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)
1066 if (len >= IFALIASZ)
1071 kfree(dev->ifalias);
1072 dev->ifalias = NULL;
1077 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1080 dev->ifalias = new_ifalias;
1082 strlcpy(dev->ifalias, alias, len+1);
1088 * netdev_features_change - device changes features
1089 * @dev: device to cause notification
1091 * Called to indicate a device has changed features.
1093 void netdev_features_change(struct net_device *dev)
1095 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1097 EXPORT_SYMBOL(netdev_features_change);
1100 * netdev_state_change - device changes state
1101 * @dev: device to cause notification
1103 * Called to indicate a device has changed state. This function calls
1104 * the notifier chains for netdev_chain and sends a NEWLINK message
1105 * to the routing socket.
1107 void netdev_state_change(struct net_device *dev)
1109 if (dev->flags & IFF_UP) {
1110 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1111 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1114 EXPORT_SYMBOL(netdev_state_change);
1116 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1118 return call_netdevice_notifiers(event, dev);
1120 EXPORT_SYMBOL(netdev_bonding_change);
1123 * dev_load - load a network module
1124 * @net: the applicable net namespace
1125 * @name: name of interface
1127 * If a network interface is not present and the process has suitable
1128 * privileges this function loads the module. If module loading is not
1129 * available in this kernel then it becomes a nop.
1132 void dev_load(struct net *net, const char *name)
1134 struct net_device *dev;
1138 dev = dev_get_by_name_rcu(net, name);
1142 if (no_module && capable(CAP_NET_ADMIN))
1143 no_module = request_module("netdev-%s", name);
1144 if (no_module && capable(CAP_SYS_MODULE)) {
1145 if (!request_module("%s", name))
1146 pr_err("Loading kernel module for a network device "
1147 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1151 EXPORT_SYMBOL(dev_load);
1153 static int __dev_open(struct net_device *dev)
1155 const struct net_device_ops *ops = dev->netdev_ops;
1160 if (!netif_device_present(dev))
1163 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1164 ret = notifier_to_errno(ret);
1168 set_bit(__LINK_STATE_START, &dev->state);
1170 if (ops->ndo_validate_addr)
1171 ret = ops->ndo_validate_addr(dev);
1173 if (!ret && ops->ndo_open)
1174 ret = ops->ndo_open(dev);
1177 clear_bit(__LINK_STATE_START, &dev->state);
1179 dev->flags |= IFF_UP;
1180 net_dmaengine_get();
1181 dev_set_rx_mode(dev);
1183 add_device_randomness(dev->dev_addr, dev->addr_len);
1190 * dev_open - prepare an interface for use.
1191 * @dev: device to open
1193 * Takes a device from down to up state. The device's private open
1194 * function is invoked and then the multicast lists are loaded. Finally
1195 * the device is moved into the up state and a %NETDEV_UP message is
1196 * sent to the netdev notifier chain.
1198 * Calling this function on an active interface is a nop. On a failure
1199 * a negative errno code is returned.
1201 int dev_open(struct net_device *dev)
1205 if (dev->flags & IFF_UP)
1208 ret = __dev_open(dev);
1212 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1213 call_netdevice_notifiers(NETDEV_UP, dev);
1217 EXPORT_SYMBOL(dev_open);
1219 static int __dev_close_many(struct list_head *head)
1221 struct net_device *dev;
1226 list_for_each_entry(dev, head, unreg_list) {
1227 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1229 clear_bit(__LINK_STATE_START, &dev->state);
1231 /* Synchronize to scheduled poll. We cannot touch poll list, it
1232 * can be even on different cpu. So just clear netif_running().
1234 * dev->stop() will invoke napi_disable() on all of it's
1235 * napi_struct instances on this device.
1237 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1240 dev_deactivate_many(head);
1242 list_for_each_entry(dev, head, unreg_list) {
1243 const struct net_device_ops *ops = dev->netdev_ops;
1246 * Call the device specific close. This cannot fail.
1247 * Only if device is UP
1249 * We allow it to be called even after a DETACH hot-plug
1255 dev->flags &= ~IFF_UP;
1256 net_dmaengine_put();
1262 static int __dev_close(struct net_device *dev)
1267 list_add(&dev->unreg_list, &single);
1268 retval = __dev_close_many(&single);
1273 static int dev_close_many(struct list_head *head)
1275 struct net_device *dev, *tmp;
1276 LIST_HEAD(tmp_list);
1278 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1279 if (!(dev->flags & IFF_UP))
1280 list_move(&dev->unreg_list, &tmp_list);
1282 __dev_close_many(head);
1284 list_for_each_entry(dev, head, unreg_list) {
1285 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1286 call_netdevice_notifiers(NETDEV_DOWN, dev);
1289 /* rollback_registered_many needs the complete original list */
1290 list_splice(&tmp_list, head);
1295 * dev_close - shutdown an interface.
1296 * @dev: device to shutdown
1298 * This function moves an active device into down state. A
1299 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1300 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1303 int dev_close(struct net_device *dev)
1305 if (dev->flags & IFF_UP) {
1308 list_add(&dev->unreg_list, &single);
1309 dev_close_many(&single);
1314 EXPORT_SYMBOL(dev_close);
1318 * dev_disable_lro - disable Large Receive Offload on a device
1321 * Disable Large Receive Offload (LRO) on a net device. Must be
1322 * called under RTNL. This is needed if received packets may be
1323 * forwarded to another interface.
1325 void dev_disable_lro(struct net_device *dev)
1330 * If we're trying to disable lro on a vlan device
1331 * use the underlying physical device instead
1333 if (is_vlan_dev(dev))
1334 dev = vlan_dev_real_dev(dev);
1336 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1337 flags = dev->ethtool_ops->get_flags(dev);
1339 flags = ethtool_op_get_flags(dev);
1341 if (!(flags & ETH_FLAG_LRO))
1344 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1345 if (unlikely(dev->features & NETIF_F_LRO))
1346 netdev_WARN(dev, "failed to disable LRO!\n");
1348 EXPORT_SYMBOL(dev_disable_lro);
1351 static int dev_boot_phase = 1;
1354 * register_netdevice_notifier - register a network notifier block
1357 * Register a notifier to be called when network device events occur.
1358 * The notifier passed is linked into the kernel structures and must
1359 * not be reused until it has been unregistered. A negative errno code
1360 * is returned on a failure.
1362 * When registered all registration and up events are replayed
1363 * to the new notifier to allow device to have a race free
1364 * view of the network device list.
1367 int register_netdevice_notifier(struct notifier_block *nb)
1369 struct net_device *dev;
1370 struct net_device *last;
1375 err = raw_notifier_chain_register(&netdev_chain, nb);
1381 for_each_netdev(net, dev) {
1382 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1383 err = notifier_to_errno(err);
1387 if (!(dev->flags & IFF_UP))
1390 nb->notifier_call(nb, NETDEV_UP, dev);
1401 for_each_netdev(net, dev) {
1405 if (dev->flags & IFF_UP) {
1406 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1407 nb->notifier_call(nb, NETDEV_DOWN, dev);
1409 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1410 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1415 raw_notifier_chain_unregister(&netdev_chain, nb);
1418 EXPORT_SYMBOL(register_netdevice_notifier);
1421 * unregister_netdevice_notifier - unregister a network notifier block
1424 * Unregister a notifier previously registered by
1425 * register_netdevice_notifier(). The notifier is unlinked into the
1426 * kernel structures and may then be reused. A negative errno code
1427 * is returned on a failure.
1429 * After unregistering unregister and down device events are synthesized
1430 * for all devices on the device list to the removed notifier to remove
1431 * the need for special case cleanup code.
1434 int unregister_netdevice_notifier(struct notifier_block *nb)
1436 struct net_device *dev;
1441 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1446 for_each_netdev(net, dev) {
1447 if (dev->flags & IFF_UP) {
1448 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1449 nb->notifier_call(nb, NETDEV_DOWN, dev);
1451 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1452 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1459 EXPORT_SYMBOL(unregister_netdevice_notifier);
1462 * call_netdevice_notifiers - call all network notifier blocks
1463 * @val: value passed unmodified to notifier function
1464 * @dev: net_device pointer passed unmodified to notifier function
1466 * Call all network notifier blocks. Parameters and return value
1467 * are as for raw_notifier_call_chain().
1470 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1473 return raw_notifier_call_chain(&netdev_chain, val, dev);
1475 EXPORT_SYMBOL(call_netdevice_notifiers);
1477 /* When > 0 there are consumers of rx skb time stamps */
1478 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1480 void net_enable_timestamp(void)
1482 atomic_inc(&netstamp_needed);
1484 EXPORT_SYMBOL(net_enable_timestamp);
1486 void net_disable_timestamp(void)
1488 atomic_dec(&netstamp_needed);
1490 EXPORT_SYMBOL(net_disable_timestamp);
1492 static inline void net_timestamp_set(struct sk_buff *skb)
1494 if (atomic_read(&netstamp_needed))
1495 __net_timestamp(skb);
1497 skb->tstamp.tv64 = 0;
1500 static inline void net_timestamp_check(struct sk_buff *skb)
1502 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1503 __net_timestamp(skb);
1506 static int net_hwtstamp_validate(struct ifreq *ifr)
1508 struct hwtstamp_config cfg;
1509 enum hwtstamp_tx_types tx_type;
1510 enum hwtstamp_rx_filters rx_filter;
1511 int tx_type_valid = 0;
1512 int rx_filter_valid = 0;
1514 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1517 if (cfg.flags) /* reserved for future extensions */
1520 tx_type = cfg.tx_type;
1521 rx_filter = cfg.rx_filter;
1524 case HWTSTAMP_TX_OFF:
1525 case HWTSTAMP_TX_ON:
1526 case HWTSTAMP_TX_ONESTEP_SYNC:
1531 switch (rx_filter) {
1532 case HWTSTAMP_FILTER_NONE:
1533 case HWTSTAMP_FILTER_ALL:
1534 case HWTSTAMP_FILTER_SOME:
1535 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1536 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1537 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1538 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1539 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1540 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1541 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1542 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1543 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1544 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1545 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1546 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1547 rx_filter_valid = 1;
1551 if (!tx_type_valid || !rx_filter_valid)
1557 static inline bool is_skb_forwardable(struct net_device *dev,
1558 struct sk_buff *skb)
1562 if (!(dev->flags & IFF_UP))
1565 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1566 if (skb->len <= len)
1569 /* if TSO is enabled, we don't care about the length as the packet
1570 * could be forwarded without being segmented before
1572 if (skb_is_gso(skb))
1579 * dev_forward_skb - loopback an skb to another netif
1581 * @dev: destination network device
1582 * @skb: buffer to forward
1585 * NET_RX_SUCCESS (no congestion)
1586 * NET_RX_DROP (packet was dropped, but freed)
1588 * dev_forward_skb can be used for injecting an skb from the
1589 * start_xmit function of one device into the receive queue
1590 * of another device.
1592 * The receiving device may be in another namespace, so
1593 * we have to clear all information in the skb that could
1594 * impact namespace isolation.
1596 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1598 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1599 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1600 atomic_long_inc(&dev->rx_dropped);
1609 if (unlikely(!is_skb_forwardable(dev, skb))) {
1610 atomic_long_inc(&dev->rx_dropped);
1616 skb->tstamp.tv64 = 0;
1617 skb->pkt_type = PACKET_HOST;
1618 skb->protocol = eth_type_trans(skb, dev);
1622 return netif_rx(skb);
1624 EXPORT_SYMBOL_GPL(dev_forward_skb);
1626 static inline int deliver_skb(struct sk_buff *skb,
1627 struct packet_type *pt_prev,
1628 struct net_device *orig_dev)
1630 atomic_inc(&skb->users);
1631 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1634 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1636 if (!ptype->af_packet_priv || !skb->sk)
1639 if (ptype->id_match)
1640 return ptype->id_match(ptype, skb->sk);
1641 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1648 * Support routine. Sends outgoing frames to any network
1649 * taps currently in use.
1652 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1654 struct packet_type *ptype;
1655 struct sk_buff *skb2 = NULL;
1656 struct packet_type *pt_prev = NULL;
1659 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1660 /* Never send packets back to the socket
1661 * they originated from - MvS (miquels@drinkel.ow.org)
1663 if ((ptype->dev == dev || !ptype->dev) &&
1664 (!skb_loop_sk(ptype, skb))) {
1666 deliver_skb(skb2, pt_prev, skb->dev);
1671 skb2 = skb_clone(skb, GFP_ATOMIC);
1675 net_timestamp_set(skb2);
1677 /* skb->nh should be correctly
1678 set by sender, so that the second statement is
1679 just protection against buggy protocols.
1681 skb_reset_mac_header(skb2);
1683 if (skb_network_header(skb2) < skb2->data ||
1684 skb2->network_header > skb2->tail) {
1685 if (net_ratelimit())
1686 printk(KERN_CRIT "protocol %04x is "
1688 ntohs(skb2->protocol),
1690 skb_reset_network_header(skb2);
1693 skb2->transport_header = skb2->network_header;
1694 skb2->pkt_type = PACKET_OUTGOING;
1699 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1703 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1704 * @dev: Network device
1705 * @txq: number of queues available
1707 * If real_num_tx_queues is changed the tc mappings may no longer be
1708 * valid. To resolve this verify the tc mapping remains valid and if
1709 * not NULL the mapping. With no priorities mapping to this
1710 * offset/count pair it will no longer be used. In the worst case TC0
1711 * is invalid nothing can be done so disable priority mappings. If is
1712 * expected that drivers will fix this mapping if they can before
1713 * calling netif_set_real_num_tx_queues.
