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>
141 #include "net-sysfs.h"
143 /* Instead of increasing this, you should create a hash table. */
144 #define MAX_GRO_SKBS 8
146 /* This should be increased if a protocol with a bigger head is added. */
147 #define GRO_MAX_HEAD (MAX_HEADER + 128)
150 * The list of packet types we will receive (as opposed to discard)
151 * and the routines to invoke.
153 * Why 16. Because with 16 the only overlap we get on a hash of the
154 * low nibble of the protocol value is RARP/SNAP/X.25.
156 * NOTE: That is no longer true with the addition of VLAN tags. Not
157 * sure which should go first, but I bet it won't make much
158 * difference if we are running VLANs. The good news is that
159 * this protocol won't be in the list unless compiled in, so
160 * the average user (w/out VLANs) will not be adversely affected.
177 #define PTYPE_HASH_SIZE (16)
178 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
180 static DEFINE_SPINLOCK(ptype_lock);
181 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
182 static struct list_head ptype_all __read_mostly; /* Taps */
185 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
188 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
190 * Writers must hold the rtnl semaphore while they loop through the
191 * dev_base_head list, and hold dev_base_lock for writing when they do the
192 * actual updates. This allows pure readers to access the list even
193 * while a writer is preparing to update it.
195 * To put it another way, dev_base_lock is held for writing only to
196 * protect against pure readers; the rtnl semaphore provides the
197 * protection against other writers.
199 * See, for example usages, register_netdevice() and
200 * unregister_netdevice(), which must be called with the rtnl
203 DEFINE_RWLOCK(dev_base_lock);
204 EXPORT_SYMBOL(dev_base_lock);
206 static inline void dev_base_seq_inc(struct net *net)
208 while (++net->dev_base_seq == 0);
211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
213 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
214 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
217 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
219 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
222 static inline void rps_lock(struct softnet_data *sd)
225 spin_lock(&sd->input_pkt_queue.lock);
229 static inline void rps_unlock(struct softnet_data *sd)
232 spin_unlock(&sd->input_pkt_queue.lock);
236 /* Device list insertion */
237 static int list_netdevice(struct net_device *dev)
239 struct net *net = dev_net(dev);
243 write_lock_bh(&dev_base_lock);
244 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
245 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
246 hlist_add_head_rcu(&dev->index_hlist,
247 dev_index_hash(net, dev->ifindex));
248 write_unlock_bh(&dev_base_lock);
250 dev_base_seq_inc(net);
255 /* Device list removal
256 * caller must respect a RCU grace period before freeing/reusing dev
258 static void unlist_netdevice(struct net_device *dev)
262 /* Unlink dev from the device chain */
263 write_lock_bh(&dev_base_lock);
264 list_del_rcu(&dev->dev_list);
265 hlist_del_rcu(&dev->name_hlist);
266 hlist_del_rcu(&dev->index_hlist);
267 write_unlock_bh(&dev_base_lock);
269 dev_base_seq_inc(dev_net(dev));
276 static RAW_NOTIFIER_HEAD(netdev_chain);
279 * Device drivers call our routines to queue packets here. We empty the
280 * queue in the local softnet handler.
283 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
284 EXPORT_PER_CPU_SYMBOL(softnet_data);
286 #ifdef CONFIG_LOCKDEP
288 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
289 * according to dev->type
291 static const unsigned short netdev_lock_type[] =
292 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
293 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
294 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
295 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
296 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
297 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
298 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
299 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
300 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
301 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
302 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
303 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
304 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
305 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
306 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
307 ARPHRD_VOID, ARPHRD_NONE};
309 static const char *const netdev_lock_name[] =
310 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
311 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
312 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
313 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
314 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
315 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
316 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
317 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
318 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
319 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
320 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
321 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
322 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
323 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
324 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
325 "_xmit_VOID", "_xmit_NONE"};
327 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
328 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
330 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
334 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
335 if (netdev_lock_type[i] == dev_type)
337 /* the last key is used by default */
338 return ARRAY_SIZE(netdev_lock_type) - 1;
341 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
342 unsigned short dev_type)
346 i = netdev_lock_pos(dev_type);
347 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
348 netdev_lock_name[i]);
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
355 i = netdev_lock_pos(dev->type);
356 lockdep_set_class_and_name(&dev->addr_list_lock,
357 &netdev_addr_lock_key[i],
358 netdev_lock_name[i]);
361 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
362 unsigned short dev_type)
365 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
370 /*******************************************************************************
372 Protocol management and registration routines
374 *******************************************************************************/
377 * Add a protocol ID to the list. Now that the input handler is
378 * smarter we can dispense with all the messy stuff that used to be
381 * BEWARE!!! Protocol handlers, mangling input packets,
382 * MUST BE last in hash buckets and checking protocol handlers
383 * MUST start from promiscuous ptype_all chain in net_bh.
384 * It is true now, do not change it.
385 * Explanation follows: if protocol handler, mangling packet, will
386 * be the first on list, it is not able to sense, that packet
387 * is cloned and should be copied-on-write, so that it will
388 * change it and subsequent readers will get broken packet.
392 static inline struct list_head *ptype_head(const struct packet_type *pt)
394 if (pt->type == htons(ETH_P_ALL))
397 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
401 * dev_add_pack - add packet handler
402 * @pt: packet type declaration
404 * Add a protocol handler to the networking stack. The passed &packet_type
405 * is linked into kernel lists and may not be freed until it has been
406 * removed from the kernel lists.
408 * This call does not sleep therefore it can not
409 * guarantee all CPU's that are in middle of receiving packets
410 * will see the new packet type (until the next received packet).
413 void dev_add_pack(struct packet_type *pt)
415 struct list_head *head = ptype_head(pt);
417 spin_lock(&ptype_lock);
418 list_add_rcu(&pt->list, head);
419 spin_unlock(&ptype_lock);
421 EXPORT_SYMBOL(dev_add_pack);
424 * __dev_remove_pack - remove packet handler
425 * @pt: packet type declaration
427 * Remove a protocol handler that was previously added to the kernel
428 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
429 * from the kernel lists and can be freed or reused once this function
432 * The packet type might still be in use by receivers
433 * and must not be freed until after all the CPU's have gone
434 * through a quiescent state.
436 void __dev_remove_pack(struct packet_type *pt)
438 struct list_head *head = ptype_head(pt);
439 struct packet_type *pt1;
441 spin_lock(&ptype_lock);
443 list_for_each_entry(pt1, head, list) {
445 list_del_rcu(&pt->list);
450 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
452 spin_unlock(&ptype_lock);
454 EXPORT_SYMBOL(__dev_remove_pack);
457 * dev_remove_pack - remove packet handler
458 * @pt: packet type declaration
460 * Remove a protocol handler that was previously added to the kernel
461 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
462 * from the kernel lists and can be freed or reused once this function
465 * This call sleeps to guarantee that no CPU is looking at the packet
468 void dev_remove_pack(struct packet_type *pt)
470 __dev_remove_pack(pt);
474 EXPORT_SYMBOL(dev_remove_pack);
476 /******************************************************************************
478 Device Boot-time Settings Routines
480 *******************************************************************************/
482 /* Boot time configuration table */
483 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
486 * netdev_boot_setup_add - add new setup entry
487 * @name: name of the device
488 * @map: configured settings for the device
490 * Adds new setup entry to the dev_boot_setup list. The function
491 * returns 0 on error and 1 on success. This is a generic routine to
494 static int netdev_boot_setup_add(char *name, struct ifmap *map)
496 struct netdev_boot_setup *s;
500 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
501 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
502 memset(s[i].name, 0, sizeof(s[i].name));
503 strlcpy(s[i].name, name, IFNAMSIZ);
504 memcpy(&s[i].map, map, sizeof(s[i].map));
509 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
513 * netdev_boot_setup_check - check boot time settings
514 * @dev: the netdevice
516 * Check boot time settings for the device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found, 1 if they are.
521 int netdev_boot_setup_check(struct net_device *dev)
523 struct netdev_boot_setup *s = dev_boot_setup;
526 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
527 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
528 !strcmp(dev->name, s[i].name)) {
529 dev->irq = s[i].map.irq;
530 dev->base_addr = s[i].map.base_addr;
531 dev->mem_start = s[i].map.mem_start;
532 dev->mem_end = s[i].map.mem_end;
538 EXPORT_SYMBOL(netdev_boot_setup_check);
542 * netdev_boot_base - get address from boot time settings
543 * @prefix: prefix for network device
544 * @unit: id for network device
546 * Check boot time settings for the base address of device.
547 * The found settings are set for the device to be used
548 * later in the device probing.
549 * Returns 0 if no settings found.
551 unsigned long netdev_boot_base(const char *prefix, int unit)
553 const struct netdev_boot_setup *s = dev_boot_setup;
557 sprintf(name, "%s%d", prefix, unit);
560 * If device already registered then return base of 1
561 * to indicate not to probe for this interface
563 if (__dev_get_by_name(&init_net, name))
566 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
567 if (!strcmp(name, s[i].name))
568 return s[i].map.base_addr;
573 * Saves at boot time configured settings for any netdevice.
575 int __init netdev_boot_setup(char *str)
580 str = get_options(str, ARRAY_SIZE(ints), ints);
585 memset(&map, 0, sizeof(map));
589 map.base_addr = ints[2];
591 map.mem_start = ints[3];
593 map.mem_end = ints[4];
595 /* Add new entry to the list */
596 return netdev_boot_setup_add(str, &map);
599 __setup("netdev=", netdev_boot_setup);
601 /*******************************************************************************
603 Device Interface Subroutines
605 *******************************************************************************/
608 * __dev_get_by_name - find a device by its name
609 * @net: the applicable net namespace
610 * @name: name to find
612 * Find an interface by name. Must be called under RTNL semaphore
613 * or @dev_base_lock. If the name is found a pointer to the device
614 * is returned. If the name is not found then %NULL is returned. The
615 * reference counters are not incremented so the caller must be
616 * careful with locks.
619 struct net_device *__dev_get_by_name(struct net *net, const char *name)
621 struct hlist_node *p;
622 struct net_device *dev;
623 struct hlist_head *head = dev_name_hash(net, name);
625 hlist_for_each_entry(dev, p, head, name_hlist)
626 if (!strncmp(dev->name, name, IFNAMSIZ))
631 EXPORT_SYMBOL(__dev_get_by_name);
634 * dev_get_by_name_rcu - find a device by its name
635 * @net: the applicable net namespace
636 * @name: name to find
638 * Find an interface by name.
639 * If the name is found a pointer to the device is returned.
640 * If the name is not found then %NULL is returned.
641 * The reference counters are not incremented so the caller must be
642 * careful with locks. The caller must hold RCU lock.
645 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
647 struct hlist_node *p;
648 struct net_device *dev;
649 struct hlist_head *head = dev_name_hash(net, name);
651 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
652 if (!strncmp(dev->name, name, IFNAMSIZ))
657 EXPORT_SYMBOL(dev_get_by_name_rcu);
660 * dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. This can be called from any
665 * context and does its own locking. The returned handle has
666 * the usage count incremented and the caller must use dev_put() to
667 * release it when it is no longer needed. %NULL is returned if no
668 * matching device is found.
671 struct net_device *dev_get_by_name(struct net *net, const char *name)
673 struct net_device *dev;
676 dev = dev_get_by_name_rcu(net, name);
682 EXPORT_SYMBOL(dev_get_by_name);
685 * __dev_get_by_index - find a device by its ifindex
686 * @net: the applicable net namespace
687 * @ifindex: index of device
689 * Search for an interface by index. Returns %NULL if the device
690 * is not found or a pointer to the device. The device has not
691 * had its reference counter increased so the caller must be careful
692 * about locking. The caller must hold either the RTNL semaphore
696 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
698 struct hlist_node *p;
699 struct net_device *dev;
700 struct hlist_head *head = dev_index_hash(net, ifindex);
702 hlist_for_each_entry(dev, p, head, index_hlist)
703 if (dev->ifindex == ifindex)
708 EXPORT_SYMBOL(__dev_get_by_index);
711 * dev_get_by_index_rcu - find a device by its ifindex
712 * @net: the applicable net namespace
713 * @ifindex: index of device
715 * Search for an interface by index. Returns %NULL if the device
716 * is not found or a pointer to the device. The device has not
717 * had its reference counter increased so the caller must be careful
718 * about locking. The caller must hold RCU lock.
721 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
723 struct hlist_node *p;
724 struct net_device *dev;
725 struct hlist_head *head = dev_index_hash(net, ifindex);
727 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
728 if (dev->ifindex == ifindex)
733 EXPORT_SYMBOL(dev_get_by_index_rcu);
737 * dev_get_by_index - find a device by its ifindex
738 * @net: the applicable net namespace
739 * @ifindex: index of device
741 * Search for an interface by index. Returns NULL if the device
742 * is not found or a pointer to the device. The device returned has
743 * had a reference added and the pointer is safe until the user calls
744 * dev_put to indicate they have finished with it.
747 struct net_device *dev_get_by_index(struct net *net, int ifindex)
749 struct net_device *dev;
752 dev = dev_get_by_index_rcu(net, ifindex);
758 EXPORT_SYMBOL(dev_get_by_index);
761 * dev_getbyhwaddr_rcu - find a device by its hardware address
762 * @net: the applicable net namespace
763 * @type: media type of device
764 * @ha: hardware address
766 * Search for an interface by MAC address. Returns NULL if the device
767 * is not found or a pointer to the device.
768 * The caller must hold RCU or RTNL.
769 * The returned device has not had its ref count increased
770 * and the caller must therefore be careful about locking
774 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
777 struct net_device *dev;
779 for_each_netdev_rcu(net, dev)
780 if (dev->type == type &&
781 !memcmp(dev->dev_addr, ha, dev->addr_len))
786 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
788 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
790 struct net_device *dev;
793 for_each_netdev(net, dev)
794 if (dev->type == type)
799 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
801 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
803 struct net_device *dev, *ret = NULL;
806 for_each_netdev_rcu(net, dev)
807 if (dev->type == type) {
815 EXPORT_SYMBOL(dev_getfirstbyhwtype);
818 * dev_get_by_flags_rcu - find any device with given flags
819 * @net: the applicable net namespace
820 * @if_flags: IFF_* values
821 * @mask: bitmask of bits in if_flags to check
823 * Search for any interface with the given flags. Returns NULL if a device
824 * is not found or a pointer to the device. Must be called inside
825 * rcu_read_lock(), and result refcount is unchanged.
828 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
831 struct net_device *dev, *ret;
834 for_each_netdev_rcu(net, dev) {
835 if (((dev->flags ^ if_flags) & mask) == 0) {
842 EXPORT_SYMBOL(dev_get_by_flags_rcu);
845 * dev_valid_name - check if name is okay for network device
848 * Network device names need to be valid file names to
849 * to allow sysfs to work. We also disallow any kind of
852 int dev_valid_name(const char *name)
856 if (strlen(name) >= IFNAMSIZ)
858 if (!strcmp(name, ".") || !strcmp(name, ".."))
862 if (*name == '/' || isspace(*name))
868 EXPORT_SYMBOL(dev_valid_name);
871 * __dev_alloc_name - allocate a name for a device
872 * @net: network namespace to allocate the device name in
873 * @name: name format string
874 * @buf: scratch buffer and result name string
876 * Passed a format string - eg "lt%d" it will try and find a suitable
877 * id. It scans list of devices to build up a free map, then chooses
878 * the first empty slot. The caller must hold the dev_base or rtnl lock
879 * while allocating the name and adding the device in order to avoid
881 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
882 * Returns the number of the unit assigned or a negative errno code.
885 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
889 const int max_netdevices = 8*PAGE_SIZE;
890 unsigned long *inuse;
891 struct net_device *d;
893 p = strnchr(name, IFNAMSIZ-1, '%');
896 * Verify the string as this thing may have come from
897 * the user. There must be either one "%d" and no other "%"
900 if (p[1] != 'd' || strchr(p + 2, '%'))
903 /* Use one page as a bit array of possible slots */
904 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
908 for_each_netdev(net, d) {
909 if (!sscanf(d->name, name, &i))
911 if (i < 0 || i >= max_netdevices)
914 /* avoid cases where sscanf is not exact inverse of printf */
915 snprintf(buf, IFNAMSIZ, name, i);
916 if (!strncmp(buf, d->name, IFNAMSIZ))
920 i = find_first_zero_bit(inuse, max_netdevices);
921 free_page((unsigned long) inuse);
925 snprintf(buf, IFNAMSIZ, name, i);
926 if (!__dev_get_by_name(net, buf))
929 /* It is possible to run out of possible slots
930 * when the name is long and there isn't enough space left
931 * for the digits, or if all bits are used.
937 * dev_alloc_name - allocate a name for a device
939 * @name: name format string
941 * Passed a format string - eg "lt%d" it will try and find a suitable
942 * id. It scans list of devices to build up a free map, then chooses
943 * the first empty slot. The caller must hold the dev_base or rtnl lock
944 * while allocating the name and adding the device in order to avoid
946 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
947 * Returns the number of the unit assigned or a negative errno code.
