212 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
214 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
215 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
218 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
220 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
223 static inline void rps_lock(struct softnet_data *sd)
226 spin_lock(&sd->input_pkt_queue.lock);
230 static inline void rps_unlock(struct softnet_data *sd)
233 spin_unlock(&sd->input_pkt_queue.lock);
237 /* Device list insertion */
238 static int list_netdevice(struct net_device *dev)
240 struct net *net = dev_net(dev);
244 write_lock_bh(&dev_base_lock);
245 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
246 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
247 hlist_add_head_rcu(&dev->index_hlist,
248 dev_index_hash(net, dev->ifindex));
249 write_unlock_bh(&dev_base_lock);
251 dev_base_seq_inc(net);
256 /* Device list removal
257 * caller must respect a RCU grace period before freeing/reusing dev
259 static void unlist_netdevice(struct net_device *dev)
263 /* Unlink dev from the device chain */
264 write_lock_bh(&dev_base_lock);
265 list_del_rcu(&dev->dev_list);
266 hlist_del_rcu(&dev->name_hlist);
267 hlist_del_rcu(&dev->index_hlist);
268 write_unlock_bh(&dev_base_lock);
270 dev_base_seq_inc(dev_net(dev));
277 static RAW_NOTIFIER_HEAD(netdev_chain);
280 * Device drivers call our routines to queue packets here. We empty the
281 * queue in the local softnet handler.
284 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
285 EXPORT_PER_CPU_SYMBOL(softnet_data);
287 #ifdef CONFIG_LOCKDEP
289 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
290 * according to dev->type
292 static const unsigned short netdev_lock_type[] =
293 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
294 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
295 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
296 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
297 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
298 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
299 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
300 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
301 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
302 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
303 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
304 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
305 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
306 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
307 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
308 ARPHRD_VOID, ARPHRD_NONE};
310 static const char *const netdev_lock_name[] =
311 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
312 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
313 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
314 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
315 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
316 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
317 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
318 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
319 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
320 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
321 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
322 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
323 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
324 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
325 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
326 "_xmit_VOID", "_xmit_NONE"};
328 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
329 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
331 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
335 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
336 if (netdev_lock_type[i] == dev_type)
338 /* the last key is used by default */
339 return ARRAY_SIZE(netdev_lock_type) - 1;
342 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
343 unsigned short dev_type)
347 i = netdev_lock_pos(dev_type);
348 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
349 netdev_lock_name[i]);
352 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
356 i = netdev_lock_pos(dev->type);
357 lockdep_set_class_and_name(&dev->addr_list_lock,
358 &netdev_addr_lock_key[i],
359 netdev_lock_name[i]);
362 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
363 unsigned short dev_type)
366 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
371 /*******************************************************************************
373 Protocol management and registration routines
375 *******************************************************************************/
378 * Add a protocol ID to the list. Now that the input handler is
379 * smarter we can dispense with all the messy stuff that used to be
382 * BEWARE!!! Protocol handlers, mangling input packets,
383 * MUST BE last in hash buckets and checking protocol handlers
384 * MUST start from promiscuous ptype_all chain in net_bh.
385 * It is true now, do not change it.
386 * Explanation follows: if protocol handler, mangling packet, will
387 * be the first on list, it is not able to sense, that packet
388 * is cloned and should be copied-on-write, so that it will
389 * change it and subsequent readers will get broken packet.
393 static inline struct list_head *ptype_head(const struct packet_type *pt)
395 if (pt->type == htons(ETH_P_ALL))
398 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
402 * dev_add_pack - add packet handler
403 * @pt: packet type declaration
405 * Add a protocol handler to the networking stack. The passed &packet_type
406 * is linked into kernel lists and may not be freed until it has been
407 * removed from the kernel lists.
409 * This call does not sleep therefore it can not
410 * guarantee all CPU's that are in middle of receiving packets
411 * will see the new packet type (until the next received packet).
414 void dev_add_pack(struct packet_type *pt)
416 struct list_head *head = ptype_head(pt);
418 spin_lock(&ptype_lock);
419 list_add_rcu(&pt->list, head);
420 spin_unlock(&ptype_lock);
422 EXPORT_SYMBOL(dev_add_pack);
425 * __dev_remove_pack - remove packet handler
426 * @pt: packet type declaration
428 * Remove a protocol handler that was previously added to the kernel
429 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
430 * from the kernel lists and can be freed or reused once this function
433 * The packet type might still be in use by receivers
434 * and must not be freed until after all the CPU's have gone
435 * through a quiescent state.
437 void __dev_remove_pack(struct packet_type *pt)
439 struct list_head *head = ptype_head(pt);
440 struct packet_type *pt1;
442 spin_lock(&ptype_lock);
444 list_for_each_entry(pt1, head, list) {
446 list_del_rcu(&pt->list);
451 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
453 spin_unlock(&ptype_lock);
455 EXPORT_SYMBOL(__dev_remove_pack);
458 * dev_remove_pack - remove packet handler
459 * @pt: packet type declaration
461 * Remove a protocol handler that was previously added to the kernel
462 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
463 * from the kernel lists and can be freed or reused once this function
466 * This call sleeps to guarantee that no CPU is looking at the packet
469 void dev_remove_pack(struct packet_type *pt)
471 __dev_remove_pack(pt);
475 EXPORT_SYMBOL(dev_remove_pack);
477 /******************************************************************************
479 Device Boot-time Settings Routines
481 *******************************************************************************/
483 /* Boot time configuration table */
484 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
487 * netdev_boot_setup_add - add new setup entry
488 * @name: name of the device
489 * @map: configured settings for the device
491 * Adds new setup entry to the dev_boot_setup list. The function
492 * returns 0 on error and 1 on success. This is a generic routine to
495 static int netdev_boot_setup_add(char *name, struct ifmap *map)
497 struct netdev_boot_setup *s;
501 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
502 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
503 memset(s[i].name, 0, sizeof(s[i].name));
504 strlcpy(s[i].name, name, IFNAMSIZ);
505 memcpy(&s[i].map, map, sizeof(s[i].map));
510 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
514 * netdev_boot_setup_check - check boot time settings
515 * @dev: the netdevice
517 * Check boot time settings for the device.
518 * The found settings are set for the device to be used
519 * later in the device probing.
520 * Returns 0 if no settings found, 1 if they are.
522 int netdev_boot_setup_check(struct net_device *dev)
524 struct netdev_boot_setup *s = dev_boot_setup;
527 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
528 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
529 !strcmp(dev->name, s[i].name)) {
530 dev->irq = s[i].map.irq;
531 dev->base_addr = s[i].map.base_addr;
532 dev->mem_start = s[i].map.mem_start;
533 dev->mem_end = s[i].map.mem_end;
539 EXPORT_SYMBOL(netdev_boot_setup_check);
543 * netdev_boot_base - get address from boot time settings
544 * @prefix: prefix for network device
545 * @unit: id for network device
547 * Check boot time settings for the base address of device.
548 * The found settings are set for the device to be used
549 * later in the device probing.
550 * Returns 0 if no settings found.
552 unsigned long netdev_boot_base(const char *prefix, int unit)
554 const struct netdev_boot_setup *s = dev_boot_setup;
558 sprintf(name, "%s%d", prefix, unit);
561 * If device already registered then return base of 1
562 * to indicate not to probe for this interface
564 if (__dev_get_by_name(&init_net, name))
567 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
568 if (!strcmp(name, s[i].name))
569 return s[i].map.base_addr;
574 * Saves at boot time configured settings for any netdevice.
576 int __init netdev_boot_setup(char *str)
581 str = get_options(str, ARRAY_SIZE(ints), ints);
586 memset(&map, 0, sizeof(map));
590 map.base_addr = ints[2];
592 map.mem_start = ints[3];
594 map.mem_end = ints[4];
596 /* Add new entry to the list */
597 return netdev_boot_setup_add(str, &map);
600 __setup("netdev=", netdev_boot_setup);
602 /*******************************************************************************
604 Device Interface Subroutines
606 *******************************************************************************/
609 * __dev_get_by_name - find a device by its name
610 * @net: the applicable net namespace
611 * @name: name to find
613 * Find an interface by name. Must be called under RTNL semaphore
614 * or @dev_base_lock. If the name is found a pointer to the device
615 * is returned. If the name is not found then %NULL is returned. The
616 * reference counters are not incremented so the caller must be
617 * careful with locks.
620 struct net_device *__dev_get_by_name(struct net *net, const char *name)
622 struct hlist_node *p;
623 struct net_device *dev;
624 struct hlist_head *head = dev_name_hash(net, name);
626 hlist_for_each_entry(dev, p, head, name_hlist)
627 if (!strncmp(dev->name, name, IFNAMSIZ))
632 EXPORT_SYMBOL(__dev_get_by_name);
635 * dev_get_by_name_rcu - find a device by its name
636 * @net: the applicable net namespace
637 * @name: name to find
639 * Find an interface by name.
640 * If the name is found a pointer to the device is returned.
641 * If the name is not found then %NULL is returned.
642 * The reference counters are not incremented so the caller must be
643 * careful with locks. The caller must hold RCU lock.
646 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
648 struct hlist_node *p;
649 struct net_device *dev;
650 struct hlist_head *head = dev_name_hash(net, name);
652 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
653 if (!strncmp(dev->name, name, IFNAMSIZ))
658 EXPORT_SYMBOL(dev_get_by_name_rcu);
661 * dev_get_by_name - find a device by its name
662 * @net: the applicable net namespace
663 * @name: name to find
665 * Find an interface by name. This can be called from any
666 * context and does its own locking. The returned handle has
667 * the usage count incremented and the caller must use dev_put() to
668 * release it when it is no longer needed. %NULL is returned if no
669 * matching device is found.
672 struct net_device *dev_get_by_name(struct net *net, const char *name)
674 struct net_device *dev;
677 dev = dev_get_by_name_rcu(net, name);
683 EXPORT_SYMBOL(dev_get_by_name);
686 * __dev_get_by_index - find a device by its ifindex
687 * @net: the applicable net namespace
688 * @ifindex: index of device
690 * Search for an interface by index. Returns %NULL if the device
691 * is not found or a pointer to the device. The device has not
692 * had its reference counter increased so the caller must be careful
693 * about locking. The caller must hold either the RTNL semaphore
697 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
699 struct hlist_node *p;
700 struct net_device *dev;
701 struct hlist_head *head = dev_index_hash(net, ifindex);
703 hlist_for_each_entry(dev, p, head, index_hlist)
704 if (dev->ifindex == ifindex)
709 EXPORT_SYMBOL(__dev_get_by_index);
712 * dev_get_by_index_rcu - find a device by its ifindex
713 * @net: the applicable net namespace
714 * @ifindex: index of device
716 * Search for an interface by index. Returns %NULL if the device
717 * is not found or a pointer to the device. The device has not
718 * had its reference counter increased so the caller must be careful
719 * about locking. The caller must hold RCU lock.
722 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
724 struct hlist_node *p;
725 struct net_device *dev;
726 struct hlist_head *head = dev_index_hash(net, ifindex);
728 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
729 if (dev->ifindex == ifindex)
734 EXPORT_SYMBOL(dev_get_by_index_rcu);
738 * dev_get_by_index - find a device by its ifindex
739 * @net: the applicable net namespace
740 * @ifindex: index of device
742 * Search for an interface by index. Returns NULL if the device
743 * is not found or a pointer to the device. The device returned has
744 * had a reference added and the pointer is safe until the user calls
745 * dev_put to indicate they have finished with it.
748 struct net_device *dev_get_by_index(struct net *net, int ifindex)
750 struct net_device *dev;
753 dev = dev_get_by_index_rcu(net, ifindex);
759 EXPORT_SYMBOL(dev_get_by_index);
762 * dev_getbyhwaddr_rcu - find a device by its hardware address
763 * @net: the applicable net namespace
764 * @type: media type of device
765 * @ha: hardware address
767 * Search for an interface by MAC address. Returns NULL if the device
768 * is not found or a pointer to the device.
769 * The caller must hold RCU or RTNL.
770 * The returned device has not had its ref count increased
771 * and the caller must therefore be careful about locking
775 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
778 struct net_device *dev;
780 for_each_netdev_rcu(net, dev)
781 if (dev->type == type &&
782 !memcmp(dev->dev_addr, ha, dev->addr_len))
787 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
789 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
791 struct net_device *dev;
794 for_each_netdev(net, dev)
795 if (dev->type == type)
800 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
802 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
804 struct net_device *dev, *ret = NULL;
807 for_each_netdev_rcu(net, dev)
808 if (dev->type == type) {
816 EXPORT_SYMBOL(dev_getfirstbyhwtype);
819 * dev_get_by_flags_rcu - find any device with given flags
820 * @net: the applicable net namespace
821 * @if_flags: IFF_* values
822 * @mask: bitmask of bits in if_flags to check
824 * Search for any interface with the given flags. Returns NULL if a device
825 * is not found or a pointer to the device. Must be called inside
826 * rcu_read_lock(), and result refcount is unchanged.
829 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
832 struct net_device *dev, *ret;
835 for_each_netdev_rcu(net, dev) {
836 if (((dev->flags ^ if_flags) & mask) == 0) {
843 EXPORT_SYMBOL(dev_get_by_flags_rcu);
846 * dev_valid_name - check if name is okay for network device
849 * Network device names need to be valid file names to
850 * to allow sysfs to work. We also disallow any kind of
853 int dev_valid_name(const char *name)
857 if (strlen(name) >= IFNAMSIZ)
859 if (!strcmp(name, ".") || !strcmp(name, ".."))
863 if (*name == '/' || isspace(*name))
869 EXPORT_SYMBOL(dev_valid_name);
872 * __dev_alloc_name - allocate a name for a device
873 * @net: network namespace to allocate the device name in
874 * @name: name format string
875 * @buf: scratch buffer and result name string
877 * Passed a format string - eg "lt%d" it will try and find a suitable
878 * id. It scans list of devices to build up a free map, then chooses
879 * the first empty slot. The caller must hold the dev_base or rtnl lock
880 * while allocating the name and adding the device in order to avoid
882 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
883 * Returns the number of the unit assigned or a negative errno code.
886 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
890 const int max_netdevices = 8*PAGE_SIZE;
891 unsigned long *inuse;
892 struct net_device *d;
894 p = strnchr(name, IFNAMSIZ-1, '%');
897 * Verify the string as this thing may have come from
898 * the user. There must be either one "%d" and no other "%"
901 if (p[1] != 'd' || strchr(p + 2, '%'))
904 /* Use one page as a bit array of possible slots */
905 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
909 for_each_netdev(net, d) {
910 if (!sscanf(d->name, name, &i))
912 if (i < 0 || i >= max_netdevices)
915 /* avoid cases where sscanf is not exact inverse of printf */
916 snprintf(buf, IFNAMSIZ, name, i);
917 if (!strncmp(buf, d->name, IFNAMSIZ))
921 i = find_first_zero_bit(inuse, max_netdevices);
922 free_page((unsigned long) inuse);
926 snprintf(buf, IFNAMSIZ, name, i);
927 if (!__dev_get_by_name(net, buf))
930 /* It is possible to run out of possible slots
931 * when the name is long and there isn't enough space left
932 * for the digits, or if all bits are used.
938 * dev_alloc_name - allocate a name for a device
940 * @name: name format string
942 * Passed a format string - eg "lt%d" it will try and find a suitable
943 * id. It scans list of devices to build up a free map, then chooses
944 * the first empty slot. The caller must hold the dev_base or rtnl lock
945 * while allocating the name and adding the device in order to avoid
947 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
948 * Returns the number of the unit assigned or a negative errno code.
951 int dev_alloc_name(struct net_device *dev, const char *name)
957 BUG_ON(!dev_net(dev));
959 ret = __dev_alloc_name(net, name, buf);
961 strlcpy(dev->name, buf, IFNAMSIZ);
964 EXPORT_SYMBOL(dev_alloc_name);
966 static int dev_get_valid_name(struct net_device *dev, const char *name)
970 BUG_ON(!dev_net(dev));
973 if (!dev_valid_name(name))
976 if (strchr(name, '%'))
977 return dev_alloc_name(dev, name);
978 else if (__dev_get_by_name(net, name))
980 else if (dev->name != name)
981 strlcpy(dev->name, name, IFNAMSIZ);
987 * dev_change_name - change name of a device
989 * @newname: name (or format string) must be at least IFNAMSIZ
991 * Change name of a device, can pass format strings "eth%d".