1715 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1718 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1720 /* If TC0 is invalidated disable TC mapping */
1721 if (tc->offset + tc->count > txq) {
1722 pr_warning("Number of in use tx queues changed "
1723 "invalidating tc mappings. Priority "
1724 "traffic classification disabled!\n");
1729 /* Invalidated prio to tc mappings set to TC0 */
1730 for (i = 1; i < TC_BITMASK + 1; i++) {
1731 int q = netdev_get_prio_tc_map(dev, i);
1733 tc = &dev->tc_to_txq[q];
1734 if (tc->offset + tc->count > txq) {
1735 pr_warning("Number of in use tx queues "
1736 "changed. Priority %i to tc "
1737 "mapping %i is no longer valid "
1738 "setting map to 0\n",
1740 netdev_set_prio_tc_map(dev, i, 0);
1746 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1747 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1749 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1753 if (txq < 1 || txq > dev->num_tx_queues)
1756 if (dev->reg_state == NETREG_REGISTERED ||
1757 dev->reg_state == NETREG_UNREGISTERING) {
1760 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1766 netif_setup_tc(dev, txq);
1768 if (txq < dev->real_num_tx_queues)
1769 qdisc_reset_all_tx_gt(dev, txq);
1772 dev->real_num_tx_queues = txq;
1775 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1779 * netif_set_real_num_rx_queues - set actual number of RX queues used
1780 * @dev: Network device
1781 * @rxq: Actual number of RX queues
1783 * This must be called either with the rtnl_lock held or before
1784 * registration of the net device. Returns 0 on success, or a
1785 * negative error code. If called before registration, it always
1788 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1792 if (rxq < 1 || rxq > dev->num_rx_queues)
1795 if (dev->reg_state == NETREG_REGISTERED) {
1798 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1804 dev->real_num_rx_queues = rxq;
1807 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1810 static inline void __netif_reschedule(struct Qdisc *q)
1812 struct softnet_data *sd;
1813 unsigned long flags;
1815 local_irq_save(flags);
1816 sd = &__get_cpu_var(softnet_data);
1817 q->next_sched = NULL;
1818 *sd->output_queue_tailp = q;
1819 sd->output_queue_tailp = &q->next_sched;
1820 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1821 local_irq_restore(flags);
1824 void __netif_schedule(struct Qdisc *q)
1826 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1827 __netif_reschedule(q);
1829 EXPORT_SYMBOL(__netif_schedule);
1831 void dev_kfree_skb_irq(struct sk_buff *skb)
1833 if (atomic_dec_and_test(&skb->users)) {
1834 struct softnet_data *sd;
1835 unsigned long flags;
1837 local_irq_save(flags);
1838 sd = &__get_cpu_var(softnet_data);
1839 skb->next = sd->completion_queue;
1840 sd->completion_queue = skb;
1841 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1842 local_irq_restore(flags);
1845 EXPORT_SYMBOL(dev_kfree_skb_irq);
1847 void dev_kfree_skb_any(struct sk_buff *skb)
1849 if (in_irq() || irqs_disabled())
1850 dev_kfree_skb_irq(skb);
1854 EXPORT_SYMBOL(dev_kfree_skb_any);
1858 * netif_device_detach - mark device as removed
1859 * @dev: network device
1861 * Mark device as removed from system and therefore no longer available.
1863 void netif_device_detach(struct net_device *dev)
1865 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1866 netif_running(dev)) {
1867 netif_tx_stop_all_queues(dev);
1870 EXPORT_SYMBOL(netif_device_detach);
1873 * netif_device_attach - mark device as attached
1874 * @dev: network device
1876 * Mark device as attached from system and restart if needed.
1878 void netif_device_attach(struct net_device *dev)
1880 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1881 netif_running(dev)) {
1882 netif_tx_wake_all_queues(dev);
1883 __netdev_watchdog_up(dev);
1886 EXPORT_SYMBOL(netif_device_attach);
1889 * Invalidate hardware checksum when packet is to be mangled, and
1890 * complete checksum manually on outgoing path.
1892 int skb_checksum_help(struct sk_buff *skb)
1895 int ret = 0, offset;
1897 if (skb->ip_summed == CHECKSUM_COMPLETE)
1898 goto out_set_summed;
1900 if (unlikely(skb_shinfo(skb)->gso_size)) {
1901 /* Let GSO fix up the checksum. */
1902 goto out_set_summed;
1905 offset = skb_checksum_start_offset(skb);
1906 BUG_ON(offset >= skb_headlen(skb));
1907 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1909 offset += skb->csum_offset;
1910 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1912 if (skb_cloned(skb) &&
1913 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1914 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1919 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1921 skb->ip_summed = CHECKSUM_NONE;
1925 EXPORT_SYMBOL(skb_checksum_help);
1928 * skb_gso_segment - Perform segmentation on skb.
1929 * @skb: buffer to segment
1930 * @features: features for the output path (see dev->features)
1932 * This function segments the given skb and returns a list of segments.
1934 * It may return NULL if the skb requires no segmentation. This is
1935 * only possible when GSO is used for verifying header integrity.
1937 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1939 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1940 struct packet_type *ptype;
1941 __be16 type = skb->protocol;
1942 int vlan_depth = ETH_HLEN;
1945 while (type == htons(ETH_P_8021Q)) {
1946 struct vlan_hdr *vh;
1948 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1949 return ERR_PTR(-EINVAL);
1951 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1952 type = vh->h_vlan_encapsulated_proto;
1953 vlan_depth += VLAN_HLEN;
1956 skb_reset_mac_header(skb);
1957 skb->mac_len = skb->network_header - skb->mac_header;
1958 __skb_pull(skb, skb->mac_len);
1960 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1961 struct net_device *dev = skb->dev;
1962 struct ethtool_drvinfo info = {};
1964 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1965 dev->ethtool_ops->get_drvinfo(dev, &info);
1967 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1968 info.driver, dev ? dev->features : 0L,
1969 skb->sk ? skb->sk->sk_route_caps : 0L,
1970 skb->len, skb->data_len, skb->ip_summed);
1972 if (skb_header_cloned(skb) &&
1973 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1974 return ERR_PTR(err);
1978 list_for_each_entry_rcu(ptype,
1979 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1980 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1981 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1982 err = ptype->gso_send_check(skb);
1983 segs = ERR_PTR(err);
1984 if (err || skb_gso_ok(skb, features))
1986 __skb_push(skb, (skb->data -
1987 skb_network_header(skb)));
1989 segs = ptype->gso_segment(skb, features);
1995 __skb_push(skb, skb->data - skb_mac_header(skb));
1999 EXPORT_SYMBOL(skb_gso_segment);
2001 /* Take action when hardware reception checksum errors are detected. */
2003 void netdev_rx_csum_fault(struct net_device *dev)
2005 if (net_ratelimit()) {
2006 printk(KERN_ERR "%s: hw csum failure.\n",
2007 dev ? dev->name : "<unknown>");
2011 EXPORT_SYMBOL(netdev_rx_csum_fault);
2014 /* Actually, we should eliminate this check as soon as we know, that:
2015 * 1. IOMMU is present and allows to map all the memory.
2016 * 2. No high memory really exists on this machine.
2019 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2021 #ifdef CONFIG_HIGHMEM
2023 if (!(dev->features & NETIF_F_HIGHDMA)) {
2024 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2025 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2026 if (PageHighMem(skb_frag_page(frag)))
2031 if (PCI_DMA_BUS_IS_PHYS) {
2032 struct device *pdev = dev->dev.parent;
2036 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2037 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2038 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2039 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2048 void (*destructor)(struct sk_buff *skb);
2051 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2053 static void dev_gso_skb_destructor(struct sk_buff *skb)
2055 struct dev_gso_cb *cb;
2058 struct sk_buff *nskb = skb->next;
2060 skb->next = nskb->next;
2063 } while (skb->next);
2065 cb = DEV_GSO_CB(skb);
2067 cb->destructor(skb);
2071 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2072 * @skb: buffer to segment
2073 * @features: device features as applicable to this skb
2075 * This function segments the given skb and stores the list of segments
2078 static int dev_gso_segment(struct sk_buff *skb, int features)
2080 struct sk_buff *segs;
2082 segs = skb_gso_segment(skb, features);
2084 /* Verifying header integrity only. */
2089 return PTR_ERR(segs);
2092 DEV_GSO_CB(skb)->destructor = skb->destructor;
2093 skb->destructor = dev_gso_skb_destructor;
2098 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2100 return ((features & NETIF_F_GEN_CSUM) ||
2101 ((features & NETIF_F_V4_CSUM) &&
2102 protocol == htons(ETH_P_IP)) ||
2103 ((features & NETIF_F_V6_CSUM) &&
2104 protocol == htons(ETH_P_IPV6)) ||
2105 ((features & NETIF_F_FCOE_CRC) &&
2106 protocol == htons(ETH_P_FCOE)));
2109 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2111 if (skb->ip_summed != CHECKSUM_NONE &&
2112 !can_checksum_protocol(features, protocol)) {
2113 features &= ~NETIF_F_ALL_CSUM;
2114 features &= ~NETIF_F_SG;
2115 } else if (illegal_highdma(skb->dev, skb)) {
2116 features &= ~NETIF_F_SG;
2122 u32 netif_skb_features(struct sk_buff *skb)
2124 __be16 protocol = skb->protocol;
2125 u32 features = skb->dev->features;
2127 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2128 features &= ~NETIF_F_GSO_MASK;
2130 if (protocol == htons(ETH_P_8021Q)) {
2131 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2132 protocol = veh->h_vlan_encapsulated_proto;
2133 } else if (!vlan_tx_tag_present(skb)) {
2134 return harmonize_features(skb, protocol, features);
2137 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2139 if (protocol != htons(ETH_P_8021Q)) {
2140 return harmonize_features(skb, protocol, features);
2142 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2143 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2144 return harmonize_features(skb, protocol, features);
2147 EXPORT_SYMBOL(netif_skb_features);
2150 * Returns true if either:
2151 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2152 * 2. skb is fragmented and the device does not support SG, or if
2153 * at least one of fragments is in highmem and device does not
2154 * support DMA from it.
2156 static inline int skb_needs_linearize(struct sk_buff *skb,
2159 return skb_is_nonlinear(skb) &&
2160 ((skb_has_frag_list(skb) &&
2161 !(features & NETIF_F_FRAGLIST)) ||
2162 (skb_shinfo(skb)->nr_frags &&
2163 !(features & NETIF_F_SG)));
2166 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2167 struct netdev_queue *txq)
2169 const struct net_device_ops *ops = dev->netdev_ops;
2170 int rc = NETDEV_TX_OK;
2171 unsigned int skb_len;
2173 if (likely(!skb->next)) {
2177 * If device doesn't need skb->dst, release it right now while
2178 * its hot in this cpu cache
2180 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2183 if (!list_empty(&ptype_all))
2184 dev_queue_xmit_nit(skb, dev);
2186 features = netif_skb_features(skb);
2188 if (vlan_tx_tag_present(skb) &&
2189 !(features & NETIF_F_HW_VLAN_TX)) {
2190 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2197 if (netif_needs_gso(skb, features)) {
2198 if (unlikely(dev_gso_segment(skb, features)))
2203 if (skb_needs_linearize(skb, features) &&
2204 __skb_linearize(skb))
2207 /* If packet is not checksummed and device does not
2208 * support checksumming for this protocol, complete
2209 * checksumming here.
2211 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2212 skb_set_transport_header(skb,
2213 skb_checksum_start_offset(skb));
2214 if (!(features & NETIF_F_ALL_CSUM) &&
2215 skb_checksum_help(skb))
2221 rc = ops->ndo_start_xmit(skb, dev);
2222 trace_net_dev_xmit(skb, rc, dev, skb_len);
2223 if (rc == NETDEV_TX_OK)
2224 txq_trans_update(txq);
2230 struct sk_buff *nskb = skb->next;
2232 skb->next = nskb->next;
2236 * If device doesn't need nskb->dst, release it right now while
2237 * its hot in this cpu cache
2239 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2242 skb_len = nskb->len;
2243 rc = ops->ndo_start_xmit(nskb, dev);
2244 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2245 if (unlikely(rc != NETDEV_TX_OK)) {
2246 if (rc & ~NETDEV_TX_MASK)
2247 goto out_kfree_gso_skb;
2248 nskb->next = skb->next;
2252 txq_trans_update(txq);
2253 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2254 return NETDEV_TX_BUSY;
2255 } while (skb->next);
2258 if (likely(skb->next == NULL))
2259 skb->destructor = DEV_GSO_CB(skb)->destructor;
2266 static u32 hashrnd __read_mostly;
2269 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2270 * to be used as a distribution range.
2272 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2273 unsigned int num_tx_queues)
2277 u16 qcount = num_tx_queues;
2279 if (skb_rx_queue_recorded(skb)) {
2280 hash = skb_get_rx_queue(skb);
2281 while (unlikely(hash >= num_tx_queues))
2282 hash -= num_tx_queues;
2287 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2288 qoffset = dev->tc_to_txq[tc].offset;
2289 qcount = dev->tc_to_txq[tc].count;
2292 if (skb->sk && skb->sk->sk_hash)
2293 hash = skb->sk->sk_hash;
2295 hash = (__force u16) skb->protocol;
2296 hash = jhash_1word(hash, hashrnd);
2298 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2300 EXPORT_SYMBOL(__skb_tx_hash);
2302 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2304 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2305 if (net_ratelimit()) {
2306 pr_warning("%s selects TX queue %d, but "
2307 "real number of TX queues is %d\n",
2308 dev->name, queue_index, dev->real_num_tx_queues);
2315 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2318 struct xps_dev_maps *dev_maps;
2319 struct xps_map *map;
2320 int queue_index = -1;
2323 dev_maps = rcu_dereference(dev->xps_maps);
2325 map = rcu_dereference(
2326 dev_maps->cpu_map[raw_smp_processor_id()]);
2329 queue_index = map->queues[0];
2332 if (skb->sk && skb->sk->sk_hash)
2333 hash = skb->sk->sk_hash;
2335 hash = (__force u16) skb->protocol ^
2337 hash = jhash_1word(hash, hashrnd);
2338 queue_index = map->queues[
2339 ((u64)hash * map->len) >> 32];
2341 if (unlikely(queue_index >= dev->real_num_tx_queues))
2353 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2354 struct sk_buff *skb)
2357 const struct net_device_ops *ops = dev->netdev_ops;
2359 if (dev->real_num_tx_queues == 1)
2361 else if (ops->ndo_select_queue) {
2362 queue_index = ops->ndo_select_queue(dev, skb);
2363 queue_index = dev_cap_txqueue(dev, queue_index);
2365 struct sock *sk = skb->sk;
2366 queue_index = sk_tx_queue_get(sk);
2368 if (queue_index < 0 || skb->ooo_okay ||
2369 queue_index >= dev->real_num_tx_queues) {
2370 int old_index = queue_index;
2372 queue_index = get_xps_queue(dev, skb);
2373 if (queue_index < 0)
2374 queue_index = skb_tx_hash(dev, skb);
2376 if (queue_index != old_index && sk) {
2377 struct dst_entry *dst =
2378 rcu_dereference_check(sk->sk_dst_cache, 1);
2380 if (dst && skb_dst(skb) == dst)
2381 sk_tx_queue_set(sk, queue_index);
2386 skb_set_queue_mapping(skb, queue_index);
2387 return netdev_get_tx_queue(dev, queue_index);
2390 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2391 struct net_device *dev,
2392 struct netdev_queue *txq)
2394 spinlock_t *root_lock = qdisc_lock(q);
2398 qdisc_skb_cb(skb)->pkt_len = skb->len;
2399 qdisc_calculate_pkt_len(skb, q);
2401 * Heuristic to force contended enqueues to serialize on a
2402 * separate lock before trying to get qdisc main lock.
2403 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2404 * and dequeue packets faster.