950 int dev_alloc_name(struct net_device *dev, const char *name)
956 BUG_ON(!dev_net(dev));
958 ret = __dev_alloc_name(net, name, buf);
960 strlcpy(dev->name, buf, IFNAMSIZ);
963 EXPORT_SYMBOL(dev_alloc_name);
965 static int dev_get_valid_name(struct net_device *dev, const char *name)
969 BUG_ON(!dev_net(dev));
972 if (!dev_valid_name(name))
975 if (strchr(name, '%'))
976 return dev_alloc_name(dev, name);
977 else if (__dev_get_by_name(net, name))
979 else if (dev->name != name)
980 strlcpy(dev->name, name, IFNAMSIZ);
986 * dev_change_name - change name of a device
988 * @newname: name (or format string) must be at least IFNAMSIZ
990 * Change name of a device, can pass format strings "eth%d".
993 int dev_change_name(struct net_device *dev, const char *newname)
995 char oldname[IFNAMSIZ];
1001 BUG_ON(!dev_net(dev));
1004 if (dev->flags & IFF_UP)
1007 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1010 memcpy(oldname, dev->name, IFNAMSIZ);
1012 err = dev_get_valid_name(dev, newname);
1017 ret = device_rename(&dev->dev, dev->name);
1019 memcpy(dev->name, oldname, IFNAMSIZ);
1023 write_lock_bh(&dev_base_lock);
1024 hlist_del_rcu(&dev->name_hlist);
1025 write_unlock_bh(&dev_base_lock);
1029 write_lock_bh(&dev_base_lock);
1030 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1031 write_unlock_bh(&dev_base_lock);
1033 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1034 ret = notifier_to_errno(ret);
1037 /* err >= 0 after dev_alloc_name() or stores the first errno */
1040 memcpy(dev->name, oldname, IFNAMSIZ);
1044 "%s: name change rollback failed: %d.\n",
1053 * dev_set_alias - change ifalias of a device
1055 * @alias: name up to IFALIASZ
1056 * @len: limit of bytes to copy from info
1058 * Set ifalias for a device,
1060 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1064 if (len >= IFALIASZ)
1069 kfree(dev->ifalias);
1070 dev->ifalias = NULL;
1075 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1079 strlcpy(dev->ifalias, alias, len+1);
1085 * netdev_features_change - device changes features
1086 * @dev: device to cause notification
1088 * Called to indicate a device has changed features.
1090 void netdev_features_change(struct net_device *dev)
1092 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1094 EXPORT_SYMBOL(netdev_features_change);
1097 * netdev_state_change - device changes state
1098 * @dev: device to cause notification
1100 * Called to indicate a device has changed state. This function calls
1101 * the notifier chains for netdev_chain and sends a NEWLINK message
1102 * to the routing socket.
1104 void netdev_state_change(struct net_device *dev)
1106 if (dev->flags & IFF_UP) {
1107 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1108 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1111 EXPORT_SYMBOL(netdev_state_change);
1113 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1115 return call_netdevice_notifiers(event, dev);
1117 EXPORT_SYMBOL(netdev_bonding_change);
1120 * dev_load - load a network module
1121 * @net: the applicable net namespace
1122 * @name: name of interface
1124 * If a network interface is not present and the process has suitable
1125 * privileges this function loads the module. If module loading is not
1126 * available in this kernel then it becomes a nop.
1129 void dev_load(struct net *net, const char *name)
1131 struct net_device *dev;
1135 dev = dev_get_by_name_rcu(net, name);
1139 if (no_module && capable(CAP_NET_ADMIN))
1140 no_module = request_module("netdev-%s", name);
1141 if (no_module && capable(CAP_SYS_MODULE)) {
1142 if (!request_module("%s", name))
1143 pr_err("Loading kernel module for a network device "
1144 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1148 EXPORT_SYMBOL(dev_load);
1150 static int __dev_open(struct net_device *dev)
1152 const struct net_device_ops *ops = dev->netdev_ops;
1157 if (!netif_device_present(dev))
1160 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1161 ret = notifier_to_errno(ret);
1165 set_bit(__LINK_STATE_START, &dev->state);
1167 if (ops->ndo_validate_addr)
1168 ret = ops->ndo_validate_addr(dev);
1170 if (!ret && ops->ndo_open)
1171 ret = ops->ndo_open(dev);
1174 clear_bit(__LINK_STATE_START, &dev->state);
1176 dev->flags |= IFF_UP;
1177 net_dmaengine_get();
1178 dev_set_rx_mode(dev);
1186 * dev_open - prepare an interface for use.
1187 * @dev: device to open
1189 * Takes a device from down to up state. The device's private open
1190 * function is invoked and then the multicast lists are loaded. Finally
1191 * the device is moved into the up state and a %NETDEV_UP message is
1192 * sent to the netdev notifier chain.
1194 * Calling this function on an active interface is a nop. On a failure
1195 * a negative errno code is returned.
1197 int dev_open(struct net_device *dev)
1201 if (dev->flags & IFF_UP)
1204 ret = __dev_open(dev);
1208 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1209 call_netdevice_notifiers(NETDEV_UP, dev);
1213 EXPORT_SYMBOL(dev_open);
1215 static int __dev_close_many(struct list_head *head)
1217 struct net_device *dev;
1222 list_for_each_entry(dev, head, unreg_list) {
1223 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1225 clear_bit(__LINK_STATE_START, &dev->state);
1227 /* Synchronize to scheduled poll. We cannot touch poll list, it
1228 * can be even on different cpu. So just clear netif_running().
1230 * dev->stop() will invoke napi_disable() on all of it's
1231 * napi_struct instances on this device.
1233 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1236 dev_deactivate_many(head);
1238 list_for_each_entry(dev, head, unreg_list) {
1239 const struct net_device_ops *ops = dev->netdev_ops;
1242 * Call the device specific close. This cannot fail.
1243 * Only if device is UP
1245 * We allow it to be called even after a DETACH hot-plug
1251 dev->flags &= ~IFF_UP;
1252 net_dmaengine_put();
1258 static int __dev_close(struct net_device *dev)
1263 list_add(&dev->unreg_list, &single);
1264 retval = __dev_close_many(&single);
1269 static int dev_close_many(struct list_head *head)
1271 struct net_device *dev, *tmp;
1272 LIST_HEAD(tmp_list);
1274 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1275 if (!(dev->flags & IFF_UP))
1276 list_move(&dev->unreg_list, &tmp_list);
1278 __dev_close_many(head);
1280 list_for_each_entry(dev, head, unreg_list) {
1281 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1282 call_netdevice_notifiers(NETDEV_DOWN, dev);
1285 /* rollback_registered_many needs the complete original list */
1286 list_splice(&tmp_list, head);
1291 * dev_close - shutdown an interface.
1292 * @dev: device to shutdown
1294 * This function moves an active device into down state. A
1295 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1296 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1299 int dev_close(struct net_device *dev)
1301 if (dev->flags & IFF_UP) {
1304 list_add(&dev->unreg_list, &single);
1305 dev_close_many(&single);
1310 EXPORT_SYMBOL(dev_close);
1314 * dev_disable_lro - disable Large Receive Offload on a device
1317 * Disable Large Receive Offload (LRO) on a net device. Must be
1318 * called under RTNL. This is needed if received packets may be
1319 * forwarded to another interface.
1321 void dev_disable_lro(struct net_device *dev)
1326 * If we're trying to disable lro on a vlan device
1327 * use the underlying physical device instead
1329 if (is_vlan_dev(dev))
1330 dev = vlan_dev_real_dev(dev);
1332 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1333 flags = dev->ethtool_ops->get_flags(dev);
1335 flags = ethtool_op_get_flags(dev);
1337 if (!(flags & ETH_FLAG_LRO))
1340 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1341 if (unlikely(dev->features & NETIF_F_LRO))
1342 netdev_WARN(dev, "failed to disable LRO!\n");
1344 EXPORT_SYMBOL(dev_disable_lro);
1347 static int dev_boot_phase = 1;
1350 * register_netdevice_notifier - register a network notifier block
1353 * Register a notifier to be called when network device events occur.
1354 * The notifier passed is linked into the kernel structures and must
1355 * not be reused until it has been unregistered. A negative errno code
1356 * is returned on a failure.
1358 * When registered all registration and up events are replayed
1359 * to the new notifier to allow device to have a race free
1360 * view of the network device list.
1363 int register_netdevice_notifier(struct notifier_block *nb)
1365 struct net_device *dev;
1366 struct net_device *last;
1371 err = raw_notifier_chain_register(&netdev_chain, nb);
1377 for_each_netdev(net, dev) {
1378 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1379 err = notifier_to_errno(err);
1383 if (!(dev->flags & IFF_UP))
1386 nb->notifier_call(nb, NETDEV_UP, dev);
1397 for_each_netdev(net, dev) {
1401 if (dev->flags & IFF_UP) {
1402 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1403 nb->notifier_call(nb, NETDEV_DOWN, dev);
1405 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1406 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1410 raw_notifier_chain_unregister(&netdev_chain, nb);
1413 EXPORT_SYMBOL(register_netdevice_notifier);
1416 * unregister_netdevice_notifier - unregister a network notifier block
1419 * Unregister a notifier previously registered by
1420 * register_netdevice_notifier(). The notifier is unlinked into the
1421 * kernel structures and may then be reused. A negative errno code
1422 * is returned on a failure.
1425 int unregister_netdevice_notifier(struct notifier_block *nb)
1430 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1434 EXPORT_SYMBOL(unregister_netdevice_notifier);
1437 * call_netdevice_notifiers - call all network notifier blocks
1438 * @val: value passed unmodified to notifier function
1439 * @dev: net_device pointer passed unmodified to notifier function
1441 * Call all network notifier blocks. Parameters and return value
1442 * are as for raw_notifier_call_chain().
1445 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1448 return raw_notifier_call_chain(&netdev_chain, val, dev);
1450 EXPORT_SYMBOL(call_netdevice_notifiers);
1452 /* When > 0 there are consumers of rx skb time stamps */
1453 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1455 void net_enable_timestamp(void)
1457 atomic_inc(&netstamp_needed);
1459 EXPORT_SYMBOL(net_enable_timestamp);
1461 void net_disable_timestamp(void)
1463 atomic_dec(&netstamp_needed);
1465 EXPORT_SYMBOL(net_disable_timestamp);
1467 static inline void net_timestamp_set(struct sk_buff *skb)
1469 if (atomic_read(&netstamp_needed))
1470 __net_timestamp(skb);
1472 skb->tstamp.tv64 = 0;
1475 static inline void net_timestamp_check(struct sk_buff *skb)
1477 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1478 __net_timestamp(skb);
1481 static int net_hwtstamp_validate(struct ifreq *ifr)
1483 struct hwtstamp_config cfg;
1484 enum hwtstamp_tx_types tx_type;
1485 enum hwtstamp_rx_filters rx_filter;
1486 int tx_type_valid = 0;
1487 int rx_filter_valid = 0;
1489 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1492 if (cfg.flags) /* reserved for future extensions */
1495 tx_type = cfg.tx_type;
1496 rx_filter = cfg.rx_filter;
1499 case HWTSTAMP_TX_OFF:
1500 case HWTSTAMP_TX_ON:
1501 case HWTSTAMP_TX_ONESTEP_SYNC:
1506 switch (rx_filter) {
1507 case HWTSTAMP_FILTER_NONE:
1508 case HWTSTAMP_FILTER_ALL:
1509 case HWTSTAMP_FILTER_SOME:
1510 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1511 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1512 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1513 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1514 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1515 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1516 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1517 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1518 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1519 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1520 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1521 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1522 rx_filter_valid = 1;
1526 if (!tx_type_valid || !rx_filter_valid)
1532 static inline bool is_skb_forwardable(struct net_device *dev,
1533 struct sk_buff *skb)
1537 if (!(dev->flags & IFF_UP))
1540 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1541 if (skb->len <= len)
1544 /* if TSO is enabled, we don't care about the length as the packet
1545 * could be forwarded without being segmented before
1547 if (skb_is_gso(skb))
1554 * dev_forward_skb - loopback an skb to another netif
1556 * @dev: destination network device
1557 * @skb: buffer to forward
1560 * NET_RX_SUCCESS (no congestion)
1561 * NET_RX_DROP (packet was dropped, but freed)
1563 * dev_forward_skb can be used for injecting an skb from the
1564 * start_xmit function of one device into the receive queue
1565 * of another device.
1567 * The receiving device may be in another namespace, so
1568 * we have to clear all information in the skb that could
1569 * impact namespace isolation.
1571 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1573 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1574 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1575 atomic_long_inc(&dev->rx_dropped);
1584 if (unlikely(!is_skb_forwardable(dev, skb))) {
1585 atomic_long_inc(&dev->rx_dropped);
1589 skb_set_dev(skb, dev);
1590 skb->tstamp.tv64 = 0;
1591 skb->pkt_type = PACKET_HOST;
1592 skb->protocol = eth_type_trans(skb, dev);
1593 return netif_rx(skb);
1595 EXPORT_SYMBOL_GPL(dev_forward_skb);
1597 static inline int deliver_skb(struct sk_buff *skb,
1598 struct packet_type *pt_prev,
1599 struct net_device *orig_dev)
1601 atomic_inc(&skb->users);
1602 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1606 * Support routine. Sends outgoing frames to any network
1607 * taps currently in use.
1610 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1612 struct packet_type *ptype;
1613 struct sk_buff *skb2 = NULL;
1614 struct packet_type *pt_prev = NULL;
1617 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1618 /* Never send packets back to the socket
1619 * they originated from - MvS (miquels@drinkel.ow.org)
1621 if ((ptype->dev == dev || !ptype->dev) &&
1622 (ptype->af_packet_priv == NULL ||
1623 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1625 deliver_skb(skb2, pt_prev, skb->dev);
1630 skb2 = skb_clone(skb, GFP_ATOMIC);
1634 net_timestamp_set(skb2);
1636 /* skb->nh should be correctly
1637 set by sender, so that the second statement is
1638 just protection against buggy protocols.
1640 skb_reset_mac_header(skb2);
1642 if (skb_network_header(skb2) < skb2->data ||
1643 skb2->network_header > skb2->tail) {
1644 if (net_ratelimit())
1645 printk(KERN_CRIT "protocol %04x is "
1647 ntohs(skb2->protocol),
1649 skb_reset_network_header(skb2);
1652 skb2->transport_header = skb2->network_header;
1653 skb2->pkt_type = PACKET_OUTGOING;
1658 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1662 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1663 * @dev: Network device
1664 * @txq: number of queues available
1666 * If real_num_tx_queues is changed the tc mappings may no longer be
1667 * valid. To resolve this verify the tc mapping remains valid and if
1668 * not NULL the mapping. With no priorities mapping to this
1669 * offset/count pair it will no longer be used. In the worst case TC0
1670 * is invalid nothing can be done so disable priority mappings. If is
1671 * expected that drivers will fix this mapping if they can before
1672 * calling netif_set_real_num_tx_queues.
1674 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1677 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1679 /* If TC0 is invalidated disable TC mapping */
1680 if (tc->offset + tc->count > txq) {
1681 pr_warning("Number of in use tx queues changed "
1682 "invalidating tc mappings. Priority "
1683 "traffic classification disabled!\n");
1688 /* Invalidated prio to tc mappings set to TC0 */
1689 for (i = 1; i < TC_BITMASK + 1; i++) {
1690 int q = netdev_get_prio_tc_map(dev, i);
1692 tc = &dev->tc_to_txq[q];
1693 if (tc->offset + tc->count > txq) {
1694 pr_warning("Number of in use tx queues "
1695 "changed. Priority %i to tc "
1696 "mapping %i is no longer valid "
1697 "setting map to 0\n",
1699 netdev_set_prio_tc_map(dev, i, 0);
1705 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1706 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1708 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1712 if (txq < 1 || txq > dev->num_tx_queues)
1715 if (dev->reg_state == NETREG_REGISTERED ||
1716 dev->reg_state == NETREG_UNREGISTERING) {
1719 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1725 netif_setup_tc(dev, txq);
1727 if (txq < dev->real_num_tx_queues)
1728 qdisc_reset_all_tx_gt(dev, txq);
1731 dev->real_num_tx_queues = txq;
1734 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1738 * netif_set_real_num_rx_queues - set actual number of RX queues used
1739 * @dev: Network device
1740 * @rxq: Actual number of RX queues
1742 * This must be called either with the rtnl_lock held or before
1743 * registration of the net device. Returns 0 on success, or a
1744 * negative error code. If called before registration, it always
1747 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1751 if (rxq < 1 || rxq > dev->num_rx_queues)
1754 if (dev->reg_state == NETREG_REGISTERED) {
1757 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1763 dev->real_num_rx_queues = rxq;
1766 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1769 static inline void __netif_reschedule(struct Qdisc *q)
1771 struct softnet_data *sd;
1772 unsigned long flags;
1774 local_irq_save(flags);
1775 sd = &__get_cpu_var(softnet_data);
1776 q->next_sched = NULL;
1777 *sd->output_queue_tailp = q;
1778 sd->output_queue_tailp = &q->next_sched;
1779 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1780 local_irq_restore(flags);
1783 void __netif_schedule(struct Qdisc *q)
1785 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1786 __netif_reschedule(q);
1788 EXPORT_SYMBOL(__netif_schedule);
1790 void dev_kfree_skb_irq(struct sk_buff *skb)
1792 if (atomic_dec_and_test(&skb->users)) {
1793 struct softnet_data *sd;
1794 unsigned long flags;
1796 local_irq_save(flags);
1797 sd = &__get_cpu_var(softnet_data);
1798 skb->next = sd->completion_queue;
1799 sd->completion_queue = skb;
1800 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1801 local_irq_restore(flags);
1804 EXPORT_SYMBOL(dev_kfree_skb_irq);
1806 void dev_kfree_skb_any(struct sk_buff *skb)
1808 if (in_irq() || irqs_disabled())
1809 dev_kfree_skb_irq(skb);
1813 EXPORT_SYMBOL(dev_kfree_skb_any);
1817 * netif_device_detach - mark device as removed
1818 * @dev: network device
1820 * Mark device as removed from system and therefore no longer available.
