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);
1411 raw_notifier_chain_unregister(&netdev_chain, nb);
1414 EXPORT_SYMBOL(register_netdevice_notifier);
1417 * unregister_netdevice_notifier - unregister a network notifier block
1420 * Unregister a notifier previously registered by
1421 * register_netdevice_notifier(). The notifier is unlinked into the
1422 * kernel structures and may then be reused. A negative errno code
1423 * is returned on a failure.
1425 * After unregistering unregister and down device events are synthesized
1426 * for all devices on the device list to the removed notifier to remove
1427 * the need for special case cleanup code.
1430 int unregister_netdevice_notifier(struct notifier_block *nb)
1432 struct net_device *dev;
1437 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1442 for_each_netdev(net, dev) {
1443 if (dev->flags & IFF_UP) {
1444 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1445 nb->notifier_call(nb, NETDEV_DOWN, dev);
1447 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1448 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1455 EXPORT_SYMBOL(unregister_netdevice_notifier);
1458 * call_netdevice_notifiers - call all network notifier blocks
1459 * @val: value passed unmodified to notifier function
1460 * @dev: net_device pointer passed unmodified to notifier function
1462 * Call all network notifier blocks. Parameters and return value
1463 * are as for raw_notifier_call_chain().
1466 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1469 return raw_notifier_call_chain(&netdev_chain, val, dev);
1471 EXPORT_SYMBOL(call_netdevice_notifiers);
1473 /* When > 0 there are consumers of rx skb time stamps */
1474 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1476 void net_enable_timestamp(void)
1478 atomic_inc(&netstamp_needed);
1480 EXPORT_SYMBOL(net_enable_timestamp);
1482 void net_disable_timestamp(void)
1484 atomic_dec(&netstamp_needed);
1486 EXPORT_SYMBOL(net_disable_timestamp);
1488 static inline void net_timestamp_set(struct sk_buff *skb)
1490 if (atomic_read(&netstamp_needed))
1491 __net_timestamp(skb);
1493 skb->tstamp.tv64 = 0;
1496 static inline void net_timestamp_check(struct sk_buff *skb)
1498 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1499 __net_timestamp(skb);
1502 static int net_hwtstamp_validate(struct ifreq *ifr)
1504 struct hwtstamp_config cfg;
1505 enum hwtstamp_tx_types tx_type;
1506 enum hwtstamp_rx_filters rx_filter;
1507 int tx_type_valid = 0;
1508 int rx_filter_valid = 0;
1510 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1513 if (cfg.flags) /* reserved for future extensions */
1516 tx_type = cfg.tx_type;
1517 rx_filter = cfg.rx_filter;
1520 case HWTSTAMP_TX_OFF:
1521 case HWTSTAMP_TX_ON:
1522 case HWTSTAMP_TX_ONESTEP_SYNC:
1527 switch (rx_filter) {
1528 case HWTSTAMP_FILTER_NONE:
1529 case HWTSTAMP_FILTER_ALL:
1530 case HWTSTAMP_FILTER_SOME:
1531 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1532 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1533 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1534 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1535 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1536 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1537 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1538 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1539 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1540 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1541 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1542 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1543 rx_filter_valid = 1;
1547 if (!tx_type_valid || !rx_filter_valid)
1553 static inline bool is_skb_forwardable(struct net_device *dev,
1554 struct sk_buff *skb)
1558 if (!(dev->flags & IFF_UP))
1561 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1562 if (skb->len <= len)
1565 /* if TSO is enabled, we don't care about the length as the packet
1566 * could be forwarded without being segmented before
1568 if (skb_is_gso(skb))
1575 * dev_forward_skb - loopback an skb to another netif
1577 * @dev: destination network device
1578 * @skb: buffer to forward
1581 * NET_RX_SUCCESS (no congestion)
1582 * NET_RX_DROP (packet was dropped, but freed)
1584 * dev_forward_skb can be used for injecting an skb from the
1585 * start_xmit function of one device into the receive queue
1586 * of another device.
1588 * The receiving device may be in another namespace, so
1589 * we have to clear all information in the skb that could
1590 * impact namespace isolation.
1592 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1594 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1595 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1596 atomic_long_inc(&dev->rx_dropped);
1605 if (unlikely(!is_skb_forwardable(dev, skb))) {
1606 atomic_long_inc(&dev->rx_dropped);
1612 skb->tstamp.tv64 = 0;
1613 skb->pkt_type = PACKET_HOST;
1614 skb->protocol = eth_type_trans(skb, dev);
1618 return netif_rx(skb);
1620 EXPORT_SYMBOL_GPL(dev_forward_skb);
1622 static inline int deliver_skb(struct sk_buff *skb,
1623 struct packet_type *pt_prev,
1624 struct net_device *orig_dev)
1626 atomic_inc(&skb->users);
1627 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1631 * Support routine. Sends outgoing frames to any network
1632 * taps currently in use.
1635 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1637 struct packet_type *ptype;
1638 struct sk_buff *skb2 = NULL;
1639 struct packet_type *pt_prev = NULL;
1642 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1643 /* Never send packets back to the socket
1644 * they originated from - MvS (miquels@drinkel.ow.org)
1646 if ((ptype->dev == dev || !ptype->dev) &&
1647 (ptype->af_packet_priv == NULL ||
1648 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1650 deliver_skb(skb2, pt_prev, skb->dev);
1655 skb2 = skb_clone(skb, GFP_ATOMIC);
1659 net_timestamp_set(skb2);
1661 /* skb->nh should be correctly
1662 set by sender, so that the second statement is
1663 just protection against buggy protocols.
1665 skb_reset_mac_header(skb2);
1667 if (skb_network_header(skb2) < skb2->data ||
1668 skb2->network_header > skb2->tail) {
1669 if (net_ratelimit())
1670 printk(KERN_CRIT "protocol %04x is "
1672 ntohs(skb2->protocol),
1674 skb_reset_network_header(skb2);
1677 skb2->transport_header = skb2->network_header;
1678 skb2->pkt_type = PACKET_OUTGOING;
1683 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1687 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1688 * @dev: Network device
1689 * @txq: number of queues available
1691 * If real_num_tx_queues is changed the tc mappings may no longer be
1692 * valid. To resolve this verify the tc mapping remains valid and if
1693 * not NULL the mapping. With no priorities mapping to this
1694 * offset/count pair it will no longer be used. In the worst case TC0
1695 * is invalid nothing can be done so disable priority mappings. If is
1696 * expected that drivers will fix this mapping if they can before
1697 * calling netif_set_real_num_tx_queues.
1699 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1702 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1704 /* If TC0 is invalidated disable TC mapping */
1705 if (tc->offset + tc->count > txq) {
1706 pr_warning("Number of in use tx queues changed "
1707 "invalidating tc mappings. Priority "
1708 "traffic classification disabled!\n");
1713 /* Invalidated prio to tc mappings set to TC0 */
1714 for (i = 1; i < TC_BITMASK + 1; i++) {
1715 int q = netdev_get_prio_tc_map(dev, i);
1717 tc = &dev->tc_to_txq[q];
1718 if (tc->offset + tc->count > txq) {
1719 pr_warning("Number of in use tx queues "
1720 "changed. Priority %i to tc "
1721 "mapping %i is no longer valid "
1722 "setting map to 0\n",
1724 netdev_set_prio_tc_map(dev, i, 0);
1730 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1731 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1733 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1737 if (txq < 1 || txq > dev->num_tx_queues)
1740 if (dev->reg_state == NETREG_REGISTERED ||
1741 dev->reg_state == NETREG_UNREGISTERING) {
1744 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1750 netif_setup_tc(dev, txq);
1752 if (txq < dev->real_num_tx_queues)
1753 qdisc_reset_all_tx_gt(dev, txq);
1756 dev->real_num_tx_queues = txq;
1759 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1763 * netif_set_real_num_rx_queues - set actual number of RX queues used
1764 * @dev: Network device
1765 * @rxq: Actual number of RX queues
1767 * This must be called either with the rtnl_lock held or before
1768 * registration of the net device. Returns 0 on success, or a
1769 * negative error code. If called before registration, it always
1772 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1776 if (rxq < 1 || rxq > dev->num_rx_queues)
1779 if (dev->reg_state == NETREG_REGISTERED) {
1782 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1788 dev->real_num_rx_queues = rxq;
1791 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1794 static inline void __netif_reschedule(struct Qdisc *q)
1796 struct softnet_data *sd;
1797 unsigned long flags;
1799 local_irq_save(flags);
1800 sd = &__get_cpu_var(softnet_data);
1801 q->next_sched = NULL;
1802 *sd->output_queue_tailp = q;
1803 sd->output_queue_tailp = &q->next_sched;
1804 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1805 local_irq_restore(flags);
1808 void __netif_schedule(struct Qdisc *q)
1810 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1811 __netif_reschedule(q);
1813 EXPORT_SYMBOL(__netif_schedule);
1815 void dev_kfree_skb_irq(struct sk_buff *skb)
1817 if (atomic_dec_and_test(&skb->users)) {
1818 struct softnet_data *sd;
1819 unsigned long flags;
1821 local_irq_save(flags);
1822 sd = &__get_cpu_var(softnet_data);
1823 skb->next = sd->completion_queue;
1824 sd->completion_queue = skb;
1825 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1826 local_irq_restore(flags);
1829 EXPORT_SYMBOL(dev_kfree_skb_irq);
1831 void dev_kfree_skb_any(struct sk_buff *skb)
1833 if (in_irq() || irqs_disabled())
1834 dev_kfree_skb_irq(skb);
1838 EXPORT_SYMBOL(dev_kfree_skb_any);
1842 * netif_device_detach - mark device as removed
1843 * @dev: network device
1845 * Mark device as removed from system and therefore no longer available.
1847 void netif_device_detach(struct net_device *dev)
1849 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1850 netif_running(dev)) {
1851 netif_tx_stop_all_queues(dev);
1854 EXPORT_SYMBOL(netif_device_detach);
1857 * netif_device_attach - mark device as attached
1858 * @dev: network device
1860 * Mark device as attached from system and restart if needed.
1862 void netif_device_attach(struct net_device *dev)
1864 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1865 netif_running(dev)) {
1866 netif_tx_wake_all_queues(dev);
1867 __netdev_watchdog_up(dev);
1870 EXPORT_SYMBOL(netif_device_attach);
1873 * Invalidate hardware checksum when packet is to be mangled, and
1874 * complete checksum manually on outgoing path.
1876 int skb_checksum_help(struct sk_buff *skb)
1879 int ret = 0, offset;
1881 if (skb->ip_summed == CHECKSUM_COMPLETE)
1882 goto out_set_summed;
1884 if (unlikely(skb_shinfo(skb)->gso_size)) {
1885 /* Let GSO fix up the checksum. */
1886 goto out_set_summed;
1889 offset = skb_checksum_start_offset(skb);
1890 BUG_ON(offset >= skb_headlen(skb));
1891 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1893 offset += skb->csum_offset;
1894 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1896 if (skb_cloned(skb) &&
1897 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1898 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1903 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1905 skb->ip_summed = CHECKSUM_NONE;
1909 EXPORT_SYMBOL(skb_checksum_help);
1912 * skb_gso_segment - Perform segmentation on skb.
1913 * @skb: buffer to segment
1914 * @features: features for the output path (see dev->features)
1916 * This function segments the given skb and returns a list of segments.
1918 * It may return NULL if the skb requires no segmentation. This is
1919 * only possible when GSO is used for verifying header integrity.
1921 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1923 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1924 struct packet_type *ptype;
1925 __be16 type = skb->protocol;
1926 int vlan_depth = ETH_HLEN;
1929 while (type == htons(ETH_P_8021Q)) {
1930 struct vlan_hdr *vh;
1932 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1933 return ERR_PTR(-EINVAL);
1935 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1936 type = vh->h_vlan_encapsulated_proto;
1937 vlan_depth += VLAN_HLEN;
1940 skb_reset_mac_header(skb);
1941 skb->mac_len = skb->network_header - skb->mac_header;
1942 __skb_pull(skb, skb->mac_len);
1944 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1945 struct net_device *dev = skb->dev;
1946 struct ethtool_drvinfo info = {};
1948 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1949 dev->ethtool_ops->get_drvinfo(dev, &info);
1951 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1952 info.driver, dev ? dev->features : 0L,
1953 skb->sk ? skb->sk->sk_route_caps : 0L,
1954 skb->len, skb->data_len, skb->ip_summed);
1956 if (skb_header_cloned(skb) &&
1957 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1958 return ERR_PTR(err);
1962 list_for_each_entry_rcu(ptype,
1963 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1964 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1965 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1966 err = ptype->gso_send_check(skb);
1967 segs = ERR_PTR(err);
1968 if (err || skb_gso_ok(skb, features))
1970 __skb_push(skb, (skb->data -
1971 skb_network_header(skb)));
1973 segs = ptype->gso_segment(skb, features);
1979 __skb_push(skb, skb->data - skb_mac_header(skb));
1983 EXPORT_SYMBOL(skb_gso_segment);
1985 /* Take action when hardware reception checksum errors are detected. */
1987 void netdev_rx_csum_fault(struct net_device *dev)
1989 if (net_ratelimit()) {
1990 printk(KERN_ERR "%s: hw csum failure.\n",
1991 dev ? dev->name : "<unknown>");
1995 EXPORT_SYMBOL(netdev_rx_csum_fault);
1998 /* Actually, we should eliminate this check as soon as we know, that:
1999 * 1. IOMMU is present and allows to map all the memory.
2000 * 2. No high memory really exists on this machine.
