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 /* Note: rx_handler_data must be set before rx_handler */
3197 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3198 rcu_assign_pointer(dev->rx_handler, rx_handler);
3202 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3205 * netdev_rx_handler_unregister - unregister receive handler
3206 * @dev: device to unregister a handler from
3208 * Unregister a receive hander from a device.
3210 * The caller must hold the rtnl_mutex.
3212 void netdev_rx_handler_unregister(struct net_device *dev)
3216 RCU_INIT_POINTER(dev->rx_handler, NULL);
3217 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3218 * section has a guarantee to see a non NULL rx_handler_data
3222 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3224 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3226 static int __netif_receive_skb(struct sk_buff *skb)
3228 struct packet_type *ptype, *pt_prev;
3229 rx_handler_func_t *rx_handler;
3230 struct net_device *orig_dev;
3231 struct net_device *null_or_dev;
3232 bool deliver_exact = false;
3233 int ret = NET_RX_DROP;
3236 if (!netdev_tstamp_prequeue)
3237 net_timestamp_check(skb);
3239 trace_netif_receive_skb(skb);
3241 /* if we've gotten here through NAPI, check netpoll */
3242 if (netpoll_receive_skb(skb))
3246 skb->skb_iif = skb->dev->ifindex;
3247 orig_dev = skb->dev;
3249 skb_reset_network_header(skb);
3250 skb_reset_transport_header(skb);
3251 skb_reset_mac_len(skb);
3259 __this_cpu_inc(softnet_data.processed);
3261 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3262 skb = vlan_untag(skb);
3267 #ifdef CONFIG_NET_CLS_ACT
3268 if (skb->tc_verd & TC_NCLS) {
3269 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3274 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3275 if (!ptype->dev || ptype->dev == skb->dev) {
3277 ret = deliver_skb(skb, pt_prev, orig_dev);
3282 #ifdef CONFIG_NET_CLS_ACT
3283 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3289 if (vlan_tx_tag_present(skb)) {
3291 ret = deliver_skb(skb, pt_prev, orig_dev);
3294 if (vlan_do_receive(&skb))
3296 else if (unlikely(!skb))
3300 rx_handler = rcu_dereference(skb->dev->rx_handler);
3303 ret = deliver_skb(skb, pt_prev, orig_dev);
3306 switch (rx_handler(&skb)) {
3307 case RX_HANDLER_CONSUMED:
3308 ret = NET_RX_SUCCESS;
3310 case RX_HANDLER_ANOTHER:
3312 case RX_HANDLER_EXACT:
3313 deliver_exact = true;
3314 case RX_HANDLER_PASS:
3321 if (vlan_tx_nonzero_tag_present(skb))
3322 skb->pkt_type = PACKET_OTHERHOST;
3324 /* deliver only exact match when indicated */
3325 null_or_dev = deliver_exact ? skb->dev : NULL;
3327 type = skb->protocol;
3328 list_for_each_entry_rcu(ptype,
3329 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3330 if (ptype->type == type &&
3331 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3332 ptype->dev == orig_dev)) {
3334 ret = deliver_skb(skb, pt_prev, orig_dev);
3340 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3342 atomic_long_inc(&skb->dev->rx_dropped);
3344 /* Jamal, now you will not able to escape explaining
3345 * me how you were going to use this. :-)
3356 * netif_receive_skb - process receive buffer from network
3357 * @skb: buffer to process
3359 * netif_receive_skb() is the main receive data processing function.
3360 * It always succeeds. The buffer may be dropped during processing
3361 * for congestion control or by the protocol layers.
3363 * This function may only be called from softirq context and interrupts
3364 * should be enabled.
3366 * Return values (usually ignored):
3367 * NET_RX_SUCCESS: no congestion
3368 * NET_RX_DROP: packet was dropped
3370 int netif_receive_skb(struct sk_buff *skb)
3372 if (netdev_tstamp_prequeue)
3373 net_timestamp_check(skb);
3375 if (skb_defer_rx_timestamp(skb))
3376 return NET_RX_SUCCESS;
3380 struct rps_dev_flow voidflow, *rflow = &voidflow;
3385 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3388 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3392 ret = __netif_receive_skb(skb);
3398 return __netif_receive_skb(skb);
3401 EXPORT_SYMBOL(netif_receive_skb);
3403 /* Network device is going away, flush any packets still pending
3404 * Called with irqs disabled.