2406 contended = qdisc_is_running(q);
2407 if (unlikely(contended))
2408 spin_lock(&q->busylock);
2410 spin_lock(root_lock);
2411 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2414 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2415 qdisc_run_begin(q)) {
2417 * This is a work-conserving queue; there are no old skbs
2418 * waiting to be sent out; and the qdisc is not running -
2419 * xmit the skb directly.
2421 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2424 qdisc_bstats_update(q, skb);
2426 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2427 if (unlikely(contended)) {
2428 spin_unlock(&q->busylock);
2435 rc = NET_XMIT_SUCCESS;
2438 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2439 if (qdisc_run_begin(q)) {
2440 if (unlikely(contended)) {
2441 spin_unlock(&q->busylock);
2447 spin_unlock(root_lock);
2448 if (unlikely(contended))
2449 spin_unlock(&q->busylock);
2453 static DEFINE_PER_CPU(int, xmit_recursion);
2454 #define RECURSION_LIMIT 10
2457 * dev_queue_xmit - transmit a buffer
2458 * @skb: buffer to transmit
2460 * Queue a buffer for transmission to a network device. The caller must
2461 * have set the device and priority and built the buffer before calling
2462 * this function. The function can be called from an interrupt.
2464 * A negative errno code is returned on a failure. A success does not
2465 * guarantee the frame will be transmitted as it may be dropped due
2466 * to congestion or traffic shaping.
2468 * -----------------------------------------------------------------------------------
2469 * I notice this method can also return errors from the queue disciplines,
2470 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2473 * Regardless of the return value, the skb is consumed, so it is currently
2474 * difficult to retry a send to this method. (You can bump the ref count
2475 * before sending to hold a reference for retry if you are careful.)
2477 * When calling this method, interrupts MUST be enabled. This is because
2478 * the BH enable code must have IRQs enabled so that it will not deadlock.
2481 int dev_queue_xmit(struct sk_buff *skb)
2483 struct net_device *dev = skb->dev;
2484 struct netdev_queue *txq;
2488 /* Disable soft irqs for various locks below. Also
2489 * stops preemption for RCU.
2493 txq = dev_pick_tx(dev, skb);
2494 q = rcu_dereference_bh(txq->qdisc);
2496 #ifdef CONFIG_NET_CLS_ACT
2497 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2499 trace_net_dev_queue(skb);
2501 rc = __dev_xmit_skb(skb, q, dev, txq);
2505 /* The device has no queue. Common case for software devices:
2506 loopback, all the sorts of tunnels...
2508 Really, it is unlikely that netif_tx_lock protection is necessary
2509 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2511 However, it is possible, that they rely on protection
2514 Check this and shot the lock. It is not prone from deadlocks.
2515 Either shot noqueue qdisc, it is even simpler 8)
2517 if (dev->flags & IFF_UP) {
2518 int cpu = smp_processor_id(); /* ok because BHs are off */
2520 if (txq->xmit_lock_owner != cpu) {
2522 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2523 goto recursion_alert;
2525 HARD_TX_LOCK(dev, txq, cpu);
2527 if (!netif_tx_queue_stopped(txq)) {
2528 __this_cpu_inc(xmit_recursion);
2529 rc = dev_hard_start_xmit(skb, dev, txq);
2530 __this_cpu_dec(xmit_recursion);
2531 if (dev_xmit_complete(rc)) {
2532 HARD_TX_UNLOCK(dev, txq);
2536 HARD_TX_UNLOCK(dev, txq);
2537 if (net_ratelimit())
2538 printk(KERN_CRIT "Virtual device %s asks to "
2539 "queue packet!\n", dev->name);
2541 /* Recursion is detected! It is possible,
2545 if (net_ratelimit())
2546 printk(KERN_CRIT "Dead loop on virtual device "
2547 "%s, fix it urgently!\n", dev->name);
2552 rcu_read_unlock_bh();
2557 rcu_read_unlock_bh();
2560 EXPORT_SYMBOL(dev_queue_xmit);
2563 /*=======================================================================
2565 =======================================================================*/
2567 int netdev_max_backlog __read_mostly = 1000;
2568 int netdev_tstamp_prequeue __read_mostly = 1;
2569 int netdev_budget __read_mostly = 300;
2570 int weight_p __read_mostly = 64; /* old backlog weight */
2572 /* Called with irq disabled */
2573 static inline void ____napi_schedule(struct softnet_data *sd,
2574 struct napi_struct *napi)
2576 list_add_tail(&napi->poll_list, &sd->poll_list);
2577 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2581 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2582 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2583 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2584 * if hash is a canonical 4-tuple hash over transport ports.
2586 void __skb_get_rxhash(struct sk_buff *skb)
2588 int nhoff, hash = 0, poff;
2589 const struct ipv6hdr *ip6;
2590 const struct iphdr *ip;
2591 const struct vlan_hdr *vlan;
2600 nhoff = skb_network_offset(skb);
2601 proto = skb->protocol;
2605 case __constant_htons(ETH_P_IP):
2607 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2610 ip = (const struct iphdr *) (skb->data + nhoff);
2611 if (ip_is_fragment(ip))
2614 ip_proto = ip->protocol;
2615 addr1 = (__force u32) ip->saddr;
2616 addr2 = (__force u32) ip->daddr;
2617 nhoff += ip->ihl * 4;
2619 case __constant_htons(ETH_P_IPV6):
2621 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2624 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2625 ip_proto = ip6->nexthdr;
2626 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2627 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2630 case __constant_htons(ETH_P_8021Q):
2631 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2633 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2634 proto = vlan->h_vlan_encapsulated_proto;
2635 nhoff += sizeof(*vlan);
2637 case __constant_htons(ETH_P_PPP_SES):
2638 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2640 proto = *((__be16 *) (skb->data + nhoff +
2641 sizeof(struct pppoe_hdr)));
2642 nhoff += PPPOE_SES_HLEN;
2644 case __constant_htons(PPP_IP):
2646 case __constant_htons(PPP_IPV6):
2657 if (pskb_may_pull(skb, nhoff + 16)) {
2658 u8 *h = skb->data + nhoff;
2659 __be16 flags = *(__be16 *)h;
2662 * Only look inside GRE if version zero and no
2665 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2666 proto = *(__be16 *)(h + 2);
2668 if (flags & GRE_CSUM)
2670 if (flags & GRE_KEY)
2672 if (flags & GRE_SEQ)
2685 poff = proto_ports_offset(ip_proto);
2688 if (pskb_may_pull(skb, nhoff + 4)) {
2689 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2694 /* get a consistent hash (same value on both flow directions) */
2695 if (addr2 < addr1 ||
2697 ports.v16[1] < ports.v16[0])) {
2699 swap(ports.v16[0], ports.v16[1]);
2701 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2708 EXPORT_SYMBOL(__skb_get_rxhash);
2712 /* One global table that all flow-based protocols share. */
2713 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2714 EXPORT_SYMBOL(rps_sock_flow_table);
2716 static struct rps_dev_flow *
2717 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2718 struct rps_dev_flow *rflow, u16 next_cpu)
2720 if (next_cpu != RPS_NO_CPU) {
2721 #ifdef CONFIG_RFS_ACCEL
2722 struct netdev_rx_queue *rxqueue;
2723 struct rps_dev_flow_table *flow_table;
2724 struct rps_dev_flow *old_rflow;
2729 /* Should we steer this flow to a different hardware queue? */
2730 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2731 !(dev->features & NETIF_F_NTUPLE))
2733 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2734 if (rxq_index == skb_get_rx_queue(skb))
2737 rxqueue = dev->_rx + rxq_index;
2738 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2741 flow_id = skb->rxhash & flow_table->mask;
2742 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2743 rxq_index, flow_id);
2747 rflow = &flow_table->flows[flow_id];
2749 if (old_rflow->filter == rflow->filter)
2750 old_rflow->filter = RPS_NO_FILTER;
2754 per_cpu(softnet_data, next_cpu).input_queue_head;
2757 rflow->cpu = next_cpu;
2762 * get_rps_cpu is called from netif_receive_skb and returns the target
2763 * CPU from the RPS map of the receiving queue for a given skb.
2764 * rcu_read_lock must be held on entry.
2766 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2767 struct rps_dev_flow **rflowp)
2769 struct netdev_rx_queue *rxqueue;
2770 struct rps_map *map;
2771 struct rps_dev_flow_table *flow_table;
2772 struct rps_sock_flow_table *sock_flow_table;
2776 if (skb_rx_queue_recorded(skb)) {
2777 u16 index = skb_get_rx_queue(skb);
2778 if (unlikely(index >= dev->real_num_rx_queues)) {
2779 WARN_ONCE(dev->real_num_rx_queues > 1,
2780 "%s received packet on queue %u, but number "
2781 "of RX queues is %u\n",
2782 dev->name, index, dev->real_num_rx_queues);
2785 rxqueue = dev->_rx + index;
2789 map = rcu_dereference(rxqueue->rps_map);
2791 if (map->len == 1 &&
2792 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2793 tcpu = map->cpus[0];
2794 if (cpu_online(tcpu))
2798 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2802 skb_reset_network_header(skb);
2803 if (!skb_get_rxhash(skb))
2806 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2807 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2808 if (flow_table && sock_flow_table) {
2810 struct rps_dev_flow *rflow;
2812 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2815 next_cpu = sock_flow_table->ents[skb->rxhash &
2816 sock_flow_table->mask];
2819 * If the desired CPU (where last recvmsg was done) is
2820 * different from current CPU (one in the rx-queue flow
2821 * table entry), switch if one of the following holds:
2822 * - Current CPU is unset (equal to RPS_NO_CPU).
2823 * - Current CPU is offline.
2824 * - The current CPU's queue tail has advanced beyond the
2825 * last packet that was enqueued using this table entry.
2826 * This guarantees that all previous packets for the flow
2827 * have been dequeued, thus preserving in order delivery.
2829 if (unlikely(tcpu != next_cpu) &&
2830 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2831 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2832 rflow->last_qtail)) >= 0)) {
2834 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2837 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2845 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2847 if (cpu_online(tcpu)) {
2857 #ifdef CONFIG_RFS_ACCEL
2860 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2861 * @dev: Device on which the filter was set
2862 * @rxq_index: RX queue index
2863 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2864 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2866 * Drivers that implement ndo_rx_flow_steer() should periodically call
2867 * this function for each installed filter and remove the filters for
2868 * which it returns %true.
2870 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2871 u32 flow_id, u16 filter_id)
2873 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2874 struct rps_dev_flow_table *flow_table;
2875 struct rps_dev_flow *rflow;
2880 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2881 if (flow_table && flow_id <= flow_table->mask) {
2882 rflow = &flow_table->flows[flow_id];
2883 cpu = ACCESS_ONCE(rflow->cpu);
2884 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2885 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2886 rflow->last_qtail) <
2887 (int)(10 * flow_table->mask)))
2893 EXPORT_SYMBOL(rps_may_expire_flow);
2895 #endif /* CONFIG_RFS_ACCEL */
2897 /* Called from hardirq (IPI) context */
2898 static void rps_trigger_softirq(void *data)
2900 struct softnet_data *sd = data;
2902 ____napi_schedule(sd, &sd->backlog);
2906 #endif /* CONFIG_RPS */
2909 * Check if this softnet_data structure is another cpu one
2910 * If yes, queue it to our IPI list and return 1
2913 static int rps_ipi_queued(struct softnet_data *sd)
2916 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2919 sd->rps_ipi_next = mysd->rps_ipi_list;
2920 mysd->rps_ipi_list = sd;
2922 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2925 #endif /* CONFIG_RPS */
2930 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2931 * queue (may be a remote CPU queue).
2933 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2934 unsigned int *qtail)
2936 struct softnet_data *sd;
2937 unsigned long flags;
2939 sd = &per_cpu(softnet_data, cpu);
2941 local_irq_save(flags);
2944 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2945 if (skb_queue_len(&sd->input_pkt_queue)) {
2947 __skb_queue_tail(&sd->input_pkt_queue, skb);
2948 input_queue_tail_incr_save(sd, qtail);
2950 local_irq_restore(flags);
2951 return NET_RX_SUCCESS;
2954 /* Schedule NAPI for backlog device
2955 * We can use non atomic operation since we own the queue lock
2957 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2958 if (!rps_ipi_queued(sd))
2959 ____napi_schedule(sd, &sd->backlog);
2967 local_irq_restore(flags);
2969 atomic_long_inc(&skb->dev->rx_dropped);
2975 * netif_rx - post buffer to the network code
2976 * @skb: buffer to post
2978 * This function receives a packet from a device driver and queues it for
2979 * the upper (protocol) levels to process. It always succeeds. The buffer
2980 * may be dropped during processing for congestion control or by the
2984 * NET_RX_SUCCESS (no congestion)
2985 * NET_RX_DROP (packet was dropped)
2989 int netif_rx(struct sk_buff *skb)
2993 /* if netpoll wants it, pretend we never saw it */
2994 if (netpoll_rx(skb))
2997 if (netdev_tstamp_prequeue)
2998 net_timestamp_check(skb);
3000 trace_netif_rx(skb);
3003 struct rps_dev_flow voidflow, *rflow = &voidflow;
3009 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3011 cpu = smp_processor_id();
3013 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3021 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3027 EXPORT_SYMBOL(netif_rx);
3029 int netif_rx_ni(struct sk_buff *skb)
3034 err = netif_rx(skb);
3035 if (local_softirq_pending())
3041 EXPORT_SYMBOL(netif_rx_ni);
3043 static void net_tx_action(struct softirq_action *h)
3045 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3047 if (sd->completion_queue) {
3048 struct sk_buff *clist;
3050 local_irq_disable();
3051 clist = sd->completion_queue;
3052 sd->completion_queue = NULL;
3056 struct sk_buff *skb = clist;
3057 clist = clist->next;
3059 WARN_ON(atomic_read(&skb->users));
3060 trace_kfree_skb(skb, net_tx_action);
3065 if (sd->output_queue) {
3068 local_irq_disable();
3069 head = sd->output_queue;
3070 sd->output_queue = NULL;
3071 sd->output_queue_tailp = &sd->output_queue;
3075 struct Qdisc *q = head;
3076 spinlock_t *root_lock;
3078 head = head->next_sched;
3080 root_lock = qdisc_lock(q);
3081 if (spin_trylock(root_lock)) {
3082 smp_mb__before_clear_bit();
3083 clear_bit(__QDISC_STATE_SCHED,
3086 spin_unlock(root_lock);
3088 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3090 __netif_reschedule(q);
3092 smp_mb__before_clear_bit();
3093 clear_bit(__QDISC_STATE_SCHED,
3101 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3102 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3103 /* This hook is defined here for ATM LANE */
3104 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3105 unsigned char *addr) __read_mostly;
3106 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3109 #ifdef CONFIG_NET_CLS_ACT
3110 /* TODO: Maybe we should just force sch_ingress to be compiled in
3111 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3112 * a compare and 2 stores extra right now if we dont have it on
3113 * but have CONFIG_NET_CLS_ACT
3114 * NOTE: This doesn't stop any functionality; if you dont have
3115 * the ingress scheduler, you just can't add policies on ingress.