1822 void netif_device_detach(struct net_device *dev)
1824 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1825 netif_running(dev)) {
1826 netif_tx_stop_all_queues(dev);
1829 EXPORT_SYMBOL(netif_device_detach);
1832 * netif_device_attach - mark device as attached
1833 * @dev: network device
1835 * Mark device as attached from system and restart if needed.
1837 void netif_device_attach(struct net_device *dev)
1839 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1840 netif_running(dev)) {
1841 netif_tx_wake_all_queues(dev);
1842 __netdev_watchdog_up(dev);
1845 EXPORT_SYMBOL(netif_device_attach);
1848 * skb_dev_set -- assign a new device to a buffer
1849 * @skb: buffer for the new device
1850 * @dev: network device
1852 * If an skb is owned by a device already, we have to reset
1853 * all data private to the namespace a device belongs to
1854 * before assigning it a new device.
1856 #ifdef CONFIG_NET_NS
1857 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1860 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1863 skb_init_secmark(skb);
1867 skb->ipvs_property = 0;
1868 #ifdef CONFIG_NET_SCHED
1874 EXPORT_SYMBOL(skb_set_dev);
1875 #endif /* CONFIG_NET_NS */
1878 * Invalidate hardware checksum when packet is to be mangled, and
1879 * complete checksum manually on outgoing path.
1881 int skb_checksum_help(struct sk_buff *skb)
1884 int ret = 0, offset;
1886 if (skb->ip_summed == CHECKSUM_COMPLETE)
1887 goto out_set_summed;
1889 if (unlikely(skb_shinfo(skb)->gso_size)) {
1890 /* Let GSO fix up the checksum. */
1891 goto out_set_summed;
1894 offset = skb_checksum_start_offset(skb);
1895 BUG_ON(offset >= skb_headlen(skb));
1896 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1898 offset += skb->csum_offset;
1899 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1901 if (skb_cloned(skb) &&
1902 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1903 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1908 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1910 skb->ip_summed = CHECKSUM_NONE;
1914 EXPORT_SYMBOL(skb_checksum_help);
1917 * skb_gso_segment - Perform segmentation on skb.
1918 * @skb: buffer to segment
1919 * @features: features for the output path (see dev->features)
1921 * This function segments the given skb and returns a list of segments.
1923 * It may return NULL if the skb requires no segmentation. This is
1924 * only possible when GSO is used for verifying header integrity.
1926 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1928 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1929 struct packet_type *ptype;
1930 __be16 type = skb->protocol;
1931 int vlan_depth = ETH_HLEN;
1934 while (type == htons(ETH_P_8021Q)) {
1935 struct vlan_hdr *vh;
1937 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1938 return ERR_PTR(-EINVAL);
1940 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1941 type = vh->h_vlan_encapsulated_proto;
1942 vlan_depth += VLAN_HLEN;
1945 skb_reset_mac_header(skb);
1946 skb->mac_len = skb->network_header - skb->mac_header;
1947 __skb_pull(skb, skb->mac_len);
1949 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1950 struct net_device *dev = skb->dev;
1951 struct ethtool_drvinfo info = {};
1953 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1954 dev->ethtool_ops->get_drvinfo(dev, &info);
1956 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1957 info.driver, dev ? dev->features : 0L,
1958 skb->sk ? skb->sk->sk_route_caps : 0L,
1959 skb->len, skb->data_len, skb->ip_summed);
1961 if (skb_header_cloned(skb) &&
1962 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1963 return ERR_PTR(err);
1967 list_for_each_entry_rcu(ptype,
1968 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1969 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1970 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1971 err = ptype->gso_send_check(skb);
1972 segs = ERR_PTR(err);
1973 if (err || skb_gso_ok(skb, features))
1975 __skb_push(skb, (skb->data -
1976 skb_network_header(skb)));
1978 segs = ptype->gso_segment(skb, features);
1984 __skb_push(skb, skb->data - skb_mac_header(skb));
1988 EXPORT_SYMBOL(skb_gso_segment);
1990 /* Take action when hardware reception checksum errors are detected. */
1992 void netdev_rx_csum_fault(struct net_device *dev)
1994 if (net_ratelimit()) {
1995 printk(KERN_ERR "%s: hw csum failure.\n",
1996 dev ? dev->name : "<unknown>");
2000 EXPORT_SYMBOL(netdev_rx_csum_fault);
2003 /* Actually, we should eliminate this check as soon as we know, that:
2004 * 1. IOMMU is present and allows to map all the memory.
2005 * 2. No high memory really exists on this machine.
2008 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2010 #ifdef CONFIG_HIGHMEM
2012 if (!(dev->features & NETIF_F_HIGHDMA)) {
2013 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2014 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2015 if (PageHighMem(skb_frag_page(frag)))
2020 if (PCI_DMA_BUS_IS_PHYS) {
2021 struct device *pdev = dev->dev.parent;
2025 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2026 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2027 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2028 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2037 void (*destructor)(struct sk_buff *skb);
2040 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2042 static void dev_gso_skb_destructor(struct sk_buff *skb)
2044 struct dev_gso_cb *cb;
2047 struct sk_buff *nskb = skb->next;
2049 skb->next = nskb->next;
2052 } while (skb->next);
2054 cb = DEV_GSO_CB(skb);
2056 cb->destructor(skb);
2060 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2061 * @skb: buffer to segment
2062 * @features: device features as applicable to this skb
2064 * This function segments the given skb and stores the list of segments
2067 static int dev_gso_segment(struct sk_buff *skb, int features)
2069 struct sk_buff *segs;
2071 segs = skb_gso_segment(skb, features);
2073 /* Verifying header integrity only. */
2078 return PTR_ERR(segs);
2081 DEV_GSO_CB(skb)->destructor = skb->destructor;
2082 skb->destructor = dev_gso_skb_destructor;
2088 * Try to orphan skb early, right before transmission by the device.
2089 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2090 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2092 static inline void skb_orphan_try(struct sk_buff *skb)
2094 struct sock *sk = skb->sk;
2096 if (sk && !skb_shinfo(skb)->tx_flags) {
2097 /* skb_tx_hash() wont be able to get sk.
2098 * We copy sk_hash into skb->rxhash
2101 skb->rxhash = sk->sk_hash;
2106 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2108 return ((features & NETIF_F_GEN_CSUM) ||
2109 ((features & NETIF_F_V4_CSUM) &&
2110 protocol == htons(ETH_P_IP)) ||
2111 ((features & NETIF_F_V6_CSUM) &&
2112 protocol == htons(ETH_P_IPV6)) ||
2113 ((features & NETIF_F_FCOE_CRC) &&
2114 protocol == htons(ETH_P_FCOE)));
2117 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2119 if (!can_checksum_protocol(features, protocol)) {
2120 features &= ~NETIF_F_ALL_CSUM;
2121 features &= ~NETIF_F_SG;
2122 } else if (illegal_highdma(skb->dev, skb)) {
2123 features &= ~NETIF_F_SG;
2129 u32 netif_skb_features(struct sk_buff *skb)
2131 __be16 protocol = skb->protocol;
2132 u32 features = skb->dev->features;
2134 if (protocol == htons(ETH_P_8021Q)) {
2135 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2136 protocol = veh->h_vlan_encapsulated_proto;
2137 } else if (!vlan_tx_tag_present(skb)) {
2138 return harmonize_features(skb, protocol, features);
2141 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2143 if (protocol != htons(ETH_P_8021Q)) {
2144 return harmonize_features(skb, protocol, features);
2146 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2147 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2148 return harmonize_features(skb, protocol, features);
2151 EXPORT_SYMBOL(netif_skb_features);
2154 * Returns true if either:
2155 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2156 * 2. skb is fragmented and the device does not support SG, or if
2157 * at least one of fragments is in highmem and device does not
2158 * support DMA from it.
2160 static inline int skb_needs_linearize(struct sk_buff *skb,
2163 return skb_is_nonlinear(skb) &&
2164 ((skb_has_frag_list(skb) &&
2165 !(features & NETIF_F_FRAGLIST)) ||
2166 (skb_shinfo(skb)->nr_frags &&
2167 !(features & NETIF_F_SG)));
2170 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2171 struct netdev_queue *txq)
2173 const struct net_device_ops *ops = dev->netdev_ops;
2174 int rc = NETDEV_TX_OK;
2175 unsigned int skb_len;
2177 if (likely(!skb->next)) {
2181 * If device doesn't need skb->dst, release it right now while
2182 * its hot in this cpu cache
2184 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2187 if (!list_empty(&ptype_all))
2188 dev_queue_xmit_nit(skb, dev);
2190 skb_orphan_try(skb);
2192 features = netif_skb_features(skb);
2194 if (vlan_tx_tag_present(skb) &&
2195 !(features & NETIF_F_HW_VLAN_TX)) {
2196 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2203 if (netif_needs_gso(skb, features)) {
2204 if (unlikely(dev_gso_segment(skb, features)))
2209 if (skb_needs_linearize(skb, features) &&
2210 __skb_linearize(skb))
2213 /* If packet is not checksummed and device does not
2214 * support checksumming for this protocol, complete
2215 * checksumming here.
2217 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2218 skb_set_transport_header(skb,
2219 skb_checksum_start_offset(skb));
2220 if (!(features & NETIF_F_ALL_CSUM) &&
2221 skb_checksum_help(skb))
2227 rc = ops->ndo_start_xmit(skb, dev);
2228 trace_net_dev_xmit(skb, rc, dev, skb_len);
2229 if (rc == NETDEV_TX_OK)
2230 txq_trans_update(txq);
2236 struct sk_buff *nskb = skb->next;
2238 skb->next = nskb->next;
2242 * If device doesn't need nskb->dst, release it right now while
2243 * its hot in this cpu cache
2245 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2248 skb_len = nskb->len;
2249 rc = ops->ndo_start_xmit(nskb, dev);
2250 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2251 if (unlikely(rc != NETDEV_TX_OK)) {
2252 if (rc & ~NETDEV_TX_MASK)
2253 goto out_kfree_gso_skb;
2254 nskb->next = skb->next;
2258 txq_trans_update(txq);
2259 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2260 return NETDEV_TX_BUSY;
2261 } while (skb->next);
2264 if (likely(skb->next == NULL))
2265 skb->destructor = DEV_GSO_CB(skb)->destructor;
2272 static u32 hashrnd __read_mostly;
2275 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2276 * to be used as a distribution range.
2278 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2279 unsigned int num_tx_queues)
2283 u16 qcount = num_tx_queues;
2285 if (skb_rx_queue_recorded(skb)) {
2286 hash = skb_get_rx_queue(skb);
2287 while (unlikely(hash >= num_tx_queues))
2288 hash -= num_tx_queues;
2293 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2294 qoffset = dev->tc_to_txq[tc].offset;
2295 qcount = dev->tc_to_txq[tc].count;
2298 if (skb->sk && skb->sk->sk_hash)
2299 hash = skb->sk->sk_hash;
2301 hash = (__force u16) skb->protocol ^ skb->rxhash;
2302 hash = jhash_1word(hash, hashrnd);
2304 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2306 EXPORT_SYMBOL(__skb_tx_hash);
2308 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2310 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2311 if (net_ratelimit()) {
2312 pr_warning("%s selects TX queue %d, but "
2313 "real number of TX queues is %d\n",
2314 dev->name, queue_index, dev->real_num_tx_queues);
2321 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2324 struct xps_dev_maps *dev_maps;
2325 struct xps_map *map;
2326 int queue_index = -1;
2329 dev_maps = rcu_dereference(dev->xps_maps);
2331 map = rcu_dereference(
2332 dev_maps->cpu_map[raw_smp_processor_id()]);
2335 queue_index = map->queues[0];
2338 if (skb->sk && skb->sk->sk_hash)
2339 hash = skb->sk->sk_hash;
2341 hash = (__force u16) skb->protocol ^
2343 hash = jhash_1word(hash, hashrnd);
2344 queue_index = map->queues[
2345 ((u64)hash * map->len) >> 32];
2347 if (unlikely(queue_index >= dev->real_num_tx_queues))
2359 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2360 struct sk_buff *skb)
2363 const struct net_device_ops *ops = dev->netdev_ops;
2365 if (dev->real_num_tx_queues == 1)
2367 else if (ops->ndo_select_queue) {
2368 queue_index = ops->ndo_select_queue(dev, skb);
2369 queue_index = dev_cap_txqueue(dev, queue_index);
2371 struct sock *sk = skb->sk;
2372 queue_index = sk_tx_queue_get(sk);
2374 if (queue_index < 0 || skb->ooo_okay ||
2375 queue_index >= dev->real_num_tx_queues) {
2376 int old_index = queue_index;
2378 queue_index = get_xps_queue(dev, skb);
2379 if (queue_index < 0)
2380 queue_index = skb_tx_hash(dev, skb);
2382 if (queue_index != old_index && sk) {
2383 struct dst_entry *dst =
2384 rcu_dereference_check(sk->sk_dst_cache, 1);
2386 if (dst && skb_dst(skb) == dst)
2387 sk_tx_queue_set(sk, queue_index);
2392 skb_set_queue_mapping(skb, queue_index);
2393 return netdev_get_tx_queue(dev, queue_index);
2396 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2397 struct net_device *dev,
2398 struct netdev_queue *txq)
2400 spinlock_t *root_lock = qdisc_lock(q);
2404 qdisc_skb_cb(skb)->pkt_len = skb->len;
2405 qdisc_calculate_pkt_len(skb, q);
2407 * Heuristic to force contended enqueues to serialize on a
2408 * separate lock before trying to get qdisc main lock.
2409 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2410 * and dequeue packets faster.
2412 contended = qdisc_is_running(q);
2413 if (unlikely(contended))
2414 spin_lock(&q->busylock);
2416 spin_lock(root_lock);
2417 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2420 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2421 qdisc_run_begin(q)) {
2423 * This is a work-conserving queue; there are no old skbs
2424 * waiting to be sent out; and the qdisc is not running -
2425 * xmit the skb directly.
2427 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2430 qdisc_bstats_update(q, skb);
2432 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2433 if (unlikely(contended)) {
2434 spin_unlock(&q->busylock);
2441 rc = NET_XMIT_SUCCESS;
2444 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2445 if (qdisc_run_begin(q)) {
2446 if (unlikely(contended)) {
2447 spin_unlock(&q->busylock);
2453 spin_unlock(root_lock);
2454 if (unlikely(contended))
2455 spin_unlock(&q->busylock);
2459 static DEFINE_PER_CPU(int, xmit_recursion);
2460 #define RECURSION_LIMIT 10
2463 * dev_queue_xmit - transmit a buffer
2464 * @skb: buffer to transmit
2466 * Queue a buffer for transmission to a network device. The caller must
2467 * have set the device and priority and built the buffer before calling
2468 * this function. The function can be called from an interrupt.
2470 * A negative errno code is returned on a failure. A success does not
2471 * guarantee the frame will be transmitted as it may be dropped due
2472 * to congestion or traffic shaping.
2474 * -----------------------------------------------------------------------------------
2475 * I notice this method can also return errors from the queue disciplines,
2476 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2479 * Regardless of the return value, the skb is consumed, so it is currently
2480 * difficult to retry a send to this method. (You can bump the ref count
2481 * before sending to hold a reference for retry if you are careful.)
2483 * When calling this method, interrupts MUST be enabled. This is because
2484 * the BH enable code must have IRQs enabled so that it will not deadlock.
2487 int dev_queue_xmit(struct sk_buff *skb)
2489 struct net_device *dev = skb->dev;
2490 struct netdev_queue *txq;
2494 /* Disable soft irqs for various locks below. Also
2495 * stops preemption for RCU.
2499 txq = dev_pick_tx(dev, skb);
2500 q = rcu_dereference_bh(txq->qdisc);
2502 #ifdef CONFIG_NET_CLS_ACT
2503 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2505 trace_net_dev_queue(skb);
2507 rc = __dev_xmit_skb(skb, q, dev, txq);
2511 /* The device has no queue. Common case for software devices:
2512 loopback, all the sorts of tunnels...