2003 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2005 #ifdef CONFIG_HIGHMEM
2007 if (!(dev->features & NETIF_F_HIGHDMA)) {
2008 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2009 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2010 if (PageHighMem(skb_frag_page(frag)))
2015 if (PCI_DMA_BUS_IS_PHYS) {
2016 struct device *pdev = dev->dev.parent;
2020 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2021 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2022 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2023 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2032 void (*destructor)(struct sk_buff *skb);
2035 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2037 static void dev_gso_skb_destructor(struct sk_buff *skb)
2039 struct dev_gso_cb *cb;
2042 struct sk_buff *nskb = skb->next;
2044 skb->next = nskb->next;
2047 } while (skb->next);
2049 cb = DEV_GSO_CB(skb);
2051 cb->destructor(skb);
2055 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2056 * @skb: buffer to segment
2057 * @features: device features as applicable to this skb
2059 * This function segments the given skb and stores the list of segments
2062 static int dev_gso_segment(struct sk_buff *skb, int features)
2064 struct sk_buff *segs;
2066 segs = skb_gso_segment(skb, features);
2068 /* Verifying header integrity only. */
2073 return PTR_ERR(segs);
2076 DEV_GSO_CB(skb)->destructor = skb->destructor;
2077 skb->destructor = dev_gso_skb_destructor;
2082 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2084 return ((features & NETIF_F_GEN_CSUM) ||
2085 ((features & NETIF_F_V4_CSUM) &&
2086 protocol == htons(ETH_P_IP)) ||
2087 ((features & NETIF_F_V6_CSUM) &&
2088 protocol == htons(ETH_P_IPV6)) ||
2089 ((features & NETIF_F_FCOE_CRC) &&
2090 protocol == htons(ETH_P_FCOE)));
2093 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2095 if (!can_checksum_protocol(features, protocol)) {
2096 features &= ~NETIF_F_ALL_CSUM;
2097 features &= ~NETIF_F_SG;
2098 } else if (illegal_highdma(skb->dev, skb)) {
2099 features &= ~NETIF_F_SG;
2105 u32 netif_skb_features(struct sk_buff *skb)
2107 __be16 protocol = skb->protocol;
2108 u32 features = skb->dev->features;
2110 if (protocol == htons(ETH_P_8021Q)) {
2111 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2112 protocol = veh->h_vlan_encapsulated_proto;
2113 } else if (!vlan_tx_tag_present(skb)) {
2114 return harmonize_features(skb, protocol, features);
2117 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2119 if (protocol != htons(ETH_P_8021Q)) {
2120 return harmonize_features(skb, protocol, features);
2122 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2123 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2124 return harmonize_features(skb, protocol, features);
2127 EXPORT_SYMBOL(netif_skb_features);
2130 * Returns true if either:
2131 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2132 * 2. skb is fragmented and the device does not support SG, or if
2133 * at least one of fragments is in highmem and device does not
2134 * support DMA from it.
2136 static inline int skb_needs_linearize(struct sk_buff *skb,
2139 return skb_is_nonlinear(skb) &&
2140 ((skb_has_frag_list(skb) &&
2141 !(features & NETIF_F_FRAGLIST)) ||
2142 (skb_shinfo(skb)->nr_frags &&
2143 !(features & NETIF_F_SG)));
2146 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2147 struct netdev_queue *txq)
2149 const struct net_device_ops *ops = dev->netdev_ops;
2150 int rc = NETDEV_TX_OK;
2151 unsigned int skb_len;
2153 if (likely(!skb->next)) {
2157 * If device doesn't need skb->dst, release it right now while
2158 * its hot in this cpu cache
2160 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2163 if (!list_empty(&ptype_all))
2164 dev_queue_xmit_nit(skb, dev);
2166 features = netif_skb_features(skb);
2168 if (vlan_tx_tag_present(skb) &&
2169 !(features & NETIF_F_HW_VLAN_TX)) {
2170 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2177 if (netif_needs_gso(skb, features)) {
2178 if (unlikely(dev_gso_segment(skb, features)))
2183 if (skb_needs_linearize(skb, features) &&
2184 __skb_linearize(skb))
2187 /* If packet is not checksummed and device does not
2188 * support checksumming for this protocol, complete
2189 * checksumming here.
2191 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2192 skb_set_transport_header(skb,
2193 skb_checksum_start_offset(skb));
2194 if (!(features & NETIF_F_ALL_CSUM) &&
2195 skb_checksum_help(skb))
2201 rc = ops->ndo_start_xmit(skb, dev);
2202 trace_net_dev_xmit(skb, rc, dev, skb_len);
2203 if (rc == NETDEV_TX_OK)
2204 txq_trans_update(txq);
2210 struct sk_buff *nskb = skb->next;
2212 skb->next = nskb->next;
2216 * If device doesn't need nskb->dst, release it right now while
2217 * its hot in this cpu cache
2219 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2222 skb_len = nskb->len;
2223 rc = ops->ndo_start_xmit(nskb, dev);
2224 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2225 if (unlikely(rc != NETDEV_TX_OK)) {
2226 if (rc & ~NETDEV_TX_MASK)
2227 goto out_kfree_gso_skb;
2228 nskb->next = skb->next;
2232 txq_trans_update(txq);
2233 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2234 return NETDEV_TX_BUSY;
2235 } while (skb->next);
2238 if (likely(skb->next == NULL))
2239 skb->destructor = DEV_GSO_CB(skb)->destructor;
2246 static u32 hashrnd __read_mostly;
2249 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2250 * to be used as a distribution range.
2252 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2253 unsigned int num_tx_queues)
2257 u16 qcount = num_tx_queues;
2259 if (skb_rx_queue_recorded(skb)) {
2260 hash = skb_get_rx_queue(skb);
2261 while (unlikely(hash >= num_tx_queues))
2262 hash -= num_tx_queues;
2267 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2268 qoffset = dev->tc_to_txq[tc].offset;
2269 qcount = dev->tc_to_txq[tc].count;
2272 if (skb->sk && skb->sk->sk_hash)
2273 hash = skb->sk->sk_hash;
2275 hash = (__force u16) skb->protocol;
2276 hash = jhash_1word(hash, hashrnd);
2278 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2280 EXPORT_SYMBOL(__skb_tx_hash);
2282 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2284 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2285 if (net_ratelimit()) {
2286 pr_warning("%s selects TX queue %d, but "
2287 "real number of TX queues is %d\n",
2288 dev->name, queue_index, dev->real_num_tx_queues);
2295 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2298 struct xps_dev_maps *dev_maps;
2299 struct xps_map *map;
2300 int queue_index = -1;
2303 dev_maps = rcu_dereference(dev->xps_maps);
2305 map = rcu_dereference(
2306 dev_maps->cpu_map[raw_smp_processor_id()]);
2309 queue_index = map->queues[0];
2312 if (skb->sk && skb->sk->sk_hash)
2313 hash = skb->sk->sk_hash;
2315 hash = (__force u16) skb->protocol ^
2317 hash = jhash_1word(hash, hashrnd);
2318 queue_index = map->queues[
2319 ((u64)hash * map->len) >> 32];
2321 if (unlikely(queue_index >= dev->real_num_tx_queues))
2333 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2334 struct sk_buff *skb)
2337 const struct net_device_ops *ops = dev->netdev_ops;
2339 if (dev->real_num_tx_queues == 1)
2341 else if (ops->ndo_select_queue) {
2342 queue_index = ops->ndo_select_queue(dev, skb);
2343 queue_index = dev_cap_txqueue(dev, queue_index);
2345 struct sock *sk = skb->sk;
2346 queue_index = sk_tx_queue_get(sk);
2348 if (queue_index < 0 || skb->ooo_okay ||
2349 queue_index >= dev->real_num_tx_queues) {
2350 int old_index = queue_index;
2352 queue_index = get_xps_queue(dev, skb);
2353 if (queue_index < 0)
2354 queue_index = skb_tx_hash(dev, skb);
2356 if (queue_index != old_index && sk) {
2357 struct dst_entry *dst =
2358 rcu_dereference_check(sk->sk_dst_cache, 1);
2360 if (dst && skb_dst(skb) == dst)
2361 sk_tx_queue_set(sk, queue_index);
2366 skb_set_queue_mapping(skb, queue_index);
2367 return netdev_get_tx_queue(dev, queue_index);
2370 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2371 struct net_device *dev,
2372 struct netdev_queue *txq)
2374 spinlock_t *root_lock = qdisc_lock(q);
2378 qdisc_skb_cb(skb)->pkt_len = skb->len;
2379 qdisc_calculate_pkt_len(skb, q);
2381 * Heuristic to force contended enqueues to serialize on a
2382 * separate lock before trying to get qdisc main lock.
2383 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2384 * and dequeue packets faster.
2386 contended = qdisc_is_running(q);
2387 if (unlikely(contended))
2388 spin_lock(&q->busylock);
2390 spin_lock(root_lock);
2391 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2394 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2395 qdisc_run_begin(q)) {
2397 * This is a work-conserving queue; there are no old skbs
2398 * waiting to be sent out; and the qdisc is not running -
2399 * xmit the skb directly.
2401 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2404 qdisc_bstats_update(q, skb);
2406 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2407 if (unlikely(contended)) {
2408 spin_unlock(&q->busylock);
2415 rc = NET_XMIT_SUCCESS;
2418 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2419 if (qdisc_run_begin(q)) {
2420 if (unlikely(contended)) {
2421 spin_unlock(&q->busylock);
2427 spin_unlock(root_lock);
2428 if (unlikely(contended))
2429 spin_unlock(&q->busylock);
2433 static DEFINE_PER_CPU(int, xmit_recursion);
2434 #define RECURSION_LIMIT 10
2437 * dev_queue_xmit - transmit a buffer
2438 * @skb: buffer to transmit
2440 * Queue a buffer for transmission to a network device. The caller must
2441 * have set the device and priority and built the buffer before calling
2442 * this function. The function can be called from an interrupt.
2444 * A negative errno code is returned on a failure. A success does not
2445 * guarantee the frame will be transmitted as it may be dropped due
2446 * to congestion or traffic shaping.
2448 * -----------------------------------------------------------------------------------
2449 * I notice this method can also return errors from the queue disciplines,
2450 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2453 * Regardless of the return value, the skb is consumed, so it is currently
2454 * difficult to retry a send to this method. (You can bump the ref count
2455 * before sending to hold a reference for retry if you are careful.)
2457 * When calling this method, interrupts MUST be enabled. This is because
2458 * the BH enable code must have IRQs enabled so that it will not deadlock.
2461 int dev_queue_xmit(struct sk_buff *skb)
2463 struct net_device *dev = skb->dev;
2464 struct netdev_queue *txq;
2468 /* Disable soft irqs for various locks below. Also
2469 * stops preemption for RCU.
2473 txq = dev_pick_tx(dev, skb);
2474 q = rcu_dereference_bh(txq->qdisc);
2476 #ifdef CONFIG_NET_CLS_ACT
2477 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2479 trace_net_dev_queue(skb);
2481 rc = __dev_xmit_skb(skb, q, dev, txq);
2485 /* The device has no queue. Common case for software devices:
2486 loopback, all the sorts of tunnels...
2488 Really, it is unlikely that netif_tx_lock protection is necessary
2489 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2491 However, it is possible, that they rely on protection
2494 Check this and shot the lock. It is not prone from deadlocks.
2495 Either shot noqueue qdisc, it is even simpler 8)
2497 if (dev->flags & IFF_UP) {
2498 int cpu = smp_processor_id(); /* ok because BHs are off */
2500 if (txq->xmit_lock_owner != cpu) {
2502 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2503 goto recursion_alert;
2505 HARD_TX_LOCK(dev, txq, cpu);
2507 if (!netif_tx_queue_stopped(txq)) {
2508 __this_cpu_inc(xmit_recursion);
2509 rc = dev_hard_start_xmit(skb, dev, txq);
2510 __this_cpu_dec(xmit_recursion);
2511 if (dev_xmit_complete(rc)) {
2512 HARD_TX_UNLOCK(dev, txq);
2516 HARD_TX_UNLOCK(dev, txq);
2517 if (net_ratelimit())
2518 printk(KERN_CRIT "Virtual device %s asks to "
2519 "queue packet!\n", dev->name);
2521 /* Recursion is detected! It is possible,
2525 if (net_ratelimit())
2526 printk(KERN_CRIT "Dead loop on virtual device "
2527 "%s, fix it urgently!\n", dev->name);
2532 rcu_read_unlock_bh();
2537 rcu_read_unlock_bh();
2540 EXPORT_SYMBOL(dev_queue_xmit);
2543 /*=======================================================================
2545 =======================================================================*/
2547 int netdev_max_backlog __read_mostly = 1000;
2548 int netdev_tstamp_prequeue __read_mostly = 1;
2549 int netdev_budget __read_mostly = 300;
2550 int weight_p __read_mostly = 64; /* old backlog weight */
2552 /* Called with irq disabled */
2553 static inline void ____napi_schedule(struct softnet_data *sd,
2554 struct napi_struct *napi)
2556 list_add_tail(&napi->poll_list, &sd->poll_list);
2557 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2561 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2562 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2563 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2564 * if hash is a canonical 4-tuple hash over transport ports.