3118 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3120 struct net_device *dev = skb->dev;
3121 u32 ttl = G_TC_RTTL(skb->tc_verd);
3122 int result = TC_ACT_OK;
3125 if (unlikely(MAX_RED_LOOP < ttl++)) {
3126 if (net_ratelimit())
3127 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3128 skb->skb_iif, dev->ifindex);
3132 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3133 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3136 if (q != &noop_qdisc) {
3137 spin_lock(qdisc_lock(q));
3138 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3139 result = qdisc_enqueue_root(skb, q);
3140 spin_unlock(qdisc_lock(q));
3146 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3147 struct packet_type **pt_prev,
3148 int *ret, struct net_device *orig_dev)
3150 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3152 if (!rxq || rxq->qdisc == &noop_qdisc)
3156 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3160 switch (ing_filter(skb, rxq)) {
3174 * netdev_rx_handler_register - register receive handler
3175 * @dev: device to register a handler for
3176 * @rx_handler: receive handler to register
3177 * @rx_handler_data: data pointer that is used by rx handler
3179 * Register a receive hander for a device. This handler will then be
3180 * called from __netif_receive_skb. A negative errno code is returned
3183 * The caller must hold the rtnl_mutex.
3185 * For a general description of rx_handler, see enum rx_handler_result.
3187 int netdev_rx_handler_register(struct net_device *dev,
3188 rx_handler_func_t *rx_handler,
3189 void *rx_handler_data)
3193 if (dev->rx_handler)
3196 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3197 rcu_assign_pointer(dev->rx_handler, rx_handler);
3201 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3204 * netdev_rx_handler_unregister - unregister receive handler
3205 * @dev: device to unregister a handler from
3207 * Unregister a receive hander from a device.
3209 * The caller must hold the rtnl_mutex.
3211 void netdev_rx_handler_unregister(struct net_device *dev)
3215 RCU_INIT_POINTER(dev->rx_handler, NULL);
3216 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3218 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3220 static int __netif_receive_skb(struct sk_buff *skb)
3222 struct packet_type *ptype, *pt_prev;
3223 rx_handler_func_t *rx_handler;
3224 struct net_device *orig_dev;
3225 struct net_device *null_or_dev;
3226 bool deliver_exact = false;
3227 int ret = NET_RX_DROP;
3230 if (!netdev_tstamp_prequeue)
3231 net_timestamp_check(skb);
3233 trace_netif_receive_skb(skb);
3235 /* if we've gotten here through NAPI, check netpoll */
3236 if (netpoll_receive_skb(skb))
3240 skb->skb_iif = skb->dev->ifindex;
3241 orig_dev = skb->dev;
3243 skb_reset_network_header(skb);
3244 skb_reset_transport_header(skb);
3245 skb_reset_mac_len(skb);
3253 __this_cpu_inc(softnet_data.processed);
3255 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3256 skb = vlan_untag(skb);
3261 #ifdef CONFIG_NET_CLS_ACT
3262 if (skb->tc_verd & TC_NCLS) {
3263 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3268 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3269 if (!ptype->dev || ptype->dev == skb->dev) {
3271 ret = deliver_skb(skb, pt_prev, orig_dev);
3276 #ifdef CONFIG_NET_CLS_ACT
3277 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3283 if (vlan_tx_tag_present(skb)) {
3285 ret = deliver_skb(skb, pt_prev, orig_dev);
3288 if (vlan_do_receive(&skb))
3290 else if (unlikely(!skb))
3294 rx_handler = rcu_dereference(skb->dev->rx_handler);
3297 ret = deliver_skb(skb, pt_prev, orig_dev);
3300 switch (rx_handler(&skb)) {
3301 case RX_HANDLER_CONSUMED:
3302 ret = NET_RX_SUCCESS;
3304 case RX_HANDLER_ANOTHER:
3306 case RX_HANDLER_EXACT:
3307 deliver_exact = true;
3308 case RX_HANDLER_PASS:
3315 if (vlan_tx_nonzero_tag_present(skb))
3316 skb->pkt_type = PACKET_OTHERHOST;
3318 /* deliver only exact match when indicated */
3319 null_or_dev = deliver_exact ? skb->dev : NULL;
3321 type = skb->protocol;
3322 list_for_each_entry_rcu(ptype,
3323 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3324 if (ptype->type == type &&
3325 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3326 ptype->dev == orig_dev)) {
3328 ret = deliver_skb(skb, pt_prev, orig_dev);
3334 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3336 atomic_long_inc(&skb->dev->rx_dropped);
3338 /* Jamal, now you will not able to escape explaining
3339 * me how you were going to use this. :-)
3350 * netif_receive_skb - process receive buffer from network
3351 * @skb: buffer to process
3353 * netif_receive_skb() is the main receive data processing function.
3354 * It always succeeds. The buffer may be dropped during processing
3355 * for congestion control or by the protocol layers.
3357 * This function may only be called from softirq context and interrupts
3358 * should be enabled.
3360 * Return values (usually ignored):
3361 * NET_RX_SUCCESS: no congestion
3362 * NET_RX_DROP: packet was dropped
3364 int netif_receive_skb(struct sk_buff *skb)
3366 if (netdev_tstamp_prequeue)
3367 net_timestamp_check(skb);
3369 if (skb_defer_rx_timestamp(skb))
3370 return NET_RX_SUCCESS;
3374 struct rps_dev_flow voidflow, *rflow = &voidflow;
3379 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3382 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3386 ret = __netif_receive_skb(skb);
3392 return __netif_receive_skb(skb);
3395 EXPORT_SYMBOL(netif_receive_skb);
3397 /* Network device is going away, flush any packets still pending
3398 * Called with irqs disabled.
3400 static void flush_backlog(void *arg)
3402 struct net_device *dev = arg;
3403 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3404 struct sk_buff *skb, *tmp;
3407 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3408 if (skb->dev == dev) {
3409 __skb_unlink(skb, &sd->input_pkt_queue);
3411 input_queue_head_incr(sd);
3416 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3417 if (skb->dev == dev) {
3418 __skb_unlink(skb, &sd->process_queue);
3420 input_queue_head_incr(sd);
3425 static int napi_gro_complete(struct sk_buff *skb)
3427 struct packet_type *ptype;
3428 __be16 type = skb->protocol;
3429 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3432 if (NAPI_GRO_CB(skb)->count == 1) {
3433 skb_shinfo(skb)->gso_size = 0;
3438 list_for_each_entry_rcu(ptype, head, list) {
3439 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3442 err = ptype->gro_complete(skb);
3448 WARN_ON(&ptype->list == head);
3450 return NET_RX_SUCCESS;
3454 return netif_receive_skb(skb);
3457 inline void napi_gro_flush(struct napi_struct *napi)
3459 struct sk_buff *skb, *next;
3461 for (skb = napi->gro_list; skb; skb = next) {
3464 napi_gro_complete(skb);
3467 napi->gro_count = 0;
3468 napi->gro_list = NULL;
3470 EXPORT_SYMBOL(napi_gro_flush);
3472 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3474 struct sk_buff **pp = NULL;
3475 struct packet_type *ptype;
3476 __be16 type = skb->protocol;
3477 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3480 enum gro_result ret;
3482 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3485 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3489 list_for_each_entry_rcu(ptype, head, list) {
3490 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3493 skb_set_network_header(skb, skb_gro_offset(skb));
3494 mac_len = skb->network_header - skb->mac_header;
3495 skb->mac_len = mac_len;
3496 NAPI_GRO_CB(skb)->same_flow = 0;
3497 NAPI_GRO_CB(skb)->flush = 0;
3498 NAPI_GRO_CB(skb)->free = 0;
3500 pp = ptype->gro_receive(&napi->gro_list, skb);
3505 if (&ptype->list == head)
3508 same_flow = NAPI_GRO_CB(skb)->same_flow;
3509 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3512 struct sk_buff *nskb = *pp;
3516 napi_gro_complete(nskb);
3523 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3527 NAPI_GRO_CB(skb)->count = 1;
3528 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3529 skb->next = napi->gro_list;
3530 napi->gro_list = skb;
3534 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3535 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3537 BUG_ON(skb->end - skb->tail < grow);
3539 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3542 skb->data_len -= grow;
3544 skb_shinfo(skb)->frags[0].page_offset += grow;
3545 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3547 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3548 skb_frag_unref(skb, 0);
3549 memmove(skb_shinfo(skb)->frags,
3550 skb_shinfo(skb)->frags + 1,
3551 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3562 EXPORT_SYMBOL(dev_gro_receive);
3564 static inline gro_result_t
3565 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3568 unsigned int maclen = skb->dev->hard_header_len;
3570 for (p = napi->gro_list; p; p = p->next) {
3571 unsigned long diffs;
3573 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3574 diffs |= p->vlan_tci ^ skb->vlan_tci;
3575 if (maclen == ETH_HLEN)
3576 diffs |= compare_ether_header(skb_mac_header(p),
3577 skb_gro_mac_header(skb));
3579 diffs = memcmp(skb_mac_header(p),
3580 skb_gro_mac_header(skb),
3582 NAPI_GRO_CB(p)->same_flow = !diffs;
3583 NAPI_GRO_CB(p)->flush = 0;
3586 return dev_gro_receive(napi, skb);
3589 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3593 if (netif_receive_skb(skb))
3598 case GRO_MERGED_FREE:
3609 EXPORT_SYMBOL(napi_skb_finish);
3611 void skb_gro_reset_offset(struct sk_buff *skb)
3613 NAPI_GRO_CB(skb)->data_offset = 0;
3614 NAPI_GRO_CB(skb)->frag0 = NULL;
3615 NAPI_GRO_CB(skb)->frag0_len = 0;
3617 if (skb->mac_header == skb->tail &&
3618 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3619 NAPI_GRO_CB(skb)->frag0 =
3620 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3621 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3624 EXPORT_SYMBOL(skb_gro_reset_offset);
3626 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3628 skb_gro_reset_offset(skb);
3630 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3632 EXPORT_SYMBOL(napi_gro_receive);
3634 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3636 __skb_pull(skb, skb_headlen(skb));
3637 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3638 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3640 skb->dev = napi->dev;
3646 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3648 struct sk_buff *skb = napi->skb;
3651 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3657 EXPORT_SYMBOL(napi_get_frags);
3659 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3665 skb->protocol = eth_type_trans(skb, skb->dev);
3667 if (ret == GRO_HELD)
3668 skb_gro_pull(skb, -ETH_HLEN);
3669 else if (netif_receive_skb(skb))
3674 case GRO_MERGED_FREE:
3675 napi_reuse_skb(napi, skb);
3684 EXPORT_SYMBOL(napi_frags_finish);
3686 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3688 struct sk_buff *skb = napi->skb;
3695 skb_reset_mac_header(skb);
3696 skb_gro_reset_offset(skb);
3698 off = skb_gro_offset(skb);
3699 hlen = off + sizeof(*eth);
3700 eth = skb_gro_header_fast(skb, off);
3701 if (skb_gro_header_hard(skb, hlen)) {
3702 eth = skb_gro_header_slow(skb, hlen, off);
3703 if (unlikely(!eth)) {
3704 napi_reuse_skb(napi, skb);
3710 skb_gro_pull(skb, sizeof(*eth));
3713 * This works because the only protocols we care about don't require
3714 * special handling. We'll fix it up properly at the end.
3716 skb->protocol = eth->h_proto;
3721 EXPORT_SYMBOL(napi_frags_skb);
3723 gro_result_t napi_gro_frags(struct napi_struct *napi)
3725 struct sk_buff *skb = napi_frags_skb(napi);
3730 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3732 EXPORT_SYMBOL(napi_gro_frags);
3735 * net_rps_action sends any pending IPI's for rps.
3736 * Note: called with local irq disabled, but exits with local irq enabled.
3738 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3741 struct softnet_data *remsd = sd->rps_ipi_list;
3744 sd->rps_ipi_list = NULL;
3748 /* Send pending IPI's to kick RPS processing on remote cpus. */
3750 struct softnet_data *next = remsd->rps_ipi_next;
3752 if (cpu_online(remsd->cpu))
3753 __smp_call_function_single(remsd->cpu,
3762 static int process_backlog(struct napi_struct *napi, int quota)
3765 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3768 /* Check if we have pending ipi, its better to send them now,
3769 * not waiting net_rx_action() end.
3771 if (sd->rps_ipi_list) {
3772 local_irq_disable();
3773 net_rps_action_and_irq_enable(sd);
3776 napi->weight = weight_p;
3777 local_irq_disable();
3778 while (work < quota) {
3779 struct sk_buff *skb;
3782 while ((skb = __skb_dequeue(&sd->process_queue))) {
3784 __netif_receive_skb(skb);
3785 local_irq_disable();
3786 input_queue_head_incr(sd);
3787 if (++work >= quota) {
3794 qlen = skb_queue_len(&sd->input_pkt_queue);
3796 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3797 &sd->process_queue);
3799 if (qlen < quota - work) {
3801 * Inline a custom version of __napi_complete().
3802 * only current cpu owns and manipulates this napi,
3803 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3804 * we can use a plain write instead of clear_bit(),
3805 * and we dont need an smp_mb() memory barrier.
3807 list_del(&napi->poll_list);
3810 quota = work + qlen;
3820 * __napi_schedule - schedule for receive
3821 * @n: entry to schedule
3823 * The entry's receive function will be scheduled to run
3825 void __napi_schedule(struct napi_struct *n)
3827 unsigned long flags;
3829 local_irq_save(flags);
3830 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3831 local_irq_restore(flags);
3833 EXPORT_SYMBOL(__napi_schedule);
3835 void __napi_complete(struct napi_struct *n)
3837 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3838 BUG_ON(n->gro_list);
3840 list_del(&n->poll_list);
3841 smp_mb__before_clear_bit();
3842 clear_bit(NAPI_STATE_SCHED, &n->state);
3844 EXPORT_SYMBOL(__napi_complete);
3846 void napi_complete(struct napi_struct *n)
3848 unsigned long flags;
3851 * don't let napi dequeue from the cpu poll list
3852 * just in case its running on a different cpu
3854 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3858 local_irq_save(flags);
3860 local_irq_restore(flags);
3862 EXPORT_SYMBOL(napi_complete);
3864 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3865 int (*poll)(struct napi_struct *, int), int weight)
3867 INIT_LIST_HEAD(&napi->poll_list);
3868 napi->gro_count = 0;
3869 napi->gro_list = NULL;
3872 napi->weight = weight;
3873 list_add(&napi->dev_list, &dev->napi_list);
3875 #ifdef CONFIG_NETPOLL
3876 spin_lock_init(&napi->poll_lock);
3877 napi->poll_owner = -1;
3879 set_bit(NAPI_STATE_SCHED, &napi->state);
3881 EXPORT_SYMBOL(netif_napi_add);
3883 void netif_napi_del(struct napi_struct *napi)
3885 struct sk_buff *skb, *next;
3887 list_del_init(&napi->dev_list);
3888 napi_free_frags(napi);
3890 for (skb = napi->gro_list; skb; skb = next) {
3896 napi->gro_list = NULL;
3897 napi->gro_count = 0;
3899 EXPORT_SYMBOL(netif_napi_del);
3901 static void net_rx_action(struct softirq_action *h)
3903 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3904 unsigned long time_limit = jiffies + 2;
3905 int budget = netdev_budget;
3908 local_irq_disable();
3910 while (!list_empty(&sd->poll_list)) {
3911 struct napi_struct *n;
3914 /* If softirq window is exhuasted then punt.