2514 Really, it is unlikely that netif_tx_lock protection is necessary
2515 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2517 However, it is possible, that they rely on protection
2520 Check this and shot the lock. It is not prone from deadlocks.
2521 Either shot noqueue qdisc, it is even simpler 8)
2523 if (dev->flags & IFF_UP) {
2524 int cpu = smp_processor_id(); /* ok because BHs are off */
2526 if (txq->xmit_lock_owner != cpu) {
2528 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2529 goto recursion_alert;
2531 HARD_TX_LOCK(dev, txq, cpu);
2533 if (!netif_tx_queue_stopped(txq)) {
2534 __this_cpu_inc(xmit_recursion);
2535 rc = dev_hard_start_xmit(skb, dev, txq);
2536 __this_cpu_dec(xmit_recursion);
2537 if (dev_xmit_complete(rc)) {
2538 HARD_TX_UNLOCK(dev, txq);
2542 HARD_TX_UNLOCK(dev, txq);
2543 if (net_ratelimit())
2544 printk(KERN_CRIT "Virtual device %s asks to "
2545 "queue packet!\n", dev->name);
2547 /* Recursion is detected! It is possible,
2551 if (net_ratelimit())
2552 printk(KERN_CRIT "Dead loop on virtual device "
2553 "%s, fix it urgently!\n", dev->name);
2558 rcu_read_unlock_bh();
2563 rcu_read_unlock_bh();
2566 EXPORT_SYMBOL(dev_queue_xmit);
2569 /*=======================================================================
2571 =======================================================================*/
2573 int netdev_max_backlog __read_mostly = 1000;
2574 int netdev_tstamp_prequeue __read_mostly = 1;
2575 int netdev_budget __read_mostly = 300;
2576 int weight_p __read_mostly = 64; /* old backlog weight */
2578 /* Called with irq disabled */
2579 static inline void ____napi_schedule(struct softnet_data *sd,
2580 struct napi_struct *napi)
2582 list_add_tail(&napi->poll_list, &sd->poll_list);
2583 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2587 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2588 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2589 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2590 * if hash is a canonical 4-tuple hash over transport ports.
2592 void __skb_get_rxhash(struct sk_buff *skb)
2594 int nhoff, hash = 0, poff;
2595 const struct ipv6hdr *ip6;
2596 const struct iphdr *ip;
2597 const struct vlan_hdr *vlan;
2606 nhoff = skb_network_offset(skb);
2607 proto = skb->protocol;
2611 case __constant_htons(ETH_P_IP):
2613 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2616 ip = (const struct iphdr *) (skb->data + nhoff);
2617 if (ip_is_fragment(ip))
2620 ip_proto = ip->protocol;
2621 addr1 = (__force u32) ip->saddr;
2622 addr2 = (__force u32) ip->daddr;
2623 nhoff += ip->ihl * 4;
2625 case __constant_htons(ETH_P_IPV6):
2627 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2630 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2631 ip_proto = ip6->nexthdr;
2632 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2633 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2636 case __constant_htons(ETH_P_8021Q):
2637 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2639 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2640 proto = vlan->h_vlan_encapsulated_proto;
2641 nhoff += sizeof(*vlan);
2643 case __constant_htons(ETH_P_PPP_SES):
2644 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2646 proto = *((__be16 *) (skb->data + nhoff +
2647 sizeof(struct pppoe_hdr)));
2648 nhoff += PPPOE_SES_HLEN;
2650 case __constant_htons(PPP_IP):
2652 case __constant_htons(PPP_IPV6):
2663 if (pskb_may_pull(skb, nhoff + 16)) {
2664 u8 *h = skb->data + nhoff;
2665 __be16 flags = *(__be16 *)h;
2668 * Only look inside GRE if version zero and no
2671 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2672 proto = *(__be16 *)(h + 2);
2674 if (flags & GRE_CSUM)
2676 if (flags & GRE_KEY)
2678 if (flags & GRE_SEQ)
2691 poff = proto_ports_offset(ip_proto);
2694 if (pskb_may_pull(skb, nhoff + 4)) {
2695 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2696 if (ports.v16[1] < ports.v16[0])
2697 swap(ports.v16[0], ports.v16[1]);
2702 /* get a consistent hash (same value on both flow directions) */
2706 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2713 EXPORT_SYMBOL(__skb_get_rxhash);
2717 /* One global table that all flow-based protocols share. */
2718 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2719 EXPORT_SYMBOL(rps_sock_flow_table);
2721 static struct rps_dev_flow *
2722 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2723 struct rps_dev_flow *rflow, u16 next_cpu)
2725 if (next_cpu != RPS_NO_CPU) {
2726 #ifdef CONFIG_RFS_ACCEL
2727 struct netdev_rx_queue *rxqueue;
2728 struct rps_dev_flow_table *flow_table;
2729 struct rps_dev_flow *old_rflow;
2734 /* Should we steer this flow to a different hardware queue? */
2735 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2736 !(dev->features & NETIF_F_NTUPLE))
2738 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2739 if (rxq_index == skb_get_rx_queue(skb))
2742 rxqueue = dev->_rx + rxq_index;
2743 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2746 flow_id = skb->rxhash & flow_table->mask;
2747 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2748 rxq_index, flow_id);
2752 rflow = &flow_table->flows[flow_id];
2754 if (old_rflow->filter == rflow->filter)
2755 old_rflow->filter = RPS_NO_FILTER;
2759 per_cpu(softnet_data, next_cpu).input_queue_head;
2762 rflow->cpu = next_cpu;
2767 * get_rps_cpu is called from netif_receive_skb and returns the target
2768 * CPU from the RPS map of the receiving queue for a given skb.
2769 * rcu_read_lock must be held on entry.
2771 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2772 struct rps_dev_flow **rflowp)
2774 struct netdev_rx_queue *rxqueue;
2775 struct rps_map *map;
2776 struct rps_dev_flow_table *flow_table;
2777 struct rps_sock_flow_table *sock_flow_table;
2781 if (skb_rx_queue_recorded(skb)) {
2782 u16 index = skb_get_rx_queue(skb);
2783 if (unlikely(index >= dev->real_num_rx_queues)) {
2784 WARN_ONCE(dev->real_num_rx_queues > 1,
2785 "%s received packet on queue %u, but number "
2786 "of RX queues is %u\n",
2787 dev->name, index, dev->real_num_rx_queues);
2790 rxqueue = dev->_rx + index;
2794 map = rcu_dereference(rxqueue->rps_map);
2796 if (map->len == 1 &&
2797 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2798 tcpu = map->cpus[0];
2799 if (cpu_online(tcpu))
2803 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2807 skb_reset_network_header(skb);
2808 if (!skb_get_rxhash(skb))
2811 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2812 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2813 if (flow_table && sock_flow_table) {
2815 struct rps_dev_flow *rflow;
2817 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2820 next_cpu = sock_flow_table->ents[skb->rxhash &
2821 sock_flow_table->mask];
2824 * If the desired CPU (where last recvmsg was done) is
2825 * different from current CPU (one in the rx-queue flow
2826 * table entry), switch if one of the following holds:
2827 * - Current CPU is unset (equal to RPS_NO_CPU).
2828 * - Current CPU is offline.
2829 * - The current CPU's queue tail has advanced beyond the
2830 * last packet that was enqueued using this table entry.
2831 * This guarantees that all previous packets for the flow
2832 * have been dequeued, thus preserving in order delivery.
2834 if (unlikely(tcpu != next_cpu) &&
2835 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2836 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2837 rflow->last_qtail)) >= 0))
2838 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2840 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2848 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2850 if (cpu_online(tcpu)) {
2860 #ifdef CONFIG_RFS_ACCEL
2863 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2864 * @dev: Device on which the filter was set
2865 * @rxq_index: RX queue index
2866 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2867 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2869 * Drivers that implement ndo_rx_flow_steer() should periodically call
2870 * this function for each installed filter and remove the filters for
2871 * which it returns %true.
2873 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2874 u32 flow_id, u16 filter_id)
2876 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2877 struct rps_dev_flow_table *flow_table;
2878 struct rps_dev_flow *rflow;
2883 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2884 if (flow_table && flow_id <= flow_table->mask) {
2885 rflow = &flow_table->flows[flow_id];
2886 cpu = ACCESS_ONCE(rflow->cpu);
2887 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2888 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2889 rflow->last_qtail) <
2890 (int)(10 * flow_table->mask)))
2896 EXPORT_SYMBOL(rps_may_expire_flow);
2898 #endif /* CONFIG_RFS_ACCEL */
2900 /* Called from hardirq (IPI) context */
2901 static void rps_trigger_softirq(void *data)
2903 struct softnet_data *sd = data;
2905 ____napi_schedule(sd, &sd->backlog);
2909 #endif /* CONFIG_RPS */
2912 * Check if this softnet_data structure is another cpu one
2913 * If yes, queue it to our IPI list and return 1
2916 static int rps_ipi_queued(struct softnet_data *sd)
2919 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2922 sd->rps_ipi_next = mysd->rps_ipi_list;
2923 mysd->rps_ipi_list = sd;
2925 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2928 #endif /* CONFIG_RPS */
2933 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2934 * queue (may be a remote CPU queue).
2936 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2937 unsigned int *qtail)
2939 struct softnet_data *sd;
2940 unsigned long flags;
2942 sd = &per_cpu(softnet_data, cpu);
2944 local_irq_save(flags);
2947 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2948 if (skb_queue_len(&sd->input_pkt_queue)) {
2950 __skb_queue_tail(&sd->input_pkt_queue, skb);
2951 input_queue_tail_incr_save(sd, qtail);
2953 local_irq_restore(flags);
2954 return NET_RX_SUCCESS;
2957 /* Schedule NAPI for backlog device
2958 * We can use non atomic operation since we own the queue lock
2960 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2961 if (!rps_ipi_queued(sd))
2962 ____napi_schedule(sd, &sd->backlog);
2970 local_irq_restore(flags);
2972 atomic_long_inc(&skb->dev->rx_dropped);
2978 * netif_rx - post buffer to the network code
2979 * @skb: buffer to post
2981 * This function receives a packet from a device driver and queues it for
2982 * the upper (protocol) levels to process. It always succeeds. The buffer
2983 * may be dropped during processing for congestion control or by the
2987 * NET_RX_SUCCESS (no congestion)
2988 * NET_RX_DROP (packet was dropped)
2992 int netif_rx(struct sk_buff *skb)
2996 /* if netpoll wants it, pretend we never saw it */
2997 if (netpoll_rx(skb))
3000 if (netdev_tstamp_prequeue)
3001 net_timestamp_check(skb);
3003 trace_netif_rx(skb);
3006 struct rps_dev_flow voidflow, *rflow = &voidflow;
3012 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3014 cpu = smp_processor_id();
3016 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3024 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3030 EXPORT_SYMBOL(netif_rx);
3032 int netif_rx_ni(struct sk_buff *skb)
3037 err = netif_rx(skb);
3038 if (local_softirq_pending())
3044 EXPORT_SYMBOL(netif_rx_ni);
3046 static void net_tx_action(struct softirq_action *h)
3048 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3050 if (sd->completion_queue) {
3051 struct sk_buff *clist;
3053 local_irq_disable();
3054 clist = sd->completion_queue;
3055 sd->completion_queue = NULL;
3059 struct sk_buff *skb = clist;
3060 clist = clist->next;
3062 WARN_ON(atomic_read(&skb->users));
3063 trace_kfree_skb(skb, net_tx_action);
3068 if (sd->output_queue) {
3071 local_irq_disable();
3072 head = sd->output_queue;
3073 sd->output_queue = NULL;
3074 sd->output_queue_tailp = &sd->output_queue;
3078 struct Qdisc *q = head;
3079 spinlock_t *root_lock;
3081 head = head->next_sched;
3083 root_lock = qdisc_lock(q);
3084 if (spin_trylock(root_lock)) {
3085 smp_mb__before_clear_bit();
3086 clear_bit(__QDISC_STATE_SCHED,
3089 spin_unlock(root_lock);
3091 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3093 __netif_reschedule(q);
3095 smp_mb__before_clear_bit();
3096 clear_bit(__QDISC_STATE_SCHED,
3104 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3105 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3106 /* This hook is defined here for ATM LANE */
3107 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3108 unsigned char *addr) __read_mostly;
3109 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3112 #ifdef CONFIG_NET_CLS_ACT
3113 /* TODO: Maybe we should just force sch_ingress to be compiled in
3114 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3115 * a compare and 2 stores extra right now if we dont have it on
3116 * but have CONFIG_NET_CLS_ACT
3117 * NOTE: This doesn't stop any functionality; if you dont have
3118 * the ingress scheduler, you just can't add policies on ingress.
3121 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3123 struct net_device *dev = skb->dev;
3124 u32 ttl = G_TC_RTTL(skb->tc_verd);
3125 int result = TC_ACT_OK;
3128 if (unlikely(MAX_RED_LOOP < ttl++)) {
3129 if (net_ratelimit())
3130 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3131 skb->skb_iif, dev->ifindex);
3135 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3136 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3139 if (q != &noop_qdisc) {
3140 spin_lock(qdisc_lock(q));
3141 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3142 result = qdisc_enqueue_root(skb, q);
3143 spin_unlock(qdisc_lock(q));
3149 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3150 struct packet_type **pt_prev,
3151 int *ret, struct net_device *orig_dev)
3153 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3155 if (!rxq || rxq->qdisc == &noop_qdisc)
3159 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3163 switch (ing_filter(skb, rxq)) {
3177 * netdev_rx_handler_register - register receive handler
3178 * @dev: device to register a handler for
3179 * @rx_handler: receive handler to register
3180 * @rx_handler_data: data pointer that is used by rx handler
3182 * Register a receive hander for a device. This handler will then be
3183 * called from __netif_receive_skb. A negative errno code is returned
3186 * The caller must hold the rtnl_mutex.
3188 * For a general description of rx_handler, see enum rx_handler_result.
3190 int netdev_rx_handler_register(struct net_device *dev,
3191 rx_handler_func_t *rx_handler,
3192 void *rx_handler_data)
3196 if (dev->rx_handler)
3199 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3200 rcu_assign_pointer(dev->rx_handler, rx_handler);
3204 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3207 * netdev_rx_handler_unregister - unregister receive handler
3208 * @dev: device to unregister a handler from
3210 * Unregister a receive hander from a device.
3212 * The caller must hold the rtnl_mutex.
3214 void netdev_rx_handler_unregister(struct net_device *dev)
3218 RCU_INIT_POINTER(dev->rx_handler, NULL);
3219 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3221 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3223 static int __netif_receive_skb(struct sk_buff *skb)
3225 struct packet_type *ptype, *pt_prev;
3226 rx_handler_func_t *rx_handler;
3227 struct net_device *orig_dev;
3228 struct net_device *null_or_dev;
3229 bool deliver_exact = false;
3230 int ret = NET_RX_DROP;
3233 if (!netdev_tstamp_prequeue)
3234 net_timestamp_check(skb);
3236 trace_netif_receive_skb(skb);
3238 /* if we've gotten here through NAPI, check netpoll */
3239 if (netpoll_receive_skb(skb))
3243 skb->skb_iif = skb->dev->ifindex;
3244 orig_dev = skb->dev;
3246 skb_reset_network_header(skb);
3247 skb_reset_transport_header(skb);
3248 skb_reset_mac_len(skb);
3256 __this_cpu_inc(softnet_data.processed);
3258 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3259 skb = vlan_untag(skb);
3264 #ifdef CONFIG_NET_CLS_ACT
3265 if (skb->tc_verd & TC_NCLS) {
3266 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3271 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3272 if (!ptype->dev || ptype->dev == skb->dev) {
3274 ret = deliver_skb(skb, pt_prev, orig_dev);
3279 #ifdef CONFIG_NET_CLS_ACT
3280 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3286 rx_handler = rcu_dereference(skb->dev->rx_handler);
3287 if (vlan_tx_tag_present(skb)) {
3289 ret = deliver_skb(skb, pt_prev, orig_dev);
3292 if (vlan_do_receive(&skb, !rx_handler))
3294 else if (unlikely(!skb))
3300 ret = deliver_skb(skb, pt_prev, orig_dev);
3303 switch (rx_handler(&skb)) {
3304 case RX_HANDLER_CONSUMED:
3306 case RX_HANDLER_ANOTHER:
3308 case RX_HANDLER_EXACT:
3309 deliver_exact = true;
3310 case RX_HANDLER_PASS:
3317 /* deliver only exact match when indicated */
3318 null_or_dev = deliver_exact ? skb->dev : NULL;
3320 type = skb->protocol;
3321 list_for_each_entry_rcu(ptype,
3322 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3323 if (ptype->type == type &&
3324 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3325 ptype->dev == orig_dev)) {
3327 ret = deliver_skb(skb, pt_prev, orig_dev);
3333 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3335 atomic_long_inc(&skb->dev->rx_dropped);
3337 /* Jamal, now you will not able to escape explaining
3338 * me how you were going to use this. :-)
3349 * netif_receive_skb - process receive buffer from network
3350 * @skb: buffer to process
3352 * netif_receive_skb() is the main receive data processing function.
3353 * It always succeeds. The buffer may be dropped during processing
3354 * for congestion control or by the protocol layers.
3356 * This function may only be called from softirq context and interrupts
3357 * should be enabled.