994 int dev_change_name(struct net_device *dev, const char *newname)
996 char oldname[IFNAMSIZ];
1002 BUG_ON(!dev_net(dev));
1005 if (dev->flags & IFF_UP)
1008 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1011 memcpy(oldname, dev->name, IFNAMSIZ);
1013 err = dev_get_valid_name(dev, newname);
1018 ret = device_rename(&dev->dev, dev->name);
1020 memcpy(dev->name, oldname, IFNAMSIZ);
1024 write_lock_bh(&dev_base_lock);
1025 hlist_del_rcu(&dev->name_hlist);
1026 write_unlock_bh(&dev_base_lock);
1030 write_lock_bh(&dev_base_lock);
1031 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1032 write_unlock_bh(&dev_base_lock);
1034 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1035 ret = notifier_to_errno(ret);
1038 /* err >= 0 after dev_alloc_name() or stores the first errno */
1041 memcpy(dev->name, oldname, IFNAMSIZ);
1045 "%s: name change rollback failed: %d.\n",
1054 * dev_set_alias - change ifalias of a device
1056 * @alias: name up to IFALIASZ
1057 * @len: limit of bytes to copy from info
1059 * Set ifalias for a device,
1061 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1065 if (len >= IFALIASZ)
1070 kfree(dev->ifalias);
1071 dev->ifalias = NULL;
1076 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1080 strlcpy(dev->ifalias, alias, len+1);
1086 * netdev_features_change - device changes features
1087 * @dev: device to cause notification
1089 * Called to indicate a device has changed features.
1091 void netdev_features_change(struct net_device *dev)
1093 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1095 EXPORT_SYMBOL(netdev_features_change);
1098 * netdev_state_change - device changes state
1099 * @dev: device to cause notification
1101 * Called to indicate a device has changed state. This function calls
1102 * the notifier chains for netdev_chain and sends a NEWLINK message
1103 * to the routing socket.
1105 void netdev_state_change(struct net_device *dev)
1107 if (dev->flags & IFF_UP) {
1108 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1109 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1112 EXPORT_SYMBOL(netdev_state_change);
1114 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1116 return call_netdevice_notifiers(event, dev);
1118 EXPORT_SYMBOL(netdev_bonding_change);
1121 * dev_load - load a network module
1122 * @net: the applicable net namespace
1123 * @name: name of interface
1125 * If a network interface is not present and the process has suitable
1126 * privileges this function loads the module. If module loading is not
1127 * available in this kernel then it becomes a nop.
1130 void dev_load(struct net *net, const char *name)
1132 struct net_device *dev;
1136 dev = dev_get_by_name_rcu(net, name);
1140 if (no_module && capable(CAP_NET_ADMIN))
1141 no_module = request_module("netdev-%s", name);
1142 if (no_module && capable(CAP_SYS_MODULE)) {
1143 if (!request_module("%s", name))
1144 pr_err("Loading kernel module for a network device "
1145 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1149 EXPORT_SYMBOL(dev_load);
1151 static int __dev_open(struct net_device *dev)
1153 const struct net_device_ops *ops = dev->netdev_ops;
1158 if (!netif_device_present(dev))
1161 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1162 ret = notifier_to_errno(ret);
1166 set_bit(__LINK_STATE_START, &dev->state);
1168 if (ops->ndo_validate_addr)
1169 ret = ops->ndo_validate_addr(dev);
1171 if (!ret && ops->ndo_open)
1172 ret = ops->ndo_open(dev);
1175 clear_bit(__LINK_STATE_START, &dev->state);
1177 dev->flags |= IFF_UP;
1178 net_dmaengine_get();
1179 dev_set_rx_mode(dev);
1187 * dev_open - prepare an interface for use.
1188 * @dev: device to open
1190 * Takes a device from down to up state. The device's private open
1191 * function is invoked and then the multicast lists are loaded. Finally
1192 * the device is moved into the up state and a %NETDEV_UP message is
1193 * sent to the netdev notifier chain.
1195 * Calling this function on an active interface is a nop. On a failure
1196 * a negative errno code is returned.
1198 int dev_open(struct net_device *dev)
1202 if (dev->flags & IFF_UP)
1205 ret = __dev_open(dev);
1209 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1210 call_netdevice_notifiers(NETDEV_UP, dev);
1214 EXPORT_SYMBOL(dev_open);
1216 static int __dev_close_many(struct list_head *head)
1218 struct net_device *dev;
1223 list_for_each_entry(dev, head, unreg_list) {
1224 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1226 clear_bit(__LINK_STATE_START, &dev->state);
1228 /* Synchronize to scheduled poll. We cannot touch poll list, it
1229 * can be even on different cpu. So just clear netif_running().
1231 * dev->stop() will invoke napi_disable() on all of it's
1232 * napi_struct instances on this device.
1234 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1237 dev_deactivate_many(head);
1239 list_for_each_entry(dev, head, unreg_list) {
1240 const struct net_device_ops *ops = dev->netdev_ops;
1243 * Call the device specific close. This cannot fail.
1244 * Only if device is UP
1246 * We allow it to be called even after a DETACH hot-plug
1252 dev->flags &= ~IFF_UP;
1253 net_dmaengine_put();
1259 static int __dev_close(struct net_device *dev)
1264 list_add(&dev->unreg_list, &single);
1265 retval = __dev_close_many(&single);
1270 static int dev_close_many(struct list_head *head)
1272 struct net_device *dev, *tmp;
1273 LIST_HEAD(tmp_list);
1275 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1276 if (!(dev->flags & IFF_UP))
1277 list_move(&dev->unreg_list, &tmp_list);
1279 __dev_close_many(head);
1281 list_for_each_entry(dev, head, unreg_list) {
1282 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1283 call_netdevice_notifiers(NETDEV_DOWN, dev);
1286 /* rollback_registered_many needs the complete original list */
1287 list_splice(&tmp_list, head);
1292 * dev_close - shutdown an interface.
1293 * @dev: device to shutdown
1295 * This function moves an active device into down state. A
1296 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1297 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1300 int dev_close(struct net_device *dev)
1302 if (dev->flags & IFF_UP) {
1305 list_add(&dev->unreg_list, &single);
1306 dev_close_many(&single);
1311 EXPORT_SYMBOL(dev_close);
1315 * dev_disable_lro - disable Large Receive Offload on a device
1318 * Disable Large Receive Offload (LRO) on a net device. Must be
1319 * called under RTNL. This is needed if received packets may be
1320 * forwarded to another interface.
1322 void dev_disable_lro(struct net_device *dev)
1325 * If we're trying to disable lro on a vlan device
1326 * use the underlying physical device instead
1328 if (is_vlan_dev(dev))
1329 dev = vlan_dev_real_dev(dev);
1331 dev->wanted_features &= ~NETIF_F_LRO;
1332 netdev_update_features(dev);
1334 if (unlikely(dev->features & NETIF_F_LRO))
1335 netdev_WARN(dev, "failed to disable LRO!\n");
1337 EXPORT_SYMBOL(dev_disable_lro);
1340 static int dev_boot_phase = 1;
1343 * register_netdevice_notifier - register a network notifier block
1346 * Register a notifier to be called when network device events occur.
1347 * The notifier passed is linked into the kernel structures and must
1348 * not be reused until it has been unregistered. A negative errno code
1349 * is returned on a failure.
1351 * When registered all registration and up events are replayed
1352 * to the new notifier to allow device to have a race free
1353 * view of the network device list.
1356 int register_netdevice_notifier(struct notifier_block *nb)
1358 struct net_device *dev;
1359 struct net_device *last;
1364 err = raw_notifier_chain_register(&netdev_chain, nb);
1370 for_each_netdev(net, dev) {
1371 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1372 err = notifier_to_errno(err);
1376 if (!(dev->flags & IFF_UP))
1379 nb->notifier_call(nb, NETDEV_UP, dev);
1390 for_each_netdev(net, dev) {
1394 if (dev->flags & IFF_UP) {
1395 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1396 nb->notifier_call(nb, NETDEV_DOWN, dev);
1398 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1399 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1403 raw_notifier_chain_unregister(&netdev_chain, nb);
1406 EXPORT_SYMBOL(register_netdevice_notifier);
1409 * unregister_netdevice_notifier - unregister a network notifier block
1412 * Unregister a notifier previously registered by
1413 * register_netdevice_notifier(). The notifier is unlinked into the
1414 * kernel structures and may then be reused. A negative errno code
1415 * is returned on a failure.
1418 int unregister_netdevice_notifier(struct notifier_block *nb)
1423 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1427 EXPORT_SYMBOL(unregister_netdevice_notifier);
1430 * call_netdevice_notifiers - call all network notifier blocks
1431 * @val: value passed unmodified to notifier function
1432 * @dev: net_device pointer passed unmodified to notifier function
1434 * Call all network notifier blocks. Parameters and return value
1435 * are as for raw_notifier_call_chain().
1438 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1441 return raw_notifier_call_chain(&netdev_chain, val, dev);
1443 EXPORT_SYMBOL(call_netdevice_notifiers);
1445 static struct jump_label_key netstamp_needed __read_mostly;
1447 void net_enable_timestamp(void)
1449 jump_label_inc(&netstamp_needed);
1451 EXPORT_SYMBOL(net_enable_timestamp);
1453 void net_disable_timestamp(void)
1455 jump_label_dec(&netstamp_needed);
1457 EXPORT_SYMBOL(net_disable_timestamp);
1459 static inline void net_timestamp_set(struct sk_buff *skb)
1461 skb->tstamp.tv64 = 0;
1462 if (static_branch(&netstamp_needed))
1463 __net_timestamp(skb);
1466 #define net_timestamp_check(COND, SKB) \
1467 if (static_branch(&netstamp_needed)) { \
1468 if ((COND) && !(SKB)->tstamp.tv64) \
1469 __net_timestamp(SKB); \
1472 static int net_hwtstamp_validate(struct ifreq *ifr)
1474 struct hwtstamp_config cfg;
1475 enum hwtstamp_tx_types tx_type;
1476 enum hwtstamp_rx_filters rx_filter;
1477 int tx_type_valid = 0;
1478 int rx_filter_valid = 0;
1480 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1483 if (cfg.flags) /* reserved for future extensions */
1486 tx_type = cfg.tx_type;
1487 rx_filter = cfg.rx_filter;
1490 case HWTSTAMP_TX_OFF:
1491 case HWTSTAMP_TX_ON:
1492 case HWTSTAMP_TX_ONESTEP_SYNC:
1497 switch (rx_filter) {
1498 case HWTSTAMP_FILTER_NONE:
1499 case HWTSTAMP_FILTER_ALL:
1500 case HWTSTAMP_FILTER_SOME:
1501 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1502 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1503 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1504 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1505 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1506 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1507 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1508 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1509 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1510 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1511 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1512 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1513 rx_filter_valid = 1;
1517 if (!tx_type_valid || !rx_filter_valid)
1523 static inline bool is_skb_forwardable(struct net_device *dev,
1524 struct sk_buff *skb)
1528 if (!(dev->flags & IFF_UP))
1531 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1532 if (skb->len <= len)
1535 /* if TSO is enabled, we don't care about the length as the packet
1536 * could be forwarded without being segmented before
1538 if (skb_is_gso(skb))
1545 * dev_forward_skb - loopback an skb to another netif
1547 * @dev: destination network device
1548 * @skb: buffer to forward
1551 * NET_RX_SUCCESS (no congestion)
1552 * NET_RX_DROP (packet was dropped, but freed)
1554 * dev_forward_skb can be used for injecting an skb from the
1555 * start_xmit function of one device into the receive queue
1556 * of another device.
1558 * The receiving device may be in another namespace, so
1559 * we have to clear all information in the skb that could
1560 * impact namespace isolation.
1562 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1564 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1565 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1566 atomic_long_inc(&dev->rx_dropped);
1575 if (unlikely(!is_skb_forwardable(dev, skb))) {
1576 atomic_long_inc(&dev->rx_dropped);
1580 skb_set_dev(skb, dev);
1581 skb->tstamp.tv64 = 0;
1582 skb->pkt_type = PACKET_HOST;
1583 skb->protocol = eth_type_trans(skb, dev);
1584 return netif_rx(skb);
1586 EXPORT_SYMBOL_GPL(dev_forward_skb);
1588 static inline int deliver_skb(struct sk_buff *skb,
1589 struct packet_type *pt_prev,
1590 struct net_device *orig_dev)
1592 atomic_inc(&skb->users);
1593 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1597 * Support routine. Sends outgoing frames to any network
1598 * taps currently in use.
1601 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1603 struct packet_type *ptype;
1604 struct sk_buff *skb2 = NULL;
1605 struct packet_type *pt_prev = NULL;
1608 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1609 /* Never send packets back to the socket
1610 * they originated from - MvS (miquels@drinkel.ow.org)
1612 if ((ptype->dev == dev || !ptype->dev) &&
1613 (ptype->af_packet_priv == NULL ||
1614 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1616 deliver_skb(skb2, pt_prev, skb->dev);
1621 skb2 = skb_clone(skb, GFP_ATOMIC);
1625 net_timestamp_set(skb2);
1627 /* skb->nh should be correctly
1628 set by sender, so that the second statement is
1629 just protection against buggy protocols.
1631 skb_reset_mac_header(skb2);
1633 if (skb_network_header(skb2) < skb2->data ||
1634 skb2->network_header > skb2->tail) {
1635 if (net_ratelimit())
1636 printk(KERN_CRIT "protocol %04x is "
1638 ntohs(skb2->protocol),
1640 skb_reset_network_header(skb2);
1643 skb2->transport_header = skb2->network_header;
1644 skb2->pkt_type = PACKET_OUTGOING;
1649 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1653 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1654 * @dev: Network device
1655 * @txq: number of queues available
1657 * If real_num_tx_queues is changed the tc mappings may no longer be
1658 * valid. To resolve this verify the tc mapping remains valid and if
1659 * not NULL the mapping. With no priorities mapping to this
1660 * offset/count pair it will no longer be used. In the worst case TC0
1661 * is invalid nothing can be done so disable priority mappings. If is
1662 * expected that drivers will fix this mapping if they can before
1663 * calling netif_set_real_num_tx_queues.
1665 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1668 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1670 /* If TC0 is invalidated disable TC mapping */
1671 if (tc->offset + tc->count > txq) {
1672 pr_warning("Number of in use tx queues changed "
1673 "invalidating tc mappings. Priority "
1674 "traffic classification disabled!\n");
1679 /* Invalidated prio to tc mappings set to TC0 */
1680 for (i = 1; i < TC_BITMASK + 1; i++) {
1681 int q = netdev_get_prio_tc_map(dev, i);
1683 tc = &dev->tc_to_txq[q];
1684 if (tc->offset + tc->count > txq) {
1685 pr_warning("Number of in use tx queues "
1686 "changed. Priority %i to tc "
1687 "mapping %i is no longer valid "
1688 "setting map to 0\n",
1690 netdev_set_prio_tc_map(dev, i, 0);
1696 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1697 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1699 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1703 if (txq < 1 || txq > dev->num_tx_queues)
1706 if (dev->reg_state == NETREG_REGISTERED ||
1707 dev->reg_state == NETREG_UNREGISTERING) {
1710 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1716 netif_setup_tc(dev, txq);
1718 if (txq < dev->real_num_tx_queues)
1719 qdisc_reset_all_tx_gt(dev, txq);
1722 dev->real_num_tx_queues = txq;
1725 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1729 * netif_set_real_num_rx_queues - set actual number of RX queues used
1730 * @dev: Network device
1731 * @rxq: Actual number of RX queues
1733 * This must be called either with the rtnl_lock held or before
1734 * registration of the net device. Returns 0 on success, or a
1735 * negative error code. If called before registration, it always
1738 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1742 if (rxq < 1 || rxq > dev->num_rx_queues)
1745 if (dev->reg_state == NETREG_REGISTERED) {
1748 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1754 dev->real_num_rx_queues = rxq;
1757 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1760 static inline void __netif_reschedule(struct Qdisc *q)
1762 struct softnet_data *sd;
1763 unsigned long flags;
1765 local_irq_save(flags);
1766 sd = &__get_cpu_var(softnet_data);
1767 q->next_sched = NULL;
1768 *sd->output_queue_tailp = q;
1769 sd->output_queue_tailp = &q->next_sched;
1770 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1771 local_irq_restore(flags);
1774 void __netif_schedule(struct Qdisc *q)
1776 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1777 __netif_reschedule(q);
1779 EXPORT_SYMBOL(__netif_schedule);
1781 void dev_kfree_skb_irq(struct sk_buff *skb)
1783 if (atomic_dec_and_test(&skb->users)) {
1784 struct softnet_data *sd;
1785 unsigned long flags;
1787 local_irq_save(flags);
1788 sd = &__get_cpu_var(softnet_data);
1789 skb->next = sd->completion_queue;
1790 sd->completion_queue = skb;
1791 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1792 local_irq_restore(flags);
1795 EXPORT_SYMBOL(dev_kfree_skb_irq);
1797 void dev_kfree_skb_any(struct sk_buff *skb)
1799 if (in_irq() || irqs_disabled())
1800 dev_kfree_skb_irq(skb);
1804 EXPORT_SYMBOL(dev_kfree_skb_any);
1808 * netif_device_detach - mark device as removed
1809 * @dev: network device
1811 * Mark device as removed from system and therefore no longer available.
1813 void netif_device_detach(struct net_device *dev)
1815 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1816 netif_running(dev)) {
1817 netif_tx_stop_all_queues(dev);
1820 EXPORT_SYMBOL(netif_device_detach);
1823 * netif_device_attach - mark device as attached
1824 * @dev: network device
1826 * Mark device as attached from system and restart if needed.