2566 void __skb_get_rxhash(struct sk_buff *skb)
2568 int nhoff, hash = 0, poff;
2569 const struct ipv6hdr *ip6;
2570 const struct iphdr *ip;
2571 const struct vlan_hdr *vlan;
2580 nhoff = skb_network_offset(skb);
2581 proto = skb->protocol;
2585 case __constant_htons(ETH_P_IP):
2587 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2590 ip = (const struct iphdr *) (skb->data + nhoff);
2591 if (ip_is_fragment(ip))
2594 ip_proto = ip->protocol;
2595 addr1 = (__force u32) ip->saddr;
2596 addr2 = (__force u32) ip->daddr;
2597 nhoff += ip->ihl * 4;
2599 case __constant_htons(ETH_P_IPV6):
2601 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2604 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2605 ip_proto = ip6->nexthdr;
2606 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2607 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2610 case __constant_htons(ETH_P_8021Q):
2611 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2613 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2614 proto = vlan->h_vlan_encapsulated_proto;
2615 nhoff += sizeof(*vlan);
2617 case __constant_htons(ETH_P_PPP_SES):
2618 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2620 proto = *((__be16 *) (skb->data + nhoff +
2621 sizeof(struct pppoe_hdr)));
2622 nhoff += PPPOE_SES_HLEN;
2624 case __constant_htons(PPP_IP):
2626 case __constant_htons(PPP_IPV6):
2637 if (pskb_may_pull(skb, nhoff + 16)) {
2638 u8 *h = skb->data + nhoff;
2639 __be16 flags = *(__be16 *)h;
2642 * Only look inside GRE if version zero and no
2645 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2646 proto = *(__be16 *)(h + 2);
2648 if (flags & GRE_CSUM)
2650 if (flags & GRE_KEY)
2652 if (flags & GRE_SEQ)
2665 poff = proto_ports_offset(ip_proto);
2668 if (pskb_may_pull(skb, nhoff + 4)) {
2669 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2670 if (ports.v16[1] < ports.v16[0])
2671 swap(ports.v16[0], ports.v16[1]);
2676 /* get a consistent hash (same value on both flow directions) */
2680 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2687 EXPORT_SYMBOL(__skb_get_rxhash);
2691 /* One global table that all flow-based protocols share. */
2692 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2693 EXPORT_SYMBOL(rps_sock_flow_table);
2695 static struct rps_dev_flow *
2696 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2697 struct rps_dev_flow *rflow, u16 next_cpu)
2699 if (next_cpu != RPS_NO_CPU) {
2700 #ifdef CONFIG_RFS_ACCEL
2701 struct netdev_rx_queue *rxqueue;
2702 struct rps_dev_flow_table *flow_table;
2703 struct rps_dev_flow *old_rflow;
2708 /* Should we steer this flow to a different hardware queue? */
2709 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2710 !(dev->features & NETIF_F_NTUPLE))
2712 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2713 if (rxq_index == skb_get_rx_queue(skb))
2716 rxqueue = dev->_rx + rxq_index;
2717 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2720 flow_id = skb->rxhash & flow_table->mask;
2721 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2722 rxq_index, flow_id);
2726 rflow = &flow_table->flows[flow_id];
2728 if (old_rflow->filter == rflow->filter)
2729 old_rflow->filter = RPS_NO_FILTER;
2733 per_cpu(softnet_data, next_cpu).input_queue_head;
2736 rflow->cpu = next_cpu;
2741 * get_rps_cpu is called from netif_receive_skb and returns the target
2742 * CPU from the RPS map of the receiving queue for a given skb.
2743 * rcu_read_lock must be held on entry.
2745 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2746 struct rps_dev_flow **rflowp)
2748 struct netdev_rx_queue *rxqueue;
2749 struct rps_map *map;
2750 struct rps_dev_flow_table *flow_table;
2751 struct rps_sock_flow_table *sock_flow_table;
2755 if (skb_rx_queue_recorded(skb)) {
2756 u16 index = skb_get_rx_queue(skb);
2757 if (unlikely(index >= dev->real_num_rx_queues)) {
2758 WARN_ONCE(dev->real_num_rx_queues > 1,
2759 "%s received packet on queue %u, but number "
2760 "of RX queues is %u\n",
2761 dev->name, index, dev->real_num_rx_queues);
2764 rxqueue = dev->_rx + index;
2768 map = rcu_dereference(rxqueue->rps_map);
2770 if (map->len == 1 &&
2771 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2772 tcpu = map->cpus[0];
2773 if (cpu_online(tcpu))
2777 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2781 skb_reset_network_header(skb);
2782 if (!skb_get_rxhash(skb))
2785 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2786 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2787 if (flow_table && sock_flow_table) {
2789 struct rps_dev_flow *rflow;
2791 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2794 next_cpu = sock_flow_table->ents[skb->rxhash &
2795 sock_flow_table->mask];
2798 * If the desired CPU (where last recvmsg was done) is
2799 * different from current CPU (one in the rx-queue flow
2800 * table entry), switch if one of the following holds:
2801 * - Current CPU is unset (equal to RPS_NO_CPU).
2802 * - Current CPU is offline.
2803 * - The current CPU's queue tail has advanced beyond the
2804 * last packet that was enqueued using this table entry.
2805 * This guarantees that all previous packets for the flow
2806 * have been dequeued, thus preserving in order delivery.
2808 if (unlikely(tcpu != next_cpu) &&
2809 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2810 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2811 rflow->last_qtail)) >= 0))
2812 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2814 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2822 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2824 if (cpu_online(tcpu)) {
2834 #ifdef CONFIG_RFS_ACCEL
2837 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2838 * @dev: Device on which the filter was set
2839 * @rxq_index: RX queue index
2840 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2841 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2843 * Drivers that implement ndo_rx_flow_steer() should periodically call
2844 * this function for each installed filter and remove the filters for
2845 * which it returns %true.
2847 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2848 u32 flow_id, u16 filter_id)
2850 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2851 struct rps_dev_flow_table *flow_table;
2852 struct rps_dev_flow *rflow;
2857 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2858 if (flow_table && flow_id <= flow_table->mask) {
2859 rflow = &flow_table->flows[flow_id];
2860 cpu = ACCESS_ONCE(rflow->cpu);
2861 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2862 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2863 rflow->last_qtail) <
2864 (int)(10 * flow_table->mask)))
2870 EXPORT_SYMBOL(rps_may_expire_flow);
2872 #endif /* CONFIG_RFS_ACCEL */
2874 /* Called from hardirq (IPI) context */
2875 static void rps_trigger_softirq(void *data)
2877 struct softnet_data *sd = data;
2879 ____napi_schedule(sd, &sd->backlog);
2883 #endif /* CONFIG_RPS */
2886 * Check if this softnet_data structure is another cpu one
2887 * If yes, queue it to our IPI list and return 1
2890 static int rps_ipi_queued(struct softnet_data *sd)
2893 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2896 sd->rps_ipi_next = mysd->rps_ipi_list;
2897 mysd->rps_ipi_list = sd;
2899 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2902 #endif /* CONFIG_RPS */
2907 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2908 * queue (may be a remote CPU queue).
2910 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2911 unsigned int *qtail)
2913 struct softnet_data *sd;
2914 unsigned long flags;
2916 sd = &per_cpu(softnet_data, cpu);
2918 local_irq_save(flags);
2921 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2922 if (skb_queue_len(&sd->input_pkt_queue)) {
2924 __skb_queue_tail(&sd->input_pkt_queue, skb);
2925 input_queue_tail_incr_save(sd, qtail);
2927 local_irq_restore(flags);
2928 return NET_RX_SUCCESS;
2931 /* Schedule NAPI for backlog device
2932 * We can use non atomic operation since we own the queue lock
2934 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2935 if (!rps_ipi_queued(sd))
2936 ____napi_schedule(sd, &sd->backlog);
2944 local_irq_restore(flags);
2946 atomic_long_inc(&skb->dev->rx_dropped);
2952 * netif_rx - post buffer to the network code
2953 * @skb: buffer to post
2955 * This function receives a packet from a device driver and queues it for
2956 * the upper (protocol) levels to process. It always succeeds. The buffer
2957 * may be dropped during processing for congestion control or by the
2961 * NET_RX_SUCCESS (no congestion)
2962 * NET_RX_DROP (packet was dropped)
2966 int netif_rx(struct sk_buff *skb)
2970 /* if netpoll wants it, pretend we never saw it */
2971 if (netpoll_rx(skb))
2974 if (netdev_tstamp_prequeue)
2975 net_timestamp_check(skb);
2977 trace_netif_rx(skb);
2980 struct rps_dev_flow voidflow, *rflow = &voidflow;
2986 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2988 cpu = smp_processor_id();
2990 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2998 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3004 EXPORT_SYMBOL(netif_rx);
3006 int netif_rx_ni(struct sk_buff *skb)
3011 err = netif_rx(skb);
3012 if (local_softirq_pending())
3018 EXPORT_SYMBOL(netif_rx_ni);
3020 static void net_tx_action(struct softirq_action *h)
3022 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3024 if (sd->completion_queue) {
3025 struct sk_buff *clist;
3027 local_irq_disable();
3028 clist = sd->completion_queue;
3029 sd->completion_queue = NULL;
3033 struct sk_buff *skb = clist;
3034 clist = clist->next;
3036 WARN_ON(atomic_read(&skb->users));
3037 trace_kfree_skb(skb, net_tx_action);
3042 if (sd->output_queue) {
3045 local_irq_disable();
3046 head = sd->output_queue;
3047 sd->output_queue = NULL;
3048 sd->output_queue_tailp = &sd->output_queue;
3052 struct Qdisc *q = head;
3053 spinlock_t *root_lock;
3055 head = head->next_sched;
3057 root_lock = qdisc_lock(q);
3058 if (spin_trylock(root_lock)) {
3059 smp_mb__before_clear_bit();
3060 clear_bit(__QDISC_STATE_SCHED,
3063 spin_unlock(root_lock);
3065 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3067 __netif_reschedule(q);
3069 smp_mb__before_clear_bit();
3070 clear_bit(__QDISC_STATE_SCHED,
3078 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3079 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3080 /* This hook is defined here for ATM LANE */
3081 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3082 unsigned char *addr) __read_mostly;
3083 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3086 #ifdef CONFIG_NET_CLS_ACT
3087 /* TODO: Maybe we should just force sch_ingress to be compiled in
3088 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3089 * a compare and 2 stores extra right now if we dont have it on
3090 * but have CONFIG_NET_CLS_ACT
3091 * NOTE: This doesn't stop any functionality; if you dont have
3092 * the ingress scheduler, you just can't add policies on ingress.
3095 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3097 struct net_device *dev = skb->dev;
3098 u32 ttl = G_TC_RTTL(skb->tc_verd);
3099 int result = TC_ACT_OK;
3102 if (unlikely(MAX_RED_LOOP < ttl++)) {
3103 if (net_ratelimit())
3104 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3105 skb->skb_iif, dev->ifindex);
3109 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3110 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3113 if (q != &noop_qdisc) {
3114 spin_lock(qdisc_lock(q));
3115 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3116 result = qdisc_enqueue_root(skb, q);
3117 spin_unlock(qdisc_lock(q));
3123 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3124 struct packet_type **pt_prev,
3125 int *ret, struct net_device *orig_dev)
3127 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3129 if (!rxq || rxq->qdisc == &noop_qdisc)
3133 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3137 switch (ing_filter(skb, rxq)) {
3151 * netdev_rx_handler_register - register receive handler
3152 * @dev: device to register a handler for
3153 * @rx_handler: receive handler to register
3154 * @rx_handler_data: data pointer that is used by rx handler
3156 * Register a receive hander for a device. This handler will then be
3157 * called from __netif_receive_skb. A negative errno code is returned
3160 * The caller must hold the rtnl_mutex.
3162 * For a general description of rx_handler, see enum rx_handler_result.
3164 int netdev_rx_handler_register(struct net_device *dev,
3165 rx_handler_func_t *rx_handler,
3166 void *rx_handler_data)
3170 if (dev->rx_handler)
3173 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3174 rcu_assign_pointer(dev->rx_handler, rx_handler);
3178 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3181 * netdev_rx_handler_unregister - unregister receive handler
3182 * @dev: device to unregister a handler from
3184 * Unregister a receive hander from a device.
3186 * The caller must hold the rtnl_mutex.
3188 void netdev_rx_handler_unregister(struct net_device *dev)
3192 RCU_INIT_POINTER(dev->rx_handler, NULL);
3193 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3195 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3197 static int __netif_receive_skb(struct sk_buff *skb)
3199 struct packet_type *ptype, *pt_prev;
3200 rx_handler_func_t *rx_handler;
3201 struct net_device *orig_dev;
3202 struct net_device *null_or_dev;
3203 bool deliver_exact = false;
3204 int ret = NET_RX_DROP;
3207 if (!netdev_tstamp_prequeue)
3208 net_timestamp_check(skb);
3210 trace_netif_receive_skb(skb);
3212 /* if we've gotten here through NAPI, check netpoll */
3213 if (netpoll_receive_skb(skb))
3217 skb->skb_iif = skb->dev->ifindex;
3218 orig_dev = skb->dev;
3220 skb_reset_network_header(skb);
3221 skb_reset_transport_header(skb);
3222 skb_reset_mac_len(skb);
3230 __this_cpu_inc(softnet_data.processed);
3232 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3233 skb = vlan_untag(skb);
3238 #ifdef CONFIG_NET_CLS_ACT
3239 if (skb->tc_verd & TC_NCLS) {
3240 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3245 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3246 if (!ptype->dev || ptype->dev == skb->dev) {
3248 ret = deliver_skb(skb, pt_prev, orig_dev);
3253 #ifdef CONFIG_NET_CLS_ACT
3254 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3260 rx_handler = rcu_dereference(skb->dev->rx_handler);
3261 if (vlan_tx_tag_present(skb)) {
3263 ret = deliver_skb(skb, pt_prev, orig_dev);
3266 if (vlan_do_receive(&skb, !rx_handler))
3268 else if (unlikely(!skb))
3274 ret = deliver_skb(skb, pt_prev, orig_dev);
3277 switch (rx_handler(&skb)) {
3278 case RX_HANDLER_CONSUMED:
3280 case RX_HANDLER_ANOTHER:
3282 case RX_HANDLER_EXACT:
3283 deliver_exact = true;
3284 case RX_HANDLER_PASS:
3291 /* deliver only exact match when indicated */
3292 null_or_dev = deliver_exact ? skb->dev : NULL;
3294 type = skb->protocol;
3295 list_for_each_entry_rcu(ptype,
3296 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3297 if (ptype->type == type &&
3298 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3299 ptype->dev == orig_dev)) {
3301 ret = deliver_skb(skb, pt_prev, orig_dev);
3307 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3309 atomic_long_inc(&skb->dev->rx_dropped);
3311 /* Jamal, now you will not able to escape explaining
3312 * me how you were going to use this. :-)
3323 * netif_receive_skb - process receive buffer from network
3324 * @skb: buffer to process
3326 * netif_receive_skb() is the main receive data processing function.