3359 * Return values (usually ignored):
3360 * NET_RX_SUCCESS: no congestion
3361 * NET_RX_DROP: packet was dropped
3363 int netif_receive_skb(struct sk_buff *skb)
3365 if (netdev_tstamp_prequeue)
3366 net_timestamp_check(skb);
3368 if (skb_defer_rx_timestamp(skb))
3369 return NET_RX_SUCCESS;
3373 struct rps_dev_flow voidflow, *rflow = &voidflow;
3378 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3381 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3385 ret = __netif_receive_skb(skb);
3391 return __netif_receive_skb(skb);
3394 EXPORT_SYMBOL(netif_receive_skb);
3396 /* Network device is going away, flush any packets still pending
3397 * Called with irqs disabled.
3399 static void flush_backlog(void *arg)
3401 struct net_device *dev = arg;
3402 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3403 struct sk_buff *skb, *tmp;
3406 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3407 if (skb->dev == dev) {
3408 __skb_unlink(skb, &sd->input_pkt_queue);
3410 input_queue_head_incr(sd);
3415 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3416 if (skb->dev == dev) {
3417 __skb_unlink(skb, &sd->process_queue);
3419 input_queue_head_incr(sd);
3424 static int napi_gro_complete(struct sk_buff *skb)
3426 struct packet_type *ptype;
3427 __be16 type = skb->protocol;
3428 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3431 if (NAPI_GRO_CB(skb)->count == 1) {
3432 skb_shinfo(skb)->gso_size = 0;
3437 list_for_each_entry_rcu(ptype, head, list) {
3438 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3441 err = ptype->gro_complete(skb);
3447 WARN_ON(&ptype->list == head);
3449 return NET_RX_SUCCESS;
3453 return netif_receive_skb(skb);
3456 inline void napi_gro_flush(struct napi_struct *napi)
3458 struct sk_buff *skb, *next;
3460 for (skb = napi->gro_list; skb; skb = next) {
3463 napi_gro_complete(skb);
3466 napi->gro_count = 0;
3467 napi->gro_list = NULL;
3469 EXPORT_SYMBOL(napi_gro_flush);
3471 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3473 struct sk_buff **pp = NULL;
3474 struct packet_type *ptype;
3475 __be16 type = skb->protocol;
3476 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3479 enum gro_result ret;
3481 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3484 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3488 list_for_each_entry_rcu(ptype, head, list) {
3489 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3492 skb_set_network_header(skb, skb_gro_offset(skb));
3493 mac_len = skb->network_header - skb->mac_header;
3494 skb->mac_len = mac_len;
3495 NAPI_GRO_CB(skb)->same_flow = 0;
3496 NAPI_GRO_CB(skb)->flush = 0;
3497 NAPI_GRO_CB(skb)->free = 0;
3499 pp = ptype->gro_receive(&napi->gro_list, skb);
3504 if (&ptype->list == head)
3507 same_flow = NAPI_GRO_CB(skb)->same_flow;
3508 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3511 struct sk_buff *nskb = *pp;
3515 napi_gro_complete(nskb);
3522 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3526 NAPI_GRO_CB(skb)->count = 1;
3527 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3528 skb->next = napi->gro_list;
3529 napi->gro_list = skb;
3533 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3534 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3536 BUG_ON(skb->end - skb->tail < grow);
3538 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3541 skb->data_len -= grow;
3543 skb_shinfo(skb)->frags[0].page_offset += grow;
3544 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3546 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3547 skb_frag_unref(skb, 0);
3548 memmove(skb_shinfo(skb)->frags,
3549 skb_shinfo(skb)->frags + 1,
3550 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3561 EXPORT_SYMBOL(dev_gro_receive);
3563 static inline gro_result_t
3564 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3568 for (p = napi->gro_list; p; p = p->next) {
3569 unsigned long diffs;
3571 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3572 diffs |= p->vlan_tci ^ skb->vlan_tci;
3573 diffs |= compare_ether_header(skb_mac_header(p),
3574 skb_gro_mac_header(skb));
3575 NAPI_GRO_CB(p)->same_flow = !diffs;
3576 NAPI_GRO_CB(p)->flush = 0;
3579 return dev_gro_receive(napi, skb);
3582 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3586 if (netif_receive_skb(skb))
3591 case GRO_MERGED_FREE:
3602 EXPORT_SYMBOL(napi_skb_finish);
3604 void skb_gro_reset_offset(struct sk_buff *skb)
3606 NAPI_GRO_CB(skb)->data_offset = 0;
3607 NAPI_GRO_CB(skb)->frag0 = NULL;
3608 NAPI_GRO_CB(skb)->frag0_len = 0;
3610 if (skb->mac_header == skb->tail &&
3611 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3612 NAPI_GRO_CB(skb)->frag0 =
3613 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3614 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3617 EXPORT_SYMBOL(skb_gro_reset_offset);
3619 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3621 skb_gro_reset_offset(skb);
3623 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3625 EXPORT_SYMBOL(napi_gro_receive);
3627 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3629 __skb_pull(skb, skb_headlen(skb));
3630 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3632 skb->dev = napi->dev;
3638 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3640 struct sk_buff *skb = napi->skb;
3643 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3649 EXPORT_SYMBOL(napi_get_frags);
3651 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3657 skb->protocol = eth_type_trans(skb, skb->dev);
3659 if (ret == GRO_HELD)
3660 skb_gro_pull(skb, -ETH_HLEN);
3661 else if (netif_receive_skb(skb))
3666 case GRO_MERGED_FREE:
3667 napi_reuse_skb(napi, skb);
3676 EXPORT_SYMBOL(napi_frags_finish);
3678 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3680 struct sk_buff *skb = napi->skb;
3687 skb_reset_mac_header(skb);
3688 skb_gro_reset_offset(skb);
3690 off = skb_gro_offset(skb);
3691 hlen = off + sizeof(*eth);
3692 eth = skb_gro_header_fast(skb, off);
3693 if (skb_gro_header_hard(skb, hlen)) {
3694 eth = skb_gro_header_slow(skb, hlen, off);
3695 if (unlikely(!eth)) {
3696 napi_reuse_skb(napi, skb);
3702 skb_gro_pull(skb, sizeof(*eth));
3705 * This works because the only protocols we care about don't require
3706 * special handling. We'll fix it up properly at the end.
3708 skb->protocol = eth->h_proto;
3713 EXPORT_SYMBOL(napi_frags_skb);
3715 gro_result_t napi_gro_frags(struct napi_struct *napi)
3717 struct sk_buff *skb = napi_frags_skb(napi);
3722 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3724 EXPORT_SYMBOL(napi_gro_frags);
3727 * net_rps_action sends any pending IPI's for rps.
3728 * Note: called with local irq disabled, but exits with local irq enabled.
3730 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3733 struct softnet_data *remsd = sd->rps_ipi_list;
3736 sd->rps_ipi_list = NULL;
3740 /* Send pending IPI's to kick RPS processing on remote cpus. */
3742 struct softnet_data *next = remsd->rps_ipi_next;
3744 if (cpu_online(remsd->cpu))
3745 __smp_call_function_single(remsd->cpu,
3754 static int process_backlog(struct napi_struct *napi, int quota)
3757 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3760 /* Check if we have pending ipi, its better to send them now,
3761 * not waiting net_rx_action() end.
3763 if (sd->rps_ipi_list) {
3764 local_irq_disable();
3765 net_rps_action_and_irq_enable(sd);
3768 napi->weight = weight_p;
3769 local_irq_disable();
3770 while (work < quota) {
3771 struct sk_buff *skb;
3774 while ((skb = __skb_dequeue(&sd->process_queue))) {
3776 __netif_receive_skb(skb);
3777 local_irq_disable();
3778 input_queue_head_incr(sd);
3779 if (++work >= quota) {
3786 qlen = skb_queue_len(&sd->input_pkt_queue);
3788 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3789 &sd->process_queue);
3791 if (qlen < quota - work) {
3793 * Inline a custom version of __napi_complete().
3794 * only current cpu owns and manipulates this napi,
3795 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3796 * we can use a plain write instead of clear_bit(),
3797 * and we dont need an smp_mb() memory barrier.
3799 list_del(&napi->poll_list);
3802 quota = work + qlen;
3812 * __napi_schedule - schedule for receive
3813 * @n: entry to schedule
3815 * The entry's receive function will be scheduled to run
3817 void __napi_schedule(struct napi_struct *n)
3819 unsigned long flags;
3821 local_irq_save(flags);
3822 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3823 local_irq_restore(flags);
3825 EXPORT_SYMBOL(__napi_schedule);
3827 void __napi_complete(struct napi_struct *n)
3829 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3830 BUG_ON(n->gro_list);
3832 list_del(&n->poll_list);
3833 smp_mb__before_clear_bit();
3834 clear_bit(NAPI_STATE_SCHED, &n->state);
3836 EXPORT_SYMBOL(__napi_complete);
3838 void napi_complete(struct napi_struct *n)
3840 unsigned long flags;
3843 * don't let napi dequeue from the cpu poll list
3844 * just in case its running on a different cpu
3846 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3850 local_irq_save(flags);
3852 local_irq_restore(flags);
3854 EXPORT_SYMBOL(napi_complete);
3856 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3857 int (*poll)(struct napi_struct *, int), int weight)
3859 INIT_LIST_HEAD(&napi->poll_list);
3860 napi->gro_count = 0;
3861 napi->gro_list = NULL;
3864 napi->weight = weight;
3865 list_add(&napi->dev_list, &dev->napi_list);
3867 #ifdef CONFIG_NETPOLL
3868 spin_lock_init(&napi->poll_lock);
3869 napi->poll_owner = -1;
3871 set_bit(NAPI_STATE_SCHED, &napi->state);
3873 EXPORT_SYMBOL(netif_napi_add);
3875 void netif_napi_del(struct napi_struct *napi)
3877 struct sk_buff *skb, *next;
3879 list_del_init(&napi->dev_list);
3880 napi_free_frags(napi);
3882 for (skb = napi->gro_list; skb; skb = next) {
3888 napi->gro_list = NULL;
3889 napi->gro_count = 0;
3891 EXPORT_SYMBOL(netif_napi_del);
3893 static void net_rx_action(struct softirq_action *h)
3895 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3896 unsigned long time_limit = jiffies + 2;
3897 int budget = netdev_budget;
3900 local_irq_disable();
3902 while (!list_empty(&sd->poll_list)) {
3903 struct napi_struct *n;
3906 /* If softirq window is exhuasted then punt.
3907 * Allow this to run for 2 jiffies since which will allow
3908 * an average latency of 1.5/HZ.
3910 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3915 /* Even though interrupts have been re-enabled, this
3916 * access is safe because interrupts can only add new
3917 * entries to the tail of this list, and only ->poll()
3918 * calls can remove this head entry from the list.
3920 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3922 have = netpoll_poll_lock(n);
3926 /* This NAPI_STATE_SCHED test is for avoiding a race
3927 * with netpoll's poll_napi(). Only the entity which
3928 * obtains the lock and sees NAPI_STATE_SCHED set will
3929 * actually make the ->poll() call. Therefore we avoid
3930 * accidentally calling ->poll() when NAPI is not scheduled.
3933 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3934 work = n->poll(n, weight);
3938 WARN_ON_ONCE(work > weight);
3942 local_irq_disable();
3944 /* Drivers must not modify the NAPI state if they
3945 * consume the entire weight. In such cases this code
3946 * still "owns" the NAPI instance and therefore can
3947 * move the instance around on the list at-will.
3949 if (unlikely(work == weight)) {
3950 if (unlikely(napi_disable_pending(n))) {
3953 local_irq_disable();
3955 list_move_tail(&n->poll_list, &sd->poll_list);
3958 netpoll_poll_unlock(have);
3961 net_rps_action_and_irq_enable(sd);
3963 #ifdef CONFIG_NET_DMA
3965 * There may not be any more sk_buffs coming right now, so push
3966 * any pending DMA copies to hardware
3968 dma_issue_pending_all();
3975 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3979 static gifconf_func_t *gifconf_list[NPROTO];
3982 * register_gifconf - register a SIOCGIF handler
3983 * @family: Address family
3984 * @gifconf: Function handler
3986 * Register protocol dependent address dumping routines. The handler
3987 * that is passed must not be freed or reused until it has been replaced
3988 * by another handler.
3990 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3992 if (family >= NPROTO)
3994 gifconf_list[family] = gifconf;
3997 EXPORT_SYMBOL(register_gifconf);
4001 * Map an interface index to its name (SIOCGIFNAME)
4005 * We need this ioctl for efficient implementation of the
4006 * if_indextoname() function required by the IPv6 API. Without
4007 * it, we would have to search all the interfaces to find a
4011 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4013 struct net_device *dev;
4017 * Fetch the caller's info block.
4020 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4024 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4030 strcpy(ifr.ifr_name, dev->name);
4033 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4039 * Perform a SIOCGIFCONF call. This structure will change
4040 * size eventually, and there is nothing I can do about it.
4041 * Thus we will need a 'compatibility mode'.
4044 static int dev_ifconf(struct net *net, char __user *arg)
4047 struct net_device *dev;
4054 * Fetch the caller's info block.
4057 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4064 * Loop over the interfaces, and write an info block for each.
4068 for_each_netdev(net, dev) {
4069 for (i = 0; i < NPROTO; i++) {
4070 if (gifconf_list[i]) {
4073 done = gifconf_list[i](dev, NULL, 0);
4075 done = gifconf_list[i](dev, pos + total,
4085 * All done. Write the updated control block back to the caller.
4087 ifc.ifc_len = total;
4090 * Both BSD and Solaris return 0 here, so we do too.