1828 void netif_device_attach(struct net_device *dev)
1830 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1831 netif_running(dev)) {
1832 netif_tx_wake_all_queues(dev);
1833 __netdev_watchdog_up(dev);
1836 EXPORT_SYMBOL(netif_device_attach);
1839 * skb_dev_set -- assign a new device to a buffer
1840 * @skb: buffer for the new device
1841 * @dev: network device
1843 * If an skb is owned by a device already, we have to reset
1844 * all data private to the namespace a device belongs to
1845 * before assigning it a new device.
1847 #ifdef CONFIG_NET_NS
1848 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1851 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1854 skb_init_secmark(skb);
1858 skb->ipvs_property = 0;
1859 #ifdef CONFIG_NET_SCHED
1865 EXPORT_SYMBOL(skb_set_dev);
1866 #endif /* CONFIG_NET_NS */
1869 * Invalidate hardware checksum when packet is to be mangled, and
1870 * complete checksum manually on outgoing path.
1872 int skb_checksum_help(struct sk_buff *skb)
1875 int ret = 0, offset;
1877 if (skb->ip_summed == CHECKSUM_COMPLETE)
1878 goto out_set_summed;
1880 if (unlikely(skb_shinfo(skb)->gso_size)) {
1881 /* Let GSO fix up the checksum. */
1882 goto out_set_summed;
1885 offset = skb_checksum_start_offset(skb);
1886 BUG_ON(offset >= skb_headlen(skb));
1887 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1889 offset += skb->csum_offset;
1890 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1892 if (skb_cloned(skb) &&
1893 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1894 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1899 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1901 skb->ip_summed = CHECKSUM_NONE;
1905 EXPORT_SYMBOL(skb_checksum_help);
1908 * skb_gso_segment - Perform segmentation on skb.
1909 * @skb: buffer to segment
1910 * @features: features for the output path (see dev->features)
1912 * This function segments the given skb and returns a list of segments.
1914 * It may return NULL if the skb requires no segmentation. This is
1915 * only possible when GSO is used for verifying header integrity.
1917 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1918 netdev_features_t features)
1920 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1921 struct packet_type *ptype;
1922 __be16 type = skb->protocol;
1923 int vlan_depth = ETH_HLEN;
1926 while (type == htons(ETH_P_8021Q)) {
1927 struct vlan_hdr *vh;
1929 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1930 return ERR_PTR(-EINVAL);
1932 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1933 type = vh->h_vlan_encapsulated_proto;
1934 vlan_depth += VLAN_HLEN;
1937 skb_reset_mac_header(skb);
1938 skb->mac_len = skb->network_header - skb->mac_header;
1939 __skb_pull(skb, skb->mac_len);
1941 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1942 struct net_device *dev = skb->dev;
1943 struct ethtool_drvinfo info = {};
1945 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1946 dev->ethtool_ops->get_drvinfo(dev, &info);
1948 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d ip_summed=%d\n",
1949 info.driver, dev ? &dev->features : NULL,
1950 skb->sk ? &skb->sk->sk_route_caps : NULL,
1951 skb->len, skb->data_len, skb->ip_summed);
1953 if (skb_header_cloned(skb) &&
1954 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1955 return ERR_PTR(err);
1959 list_for_each_entry_rcu(ptype,
1960 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1961 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1962 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1963 err = ptype->gso_send_check(skb);
1964 segs = ERR_PTR(err);
1965 if (err || skb_gso_ok(skb, features))
1967 __skb_push(skb, (skb->data -
1968 skb_network_header(skb)));
1970 segs = ptype->gso_segment(skb, features);
1976 __skb_push(skb, skb->data - skb_mac_header(skb));
1980 EXPORT_SYMBOL(skb_gso_segment);
1982 /* Take action when hardware reception checksum errors are detected. */
1984 void netdev_rx_csum_fault(struct net_device *dev)
1986 if (net_ratelimit()) {
1987 printk(KERN_ERR "%s: hw csum failure.\n",
1988 dev ? dev->name : "<unknown>");
1992 EXPORT_SYMBOL(netdev_rx_csum_fault);
1995 /* Actually, we should eliminate this check as soon as we know, that:
1996 * 1. IOMMU is present and allows to map all the memory.
1997 * 2. No high memory really exists on this machine.
2000 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2002 #ifdef CONFIG_HIGHMEM
2004 if (!(dev->features & NETIF_F_HIGHDMA)) {
2005 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2006 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2007 if (PageHighMem(skb_frag_page(frag)))
2012 if (PCI_DMA_BUS_IS_PHYS) {
2013 struct device *pdev = dev->dev.parent;
2017 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2018 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2019 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2020 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2029 void (*destructor)(struct sk_buff *skb);
2032 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2034 static void dev_gso_skb_destructor(struct sk_buff *skb)
2036 struct dev_gso_cb *cb;
2039 struct sk_buff *nskb = skb->next;
2041 skb->next = nskb->next;
2044 } while (skb->next);
2046 cb = DEV_GSO_CB(skb);
2048 cb->destructor(skb);
2052 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2053 * @skb: buffer to segment
2054 * @features: device features as applicable to this skb
2056 * This function segments the given skb and stores the list of segments
2059 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2061 struct sk_buff *segs;
2063 segs = skb_gso_segment(skb, features);
2065 /* Verifying header integrity only. */
2070 return PTR_ERR(segs);
2073 DEV_GSO_CB(skb)->destructor = skb->destructor;
2074 skb->destructor = dev_gso_skb_destructor;
2080 * Try to orphan skb early, right before transmission by the device.
2081 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2082 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2084 static inline void skb_orphan_try(struct sk_buff *skb)
2086 struct sock *sk = skb->sk;
2088 if (sk && !skb_shinfo(skb)->tx_flags) {
2089 /* skb_tx_hash() wont be able to get sk.
2090 * We copy sk_hash into skb->rxhash
2093 skb->rxhash = sk->sk_hash;
2098 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2100 return ((features & NETIF_F_GEN_CSUM) ||
2101 ((features & NETIF_F_V4_CSUM) &&
2102 protocol == htons(ETH_P_IP)) ||
2103 ((features & NETIF_F_V6_CSUM) &&
2104 protocol == htons(ETH_P_IPV6)) ||
2105 ((features & NETIF_F_FCOE_CRC) &&
2106 protocol == htons(ETH_P_FCOE)));
2109 static netdev_features_t harmonize_features(struct sk_buff *skb,
2110 __be16 protocol, netdev_features_t features)
2112 if (!can_checksum_protocol(features, protocol)) {
2113 features &= ~NETIF_F_ALL_CSUM;
2114 features &= ~NETIF_F_SG;
2115 } else if (illegal_highdma(skb->dev, skb)) {
2116 features &= ~NETIF_F_SG;
2122 netdev_features_t netif_skb_features(struct sk_buff *skb)
2124 __be16 protocol = skb->protocol;
2125 netdev_features_t features = skb->dev->features;
2127 if (protocol == htons(ETH_P_8021Q)) {
2128 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2129 protocol = veh->h_vlan_encapsulated_proto;
2130 } else if (!vlan_tx_tag_present(skb)) {
2131 return harmonize_features(skb, protocol, features);
2134 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2136 if (protocol != htons(ETH_P_8021Q)) {
2137 return harmonize_features(skb, protocol, features);
2139 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2140 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2141 return harmonize_features(skb, protocol, features);
2144 EXPORT_SYMBOL(netif_skb_features);
2147 * Returns true if either:
2148 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2149 * 2. skb is fragmented and the device does not support SG, or if
2150 * at least one of fragments is in highmem and device does not
2151 * support DMA from it.
2153 static inline int skb_needs_linearize(struct sk_buff *skb,
2156 return skb_is_nonlinear(skb) &&
2157 ((skb_has_frag_list(skb) &&
2158 !(features & NETIF_F_FRAGLIST)) ||
2159 (skb_shinfo(skb)->nr_frags &&
2160 !(features & NETIF_F_SG)));
2163 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2164 struct netdev_queue *txq)
2166 const struct net_device_ops *ops = dev->netdev_ops;
2167 int rc = NETDEV_TX_OK;
2168 unsigned int skb_len;
2170 if (likely(!skb->next)) {
2171 netdev_features_t features;
2174 * If device doesn't need skb->dst, release it right now while
2175 * its hot in this cpu cache
2177 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2180 if (!list_empty(&ptype_all))
2181 dev_queue_xmit_nit(skb, dev);
2183 skb_orphan_try(skb);
2185 features = netif_skb_features(skb);
2187 if (vlan_tx_tag_present(skb) &&
2188 !(features & NETIF_F_HW_VLAN_TX)) {
2189 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2196 if (netif_needs_gso(skb, features)) {
2197 if (unlikely(dev_gso_segment(skb, features)))
2202 if (skb_needs_linearize(skb, features) &&
2203 __skb_linearize(skb))
2206 /* If packet is not checksummed and device does not
2207 * support checksumming for this protocol, complete
2208 * checksumming here.
2210 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2211 skb_set_transport_header(skb,
2212 skb_checksum_start_offset(skb));
2213 if (!(features & NETIF_F_ALL_CSUM) &&
2214 skb_checksum_help(skb))
2220 rc = ops->ndo_start_xmit(skb, dev);
2221 trace_net_dev_xmit(skb, rc, dev, skb_len);
2222 if (rc == NETDEV_TX_OK)
2223 txq_trans_update(txq);
2229 struct sk_buff *nskb = skb->next;
2231 skb->next = nskb->next;
2235 * If device doesn't need nskb->dst, release it right now while
2236 * its hot in this cpu cache
2238 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2241 skb_len = nskb->len;
2242 rc = ops->ndo_start_xmit(nskb, dev);
2243 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2244 if (unlikely(rc != NETDEV_TX_OK)) {
2245 if (rc & ~NETDEV_TX_MASK)
2246 goto out_kfree_gso_skb;
2247 nskb->next = skb->next;
2251 txq_trans_update(txq);
2252 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2253 return NETDEV_TX_BUSY;
2254 } while (skb->next);
2257 if (likely(skb->next == NULL))
2258 skb->destructor = DEV_GSO_CB(skb)->destructor;
2265 static u32 hashrnd __read_mostly;
2268 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2269 * to be used as a distribution range.
2271 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2272 unsigned int num_tx_queues)
2276 u16 qcount = num_tx_queues;
2278 if (skb_rx_queue_recorded(skb)) {
2279 hash = skb_get_rx_queue(skb);
2280 while (unlikely(hash >= num_tx_queues))
2281 hash -= num_tx_queues;
2286 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2287 qoffset = dev->tc_to_txq[tc].offset;
2288 qcount = dev->tc_to_txq[tc].count;
2291 if (skb->sk && skb->sk->sk_hash)
2292 hash = skb->sk->sk_hash;
2294 hash = (__force u16) skb->protocol ^ skb->rxhash;
2295 hash = jhash_1word(hash, hashrnd);
2297 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2299 EXPORT_SYMBOL(__skb_tx_hash);
2301 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2303 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2304 if (net_ratelimit()) {
2305 pr_warning("%s selects TX queue %d, but "
2306 "real number of TX queues is %d\n",
2307 dev->name, queue_index, dev->real_num_tx_queues);
2314 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2317 struct xps_dev_maps *dev_maps;
2318 struct xps_map *map;
2319 int queue_index = -1;
2322 dev_maps = rcu_dereference(dev->xps_maps);
2324 map = rcu_dereference(
2325 dev_maps->cpu_map[raw_smp_processor_id()]);
2328 queue_index = map->queues[0];
2331 if (skb->sk && skb->sk->sk_hash)
2332 hash = skb->sk->sk_hash;
2334 hash = (__force u16) skb->protocol ^
2336 hash = jhash_1word(hash, hashrnd);
2337 queue_index = map->queues[
2338 ((u64)hash * map->len) >> 32];
2340 if (unlikely(queue_index >= dev->real_num_tx_queues))
2352 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2353 struct sk_buff *skb)
2356 const struct net_device_ops *ops = dev->netdev_ops;
2358 if (dev->real_num_tx_queues == 1)
2360 else if (ops->ndo_select_queue) {
2361 queue_index = ops->ndo_select_queue(dev, skb);
2362 queue_index = dev_cap_txqueue(dev, queue_index);
2364 struct sock *sk = skb->sk;
2365 queue_index = sk_tx_queue_get(sk);
2367 if (queue_index < 0 || skb->ooo_okay ||
2368 queue_index >= dev->real_num_tx_queues) {
2369 int old_index = queue_index;
2371 queue_index = get_xps_queue(dev, skb);
2372 if (queue_index < 0)
2373 queue_index = skb_tx_hash(dev, skb);
2375 if (queue_index != old_index && sk) {
2376 struct dst_entry *dst =
2377 rcu_dereference_check(sk->sk_dst_cache, 1);
2379 if (dst && skb_dst(skb) == dst)
2380 sk_tx_queue_set(sk, queue_index);
2385 skb_set_queue_mapping(skb, queue_index);
2386 return netdev_get_tx_queue(dev, queue_index);
2389 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2390 struct net_device *dev,
2391 struct netdev_queue *txq)
2393 spinlock_t *root_lock = qdisc_lock(q);
2397 qdisc_skb_cb(skb)->pkt_len = skb->len;
2398 qdisc_calculate_pkt_len(skb, q);
2400 * Heuristic to force contended enqueues to serialize on a
2401 * separate lock before trying to get qdisc main lock.
2402 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2403 * and dequeue packets faster.
2405 contended = qdisc_is_running(q);
2406 if (unlikely(contended))
2407 spin_lock(&q->busylock);
2409 spin_lock(root_lock);
2410 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2413 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2414 qdisc_run_begin(q)) {
2416 * This is a work-conserving queue; there are no old skbs
2417 * waiting to be sent out; and the qdisc is not running -
2418 * xmit the skb directly.
2420 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2423 qdisc_bstats_update(q, skb);
2425 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2426 if (unlikely(contended)) {
2427 spin_unlock(&q->busylock);
2434 rc = NET_XMIT_SUCCESS;
2437 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2438 if (qdisc_run_begin(q)) {
2439 if (unlikely(contended)) {
2440 spin_unlock(&q->busylock);
2446 spin_unlock(root_lock);
2447 if (unlikely(contended))
2448 spin_unlock(&q->busylock);
2452 static DEFINE_PER_CPU(int, xmit_recursion);
2453 #define RECURSION_LIMIT 10
2456 * dev_queue_xmit - transmit a buffer
2457 * @skb: buffer to transmit
2459 * Queue a buffer for transmission to a network device. The caller must
2460 * have set the device and priority and built the buffer before calling
2461 * this function. The function can be called from an interrupt.
2463 * A negative errno code is returned on a failure. A success does not
2464 * guarantee the frame will be transmitted as it may be dropped due
2465 * to congestion or traffic shaping.
2467 * -----------------------------------------------------------------------------------
2468 * I notice this method can also return errors from the queue disciplines,
2469 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2472 * Regardless of the return value, the skb is consumed, so it is currently
2473 * difficult to retry a send to this method. (You can bump the ref count
2474 * before sending to hold a reference for retry if you are careful.)
2476 * When calling this method, interrupts MUST be enabled. This is because
2477 * the BH enable code must have IRQs enabled so that it will not deadlock.
2480 int dev_queue_xmit(struct sk_buff *skb)
2482 struct net_device *dev = skb->dev;
2483 struct netdev_queue *txq;
2487 /* Disable soft irqs for various locks below. Also
2488 * stops preemption for RCU.
2492 txq = dev_pick_tx(dev, skb);
2493 q = rcu_dereference_bh(txq->qdisc);
2495 #ifdef CONFIG_NET_CLS_ACT
2496 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2498 trace_net_dev_queue(skb);
2500 rc = __dev_xmit_skb(skb, q, dev, txq);
2504 /* The device has no queue. Common case for software devices:
2505 loopback, all the sorts of tunnels...
2507 Really, it is unlikely that netif_tx_lock protection is necessary
2508 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2510 However, it is possible, that they rely on protection
2513 Check this and shot the lock. It is not prone from deadlocks.
2514 Either shot noqueue qdisc, it is even simpler 8)
2516 if (dev->flags & IFF_UP) {
2517 int cpu = smp_processor_id(); /* ok because BHs are off */
2519 if (txq->xmit_lock_owner != cpu) {
2521 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2522 goto recursion_alert;
2524 HARD_TX_LOCK(dev, txq, cpu);
2526 if (!netif_tx_queue_stopped(txq)) {
2527 __this_cpu_inc(xmit_recursion);
2528 rc = dev_hard_start_xmit(skb, dev, txq);
2529 __this_cpu_dec(xmit_recursion);
2530 if (dev_xmit_complete(rc)) {
2531 HARD_TX_UNLOCK(dev, txq);
2535 HARD_TX_UNLOCK(dev, txq);
2536 if (net_ratelimit())
2537 printk(KERN_CRIT "Virtual device %s asks to "
2538 "queue packet!\n", dev->name);
2540 /* Recursion is detected! It is possible,
2544 if (net_ratelimit())
2545 printk(KERN_CRIT "Dead loop on virtual device "
2546 "%s, fix it urgently!\n", dev->name);
2551 rcu_read_unlock_bh();
2556 rcu_read_unlock_bh();
2559 EXPORT_SYMBOL(dev_queue_xmit);
2562 /*=======================================================================
2564 =======================================================================*/
2566 int netdev_max_backlog __read_mostly = 1000;
2567 int netdev_tstamp_prequeue __read_mostly = 1;
2568 int netdev_budget __read_mostly = 300;
2569 int weight_p __read_mostly = 64; /* old backlog weight */
2571 /* Called with irq disabled */
2572 static inline void ____napi_schedule(struct softnet_data *sd,
2573 struct napi_struct *napi)
2575 list_add_tail(&napi->poll_list, &sd->poll_list);
2576 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2580 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2581 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2582 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2583 * if hash is a canonical 4-tuple hash over transport ports.