3327 * It always succeeds. The buffer may be dropped during processing
3328 * for congestion control or by the protocol layers.
3330 * This function may only be called from softirq context and interrupts
3331 * should be enabled.
3333 * Return values (usually ignored):
3334 * NET_RX_SUCCESS: no congestion
3335 * NET_RX_DROP: packet was dropped
3337 int netif_receive_skb(struct sk_buff *skb)
3339 if (netdev_tstamp_prequeue)
3340 net_timestamp_check(skb);
3342 if (skb_defer_rx_timestamp(skb))
3343 return NET_RX_SUCCESS;
3347 struct rps_dev_flow voidflow, *rflow = &voidflow;
3352 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3355 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3359 ret = __netif_receive_skb(skb);
3365 return __netif_receive_skb(skb);
3368 EXPORT_SYMBOL(netif_receive_skb);
3370 /* Network device is going away, flush any packets still pending
3371 * Called with irqs disabled.
3373 static void flush_backlog(void *arg)
3375 struct net_device *dev = arg;
3376 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3377 struct sk_buff *skb, *tmp;
3380 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3381 if (skb->dev == dev) {
3382 __skb_unlink(skb, &sd->input_pkt_queue);
3384 input_queue_head_incr(sd);
3389 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3390 if (skb->dev == dev) {
3391 __skb_unlink(skb, &sd->process_queue);
3393 input_queue_head_incr(sd);
3398 static int napi_gro_complete(struct sk_buff *skb)
3400 struct packet_type *ptype;
3401 __be16 type = skb->protocol;
3402 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3405 if (NAPI_GRO_CB(skb)->count == 1) {
3406 skb_shinfo(skb)->gso_size = 0;
3411 list_for_each_entry_rcu(ptype, head, list) {
3412 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3415 err = ptype->gro_complete(skb);
3421 WARN_ON(&ptype->list == head);
3423 return NET_RX_SUCCESS;
3427 return netif_receive_skb(skb);
3430 inline void napi_gro_flush(struct napi_struct *napi)
3432 struct sk_buff *skb, *next;
3434 for (skb = napi->gro_list; skb; skb = next) {
3437 napi_gro_complete(skb);
3440 napi->gro_count = 0;
3441 napi->gro_list = NULL;
3443 EXPORT_SYMBOL(napi_gro_flush);
3445 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3447 struct sk_buff **pp = NULL;
3448 struct packet_type *ptype;
3449 __be16 type = skb->protocol;
3450 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3453 enum gro_result ret;
3455 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3458 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3462 list_for_each_entry_rcu(ptype, head, list) {
3463 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3466 skb_set_network_header(skb, skb_gro_offset(skb));
3467 mac_len = skb->network_header - skb->mac_header;
3468 skb->mac_len = mac_len;
3469 NAPI_GRO_CB(skb)->same_flow = 0;
3470 NAPI_GRO_CB(skb)->flush = 0;
3471 NAPI_GRO_CB(skb)->free = 0;
3473 pp = ptype->gro_receive(&napi->gro_list, skb);
3478 if (&ptype->list == head)
3481 same_flow = NAPI_GRO_CB(skb)->same_flow;
3482 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3485 struct sk_buff *nskb = *pp;
3489 napi_gro_complete(nskb);
3496 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3500 NAPI_GRO_CB(skb)->count = 1;
3501 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3502 skb->next = napi->gro_list;
3503 napi->gro_list = skb;
3507 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3508 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3510 BUG_ON(skb->end - skb->tail < grow);
3512 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3515 skb->data_len -= grow;
3517 skb_shinfo(skb)->frags[0].page_offset += grow;
3518 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3520 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3521 skb_frag_unref(skb, 0);
3522 memmove(skb_shinfo(skb)->frags,
3523 skb_shinfo(skb)->frags + 1,
3524 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3535 EXPORT_SYMBOL(dev_gro_receive);
3537 static inline gro_result_t
3538 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3541 unsigned int maclen = skb->dev->hard_header_len;
3543 for (p = napi->gro_list; p; p = p->next) {
3544 unsigned long diffs;
3546 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3547 diffs |= p->vlan_tci ^ skb->vlan_tci;
3548 if (maclen == ETH_HLEN)
3549 diffs |= compare_ether_header(skb_mac_header(p),
3550 skb_gro_mac_header(skb));
3552 diffs = memcmp(skb_mac_header(p),
3553 skb_gro_mac_header(skb),
3555 NAPI_GRO_CB(p)->same_flow = !diffs;
3556 NAPI_GRO_CB(p)->flush = 0;
3559 return dev_gro_receive(napi, skb);
3562 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3566 if (netif_receive_skb(skb))
3571 case GRO_MERGED_FREE:
3582 EXPORT_SYMBOL(napi_skb_finish);
3584 void skb_gro_reset_offset(struct sk_buff *skb)
3586 NAPI_GRO_CB(skb)->data_offset = 0;
3587 NAPI_GRO_CB(skb)->frag0 = NULL;
3588 NAPI_GRO_CB(skb)->frag0_len = 0;
3590 if (skb->mac_header == skb->tail &&
3591 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3592 NAPI_GRO_CB(skb)->frag0 =
3593 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3594 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3597 EXPORT_SYMBOL(skb_gro_reset_offset);
3599 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3601 skb_gro_reset_offset(skb);
3603 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3605 EXPORT_SYMBOL(napi_gro_receive);
3607 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3609 __skb_pull(skb, skb_headlen(skb));
3610 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3611 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3613 skb->dev = napi->dev;
3619 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3621 struct sk_buff *skb = napi->skb;
3624 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3630 EXPORT_SYMBOL(napi_get_frags);
3632 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3638 skb->protocol = eth_type_trans(skb, skb->dev);
3640 if (ret == GRO_HELD)
3641 skb_gro_pull(skb, -ETH_HLEN);
3642 else if (netif_receive_skb(skb))
3647 case GRO_MERGED_FREE:
3648 napi_reuse_skb(napi, skb);
3657 EXPORT_SYMBOL(napi_frags_finish);
3659 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3661 struct sk_buff *skb = napi->skb;
3668 skb_reset_mac_header(skb);
3669 skb_gro_reset_offset(skb);
3671 off = skb_gro_offset(skb);
3672 hlen = off + sizeof(*eth);
3673 eth = skb_gro_header_fast(skb, off);
3674 if (skb_gro_header_hard(skb, hlen)) {
3675 eth = skb_gro_header_slow(skb, hlen, off);
3676 if (unlikely(!eth)) {
3677 napi_reuse_skb(napi, skb);
3683 skb_gro_pull(skb, sizeof(*eth));
3686 * This works because the only protocols we care about don't require
3687 * special handling. We'll fix it up properly at the end.
3689 skb->protocol = eth->h_proto;
3694 EXPORT_SYMBOL(napi_frags_skb);
3696 gro_result_t napi_gro_frags(struct napi_struct *napi)
3698 struct sk_buff *skb = napi_frags_skb(napi);
3703 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3705 EXPORT_SYMBOL(napi_gro_frags);
3708 * net_rps_action sends any pending IPI's for rps.
3709 * Note: called with local irq disabled, but exits with local irq enabled.
3711 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3714 struct softnet_data *remsd = sd->rps_ipi_list;
3717 sd->rps_ipi_list = NULL;
3721 /* Send pending IPI's to kick RPS processing on remote cpus. */
3723 struct softnet_data *next = remsd->rps_ipi_next;
3725 if (cpu_online(remsd->cpu))
3726 __smp_call_function_single(remsd->cpu,
3735 static int process_backlog(struct napi_struct *napi, int quota)
3738 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3741 /* Check if we have pending ipi, its better to send them now,
3742 * not waiting net_rx_action() end.
3744 if (sd->rps_ipi_list) {
3745 local_irq_disable();
3746 net_rps_action_and_irq_enable(sd);
3749 napi->weight = weight_p;
3750 local_irq_disable();
3751 while (work < quota) {
3752 struct sk_buff *skb;
3755 while ((skb = __skb_dequeue(&sd->process_queue))) {
3757 __netif_receive_skb(skb);
3758 local_irq_disable();
3759 input_queue_head_incr(sd);
3760 if (++work >= quota) {
3767 qlen = skb_queue_len(&sd->input_pkt_queue);
3769 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3770 &sd->process_queue);
3772 if (qlen < quota - work) {
3774 * Inline a custom version of __napi_complete().
3775 * only current cpu owns and manipulates this napi,
3776 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3777 * we can use a plain write instead of clear_bit(),
3778 * and we dont need an smp_mb() memory barrier.
3780 list_del(&napi->poll_list);
3783 quota = work + qlen;
3793 * __napi_schedule - schedule for receive
3794 * @n: entry to schedule
3796 * The entry's receive function will be scheduled to run
3798 void __napi_schedule(struct napi_struct *n)
3800 unsigned long flags;
3802 local_irq_save(flags);
3803 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3804 local_irq_restore(flags);
3806 EXPORT_SYMBOL(__napi_schedule);
3808 void __napi_complete(struct napi_struct *n)
3810 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3811 BUG_ON(n->gro_list);
3813 list_del(&n->poll_list);
3814 smp_mb__before_clear_bit();
3815 clear_bit(NAPI_STATE_SCHED, &n->state);
3817 EXPORT_SYMBOL(__napi_complete);
3819 void napi_complete(struct napi_struct *n)
3821 unsigned long flags;
3824 * don't let napi dequeue from the cpu poll list
3825 * just in case its running on a different cpu
3827 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3831 local_irq_save(flags);
3833 local_irq_restore(flags);
3835 EXPORT_SYMBOL(napi_complete);
3837 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3838 int (*poll)(struct napi_struct *, int), int weight)
3840 INIT_LIST_HEAD(&napi->poll_list);
3841 napi->gro_count = 0;
3842 napi->gro_list = NULL;
3845 napi->weight = weight;
3846 list_add(&napi->dev_list, &dev->napi_list);
3848 #ifdef CONFIG_NETPOLL
3849 spin_lock_init(&napi->poll_lock);
3850 napi->poll_owner = -1;
3852 set_bit(NAPI_STATE_SCHED, &napi->state);
3854 EXPORT_SYMBOL(netif_napi_add);
3856 void netif_napi_del(struct napi_struct *napi)
3858 struct sk_buff *skb, *next;
3860 list_del_init(&napi->dev_list);
3861 napi_free_frags(napi);
3863 for (skb = napi->gro_list; skb; skb = next) {
3869 napi->gro_list = NULL;
3870 napi->gro_count = 0;
3872 EXPORT_SYMBOL(netif_napi_del);
3874 static void net_rx_action(struct softirq_action *h)
3876 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3877 unsigned long time_limit = jiffies + 2;
3878 int budget = netdev_budget;
3881 local_irq_disable();
3883 while (!list_empty(&sd->poll_list)) {
3884 struct napi_struct *n;
3887 /* If softirq window is exhuasted then punt.
3888 * Allow this to run for 2 jiffies since which will allow
3889 * an average latency of 1.5/HZ.
3891 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3896 /* Even though interrupts have been re-enabled, this
3897 * access is safe because interrupts can only add new
3898 * entries to the tail of this list, and only ->poll()
3899 * calls can remove this head entry from the list.
3901 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3903 have = netpoll_poll_lock(n);
3907 /* This NAPI_STATE_SCHED test is for avoiding a race
3908 * with netpoll's poll_napi(). Only the entity which
3909 * obtains the lock and sees NAPI_STATE_SCHED set will
3910 * actually make the ->poll() call. Therefore we avoid
3911 * accidentally calling ->poll() when NAPI is not scheduled.
3914 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3915 work = n->poll(n, weight);
3919 WARN_ON_ONCE(work > weight);
3923 local_irq_disable();
3925 /* Drivers must not modify the NAPI state if they
3926 * consume the entire weight. In such cases this code
3927 * still "owns" the NAPI instance and therefore can
3928 * move the instance around on the list at-will.
3930 if (unlikely(work == weight)) {
3931 if (unlikely(napi_disable_pending(n))) {
3934 local_irq_disable();
3936 list_move_tail(&n->poll_list, &sd->poll_list);
3939 netpoll_poll_unlock(have);
3942 net_rps_action_and_irq_enable(sd);
3944 #ifdef CONFIG_NET_DMA
3946 * There may not be any more sk_buffs coming right now, so push
3947 * any pending DMA copies to hardware
3949 dma_issue_pending_all();
3956 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3960 static gifconf_func_t *gifconf_list[NPROTO];
3963 * register_gifconf - register a SIOCGIF handler
3964 * @family: Address family
3965 * @gifconf: Function handler
3967 * Register protocol dependent address dumping routines. The handler
3968 * that is passed must not be freed or reused until it has been replaced
3969 * by another handler.
3971 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3973 if (family >= NPROTO)
3975 gifconf_list[family] = gifconf;
3978 EXPORT_SYMBOL(register_gifconf);
3982 * Map an interface index to its name (SIOCGIFNAME)
3986 * We need this ioctl for efficient implementation of the
3987 * if_indextoname() function required by the IPv6 API. Without
3988 * it, we would have to search all the interfaces to find a
3992 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3994 struct net_device *dev;
3998 * Fetch the caller's info block.
4001 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4005 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4011 strcpy(ifr.ifr_name, dev->name);
4014 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4020 * Perform a SIOCGIFCONF call. This structure will change
4021 * size eventually, and there is nothing I can do about it.
4022 * Thus we will need a 'compatibility mode'.