3915 * Allow this to run for 2 jiffies since which will allow
3916 * an average latency of 1.5/HZ.
3918 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3923 /* Even though interrupts have been re-enabled, this
3924 * access is safe because interrupts can only add new
3925 * entries to the tail of this list, and only ->poll()
3926 * calls can remove this head entry from the list.
3928 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3930 have = netpoll_poll_lock(n);
3934 /* This NAPI_STATE_SCHED test is for avoiding a race
3935 * with netpoll's poll_napi(). Only the entity which
3936 * obtains the lock and sees NAPI_STATE_SCHED set will
3937 * actually make the ->poll() call. Therefore we avoid
3938 * accidentally calling ->poll() when NAPI is not scheduled.
3941 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3942 work = n->poll(n, weight);
3946 WARN_ON_ONCE(work > weight);
3950 local_irq_disable();
3952 /* Drivers must not modify the NAPI state if they
3953 * consume the entire weight. In such cases this code
3954 * still "owns" the NAPI instance and therefore can
3955 * move the instance around on the list at-will.
3957 if (unlikely(work == weight)) {
3958 if (unlikely(napi_disable_pending(n))) {
3961 local_irq_disable();
3963 list_move_tail(&n->poll_list, &sd->poll_list);
3966 netpoll_poll_unlock(have);
3969 net_rps_action_and_irq_enable(sd);
3971 #ifdef CONFIG_NET_DMA
3973 * There may not be any more sk_buffs coming right now, so push
3974 * any pending DMA copies to hardware
3976 dma_issue_pending_all();
3983 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3987 static gifconf_func_t *gifconf_list[NPROTO];
3990 * register_gifconf - register a SIOCGIF handler
3991 * @family: Address family
3992 * @gifconf: Function handler
3994 * Register protocol dependent address dumping routines. The handler
3995 * that is passed must not be freed or reused until it has been replaced
3996 * by another handler.
3998 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4000 if (family >= NPROTO)
4002 gifconf_list[family] = gifconf;
4005 EXPORT_SYMBOL(register_gifconf);
4009 * Map an interface index to its name (SIOCGIFNAME)
4013 * We need this ioctl for efficient implementation of the
4014 * if_indextoname() function required by the IPv6 API. Without
4015 * it, we would have to search all the interfaces to find a
4019 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4021 struct net_device *dev;
4025 * Fetch the caller's info block.
4028 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4032 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4038 strcpy(ifr.ifr_name, dev->name);
4041 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4047 * Perform a SIOCGIFCONF call. This structure will change
4048 * size eventually, and there is nothing I can do about it.
4049 * Thus we will need a 'compatibility mode'.
4052 static int dev_ifconf(struct net *net, char __user *arg)
4055 struct net_device *dev;
4062 * Fetch the caller's info block.
4065 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4072 * Loop over the interfaces, and write an info block for each.
4076 for_each_netdev(net, dev) {
4077 for (i = 0; i < NPROTO; i++) {
4078 if (gifconf_list[i]) {
4081 done = gifconf_list[i](dev, NULL, 0);
4083 done = gifconf_list[i](dev, pos + total,
4093 * All done. Write the updated control block back to the caller.
4095 ifc.ifc_len = total;
4098 * Both BSD and Solaris return 0 here, so we do too.
4100 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4103 #ifdef CONFIG_PROC_FS
4105 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4107 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4108 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4109 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4111 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4113 struct net *net = seq_file_net(seq);
4114 struct net_device *dev;
4115 struct hlist_node *p;
4116 struct hlist_head *h;
4117 unsigned int count = 0, offset = get_offset(*pos);
4119 h = &net->dev_name_head[get_bucket(*pos)];
4120 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4121 if (++count == offset)
4128 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4130 struct net_device *dev;
4131 unsigned int bucket;
4134 dev = dev_from_same_bucket(seq, pos);
4138 bucket = get_bucket(*pos) + 1;
4139 *pos = set_bucket_offset(bucket, 1);
4140 } while (bucket < NETDEV_HASHENTRIES);
4146 * This is invoked by the /proc filesystem handler to display a device
4149 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4154 return SEQ_START_TOKEN;
4156 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4159 return dev_from_bucket(seq, pos);
4162 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4165 return dev_from_bucket(seq, pos);
4168 void dev_seq_stop(struct seq_file *seq, void *v)
4174 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4176 struct rtnl_link_stats64 temp;
4177 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4179 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4180 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4181 dev->name, stats->rx_bytes, stats->rx_packets,
4183 stats->rx_dropped + stats->rx_missed_errors,
4184 stats->rx_fifo_errors,
4185 stats->rx_length_errors + stats->rx_over_errors +
4186 stats->rx_crc_errors + stats->rx_frame_errors,
4187 stats->rx_compressed, stats->multicast,
4188 stats->tx_bytes, stats->tx_packets,
4189 stats->tx_errors, stats->tx_dropped,
4190 stats->tx_fifo_errors, stats->collisions,
4191 stats->tx_carrier_errors +
4192 stats->tx_aborted_errors +
4193 stats->tx_window_errors +
4194 stats->tx_heartbeat_errors,
4195 stats->tx_compressed);
4199 * Called from the PROCfs module. This now uses the new arbitrary sized
4200 * /proc/net interface to create /proc/net/dev
4202 static int dev_seq_show(struct seq_file *seq, void *v)
4204 if (v == SEQ_START_TOKEN)
4205 seq_puts(seq, "Inter-| Receive "
4207 " face |bytes packets errs drop fifo frame "
4208 "compressed multicast|bytes packets errs "
4209 "drop fifo colls carrier compressed\n");
4211 dev_seq_printf_stats(seq, v);
4215 static struct softnet_data *softnet_get_online(loff_t *pos)
4217 struct softnet_data *sd = NULL;
4219 while (*pos < nr_cpu_ids)
4220 if (cpu_online(*pos)) {
4221 sd = &per_cpu(softnet_data, *pos);
4228 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4230 return softnet_get_online(pos);
4233 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4236 return softnet_get_online(pos);
4239 static void softnet_seq_stop(struct seq_file *seq, void *v)
4243 static int softnet_seq_show(struct seq_file *seq, void *v)
4245 struct softnet_data *sd = v;
4247 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4248 sd->processed, sd->dropped, sd->time_squeeze, 0,
4249 0, 0, 0, 0, /* was fastroute */
4250 sd->cpu_collision, sd->received_rps);
4254 static const struct seq_operations dev_seq_ops = {
4255 .start = dev_seq_start,
4256 .next = dev_seq_next,
4257 .stop = dev_seq_stop,
4258 .show = dev_seq_show,
4261 static int dev_seq_open(struct inode *inode, struct file *file)
4263 return seq_open_net(inode, file, &dev_seq_ops,
4264 sizeof(struct seq_net_private));
4267 static const struct file_operations dev_seq_fops = {
4268 .owner = THIS_MODULE,
4269 .open = dev_seq_open,
4271 .llseek = seq_lseek,
4272 .release = seq_release_net,
4275 static const struct seq_operations softnet_seq_ops = {
4276 .start = softnet_seq_start,
4277 .next = softnet_seq_next,
4278 .stop = softnet_seq_stop,
4279 .show = softnet_seq_show,
4282 static int softnet_seq_open(struct inode *inode, struct file *file)
4284 return seq_open(file, &softnet_seq_ops);
4287 static const struct file_operations softnet_seq_fops = {
4288 .owner = THIS_MODULE,
4289 .open = softnet_seq_open,
4291 .llseek = seq_lseek,
4292 .release = seq_release,
4295 static void *ptype_get_idx(loff_t pos)
4297 struct packet_type *pt = NULL;
4301 list_for_each_entry_rcu(pt, &ptype_all, list) {
4307 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4308 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4317 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4321 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4324 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4326 struct packet_type *pt;
4327 struct list_head *nxt;
4331 if (v == SEQ_START_TOKEN)
4332 return ptype_get_idx(0);
4335 nxt = pt->list.next;
4336 if (pt->type == htons(ETH_P_ALL)) {
4337 if (nxt != &ptype_all)
4340 nxt = ptype_base[0].next;
4342 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4344 while (nxt == &ptype_base[hash]) {
4345 if (++hash >= PTYPE_HASH_SIZE)
4347 nxt = ptype_base[hash].next;
4350 return list_entry(nxt, struct packet_type, list);
4353 static void ptype_seq_stop(struct seq_file *seq, void *v)
4359 static int ptype_seq_show(struct seq_file *seq, void *v)
4361 struct packet_type *pt = v;
4363 if (v == SEQ_START_TOKEN)
4364 seq_puts(seq, "Type Device Function\n");
4365 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4366 if (pt->type == htons(ETH_P_ALL))
4367 seq_puts(seq, "ALL ");
4369 seq_printf(seq, "%04x", ntohs(pt->type));
4371 seq_printf(seq, " %-8s %pF\n",
4372 pt->dev ? pt->dev->name : "", pt->func);
4378 static const struct seq_operations ptype_seq_ops = {
4379 .start = ptype_seq_start,
4380 .next = ptype_seq_next,
4381 .stop = ptype_seq_stop,
4382 .show = ptype_seq_show,
4385 static int ptype_seq_open(struct inode *inode, struct file *file)
4387 return seq_open_net(inode, file, &ptype_seq_ops,
4388 sizeof(struct seq_net_private));
4391 static const struct file_operations ptype_seq_fops = {
4392 .owner = THIS_MODULE,
4393 .open = ptype_seq_open,
4395 .llseek = seq_lseek,
4396 .release = seq_release_net,
4400 static int __net_init dev_proc_net_init(struct net *net)
4404 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4406 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4408 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4411 if (wext_proc_init(net))
4417 proc_net_remove(net, "ptype");
4419 proc_net_remove(net, "softnet_stat");
4421 proc_net_remove(net, "dev");
4425 static void __net_exit dev_proc_net_exit(struct net *net)
4427 wext_proc_exit(net);
4429 proc_net_remove(net, "ptype");
4430 proc_net_remove(net, "softnet_stat");
4431 proc_net_remove(net, "dev");
4434 static struct pernet_operations __net_initdata dev_proc_ops = {
4435 .init = dev_proc_net_init,
4436 .exit = dev_proc_net_exit,
4439 static int __init dev_proc_init(void)
4441 return register_pernet_subsys(&dev_proc_ops);
4444 #define dev_proc_init() 0
4445 #endif /* CONFIG_PROC_FS */
4449 * netdev_set_master - set up master pointer
4450 * @slave: slave device
4451 * @master: new master device
4453 * Changes the master device of the slave. Pass %NULL to break the
4454 * bonding. The caller must hold the RTNL semaphore. On a failure
4455 * a negative errno code is returned. On success the reference counts
4456 * are adjusted and the function returns zero.
4458 int netdev_set_master(struct net_device *slave, struct net_device *master)
4460 struct net_device *old = slave->master;
4470 slave->master = master;
4476 EXPORT_SYMBOL(netdev_set_master);
4479 * netdev_set_bond_master - set up bonding master/slave pair
4480 * @slave: slave device
4481 * @master: new master device
4483 * Changes the master device of the slave. Pass %NULL to break the
4484 * bonding. The caller must hold the RTNL semaphore. On a failure
4485 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4486 * to the routing socket and the function returns zero.
4488 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4494 err = netdev_set_master(slave, master);
4498 slave->flags |= IFF_SLAVE;
4500 slave->flags &= ~IFF_SLAVE;
4502 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4505 EXPORT_SYMBOL(netdev_set_bond_master);
4507 static void dev_change_rx_flags(struct net_device *dev, int flags)
4509 const struct net_device_ops *ops = dev->netdev_ops;
4511 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4512 ops->ndo_change_rx_flags(dev, flags);
4515 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4517 unsigned short old_flags = dev->flags;
4523 dev->flags |= IFF_PROMISC;
4524 dev->promiscuity += inc;
4525 if (dev->promiscuity == 0) {
4528 * If inc causes overflow, untouch promisc and return error.
4531 dev->flags &= ~IFF_PROMISC;
4533 dev->promiscuity -= inc;
4534 printk(KERN_WARNING "%s: promiscuity touches roof, "
4535 "set promiscuity failed, promiscuity feature "
4536 "of device might be broken.\n", dev->name);
4540 if (dev->flags != old_flags) {
4541 printk(KERN_INFO "device %s %s promiscuous mode\n",
4542 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4544 if (audit_enabled) {
4545 current_uid_gid(&uid, &gid);
4546 audit_log(current->audit_context, GFP_ATOMIC,
4547 AUDIT_ANOM_PROMISCUOUS,
4548 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4549 dev->name, (dev->flags & IFF_PROMISC),
4550 (old_flags & IFF_PROMISC),
4551 audit_get_loginuid(current),
4553 audit_get_sessionid(current));
4556 dev_change_rx_flags(dev, IFF_PROMISC);
4562 * dev_set_promiscuity - update promiscuity count on a device
4566 * Add or remove promiscuity from a device. While the count in the device
4567 * remains above zero the interface remains promiscuous. Once it hits zero
4568 * the device reverts back to normal filtering operation. A negative inc
4569 * value is used to drop promiscuity on the device.
4570 * Return 0 if successful or a negative errno code on error.
4572 int dev_set_promiscuity(struct net_device *dev, int inc)
4574 unsigned short old_flags = dev->flags;
4577 err = __dev_set_promiscuity(dev, inc);
4580 if (dev->flags != old_flags)
4581 dev_set_rx_mode(dev);
4584 EXPORT_SYMBOL(dev_set_promiscuity);
4587 * dev_set_allmulti - update allmulti count on a device
4591 * Add or remove reception of all multicast frames to a device. While the
4592 * count in the device remains above zero the interface remains listening
4593 * to all interfaces. Once it hits zero the device reverts back to normal
4594 * filtering operation. A negative @inc value is used to drop the counter
4595 * when releasing a resource needing all multicasts.