2585 void __skb_get_rxhash(struct sk_buff *skb)
2587 int nhoff, hash = 0, poff;
2588 const struct ipv6hdr *ip6;
2589 const struct iphdr *ip;
2590 const struct vlan_hdr *vlan;
2599 nhoff = skb_network_offset(skb);
2600 proto = skb->protocol;
2604 case __constant_htons(ETH_P_IP):
2606 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2609 ip = (const struct iphdr *) (skb->data + nhoff);
2610 if (ip_is_fragment(ip))
2613 ip_proto = ip->protocol;
2614 addr1 = (__force u32) ip->saddr;
2615 addr2 = (__force u32) ip->daddr;
2616 nhoff += ip->ihl * 4;
2618 case __constant_htons(ETH_P_IPV6):
2620 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2623 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2624 ip_proto = ip6->nexthdr;
2625 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2626 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2629 case __constant_htons(ETH_P_8021Q):
2630 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2632 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2633 proto = vlan->h_vlan_encapsulated_proto;
2634 nhoff += sizeof(*vlan);
2636 case __constant_htons(ETH_P_PPP_SES):
2637 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2639 proto = *((__be16 *) (skb->data + nhoff +
2640 sizeof(struct pppoe_hdr)));
2641 nhoff += PPPOE_SES_HLEN;
2643 case __constant_htons(PPP_IP):
2645 case __constant_htons(PPP_IPV6):
2656 if (pskb_may_pull(skb, nhoff + 16)) {
2657 u8 *h = skb->data + nhoff;
2658 __be16 flags = *(__be16 *)h;
2661 * Only look inside GRE if version zero and no
2664 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2665 proto = *(__be16 *)(h + 2);
2667 if (flags & GRE_CSUM)
2669 if (flags & GRE_KEY)
2671 if (flags & GRE_SEQ)
2684 poff = proto_ports_offset(ip_proto);
2687 if (pskb_may_pull(skb, nhoff + 4)) {
2688 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2689 if (ports.v16[1] < ports.v16[0])
2690 swap(ports.v16[0], ports.v16[1]);
2695 /* get a consistent hash (same value on both flow directions) */
2699 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2706 EXPORT_SYMBOL(__skb_get_rxhash);
2710 /* One global table that all flow-based protocols share. */
2711 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2712 EXPORT_SYMBOL(rps_sock_flow_table);
2714 static struct rps_dev_flow *
2715 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2716 struct rps_dev_flow *rflow, u16 next_cpu)
2718 if (next_cpu != RPS_NO_CPU) {
2719 #ifdef CONFIG_RFS_ACCEL
2720 struct netdev_rx_queue *rxqueue;
2721 struct rps_dev_flow_table *flow_table;
2722 struct rps_dev_flow *old_rflow;
2727 /* Should we steer this flow to a different hardware queue? */
2728 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2729 !(dev->features & NETIF_F_NTUPLE))
2731 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2732 if (rxq_index == skb_get_rx_queue(skb))
2735 rxqueue = dev->_rx + rxq_index;
2736 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2739 flow_id = skb->rxhash & flow_table->mask;
2740 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2741 rxq_index, flow_id);
2745 rflow = &flow_table->flows[flow_id];
2747 if (old_rflow->filter == rflow->filter)
2748 old_rflow->filter = RPS_NO_FILTER;
2752 per_cpu(softnet_data, next_cpu).input_queue_head;
2755 rflow->cpu = next_cpu;
2760 * get_rps_cpu is called from netif_receive_skb and returns the target
2761 * CPU from the RPS map of the receiving queue for a given skb.
2762 * rcu_read_lock must be held on entry.
2764 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2765 struct rps_dev_flow **rflowp)
2767 struct netdev_rx_queue *rxqueue;
2768 struct rps_map *map;
2769 struct rps_dev_flow_table *flow_table;
2770 struct rps_sock_flow_table *sock_flow_table;
2774 if (skb_rx_queue_recorded(skb)) {
2775 u16 index = skb_get_rx_queue(skb);
2776 if (unlikely(index >= dev->real_num_rx_queues)) {
2777 WARN_ONCE(dev->real_num_rx_queues > 1,
2778 "%s received packet on queue %u, but number "
2779 "of RX queues is %u\n",
2780 dev->name, index, dev->real_num_rx_queues);
2783 rxqueue = dev->_rx + index;
2787 map = rcu_dereference(rxqueue->rps_map);
2789 if (map->len == 1 &&
2790 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2791 tcpu = map->cpus[0];
2792 if (cpu_online(tcpu))
2796 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2800 skb_reset_network_header(skb);
2801 if (!skb_get_rxhash(skb))
2804 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2805 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2806 if (flow_table && sock_flow_table) {
2808 struct rps_dev_flow *rflow;
2810 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2813 next_cpu = sock_flow_table->ents[skb->rxhash &
2814 sock_flow_table->mask];
2817 * If the desired CPU (where last recvmsg was done) is
2818 * different from current CPU (one in the rx-queue flow
2819 * table entry), switch if one of the following holds:
2820 * - Current CPU is unset (equal to RPS_NO_CPU).
2821 * - Current CPU is offline.
2822 * - The current CPU's queue tail has advanced beyond the
2823 * last packet that was enqueued using this table entry.
2824 * This guarantees that all previous packets for the flow
2825 * have been dequeued, thus preserving in order delivery.
2827 if (unlikely(tcpu != next_cpu) &&
2828 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2829 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2830 rflow->last_qtail)) >= 0))
2831 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2833 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2841 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2843 if (cpu_online(tcpu)) {
2853 #ifdef CONFIG_RFS_ACCEL
2856 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2857 * @dev: Device on which the filter was set
2858 * @rxq_index: RX queue index
2859 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2860 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2862 * Drivers that implement ndo_rx_flow_steer() should periodically call
2863 * this function for each installed filter and remove the filters for
2864 * which it returns %true.
2866 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2867 u32 flow_id, u16 filter_id)
2869 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2870 struct rps_dev_flow_table *flow_table;
2871 struct rps_dev_flow *rflow;
2876 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2877 if (flow_table && flow_id <= flow_table->mask) {
2878 rflow = &flow_table->flows[flow_id];
2879 cpu = ACCESS_ONCE(rflow->cpu);
2880 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2881 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2882 rflow->last_qtail) <
2883 (int)(10 * flow_table->mask)))
2889 EXPORT_SYMBOL(rps_may_expire_flow);
2891 #endif /* CONFIG_RFS_ACCEL */
2893 /* Called from hardirq (IPI) context */
2894 static void rps_trigger_softirq(void *data)
2896 struct softnet_data *sd = data;
2898 ____napi_schedule(sd, &sd->backlog);
2902 #endif /* CONFIG_RPS */
2905 * Check if this softnet_data structure is another cpu one
2906 * If yes, queue it to our IPI list and return 1
2909 static int rps_ipi_queued(struct softnet_data *sd)
2912 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2915 sd->rps_ipi_next = mysd->rps_ipi_list;
2916 mysd->rps_ipi_list = sd;
2918 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2921 #endif /* CONFIG_RPS */
2926 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2927 * queue (may be a remote CPU queue).
2929 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2930 unsigned int *qtail)
2932 struct softnet_data *sd;
2933 unsigned long flags;
2935 sd = &per_cpu(softnet_data, cpu);
2937 local_irq_save(flags);
2940 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2941 if (skb_queue_len(&sd->input_pkt_queue)) {
2943 __skb_queue_tail(&sd->input_pkt_queue, skb);
2944 input_queue_tail_incr_save(sd, qtail);
2946 local_irq_restore(flags);
2947 return NET_RX_SUCCESS;
2950 /* Schedule NAPI for backlog device
2951 * We can use non atomic operation since we own the queue lock
2953 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2954 if (!rps_ipi_queued(sd))
2955 ____napi_schedule(sd, &sd->backlog);
2963 local_irq_restore(flags);
2965 atomic_long_inc(&skb->dev->rx_dropped);
2971 * netif_rx - post buffer to the network code
2972 * @skb: buffer to post
2974 * This function receives a packet from a device driver and queues it for
2975 * the upper (protocol) levels to process. It always succeeds. The buffer
2976 * may be dropped during processing for congestion control or by the
2980 * NET_RX_SUCCESS (no congestion)
2981 * NET_RX_DROP (packet was dropped)
2985 int netif_rx(struct sk_buff *skb)
2989 /* if netpoll wants it, pretend we never saw it */
2990 if (netpoll_rx(skb))
2993 net_timestamp_check(netdev_tstamp_prequeue, skb);
2995 trace_netif_rx(skb);
2998 struct rps_dev_flow voidflow, *rflow = &voidflow;
3004 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3006 cpu = smp_processor_id();
3008 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3016 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3022 EXPORT_SYMBOL(netif_rx);
3024 int netif_rx_ni(struct sk_buff *skb)
3029 err = netif_rx(skb);
3030 if (local_softirq_pending())
3036 EXPORT_SYMBOL(netif_rx_ni);
3038 static void net_tx_action(struct softirq_action *h)
3040 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3042 if (sd->completion_queue) {
3043 struct sk_buff *clist;
3045 local_irq_disable();
3046 clist = sd->completion_queue;
3047 sd->completion_queue = NULL;
3051 struct sk_buff *skb = clist;
3052 clist = clist->next;
3054 WARN_ON(atomic_read(&skb->users));
3055 trace_kfree_skb(skb, net_tx_action);
3060 if (sd->output_queue) {
3063 local_irq_disable();
3064 head = sd->output_queue;
3065 sd->output_queue = NULL;
3066 sd->output_queue_tailp = &sd->output_queue;
3070 struct Qdisc *q = head;
3071 spinlock_t *root_lock;
3073 head = head->next_sched;
3075 root_lock = qdisc_lock(q);
3076 if (spin_trylock(root_lock)) {
3077 smp_mb__before_clear_bit();
3078 clear_bit(__QDISC_STATE_SCHED,
3081 spin_unlock(root_lock);
3083 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3085 __netif_reschedule(q);
3087 smp_mb__before_clear_bit();
3088 clear_bit(__QDISC_STATE_SCHED,
3096 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3097 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3098 /* This hook is defined here for ATM LANE */
3099 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3100 unsigned char *addr) __read_mostly;
3101 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3104 #ifdef CONFIG_NET_CLS_ACT
3105 /* TODO: Maybe we should just force sch_ingress to be compiled in
3106 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3107 * a compare and 2 stores extra right now if we dont have it on
3108 * but have CONFIG_NET_CLS_ACT
3109 * NOTE: This doesn't stop any functionality; if you dont have
3110 * the ingress scheduler, you just can't add policies on ingress.
3113 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3115 struct net_device *dev = skb->dev;
3116 u32 ttl = G_TC_RTTL(skb->tc_verd);
3117 int result = TC_ACT_OK;
3120 if (unlikely(MAX_RED_LOOP < ttl++)) {
3121 if (net_ratelimit())
3122 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3123 skb->skb_iif, dev->ifindex);
3127 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3128 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3131 if (q != &noop_qdisc) {
3132 spin_lock(qdisc_lock(q));
3133 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3134 result = qdisc_enqueue_root(skb, q);
3135 spin_unlock(qdisc_lock(q));
3141 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3142 struct packet_type **pt_prev,
3143 int *ret, struct net_device *orig_dev)
3145 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3147 if (!rxq || rxq->qdisc == &noop_qdisc)
3151 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3155 switch (ing_filter(skb, rxq)) {
3169 * netdev_rx_handler_register - register receive handler
3170 * @dev: device to register a handler for
3171 * @rx_handler: receive handler to register
3172 * @rx_handler_data: data pointer that is used by rx handler
3174 * Register a receive hander for a device. This handler will then be
3175 * called from __netif_receive_skb. A negative errno code is returned
3178 * The caller must hold the rtnl_mutex.
3180 * For a general description of rx_handler, see enum rx_handler_result.
3182 int netdev_rx_handler_register(struct net_device *dev,
3183 rx_handler_func_t *rx_handler,
3184 void *rx_handler_data)
3188 if (dev->rx_handler)
3191 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3192 rcu_assign_pointer(dev->rx_handler, rx_handler);
3196 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3199 * netdev_rx_handler_unregister - unregister receive handler
3200 * @dev: device to unregister a handler from
3202 * Unregister a receive hander from a device.
3204 * The caller must hold the rtnl_mutex.
3206 void netdev_rx_handler_unregister(struct net_device *dev)
3210 RCU_INIT_POINTER(dev->rx_handler, NULL);
3211 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3213 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3215 static int __netif_receive_skb(struct sk_buff *skb)
3217 struct packet_type *ptype, *pt_prev;
3218 rx_handler_func_t *rx_handler;
3219 struct net_device *orig_dev;
3220 struct net_device *null_or_dev;
3221 bool deliver_exact = false;
3222 int ret = NET_RX_DROP;
3225 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3227 trace_netif_receive_skb(skb);
3229 /* if we've gotten here through NAPI, check netpoll */
3230 if (netpoll_receive_skb(skb))
3234 skb->skb_iif = skb->dev->ifindex;
3235 orig_dev = skb->dev;
3237 skb_reset_network_header(skb);
3238 skb_reset_transport_header(skb);
3239 skb_reset_mac_len(skb);
3247 __this_cpu_inc(softnet_data.processed);
3249 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3250 skb = vlan_untag(skb);
3255 #ifdef CONFIG_NET_CLS_ACT
3256 if (skb->tc_verd & TC_NCLS) {
3257 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3262 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3263 if (!ptype->dev || ptype->dev == skb->dev) {
3265 ret = deliver_skb(skb, pt_prev, orig_dev);
3270 #ifdef CONFIG_NET_CLS_ACT
3271 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3277 rx_handler = rcu_dereference(skb->dev->rx_handler);
3278 if (vlan_tx_tag_present(skb)) {
3280 ret = deliver_skb(skb, pt_prev, orig_dev);
3283 if (vlan_do_receive(&skb, !rx_handler))
3285 else if (unlikely(!skb))
3291 ret = deliver_skb(skb, pt_prev, orig_dev);
3294 switch (rx_handler(&skb)) {
3295 case RX_HANDLER_CONSUMED:
3297 case RX_HANDLER_ANOTHER:
3299 case RX_HANDLER_EXACT:
3300 deliver_exact = true;
3301 case RX_HANDLER_PASS:
3308 /* deliver only exact match when indicated */
3309 null_or_dev = deliver_exact ? skb->dev : NULL;
3311 type = skb->protocol;
3312 list_for_each_entry_rcu(ptype,
3313 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3314 if (ptype->type == type &&
3315 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3316 ptype->dev == orig_dev)) {
3318 ret = deliver_skb(skb, pt_prev, orig_dev);
3324 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3326 atomic_long_inc(&skb->dev->rx_dropped);
3328 /* Jamal, now you will not able to escape explaining
3329 * me how you were going to use this. :-)
3340 * netif_receive_skb - process receive buffer from network
3341 * @skb: buffer to process
3343 * netif_receive_skb() is the main receive data processing function.
3344 * It always succeeds. The buffer may be dropped during processing
3345 * for congestion control or by the protocol layers.
3347 * This function may only be called from softirq context and interrupts
3348 * should be enabled.
3350 * Return values (usually ignored):
3351 * NET_RX_SUCCESS: no congestion
3352 * NET_RX_DROP: packet was dropped
3354 int netif_receive_skb(struct sk_buff *skb)
3356 net_timestamp_check(netdev_tstamp_prequeue, skb);
3358 if (skb_defer_rx_timestamp(skb))
3359 return NET_RX_SUCCESS;
3363 struct rps_dev_flow voidflow, *rflow = &voidflow;
3368 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3371 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3375 ret = __netif_receive_skb(skb);
3381 return __netif_receive_skb(skb);
3384 EXPORT_SYMBOL(netif_receive_skb);
3386 /* Network device is going away, flush any packets still pending
3387 * Called with irqs disabled.