4025 static int dev_ifconf(struct net *net, char __user *arg)
4028 struct net_device *dev;
4035 * Fetch the caller's info block.
4038 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4045 * Loop over the interfaces, and write an info block for each.
4049 for_each_netdev(net, dev) {
4050 for (i = 0; i < NPROTO; i++) {
4051 if (gifconf_list[i]) {
4054 done = gifconf_list[i](dev, NULL, 0);
4056 done = gifconf_list[i](dev, pos + total,
4066 * All done. Write the updated control block back to the caller.
4068 ifc.ifc_len = total;
4071 * Both BSD and Solaris return 0 here, so we do too.
4073 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4076 #ifdef CONFIG_PROC_FS
4078 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4080 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4081 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4082 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4084 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4086 struct net *net = seq_file_net(seq);
4087 struct net_device *dev;
4088 struct hlist_node *p;
4089 struct hlist_head *h;
4090 unsigned int count = 0, offset = get_offset(*pos);
4092 h = &net->dev_name_head[get_bucket(*pos)];
4093 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4094 if (++count == offset)
4101 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4103 struct net_device *dev;
4104 unsigned int bucket;
4107 dev = dev_from_same_bucket(seq, pos);
4111 bucket = get_bucket(*pos) + 1;
4112 *pos = set_bucket_offset(bucket, 1);
4113 } while (bucket < NETDEV_HASHENTRIES);
4119 * This is invoked by the /proc filesystem handler to display a device
4122 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4127 return SEQ_START_TOKEN;
4129 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4132 return dev_from_bucket(seq, pos);
4135 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4138 return dev_from_bucket(seq, pos);
4141 void dev_seq_stop(struct seq_file *seq, void *v)
4147 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4149 struct rtnl_link_stats64 temp;
4150 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4152 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4153 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4154 dev->name, stats->rx_bytes, stats->rx_packets,
4156 stats->rx_dropped + stats->rx_missed_errors,
4157 stats->rx_fifo_errors,
4158 stats->rx_length_errors + stats->rx_over_errors +
4159 stats->rx_crc_errors + stats->rx_frame_errors,
4160 stats->rx_compressed, stats->multicast,
4161 stats->tx_bytes, stats->tx_packets,
4162 stats->tx_errors, stats->tx_dropped,
4163 stats->tx_fifo_errors, stats->collisions,
4164 stats->tx_carrier_errors +
4165 stats->tx_aborted_errors +
4166 stats->tx_window_errors +
4167 stats->tx_heartbeat_errors,
4168 stats->tx_compressed);
4172 * Called from the PROCfs module. This now uses the new arbitrary sized
4173 * /proc/net interface to create /proc/net/dev
4175 static int dev_seq_show(struct seq_file *seq, void *v)
4177 if (v == SEQ_START_TOKEN)
4178 seq_puts(seq, "Inter-| Receive "
4180 " face |bytes packets errs drop fifo frame "
4181 "compressed multicast|bytes packets errs "
4182 "drop fifo colls carrier compressed\n");
4184 dev_seq_printf_stats(seq, v);
4188 static struct softnet_data *softnet_get_online(loff_t *pos)
4190 struct softnet_data *sd = NULL;
4192 while (*pos < nr_cpu_ids)
4193 if (cpu_online(*pos)) {
4194 sd = &per_cpu(softnet_data, *pos);
4201 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4203 return softnet_get_online(pos);
4206 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4209 return softnet_get_online(pos);
4212 static void softnet_seq_stop(struct seq_file *seq, void *v)
4216 static int softnet_seq_show(struct seq_file *seq, void *v)
4218 struct softnet_data *sd = v;
4220 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4221 sd->processed, sd->dropped, sd->time_squeeze, 0,
4222 0, 0, 0, 0, /* was fastroute */
4223 sd->cpu_collision, sd->received_rps);
4227 static const struct seq_operations dev_seq_ops = {
4228 .start = dev_seq_start,
4229 .next = dev_seq_next,
4230 .stop = dev_seq_stop,
4231 .show = dev_seq_show,
4234 static int dev_seq_open(struct inode *inode, struct file *file)
4236 return seq_open_net(inode, file, &dev_seq_ops,
4237 sizeof(struct seq_net_private));
4240 static const struct file_operations dev_seq_fops = {
4241 .owner = THIS_MODULE,
4242 .open = dev_seq_open,
4244 .llseek = seq_lseek,
4245 .release = seq_release_net,
4248 static const struct seq_operations softnet_seq_ops = {
4249 .start = softnet_seq_start,
4250 .next = softnet_seq_next,
4251 .stop = softnet_seq_stop,
4252 .show = softnet_seq_show,
4255 static int softnet_seq_open(struct inode *inode, struct file *file)
4257 return seq_open(file, &softnet_seq_ops);
4260 static const struct file_operations softnet_seq_fops = {
4261 .owner = THIS_MODULE,
4262 .open = softnet_seq_open,
4264 .llseek = seq_lseek,
4265 .release = seq_release,
4268 static void *ptype_get_idx(loff_t pos)
4270 struct packet_type *pt = NULL;
4274 list_for_each_entry_rcu(pt, &ptype_all, list) {
4280 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4281 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4290 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4294 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4297 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4299 struct packet_type *pt;
4300 struct list_head *nxt;
4304 if (v == SEQ_START_TOKEN)
4305 return ptype_get_idx(0);
4308 nxt = pt->list.next;
4309 if (pt->type == htons(ETH_P_ALL)) {
4310 if (nxt != &ptype_all)
4313 nxt = ptype_base[0].next;
4315 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4317 while (nxt == &ptype_base[hash]) {
4318 if (++hash >= PTYPE_HASH_SIZE)
4320 nxt = ptype_base[hash].next;
4323 return list_entry(nxt, struct packet_type, list);
4326 static void ptype_seq_stop(struct seq_file *seq, void *v)
4332 static int ptype_seq_show(struct seq_file *seq, void *v)
4334 struct packet_type *pt = v;
4336 if (v == SEQ_START_TOKEN)
4337 seq_puts(seq, "Type Device Function\n");
4338 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4339 if (pt->type == htons(ETH_P_ALL))
4340 seq_puts(seq, "ALL ");
4342 seq_printf(seq, "%04x", ntohs(pt->type));
4344 seq_printf(seq, " %-8s %pF\n",
4345 pt->dev ? pt->dev->name : "", pt->func);
4351 static const struct seq_operations ptype_seq_ops = {
4352 .start = ptype_seq_start,
4353 .next = ptype_seq_next,
4354 .stop = ptype_seq_stop,
4355 .show = ptype_seq_show,
4358 static int ptype_seq_open(struct inode *inode, struct file *file)
4360 return seq_open_net(inode, file, &ptype_seq_ops,
4361 sizeof(struct seq_net_private));
4364 static const struct file_operations ptype_seq_fops = {
4365 .owner = THIS_MODULE,
4366 .open = ptype_seq_open,
4368 .llseek = seq_lseek,
4369 .release = seq_release_net,
4373 static int __net_init dev_proc_net_init(struct net *net)
4377 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4379 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4381 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4384 if (wext_proc_init(net))
4390 proc_net_remove(net, "ptype");
4392 proc_net_remove(net, "softnet_stat");
4394 proc_net_remove(net, "dev");
4398 static void __net_exit dev_proc_net_exit(struct net *net)
4400 wext_proc_exit(net);
4402 proc_net_remove(net, "ptype");
4403 proc_net_remove(net, "softnet_stat");
4404 proc_net_remove(net, "dev");
4407 static struct pernet_operations __net_initdata dev_proc_ops = {
4408 .init = dev_proc_net_init,
4409 .exit = dev_proc_net_exit,
4412 static int __init dev_proc_init(void)
4414 return register_pernet_subsys(&dev_proc_ops);
4417 #define dev_proc_init() 0
4418 #endif /* CONFIG_PROC_FS */
4422 * netdev_set_master - set up master pointer
4423 * @slave: slave device
4424 * @master: new master device
4426 * Changes the master device of the slave. Pass %NULL to break the
4427 * bonding. The caller must hold the RTNL semaphore. On a failure
4428 * a negative errno code is returned. On success the reference counts
4429 * are adjusted and the function returns zero.
4431 int netdev_set_master(struct net_device *slave, struct net_device *master)
4433 struct net_device *old = slave->master;
4443 slave->master = master;
4449 EXPORT_SYMBOL(netdev_set_master);
4452 * netdev_set_bond_master - set up bonding master/slave pair
4453 * @slave: slave device
4454 * @master: new master device
4456 * Changes the master device of the slave. Pass %NULL to break the
4457 * bonding. The caller must hold the RTNL semaphore. On a failure
4458 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4459 * to the routing socket and the function returns zero.
4461 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4467 err = netdev_set_master(slave, master);
4471 slave->flags |= IFF_SLAVE;
4473 slave->flags &= ~IFF_SLAVE;
4475 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4478 EXPORT_SYMBOL(netdev_set_bond_master);
4480 static void dev_change_rx_flags(struct net_device *dev, int flags)
4482 const struct net_device_ops *ops = dev->netdev_ops;
4484 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4485 ops->ndo_change_rx_flags(dev, flags);
4488 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4490 unsigned short old_flags = dev->flags;
4496 dev->flags |= IFF_PROMISC;
4497 dev->promiscuity += inc;
4498 if (dev->promiscuity == 0) {
4501 * If inc causes overflow, untouch promisc and return error.
4504 dev->flags &= ~IFF_PROMISC;
4506 dev->promiscuity -= inc;
4507 printk(KERN_WARNING "%s: promiscuity touches roof, "
4508 "set promiscuity failed, promiscuity feature "
4509 "of device might be broken.\n", dev->name);
4513 if (dev->flags != old_flags) {
4514 printk(KERN_INFO "device %s %s promiscuous mode\n",
4515 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4517 if (audit_enabled) {
4518 current_uid_gid(&uid, &gid);
4519 audit_log(current->audit_context, GFP_ATOMIC,
4520 AUDIT_ANOM_PROMISCUOUS,
4521 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4522 dev->name, (dev->flags & IFF_PROMISC),
4523 (old_flags & IFF_PROMISC),
4524 audit_get_loginuid(current),
4526 audit_get_sessionid(current));
4529 dev_change_rx_flags(dev, IFF_PROMISC);
4535 * dev_set_promiscuity - update promiscuity count on a device
4539 * Add or remove promiscuity from a device. While the count in the device
4540 * remains above zero the interface remains promiscuous. Once it hits zero
4541 * the device reverts back to normal filtering operation. A negative inc
4542 * value is used to drop promiscuity on the device.
4543 * Return 0 if successful or a negative errno code on error.
4545 int dev_set_promiscuity(struct net_device *dev, int inc)
4547 unsigned short old_flags = dev->flags;
4550 err = __dev_set_promiscuity(dev, inc);
4553 if (dev->flags != old_flags)
4554 dev_set_rx_mode(dev);
4557 EXPORT_SYMBOL(dev_set_promiscuity);
4560 * dev_set_allmulti - update allmulti count on a device
4564 * Add or remove reception of all multicast frames to a device. While the
4565 * count in the device remains above zero the interface remains listening
4566 * to all interfaces. Once it hits zero the device reverts back to normal
4567 * filtering operation. A negative @inc value is used to drop the counter
4568 * when releasing a resource needing all multicasts.
4569 * Return 0 if successful or a negative errno code on error.
4572 int dev_set_allmulti(struct net_device *dev, int inc)
4574 unsigned short old_flags = dev->flags;
4578 dev->flags |= IFF_ALLMULTI;
4579 dev->allmulti += inc;
4580 if (dev->allmulti == 0) {
4583 * If inc causes overflow, untouch allmulti and return error.
4586 dev->flags &= ~IFF_ALLMULTI;
4588 dev->allmulti -= inc;
4589 printk(KERN_WARNING "%s: allmulti touches roof, "
4590 "set allmulti failed, allmulti feature of "
4591 "device might be broken.\n", dev->name);
4595 if (dev->flags ^ old_flags) {
4596 dev_change_rx_flags(dev, IFF_ALLMULTI);
4597 dev_set_rx_mode(dev);
4601 EXPORT_SYMBOL(dev_set_allmulti);
4604 * Upload unicast and multicast address lists to device and
4605 * configure RX filtering. When the device doesn't support unicast
4606 * filtering it is put in promiscuous mode while unicast addresses
4609 void __dev_set_rx_mode(struct net_device *dev)
4611 const struct net_device_ops *ops = dev->netdev_ops;
4613 /* dev_open will call this function so the list will stay sane. */
4614 if (!(dev->flags&IFF_UP))
4617 if (!netif_device_present(dev))
4620 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4621 /* Unicast addresses changes may only happen under the rtnl,
4622 * therefore calling __dev_set_promiscuity here is safe.
4624 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4625 __dev_set_promiscuity(dev, 1);
4626 dev->uc_promisc = true;
4627 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4628 __dev_set_promiscuity(dev, -1);
4629 dev->uc_promisc = false;
4633 if (ops->ndo_set_rx_mode)
4634 ops->ndo_set_rx_mode(dev);
4637 void dev_set_rx_mode(struct net_device *dev)
4639 netif_addr_lock_bh(dev);
4640 __dev_set_rx_mode(dev);
4641 netif_addr_unlock_bh(dev);
4645 * dev_get_flags - get flags reported to userspace
4648 * Get the combination of flag bits exported through APIs to userspace.
4650 unsigned dev_get_flags(const struct net_device *dev)
4654 flags = (dev->flags & ~(IFF_PROMISC |
4659 (dev->gflags & (IFF_PROMISC |
4662 if (netif_running(dev)) {
4663 if (netif_oper_up(dev))
4664 flags |= IFF_RUNNING;
4665 if (netif_carrier_ok(dev))
4666 flags |= IFF_LOWER_UP;
4667 if (netif_dormant(dev))
4668 flags |= IFF_DORMANT;
4673 EXPORT_SYMBOL(dev_get_flags);
4675 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4677 int old_flags = dev->flags;
4683 * Set the flags on our device.