4092 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4095 #ifdef CONFIG_PROC_FS
4097 #define BUCKET_SPACE (32 - NETDEV_HASHBITS)
4099 struct dev_iter_state {
4100 struct seq_net_private p;
4101 unsigned int pos; /* bucket << BUCKET_SPACE + offset */
4104 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4105 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4106 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4108 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq)
4110 struct dev_iter_state *state = seq->private;
4111 struct net *net = seq_file_net(seq);
4112 struct net_device *dev;
4113 struct hlist_node *p;
4114 struct hlist_head *h;
4115 unsigned int count, bucket, offset;
4117 bucket = get_bucket(state->pos);
4118 offset = get_offset(state->pos);
4119 h = &net->dev_name_head[bucket];
4121 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4122 if (count++ == offset) {
4123 state->pos = set_bucket_offset(bucket, count);
4131 static inline struct net_device *dev_from_new_bucket(struct seq_file *seq)
4133 struct dev_iter_state *state = seq->private;
4134 struct net_device *dev;
4135 unsigned int bucket;
4137 bucket = get_bucket(state->pos);
4139 dev = dev_from_same_bucket(seq);
4144 state->pos = set_bucket_offset(bucket, 0);
4145 } while (bucket < NETDEV_HASHENTRIES);
4151 * This is invoked by the /proc filesystem handler to display a device
4154 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4157 struct dev_iter_state *state = seq->private;
4161 return SEQ_START_TOKEN;
4163 /* check for end of the hash */
4164 if (state->pos == 0 && *pos > 1)
4167 return dev_from_new_bucket(seq);
4170 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4172 struct net_device *dev;
4176 if (v == SEQ_START_TOKEN)
4177 return dev_from_new_bucket(seq);
4179 dev = dev_from_same_bucket(seq);
4183 return dev_from_new_bucket(seq);
4186 void dev_seq_stop(struct seq_file *seq, void *v)
4192 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4194 struct rtnl_link_stats64 temp;
4195 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4197 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4198 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4199 dev->name, stats->rx_bytes, stats->rx_packets,
4201 stats->rx_dropped + stats->rx_missed_errors,
4202 stats->rx_fifo_errors,
4203 stats->rx_length_errors + stats->rx_over_errors +
4204 stats->rx_crc_errors + stats->rx_frame_errors,
4205 stats->rx_compressed, stats->multicast,
4206 stats->tx_bytes, stats->tx_packets,
4207 stats->tx_errors, stats->tx_dropped,
4208 stats->tx_fifo_errors, stats->collisions,
4209 stats->tx_carrier_errors +
4210 stats->tx_aborted_errors +
4211 stats->tx_window_errors +
4212 stats->tx_heartbeat_errors,
4213 stats->tx_compressed);
4217 * Called from the PROCfs module. This now uses the new arbitrary sized
4218 * /proc/net interface to create /proc/net/dev
4220 static int dev_seq_show(struct seq_file *seq, void *v)
4222 if (v == SEQ_START_TOKEN)
4223 seq_puts(seq, "Inter-| Receive "
4225 " face |bytes packets errs drop fifo frame "
4226 "compressed multicast|bytes packets errs "
4227 "drop fifo colls carrier compressed\n");
4229 dev_seq_printf_stats(seq, v);
4233 static struct softnet_data *softnet_get_online(loff_t *pos)
4235 struct softnet_data *sd = NULL;
4237 while (*pos < nr_cpu_ids)
4238 if (cpu_online(*pos)) {
4239 sd = &per_cpu(softnet_data, *pos);
4246 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4248 return softnet_get_online(pos);
4251 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4254 return softnet_get_online(pos);
4257 static void softnet_seq_stop(struct seq_file *seq, void *v)
4261 static int softnet_seq_show(struct seq_file *seq, void *v)
4263 struct softnet_data *sd = v;
4265 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4266 sd->processed, sd->dropped, sd->time_squeeze, 0,
4267 0, 0, 0, 0, /* was fastroute */
4268 sd->cpu_collision, sd->received_rps);
4272 static const struct seq_operations dev_seq_ops = {
4273 .start = dev_seq_start,
4274 .next = dev_seq_next,
4275 .stop = dev_seq_stop,
4276 .show = dev_seq_show,
4279 static int dev_seq_open(struct inode *inode, struct file *file)
4281 return seq_open_net(inode, file, &dev_seq_ops,
4282 sizeof(struct dev_iter_state));
4285 int dev_seq_open_ops(struct inode *inode, struct file *file,
4286 const struct seq_operations *ops)
4288 return seq_open_net(inode, file, ops, sizeof(struct dev_iter_state));
4291 static const struct file_operations dev_seq_fops = {
4292 .owner = THIS_MODULE,
4293 .open = dev_seq_open,
4295 .llseek = seq_lseek,
4296 .release = seq_release_net,
4299 static const struct seq_operations softnet_seq_ops = {
4300 .start = softnet_seq_start,
4301 .next = softnet_seq_next,
4302 .stop = softnet_seq_stop,
4303 .show = softnet_seq_show,
4306 static int softnet_seq_open(struct inode *inode, struct file *file)
4308 return seq_open(file, &softnet_seq_ops);
4311 static const struct file_operations softnet_seq_fops = {
4312 .owner = THIS_MODULE,
4313 .open = softnet_seq_open,
4315 .llseek = seq_lseek,
4316 .release = seq_release,
4319 static void *ptype_get_idx(loff_t pos)
4321 struct packet_type *pt = NULL;
4325 list_for_each_entry_rcu(pt, &ptype_all, list) {
4331 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4332 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4341 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4345 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4348 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4350 struct packet_type *pt;
4351 struct list_head *nxt;
4355 if (v == SEQ_START_TOKEN)
4356 return ptype_get_idx(0);
4359 nxt = pt->list.next;
4360 if (pt->type == htons(ETH_P_ALL)) {
4361 if (nxt != &ptype_all)
4364 nxt = ptype_base[0].next;
4366 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4368 while (nxt == &ptype_base[hash]) {
4369 if (++hash >= PTYPE_HASH_SIZE)
4371 nxt = ptype_base[hash].next;
4374 return list_entry(nxt, struct packet_type, list);
4377 static void ptype_seq_stop(struct seq_file *seq, void *v)
4383 static int ptype_seq_show(struct seq_file *seq, void *v)
4385 struct packet_type *pt = v;
4387 if (v == SEQ_START_TOKEN)
4388 seq_puts(seq, "Type Device Function\n");
4389 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4390 if (pt->type == htons(ETH_P_ALL))
4391 seq_puts(seq, "ALL ");
4393 seq_printf(seq, "%04x", ntohs(pt->type));
4395 seq_printf(seq, " %-8s %pF\n",
4396 pt->dev ? pt->dev->name : "", pt->func);
4402 static const struct seq_operations ptype_seq_ops = {
4403 .start = ptype_seq_start,
4404 .next = ptype_seq_next,
4405 .stop = ptype_seq_stop,
4406 .show = ptype_seq_show,
4409 static int ptype_seq_open(struct inode *inode, struct file *file)
4411 return seq_open_net(inode, file, &ptype_seq_ops,
4412 sizeof(struct seq_net_private));
4415 static const struct file_operations ptype_seq_fops = {
4416 .owner = THIS_MODULE,
4417 .open = ptype_seq_open,
4419 .llseek = seq_lseek,
4420 .release = seq_release_net,
4424 static int __net_init dev_proc_net_init(struct net *net)
4428 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4430 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4432 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4435 if (wext_proc_init(net))
4441 proc_net_remove(net, "ptype");
4443 proc_net_remove(net, "softnet_stat");
4445 proc_net_remove(net, "dev");
4449 static void __net_exit dev_proc_net_exit(struct net *net)
4451 wext_proc_exit(net);
4453 proc_net_remove(net, "ptype");
4454 proc_net_remove(net, "softnet_stat");
4455 proc_net_remove(net, "dev");
4458 static struct pernet_operations __net_initdata dev_proc_ops = {
4459 .init = dev_proc_net_init,
4460 .exit = dev_proc_net_exit,
4463 static int __init dev_proc_init(void)
4465 return register_pernet_subsys(&dev_proc_ops);
4468 #define dev_proc_init() 0
4469 #endif /* CONFIG_PROC_FS */
4473 * netdev_set_master - set up master pointer
4474 * @slave: slave device
4475 * @master: new master device
4477 * Changes the master device of the slave. Pass %NULL to break the
4478 * bonding. The caller must hold the RTNL semaphore. On a failure
4479 * a negative errno code is returned. On success the reference counts
4480 * are adjusted and the function returns zero.
4482 int netdev_set_master(struct net_device *slave, struct net_device *master)
4484 struct net_device *old = slave->master;
4494 slave->master = master;
4500 EXPORT_SYMBOL(netdev_set_master);
4503 * netdev_set_bond_master - set up bonding master/slave pair
4504 * @slave: slave device
4505 * @master: new master device
4507 * Changes the master device of the slave. Pass %NULL to break the
4508 * bonding. The caller must hold the RTNL semaphore. On a failure
4509 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4510 * to the routing socket and the function returns zero.
4512 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4518 err = netdev_set_master(slave, master);
4522 slave->flags |= IFF_SLAVE;
4524 slave->flags &= ~IFF_SLAVE;
4526 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4529 EXPORT_SYMBOL(netdev_set_bond_master);
4531 static void dev_change_rx_flags(struct net_device *dev, int flags)
4533 const struct net_device_ops *ops = dev->netdev_ops;
4535 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4536 ops->ndo_change_rx_flags(dev, flags);
4539 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4541 unsigned short old_flags = dev->flags;
4547 dev->flags |= IFF_PROMISC;
4548 dev->promiscuity += inc;
4549 if (dev->promiscuity == 0) {
4552 * If inc causes overflow, untouch promisc and return error.
4555 dev->flags &= ~IFF_PROMISC;
4557 dev->promiscuity -= inc;
4558 printk(KERN_WARNING "%s: promiscuity touches roof, "
4559 "set promiscuity failed, promiscuity feature "
4560 "of device might be broken.\n", dev->name);
4564 if (dev->flags != old_flags) {
4565 printk(KERN_INFO "device %s %s promiscuous mode\n",
4566 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4568 if (audit_enabled) {
4569 current_uid_gid(&uid, &gid);
4570 audit_log(current->audit_context, GFP_ATOMIC,
4571 AUDIT_ANOM_PROMISCUOUS,
4572 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4573 dev->name, (dev->flags & IFF_PROMISC),
4574 (old_flags & IFF_PROMISC),
4575 audit_get_loginuid(current),
4577 audit_get_sessionid(current));
4580 dev_change_rx_flags(dev, IFF_PROMISC);
4586 * dev_set_promiscuity - update promiscuity count on a device
4590 * Add or remove promiscuity from a device. While the count in the device
4591 * remains above zero the interface remains promiscuous. Once it hits zero
4592 * the device reverts back to normal filtering operation. A negative inc
4593 * value is used to drop promiscuity on the device.
4594 * Return 0 if successful or a negative errno code on error.
4596 int dev_set_promiscuity(struct net_device *dev, int inc)
4598 unsigned short old_flags = dev->flags;
4601 err = __dev_set_promiscuity(dev, inc);
4604 if (dev->flags != old_flags)
4605 dev_set_rx_mode(dev);
4608 EXPORT_SYMBOL(dev_set_promiscuity);
4611 * dev_set_allmulti - update allmulti count on a device
4615 * Add or remove reception of all multicast frames to a device. While the
4616 * count in the device remains above zero the interface remains listening
4617 * to all interfaces. Once it hits zero the device reverts back to normal
4618 * filtering operation. A negative @inc value is used to drop the counter
4619 * when releasing a resource needing all multicasts.
4620 * Return 0 if successful or a negative errno code on error.
4623 int dev_set_allmulti(struct net_device *dev, int inc)
4625 unsigned short old_flags = dev->flags;
4629 dev->flags |= IFF_ALLMULTI;
4630 dev->allmulti += inc;
4631 if (dev->allmulti == 0) {
4634 * If inc causes overflow, untouch allmulti and return error.
4637 dev->flags &= ~IFF_ALLMULTI;
4639 dev->allmulti -= inc;
4640 printk(KERN_WARNING "%s: allmulti touches roof, "
4641 "set allmulti failed, allmulti feature of "
4642 "device might be broken.\n", dev->name);
4646 if (dev->flags ^ old_flags) {
4647 dev_change_rx_flags(dev, IFF_ALLMULTI);
4648 dev_set_rx_mode(dev);
4652 EXPORT_SYMBOL(dev_set_allmulti);
4655 * Upload unicast and multicast address lists to device and
4656 * configure RX filtering. When the device doesn't support unicast
4657 * filtering it is put in promiscuous mode while unicast addresses
4660 void __dev_set_rx_mode(struct net_device *dev)
4662 const struct net_device_ops *ops = dev->netdev_ops;
4664 /* dev_open will call this function so the list will stay sane. */
4665 if (!(dev->flags&IFF_UP))
4668 if (!netif_device_present(dev))
4671 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4672 /* Unicast addresses changes may only happen under the rtnl,
4673 * therefore calling __dev_set_promiscuity here is safe.
4675 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4676 __dev_set_promiscuity(dev, 1);
4677 dev->uc_promisc = true;
4678 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4679 __dev_set_promiscuity(dev, -1);
4680 dev->uc_promisc = false;
4684 if (ops->ndo_set_rx_mode)
4685 ops->ndo_set_rx_mode(dev);
4688 void dev_set_rx_mode(struct net_device *dev)
4690 netif_addr_lock_bh(dev);
4691 __dev_set_rx_mode(dev);
4692 netif_addr_unlock_bh(dev);
4696 * dev_get_flags - get flags reported to userspace
4699 * Get the combination of flag bits exported through APIs to userspace.
4701 unsigned dev_get_flags(const struct net_device *dev)
4705 flags = (dev->flags & ~(IFF_PROMISC |
4710 (dev->gflags & (IFF_PROMISC |
4713 if (netif_running(dev)) {
4714 if (netif_oper_up(dev))
4715 flags |= IFF_RUNNING;
4716 if (netif_carrier_ok(dev))
4717 flags |= IFF_LOWER_UP;
4718 if (netif_dormant(dev))
4719 flags |= IFF_DORMANT;
4724 EXPORT_SYMBOL(dev_get_flags);
4726 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4728 int old_flags = dev->flags;
4734 * Set the flags on our device.
4737 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4738 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4740 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4744 * Load in the correct multicast list now the flags have changed.
4747 if ((old_flags ^ flags) & IFF_MULTICAST)
4748 dev_change_rx_flags(dev, IFF_MULTICAST);
4750 dev_set_rx_mode(dev);
4753 * Have we downed the interface. We handle IFF_UP ourselves
4754 * according to user attempts to set it, rather than blindly
4759 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4760 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4763 dev_set_rx_mode(dev);
4766 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4767 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4769 dev->gflags ^= IFF_PROMISC;
4770 dev_set_promiscuity(dev, inc);
4773 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4774 is important. Some (broken) drivers set IFF_PROMISC, when
4775 IFF_ALLMULTI is requested not asking us and not reporting.
4777 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4778 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4780 dev->gflags ^= IFF_ALLMULTI;
4781 dev_set_allmulti(dev, inc);
4787 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4789 unsigned int changes = dev->flags ^ old_flags;
4791 if (changes & IFF_UP) {
4792 if (dev->flags & IFF_UP)
4793 call_netdevice_notifiers(NETDEV_UP, dev);
4795 call_netdevice_notifiers(NETDEV_DOWN, dev);
4798 if (dev->flags & IFF_UP &&
4799 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4800 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4804 * dev_change_flags - change device settings
4806 * @flags: device state flags
4808 * Change settings on device based state flags. The flags are
4809 * in the userspace exported format.
4811 int dev_change_flags(struct net_device *dev, unsigned flags)
4814 int old_flags = dev->flags;
4816 ret = __dev_change_flags(dev, flags);
4820 changes = old_flags ^ dev->flags;
4822 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4824 __dev_notify_flags(dev, old_flags);
4827 EXPORT_SYMBOL(dev_change_flags);
4830 * dev_set_mtu - Change maximum transfer unit
4832 * @new_mtu: new transfer unit
4834 * Change the maximum transfer size of the network device.
4836 int dev_set_mtu(struct net_device *dev, int new_mtu)
4838 const struct net_device_ops *ops = dev->netdev_ops;
4841 if (new_mtu == dev->mtu)
4844 /* MTU must be positive. */
4848 if (!netif_device_present(dev))
4852 if (ops->ndo_change_mtu)
4853 err = ops->ndo_change_mtu(dev, new_mtu);
4857 if (!err && dev->flags & IFF_UP)
4858 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4861 EXPORT_SYMBOL(dev_set_mtu);
4864 * dev_set_group - Change group this device belongs to
4866 * @new_group: group this device should belong to
4868 void dev_set_group(struct net_device *dev, int new_group)
4870 dev->group = new_group;
4872 EXPORT_SYMBOL(dev_set_group);
4875 * dev_set_mac_address - Change Media Access Control Address
4879 * Change the hardware (MAC) address of the device
4881 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4883 const struct net_device_ops *ops = dev->netdev_ops;
4886 if (!ops->ndo_set_mac_address)
4888 if (sa->sa_family != dev->type)
4890 if (!netif_device_present(dev))
4892 err = ops->ndo_set_mac_address(dev, sa);
4894 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4897 EXPORT_SYMBOL(dev_set_mac_address);
4900 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4902 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4905 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4911 case SIOCGIFFLAGS: /* Get interface flags */
4912 ifr->ifr_flags = (short) dev_get_flags(dev);
4915 case SIOCGIFMETRIC: /* Get the metric on the interface
4916 (currently unused) */
4917 ifr->ifr_metric = 0;
4920 case SIOCGIFMTU: /* Get the MTU of a device */
4921 ifr->ifr_mtu = dev->mtu;
4926 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4928 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4929 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4930 ifr->ifr_hwaddr.sa_family = dev->type;
4938 ifr->ifr_map.mem_start = dev->mem_start;
4939 ifr->ifr_map.mem_end = dev->mem_end;
4940 ifr->ifr_map.base_addr = dev->base_addr;
4941 ifr->ifr_map.irq = dev->irq;
4942 ifr->ifr_map.dma = dev->dma;
4943 ifr->ifr_map.port = dev->if_port;
4947 ifr->ifr_ifindex = dev->ifindex;
4951 ifr->ifr_qlen = dev->tx_queue_len;
4955 /* dev_ioctl() should ensure this case
4967 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4969 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4972 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4973 const struct net_device_ops *ops;
4978 ops = dev->netdev_ops;
4981 case SIOCSIFFLAGS: /* Set interface flags */
4982 return dev_change_flags(dev, ifr->ifr_flags);
4984 case SIOCSIFMETRIC: /* Set the metric on the interface
4985 (currently unused) */
4988 case SIOCSIFMTU: /* Set the MTU of a device */
4989 return dev_set_mtu(dev, ifr->ifr_mtu);
4992 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4994 case SIOCSIFHWBROADCAST:
4995 if (ifr->ifr_hwaddr.sa_family != dev->type)
4997 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4998 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4999 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5003 if (ops->ndo_set_config) {
5004 if (!netif_device_present(dev))
5006 return ops->ndo_set_config(dev, &ifr->ifr_map);
5011 if (!ops->ndo_set_rx_mode ||
5012 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5014 if (!netif_device_present(dev))
5016 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5019 if (!ops->ndo_set_rx_mode ||
5020 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5022 if (!netif_device_present(dev))
5024 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5027 if (ifr->ifr_qlen < 0)
5029 dev->tx_queue_len = ifr->ifr_qlen;
5033 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5034 return dev_change_name(dev, ifr->ifr_newname);
5037 err = net_hwtstamp_validate(ifr);
5043 * Unknown or private ioctl
5046 if ((cmd >= SIOCDEVPRIVATE &&
5047 cmd <= SIOCDEVPRIVATE + 15) ||
5048 cmd == SIOCBONDENSLAVE ||
5049 cmd == SIOCBONDRELEASE ||
5050 cmd == SIOCBONDSETHWADDR ||
5051 cmd == SIOCBONDSLAVEINFOQUERY ||
5052 cmd == SIOCBONDINFOQUERY ||
5053 cmd == SIOCBONDCHANGEACTIVE ||
5054 cmd == SIOCGMIIPHY ||
5055 cmd == SIOCGMIIREG ||
5056 cmd == SIOCSMIIREG ||
5057 cmd == SIOCBRADDIF ||
5058 cmd == SIOCBRDELIF ||
5059 cmd == SIOCSHWTSTAMP ||
5060 cmd == SIOCWANDEV) {
5062 if (ops->ndo_do_ioctl) {
5063 if (netif_device_present(dev))
5064 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5076 * This function handles all "interface"-type I/O control requests. The actual
5077 * 'doing' part of this is dev_ifsioc above.
5081 * dev_ioctl - network device ioctl
5082 * @net: the applicable net namespace
5083 * @cmd: command to issue
5084 * @arg: pointer to a struct ifreq in user space
5086 * Issue ioctl functions to devices. This is normally called by the
5087 * user space syscall interfaces but can sometimes be useful for
5088 * other purposes. The return value is the return from the syscall if
5089 * positive or a negative errno code on error.