3389 static void flush_backlog(void *arg)
3391 struct net_device *dev = arg;
3392 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3393 struct sk_buff *skb, *tmp;
3396 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3397 if (skb->dev == dev) {
3398 __skb_unlink(skb, &sd->input_pkt_queue);
3400 input_queue_head_incr(sd);
3405 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3406 if (skb->dev == dev) {
3407 __skb_unlink(skb, &sd->process_queue);
3409 input_queue_head_incr(sd);
3414 static int napi_gro_complete(struct sk_buff *skb)
3416 struct packet_type *ptype;
3417 __be16 type = skb->protocol;
3418 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3421 if (NAPI_GRO_CB(skb)->count == 1) {
3422 skb_shinfo(skb)->gso_size = 0;
3427 list_for_each_entry_rcu(ptype, head, list) {
3428 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3431 err = ptype->gro_complete(skb);
3437 WARN_ON(&ptype->list == head);
3439 return NET_RX_SUCCESS;
3443 return netif_receive_skb(skb);
3446 inline void napi_gro_flush(struct napi_struct *napi)
3448 struct sk_buff *skb, *next;
3450 for (skb = napi->gro_list; skb; skb = next) {
3453 napi_gro_complete(skb);
3456 napi->gro_count = 0;
3457 napi->gro_list = NULL;
3459 EXPORT_SYMBOL(napi_gro_flush);
3461 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3463 struct sk_buff **pp = NULL;
3464 struct packet_type *ptype;
3465 __be16 type = skb->protocol;
3466 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3469 enum gro_result ret;
3471 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3474 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3478 list_for_each_entry_rcu(ptype, head, list) {
3479 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3482 skb_set_network_header(skb, skb_gro_offset(skb));
3483 mac_len = skb->network_header - skb->mac_header;
3484 skb->mac_len = mac_len;
3485 NAPI_GRO_CB(skb)->same_flow = 0;
3486 NAPI_GRO_CB(skb)->flush = 0;
3487 NAPI_GRO_CB(skb)->free = 0;
3489 pp = ptype->gro_receive(&napi->gro_list, skb);
3494 if (&ptype->list == head)
3497 same_flow = NAPI_GRO_CB(skb)->same_flow;
3498 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3501 struct sk_buff *nskb = *pp;
3505 napi_gro_complete(nskb);
3512 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3516 NAPI_GRO_CB(skb)->count = 1;
3517 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3518 skb->next = napi->gro_list;
3519 napi->gro_list = skb;
3523 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3524 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3526 BUG_ON(skb->end - skb->tail < grow);
3528 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3531 skb->data_len -= grow;
3533 skb_shinfo(skb)->frags[0].page_offset += grow;
3534 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3536 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3537 skb_frag_unref(skb, 0);
3538 memmove(skb_shinfo(skb)->frags,
3539 skb_shinfo(skb)->frags + 1,
3540 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3551 EXPORT_SYMBOL(dev_gro_receive);
3553 static inline gro_result_t
3554 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3558 for (p = napi->gro_list; p; p = p->next) {
3559 unsigned long diffs;
3561 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3562 diffs |= p->vlan_tci ^ skb->vlan_tci;
3563 diffs |= compare_ether_header(skb_mac_header(p),
3564 skb_gro_mac_header(skb));
3565 NAPI_GRO_CB(p)->same_flow = !diffs;
3566 NAPI_GRO_CB(p)->flush = 0;
3569 return dev_gro_receive(napi, skb);
3572 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3576 if (netif_receive_skb(skb))
3581 case GRO_MERGED_FREE:
3592 EXPORT_SYMBOL(napi_skb_finish);
3594 void skb_gro_reset_offset(struct sk_buff *skb)
3596 NAPI_GRO_CB(skb)->data_offset = 0;
3597 NAPI_GRO_CB(skb)->frag0 = NULL;
3598 NAPI_GRO_CB(skb)->frag0_len = 0;
3600 if (skb->mac_header == skb->tail &&
3601 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3602 NAPI_GRO_CB(skb)->frag0 =
3603 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3604 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3607 EXPORT_SYMBOL(skb_gro_reset_offset);
3609 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3611 skb_gro_reset_offset(skb);
3613 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3615 EXPORT_SYMBOL(napi_gro_receive);
3617 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3619 __skb_pull(skb, skb_headlen(skb));
3620 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3622 skb->dev = napi->dev;
3628 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3630 struct sk_buff *skb = napi->skb;
3633 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3639 EXPORT_SYMBOL(napi_get_frags);
3641 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3647 skb->protocol = eth_type_trans(skb, skb->dev);
3649 if (ret == GRO_HELD)
3650 skb_gro_pull(skb, -ETH_HLEN);
3651 else if (netif_receive_skb(skb))
3656 case GRO_MERGED_FREE:
3657 napi_reuse_skb(napi, skb);
3666 EXPORT_SYMBOL(napi_frags_finish);
3668 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3670 struct sk_buff *skb = napi->skb;
3677 skb_reset_mac_header(skb);
3678 skb_gro_reset_offset(skb);
3680 off = skb_gro_offset(skb);
3681 hlen = off + sizeof(*eth);
3682 eth = skb_gro_header_fast(skb, off);
3683 if (skb_gro_header_hard(skb, hlen)) {
3684 eth = skb_gro_header_slow(skb, hlen, off);
3685 if (unlikely(!eth)) {
3686 napi_reuse_skb(napi, skb);
3692 skb_gro_pull(skb, sizeof(*eth));
3695 * This works because the only protocols we care about don't require
3696 * special handling. We'll fix it up properly at the end.
3698 skb->protocol = eth->h_proto;
3703 EXPORT_SYMBOL(napi_frags_skb);
3705 gro_result_t napi_gro_frags(struct napi_struct *napi)
3707 struct sk_buff *skb = napi_frags_skb(napi);
3712 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3714 EXPORT_SYMBOL(napi_gro_frags);
3717 * net_rps_action sends any pending IPI's for rps.
3718 * Note: called with local irq disabled, but exits with local irq enabled.
3720 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3723 struct softnet_data *remsd = sd->rps_ipi_list;
3726 sd->rps_ipi_list = NULL;
3730 /* Send pending IPI's to kick RPS processing on remote cpus. */
3732 struct softnet_data *next = remsd->rps_ipi_next;
3734 if (cpu_online(remsd->cpu))
3735 __smp_call_function_single(remsd->cpu,
3744 static int process_backlog(struct napi_struct *napi, int quota)
3747 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3750 /* Check if we have pending ipi, its better to send them now,
3751 * not waiting net_rx_action() end.
3753 if (sd->rps_ipi_list) {
3754 local_irq_disable();
3755 net_rps_action_and_irq_enable(sd);
3758 napi->weight = weight_p;
3759 local_irq_disable();
3760 while (work < quota) {
3761 struct sk_buff *skb;
3764 while ((skb = __skb_dequeue(&sd->process_queue))) {
3766 __netif_receive_skb(skb);
3767 local_irq_disable();
3768 input_queue_head_incr(sd);
3769 if (++work >= quota) {
3776 qlen = skb_queue_len(&sd->input_pkt_queue);
3778 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3779 &sd->process_queue);
3781 if (qlen < quota - work) {
3783 * Inline a custom version of __napi_complete().
3784 * only current cpu owns and manipulates this napi,
3785 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3786 * we can use a plain write instead of clear_bit(),
3787 * and we dont need an smp_mb() memory barrier.
3789 list_del(&napi->poll_list);
3792 quota = work + qlen;
3802 * __napi_schedule - schedule for receive
3803 * @n: entry to schedule
3805 * The entry's receive function will be scheduled to run
3807 void __napi_schedule(struct napi_struct *n)
3809 unsigned long flags;
3811 local_irq_save(flags);
3812 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3813 local_irq_restore(flags);
3815 EXPORT_SYMBOL(__napi_schedule);
3817 void __napi_complete(struct napi_struct *n)
3819 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3820 BUG_ON(n->gro_list);
3822 list_del(&n->poll_list);
3823 smp_mb__before_clear_bit();
3824 clear_bit(NAPI_STATE_SCHED, &n->state);
3826 EXPORT_SYMBOL(__napi_complete);
3828 void napi_complete(struct napi_struct *n)
3830 unsigned long flags;
3833 * don't let napi dequeue from the cpu poll list
3834 * just in case its running on a different cpu
3836 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3840 local_irq_save(flags);
3842 local_irq_restore(flags);
3844 EXPORT_SYMBOL(napi_complete);
3846 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3847 int (*poll)(struct napi_struct *, int), int weight)
3849 INIT_LIST_HEAD(&napi->poll_list);
3850 napi->gro_count = 0;
3851 napi->gro_list = NULL;
3854 napi->weight = weight;
3855 list_add(&napi->dev_list, &dev->napi_list);
3857 #ifdef CONFIG_NETPOLL
3858 spin_lock_init(&napi->poll_lock);
3859 napi->poll_owner = -1;
3861 set_bit(NAPI_STATE_SCHED, &napi->state);
3863 EXPORT_SYMBOL(netif_napi_add);
3865 void netif_napi_del(struct napi_struct *napi)
3867 struct sk_buff *skb, *next;
3869 list_del_init(&napi->dev_list);
3870 napi_free_frags(napi);
3872 for (skb = napi->gro_list; skb; skb = next) {
3878 napi->gro_list = NULL;
3879 napi->gro_count = 0;
3881 EXPORT_SYMBOL(netif_napi_del);
3883 static void net_rx_action(struct softirq_action *h)
3885 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3886 unsigned long time_limit = jiffies + 2;
3887 int budget = netdev_budget;
3890 local_irq_disable();
3892 while (!list_empty(&sd->poll_list)) {
3893 struct napi_struct *n;
3896 /* If softirq window is exhuasted then punt.
3897 * Allow this to run for 2 jiffies since which will allow
3898 * an average latency of 1.5/HZ.
3900 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3905 /* Even though interrupts have been re-enabled, this
3906 * access is safe because interrupts can only add new
3907 * entries to the tail of this list, and only ->poll()
3908 * calls can remove this head entry from the list.
3910 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3912 have = netpoll_poll_lock(n);
3916 /* This NAPI_STATE_SCHED test is for avoiding a race
3917 * with netpoll's poll_napi(). Only the entity which
3918 * obtains the lock and sees NAPI_STATE_SCHED set will
3919 * actually make the ->poll() call. Therefore we avoid
3920 * accidentally calling ->poll() when NAPI is not scheduled.
3923 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3924 work = n->poll(n, weight);
3928 WARN_ON_ONCE(work > weight);
3932 local_irq_disable();
3934 /* Drivers must not modify the NAPI state if they
3935 * consume the entire weight. In such cases this code
3936 * still "owns" the NAPI instance and therefore can
3937 * move the instance around on the list at-will.
3939 if (unlikely(work == weight)) {
3940 if (unlikely(napi_disable_pending(n))) {
3943 local_irq_disable();
3945 list_move_tail(&n->poll_list, &sd->poll_list);
3948 netpoll_poll_unlock(have);
3951 net_rps_action_and_irq_enable(sd);
3953 #ifdef CONFIG_NET_DMA
3955 * There may not be any more sk_buffs coming right now, so push
3956 * any pending DMA copies to hardware
3958 dma_issue_pending_all();
3965 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3969 static gifconf_func_t *gifconf_list[NPROTO];
3972 * register_gifconf - register a SIOCGIF handler
3973 * @family: Address family
3974 * @gifconf: Function handler
3976 * Register protocol dependent address dumping routines. The handler
3977 * that is passed must not be freed or reused until it has been replaced
3978 * by another handler.
3980 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3982 if (family >= NPROTO)
3984 gifconf_list[family] = gifconf;
3987 EXPORT_SYMBOL(register_gifconf);
3991 * Map an interface index to its name (SIOCGIFNAME)
3995 * We need this ioctl for efficient implementation of the
3996 * if_indextoname() function required by the IPv6 API. Without
3997 * it, we would have to search all the interfaces to find a
4001 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4003 struct net_device *dev;
4007 * Fetch the caller's info block.
4010 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4014 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4020 strcpy(ifr.ifr_name, dev->name);
4023 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4029 * Perform a SIOCGIFCONF call. This structure will change
4030 * size eventually, and there is nothing I can do about it.
4031 * Thus we will need a 'compatibility mode'.
4034 static int dev_ifconf(struct net *net, char __user *arg)
4037 struct net_device *dev;
4044 * Fetch the caller's info block.
4047 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4054 * Loop over the interfaces, and write an info block for each.
4058 for_each_netdev(net, dev) {
4059 for (i = 0; i < NPROTO; i++) {
4060 if (gifconf_list[i]) {
4063 done = gifconf_list[i](dev, NULL, 0);
4065 done = gifconf_list[i](dev, pos + total,
4075 * All done. Write the updated control block back to the caller.
4077 ifc.ifc_len = total;
4080 * Both BSD and Solaris return 0 here, so we do too.
4082 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4085 #ifdef CONFIG_PROC_FS
4087 #define BUCKET_SPACE (32 - NETDEV_HASHBITS)
4089 struct dev_iter_state {
4090 struct seq_net_private p;
4091 unsigned int pos; /* bucket << BUCKET_SPACE + offset */
4094 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4095 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4096 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4098 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq)
4100 struct dev_iter_state *state = seq->private;
4101 struct net *net = seq_file_net(seq);
4102 struct net_device *dev;
4103 struct hlist_node *p;
4104 struct hlist_head *h;
4105 unsigned int count, bucket, offset;
4107 bucket = get_bucket(state->pos);
4108 offset = get_offset(state->pos);
4109 h = &net->dev_name_head[bucket];
4111 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4112 if (count++ == offset) {
4113 state->pos = set_bucket_offset(bucket, count);
4121 static inline struct net_device *dev_from_new_bucket(struct seq_file *seq)
4123 struct dev_iter_state *state = seq->private;
4124 struct net_device *dev;
4125 unsigned int bucket;
4127 bucket = get_bucket(state->pos);
4129 dev = dev_from_same_bucket(seq);
4134 state->pos = set_bucket_offset(bucket, 0);
4135 } while (bucket < NETDEV_HASHENTRIES);
4141 * This is invoked by the /proc filesystem handler to display a device
4144 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4147 struct dev_iter_state *state = seq->private;
4151 return SEQ_START_TOKEN;
4153 /* check for end of the hash */
4154 if (state->pos == 0 && *pos > 1)
4157 return dev_from_new_bucket(seq);
4160 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4162 struct net_device *dev;
4166 if (v == SEQ_START_TOKEN)
4167 return dev_from_new_bucket(seq);
4169 dev = dev_from_same_bucket(seq);
4173 return dev_from_new_bucket(seq);
4176 void dev_seq_stop(struct seq_file *seq, void *v)
4182 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4184 struct rtnl_link_stats64 temp;
4185 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4187 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4188 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4189 dev->name, stats->rx_bytes, stats->rx_packets,
4191 stats->rx_dropped + stats->rx_missed_errors,
4192 stats->rx_fifo_errors,
4193 stats->rx_length_errors + stats->rx_over_errors +
4194 stats->rx_crc_errors + stats->rx_frame_errors,
4195 stats->rx_compressed, stats->multicast,
4196 stats->tx_bytes, stats->tx_packets,
4197 stats->tx_errors, stats->tx_dropped,
4198 stats->tx_fifo_errors, stats->collisions,
4199 stats->tx_carrier_errors +
4200 stats->tx_aborted_errors +
4201 stats->tx_window_errors +
4202 stats->tx_heartbeat_errors,
4203 stats->tx_compressed);
4207 * Called from the PROCfs module. This now uses the new arbitrary sized
4208 * /proc/net interface to create /proc/net/dev
4210 static int dev_seq_show(struct seq_file *seq, void *v)
4212 if (v == SEQ_START_TOKEN)
4213 seq_puts(seq, "Inter-| Receive "
4215 " face |bytes packets errs drop fifo frame "
4216 "compressed multicast|bytes packets errs "
4217 "drop fifo colls carrier compressed\n");
4219 dev_seq_printf_stats(seq, v);
4223 static struct softnet_data *softnet_get_online(loff_t *pos)
4225 struct softnet_data *sd = NULL;
4227 while (*pos < nr_cpu_ids)
4228 if (cpu_online(*pos)) {
4229 sd = &per_cpu(softnet_data, *pos);
4236 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4238 return softnet_get_online(pos);
4241 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4244 return softnet_get_online(pos);
4247 static void softnet_seq_stop(struct seq_file *seq, void *v)
4251 static int softnet_seq_show(struct seq_file *seq, void *v)
4253 struct softnet_data *sd = v;
4255 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4256 sd->processed, sd->dropped, sd->time_squeeze, 0,
4257 0, 0, 0, 0, /* was fastroute */
4258 sd->cpu_collision, sd->received_rps);
4262 static const struct seq_operations dev_seq_ops = {
4263 .start = dev_seq_start,
4264 .next = dev_seq_next,
4265 .stop = dev_seq_stop,
4266 .show = dev_seq_show,
4269 static int dev_seq_open(struct inode *inode, struct file *file)
4271 return seq_open_net(inode, file, &dev_seq_ops,
4272 sizeof(struct dev_iter_state));
4275 static const struct file_operations dev_seq_fops = {
4276 .owner = THIS_MODULE,
4277 .open = dev_seq_open,
4279 .llseek = seq_lseek,
4280 .release = seq_release_net,
4283 static const struct seq_operations softnet_seq_ops = {
4284 .start = softnet_seq_start,
4285 .next = softnet_seq_next,
4286 .stop = softnet_seq_stop,
4287 .show = softnet_seq_show,
4290 static int softnet_seq_open(struct inode *inode, struct file *file)
4292 return seq_open(file, &softnet_seq_ops);
4295 static const struct file_operations softnet_seq_fops = {
4296 .owner = THIS_MODULE,
4297 .open = softnet_seq_open,
4299 .llseek = seq_lseek,
4300 .release = seq_release,
4303 static void *ptype_get_idx(loff_t pos)
4305 struct packet_type *pt = NULL;
4309 list_for_each_entry_rcu(pt, &ptype_all, list) {
4315 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4316 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4325 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4329 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4332 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4334 struct packet_type *pt;
4335 struct list_head *nxt;
4339 if (v == SEQ_START_TOKEN)
4340 return ptype_get_idx(0);
4343 nxt = pt->list.next;
4344 if (pt->type == htons(ETH_P_ALL)) {
4345 if (nxt != &ptype_all)
4348 nxt = ptype_base[0].next;
4350 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4352 while (nxt == &ptype_base[hash]) {
4353 if (++hash >= PTYPE_HASH_SIZE)
4355 nxt = ptype_base[hash].next;
4358 return list_entry(nxt, struct packet_type, list);
4361 static void ptype_seq_stop(struct seq_file *seq, void *v)
4367 static int ptype_seq_show(struct seq_file *seq, void *v)
4369 struct packet_type *pt = v;
4371 if (v == SEQ_START_TOKEN)
4372 seq_puts(seq, "Type Device Function\n");
4373 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4374 if (pt->type == htons(ETH_P_ALL))
4375 seq_puts(seq, "ALL ");
4377 seq_printf(seq, "%04x", ntohs(pt->type));
4379 seq_printf(seq, " %-8s %pF\n",
4380 pt->dev ? pt->dev->name : "", pt->func);
4386 static const struct seq_operations ptype_seq_ops = {
4387 .start = ptype_seq_start,
4388 .next = ptype_seq_next,
4389 .stop = ptype_seq_stop,
4390 .show = ptype_seq_show,
4393 static int ptype_seq_open(struct inode *inode, struct file *file)
4395 return seq_open_net(inode, file, &ptype_seq_ops,
4396 sizeof(struct seq_net_private));
4399 static const struct file_operations ptype_seq_fops = {
4400 .owner = THIS_MODULE,
4401 .open = ptype_seq_open,
4403 .llseek = seq_lseek,
4404 .release = seq_release_net,
4408 static int __net_init dev_proc_net_init(struct net *net)
4412 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4414 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4416 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4419 if (wext_proc_init(net))
4425 proc_net_remove(net, "ptype");
4427 proc_net_remove(net, "softnet_stat");
4429 proc_net_remove(net, "dev");
4433 static void __net_exit dev_proc_net_exit(struct net *net)
4435 wext_proc_exit(net);
4437 proc_net_remove(net, "ptype");
4438 proc_net_remove(net, "softnet_stat");
4439 proc_net_remove(net, "dev");
4442 static struct pernet_operations __net_initdata dev_proc_ops = {
4443 .init = dev_proc_net_init,
4444 .exit = dev_proc_net_exit,
4447 static int __init dev_proc_init(void)
4449 return register_pernet_subsys(&dev_proc_ops);
4452 #define dev_proc_init() 0
4453 #endif /* CONFIG_PROC_FS */
4457 * netdev_set_master - set up master pointer
4458 * @slave: slave device
4459 * @master: new master device
4461 * Changes the master device of the slave. Pass %NULL to break the
4462 * bonding. The caller must hold the RTNL semaphore. On a failure
4463 * a negative errno code is returned. On success the reference counts
4464 * are adjusted and the function returns zero.