4686 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4687 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4689 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4693 * Load in the correct multicast list now the flags have changed.
4696 if ((old_flags ^ flags) & IFF_MULTICAST)
4697 dev_change_rx_flags(dev, IFF_MULTICAST);
4699 dev_set_rx_mode(dev);
4702 * Have we downed the interface. We handle IFF_UP ourselves
4703 * according to user attempts to set it, rather than blindly
4708 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4709 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4712 dev_set_rx_mode(dev);
4715 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4716 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4718 dev->gflags ^= IFF_PROMISC;
4719 dev_set_promiscuity(dev, inc);
4722 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4723 is important. Some (broken) drivers set IFF_PROMISC, when
4724 IFF_ALLMULTI is requested not asking us and not reporting.
4726 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4727 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4729 dev->gflags ^= IFF_ALLMULTI;
4730 dev_set_allmulti(dev, inc);
4736 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4738 unsigned int changes = dev->flags ^ old_flags;
4740 if (changes & IFF_UP) {
4741 if (dev->flags & IFF_UP)
4742 call_netdevice_notifiers(NETDEV_UP, dev);
4744 call_netdevice_notifiers(NETDEV_DOWN, dev);
4747 if (dev->flags & IFF_UP &&
4748 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4749 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4753 * dev_change_flags - change device settings
4755 * @flags: device state flags
4757 * Change settings on device based state flags. The flags are
4758 * in the userspace exported format.
4760 int dev_change_flags(struct net_device *dev, unsigned flags)
4763 int old_flags = dev->flags;
4765 ret = __dev_change_flags(dev, flags);
4769 changes = old_flags ^ dev->flags;
4771 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4773 __dev_notify_flags(dev, old_flags);
4776 EXPORT_SYMBOL(dev_change_flags);
4779 * dev_set_mtu - Change maximum transfer unit
4781 * @new_mtu: new transfer unit
4783 * Change the maximum transfer size of the network device.
4785 int dev_set_mtu(struct net_device *dev, int new_mtu)
4787 const struct net_device_ops *ops = dev->netdev_ops;
4790 if (new_mtu == dev->mtu)
4793 /* MTU must be positive. */
4797 if (!netif_device_present(dev))
4801 if (ops->ndo_change_mtu)
4802 err = ops->ndo_change_mtu(dev, new_mtu);
4806 if (!err && dev->flags & IFF_UP)
4807 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4810 EXPORT_SYMBOL(dev_set_mtu);
4813 * dev_set_group - Change group this device belongs to
4815 * @new_group: group this device should belong to
4817 void dev_set_group(struct net_device *dev, int new_group)
4819 dev->group = new_group;
4821 EXPORT_SYMBOL(dev_set_group);
4824 * dev_set_mac_address - Change Media Access Control Address
4828 * Change the hardware (MAC) address of the device
4830 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4832 const struct net_device_ops *ops = dev->netdev_ops;
4835 if (!ops->ndo_set_mac_address)
4837 if (sa->sa_family != dev->type)
4839 if (!netif_device_present(dev))
4841 err = ops->ndo_set_mac_address(dev, sa);
4843 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4846 EXPORT_SYMBOL(dev_set_mac_address);
4849 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4851 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4854 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4860 case SIOCGIFFLAGS: /* Get interface flags */
4861 ifr->ifr_flags = (short) dev_get_flags(dev);
4864 case SIOCGIFMETRIC: /* Get the metric on the interface
4865 (currently unused) */
4866 ifr->ifr_metric = 0;
4869 case SIOCGIFMTU: /* Get the MTU of a device */
4870 ifr->ifr_mtu = dev->mtu;
4875 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4877 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4878 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4879 ifr->ifr_hwaddr.sa_family = dev->type;
4887 ifr->ifr_map.mem_start = dev->mem_start;
4888 ifr->ifr_map.mem_end = dev->mem_end;
4889 ifr->ifr_map.base_addr = dev->base_addr;
4890 ifr->ifr_map.irq = dev->irq;
4891 ifr->ifr_map.dma = dev->dma;
4892 ifr->ifr_map.port = dev->if_port;
4896 ifr->ifr_ifindex = dev->ifindex;
4900 ifr->ifr_qlen = dev->tx_queue_len;
4904 /* dev_ioctl() should ensure this case
4916 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4918 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4921 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4922 const struct net_device_ops *ops;
4927 ops = dev->netdev_ops;
4930 case SIOCSIFFLAGS: /* Set interface flags */
4931 return dev_change_flags(dev, ifr->ifr_flags);
4933 case SIOCSIFMETRIC: /* Set the metric on the interface
4934 (currently unused) */
4937 case SIOCSIFMTU: /* Set the MTU of a device */
4938 return dev_set_mtu(dev, ifr->ifr_mtu);
4941 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4943 case SIOCSIFHWBROADCAST:
4944 if (ifr->ifr_hwaddr.sa_family != dev->type)
4946 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4947 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4948 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4952 if (ops->ndo_set_config) {
4953 if (!netif_device_present(dev))
4955 return ops->ndo_set_config(dev, &ifr->ifr_map);
4960 if (!ops->ndo_set_rx_mode ||
4961 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4963 if (!netif_device_present(dev))
4965 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4968 if (!ops->ndo_set_rx_mode ||
4969 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4971 if (!netif_device_present(dev))
4973 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4976 if (ifr->ifr_qlen < 0)
4978 dev->tx_queue_len = ifr->ifr_qlen;
4982 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4983 return dev_change_name(dev, ifr->ifr_newname);
4986 err = net_hwtstamp_validate(ifr);
4992 * Unknown or private ioctl
4995 if ((cmd >= SIOCDEVPRIVATE &&
4996 cmd <= SIOCDEVPRIVATE + 15) ||
4997 cmd == SIOCBONDENSLAVE ||
4998 cmd == SIOCBONDRELEASE ||
4999 cmd == SIOCBONDSETHWADDR ||
5000 cmd == SIOCBONDSLAVEINFOQUERY ||
5001 cmd == SIOCBONDINFOQUERY ||
5002 cmd == SIOCBONDCHANGEACTIVE ||
5003 cmd == SIOCGMIIPHY ||
5004 cmd == SIOCGMIIREG ||
5005 cmd == SIOCSMIIREG ||
5006 cmd == SIOCBRADDIF ||
5007 cmd == SIOCBRDELIF ||
5008 cmd == SIOCSHWTSTAMP ||
5009 cmd == SIOCWANDEV) {
5011 if (ops->ndo_do_ioctl) {
5012 if (netif_device_present(dev))
5013 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5025 * This function handles all "interface"-type I/O control requests. The actual
5026 * 'doing' part of this is dev_ifsioc above.
5030 * dev_ioctl - network device ioctl
5031 * @net: the applicable net namespace
5032 * @cmd: command to issue
5033 * @arg: pointer to a struct ifreq in user space
5035 * Issue ioctl functions to devices. This is normally called by the
5036 * user space syscall interfaces but can sometimes be useful for
5037 * other purposes. The return value is the return from the syscall if
5038 * positive or a negative errno code on error.
5041 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5047 /* One special case: SIOCGIFCONF takes ifconf argument
5048 and requires shared lock, because it sleeps writing
5052 if (cmd == SIOCGIFCONF) {
5054 ret = dev_ifconf(net, (char __user *) arg);
5058 if (cmd == SIOCGIFNAME)
5059 return dev_ifname(net, (struct ifreq __user *)arg);
5061 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5064 ifr.ifr_name[IFNAMSIZ-1] = 0;
5066 colon = strchr(ifr.ifr_name, ':');
5071 * See which interface the caller is talking about.
5076 * These ioctl calls:
5077 * - can be done by all.
5078 * - atomic and do not require locking.
5089 dev_load(net, ifr.ifr_name);
5091 ret = dev_ifsioc_locked(net, &ifr, cmd);
5096 if (copy_to_user(arg, &ifr,
5097 sizeof(struct ifreq)))
5103 dev_load(net, ifr.ifr_name);
5105 ret = dev_ethtool(net, &ifr);
5110 if (copy_to_user(arg, &ifr,
5111 sizeof(struct ifreq)))
5117 * These ioctl calls:
5118 * - require superuser power.
5119 * - require strict serialization.
5125 if (!capable(CAP_NET_ADMIN))
5127 dev_load(net, ifr.ifr_name);
5129 ret = dev_ifsioc(net, &ifr, cmd);
5134 if (copy_to_user(arg, &ifr,
5135 sizeof(struct ifreq)))
5141 * These ioctl calls:
5142 * - require superuser power.
5143 * - require strict serialization.
5144 * - do not return a value
5154 case SIOCSIFHWBROADCAST:
5157 case SIOCBONDENSLAVE:
5158 case SIOCBONDRELEASE:
5159 case SIOCBONDSETHWADDR:
5160 case SIOCBONDCHANGEACTIVE:
5164 if (!capable(CAP_NET_ADMIN))
5167 case SIOCBONDSLAVEINFOQUERY:
5168 case SIOCBONDINFOQUERY:
5169 dev_load(net, ifr.ifr_name);
5171 ret = dev_ifsioc(net, &ifr, cmd);
5176 /* Get the per device memory space. We can add this but
5177 * currently do not support it */
5179 /* Set the per device memory buffer space.
5180 * Not applicable in our case */
5185 * Unknown or private ioctl.
5188 if (cmd == SIOCWANDEV ||
5189 (cmd >= SIOCDEVPRIVATE &&
5190 cmd <= SIOCDEVPRIVATE + 15)) {
5191 dev_load(net, ifr.ifr_name);
5193 ret = dev_ifsioc(net, &ifr, cmd);
5195 if (!ret && copy_to_user(arg, &ifr,
5196 sizeof(struct ifreq)))
5200 /* Take care of Wireless Extensions */
5201 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5202 return wext_handle_ioctl(net, &ifr, cmd, arg);
5209 * dev_new_index - allocate an ifindex
5210 * @net: the applicable net namespace
5212 * Returns a suitable unique value for a new device interface
5213 * number. The caller must hold the rtnl semaphore or the
5214 * dev_base_lock to be sure it remains unique.
5216 static int dev_new_index(struct net *net)
5222 if (!__dev_get_by_index(net, ifindex))
5227 /* Delayed registration/unregisteration */
5228 static LIST_HEAD(net_todo_list);
5230 static void net_set_todo(struct net_device *dev)
5232 list_add_tail(&dev->todo_list, &net_todo_list);
5235 static void rollback_registered_many(struct list_head *head)
5237 struct net_device *dev, *tmp;
5239 BUG_ON(dev_boot_phase);
5242 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5243 /* Some devices call without registering
5244 * for initialization unwind. Remove those
5245 * devices and proceed with the remaining.
5247 if (dev->reg_state == NETREG_UNINITIALIZED) {
5248 pr_debug("unregister_netdevice: device %s/%p never "
5249 "was registered\n", dev->name, dev);
5252 list_del(&dev->unreg_list);
5255 dev->dismantle = true;
5256 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5259 /* If device is running, close it first. */
5260 dev_close_many(head);
5262 list_for_each_entry(dev, head, unreg_list) {
5263 /* And unlink it from device chain. */
5264 unlist_netdevice(dev);
5266 dev->reg_state = NETREG_UNREGISTERING;
5271 list_for_each_entry(dev, head, unreg_list) {
5272 /* Shutdown queueing discipline. */
5276 /* Notify protocols, that we are about to destroy
5277 this device. They should clean all the things.
5279 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5281 if (!dev->rtnl_link_ops ||
5282 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5283 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5286 * Flush the unicast and multicast chains
5291 if (dev->netdev_ops->ndo_uninit)
5292 dev->netdev_ops->ndo_uninit(dev);
5294 /* Notifier chain MUST detach us from master device. */
5295 WARN_ON(dev->master);
5297 /* Remove entries from kobject tree */
5298 netdev_unregister_kobject(dev);
5301 /* Process any work delayed until the end of the batch */
5302 dev = list_first_entry(head, struct net_device, unreg_list);
5303 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5307 list_for_each_entry(dev, head, unreg_list)
5311 static void rollback_registered(struct net_device *dev)
5315 list_add(&dev->unreg_list, &single);
5316 rollback_registered_many(&single);
5320 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5322 /* Fix illegal checksum combinations */
5323 if ((features & NETIF_F_HW_CSUM) &&
5324 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5325 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5326 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5329 if ((features & NETIF_F_NO_CSUM) &&
5330 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5331 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5332 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5335 /* Fix illegal SG+CSUM combinations. */
5336 if ((features & NETIF_F_SG) &&
5337 !(features & NETIF_F_ALL_CSUM)) {
5339 "Dropping NETIF_F_SG since no checksum feature.\n");
5340 features &= ~NETIF_F_SG;
5343 /* TSO requires that SG is present as well. */
5344 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5345 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5346 features &= ~NETIF_F_ALL_TSO;
5349 /* TSO ECN requires that TSO is present as well. */
5350 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5351 features &= ~NETIF_F_TSO_ECN;
5353 /* Software GSO depends on SG. */
5354 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5355 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5356 features &= ~NETIF_F_GSO;
5359 /* UFO needs SG and checksumming */
5360 if (features & NETIF_F_UFO) {
5361 /* maybe split UFO into V4 and V6? */
5362 if (!((features & NETIF_F_GEN_CSUM) ||
5363 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5364 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5366 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5367 features &= ~NETIF_F_UFO;
5370 if (!(features & NETIF_F_SG)) {
5372 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5373 features &= ~NETIF_F_UFO;
5380 int __netdev_update_features(struct net_device *dev)
5387 features = netdev_get_wanted_features(dev);
5389 if (dev->netdev_ops->ndo_fix_features)
5390 features = dev->netdev_ops->ndo_fix_features(dev, features);
5392 /* driver might be less strict about feature dependencies */
5393 features = netdev_fix_features(dev, features);
5395 if (dev->features == features)
5398 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5399 dev->features, features);
5401 if (dev->netdev_ops->ndo_set_features)
5402 err = dev->netdev_ops->ndo_set_features(dev, features);
5404 if (unlikely(err < 0)) {
5406 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5407 err, features, dev->features);
5412 dev->features = features;
5418 * netdev_update_features - recalculate device features
5419 * @dev: the device to check
5421 * Recalculate dev->features set and send notifications if it
5422 * has changed. Should be called after driver or hardware dependent
5423 * conditions might have changed that influence the features.