5092 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5098 /* One special case: SIOCGIFCONF takes ifconf argument
5099 and requires shared lock, because it sleeps writing
5103 if (cmd == SIOCGIFCONF) {
5105 ret = dev_ifconf(net, (char __user *) arg);
5109 if (cmd == SIOCGIFNAME)
5110 return dev_ifname(net, (struct ifreq __user *)arg);
5112 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5115 ifr.ifr_name[IFNAMSIZ-1] = 0;
5117 colon = strchr(ifr.ifr_name, ':');
5122 * See which interface the caller is talking about.
5127 * These ioctl calls:
5128 * - can be done by all.
5129 * - atomic and do not require locking.
5140 dev_load(net, ifr.ifr_name);
5142 ret = dev_ifsioc_locked(net, &ifr, cmd);
5147 if (copy_to_user(arg, &ifr,
5148 sizeof(struct ifreq)))
5154 dev_load(net, ifr.ifr_name);
5156 ret = dev_ethtool(net, &ifr);
5161 if (copy_to_user(arg, &ifr,
5162 sizeof(struct ifreq)))
5168 * These ioctl calls:
5169 * - require superuser power.
5170 * - require strict serialization.
5176 if (!capable(CAP_NET_ADMIN))
5178 dev_load(net, ifr.ifr_name);
5180 ret = dev_ifsioc(net, &ifr, cmd);
5185 if (copy_to_user(arg, &ifr,
5186 sizeof(struct ifreq)))
5192 * These ioctl calls:
5193 * - require superuser power.
5194 * - require strict serialization.
5195 * - do not return a value
5205 case SIOCSIFHWBROADCAST:
5208 case SIOCBONDENSLAVE:
5209 case SIOCBONDRELEASE:
5210 case SIOCBONDSETHWADDR:
5211 case SIOCBONDCHANGEACTIVE:
5215 if (!capable(CAP_NET_ADMIN))
5218 case SIOCBONDSLAVEINFOQUERY:
5219 case SIOCBONDINFOQUERY:
5220 dev_load(net, ifr.ifr_name);
5222 ret = dev_ifsioc(net, &ifr, cmd);
5227 /* Get the per device memory space. We can add this but
5228 * currently do not support it */
5230 /* Set the per device memory buffer space.
5231 * Not applicable in our case */
5236 * Unknown or private ioctl.
5239 if (cmd == SIOCWANDEV ||
5240 (cmd >= SIOCDEVPRIVATE &&
5241 cmd <= SIOCDEVPRIVATE + 15)) {
5242 dev_load(net, ifr.ifr_name);
5244 ret = dev_ifsioc(net, &ifr, cmd);
5246 if (!ret && copy_to_user(arg, &ifr,
5247 sizeof(struct ifreq)))
5251 /* Take care of Wireless Extensions */
5252 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5253 return wext_handle_ioctl(net, &ifr, cmd, arg);
5260 * dev_new_index - allocate an ifindex
5261 * @net: the applicable net namespace
5263 * Returns a suitable unique value for a new device interface
5264 * number. The caller must hold the rtnl semaphore or the
5265 * dev_base_lock to be sure it remains unique.
5267 static int dev_new_index(struct net *net)
5273 if (!__dev_get_by_index(net, ifindex))
5278 /* Delayed registration/unregisteration */
5279 static LIST_HEAD(net_todo_list);
5281 static void net_set_todo(struct net_device *dev)
5283 list_add_tail(&dev->todo_list, &net_todo_list);
5286 static void rollback_registered_many(struct list_head *head)
5288 struct net_device *dev, *tmp;
5290 BUG_ON(dev_boot_phase);
5293 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5294 /* Some devices call without registering
5295 * for initialization unwind. Remove those
5296 * devices and proceed with the remaining.
5298 if (dev->reg_state == NETREG_UNINITIALIZED) {
5299 pr_debug("unregister_netdevice: device %s/%p never "
5300 "was registered\n", dev->name, dev);
5303 list_del(&dev->unreg_list);
5306 dev->dismantle = true;
5307 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5310 /* If device is running, close it first. */
5311 dev_close_many(head);
5313 list_for_each_entry(dev, head, unreg_list) {
5314 /* And unlink it from device chain. */
5315 unlist_netdevice(dev);
5317 dev->reg_state = NETREG_UNREGISTERING;
5322 list_for_each_entry(dev, head, unreg_list) {
5323 /* Shutdown queueing discipline. */
5327 /* Notify protocols, that we are about to destroy
5328 this device. They should clean all the things.
5330 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5332 if (!dev->rtnl_link_ops ||
5333 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5334 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5337 * Flush the unicast and multicast chains
5342 if (dev->netdev_ops->ndo_uninit)
5343 dev->netdev_ops->ndo_uninit(dev);
5345 /* Notifier chain MUST detach us from master device. */
5346 WARN_ON(dev->master);
5348 /* Remove entries from kobject tree */
5349 netdev_unregister_kobject(dev);
5352 /* Process any work delayed until the end of the batch */
5353 dev = list_first_entry(head, struct net_device, unreg_list);
5354 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5358 list_for_each_entry(dev, head, unreg_list)
5362 static void rollback_registered(struct net_device *dev)
5366 list_add(&dev->unreg_list, &single);
5367 rollback_registered_many(&single);
5371 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5373 /* Fix illegal checksum combinations */
5374 if ((features & NETIF_F_HW_CSUM) &&
5375 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5376 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5377 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5380 if ((features & NETIF_F_NO_CSUM) &&
5381 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5382 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5383 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5386 /* Fix illegal SG+CSUM combinations. */
5387 if ((features & NETIF_F_SG) &&
5388 !(features & NETIF_F_ALL_CSUM)) {
5390 "Dropping NETIF_F_SG since no checksum feature.\n");
5391 features &= ~NETIF_F_SG;
5394 /* TSO requires that SG is present as well. */
5395 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5396 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5397 features &= ~NETIF_F_ALL_TSO;
5400 /* TSO ECN requires that TSO is present as well. */
5401 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5402 features &= ~NETIF_F_TSO_ECN;
5404 /* Software GSO depends on SG. */
5405 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5406 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5407 features &= ~NETIF_F_GSO;
5410 /* UFO needs SG and checksumming */
5411 if (features & NETIF_F_UFO) {
5412 /* maybe split UFO into V4 and V6? */
5413 if (!((features & NETIF_F_GEN_CSUM) ||
5414 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5415 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5417 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5418 features &= ~NETIF_F_UFO;
5421 if (!(features & NETIF_F_SG)) {
5423 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5424 features &= ~NETIF_F_UFO;
5431 int __netdev_update_features(struct net_device *dev)
5438 features = netdev_get_wanted_features(dev);
5440 if (dev->netdev_ops->ndo_fix_features)
5441 features = dev->netdev_ops->ndo_fix_features(dev, features);
5443 /* driver might be less strict about feature dependencies */
5444 features = netdev_fix_features(dev, features);
5446 if (dev->features == features)
5449 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5450 dev->features, features);
5452 if (dev->netdev_ops->ndo_set_features)
5453 err = dev->netdev_ops->ndo_set_features(dev, features);
5455 if (unlikely(err < 0)) {
5457 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5458 err, features, dev->features);
5463 dev->features = features;
5469 * netdev_update_features - recalculate device features
5470 * @dev: the device to check
5472 * Recalculate dev->features set and send notifications if it
5473 * has changed. Should be called after driver or hardware dependent
5474 * conditions might have changed that influence the features.
5476 void netdev_update_features(struct net_device *dev)
5478 if (__netdev_update_features(dev))
5479 netdev_features_change(dev);
5481 EXPORT_SYMBOL(netdev_update_features);
5484 * netdev_change_features - recalculate device features
5485 * @dev: the device to check
5487 * Recalculate dev->features set and send notifications even
5488 * if they have not changed. Should be called instead of
5489 * netdev_update_features() if also dev->vlan_features might
5490 * have changed to allow the changes to be propagated to stacked
5493 void netdev_change_features(struct net_device *dev)
5495 __netdev_update_features(dev);
5496 netdev_features_change(dev);
5498 EXPORT_SYMBOL(netdev_change_features);
5501 * netif_stacked_transfer_operstate - transfer operstate
5502 * @rootdev: the root or lower level device to transfer state from
5503 * @dev: the device to transfer operstate to
5505 * Transfer operational state from root to device. This is normally
5506 * called when a stacking relationship exists between the root
5507 * device and the device(a leaf device).
5509 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5510 struct net_device *dev)
5512 if (rootdev->operstate == IF_OPER_DORMANT)
5513 netif_dormant_on(dev);
5515 netif_dormant_off(dev);
5517 if (netif_carrier_ok(rootdev)) {
5518 if (!netif_carrier_ok(dev))
5519 netif_carrier_on(dev);
5521 if (netif_carrier_ok(dev))
5522 netif_carrier_off(dev);
5525 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5528 static int netif_alloc_rx_queues(struct net_device *dev)
5530 unsigned int i, count = dev->num_rx_queues;
5531 struct netdev_rx_queue *rx;
5535 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5537 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5542 for (i = 0; i < count; i++)
5548 static void netdev_init_one_queue(struct net_device *dev,
5549 struct netdev_queue *queue, void *_unused)
5551 /* Initialize queue lock */
5552 spin_lock_init(&queue->_xmit_lock);
5553 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5554 queue->xmit_lock_owner = -1;
5555 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5559 static int netif_alloc_netdev_queues(struct net_device *dev)
5561 unsigned int count = dev->num_tx_queues;
5562 struct netdev_queue *tx;
5566 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5568 pr_err("netdev: Unable to allocate %u tx queues.\n",
5574 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5575 spin_lock_init(&dev->tx_global_lock);
5581 * register_netdevice - register a network device
5582 * @dev: device to register
5584 * Take a completed network device structure and add it to the kernel
5585 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5586 * chain. 0 is returned on success. A negative errno code is returned
5587 * on a failure to set up the device, or if the name is a duplicate.
5589 * Callers must hold the rtnl semaphore. You may want
5590 * register_netdev() instead of this.
5593 * The locking appears insufficient to guarantee two parallel registers
5594 * will not get the same name.
5597 int register_netdevice(struct net_device *dev)
5600 struct net *net = dev_net(dev);
5602 BUG_ON(dev_boot_phase);
5607 /* When net_device's are persistent, this will be fatal. */
5608 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5611 spin_lock_init(&dev->addr_list_lock);
5612 netdev_set_addr_lockdep_class(dev);
5616 ret = dev_get_valid_name(dev, dev->name);
5620 /* Init, if this function is available */
5621 if (dev->netdev_ops->ndo_init) {
5622 ret = dev->netdev_ops->ndo_init(dev);
5630 dev->ifindex = dev_new_index(net);
5631 if (dev->iflink == -1)
5632 dev->iflink = dev->ifindex;
5634 /* Transfer changeable features to wanted_features and enable
5635 * software offloads (GSO and GRO).
5637 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5638 dev->features |= NETIF_F_SOFT_FEATURES;
5639 dev->wanted_features = dev->features & dev->hw_features;
5641 /* Turn on no cache copy if HW is doing checksum */
5642 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5643 if ((dev->features & NETIF_F_ALL_CSUM) &&
5644 !(dev->features & NETIF_F_NO_CSUM)) {
5645 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5646 dev->features |= NETIF_F_NOCACHE_COPY;
5649 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5651 dev->vlan_features |= NETIF_F_HIGHDMA;
5653 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5654 ret = notifier_to_errno(ret);
5658 ret = netdev_register_kobject(dev);
5661 dev->reg_state = NETREG_REGISTERED;
5663 __netdev_update_features(dev);
5666 * Default initial state at registry is that the
5667 * device is present.
5670 set_bit(__LINK_STATE_PRESENT, &dev->state);
5672 dev_init_scheduler(dev);
5674 list_netdevice(dev);
5676 /* Notify protocols, that a new device appeared. */
5677 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5678 ret = notifier_to_errno(ret);
5680 rollback_registered(dev);
5681 dev->reg_state = NETREG_UNREGISTERED;
5684 * Prevent userspace races by waiting until the network
5685 * device is fully setup before sending notifications.
5687 if (!dev->rtnl_link_ops ||
5688 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5689 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5695 if (dev->netdev_ops->ndo_uninit)
5696 dev->netdev_ops->ndo_uninit(dev);
5699 EXPORT_SYMBOL(register_netdevice);
5702 * init_dummy_netdev - init a dummy network device for NAPI
5703 * @dev: device to init
5705 * This takes a network device structure and initialize the minimum
5706 * amount of fields so it can be used to schedule NAPI polls without
5707 * registering a full blown interface. This is to be used by drivers
5708 * that need to tie several hardware interfaces to a single NAPI
5709 * poll scheduler due to HW limitations.
5711 int init_dummy_netdev(struct net_device *dev)
5713 /* Clear everything. Note we don't initialize spinlocks
5714 * are they aren't supposed to be taken by any of the
5715 * NAPI code and this dummy netdev is supposed to be
5716 * only ever used for NAPI polls
5718 memset(dev, 0, sizeof(struct net_device));
5720 /* make sure we BUG if trying to hit standard
5721 * register/unregister code path
5723 dev->reg_state = NETREG_DUMMY;
5725 /* NAPI wants this */
5726 INIT_LIST_HEAD(&dev->napi_list);
5728 /* a dummy interface is started by default */
5729 set_bit(__LINK_STATE_PRESENT, &dev->state);
5730 set_bit(__LINK_STATE_START, &dev->state);
5732 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5733 * because users of this 'device' dont need to change
5739 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5743 * register_netdev - register a network device
5744 * @dev: device to register
5746 * Take a completed network device structure and add it to the kernel
5747 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5748 * chain. 0 is returned on success. A negative errno code is returned
5749 * on a failure to set up the device, or if the name is a duplicate.
5751 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5752 * and expands the device name if you passed a format string to
5755 int register_netdev(struct net_device *dev)
5760 err = register_netdevice(dev);
5764 EXPORT_SYMBOL(register_netdev);
5766 int netdev_refcnt_read(const struct net_device *dev)
5770 for_each_possible_cpu(i)
5771 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5774 EXPORT_SYMBOL(netdev_refcnt_read);
5777 * netdev_wait_allrefs - wait until all references are gone.
5779 * This is called when unregistering network devices.
5781 * Any protocol or device that holds a reference should register
5782 * for netdevice notification, and cleanup and put back the
5783 * reference if they receive an UNREGISTER event.
5784 * We can get stuck here if buggy protocols don't correctly
5787 static void netdev_wait_allrefs(struct net_device *dev)
5789 unsigned long rebroadcast_time, warning_time;
5792 linkwatch_forget_dev(dev);
5794 rebroadcast_time = warning_time = jiffies;
5795 refcnt = netdev_refcnt_read(dev);
5797 while (refcnt != 0) {
5798 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5801 /* Rebroadcast unregister notification */
5802 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5803 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5804 * should have already handle it the first time */
5806 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5808 /* We must not have linkwatch events
5809 * pending on unregister. If this
5810 * happens, we simply run the queue
5811 * unscheduled, resulting in a noop
5814 linkwatch_run_queue();
5819 rebroadcast_time = jiffies;
5824 refcnt = netdev_refcnt_read(dev);
5826 if (time_after(jiffies, warning_time + 10 * HZ)) {
5827 printk(KERN_EMERG "unregister_netdevice: "
5828 "waiting for %s to become free. Usage "
5831 warning_time = jiffies;
5840 * register_netdevice(x1);
5841 * register_netdevice(x2);
5843 * unregister_netdevice(y1);
5844 * unregister_netdevice(y2);
5850 * We are invoked by rtnl_unlock().
5851 * This allows us to deal with problems:
5852 * 1) We can delete sysfs objects which invoke hotplug
5853 * without deadlocking with linkwatch via keventd.
5854 * 2) Since we run with the RTNL semaphore not held, we can sleep
5855 * safely in order to wait for the netdev refcnt to drop to zero.
5857 * We must not return until all unregister events added during
5858 * the interval the lock was held have been completed.
5860 void netdev_run_todo(void)
5862 struct list_head list;
5864 /* Snapshot list, allow later requests */
5865 list_replace_init(&net_todo_list, &list);
5869 /* Wait for rcu callbacks to finish before attempting to drain
5870 * the device list. This usually avoids a 250ms wait.
5872 if (!list_empty(&list))
5875 while (!list_empty(&list)) {
5876 struct net_device *dev
5877 = list_first_entry(&list, struct net_device, todo_list);
5878 list_del(&dev->todo_list);
5880 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5881 printk(KERN_ERR "network todo '%s' but state %d\n",
5882 dev->name, dev->reg_state);
5887 dev->reg_state = NETREG_UNREGISTERED;
5889 on_each_cpu(flush_backlog, dev, 1);
5891 netdev_wait_allrefs(dev);
5894 BUG_ON(netdev_refcnt_read(dev));
5895 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5896 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5897 WARN_ON(dev->dn_ptr);
5899 if (dev->destructor)
5900 dev->destructor(dev);
5902 /* Free network device */
5903 kobject_put(&dev->dev.kobj);
5907 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5908 * fields in the same order, with only the type differing.