4466 int netdev_set_master(struct net_device *slave, struct net_device *master)
4468 struct net_device *old = slave->master;
4478 slave->master = master;
4484 EXPORT_SYMBOL(netdev_set_master);
4487 * netdev_set_bond_master - set up bonding master/slave pair
4488 * @slave: slave device
4489 * @master: new master device
4491 * Changes the master device of the slave. Pass %NULL to break the
4492 * bonding. The caller must hold the RTNL semaphore. On a failure
4493 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4494 * to the routing socket and the function returns zero.
4496 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4502 err = netdev_set_master(slave, master);
4506 slave->flags |= IFF_SLAVE;
4508 slave->flags &= ~IFF_SLAVE;
4510 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4513 EXPORT_SYMBOL(netdev_set_bond_master);
4515 static void dev_change_rx_flags(struct net_device *dev, int flags)
4517 const struct net_device_ops *ops = dev->netdev_ops;
4519 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4520 ops->ndo_change_rx_flags(dev, flags);
4523 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4525 unsigned short old_flags = dev->flags;
4531 dev->flags |= IFF_PROMISC;
4532 dev->promiscuity += inc;
4533 if (dev->promiscuity == 0) {
4536 * If inc causes overflow, untouch promisc and return error.
4539 dev->flags &= ~IFF_PROMISC;
4541 dev->promiscuity -= inc;
4542 printk(KERN_WARNING "%s: promiscuity touches roof, "
4543 "set promiscuity failed, promiscuity feature "
4544 "of device might be broken.\n", dev->name);
4548 if (dev->flags != old_flags) {
4549 printk(KERN_INFO "device %s %s promiscuous mode\n",
4550 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4552 if (audit_enabled) {
4553 current_uid_gid(&uid, &gid);
4554 audit_log(current->audit_context, GFP_ATOMIC,
4555 AUDIT_ANOM_PROMISCUOUS,
4556 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4557 dev->name, (dev->flags & IFF_PROMISC),
4558 (old_flags & IFF_PROMISC),
4559 audit_get_loginuid(current),
4561 audit_get_sessionid(current));
4564 dev_change_rx_flags(dev, IFF_PROMISC);
4570 * dev_set_promiscuity - update promiscuity count on a device
4574 * Add or remove promiscuity from a device. While the count in the device
4575 * remains above zero the interface remains promiscuous. Once it hits zero
4576 * the device reverts back to normal filtering operation. A negative inc
4577 * value is used to drop promiscuity on the device.
4578 * Return 0 if successful or a negative errno code on error.
4580 int dev_set_promiscuity(struct net_device *dev, int inc)
4582 unsigned short old_flags = dev->flags;
4585 err = __dev_set_promiscuity(dev, inc);
4588 if (dev->flags != old_flags)
4589 dev_set_rx_mode(dev);
4592 EXPORT_SYMBOL(dev_set_promiscuity);
4595 * dev_set_allmulti - update allmulti count on a device
4599 * Add or remove reception of all multicast frames to a device. While the
4600 * count in the device remains above zero the interface remains listening
4601 * to all interfaces. Once it hits zero the device reverts back to normal
4602 * filtering operation. A negative @inc value is used to drop the counter
4603 * when releasing a resource needing all multicasts.
4604 * Return 0 if successful or a negative errno code on error.
4607 int dev_set_allmulti(struct net_device *dev, int inc)
4609 unsigned short old_flags = dev->flags;
4613 dev->flags |= IFF_ALLMULTI;
4614 dev->allmulti += inc;
4615 if (dev->allmulti == 0) {
4618 * If inc causes overflow, untouch allmulti and return error.
4621 dev->flags &= ~IFF_ALLMULTI;
4623 dev->allmulti -= inc;
4624 printk(KERN_WARNING "%s: allmulti touches roof, "
4625 "set allmulti failed, allmulti feature of "
4626 "device might be broken.\n", dev->name);
4630 if (dev->flags ^ old_flags) {
4631 dev_change_rx_flags(dev, IFF_ALLMULTI);
4632 dev_set_rx_mode(dev);
4636 EXPORT_SYMBOL(dev_set_allmulti);
4639 * Upload unicast and multicast address lists to device and
4640 * configure RX filtering. When the device doesn't support unicast
4641 * filtering it is put in promiscuous mode while unicast addresses
4644 void __dev_set_rx_mode(struct net_device *dev)
4646 const struct net_device_ops *ops = dev->netdev_ops;
4648 /* dev_open will call this function so the list will stay sane. */
4649 if (!(dev->flags&IFF_UP))
4652 if (!netif_device_present(dev))
4655 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4656 /* Unicast addresses changes may only happen under the rtnl,
4657 * therefore calling __dev_set_promiscuity here is safe.
4659 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4660 __dev_set_promiscuity(dev, 1);
4661 dev->uc_promisc = true;
4662 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4663 __dev_set_promiscuity(dev, -1);
4664 dev->uc_promisc = false;
4668 if (ops->ndo_set_rx_mode)
4669 ops->ndo_set_rx_mode(dev);
4672 void dev_set_rx_mode(struct net_device *dev)
4674 netif_addr_lock_bh(dev);
4675 __dev_set_rx_mode(dev);
4676 netif_addr_unlock_bh(dev);
4680 * dev_get_flags - get flags reported to userspace
4683 * Get the combination of flag bits exported through APIs to userspace.
4685 unsigned dev_get_flags(const struct net_device *dev)
4689 flags = (dev->flags & ~(IFF_PROMISC |
4694 (dev->gflags & (IFF_PROMISC |
4697 if (netif_running(dev)) {
4698 if (netif_oper_up(dev))
4699 flags |= IFF_RUNNING;
4700 if (netif_carrier_ok(dev))
4701 flags |= IFF_LOWER_UP;
4702 if (netif_dormant(dev))
4703 flags |= IFF_DORMANT;
4708 EXPORT_SYMBOL(dev_get_flags);
4710 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4712 int old_flags = dev->flags;
4718 * Set the flags on our device.
4721 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4722 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4724 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4728 * Load in the correct multicast list now the flags have changed.
4731 if ((old_flags ^ flags) & IFF_MULTICAST)
4732 dev_change_rx_flags(dev, IFF_MULTICAST);
4734 dev_set_rx_mode(dev);
4737 * Have we downed the interface. We handle IFF_UP ourselves
4738 * according to user attempts to set it, rather than blindly
4743 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4744 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4747 dev_set_rx_mode(dev);
4750 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4751 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4753 dev->gflags ^= IFF_PROMISC;
4754 dev_set_promiscuity(dev, inc);
4757 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4758 is important. Some (broken) drivers set IFF_PROMISC, when
4759 IFF_ALLMULTI is requested not asking us and not reporting.
4761 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4762 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4764 dev->gflags ^= IFF_ALLMULTI;
4765 dev_set_allmulti(dev, inc);
4771 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4773 unsigned int changes = dev->flags ^ old_flags;
4775 if (changes & IFF_UP) {
4776 if (dev->flags & IFF_UP)
4777 call_netdevice_notifiers(NETDEV_UP, dev);
4779 call_netdevice_notifiers(NETDEV_DOWN, dev);
4782 if (dev->flags & IFF_UP &&
4783 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4784 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4788 * dev_change_flags - change device settings
4790 * @flags: device state flags
4792 * Change settings on device based state flags. The flags are
4793 * in the userspace exported format.
4795 int dev_change_flags(struct net_device *dev, unsigned flags)
4798 int old_flags = dev->flags;
4800 ret = __dev_change_flags(dev, flags);
4804 changes = old_flags ^ dev->flags;
4806 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4808 __dev_notify_flags(dev, old_flags);
4811 EXPORT_SYMBOL(dev_change_flags);
4814 * dev_set_mtu - Change maximum transfer unit
4816 * @new_mtu: new transfer unit
4818 * Change the maximum transfer size of the network device.
4820 int dev_set_mtu(struct net_device *dev, int new_mtu)
4822 const struct net_device_ops *ops = dev->netdev_ops;
4825 if (new_mtu == dev->mtu)
4828 /* MTU must be positive. */
4832 if (!netif_device_present(dev))
4836 if (ops->ndo_change_mtu)
4837 err = ops->ndo_change_mtu(dev, new_mtu);
4841 if (!err && dev->flags & IFF_UP)
4842 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4845 EXPORT_SYMBOL(dev_set_mtu);
4848 * dev_set_group - Change group this device belongs to
4850 * @new_group: group this device should belong to
4852 void dev_set_group(struct net_device *dev, int new_group)
4854 dev->group = new_group;
4856 EXPORT_SYMBOL(dev_set_group);
4859 * dev_set_mac_address - Change Media Access Control Address
4863 * Change the hardware (MAC) address of the device
4865 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4867 const struct net_device_ops *ops = dev->netdev_ops;
4870 if (!ops->ndo_set_mac_address)
4872 if (sa->sa_family != dev->type)
4874 if (!netif_device_present(dev))
4876 err = ops->ndo_set_mac_address(dev, sa);
4878 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4881 EXPORT_SYMBOL(dev_set_mac_address);
4884 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4886 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4889 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4895 case SIOCGIFFLAGS: /* Get interface flags */
4896 ifr->ifr_flags = (short) dev_get_flags(dev);
4899 case SIOCGIFMETRIC: /* Get the metric on the interface
4900 (currently unused) */
4901 ifr->ifr_metric = 0;
4904 case SIOCGIFMTU: /* Get the MTU of a device */
4905 ifr->ifr_mtu = dev->mtu;
4910 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4912 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4913 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4914 ifr->ifr_hwaddr.sa_family = dev->type;
4922 ifr->ifr_map.mem_start = dev->mem_start;
4923 ifr->ifr_map.mem_end = dev->mem_end;
4924 ifr->ifr_map.base_addr = dev->base_addr;
4925 ifr->ifr_map.irq = dev->irq;
4926 ifr->ifr_map.dma = dev->dma;
4927 ifr->ifr_map.port = dev->if_port;
4931 ifr->ifr_ifindex = dev->ifindex;
4935 ifr->ifr_qlen = dev->tx_queue_len;
4939 /* dev_ioctl() should ensure this case
4951 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4953 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4956 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4957 const struct net_device_ops *ops;
4962 ops = dev->netdev_ops;
4965 case SIOCSIFFLAGS: /* Set interface flags */
4966 return dev_change_flags(dev, ifr->ifr_flags);
4968 case SIOCSIFMETRIC: /* Set the metric on the interface
4969 (currently unused) */
4972 case SIOCSIFMTU: /* Set the MTU of a device */
4973 return dev_set_mtu(dev, ifr->ifr_mtu);
4976 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4978 case SIOCSIFHWBROADCAST:
4979 if (ifr->ifr_hwaddr.sa_family != dev->type)
4981 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4982 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4983 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4987 if (ops->ndo_set_config) {
4988 if (!netif_device_present(dev))
4990 return ops->ndo_set_config(dev, &ifr->ifr_map);
4995 if (!ops->ndo_set_rx_mode ||
4996 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4998 if (!netif_device_present(dev))
5000 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5003 if (!ops->ndo_set_rx_mode ||
5004 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5006 if (!netif_device_present(dev))
5008 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5011 if (ifr->ifr_qlen < 0)
5013 dev->tx_queue_len = ifr->ifr_qlen;
5017 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5018 return dev_change_name(dev, ifr->ifr_newname);
5021 err = net_hwtstamp_validate(ifr);
5027 * Unknown or private ioctl
5030 if ((cmd >= SIOCDEVPRIVATE &&
5031 cmd <= SIOCDEVPRIVATE + 15) ||
5032 cmd == SIOCBONDENSLAVE ||
5033 cmd == SIOCBONDRELEASE ||
5034 cmd == SIOCBONDSETHWADDR ||
5035 cmd == SIOCBONDSLAVEINFOQUERY ||
5036 cmd == SIOCBONDINFOQUERY ||
5037 cmd == SIOCBONDCHANGEACTIVE ||
5038 cmd == SIOCGMIIPHY ||
5039 cmd == SIOCGMIIREG ||
5040 cmd == SIOCSMIIREG ||
5041 cmd == SIOCBRADDIF ||
5042 cmd == SIOCBRDELIF ||
5043 cmd == SIOCSHWTSTAMP ||
5044 cmd == SIOCWANDEV) {
5046 if (ops->ndo_do_ioctl) {
5047 if (netif_device_present(dev))
5048 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5060 * This function handles all "interface"-type I/O control requests. The actual
5061 * 'doing' part of this is dev_ifsioc above.
5065 * dev_ioctl - network device ioctl
5066 * @net: the applicable net namespace
5067 * @cmd: command to issue
5068 * @arg: pointer to a struct ifreq in user space
5070 * Issue ioctl functions to devices. This is normally called by the
5071 * user space syscall interfaces but can sometimes be useful for
5072 * other purposes. The return value is the return from the syscall if
5073 * positive or a negative errno code on error.