4596 * Return 0 if successful or a negative errno code on error.
4599 int dev_set_allmulti(struct net_device *dev, int inc)
4601 unsigned short old_flags = dev->flags;
4605 dev->flags |= IFF_ALLMULTI;
4606 dev->allmulti += inc;
4607 if (dev->allmulti == 0) {
4610 * If inc causes overflow, untouch allmulti and return error.
4613 dev->flags &= ~IFF_ALLMULTI;
4615 dev->allmulti -= inc;
4616 printk(KERN_WARNING "%s: allmulti touches roof, "
4617 "set allmulti failed, allmulti feature of "
4618 "device might be broken.\n", dev->name);
4622 if (dev->flags ^ old_flags) {
4623 dev_change_rx_flags(dev, IFF_ALLMULTI);
4624 dev_set_rx_mode(dev);
4628 EXPORT_SYMBOL(dev_set_allmulti);
4631 * Upload unicast and multicast address lists to device and
4632 * configure RX filtering. When the device doesn't support unicast
4633 * filtering it is put in promiscuous mode while unicast addresses
4636 void __dev_set_rx_mode(struct net_device *dev)
4638 const struct net_device_ops *ops = dev->netdev_ops;
4640 /* dev_open will call this function so the list will stay sane. */
4641 if (!(dev->flags&IFF_UP))
4644 if (!netif_device_present(dev))
4647 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4648 /* Unicast addresses changes may only happen under the rtnl,
4649 * therefore calling __dev_set_promiscuity here is safe.
4651 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4652 __dev_set_promiscuity(dev, 1);
4653 dev->uc_promisc = true;
4654 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4655 __dev_set_promiscuity(dev, -1);
4656 dev->uc_promisc = false;
4660 if (ops->ndo_set_rx_mode)
4661 ops->ndo_set_rx_mode(dev);
4664 void dev_set_rx_mode(struct net_device *dev)
4666 netif_addr_lock_bh(dev);
4667 __dev_set_rx_mode(dev);
4668 netif_addr_unlock_bh(dev);
4672 * dev_get_flags - get flags reported to userspace
4675 * Get the combination of flag bits exported through APIs to userspace.
4677 unsigned dev_get_flags(const struct net_device *dev)
4681 flags = (dev->flags & ~(IFF_PROMISC |
4686 (dev->gflags & (IFF_PROMISC |
4689 if (netif_running(dev)) {
4690 if (netif_oper_up(dev))
4691 flags |= IFF_RUNNING;
4692 if (netif_carrier_ok(dev))
4693 flags |= IFF_LOWER_UP;
4694 if (netif_dormant(dev))
4695 flags |= IFF_DORMANT;
4700 EXPORT_SYMBOL(dev_get_flags);
4702 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4704 int old_flags = dev->flags;
4710 * Set the flags on our device.
4713 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4714 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4716 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4720 * Load in the correct multicast list now the flags have changed.
4723 if ((old_flags ^ flags) & IFF_MULTICAST)
4724 dev_change_rx_flags(dev, IFF_MULTICAST);
4726 dev_set_rx_mode(dev);
4729 * Have we downed the interface. We handle IFF_UP ourselves
4730 * according to user attempts to set it, rather than blindly
4735 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4736 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4739 dev_set_rx_mode(dev);
4742 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4743 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4745 dev->gflags ^= IFF_PROMISC;
4746 dev_set_promiscuity(dev, inc);
4749 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4750 is important. Some (broken) drivers set IFF_PROMISC, when
4751 IFF_ALLMULTI is requested not asking us and not reporting.
4753 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4754 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4756 dev->gflags ^= IFF_ALLMULTI;
4757 dev_set_allmulti(dev, inc);
4763 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4765 unsigned int changes = dev->flags ^ old_flags;
4767 if (changes & IFF_UP) {
4768 if (dev->flags & IFF_UP)
4769 call_netdevice_notifiers(NETDEV_UP, dev);
4771 call_netdevice_notifiers(NETDEV_DOWN, dev);
4774 if (dev->flags & IFF_UP &&
4775 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4776 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4780 * dev_change_flags - change device settings
4782 * @flags: device state flags
4784 * Change settings on device based state flags. The flags are
4785 * in the userspace exported format.
4787 int dev_change_flags(struct net_device *dev, unsigned flags)
4790 int old_flags = dev->flags;
4792 ret = __dev_change_flags(dev, flags);
4796 changes = old_flags ^ dev->flags;
4798 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4800 __dev_notify_flags(dev, old_flags);
4803 EXPORT_SYMBOL(dev_change_flags);
4806 * dev_set_mtu - Change maximum transfer unit
4808 * @new_mtu: new transfer unit
4810 * Change the maximum transfer size of the network device.
4812 int dev_set_mtu(struct net_device *dev, int new_mtu)
4814 const struct net_device_ops *ops = dev->netdev_ops;
4817 if (new_mtu == dev->mtu)
4820 /* MTU must be positive. */
4824 if (!netif_device_present(dev))
4828 if (ops->ndo_change_mtu)
4829 err = ops->ndo_change_mtu(dev, new_mtu);
4833 if (!err && dev->flags & IFF_UP)
4834 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4837 EXPORT_SYMBOL(dev_set_mtu);
4840 * dev_set_group - Change group this device belongs to
4842 * @new_group: group this device should belong to
4844 void dev_set_group(struct net_device *dev, int new_group)
4846 dev->group = new_group;
4848 EXPORT_SYMBOL(dev_set_group);
4851 * dev_set_mac_address - Change Media Access Control Address
4855 * Change the hardware (MAC) address of the device
4857 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4859 const struct net_device_ops *ops = dev->netdev_ops;
4862 if (!ops->ndo_set_mac_address)
4864 if (sa->sa_family != dev->type)
4866 if (!netif_device_present(dev))
4868 err = ops->ndo_set_mac_address(dev, sa);
4870 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4871 add_device_randomness(dev->dev_addr, dev->addr_len);
4874 EXPORT_SYMBOL(dev_set_mac_address);
4877 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4879 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4882 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4888 case SIOCGIFFLAGS: /* Get interface flags */
4889 ifr->ifr_flags = (short) dev_get_flags(dev);
4892 case SIOCGIFMETRIC: /* Get the metric on the interface
4893 (currently unused) */
4894 ifr->ifr_metric = 0;
4897 case SIOCGIFMTU: /* Get the MTU of a device */
4898 ifr->ifr_mtu = dev->mtu;
4903 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4905 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4906 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4907 ifr->ifr_hwaddr.sa_family = dev->type;
4915 ifr->ifr_map.mem_start = dev->mem_start;
4916 ifr->ifr_map.mem_end = dev->mem_end;
4917 ifr->ifr_map.base_addr = dev->base_addr;
4918 ifr->ifr_map.irq = dev->irq;
4919 ifr->ifr_map.dma = dev->dma;
4920 ifr->ifr_map.port = dev->if_port;
4924 ifr->ifr_ifindex = dev->ifindex;
4928 ifr->ifr_qlen = dev->tx_queue_len;
4932 /* dev_ioctl() should ensure this case
4944 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4946 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4949 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4950 const struct net_device_ops *ops;
4955 ops = dev->netdev_ops;
4958 case SIOCSIFFLAGS: /* Set interface flags */
4959 return dev_change_flags(dev, ifr->ifr_flags);
4961 case SIOCSIFMETRIC: /* Set the metric on the interface
4962 (currently unused) */
4965 case SIOCSIFMTU: /* Set the MTU of a device */
4966 return dev_set_mtu(dev, ifr->ifr_mtu);
4969 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4971 case SIOCSIFHWBROADCAST:
4972 if (ifr->ifr_hwaddr.sa_family != dev->type)
4974 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4975 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4976 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4980 if (ops->ndo_set_config) {
4981 if (!netif_device_present(dev))
4983 return ops->ndo_set_config(dev, &ifr->ifr_map);
4988 if (!ops->ndo_set_rx_mode ||
4989 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4991 if (!netif_device_present(dev))
4993 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4996 if (!ops->ndo_set_rx_mode ||
4997 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4999 if (!netif_device_present(dev))
5001 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5004 if (ifr->ifr_qlen < 0)
5006 dev->tx_queue_len = ifr->ifr_qlen;
5010 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5011 return dev_change_name(dev, ifr->ifr_newname);
5014 err = net_hwtstamp_validate(ifr);
5020 * Unknown or private ioctl
5023 if ((cmd >= SIOCDEVPRIVATE &&
5024 cmd <= SIOCDEVPRIVATE + 15) ||
5025 cmd == SIOCBONDENSLAVE ||
5026 cmd == SIOCBONDRELEASE ||
5027 cmd == SIOCBONDSETHWADDR ||
5028 cmd == SIOCBONDSLAVEINFOQUERY ||
5029 cmd == SIOCBONDINFOQUERY ||
5030 cmd == SIOCBONDCHANGEACTIVE ||
5031 cmd == SIOCGMIIPHY ||
5032 cmd == SIOCGMIIREG ||
5033 cmd == SIOCSMIIREG ||
5034 cmd == SIOCBRADDIF ||
5035 cmd == SIOCBRDELIF ||
5036 cmd == SIOCSHWTSTAMP ||
5037 cmd == SIOCWANDEV) {
5039 if (ops->ndo_do_ioctl) {
5040 if (netif_device_present(dev))
5041 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5053 * This function handles all "interface"-type I/O control requests. The actual
5054 * 'doing' part of this is dev_ifsioc above.
5058 * dev_ioctl - network device ioctl
5059 * @net: the applicable net namespace
5060 * @cmd: command to issue
5061 * @arg: pointer to a struct ifreq in user space
5063 * Issue ioctl functions to devices. This is normally called by the
5064 * user space syscall interfaces but can sometimes be useful for
5065 * other purposes. The return value is the return from the syscall if
5066 * positive or a negative errno code on error.
5069 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5075 /* One special case: SIOCGIFCONF takes ifconf argument
5076 and requires shared lock, because it sleeps writing
5080 if (cmd == SIOCGIFCONF) {
5082 ret = dev_ifconf(net, (char __user *) arg);
5086 if (cmd == SIOCGIFNAME)
5087 return dev_ifname(net, (struct ifreq __user *)arg);
5089 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5092 ifr.ifr_name[IFNAMSIZ-1] = 0;
5094 colon = strchr(ifr.ifr_name, ':');
5099 * See which interface the caller is talking about.
5104 * These ioctl calls:
5105 * - can be done by all.
5106 * - atomic and do not require locking.
5117 dev_load(net, ifr.ifr_name);
5119 ret = dev_ifsioc_locked(net, &ifr, cmd);
5124 if (copy_to_user(arg, &ifr,
5125 sizeof(struct ifreq)))
5131 dev_load(net, ifr.ifr_name);
5133 ret = dev_ethtool(net, &ifr);
5138 if (copy_to_user(arg, &ifr,
5139 sizeof(struct ifreq)))
5145 * These ioctl calls:
5146 * - require superuser power.
5147 * - require strict serialization.
5153 if (!capable(CAP_NET_ADMIN))
5155 dev_load(net, ifr.ifr_name);
5157 ret = dev_ifsioc(net, &ifr, cmd);
5162 if (copy_to_user(arg, &ifr,
5163 sizeof(struct ifreq)))
5169 * These ioctl calls:
5170 * - require superuser power.
5171 * - require strict serialization.
5172 * - do not return a value
5182 case SIOCSIFHWBROADCAST:
5185 case SIOCBONDENSLAVE:
5186 case SIOCBONDRELEASE:
5187 case SIOCBONDSETHWADDR:
5188 case SIOCBONDCHANGEACTIVE:
5192 if (!capable(CAP_NET_ADMIN))
5195 case SIOCBONDSLAVEINFOQUERY:
5196 case SIOCBONDINFOQUERY:
5197 dev_load(net, ifr.ifr_name);
5199 ret = dev_ifsioc(net, &ifr, cmd);
5204 /* Get the per device memory space. We can add this but
5205 * currently do not support it */
5207 /* Set the per device memory buffer space.
5208 * Not applicable in our case */
5213 * Unknown or private ioctl.
5216 if (cmd == SIOCWANDEV ||
5217 (cmd >= SIOCDEVPRIVATE &&
5218 cmd <= SIOCDEVPRIVATE + 15)) {
5219 dev_load(net, ifr.ifr_name);
5221 ret = dev_ifsioc(net, &ifr, cmd);
5223 if (!ret && copy_to_user(arg, &ifr,
5224 sizeof(struct ifreq)))
5228 /* Take care of Wireless Extensions */
5229 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5230 return wext_handle_ioctl(net, &ifr, cmd, arg);
5237 * dev_new_index - allocate an ifindex
5238 * @net: the applicable net namespace
5240 * Returns a suitable unique value for a new device interface
5241 * number. The caller must hold the rtnl semaphore or the
5242 * dev_base_lock to be sure it remains unique.
5244 static int dev_new_index(struct net *net)
5250 if (!__dev_get_by_index(net, ifindex))
5255 /* Delayed registration/unregisteration */
5256 static LIST_HEAD(net_todo_list);
5258 static void net_set_todo(struct net_device *dev)
5260 list_add_tail(&dev->todo_list, &net_todo_list);
5263 static void rollback_registered_many(struct list_head *head)
5265 struct net_device *dev, *tmp;
5267 BUG_ON(dev_boot_phase);
5270 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5271 /* Some devices call without registering
5272 * for initialization unwind. Remove those
5273 * devices and proceed with the remaining.
5275 if (dev->reg_state == NETREG_UNINITIALIZED) {
5276 pr_debug("unregister_netdevice: device %s/%p never "
5277 "was registered\n", dev->name, dev);
5280 list_del(&dev->unreg_list);
5283 dev->dismantle = true;
5284 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5287 /* If device is running, close it first. */
5288 dev_close_many(head);
5290 list_for_each_entry(dev, head, unreg_list) {
5291 /* And unlink it from device chain. */
5292 unlist_netdevice(dev);
5294 dev->reg_state = NETREG_UNREGISTERING;
5299 list_for_each_entry(dev, head, unreg_list) {
5300 /* Shutdown queueing discipline. */
5304 /* Notify protocols, that we are about to destroy
5305 this device. They should clean all the things.