5425 void netdev_update_features(struct net_device *dev)
5427 if (__netdev_update_features(dev))
5428 netdev_features_change(dev);
5430 EXPORT_SYMBOL(netdev_update_features);
5433 * netdev_change_features - recalculate device features
5434 * @dev: the device to check
5436 * Recalculate dev->features set and send notifications even
5437 * if they have not changed. Should be called instead of
5438 * netdev_update_features() if also dev->vlan_features might
5439 * have changed to allow the changes to be propagated to stacked
5442 void netdev_change_features(struct net_device *dev)
5444 __netdev_update_features(dev);
5445 netdev_features_change(dev);
5447 EXPORT_SYMBOL(netdev_change_features);
5450 * netif_stacked_transfer_operstate - transfer operstate
5451 * @rootdev: the root or lower level device to transfer state from
5452 * @dev: the device to transfer operstate to
5454 * Transfer operational state from root to device. This is normally
5455 * called when a stacking relationship exists between the root
5456 * device and the device(a leaf device).
5458 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5459 struct net_device *dev)
5461 if (rootdev->operstate == IF_OPER_DORMANT)
5462 netif_dormant_on(dev);
5464 netif_dormant_off(dev);
5466 if (netif_carrier_ok(rootdev)) {
5467 if (!netif_carrier_ok(dev))
5468 netif_carrier_on(dev);
5470 if (netif_carrier_ok(dev))
5471 netif_carrier_off(dev);
5474 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5477 static int netif_alloc_rx_queues(struct net_device *dev)
5479 unsigned int i, count = dev->num_rx_queues;
5480 struct netdev_rx_queue *rx;
5484 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5486 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5491 for (i = 0; i < count; i++)
5497 static void netdev_init_one_queue(struct net_device *dev,
5498 struct netdev_queue *queue, void *_unused)
5500 /* Initialize queue lock */
5501 spin_lock_init(&queue->_xmit_lock);
5502 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5503 queue->xmit_lock_owner = -1;
5504 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5508 static int netif_alloc_netdev_queues(struct net_device *dev)
5510 unsigned int count = dev->num_tx_queues;
5511 struct netdev_queue *tx;
5515 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5517 pr_err("netdev: Unable to allocate %u tx queues.\n",
5523 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5524 spin_lock_init(&dev->tx_global_lock);
5530 * register_netdevice - register a network device
5531 * @dev: device to register
5533 * Take a completed network device structure and add it to the kernel
5534 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5535 * chain. 0 is returned on success. A negative errno code is returned
5536 * on a failure to set up the device, or if the name is a duplicate.
5538 * Callers must hold the rtnl semaphore. You may want
5539 * register_netdev() instead of this.
5542 * The locking appears insufficient to guarantee two parallel registers
5543 * will not get the same name.
5546 int register_netdevice(struct net_device *dev)
5549 struct net *net = dev_net(dev);
5551 BUG_ON(dev_boot_phase);
5556 /* When net_device's are persistent, this will be fatal. */
5557 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5560 spin_lock_init(&dev->addr_list_lock);
5561 netdev_set_addr_lockdep_class(dev);
5565 ret = dev_get_valid_name(dev, dev->name);
5569 /* Init, if this function is available */
5570 if (dev->netdev_ops->ndo_init) {
5571 ret = dev->netdev_ops->ndo_init(dev);
5579 dev->ifindex = dev_new_index(net);
5580 if (dev->iflink == -1)
5581 dev->iflink = dev->ifindex;
5583 /* Transfer changeable features to wanted_features and enable
5584 * software offloads (GSO and GRO).
5586 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5587 dev->features |= NETIF_F_SOFT_FEATURES;
5588 dev->wanted_features = dev->features & dev->hw_features;
5590 /* Turn on no cache copy if HW is doing checksum */
5591 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5592 if ((dev->features & NETIF_F_ALL_CSUM) &&
5593 !(dev->features & NETIF_F_NO_CSUM)) {
5594 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5595 dev->features |= NETIF_F_NOCACHE_COPY;
5598 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5600 dev->vlan_features |= NETIF_F_HIGHDMA;
5602 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5603 ret = notifier_to_errno(ret);
5607 ret = netdev_register_kobject(dev);
5610 dev->reg_state = NETREG_REGISTERED;
5612 __netdev_update_features(dev);
5615 * Default initial state at registry is that the
5616 * device is present.
5619 set_bit(__LINK_STATE_PRESENT, &dev->state);
5621 dev_init_scheduler(dev);
5623 list_netdevice(dev);
5625 /* Notify protocols, that a new device appeared. */
5626 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5627 ret = notifier_to_errno(ret);
5629 rollback_registered(dev);
5630 dev->reg_state = NETREG_UNREGISTERED;
5633 * Prevent userspace races by waiting until the network
5634 * device is fully setup before sending notifications.
5636 if (!dev->rtnl_link_ops ||
5637 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5638 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5644 if (dev->netdev_ops->ndo_uninit)
5645 dev->netdev_ops->ndo_uninit(dev);
5648 EXPORT_SYMBOL(register_netdevice);
5651 * init_dummy_netdev - init a dummy network device for NAPI
5652 * @dev: device to init
5654 * This takes a network device structure and initialize the minimum
5655 * amount of fields so it can be used to schedule NAPI polls without
5656 * registering a full blown interface. This is to be used by drivers
5657 * that need to tie several hardware interfaces to a single NAPI
5658 * poll scheduler due to HW limitations.
5660 int init_dummy_netdev(struct net_device *dev)
5662 /* Clear everything. Note we don't initialize spinlocks
5663 * are they aren't supposed to be taken by any of the
5664 * NAPI code and this dummy netdev is supposed to be
5665 * only ever used for NAPI polls
5667 memset(dev, 0, sizeof(struct net_device));
5669 /* make sure we BUG if trying to hit standard
5670 * register/unregister code path
5672 dev->reg_state = NETREG_DUMMY;
5674 /* NAPI wants this */
5675 INIT_LIST_HEAD(&dev->napi_list);
5677 /* a dummy interface is started by default */
5678 set_bit(__LINK_STATE_PRESENT, &dev->state);
5679 set_bit(__LINK_STATE_START, &dev->state);
5681 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5682 * because users of this 'device' dont need to change
5688 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5692 * register_netdev - register a network device
5693 * @dev: device to register
5695 * Take a completed network device structure and add it to the kernel
5696 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5697 * chain. 0 is returned on success. A negative errno code is returned
5698 * on a failure to set up the device, or if the name is a duplicate.
5700 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5701 * and expands the device name if you passed a format string to
5704 int register_netdev(struct net_device *dev)
5709 err = register_netdevice(dev);
5713 EXPORT_SYMBOL(register_netdev);
5715 int netdev_refcnt_read(const struct net_device *dev)
5719 for_each_possible_cpu(i)
5720 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5723 EXPORT_SYMBOL(netdev_refcnt_read);
5726 * netdev_wait_allrefs - wait until all references are gone.
5728 * This is called when unregistering network devices.
5730 * Any protocol or device that holds a reference should register
5731 * for netdevice notification, and cleanup and put back the
5732 * reference if they receive an UNREGISTER event.
5733 * We can get stuck here if buggy protocols don't correctly
5736 static void netdev_wait_allrefs(struct net_device *dev)
5738 unsigned long rebroadcast_time, warning_time;
5741 linkwatch_forget_dev(dev);
5743 rebroadcast_time = warning_time = jiffies;
5744 refcnt = netdev_refcnt_read(dev);
5746 while (refcnt != 0) {
5747 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5750 /* Rebroadcast unregister notification */
5751 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5752 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5753 * should have already handle it the first time */
5755 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5757 /* We must not have linkwatch events
5758 * pending on unregister. If this
5759 * happens, we simply run the queue
5760 * unscheduled, resulting in a noop
5763 linkwatch_run_queue();
5768 rebroadcast_time = jiffies;
5773 refcnt = netdev_refcnt_read(dev);
5775 if (time_after(jiffies, warning_time + 10 * HZ)) {
5776 printk(KERN_EMERG "unregister_netdevice: "
5777 "waiting for %s to become free. Usage "
5780 warning_time = jiffies;
5789 * register_netdevice(x1);
5790 * register_netdevice(x2);
5792 * unregister_netdevice(y1);
5793 * unregister_netdevice(y2);
5799 * We are invoked by rtnl_unlock().
5800 * This allows us to deal with problems:
5801 * 1) We can delete sysfs objects which invoke hotplug
5802 * without deadlocking with linkwatch via keventd.
5803 * 2) Since we run with the RTNL semaphore not held, we can sleep
5804 * safely in order to wait for the netdev refcnt to drop to zero.
5806 * We must not return until all unregister events added during
5807 * the interval the lock was held have been completed.
5809 void netdev_run_todo(void)
5811 struct list_head list;
5813 /* Snapshot list, allow later requests */
5814 list_replace_init(&net_todo_list, &list);
5818 /* Wait for rcu callbacks to finish before attempting to drain
5819 * the device list. This usually avoids a 250ms wait.
5821 if (!list_empty(&list))
5824 while (!list_empty(&list)) {
5825 struct net_device *dev
5826 = list_first_entry(&list, struct net_device, todo_list);
5827 list_del(&dev->todo_list);
5829 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5830 printk(KERN_ERR "network todo '%s' but state %d\n",
5831 dev->name, dev->reg_state);
5836 dev->reg_state = NETREG_UNREGISTERED;
5838 on_each_cpu(flush_backlog, dev, 1);
5840 netdev_wait_allrefs(dev);
5843 BUG_ON(netdev_refcnt_read(dev));
5844 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5845 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5846 WARN_ON(dev->dn_ptr);
5848 if (dev->destructor)
5849 dev->destructor(dev);
5851 /* Free network device */
5852 kobject_put(&dev->dev.kobj);
5856 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5857 * fields in the same order, with only the type differing.
5859 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5860 const struct net_device_stats *netdev_stats)
5862 #if BITS_PER_LONG == 64
5863 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5864 memcpy(stats64, netdev_stats, sizeof(*stats64));
5866 size_t i, n = sizeof(*stats64) / sizeof(u64);
5867 const unsigned long *src = (const unsigned long *)netdev_stats;
5868 u64 *dst = (u64 *)stats64;
5870 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5871 sizeof(*stats64) / sizeof(u64));
5872 for (i = 0; i < n; i++)
5878 * dev_get_stats - get network device statistics
5879 * @dev: device to get statistics from
5880 * @storage: place to store stats
5882 * Get network statistics from device. Return @storage.
5883 * The device driver may provide its own method by setting
5884 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5885 * otherwise the internal statistics structure is used.
5887 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5888 struct rtnl_link_stats64 *storage)
5890 const struct net_device_ops *ops = dev->netdev_ops;
5892 if (ops->ndo_get_stats64) {
5893 memset(storage, 0, sizeof(*storage));
5894 ops->ndo_get_stats64(dev, storage);
5895 } else if (ops->ndo_get_stats) {
5896 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5898 netdev_stats_to_stats64(storage, &dev->stats);
5900 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5903 EXPORT_SYMBOL(dev_get_stats);
5905 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5907 struct netdev_queue *queue = dev_ingress_queue(dev);
5909 #ifdef CONFIG_NET_CLS_ACT
5912 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5915 netdev_init_one_queue(dev, queue, NULL);
5916 queue->qdisc = &noop_qdisc;
5917 queue->qdisc_sleeping = &noop_qdisc;
5918 rcu_assign_pointer(dev->ingress_queue, queue);
5924 * alloc_netdev_mqs - allocate network device
5925 * @sizeof_priv: size of private data to allocate space for
5926 * @name: device name format string
5927 * @setup: callback to initialize device
5928 * @txqs: the number of TX subqueues to allocate
5929 * @rxqs: the number of RX subqueues to allocate
5931 * Allocates a struct net_device with private data area for driver use
5932 * and performs basic initialization. Also allocates subquue structs
5933 * for each queue on the device.
5935 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5936 void (*setup)(struct net_device *),
5937 unsigned int txqs, unsigned int rxqs)
5939 struct net_device *dev;
5941 struct net_device *p;
5943 BUG_ON(strlen(name) >= sizeof(dev->name));
5946 pr_err("alloc_netdev: Unable to allocate device "
5947 "with zero queues.\n");
5953 pr_err("alloc_netdev: Unable to allocate device "
5954 "with zero RX queues.\n");
5959 alloc_size = sizeof(struct net_device);
5961 /* ensure 32-byte alignment of private area */
5962 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5963 alloc_size += sizeof_priv;
5965 /* ensure 32-byte alignment of whole construct */
5966 alloc_size += NETDEV_ALIGN - 1;
5968 p = kzalloc(alloc_size, GFP_KERNEL);
5970 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5974 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5975 dev->padded = (char *)dev - (char *)p;
5977 dev->pcpu_refcnt = alloc_percpu(int);
5978 if (!dev->pcpu_refcnt)
5981 if (dev_addr_init(dev))
5987 dev_net_set(dev, &init_net);
5989 dev->gso_max_size = GSO_MAX_SIZE;
5991 INIT_LIST_HEAD(&dev->napi_list);
5992 INIT_LIST_HEAD(&dev->unreg_list);
5993 INIT_LIST_HEAD(&dev->link_watch_list);
5994 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5997 dev->num_tx_queues = txqs;
5998 dev->real_num_tx_queues = txqs;
5999 if (netif_alloc_netdev_queues(dev))
6003 dev->num_rx_queues = rxqs;
6004 dev->real_num_rx_queues = rxqs;
6005 if (netif_alloc_rx_queues(dev))
6009 strcpy(dev->name, name);
6010 dev->group = INIT_NETDEV_GROUP;
6018 free_percpu(dev->pcpu_refcnt);
6028 EXPORT_SYMBOL(alloc_netdev_mqs);
6031 * free_netdev - free network device
6034 * This function does the last stage of destroying an allocated device
6035 * interface. The reference to the device object is released.