5076 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5082 /* One special case: SIOCGIFCONF takes ifconf argument
5083 and requires shared lock, because it sleeps writing
5087 if (cmd == SIOCGIFCONF) {
5089 ret = dev_ifconf(net, (char __user *) arg);
5093 if (cmd == SIOCGIFNAME)
5094 return dev_ifname(net, (struct ifreq __user *)arg);
5096 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5099 ifr.ifr_name[IFNAMSIZ-1] = 0;
5101 colon = strchr(ifr.ifr_name, ':');
5106 * See which interface the caller is talking about.
5111 * These ioctl calls:
5112 * - can be done by all.
5113 * - atomic and do not require locking.
5124 dev_load(net, ifr.ifr_name);
5126 ret = dev_ifsioc_locked(net, &ifr, cmd);
5131 if (copy_to_user(arg, &ifr,
5132 sizeof(struct ifreq)))
5138 dev_load(net, ifr.ifr_name);
5140 ret = dev_ethtool(net, &ifr);
5145 if (copy_to_user(arg, &ifr,
5146 sizeof(struct ifreq)))
5152 * These ioctl calls:
5153 * - require superuser power.
5154 * - require strict serialization.
5160 if (!capable(CAP_NET_ADMIN))
5162 dev_load(net, ifr.ifr_name);
5164 ret = dev_ifsioc(net, &ifr, cmd);
5169 if (copy_to_user(arg, &ifr,
5170 sizeof(struct ifreq)))
5176 * These ioctl calls:
5177 * - require superuser power.
5178 * - require strict serialization.
5179 * - do not return a value
5189 case SIOCSIFHWBROADCAST:
5192 case SIOCBONDENSLAVE:
5193 case SIOCBONDRELEASE:
5194 case SIOCBONDSETHWADDR:
5195 case SIOCBONDCHANGEACTIVE:
5199 if (!capable(CAP_NET_ADMIN))
5202 case SIOCBONDSLAVEINFOQUERY:
5203 case SIOCBONDINFOQUERY:
5204 dev_load(net, ifr.ifr_name);
5206 ret = dev_ifsioc(net, &ifr, cmd);
5211 /* Get the per device memory space. We can add this but
5212 * currently do not support it */
5214 /* Set the per device memory buffer space.
5215 * Not applicable in our case */
5220 * Unknown or private ioctl.
5223 if (cmd == SIOCWANDEV ||
5224 (cmd >= SIOCDEVPRIVATE &&
5225 cmd <= SIOCDEVPRIVATE + 15)) {
5226 dev_load(net, ifr.ifr_name);
5228 ret = dev_ifsioc(net, &ifr, cmd);
5230 if (!ret && copy_to_user(arg, &ifr,
5231 sizeof(struct ifreq)))
5235 /* Take care of Wireless Extensions */
5236 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5237 return wext_handle_ioctl(net, &ifr, cmd, arg);
5244 * dev_new_index - allocate an ifindex
5245 * @net: the applicable net namespace
5247 * Returns a suitable unique value for a new device interface
5248 * number. The caller must hold the rtnl semaphore or the
5249 * dev_base_lock to be sure it remains unique.
5251 static int dev_new_index(struct net *net)
5257 if (!__dev_get_by_index(net, ifindex))
5262 /* Delayed registration/unregisteration */
5263 static LIST_HEAD(net_todo_list);
5265 static void net_set_todo(struct net_device *dev)
5267 list_add_tail(&dev->todo_list, &net_todo_list);
5270 static void rollback_registered_many(struct list_head *head)
5272 struct net_device *dev, *tmp;
5274 BUG_ON(dev_boot_phase);
5277 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5278 /* Some devices call without registering
5279 * for initialization unwind. Remove those
5280 * devices and proceed with the remaining.
5282 if (dev->reg_state == NETREG_UNINITIALIZED) {
5283 pr_debug("unregister_netdevice: device %s/%p never "
5284 "was registered\n", dev->name, dev);
5287 list_del(&dev->unreg_list);
5290 dev->dismantle = true;
5291 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5294 /* If device is running, close it first. */
5295 dev_close_many(head);
5297 list_for_each_entry(dev, head, unreg_list) {
5298 /* And unlink it from device chain. */
5299 unlist_netdevice(dev);
5301 dev->reg_state = NETREG_UNREGISTERING;
5306 list_for_each_entry(dev, head, unreg_list) {
5307 /* Shutdown queueing discipline. */
5311 /* Notify protocols, that we are about to destroy
5312 this device. They should clean all the things.
5314 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5316 if (!dev->rtnl_link_ops ||
5317 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5318 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5321 * Flush the unicast and multicast chains
5326 if (dev->netdev_ops->ndo_uninit)
5327 dev->netdev_ops->ndo_uninit(dev);
5329 /* Notifier chain MUST detach us from master device. */
5330 WARN_ON(dev->master);
5332 /* Remove entries from kobject tree */
5333 netdev_unregister_kobject(dev);
5336 /* Process any work delayed until the end of the batch */
5337 dev = list_first_entry(head, struct net_device, unreg_list);
5338 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5342 list_for_each_entry(dev, head, unreg_list)
5346 static void rollback_registered(struct net_device *dev)
5350 list_add(&dev->unreg_list, &single);
5351 rollback_registered_many(&single);
5355 static netdev_features_t netdev_fix_features(struct net_device *dev,
5356 netdev_features_t features)
5358 /* Fix illegal checksum combinations */
5359 if ((features & NETIF_F_HW_CSUM) &&
5360 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5361 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5362 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5365 /* Fix illegal SG+CSUM combinations. */
5366 if ((features & NETIF_F_SG) &&
5367 !(features & NETIF_F_ALL_CSUM)) {
5369 "Dropping NETIF_F_SG since no checksum feature.\n");
5370 features &= ~NETIF_F_SG;
5373 /* TSO requires that SG is present as well. */
5374 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5375 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5376 features &= ~NETIF_F_ALL_TSO;
5379 /* TSO ECN requires that TSO is present as well. */
5380 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5381 features &= ~NETIF_F_TSO_ECN;
5383 /* Software GSO depends on SG. */
5384 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5385 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5386 features &= ~NETIF_F_GSO;
5389 /* UFO needs SG and checksumming */
5390 if (features & NETIF_F_UFO) {
5391 /* maybe split UFO into V4 and V6? */
5392 if (!((features & NETIF_F_GEN_CSUM) ||
5393 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5394 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5396 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5397 features &= ~NETIF_F_UFO;
5400 if (!(features & NETIF_F_SG)) {
5402 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5403 features &= ~NETIF_F_UFO;
5410 int __netdev_update_features(struct net_device *dev)
5412 netdev_features_t features;
5417 features = netdev_get_wanted_features(dev);
5419 if (dev->netdev_ops->ndo_fix_features)
5420 features = dev->netdev_ops->ndo_fix_features(dev, features);
5422 /* driver might be less strict about feature dependencies */
5423 features = netdev_fix_features(dev, features);
5425 if (dev->features == features)
5428 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5429 &dev->features, &features);
5431 if (dev->netdev_ops->ndo_set_features)
5432 err = dev->netdev_ops->ndo_set_features(dev, features);
5434 if (unlikely(err < 0)) {
5436 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5437 err, &features, &dev->features);
5442 dev->features = features;
5448 * netdev_update_features - recalculate device features
5449 * @dev: the device to check
5451 * Recalculate dev->features set and send notifications if it
5452 * has changed. Should be called after driver or hardware dependent
5453 * conditions might have changed that influence the features.
5455 void netdev_update_features(struct net_device *dev)
5457 if (__netdev_update_features(dev))
5458 netdev_features_change(dev);
5460 EXPORT_SYMBOL(netdev_update_features);
5463 * netdev_change_features - recalculate device features
5464 * @dev: the device to check
5466 * Recalculate dev->features set and send notifications even
5467 * if they have not changed. Should be called instead of
5468 * netdev_update_features() if also dev->vlan_features might
5469 * have changed to allow the changes to be propagated to stacked
5472 void netdev_change_features(struct net_device *dev)
5474 __netdev_update_features(dev);
5475 netdev_features_change(dev);
5477 EXPORT_SYMBOL(netdev_change_features);
5480 * netif_stacked_transfer_operstate - transfer operstate
5481 * @rootdev: the root or lower level device to transfer state from
5482 * @dev: the device to transfer operstate to
5484 * Transfer operational state from root to device. This is normally
5485 * called when a stacking relationship exists between the root
5486 * device and the device(a leaf device).
5488 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5489 struct net_device *dev)
5491 if (rootdev->operstate == IF_OPER_DORMANT)
5492 netif_dormant_on(dev);
5494 netif_dormant_off(dev);
5496 if (netif_carrier_ok(rootdev)) {
5497 if (!netif_carrier_ok(dev))
5498 netif_carrier_on(dev);
5500 if (netif_carrier_ok(dev))
5501 netif_carrier_off(dev);
5504 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5507 static int netif_alloc_rx_queues(struct net_device *dev)
5509 unsigned int i, count = dev->num_rx_queues;
5510 struct netdev_rx_queue *rx;
5514 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5516 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5521 for (i = 0; i < count; i++)
5527 static void netdev_init_one_queue(struct net_device *dev,
5528 struct netdev_queue *queue, void *_unused)
5530 /* Initialize queue lock */
5531 spin_lock_init(&queue->_xmit_lock);
5532 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5533 queue->xmit_lock_owner = -1;
5534 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5538 static int netif_alloc_netdev_queues(struct net_device *dev)
5540 unsigned int count = dev->num_tx_queues;
5541 struct netdev_queue *tx;
5545 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5547 pr_err("netdev: Unable to allocate %u tx queues.\n",
5553 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5554 spin_lock_init(&dev->tx_global_lock);
5560 * register_netdevice - register a network device
5561 * @dev: device to register
5563 * Take a completed network device structure and add it to the kernel
5564 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5565 * chain. 0 is returned on success. A negative errno code is returned
5566 * on a failure to set up the device, or if the name is a duplicate.
5568 * Callers must hold the rtnl semaphore. You may want
5569 * register_netdev() instead of this.
5572 * The locking appears insufficient to guarantee two parallel registers
5573 * will not get the same name.
5576 int register_netdevice(struct net_device *dev)
5579 struct net *net = dev_net(dev);
5581 BUG_ON(dev_boot_phase);
5586 /* When net_device's are persistent, this will be fatal. */
5587 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5590 spin_lock_init(&dev->addr_list_lock);
5591 netdev_set_addr_lockdep_class(dev);
5595 ret = dev_get_valid_name(dev, dev->name);
5599 /* Init, if this function is available */
5600 if (dev->netdev_ops->ndo_init) {
5601 ret = dev->netdev_ops->ndo_init(dev);
5609 dev->ifindex = dev_new_index(net);
5610 if (dev->iflink == -1)
5611 dev->iflink = dev->ifindex;
5613 /* Transfer changeable features to wanted_features and enable
5614 * software offloads (GSO and GRO).
5616 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5617 dev->features |= NETIF_F_SOFT_FEATURES;
5618 dev->wanted_features = dev->features & dev->hw_features;
5620 /* Turn on no cache copy if HW is doing checksum */
5621 if (!(dev->flags & IFF_LOOPBACK)) {
5622 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5623 if (dev->features & NETIF_F_ALL_CSUM) {
5624 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5625 dev->features |= NETIF_F_NOCACHE_COPY;
5629 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5631 dev->vlan_features |= NETIF_F_HIGHDMA;
5633 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5634 ret = notifier_to_errno(ret);
5638 ret = netdev_register_kobject(dev);
5641 dev->reg_state = NETREG_REGISTERED;
5643 __netdev_update_features(dev);
5646 * Default initial state at registry is that the
5647 * device is present.
5650 set_bit(__LINK_STATE_PRESENT, &dev->state);
5652 dev_init_scheduler(dev);
5654 list_netdevice(dev);
5656 /* Notify protocols, that a new device appeared. */
5657 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5658 ret = notifier_to_errno(ret);
5660 rollback_registered(dev);
5661 dev->reg_state = NETREG_UNREGISTERED;
5664 * Prevent userspace races by waiting until the network
5665 * device is fully setup before sending notifications.
5667 if (!dev->rtnl_link_ops ||
5668 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5669 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5675 if (dev->netdev_ops->ndo_uninit)
5676 dev->netdev_ops->ndo_uninit(dev);
5679 EXPORT_SYMBOL(register_netdevice);
5682 * init_dummy_netdev - init a dummy network device for NAPI
5683 * @dev: device to init
5685 * This takes a network device structure and initialize the minimum
5686 * amount of fields so it can be used to schedule NAPI polls without
5687 * registering a full blown interface. This is to be used by drivers
5688 * that need to tie several hardware interfaces to a single NAPI
5689 * poll scheduler due to HW limitations.
5691 int init_dummy_netdev(struct net_device *dev)
5693 /* Clear everything. Note we don't initialize spinlocks
5694 * are they aren't supposed to be taken by any of the
5695 * NAPI code and this dummy netdev is supposed to be
5696 * only ever used for NAPI polls
5698 memset(dev, 0, sizeof(struct net_device));
5700 /* make sure we BUG if trying to hit standard
5701 * register/unregister code path
5703 dev->reg_state = NETREG_DUMMY;
5705 /* NAPI wants this */
5706 INIT_LIST_HEAD(&dev->napi_list);
5708 /* a dummy interface is started by default */
5709 set_bit(__LINK_STATE_PRESENT, &dev->state);
5710 set_bit(__LINK_STATE_START, &dev->state);
5712 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5713 * because users of this 'device' dont need to change
5719 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5723 * register_netdev - register a network device
5724 * @dev: device to register
5726 * Take a completed network device structure and add it to the kernel
5727 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5728 * chain. 0 is returned on success. A negative errno code is returned
5729 * on a failure to set up the device, or if the name is a duplicate.
5731 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5732 * and expands the device name if you passed a format string to
5735 int register_netdev(struct net_device *dev)
5740 err = register_netdevice(dev);
5744 EXPORT_SYMBOL(register_netdev);
5746 int netdev_refcnt_read(const struct net_device *dev)
5750 for_each_possible_cpu(i)
5751 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5754 EXPORT_SYMBOL(netdev_refcnt_read);
5757 * netdev_wait_allrefs - wait until all references are gone.
5759 * This is called when unregistering network devices.
5761 * Any protocol or device that holds a reference should register
5762 * for netdevice notification, and cleanup and put back the
5763 * reference if they receive an UNREGISTER event.
5764 * We can get stuck here if buggy protocols don't correctly
5767 static void netdev_wait_allrefs(struct net_device *dev)
5769 unsigned long rebroadcast_time, warning_time;
5772 linkwatch_forget_dev(dev);
5774 rebroadcast_time = warning_time = jiffies;
5775 refcnt = netdev_refcnt_read(dev);
5777 while (refcnt != 0) {
5778 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5781 /* Rebroadcast unregister notification */
5782 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5783 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5784 * should have already handle it the first time */
5786 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5788 /* We must not have linkwatch events
5789 * pending on unregister. If this
5790 * happens, we simply run the queue
5791 * unscheduled, resulting in a noop
5794 linkwatch_run_queue();
5799 rebroadcast_time = jiffies;
5804 refcnt = netdev_refcnt_read(dev);
5806 if (time_after(jiffies, warning_time + 10 * HZ)) {
5807 printk(KERN_EMERG "unregister_netdevice: "
5808 "waiting for %s to become free. Usage "
5811 warning_time = jiffies;
5820 * register_netdevice(x1);
5821 * register_netdevice(x2);
5823 * unregister_netdevice(y1);
5824 * unregister_netdevice(y2);
5830 * We are invoked by rtnl_unlock().
5831 * This allows us to deal with problems:
5832 * 1) We can delete sysfs objects which invoke hotplug
5833 * without deadlocking with linkwatch via keventd.
5834 * 2) Since we run with the RTNL semaphore not held, we can sleep
5835 * safely in order to wait for the netdev refcnt to drop to zero.
5837 * We must not return until all unregister events added during
5838 * the interval the lock was held have been completed.
5840 void netdev_run_todo(void)
5842 struct list_head list;
5844 /* Snapshot list, allow later requests */
5845 list_replace_init(&net_todo_list, &list);
5849 /* Wait for rcu callbacks to finish before attempting to drain
5850 * the device list. This usually avoids a 250ms wait.
5852 if (!list_empty(&list))
5855 while (!list_empty(&list)) {
5856 struct net_device *dev
5857 = list_first_entry(&list, struct net_device, todo_list);
5858 list_del(&dev->todo_list);
5860 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5861 printk(KERN_ERR "network todo '%s' but state %d\n",
5862 dev->name, dev->reg_state);
5867 dev->reg_state = NETREG_UNREGISTERED;
5869 on_each_cpu(flush_backlog, dev, 1);
5871 netdev_wait_allrefs(dev);
5874 BUG_ON(netdev_refcnt_read(dev));
5875 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5876 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5877 WARN_ON(dev->dn_ptr);
5879 if (dev->destructor)
5880 dev->destructor(dev);
5882 /* Free network device */
5883 kobject_put(&dev->dev.kobj);
5887 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5888 * fields in the same order, with only the type differing.