5307 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5309 if (!dev->rtnl_link_ops ||
5310 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5311 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5314 * Flush the unicast and multicast chains
5319 if (dev->netdev_ops->ndo_uninit)
5320 dev->netdev_ops->ndo_uninit(dev);
5322 /* Notifier chain MUST detach us from master device. */
5323 WARN_ON(dev->master);
5325 /* Remove entries from kobject tree */
5326 netdev_unregister_kobject(dev);
5329 /* Process any work delayed until the end of the batch */
5330 dev = list_first_entry(head, struct net_device, unreg_list);
5331 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5335 list_for_each_entry(dev, head, unreg_list)
5339 static void rollback_registered(struct net_device *dev)
5343 list_add(&dev->unreg_list, &single);
5344 rollback_registered_many(&single);
5348 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5350 /* Fix illegal checksum combinations */
5351 if ((features & NETIF_F_HW_CSUM) &&
5352 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5353 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5354 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5357 if ((features & NETIF_F_NO_CSUM) &&
5358 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5359 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5360 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5363 /* Fix illegal SG+CSUM combinations. */
5364 if ((features & NETIF_F_SG) &&
5365 !(features & NETIF_F_ALL_CSUM)) {
5367 "Dropping NETIF_F_SG since no checksum feature.\n");
5368 features &= ~NETIF_F_SG;
5371 /* TSO requires that SG is present as well. */
5372 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5373 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5374 features &= ~NETIF_F_ALL_TSO;
5377 /* TSO ECN requires that TSO is present as well. */
5378 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5379 features &= ~NETIF_F_TSO_ECN;
5381 /* Software GSO depends on SG. */
5382 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5383 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5384 features &= ~NETIF_F_GSO;
5387 /* UFO needs SG and checksumming */
5388 if (features & NETIF_F_UFO) {
5389 /* maybe split UFO into V4 and V6? */
5390 if (!((features & NETIF_F_GEN_CSUM) ||
5391 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5392 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5394 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5395 features &= ~NETIF_F_UFO;
5398 if (!(features & NETIF_F_SG)) {
5400 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5401 features &= ~NETIF_F_UFO;
5408 int __netdev_update_features(struct net_device *dev)
5415 features = netdev_get_wanted_features(dev);
5417 if (dev->netdev_ops->ndo_fix_features)
5418 features = dev->netdev_ops->ndo_fix_features(dev, features);
5420 /* driver might be less strict about feature dependencies */
5421 features = netdev_fix_features(dev, features);
5423 if (dev->features == features)
5426 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5427 dev->features, features);
5429 if (dev->netdev_ops->ndo_set_features)
5430 err = dev->netdev_ops->ndo_set_features(dev, features);
5432 if (unlikely(err < 0)) {
5434 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5435 err, features, dev->features);
5440 dev->features = features;
5446 * netdev_update_features - recalculate device features
5447 * @dev: the device to check
5449 * Recalculate dev->features set and send notifications if it
5450 * has changed. Should be called after driver or hardware dependent
5451 * conditions might have changed that influence the features.
5453 void netdev_update_features(struct net_device *dev)
5455 if (__netdev_update_features(dev))
5456 netdev_features_change(dev);
5458 EXPORT_SYMBOL(netdev_update_features);
5461 * netdev_change_features - recalculate device features
5462 * @dev: the device to check
5464 * Recalculate dev->features set and send notifications even
5465 * if they have not changed. Should be called instead of
5466 * netdev_update_features() if also dev->vlan_features might
5467 * have changed to allow the changes to be propagated to stacked
5470 void netdev_change_features(struct net_device *dev)
5472 __netdev_update_features(dev);
5473 netdev_features_change(dev);
5475 EXPORT_SYMBOL(netdev_change_features);
5478 * netif_stacked_transfer_operstate - transfer operstate
5479 * @rootdev: the root or lower level device to transfer state from
5480 * @dev: the device to transfer operstate to
5482 * Transfer operational state from root to device. This is normally
5483 * called when a stacking relationship exists between the root
5484 * device and the device(a leaf device).
5486 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5487 struct net_device *dev)
5489 if (rootdev->operstate == IF_OPER_DORMANT)
5490 netif_dormant_on(dev);
5492 netif_dormant_off(dev);
5494 if (netif_carrier_ok(rootdev)) {
5495 if (!netif_carrier_ok(dev))
5496 netif_carrier_on(dev);
5498 if (netif_carrier_ok(dev))
5499 netif_carrier_off(dev);
5502 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5505 static int netif_alloc_rx_queues(struct net_device *dev)
5507 unsigned int i, count = dev->num_rx_queues;
5508 struct netdev_rx_queue *rx;
5512 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5514 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5519 for (i = 0; i < count; i++)
5525 static void netdev_init_one_queue(struct net_device *dev,
5526 struct netdev_queue *queue, void *_unused)
5528 /* Initialize queue lock */
5529 spin_lock_init(&queue->_xmit_lock);
5530 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5531 queue->xmit_lock_owner = -1;
5532 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5536 static int netif_alloc_netdev_queues(struct net_device *dev)
5538 unsigned int count = dev->num_tx_queues;
5539 struct netdev_queue *tx;
5543 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5545 pr_err("netdev: Unable to allocate %u tx queues.\n",
5551 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5552 spin_lock_init(&dev->tx_global_lock);
5558 * register_netdevice - register a network device
5559 * @dev: device to register
5561 * Take a completed network device structure and add it to the kernel
5562 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5563 * chain. 0 is returned on success. A negative errno code is returned
5564 * on a failure to set up the device, or if the name is a duplicate.
5566 * Callers must hold the rtnl semaphore. You may want
5567 * register_netdev() instead of this.
5570 * The locking appears insufficient to guarantee two parallel registers
5571 * will not get the same name.
5574 int register_netdevice(struct net_device *dev)
5577 struct net *net = dev_net(dev);
5579 BUG_ON(dev_boot_phase);
5584 /* When net_device's are persistent, this will be fatal. */
5585 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5588 spin_lock_init(&dev->addr_list_lock);
5589 netdev_set_addr_lockdep_class(dev);
5593 ret = dev_get_valid_name(dev, dev->name);
5597 /* Init, if this function is available */
5598 if (dev->netdev_ops->ndo_init) {
5599 ret = dev->netdev_ops->ndo_init(dev);
5607 dev->ifindex = dev_new_index(net);
5608 if (dev->iflink == -1)
5609 dev->iflink = dev->ifindex;
5611 /* Transfer changeable features to wanted_features and enable
5612 * software offloads (GSO and GRO).
5614 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5615 dev->features |= NETIF_F_SOFT_FEATURES;
5616 dev->wanted_features = dev->features & dev->hw_features;
5618 /* Turn on no cache copy if HW is doing checksum */
5619 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5620 if ((dev->features & NETIF_F_ALL_CSUM) &&
5621 !(dev->features & NETIF_F_NO_CSUM)) {
5622 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5623 dev->features |= NETIF_F_NOCACHE_COPY;
5626 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5628 dev->vlan_features |= NETIF_F_HIGHDMA;
5630 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5631 ret = notifier_to_errno(ret);
5635 ret = netdev_register_kobject(dev);
5638 dev->reg_state = NETREG_REGISTERED;
5640 __netdev_update_features(dev);
5643 * Default initial state at registry is that the
5644 * device is present.
5647 set_bit(__LINK_STATE_PRESENT, &dev->state);
5649 dev_init_scheduler(dev);
5651 list_netdevice(dev);
5652 add_device_randomness(dev->dev_addr, dev->addr_len);
5654 /* Notify protocols, that a new device appeared. */
5655 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5656 ret = notifier_to_errno(ret);
5658 rollback_registered(dev);
5659 dev->reg_state = NETREG_UNREGISTERED;
5662 * Prevent userspace races by waiting until the network
5663 * device is fully setup before sending notifications.
5665 if (!dev->rtnl_link_ops ||
5666 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5667 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5673 if (dev->netdev_ops->ndo_uninit)
5674 dev->netdev_ops->ndo_uninit(dev);
5677 EXPORT_SYMBOL(register_netdevice);
5680 * init_dummy_netdev - init a dummy network device for NAPI
5681 * @dev: device to init
5683 * This takes a network device structure and initialize the minimum
5684 * amount of fields so it can be used to schedule NAPI polls without
5685 * registering a full blown interface. This is to be used by drivers
5686 * that need to tie several hardware interfaces to a single NAPI
5687 * poll scheduler due to HW limitations.
5689 int init_dummy_netdev(struct net_device *dev)
5691 /* Clear everything. Note we don't initialize spinlocks
5692 * are they aren't supposed to be taken by any of the
5693 * NAPI code and this dummy netdev is supposed to be
5694 * only ever used for NAPI polls
5696 memset(dev, 0, sizeof(struct net_device));
5698 /* make sure we BUG if trying to hit standard
5699 * register/unregister code path
5701 dev->reg_state = NETREG_DUMMY;
5703 /* NAPI wants this */
5704 INIT_LIST_HEAD(&dev->napi_list);
5706 /* a dummy interface is started by default */
5707 set_bit(__LINK_STATE_PRESENT, &dev->state);
5708 set_bit(__LINK_STATE_START, &dev->state);
5710 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5711 * because users of this 'device' dont need to change
5717 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5721 * register_netdev - register a network device
5722 * @dev: device to register
5724 * Take a completed network device structure and add it to the kernel
5725 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5726 * chain. 0 is returned on success. A negative errno code is returned
5727 * on a failure to set up the device, or if the name is a duplicate.
5729 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5730 * and expands the device name if you passed a format string to
5733 int register_netdev(struct net_device *dev)
5738 err = register_netdevice(dev);
5742 EXPORT_SYMBOL(register_netdev);
5744 int netdev_refcnt_read(const struct net_device *dev)
5748 for_each_possible_cpu(i)
5749 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5752 EXPORT_SYMBOL(netdev_refcnt_read);
5755 * netdev_wait_allrefs - wait until all references are gone.
5757 * This is called when unregistering network devices.
5759 * Any protocol or device that holds a reference should register
5760 * for netdevice notification, and cleanup and put back the
5761 * reference if they receive an UNREGISTER event.
5762 * We can get stuck here if buggy protocols don't correctly
5765 static void netdev_wait_allrefs(struct net_device *dev)
5767 unsigned long rebroadcast_time, warning_time;
5770 linkwatch_forget_dev(dev);
5772 rebroadcast_time = warning_time = jiffies;
5773 refcnt = netdev_refcnt_read(dev);
5775 while (refcnt != 0) {
5776 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5779 /* Rebroadcast unregister notification */
5780 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5781 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5782 * should have already handle it the first time */
5784 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5786 /* We must not have linkwatch events
5787 * pending on unregister. If this
5788 * happens, we simply run the queue
5789 * unscheduled, resulting in a noop
5792 linkwatch_run_queue();
5797 rebroadcast_time = jiffies;
5802 refcnt = netdev_refcnt_read(dev);
5804 if (time_after(jiffies, warning_time + 10 * HZ)) {
5805 printk(KERN_EMERG "unregister_netdevice: "
5806 "waiting for %s to become free. Usage "
5809 warning_time = jiffies;
5818 * register_netdevice(x1);
5819 * register_netdevice(x2);
5821 * unregister_netdevice(y1);
5822 * unregister_netdevice(y2);
5828 * We are invoked by rtnl_unlock().
5829 * This allows us to deal with problems:
5830 * 1) We can delete sysfs objects which invoke hotplug
5831 * without deadlocking with linkwatch via keventd.
5832 * 2) Since we run with the RTNL semaphore not held, we can sleep
5833 * safely in order to wait for the netdev refcnt to drop to zero.
5835 * We must not return until all unregister events added during
5836 * the interval the lock was held have been completed.
5838 void netdev_run_todo(void)
5840 struct list_head list;
5842 /* Snapshot list, allow later requests */
5843 list_replace_init(&net_todo_list, &list);
5847 /* Wait for rcu callbacks to finish before attempting to drain
5848 * the device list. This usually avoids a 250ms wait.
5850 if (!list_empty(&list))
5853 while (!list_empty(&list)) {
5854 struct net_device *dev
5855 = list_first_entry(&list, struct net_device, todo_list);
5856 list_del(&dev->todo_list);
5858 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5859 printk(KERN_ERR "network todo '%s' but state %d\n",
5860 dev->name, dev->reg_state);
5865 dev->reg_state = NETREG_UNREGISTERED;
5867 on_each_cpu(flush_backlog, dev, 1);
5869 netdev_wait_allrefs(dev);
5872 BUG_ON(netdev_refcnt_read(dev));
5873 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5874 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5875 WARN_ON(dev->dn_ptr);
5877 if (dev->destructor)
5878 dev->destructor(dev);
5880 /* Free network device */
5881 kobject_put(&dev->dev.kobj);
5885 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5886 * fields in the same order, with only the type differing.
5888 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5889 const struct net_device_stats *netdev_stats)
5891 #if BITS_PER_LONG == 64
5892 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5893 memcpy(stats64, netdev_stats, sizeof(*stats64));
5895 size_t i, n = sizeof(*stats64) / sizeof(u64);
5896 const unsigned long *src = (const unsigned long *)netdev_stats;
5897 u64 *dst = (u64 *)stats64;
5899 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5900 sizeof(*stats64) / sizeof(u64));
5901 for (i = 0; i < n; i++)
5907 * dev_get_stats - get network device statistics
5908 * @dev: device to get statistics from
5909 * @storage: place to store stats
5911 * Get network statistics from device. Return @storage.
5912 * The device driver may provide its own method by setting
5913 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5914 * otherwise the internal statistics structure is used.
5916 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5917 struct rtnl_link_stats64 *storage)
5919 const struct net_device_ops *ops = dev->netdev_ops;
5921 if (ops->ndo_get_stats64) {
5922 memset(storage, 0, sizeof(*storage));
5923 ops->ndo_get_stats64(dev, storage);
5924 } else if (ops->ndo_get_stats) {
5925 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5927 netdev_stats_to_stats64(storage, &dev->stats);
5929 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5932 EXPORT_SYMBOL(dev_get_stats);
5934 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5936 struct netdev_queue *queue = dev_ingress_queue(dev);
5938 #ifdef CONFIG_NET_CLS_ACT
5941 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5944 netdev_init_one_queue(dev, queue, NULL);
5945 queue->qdisc = &noop_qdisc;
5946 queue->qdisc_sleeping = &noop_qdisc;
5947 rcu_assign_pointer(dev->ingress_queue, queue);
5953 * alloc_netdev_mqs - allocate network device
5954 * @sizeof_priv: size of private data to allocate space for
5955 * @name: device name format string
5956 * @setup: callback to initialize device
5957 * @txqs: the number of TX subqueues to allocate
5958 * @rxqs: the number of RX subqueues to allocate
5960 * Allocates a struct net_device with private data area for driver use
5961 * and performs basic initialization. Also allocates subquue structs
5962 * for each queue on the device.