6036 * If this is the last reference then it will be freed.
6038 void free_netdev(struct net_device *dev)
6040 struct napi_struct *p, *n;
6042 release_net(dev_net(dev));
6049 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6051 /* Flush device addresses */
6052 dev_addr_flush(dev);
6054 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6057 free_percpu(dev->pcpu_refcnt);
6058 dev->pcpu_refcnt = NULL;
6060 /* Compatibility with error handling in drivers */
6061 if (dev->reg_state == NETREG_UNINITIALIZED) {
6062 kfree((char *)dev - dev->padded);
6066 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6067 dev->reg_state = NETREG_RELEASED;
6069 /* will free via device release */
6070 put_device(&dev->dev);
6072 EXPORT_SYMBOL(free_netdev);
6075 * synchronize_net - Synchronize with packet receive processing
6077 * Wait for packets currently being received to be done.
6078 * Does not block later packets from starting.
6080 void synchronize_net(void)
6083 if (rtnl_is_locked())
6084 synchronize_rcu_expedited();
6088 EXPORT_SYMBOL(synchronize_net);
6091 * unregister_netdevice_queue - remove device from the kernel
6095 * This function shuts down a device interface and removes it
6096 * from the kernel tables.
6097 * If head not NULL, device is queued to be unregistered later.
6099 * Callers must hold the rtnl semaphore. You may want
6100 * unregister_netdev() instead of this.
6103 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6108 list_move_tail(&dev->unreg_list, head);
6110 rollback_registered(dev);
6111 /* Finish processing unregister after unlock */
6115 EXPORT_SYMBOL(unregister_netdevice_queue);
6118 * unregister_netdevice_many - unregister many devices
6119 * @head: list of devices
6121 void unregister_netdevice_many(struct list_head *head)
6123 struct net_device *dev;
6125 if (!list_empty(head)) {
6126 rollback_registered_many(head);
6127 list_for_each_entry(dev, head, unreg_list)
6131 EXPORT_SYMBOL(unregister_netdevice_many);
6134 * unregister_netdev - remove device from the kernel
6137 * This function shuts down a device interface and removes it
6138 * from the kernel tables.
6140 * This is just a wrapper for unregister_netdevice that takes
6141 * the rtnl semaphore. In general you want to use this and not
6142 * unregister_netdevice.
6144 void unregister_netdev(struct net_device *dev)
6147 unregister_netdevice(dev);
6150 EXPORT_SYMBOL(unregister_netdev);
6153 * dev_change_net_namespace - move device to different nethost namespace
6155 * @net: network namespace
6156 * @pat: If not NULL name pattern to try if the current device name
6157 * is already taken in the destination network namespace.
6159 * This function shuts down a device interface and moves it
6160 * to a new network namespace. On success 0 is returned, on
6161 * a failure a netagive errno code is returned.
6163 * Callers must hold the rtnl semaphore.
6166 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6172 /* Don't allow namespace local devices to be moved. */
6174 if (dev->features & NETIF_F_NETNS_LOCAL)
6177 /* Ensure the device has been registrered */
6179 if (dev->reg_state != NETREG_REGISTERED)
6182 /* Get out if there is nothing todo */
6184 if (net_eq(dev_net(dev), net))
6187 /* Pick the destination device name, and ensure
6188 * we can use it in the destination network namespace.
6191 if (__dev_get_by_name(net, dev->name)) {
6192 /* We get here if we can't use the current device name */
6195 if (dev_get_valid_name(dev, pat) < 0)
6200 * And now a mini version of register_netdevice unregister_netdevice.
6203 /* If device is running close it first. */
6206 /* And unlink it from device chain */
6208 unlist_netdevice(dev);
6212 /* Shutdown queueing discipline. */
6215 /* Notify protocols, that we are about to destroy
6216 this device. They should clean all the things.
6218 Note that dev->reg_state stays at NETREG_REGISTERED.
6219 This is wanted because this way 8021q and macvlan know
6220 the device is just moving and can keep their slaves up.
6222 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6223 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6224 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6227 * Flush the unicast and multicast chains
6232 /* Actually switch the network namespace */
6233 dev_net_set(dev, net);
6235 /* If there is an ifindex conflict assign a new one */
6236 if (__dev_get_by_index(net, dev->ifindex)) {
6237 int iflink = (dev->iflink == dev->ifindex);
6238 dev->ifindex = dev_new_index(net);
6240 dev->iflink = dev->ifindex;
6243 /* Fixup kobjects */
6244 err = device_rename(&dev->dev, dev->name);
6247 /* Add the device back in the hashes */
6248 list_netdevice(dev);
6250 /* Notify protocols, that a new device appeared. */
6251 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6254 * Prevent userspace races by waiting until the network
6255 * device is fully setup before sending notifications.
6257 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6264 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6266 static int dev_cpu_callback(struct notifier_block *nfb,
6267 unsigned long action,
6270 struct sk_buff **list_skb;
6271 struct sk_buff *skb;
6272 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6273 struct softnet_data *sd, *oldsd;
6275 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6278 local_irq_disable();
6279 cpu = smp_processor_id();
6280 sd = &per_cpu(softnet_data, cpu);
6281 oldsd = &per_cpu(softnet_data, oldcpu);
6283 /* Find end of our completion_queue. */
6284 list_skb = &sd->completion_queue;
6286 list_skb = &(*list_skb)->next;
6287 /* Append completion queue from offline CPU. */
6288 *list_skb = oldsd->completion_queue;
6289 oldsd->completion_queue = NULL;
6291 /* Append output queue from offline CPU. */
6292 if (oldsd->output_queue) {
6293 *sd->output_queue_tailp = oldsd->output_queue;
6294 sd->output_queue_tailp = oldsd->output_queue_tailp;
6295 oldsd->output_queue = NULL;
6296 oldsd->output_queue_tailp = &oldsd->output_queue;
6298 /* Append NAPI poll list from offline CPU. */
6299 if (!list_empty(&oldsd->poll_list)) {
6300 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6301 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6304 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6307 /* Process offline CPU's input_pkt_queue */
6308 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6310 input_queue_head_incr(oldsd);
6312 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6314 input_queue_head_incr(oldsd);
6322 * netdev_increment_features - increment feature set by one
6323 * @all: current feature set
6324 * @one: new feature set
6325 * @mask: mask feature set
6327 * Computes a new feature set after adding a device with feature set
6328 * @one to the master device with current feature set @all. Will not
6329 * enable anything that is off in @mask. Returns the new feature set.
6331 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6333 if (mask & NETIF_F_GEN_CSUM)
6334 mask |= NETIF_F_ALL_CSUM;
6335 mask |= NETIF_F_VLAN_CHALLENGED;
6337 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6338 all &= one | ~NETIF_F_ALL_FOR_ALL;
6340 /* If device needs checksumming, downgrade to it. */
6341 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6342 all &= ~NETIF_F_NO_CSUM;
6344 /* If one device supports hw checksumming, set for all. */
6345 if (all & NETIF_F_GEN_CSUM)
6346 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6350 EXPORT_SYMBOL(netdev_increment_features);
6352 static struct hlist_head *netdev_create_hash(void)
6355 struct hlist_head *hash;
6357 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6359 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6360 INIT_HLIST_HEAD(&hash[i]);
6365 /* Initialize per network namespace state */
6366 static int __net_init netdev_init(struct net *net)
6368 INIT_LIST_HEAD(&net->dev_base_head);
6370 net->dev_name_head = netdev_create_hash();
6371 if (net->dev_name_head == NULL)
6374 net->dev_index_head = netdev_create_hash();
6375 if (net->dev_index_head == NULL)
6381 kfree(net->dev_name_head);
6387 * netdev_drivername - network driver for the device
6388 * @dev: network device
6390 * Determine network driver for device.
6392 const char *netdev_drivername(const struct net_device *dev)
6394 const struct device_driver *driver;
6395 const struct device *parent;
6396 const char *empty = "";
6398 parent = dev->dev.parent;
6402 driver = parent->driver;
6403 if (driver && driver->name)
6404 return driver->name;
6408 int __netdev_printk(const char *level, const struct net_device *dev,
6409 struct va_format *vaf)
6413 if (dev && dev->dev.parent)
6414 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6415 netdev_name(dev), vaf);
6417 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6419 r = printk("%s(NULL net_device): %pV", level, vaf);
6423 EXPORT_SYMBOL(__netdev_printk);
6425 int netdev_printk(const char *level, const struct net_device *dev,
6426 const char *format, ...)
6428 struct va_format vaf;
6432 va_start(args, format);
6437 r = __netdev_printk(level, dev, &vaf);
6442 EXPORT_SYMBOL(netdev_printk);
6444 #define define_netdev_printk_level(func, level) \
6445 int func(const struct net_device *dev, const char *fmt, ...) \
6448 struct va_format vaf; \
6451 va_start(args, fmt); \
6456 r = __netdev_printk(level, dev, &vaf); \
6461 EXPORT_SYMBOL(func);
6463 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6464 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6465 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6466 define_netdev_printk_level(netdev_err, KERN_ERR);
6467 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6468 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6469 define_netdev_printk_level(netdev_info, KERN_INFO);
6471 static void __net_exit netdev_exit(struct net *net)
6473 kfree(net->dev_name_head);
6474 kfree(net->dev_index_head);
6477 static struct pernet_operations __net_initdata netdev_net_ops = {
6478 .init = netdev_init,
6479 .exit = netdev_exit,
6482 static void __net_exit default_device_exit(struct net *net)
6484 struct net_device *dev, *aux;
6486 * Push all migratable network devices back to the
6487 * initial network namespace
6490 for_each_netdev_safe(net, dev, aux) {
6492 char fb_name[IFNAMSIZ];
6494 /* Ignore unmoveable devices (i.e. loopback) */
6495 if (dev->features & NETIF_F_NETNS_LOCAL)
6498 /* Leave virtual devices for the generic cleanup */
6499 if (dev->rtnl_link_ops)
6502 /* Push remaining network devices to init_net */
6503 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6504 err = dev_change_net_namespace(dev, &init_net, fb_name);
6506 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6507 __func__, dev->name, err);
6514 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6516 /* At exit all network devices most be removed from a network
6517 * namespace. Do this in the reverse order of registration.
6518 * Do this across as many network namespaces as possible to
6519 * improve batching efficiency.
6521 struct net_device *dev;
6523 LIST_HEAD(dev_kill_list);
6526 list_for_each_entry(net, net_list, exit_list) {
6527 for_each_netdev_reverse(net, dev) {
6528 if (dev->rtnl_link_ops)
6529 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6531 unregister_netdevice_queue(dev, &dev_kill_list);
6534 unregister_netdevice_many(&dev_kill_list);
6535 list_del(&dev_kill_list);
6539 static struct pernet_operations __net_initdata default_device_ops = {
6540 .exit = default_device_exit,
6541 .exit_batch = default_device_exit_batch,
6545 * Initialize the DEV module. At boot time this walks the device list and
6546 * unhooks any devices that fail to initialise (normally hardware not
6547 * present) and leaves us with a valid list of present and active devices.
6552 * This is called single threaded during boot, so no need
6553 * to take the rtnl semaphore.
6555 static int __init net_dev_init(void)
6557 int i, rc = -ENOMEM;
6559 BUG_ON(!dev_boot_phase);
6561 if (dev_proc_init())
6564 if (netdev_kobject_init())
6567 INIT_LIST_HEAD(&ptype_all);
6568 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6569 INIT_LIST_HEAD(&ptype_base[i]);
6571 if (register_pernet_subsys(&netdev_net_ops))
6575 * Initialise the packet receive queues.
6578 for_each_possible_cpu(i) {
6579 struct softnet_data *sd = &per_cpu(softnet_data, i);
6581 memset(sd, 0, sizeof(*sd));
6582 skb_queue_head_init(&sd->input_pkt_queue);
6583 skb_queue_head_init(&sd->process_queue);
6584 sd->completion_queue = NULL;
6585 INIT_LIST_HEAD(&sd->poll_list);
6586 sd->output_queue = NULL;
6587 sd->output_queue_tailp = &sd->output_queue;
6589 sd->csd.func = rps_trigger_softirq;
6595 sd->backlog.poll = process_backlog;
6596 sd->backlog.weight = weight_p;
6597 sd->backlog.gro_list = NULL;
6598 sd->backlog.gro_count = 0;
6603 /* The loopback device is special if any other network devices
6604 * is present in a network namespace the loopback device must
6605 * be present. Since we now dynamically allocate and free the
6606 * loopback device ensure this invariant is maintained by
6607 * keeping the loopback device as the first device on the
6608 * list of network devices. Ensuring the loopback devices
6609 * is the first device that appears and the last network device
6612 if (register_pernet_device(&loopback_net_ops))
6615 if (register_pernet_device(&default_device_ops))
6618 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6619 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6621 hotcpu_notifier(dev_cpu_callback, 0);
6629 subsys_initcall(net_dev_init);
6631 static int __init initialize_hashrnd(void)
6633 get_random_bytes(&hashrnd, sizeof(hashrnd));
6637 late_initcall_sync(initialize_hashrnd);