3406 static void flush_backlog(void *arg)
3408 struct net_device *dev = arg;
3409 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3410 struct sk_buff *skb, *tmp;
3413 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3414 if (skb->dev == dev) {
3415 __skb_unlink(skb, &sd->input_pkt_queue);
3417 input_queue_head_incr(sd);
3422 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3423 if (skb->dev == dev) {
3424 __skb_unlink(skb, &sd->process_queue);
3426 input_queue_head_incr(sd);
3431 static int napi_gro_complete(struct sk_buff *skb)
3433 struct packet_type *ptype;
3434 __be16 type = skb->protocol;
3435 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3438 if (NAPI_GRO_CB(skb)->count == 1) {
3439 skb_shinfo(skb)->gso_size = 0;
3444 list_for_each_entry_rcu(ptype, head, list) {
3445 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3448 err = ptype->gro_complete(skb);
3454 WARN_ON(&ptype->list == head);
3456 return NET_RX_SUCCESS;
3460 return netif_receive_skb(skb);
3463 inline void napi_gro_flush(struct napi_struct *napi)
3465 struct sk_buff *skb, *next;
3467 for (skb = napi->gro_list; skb; skb = next) {
3470 napi_gro_complete(skb);
3473 napi->gro_count = 0;
3474 napi->gro_list = NULL;
3476 EXPORT_SYMBOL(napi_gro_flush);
3478 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3480 struct sk_buff **pp = NULL;
3481 struct packet_type *ptype;
3482 __be16 type = skb->protocol;
3483 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3486 enum gro_result ret;
3488 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3491 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3495 list_for_each_entry_rcu(ptype, head, list) {
3496 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3499 skb_set_network_header(skb, skb_gro_offset(skb));
3500 mac_len = skb->network_header - skb->mac_header;
3501 skb->mac_len = mac_len;
3502 NAPI_GRO_CB(skb)->same_flow = 0;
3503 NAPI_GRO_CB(skb)->flush = 0;
3504 NAPI_GRO_CB(skb)->free = 0;
3506 pp = ptype->gro_receive(&napi->gro_list, skb);
3511 if (&ptype->list == head)
3514 same_flow = NAPI_GRO_CB(skb)->same_flow;
3515 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3518 struct sk_buff *nskb = *pp;
3522 napi_gro_complete(nskb);
3529 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3533 NAPI_GRO_CB(skb)->count = 1;
3534 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3535 skb->next = napi->gro_list;
3536 napi->gro_list = skb;
3540 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3541 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3543 BUG_ON(skb->end - skb->tail < grow);
3545 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3548 skb->data_len -= grow;
3550 skb_shinfo(skb)->frags[0].page_offset += grow;
3551 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3553 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3554 skb_frag_unref(skb, 0);
3555 memmove(skb_shinfo(skb)->frags,
3556 skb_shinfo(skb)->frags + 1,
3557 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3568 EXPORT_SYMBOL(dev_gro_receive);
3570 static inline gro_result_t
3571 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3574 unsigned int maclen = skb->dev->hard_header_len;
3576 for (p = napi->gro_list; p; p = p->next) {
3577 unsigned long diffs;
3579 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3580 diffs |= p->vlan_tci ^ skb->vlan_tci;
3581 if (maclen == ETH_HLEN)
3582 diffs |= compare_ether_header(skb_mac_header(p),
3583 skb_gro_mac_header(skb));
3585 diffs = memcmp(skb_mac_header(p),
3586 skb_gro_mac_header(skb),
3588 NAPI_GRO_CB(p)->same_flow = !diffs;
3589 NAPI_GRO_CB(p)->flush = 0;
3592 return dev_gro_receive(napi, skb);
3595 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3599 if (netif_receive_skb(skb))
3604 case GRO_MERGED_FREE:
3615 EXPORT_SYMBOL(napi_skb_finish);