5964 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5965 void (*setup)(struct net_device *),
5966 unsigned int txqs, unsigned int rxqs)
5968 struct net_device *dev;
5970 struct net_device *p;
5972 BUG_ON(strlen(name) >= sizeof(dev->name));
5975 pr_err("alloc_netdev: Unable to allocate device "
5976 "with zero queues.\n");
5982 pr_err("alloc_netdev: Unable to allocate device "
5983 "with zero RX queues.\n");
5988 alloc_size = sizeof(struct net_device);
5990 /* ensure 32-byte alignment of private area */
5991 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5992 alloc_size += sizeof_priv;
5994 /* ensure 32-byte alignment of whole construct */
5995 alloc_size += NETDEV_ALIGN - 1;
5997 p = kzalloc(alloc_size, GFP_KERNEL);
5999 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
6003 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6004 dev->padded = (char *)dev - (char *)p;
6006 dev->pcpu_refcnt = alloc_percpu(int);
6007 if (!dev->pcpu_refcnt)
6010 if (dev_addr_init(dev))
6016 dev_net_set(dev, &init_net);
6018 dev->gso_max_size = GSO_MAX_SIZE;
6019 dev->gso_max_segs = GSO_MAX_SEGS;
6021 INIT_LIST_HEAD(&dev->napi_list);
6022 INIT_LIST_HEAD(&dev->unreg_list);
6023 INIT_LIST_HEAD(&dev->link_watch_list);
6024 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6027 dev->num_tx_queues = txqs;
6028 dev->real_num_tx_queues = txqs;
6029 if (netif_alloc_netdev_queues(dev))
6033 dev->num_rx_queues = rxqs;
6034 dev->real_num_rx_queues = rxqs;
6035 if (netif_alloc_rx_queues(dev))
6039 strcpy(dev->name, name);
6040 dev->group = INIT_NETDEV_GROUP;
6048 free_percpu(dev->pcpu_refcnt);
6058 EXPORT_SYMBOL(alloc_netdev_mqs);
6061 * free_netdev - free network device
6064 * This function does the last stage of destroying an allocated device
6065 * interface. The reference to the device object is released.
6066 * If this is the last reference then it will be freed.
6068 void free_netdev(struct net_device *dev)
6070 struct napi_struct *p, *n;
6072 release_net(dev_net(dev));
6079 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6081 /* Flush device addresses */
6082 dev_addr_flush(dev);
6084 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6087 free_percpu(dev->pcpu_refcnt);
6088 dev->pcpu_refcnt = NULL;
6090 /* Compatibility with error handling in drivers */
6091 if (dev->reg_state == NETREG_UNINITIALIZED) {
6092 kfree((char *)dev - dev->padded);
6096 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6097 dev->reg_state = NETREG_RELEASED;
6099 /* will free via device release */
6100 put_device(&dev->dev);
6102 EXPORT_SYMBOL(free_netdev);
6105 * synchronize_net - Synchronize with packet receive processing
6107 * Wait for packets currently being received to be done.
6108 * Does not block later packets from starting.
6110 void synchronize_net(void)
6113 if (rtnl_is_locked())
6114 synchronize_rcu_expedited();
6118 EXPORT_SYMBOL(synchronize_net);
6121 * unregister_netdevice_queue - remove device from the kernel
6125 * This function shuts down a device interface and removes it
6126 * from the kernel tables.
6127 * If head not NULL, device is queued to be unregistered later.
6129 * Callers must hold the rtnl semaphore. You may want
6130 * unregister_netdev() instead of this.
6133 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6138 list_move_tail(&dev->unreg_list, head);
6140 rollback_registered(dev);
6141 /* Finish processing unregister after unlock */
6145 EXPORT_SYMBOL(unregister_netdevice_queue);
6148 * unregister_netdevice_many - unregister many devices
6149 * @head: list of devices
6151 void unregister_netdevice_many(struct list_head *head)
6153 struct net_device *dev;
6155 if (!list_empty(head)) {
6156 rollback_registered_many(head);
6157 list_for_each_entry(dev, head, unreg_list)
6161 EXPORT_SYMBOL(unregister_netdevice_many);
6164 * unregister_netdev - remove device from the kernel
6167 * This function shuts down a device interface and removes it
6168 * from the kernel tables.
6170 * This is just a wrapper for unregister_netdevice that takes
6171 * the rtnl semaphore. In general you want to use this and not
6172 * unregister_netdevice.
6174 void unregister_netdev(struct net_device *dev)
6177 unregister_netdevice(dev);
6180 EXPORT_SYMBOL(unregister_netdev);
6183 * dev_change_net_namespace - move device to different nethost namespace
6185 * @net: network namespace
6186 * @pat: If not NULL name pattern to try if the current device name
6187 * is already taken in the destination network namespace.
6189 * This function shuts down a device interface and moves it
6190 * to a new network namespace. On success 0 is returned, on
6191 * a failure a netagive errno code is returned.
6193 * Callers must hold the rtnl semaphore.
6196 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6202 /* Don't allow namespace local devices to be moved. */
6204 if (dev->features & NETIF_F_NETNS_LOCAL)
6207 /* Ensure the device has been registrered */
6209 if (dev->reg_state != NETREG_REGISTERED)
6212 /* Get out if there is nothing todo */
6214 if (net_eq(dev_net(dev), net))
6217 /* Pick the destination device name, and ensure
6218 * we can use it in the destination network namespace.
6221 if (__dev_get_by_name(net, dev->name)) {
6222 /* We get here if we can't use the current device name */
6225 if (dev_get_valid_name(dev, pat) < 0)
6230 * And now a mini version of register_netdevice unregister_netdevice.
6233 /* If device is running close it first. */
6236 /* And unlink it from device chain */
6238 unlist_netdevice(dev);
6242 /* Shutdown queueing discipline. */
6245 /* Notify protocols, that we are about to destroy
6246 this device. They should clean all the things.
6248 Note that dev->reg_state stays at NETREG_REGISTERED.
6249 This is wanted because this way 8021q and macvlan know
6250 the device is just moving and can keep their slaves up.
6252 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6253 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6254 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6257 * Flush the unicast and multicast chains
6262 /* Actually switch the network namespace */
6263 dev_net_set(dev, net);
6265 /* If there is an ifindex conflict assign a new one */
6266 if (__dev_get_by_index(net, dev->ifindex)) {
6267 int iflink = (dev->iflink == dev->ifindex);
6268 dev->ifindex = dev_new_index(net);
6270 dev->iflink = dev->ifindex;
6273 /* Fixup kobjects */
6274 err = device_rename(&dev->dev, dev->name);
6277 /* Add the device back in the hashes */
6278 list_netdevice(dev);
6280 /* Notify protocols, that a new device appeared. */
6281 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6284 * Prevent userspace races by waiting until the network
6285 * device is fully setup before sending notifications.
6287 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6294 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6296 static int dev_cpu_callback(struct notifier_block *nfb,
6297 unsigned long action,
6300 struct sk_buff **list_skb;
6301 struct sk_buff *skb;
6302 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6303 struct softnet_data *sd, *oldsd;
6305 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6308 local_irq_disable();
6309 cpu = smp_processor_id();
6310 sd = &per_cpu(softnet_data, cpu);
6311 oldsd = &per_cpu(softnet_data, oldcpu);
6313 /* Find end of our completion_queue. */
6314 list_skb = &sd->completion_queue;
6316 list_skb = &(*list_skb)->next;
6317 /* Append completion queue from offline CPU. */
6318 *list_skb = oldsd->completion_queue;
6319 oldsd->completion_queue = NULL;
6321 /* Append output queue from offline CPU. */
6322 if (oldsd->output_queue) {
6323 *sd->output_queue_tailp = oldsd->output_queue;
6324 sd->output_queue_tailp = oldsd->output_queue_tailp;
6325 oldsd->output_queue = NULL;
6326 oldsd->output_queue_tailp = &oldsd->output_queue;
6328 /* Append NAPI poll list from offline CPU. */
6329 if (!list_empty(&oldsd->poll_list)) {
6330 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6331 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6334 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6337 /* Process offline CPU's input_pkt_queue */
6338 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6340 input_queue_head_incr(oldsd);
6342 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6344 input_queue_head_incr(oldsd);
6352 * netdev_increment_features - increment feature set by one
6353 * @all: current feature set
6354 * @one: new feature set
6355 * @mask: mask feature set
6357 * Computes a new feature set after adding a device with feature set
6358 * @one to the master device with current feature set @all. Will not
6359 * enable anything that is off in @mask. Returns the new feature set.
6361 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6363 if (mask & NETIF_F_GEN_CSUM)
6364 mask |= NETIF_F_ALL_CSUM;
6365 mask |= NETIF_F_VLAN_CHALLENGED;
6367 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6368 all &= one | ~NETIF_F_ALL_FOR_ALL;
6370 /* If device needs checksumming, downgrade to it. */
6371 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6372 all &= ~NETIF_F_NO_CSUM;
6374 /* If one device supports hw checksumming, set for all. */
6375 if (all & NETIF_F_GEN_CSUM)
6376 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6380 EXPORT_SYMBOL(netdev_increment_features);
6382 static struct hlist_head *netdev_create_hash(void)
6385 struct hlist_head *hash;
6387 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6389 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6390 INIT_HLIST_HEAD(&hash[i]);
6395 /* Initialize per network namespace state */
6396 static int __net_init netdev_init(struct net *net)
6398 if (net != &init_net)
6399 INIT_LIST_HEAD(&net->dev_base_head);
6401 net->dev_name_head = netdev_create_hash();
6402 if (net->dev_name_head == NULL)
6405 net->dev_index_head = netdev_create_hash();
6406 if (net->dev_index_head == NULL)
6412 kfree(net->dev_name_head);
6418 * netdev_drivername - network driver for the device
6419 * @dev: network device
6421 * Determine network driver for device.
6423 const char *netdev_drivername(const struct net_device *dev)
6425 const struct device_driver *driver;
6426 const struct device *parent;
6427 const char *empty = "";
6429 parent = dev->dev.parent;
6433 driver = parent->driver;
6434 if (driver && driver->name)
6435 return driver->name;
6439 int __netdev_printk(const char *level, const struct net_device *dev,
6440 struct va_format *vaf)
6444 if (dev && dev->dev.parent)
6445 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6446 netdev_name(dev), vaf);
6448 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6450 r = printk("%s(NULL net_device): %pV", level, vaf);
6454 EXPORT_SYMBOL(__netdev_printk);
6456 int netdev_printk(const char *level, const struct net_device *dev,
6457 const char *format, ...)
6459 struct va_format vaf;
6463 va_start(args, format);
6468 r = __netdev_printk(level, dev, &vaf);
6473 EXPORT_SYMBOL(netdev_printk);
6475 #define define_netdev_printk_level(func, level) \
6476 int func(const struct net_device *dev, const char *fmt, ...) \
6479 struct va_format vaf; \
6482 va_start(args, fmt); \
6487 r = __netdev_printk(level, dev, &vaf); \
6492 EXPORT_SYMBOL(func);
6494 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6495 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6496 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6497 define_netdev_printk_level(netdev_err, KERN_ERR);
6498 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6499 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6500 define_netdev_printk_level(netdev_info, KERN_INFO);
6502 static void __net_exit netdev_exit(struct net *net)
6504 kfree(net->dev_name_head);
6505 kfree(net->dev_index_head);
6508 static struct pernet_operations __net_initdata netdev_net_ops = {
6509 .init = netdev_init,
6510 .exit = netdev_exit,
6513 static void __net_exit default_device_exit(struct net *net)
6515 struct net_device *dev, *aux;
6517 * Push all migratable network devices back to the
6518 * initial network namespace
6521 for_each_netdev_safe(net, dev, aux) {
6523 char fb_name[IFNAMSIZ];
6525 /* Ignore unmoveable devices (i.e. loopback) */
6526 if (dev->features & NETIF_F_NETNS_LOCAL)
6529 /* Leave virtual devices for the generic cleanup */
6530 if (dev->rtnl_link_ops)
6533 /* Push remaining network devices to init_net */
6534 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6535 err = dev_change_net_namespace(dev, &init_net, fb_name);
6537 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6538 __func__, dev->name, err);
6545 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6547 /* At exit all network devices most be removed from a network
6548 * namespace. Do this in the reverse order of registration.
6549 * Do this across as many network namespaces as possible to
6550 * improve batching efficiency.
6552 struct net_device *dev;
6554 LIST_HEAD(dev_kill_list);
6557 list_for_each_entry(net, net_list, exit_list) {
6558 for_each_netdev_reverse(net, dev) {
6559 if (dev->rtnl_link_ops)
6560 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6562 unregister_netdevice_queue(dev, &dev_kill_list);
6565 unregister_netdevice_many(&dev_kill_list);
6566 list_del(&dev_kill_list);
6570 static struct pernet_operations __net_initdata default_device_ops = {
6571 .exit = default_device_exit,
6572 .exit_batch = default_device_exit_batch,
6576 * Initialize the DEV module. At boot time this walks the device list and
6577 * unhooks any devices that fail to initialise (normally hardware not
6578 * present) and leaves us with a valid list of present and active devices.
6583 * This is called single threaded during boot, so no need
6584 * to take the rtnl semaphore.
6586 static int __init net_dev_init(void)
6588 int i, rc = -ENOMEM;
6590 BUG_ON(!dev_boot_phase);
6592 if (dev_proc_init())
6595 if (netdev_kobject_init())
6598 INIT_LIST_HEAD(&ptype_all);
6599 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6600 INIT_LIST_HEAD(&ptype_base[i]);
6602 if (register_pernet_subsys(&netdev_net_ops))
6606 * Initialise the packet receive queues.
6609 for_each_possible_cpu(i) {
6610 struct softnet_data *sd = &per_cpu(softnet_data, i);
6612 memset(sd, 0, sizeof(*sd));
6613 skb_queue_head_init(&sd->input_pkt_queue);
6614 skb_queue_head_init(&sd->process_queue);
6615 sd->completion_queue = NULL;
6616 INIT_LIST_HEAD(&sd->poll_list);
6617 sd->output_queue = NULL;
6618 sd->output_queue_tailp = &sd->output_queue;
6620 sd->csd.func = rps_trigger_softirq;
6626 sd->backlog.poll = process_backlog;
6627 sd->backlog.weight = weight_p;
6628 sd->backlog.gro_list = NULL;
6629 sd->backlog.gro_count = 0;
6634 /* The loopback device is special if any other network devices
6635 * is present in a network namespace the loopback device must
6636 * be present. Since we now dynamically allocate and free the
6637 * loopback device ensure this invariant is maintained by
6638 * keeping the loopback device as the first device on the
6639 * list of network devices. Ensuring the loopback devices
6640 * is the first device that appears and the last network device
6643 if (register_pernet_device(&loopback_net_ops))
6646 if (register_pernet_device(&default_device_ops))
6649 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6650 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6652 hotcpu_notifier(dev_cpu_callback, 0);
6660 subsys_initcall(net_dev_init);
6662 static int __init initialize_hashrnd(void)
6664 get_random_bytes(&hashrnd, sizeof(hashrnd));
6668 late_initcall_sync(initialize_hashrnd);