5910 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5911 const struct net_device_stats *netdev_stats)
5913 #if BITS_PER_LONG == 64
5914 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5915 memcpy(stats64, netdev_stats, sizeof(*stats64));
5917 size_t i, n = sizeof(*stats64) / sizeof(u64);
5918 const unsigned long *src = (const unsigned long *)netdev_stats;
5919 u64 *dst = (u64 *)stats64;
5921 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5922 sizeof(*stats64) / sizeof(u64));
5923 for (i = 0; i < n; i++)
5929 * dev_get_stats - get network device statistics
5930 * @dev: device to get statistics from
5931 * @storage: place to store stats
5933 * Get network statistics from device. Return @storage.
5934 * The device driver may provide its own method by setting
5935 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5936 * otherwise the internal statistics structure is used.
5938 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5939 struct rtnl_link_stats64 *storage)
5941 const struct net_device_ops *ops = dev->netdev_ops;
5943 if (ops->ndo_get_stats64) {
5944 memset(storage, 0, sizeof(*storage));
5945 ops->ndo_get_stats64(dev, storage);
5946 } else if (ops->ndo_get_stats) {
5947 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5949 netdev_stats_to_stats64(storage, &dev->stats);
5951 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5954 EXPORT_SYMBOL(dev_get_stats);
5956 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5958 struct netdev_queue *queue = dev_ingress_queue(dev);
5960 #ifdef CONFIG_NET_CLS_ACT
5963 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5966 netdev_init_one_queue(dev, queue, NULL);
5967 queue->qdisc = &noop_qdisc;
5968 queue->qdisc_sleeping = &noop_qdisc;
5969 rcu_assign_pointer(dev->ingress_queue, queue);
5975 * alloc_netdev_mqs - allocate network device
5976 * @sizeof_priv: size of private data to allocate space for
5977 * @name: device name format string
5978 * @setup: callback to initialize device
5979 * @txqs: the number of TX subqueues to allocate
5980 * @rxqs: the number of RX subqueues to allocate
5982 * Allocates a struct net_device with private data area for driver use
5983 * and performs basic initialization. Also allocates subquue structs
5984 * for each queue on the device.
5986 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5987 void (*setup)(struct net_device *),
5988 unsigned int txqs, unsigned int rxqs)
5990 struct net_device *dev;
5992 struct net_device *p;
5994 BUG_ON(strlen(name) >= sizeof(dev->name));
5997 pr_err("alloc_netdev: Unable to allocate device "
5998 "with zero queues.\n");
6004 pr_err("alloc_netdev: Unable to allocate device "
6005 "with zero RX queues.\n");
6010 alloc_size = sizeof(struct net_device);
6012 /* ensure 32-byte alignment of private area */
6013 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6014 alloc_size += sizeof_priv;
6016 /* ensure 32-byte alignment of whole construct */
6017 alloc_size += NETDEV_ALIGN - 1;
6019 p = kzalloc(alloc_size, GFP_KERNEL);
6021 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
6025 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6026 dev->padded = (char *)dev - (char *)p;
6028 dev->pcpu_refcnt = alloc_percpu(int);
6029 if (!dev->pcpu_refcnt)
6032 if (dev_addr_init(dev))
6038 dev_net_set(dev, &init_net);
6040 dev->gso_max_size = GSO_MAX_SIZE;
6042 INIT_LIST_HEAD(&dev->napi_list);
6043 INIT_LIST_HEAD(&dev->unreg_list);
6044 INIT_LIST_HEAD(&dev->link_watch_list);
6045 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6048 dev->num_tx_queues = txqs;
6049 dev->real_num_tx_queues = txqs;
6050 if (netif_alloc_netdev_queues(dev))
6054 dev->num_rx_queues = rxqs;
6055 dev->real_num_rx_queues = rxqs;
6056 if (netif_alloc_rx_queues(dev))
6060 strcpy(dev->name, name);
6061 dev->group = INIT_NETDEV_GROUP;
6069 free_percpu(dev->pcpu_refcnt);
6079 EXPORT_SYMBOL(alloc_netdev_mqs);
6082 * free_netdev - free network device
6085 * This function does the last stage of destroying an allocated device
6086 * interface. The reference to the device object is released.
6087 * If this is the last reference then it will be freed.
6089 void free_netdev(struct net_device *dev)
6091 struct napi_struct *p, *n;
6093 release_net(dev_net(dev));
6100 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6102 /* Flush device addresses */
6103 dev_addr_flush(dev);
6105 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6108 free_percpu(dev->pcpu_refcnt);
6109 dev->pcpu_refcnt = NULL;
6111 /* Compatibility with error handling in drivers */
6112 if (dev->reg_state == NETREG_UNINITIALIZED) {
6113 kfree((char *)dev - dev->padded);
6117 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6118 dev->reg_state = NETREG_RELEASED;
6120 /* will free via device release */
6121 put_device(&dev->dev);
6123 EXPORT_SYMBOL(free_netdev);
6126 * synchronize_net - Synchronize with packet receive processing
6128 * Wait for packets currently being received to be done.
6129 * Does not block later packets from starting.
6131 void synchronize_net(void)
6134 if (rtnl_is_locked())
6135 synchronize_rcu_expedited();
6139 EXPORT_SYMBOL(synchronize_net);
6142 * unregister_netdevice_queue - remove device from the kernel
6146 * This function shuts down a device interface and removes it
6147 * from the kernel tables.
6148 * If head not NULL, device is queued to be unregistered later.
6150 * Callers must hold the rtnl semaphore. You may want
6151 * unregister_netdev() instead of this.
6154 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6159 list_move_tail(&dev->unreg_list, head);
6161 rollback_registered(dev);
6162 /* Finish processing unregister after unlock */
6166 EXPORT_SYMBOL(unregister_netdevice_queue);
6169 * unregister_netdevice_many - unregister many devices
6170 * @head: list of devices
6172 void unregister_netdevice_many(struct list_head *head)
6174 struct net_device *dev;
6176 if (!list_empty(head)) {
6177 rollback_registered_many(head);
6178 list_for_each_entry(dev, head, unreg_list)
6182 EXPORT_SYMBOL(unregister_netdevice_many);
6185 * unregister_netdev - remove device from the kernel
6188 * This function shuts down a device interface and removes it
6189 * from the kernel tables.
6191 * This is just a wrapper for unregister_netdevice that takes
6192 * the rtnl semaphore. In general you want to use this and not
6193 * unregister_netdevice.
6195 void unregister_netdev(struct net_device *dev)
6198 unregister_netdevice(dev);
6201 EXPORT_SYMBOL(unregister_netdev);
6204 * dev_change_net_namespace - move device to different nethost namespace
6206 * @net: network namespace
6207 * @pat: If not NULL name pattern to try if the current device name
6208 * is already taken in the destination network namespace.
6210 * This function shuts down a device interface and moves it
6211 * to a new network namespace. On success 0 is returned, on
6212 * a failure a netagive errno code is returned.
6214 * Callers must hold the rtnl semaphore.
6217 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6223 /* Don't allow namespace local devices to be moved. */
6225 if (dev->features & NETIF_F_NETNS_LOCAL)
6228 /* Ensure the device has been registrered */
6230 if (dev->reg_state != NETREG_REGISTERED)
6233 /* Get out if there is nothing todo */
6235 if (net_eq(dev_net(dev), net))
6238 /* Pick the destination device name, and ensure
6239 * we can use it in the destination network namespace.
6242 if (__dev_get_by_name(net, dev->name)) {
6243 /* We get here if we can't use the current device name */
6246 if (dev_get_valid_name(dev, pat) < 0)
6251 * And now a mini version of register_netdevice unregister_netdevice.
6254 /* If device is running close it first. */
6257 /* And unlink it from device chain */
6259 unlist_netdevice(dev);
6263 /* Shutdown queueing discipline. */
6266 /* Notify protocols, that we are about to destroy
6267 this device. They should clean all the things.
6269 Note that dev->reg_state stays at NETREG_REGISTERED.
6270 This is wanted because this way 8021q and macvlan know
6271 the device is just moving and can keep their slaves up.
6273 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6274 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6275 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6278 * Flush the unicast and multicast chains
6283 /* Actually switch the network namespace */
6284 dev_net_set(dev, net);
6286 /* If there is an ifindex conflict assign a new one */
6287 if (__dev_get_by_index(net, dev->ifindex)) {
6288 int iflink = (dev->iflink == dev->ifindex);
6289 dev->ifindex = dev_new_index(net);
6291 dev->iflink = dev->ifindex;
6294 /* Fixup kobjects */
6295 err = device_rename(&dev->dev, dev->name);
6298 /* Add the device back in the hashes */
6299 list_netdevice(dev);
6301 /* Notify protocols, that a new device appeared. */
6302 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6305 * Prevent userspace races by waiting until the network
6306 * device is fully setup before sending notifications.
6308 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6315 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6317 static int dev_cpu_callback(struct notifier_block *nfb,
6318 unsigned long action,
6321 struct sk_buff **list_skb;
6322 struct sk_buff *skb;
6323 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6324 struct softnet_data *sd, *oldsd;
6326 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6329 local_irq_disable();
6330 cpu = smp_processor_id();
6331 sd = &per_cpu(softnet_data, cpu);
6332 oldsd = &per_cpu(softnet_data, oldcpu);
6334 /* Find end of our completion_queue. */
6335 list_skb = &sd->completion_queue;
6337 list_skb = &(*list_skb)->next;
6338 /* Append completion queue from offline CPU. */
6339 *list_skb = oldsd->completion_queue;
6340 oldsd->completion_queue = NULL;
6342 /* Append output queue from offline CPU. */
6343 if (oldsd->output_queue) {
6344 *sd->output_queue_tailp = oldsd->output_queue;
6345 sd->output_queue_tailp = oldsd->output_queue_tailp;
6346 oldsd->output_queue = NULL;
6347 oldsd->output_queue_tailp = &oldsd->output_queue;
6349 /* Append NAPI poll list from offline CPU. */
6350 if (!list_empty(&oldsd->poll_list)) {
6351 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6352 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6355 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6358 /* Process offline CPU's input_pkt_queue */
6359 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6361 input_queue_head_incr(oldsd);
6363 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6365 input_queue_head_incr(oldsd);
6373 * netdev_increment_features - increment feature set by one
6374 * @all: current feature set
6375 * @one: new feature set
6376 * @mask: mask feature set
6378 * Computes a new feature set after adding a device with feature set
6379 * @one to the master device with current feature set @all. Will not
6380 * enable anything that is off in @mask. Returns the new feature set.
6382 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6384 if (mask & NETIF_F_GEN_CSUM)
6385 mask |= NETIF_F_ALL_CSUM;
6386 mask |= NETIF_F_VLAN_CHALLENGED;
6388 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6389 all &= one | ~NETIF_F_ALL_FOR_ALL;
6391 /* If device needs checksumming, downgrade to it. */
6392 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6393 all &= ~NETIF_F_NO_CSUM;
6395 /* If one device supports hw checksumming, set for all. */
6396 if (all & NETIF_F_GEN_CSUM)
6397 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6401 EXPORT_SYMBOL(netdev_increment_features);
6403 static struct hlist_head *netdev_create_hash(void)
6406 struct hlist_head *hash;
6408 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6410 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6411 INIT_HLIST_HEAD(&hash[i]);
6416 /* Initialize per network namespace state */
6417 static int __net_init netdev_init(struct net *net)
6419 INIT_LIST_HEAD(&net->dev_base_head);
6421 net->dev_name_head = netdev_create_hash();
6422 if (net->dev_name_head == NULL)
6425 net->dev_index_head = netdev_create_hash();
6426 if (net->dev_index_head == NULL)
6432 kfree(net->dev_name_head);
6438 * netdev_drivername - network driver for the device
6439 * @dev: network device
6441 * Determine network driver for device.
6443 const char *netdev_drivername(const struct net_device *dev)
6445 const struct device_driver *driver;
6446 const struct device *parent;
6447 const char *empty = "";
6449 parent = dev->dev.parent;
6453 driver = parent->driver;
6454 if (driver && driver->name)
6455 return driver->name;
6459 int __netdev_printk(const char *level, const struct net_device *dev,
6460 struct va_format *vaf)
6464 if (dev && dev->dev.parent)
6465 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6466 netdev_name(dev), vaf);
6468 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6470 r = printk("%s(NULL net_device): %pV", level, vaf);
6474 EXPORT_SYMBOL(__netdev_printk);
6476 int netdev_printk(const char *level, const struct net_device *dev,
6477 const char *format, ...)
6479 struct va_format vaf;
6483 va_start(args, format);
6488 r = __netdev_printk(level, dev, &vaf);
6493 EXPORT_SYMBOL(netdev_printk);
6495 #define define_netdev_printk_level(func, level) \
6496 int func(const struct net_device *dev, const char *fmt, ...) \
6499 struct va_format vaf; \
6502 va_start(args, fmt); \
6507 r = __netdev_printk(level, dev, &vaf); \
6512 EXPORT_SYMBOL(func);
6514 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6515 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6516 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6517 define_netdev_printk_level(netdev_err, KERN_ERR);
6518 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6519 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6520 define_netdev_printk_level(netdev_info, KERN_INFO);
6522 static void __net_exit netdev_exit(struct net *net)
6524 kfree(net->dev_name_head);
6525 kfree(net->dev_index_head);
6528 static struct pernet_operations __net_initdata netdev_net_ops = {
6529 .init = netdev_init,
6530 .exit = netdev_exit,
6533 static void __net_exit default_device_exit(struct net *net)
6535 struct net_device *dev, *aux;
6537 * Push all migratable network devices back to the
6538 * initial network namespace
6541 for_each_netdev_safe(net, dev, aux) {
6543 char fb_name[IFNAMSIZ];
6545 /* Ignore unmoveable devices (i.e. loopback) */
6546 if (dev->features & NETIF_F_NETNS_LOCAL)
6549 /* Leave virtual devices for the generic cleanup */
6550 if (dev->rtnl_link_ops)
6553 /* Push remaining network devices to init_net */
6554 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6555 err = dev_change_net_namespace(dev, &init_net, fb_name);
6557 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6558 __func__, dev->name, err);
6565 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6567 /* At exit all network devices most be removed from a network
6568 * namespace. Do this in the reverse order of registration.
6569 * Do this across as many network namespaces as possible to
6570 * improve batching efficiency.
6572 struct net_device *dev;
6574 LIST_HEAD(dev_kill_list);
6577 list_for_each_entry(net, net_list, exit_list) {
6578 for_each_netdev_reverse(net, dev) {
6579 if (dev->rtnl_link_ops)
6580 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6582 unregister_netdevice_queue(dev, &dev_kill_list);
6585 unregister_netdevice_many(&dev_kill_list);
6586 list_del(&dev_kill_list);
6590 static struct pernet_operations __net_initdata default_device_ops = {
6591 .exit = default_device_exit,
6592 .exit_batch = default_device_exit_batch,
6596 * Initialize the DEV module. At boot time this walks the device list and
6597 * unhooks any devices that fail to initialise (normally hardware not
6598 * present) and leaves us with a valid list of present and active devices.
6603 * This is called single threaded during boot, so no need
6604 * to take the rtnl semaphore.
6606 static int __init net_dev_init(void)
6608 int i, rc = -ENOMEM;
6610 BUG_ON(!dev_boot_phase);
6612 if (dev_proc_init())
6615 if (netdev_kobject_init())
6618 INIT_LIST_HEAD(&ptype_all);
6619 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6620 INIT_LIST_HEAD(&ptype_base[i]);
6622 if (register_pernet_subsys(&netdev_net_ops))
6626 * Initialise the packet receive queues.
6629 for_each_possible_cpu(i) {
6630 struct softnet_data *sd = &per_cpu(softnet_data, i);
6632 memset(sd, 0, sizeof(*sd));
6633 skb_queue_head_init(&sd->input_pkt_queue);
6634 skb_queue_head_init(&sd->process_queue);
6635 sd->completion_queue = NULL;
6636 INIT_LIST_HEAD(&sd->poll_list);
6637 sd->output_queue = NULL;
6638 sd->output_queue_tailp = &sd->output_queue;
6640 sd->csd.func = rps_trigger_softirq;
6646 sd->backlog.poll = process_backlog;
6647 sd->backlog.weight = weight_p;
6648 sd->backlog.gro_list = NULL;
6649 sd->backlog.gro_count = 0;
6654 /* The loopback device is special if any other network devices
6655 * is present in a network namespace the loopback device must
6656 * be present. Since we now dynamically allocate and free the
6657 * loopback device ensure this invariant is maintained by
6658 * keeping the loopback device as the first device on the
6659 * list of network devices. Ensuring the loopback devices
6660 * is the first device that appears and the last network device
6663 if (register_pernet_device(&loopback_net_ops))
6666 if (register_pernet_device(&default_device_ops))
6669 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6670 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6672 hotcpu_notifier(dev_cpu_callback, 0);
6680 subsys_initcall(net_dev_init);
6682 static int __init initialize_hashrnd(void)
6684 get_random_bytes(&hashrnd, sizeof(hashrnd));
6688 late_initcall_sync(initialize_hashrnd);