3617 void skb_gro_reset_offset(struct sk_buff *skb)
3619 NAPI_GRO_CB(skb)->data_offset = 0;
3620 NAPI_GRO_CB(skb)->frag0 = NULL;
3621 NAPI_GRO_CB(skb)->frag0_len = 0;
3623 if (skb->mac_header == skb->tail &&
3624 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3625 NAPI_GRO_CB(skb)->frag0 =
3626 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3627 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3630 EXPORT_SYMBOL(skb_gro_reset_offset);
3632 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3634 skb_gro_reset_offset(skb);
3636 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3638 EXPORT_SYMBOL(napi_gro_receive);
3640 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3642 __skb_pull(skb, skb_headlen(skb));
3643 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3644 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3646 skb->dev = napi->dev;
3652 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3654 struct sk_buff *skb = napi->skb;
3657 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3663 EXPORT_SYMBOL(napi_get_frags);
3665 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3671 skb->protocol = eth_type_trans(skb, skb->dev);
3673 if (ret == GRO_HELD)
3674 skb_gro_pull(skb, -ETH_HLEN);
3675 else if (netif_receive_skb(skb))
3680 case GRO_MERGED_FREE:
3681 napi_reuse_skb(napi, skb);
3690 EXPORT_SYMBOL(napi_frags_finish);
3692 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3694 struct sk_buff *skb = napi->skb;
3701 skb_reset_mac_header(skb);
3702 skb_gro_reset_offset(skb);
3704 off = skb_gro_offset(skb);
3705 hlen = off + sizeof(*eth);
3706 eth = skb_gro_header_fast(skb, off);
3707 if (skb_gro_header_hard(skb, hlen)) {
3708 eth = skb_gro_header_slow(skb, hlen, off);
3709 if (unlikely(!eth)) {
3710 napi_reuse_skb(napi, skb);
3716 skb_gro_pull(skb, sizeof(*eth));
3719 * This works because the only protocols we care about don't require
3720 * special handling. We'll fix it up properly at the end.
3722 skb->protocol = eth->h_proto;
3727 EXPORT_SYMBOL(napi_frags_skb);
3729 gro_result_t napi_gro_frags(struct napi_struct *napi)
3731 struct sk_buff *skb = napi_frags_skb(napi);
3736 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3738 EXPORT_SYMBOL(napi_gro_frags);
3741 * net_rps_action sends any pending IPI's for rps.
3742 * Note: called with local irq disabled, but exits with local irq enabled.
3744 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3747 struct softnet_data *remsd = sd->rps_ipi_list;
3750 sd->rps_ipi_list = NULL;
3754 /* Send pending IPI's to kick RPS processing on remote cpus. */
3756 struct softnet_data *next = remsd->rps_ipi_next;
3758 if (cpu_online(remsd->cpu))
3759 __smp_call_function_single(remsd->cpu,
3768 static int process_backlog(struct napi_struct *napi, int quota)
3771 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3774 /* Check if we have pending ipi, its better to send them now,
3775 * not waiting net_rx_action() end.
3777 if (sd->rps_ipi_list) {
3778 local_irq_disable();
3779 net_rps_action_and_irq_enable(sd);
3782 napi->weight = weight_p;
3783 local_irq_disable();
3784 while (work < quota) {
3785 struct sk_buff *skb;
3788 while ((skb = __skb_dequeue(&sd->process_queue))) {
3790 __netif_receive_skb(skb);
3791 local_irq_disable();
3792 input_queue_head_incr(sd);
3793 if (++work >= quota) {
3800 qlen = skb_queue_len(&sd->input_pkt_queue);
3802 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3803 &sd->process_queue);
3805 if (qlen < quota - work) {
3807 * Inline a custom version of __napi_complete().
3808 * only current cpu owns and manipulates this napi,
3809 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3810 * we can use a plain write instead of clear_bit(),
3811 * and we dont need an smp_mb() memory barrier.