5890 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5891 const struct net_device_stats *netdev_stats)
5893 #if BITS_PER_LONG == 64
5894 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5895 memcpy(stats64, netdev_stats, sizeof(*stats64));
5897 size_t i, n = sizeof(*stats64) / sizeof(u64);
5898 const unsigned long *src = (const unsigned long *)netdev_stats;
5899 u64 *dst = (u64 *)stats64;
5901 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5902 sizeof(*stats64) / sizeof(u64));
5903 for (i = 0; i < n; i++)
5909 * dev_get_stats - get network device statistics
5910 * @dev: device to get statistics from
5911 * @storage: place to store stats
5913 * Get network statistics from device. Return @storage.
5914 * The device driver may provide its own method by setting
5915 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5916 * otherwise the internal statistics structure is used.
5918 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5919 struct rtnl_link_stats64 *storage)
5921 const struct net_device_ops *ops = dev->netdev_ops;
5923 if (ops->ndo_get_stats64) {
5924 memset(storage, 0, sizeof(*storage));
5925 ops->ndo_get_stats64(dev, storage);
5926 } else if (ops->ndo_get_stats) {
5927 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5929 netdev_stats_to_stats64(storage, &dev->stats);
5931 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5934 EXPORT_SYMBOL(dev_get_stats);
5936 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5938 struct netdev_queue *queue = dev_ingress_queue(dev);
5940 #ifdef CONFIG_NET_CLS_ACT
5943 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5946 netdev_init_one_queue(dev, queue, NULL);
5947 queue->qdisc = &noop_qdisc;
5948 queue->qdisc_sleeping = &noop_qdisc;
5949 rcu_assign_pointer(dev->ingress_queue, queue);
5955 * alloc_netdev_mqs - allocate network device
5956 * @sizeof_priv: size of private data to allocate space for
5957 * @name: device name format string
5958 * @setup: callback to initialize device
5959 * @txqs: the number of TX subqueues to allocate
5960 * @rxqs: the number of RX subqueues to allocate
5962 * Allocates a struct net_device with private data area for driver use
5963 * and performs basic initialization. Also allocates subquue structs
5964 * for each queue on the device.
5966 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5967 void (*setup)(struct net_device *),
5968 unsigned int txqs, unsigned int rxqs)
5970 struct net_device *dev;
5972 struct net_device *p;
5974 BUG_ON(strlen(name) >= sizeof(dev->name));
5977 pr_err("alloc_netdev: Unable to allocate device "
5978 "with zero queues.\n");
5984 pr_err("alloc_netdev: Unable to allocate device "
5985 "with zero RX queues.\n");
5990 alloc_size = sizeof(struct net_device);
5992 /* ensure 32-byte alignment of private area */
5993 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5994 alloc_size += sizeof_priv;
5996 /* ensure 32-byte alignment of whole construct */
5997 alloc_size += NETDEV_ALIGN - 1;
5999 p = kzalloc(alloc_size, GFP_KERNEL);
6001 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
6005 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6006 dev->padded = (char *)dev - (char *)p;
6008 dev->pcpu_refcnt = alloc_percpu(int);
6009 if (!dev->pcpu_refcnt)
6012 if (dev_addr_init(dev))
6018 dev_net_set(dev, &init_net);
6020 dev->gso_max_size = GSO_MAX_SIZE;
6022 INIT_LIST_HEAD(&dev->napi_list);
6023 INIT_LIST_HEAD(&dev->unreg_list);
6024 INIT_LIST_HEAD(&dev->link_watch_list);
6025 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6028 dev->num_tx_queues = txqs;
6029 dev->real_num_tx_queues = txqs;
6030 if (netif_alloc_netdev_queues(dev))
6034 dev->num_rx_queues = rxqs;
6035 dev->real_num_rx_queues = rxqs;
6036 if (netif_alloc_rx_queues(dev))
6040 strcpy(dev->name, name);
6041 dev->group = INIT_NETDEV_GROUP;
6049 free_percpu(dev->pcpu_refcnt);
6059 EXPORT_SYMBOL(alloc_netdev_mqs);
6062 * free_netdev - free network device
6065 * This function does the last stage of destroying an allocated device
6066 * interface. The reference to the device object is released.
6067 * If this is the last reference then it will be freed.
6069 void free_netdev(struct net_device *dev)
6071 struct napi_struct *p, *n;
6073 release_net(dev_net(dev));
6080 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6082 /* Flush device addresses */
6083 dev_addr_flush(dev);
6085 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6088 free_percpu(dev->pcpu_refcnt);
6089 dev->pcpu_refcnt = NULL;
6091 /* Compatibility with error handling in drivers */
6092 if (dev->reg_state == NETREG_UNINITIALIZED) {
6093 kfree((char *)dev - dev->padded);
6097 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6098 dev->reg_state = NETREG_RELEASED;
6100 /* will free via device release */
6101 put_device(&dev->dev);
6103 EXPORT_SYMBOL(free_netdev);
6106 * synchronize_net - Synchronize with packet receive processing
6108 * Wait for packets currently being received to be done.
6109 * Does not block later packets from starting.
6111 void synchronize_net(void)
6114 if (rtnl_is_locked())
6115 synchronize_rcu_expedited();
6119 EXPORT_SYMBOL(synchronize_net);
6122 * unregister_netdevice_queue - remove device from the kernel
6126 * This function shuts down a device interface and removes it
6127 * from the kernel tables.
6128 * If head not NULL, device is queued to be unregistered later.
6130 * Callers must hold the rtnl semaphore. You may want
6131 * unregister_netdev() instead of this.
6134 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6139 list_move_tail(&dev->unreg_list, head);
6141 rollback_registered(dev);
6142 /* Finish processing unregister after unlock */
6146 EXPORT_SYMBOL(unregister_netdevice_queue);
6149 * unregister_netdevice_many - unregister many devices
6150 * @head: list of devices
6152 void unregister_netdevice_many(struct list_head *head)
6154 struct net_device *dev;
6156 if (!list_empty(head)) {
6157 rollback_registered_many(head);
6158 list_for_each_entry(dev, head, unreg_list)
6162 EXPORT_SYMBOL(unregister_netdevice_many);
6165 * unregister_netdev - remove device from the kernel
6168 * This function shuts down a device interface and removes it
6169 * from the kernel tables.
6171 * This is just a wrapper for unregister_netdevice that takes
6172 * the rtnl semaphore. In general you want to use this and not
6173 * unregister_netdevice.
6175 void unregister_netdev(struct net_device *dev)
6178 unregister_netdevice(dev);
6181 EXPORT_SYMBOL(unregister_netdev);
6184 * dev_change_net_namespace - move device to different nethost namespace
6186 * @net: network namespace
6187 * @pat: If not NULL name pattern to try if the current device name
6188 * is already taken in the destination network namespace.
6190 * This function shuts down a device interface and moves it
6191 * to a new network namespace. On success 0 is returned, on
6192 * a failure a netagive errno code is returned.
6194 * Callers must hold the rtnl semaphore.
6197 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6203 /* Don't allow namespace local devices to be moved. */
6205 if (dev->features & NETIF_F_NETNS_LOCAL)
6208 /* Ensure the device has been registrered */
6210 if (dev->reg_state != NETREG_REGISTERED)
6213 /* Get out if there is nothing todo */
6215 if (net_eq(dev_net(dev), net))
6218 /* Pick the destination device name, and ensure
6219 * we can use it in the destination network namespace.
6222 if (__dev_get_by_name(net, dev->name)) {
6223 /* We get here if we can't use the current device name */
6226 if (dev_get_valid_name(dev, pat) < 0)
6231 * And now a mini version of register_netdevice unregister_netdevice.
6234 /* If device is running close it first. */
6237 /* And unlink it from device chain */
6239 unlist_netdevice(dev);
6243 /* Shutdown queueing discipline. */
6246 /* Notify protocols, that we are about to destroy
6247 this device. They should clean all the things.
6249 Note that dev->reg_state stays at NETREG_REGISTERED.
6250 This is wanted because this way 8021q and macvlan know
6251 the device is just moving and can keep their slaves up.
6253 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6254 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6255 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6258 * Flush the unicast and multicast chains
6263 /* Actually switch the network namespace */
6264 dev_net_set(dev, net);
6266 /* If there is an ifindex conflict assign a new one */
6267 if (__dev_get_by_index(net, dev->ifindex)) {
6268 int iflink = (dev->iflink == dev->ifindex);
6269 dev->ifindex = dev_new_index(net);
6271 dev->iflink = dev->ifindex;
6274 /* Fixup kobjects */
6275 err = device_rename(&dev->dev, dev->name);
6278 /* Add the device back in the hashes */
6279 list_netdevice(dev);
6281 /* Notify protocols, that a new device appeared. */
6282 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6285 * Prevent userspace races by waiting until the network
6286 * device is fully setup before sending notifications.
6288 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6295 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6297 static int dev_cpu_callback(struct notifier_block *nfb,
6298 unsigned long action,
6301 struct sk_buff **list_skb;
6302 struct sk_buff *skb;
6303 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6304 struct softnet_data *sd, *oldsd;
6306 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6309 local_irq_disable();
6310 cpu = smp_processor_id();
6311 sd = &per_cpu(softnet_data, cpu);
6312 oldsd = &per_cpu(softnet_data, oldcpu);
6314 /* Find end of our completion_queue. */
6315 list_skb = &sd->completion_queue;
6317 list_skb = &(*list_skb)->next;
6318 /* Append completion queue from offline CPU. */
6319 *list_skb = oldsd->completion_queue;
6320 oldsd->completion_queue = NULL;
6322 /* Append output queue from offline CPU. */
6323 if (oldsd->output_queue) {
6324 *sd->output_queue_tailp = oldsd->output_queue;
6325 sd->output_queue_tailp = oldsd->output_queue_tailp;
6326 oldsd->output_queue = NULL;
6327 oldsd->output_queue_tailp = &oldsd->output_queue;
6329 /* Append NAPI poll list from offline CPU. */
6330 if (!list_empty(&oldsd->poll_list)) {
6331 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6332 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6335 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6338 /* Process offline CPU's input_pkt_queue */
6339 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6341 input_queue_head_incr(oldsd);
6343 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6345 input_queue_head_incr(oldsd);
6353 * netdev_increment_features - increment feature set by one
6354 * @all: current feature set
6355 * @one: new feature set
6356 * @mask: mask feature set
6358 * Computes a new feature set after adding a device with feature set
6359 * @one to the master device with current feature set @all. Will not
6360 * enable anything that is off in @mask. Returns the new feature set.
6362 netdev_features_t netdev_increment_features(netdev_features_t all,
6363 netdev_features_t one, netdev_features_t mask)
6365 if (mask & NETIF_F_GEN_CSUM)
6366 mask |= NETIF_F_ALL_CSUM;
6367 mask |= NETIF_F_VLAN_CHALLENGED;
6369 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6370 all &= one | ~NETIF_F_ALL_FOR_ALL;
6372 /* If one device supports hw checksumming, set for all. */
6373 if (all & NETIF_F_GEN_CSUM)
6374 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6378 EXPORT_SYMBOL(netdev_increment_features);
6380 static struct hlist_head *netdev_create_hash(void)
6383 struct hlist_head *hash;
6385 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6387 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6388 INIT_HLIST_HEAD(&hash[i]);
6393 /* Initialize per network namespace state */
6394 static int __net_init netdev_init(struct net *net)
6396 INIT_LIST_HEAD(&net->dev_base_head);
6398 net->dev_name_head = netdev_create_hash();
6399 if (net->dev_name_head == NULL)
6402 net->dev_index_head = netdev_create_hash();
6403 if (net->dev_index_head == NULL)
6409 kfree(net->dev_name_head);
6415 * netdev_drivername - network driver for the device
6416 * @dev: network device
6418 * Determine network driver for device.
6420 const char *netdev_drivername(const struct net_device *dev)
6422 const struct device_driver *driver;
6423 const struct device *parent;
6424 const char *empty = "";
6426 parent = dev->dev.parent;
6430 driver = parent->driver;
6431 if (driver && driver->name)
6432 return driver->name;
6436 int __netdev_printk(const char *level, const struct net_device *dev,
6437 struct va_format *vaf)
6441 if (dev && dev->dev.parent)
6442 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6443 netdev_name(dev), vaf);
6445 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6447 r = printk("%s(NULL net_device): %pV", level, vaf);
6451 EXPORT_SYMBOL(__netdev_printk);
6453 int netdev_printk(const char *level, const struct net_device *dev,
6454 const char *format, ...)
6456 struct va_format vaf;
6460 va_start(args, format);
6465 r = __netdev_printk(level, dev, &vaf);
6470 EXPORT_SYMBOL(netdev_printk);
6472 #define define_netdev_printk_level(func, level) \
6473 int func(const struct net_device *dev, const char *fmt, ...) \
6476 struct va_format vaf; \
6479 va_start(args, fmt); \
6484 r = __netdev_printk(level, dev, &vaf); \
6489 EXPORT_SYMBOL(func);
6491 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6492 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6493 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6494 define_netdev_printk_level(netdev_err, KERN_ERR);
6495 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6496 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6497 define_netdev_printk_level(netdev_info, KERN_INFO);
6499 static void __net_exit netdev_exit(struct net *net)
6501 kfree(net->dev_name_head);
6502 kfree(net->dev_index_head);
6505 static struct pernet_operations __net_initdata netdev_net_ops = {
6506 .init = netdev_init,
6507 .exit = netdev_exit,
6510 static void __net_exit default_device_exit(struct net *net)
6512 struct net_device *dev, *aux;
6514 * Push all migratable network devices back to the
6515 * initial network namespace
6518 for_each_netdev_safe(net, dev, aux) {
6520 char fb_name[IFNAMSIZ];
6522 /* Ignore unmoveable devices (i.e. loopback) */
6523 if (dev->features & NETIF_F_NETNS_LOCAL)
6526 /* Leave virtual devices for the generic cleanup */
6527 if (dev->rtnl_link_ops)
6530 /* Push remaining network devices to init_net */
6531 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6532 err = dev_change_net_namespace(dev, &init_net, fb_name);
6534 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6535 __func__, dev->name, err);
6542 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6544 /* At exit all network devices most be removed from a network
6545 * namespace. Do this in the reverse order of registration.
6546 * Do this across as many network namespaces as possible to
6547 * improve batching efficiency.
6549 struct net_device *dev;
6551 LIST_HEAD(dev_kill_list);
6554 list_for_each_entry(net, net_list, exit_list) {
6555 for_each_netdev_reverse(net, dev) {
6556 if (dev->rtnl_link_ops)
6557 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6559 unregister_netdevice_queue(dev, &dev_kill_list);
6562 unregister_netdevice_many(&dev_kill_list);
6563 list_del(&dev_kill_list);
6567 static struct pernet_operations __net_initdata default_device_ops = {
6568 .exit = default_device_exit,
6569 .exit_batch = default_device_exit_batch,
6573 * Initialize the DEV module. At boot time this walks the device list and
6574 * unhooks any devices that fail to initialise (normally hardware not
6575 * present) and leaves us with a valid list of present and active devices.
6580 * This is called single threaded during boot, so no need
6581 * to take the rtnl semaphore.
6583 static int __init net_dev_init(void)
6585 int i, rc = -ENOMEM;
6587 BUG_ON(!dev_boot_phase);
6589 if (dev_proc_init())
6592 if (netdev_kobject_init())
6595 INIT_LIST_HEAD(&ptype_all);
6596 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6597 INIT_LIST_HEAD(&ptype_base[i]);
6599 if (register_pernet_subsys(&netdev_net_ops))
6603 * Initialise the packet receive queues.
6606 for_each_possible_cpu(i) {
6607 struct softnet_data *sd = &per_cpu(softnet_data, i);
6609 memset(sd, 0, sizeof(*sd));
6610 skb_queue_head_init(&sd->input_pkt_queue);
6611 skb_queue_head_init(&sd->process_queue);
6612 sd->completion_queue = NULL;
6613 INIT_LIST_HEAD(&sd->poll_list);
6614 sd->output_queue = NULL;
6615 sd->output_queue_tailp = &sd->output_queue;
6617 sd->csd.func = rps_trigger_softirq;
6623 sd->backlog.poll = process_backlog;
6624 sd->backlog.weight = weight_p;
6625 sd->backlog.gro_list = NULL;
6626 sd->backlog.gro_count = 0;
6631 /* The loopback device is special if any other network devices
6632 * is present in a network namespace the loopback device must
6633 * be present. Since we now dynamically allocate and free the
6634 * loopback device ensure this invariant is maintained by
6635 * keeping the loopback device as the first device on the
6636 * list of network devices. Ensuring the loopback devices
6637 * is the first device that appears and the last network device
6640 if (register_pernet_device(&loopback_net_ops))
6643 if (register_pernet_device(&default_device_ops))
6646 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6647 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6649 hotcpu_notifier(dev_cpu_callback, 0);
6657 subsys_initcall(net_dev_init);
6659 static int __init initialize_hashrnd(void)
6661 get_random_bytes(&hashrnd, sizeof(hashrnd));
6665 late_initcall_sync(initialize_hashrnd);