3813 list_del(&napi->poll_list);
3816 quota = work + qlen;
3826 * __napi_schedule - schedule for receive
3827 * @n: entry to schedule
3829 * The entry's receive function will be scheduled to run
3831 void __napi_schedule(struct napi_struct *n)
3833 unsigned long flags;
3835 local_irq_save(flags);
3836 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3837 local_irq_restore(flags);
3839 EXPORT_SYMBOL(__napi_schedule);
3841 void __napi_complete(struct napi_struct *n)
3843 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3844 BUG_ON(n->gro_list);
3846 list_del(&n->poll_list);
3847 smp_mb__before_clear_bit();
3848 clear_bit(NAPI_STATE_SCHED, &n->state);
3850 EXPORT_SYMBOL(__napi_complete);
3852 void napi_complete(struct napi_struct *n)
3854 unsigned long flags;
3857 * don't let napi dequeue from the cpu poll list
3858 * just in case its running on a different cpu
3860 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3864 local_irq_save(flags);
3866 local_irq_restore(flags);
3868 EXPORT_SYMBOL(napi_complete);
3870 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3871 int (*poll)(struct napi_struct *, int), int weight)
3873 INIT_LIST_HEAD(&napi->poll_list);
3874 napi->gro_count = 0;
3875 napi->gro_list = NULL;
3878 napi->weight = weight;
3879 list_add(&napi->dev_list, &dev->napi_list);
3881 #ifdef CONFIG_NETPOLL
3882 spin_lock_init(&napi->poll_lock);
3883 napi->poll_owner = -1;
3885 set_bit(NAPI_STATE_SCHED, &napi->state);
3887 EXPORT_SYMBOL(netif_napi_add);
3889 void netif_napi_del(struct napi_struct *napi)
3891 struct sk_buff *skb, *next;
3893 list_del_init(&napi->dev_list);
3894 napi_free_frags(napi);
3896 for (skb = napi->gro_list; skb; skb = next) {
3902 napi->gro_list = NULL;
3903 napi->gro_count = 0;
3905 EXPORT_SYMBOL(netif_napi_del);
3907 static void net_rx_action(struct softirq_action *h)
3909 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3910 unsigned long time_limit = jiffies + 2;
3911 int budget = netdev_budget;
3914 local_irq_disable();
3916 while (!list_empty(&sd->poll_list)) {
3917 struct napi_struct *n;
3920 /* If softirq window is exhuasted then punt.
3921 * Allow this to run for 2 jiffies since which will allow
3922 * an average latency of 1.5/HZ.
3924 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3929 /* Even though interrupts have been re-enabled, this
3930 * access is safe because interrupts can only add new
3931 * entries to the tail of this list, and only ->poll()
3932 * calls can remove this head entry from the list.
3934 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3936 have = netpoll_poll_lock(n);
3940 /* This NAPI_STATE_SCHED test is for avoiding a race
3941 * with netpoll's poll_napi(). Only the entity which
3942 * obtains the lock and sees NAPI_STATE_SCHED set will
3943 * actually make the ->poll() call. Therefore we avoid
3944 * accidentally calling ->poll() when NAPI is not scheduled.
3947 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3948 work = n->poll(n, weight);
3952 WARN_ON_ONCE(work > weight);
3956 local_irq_disable();
3958 /* Drivers must not modify the NAPI state if they
3959 * consume the entire weight. In such cases this code
3960 * still "owns" the NAPI instance and therefore can
3961 * move the instance around on the list at-will.
3963 if (unlikely(work == weight)) {
3964 if (unlikely(napi_disable_pending(n))) {
3967 local_irq_disable();
3969 list_move_tail(&n->poll_list, &sd->poll_list);
3972 netpoll_poll_unlock(have);
3975 net_rps_action_and_irq_enable(sd);
3977 #ifdef CONFIG_NET_DMA
3979 * There may not be any more sk_buffs coming right now, so push
3980 * any pending DMA copies to hardware
3982 dma_issue_pending_all();
3989 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3993 static gifconf_func_t *gifconf_list[NPROTO];
3996 * register_gifconf - register a SIOCGIF handler
3997 * @family: Address family
3998 * @gifconf: Function handler
4000 * Register protocol dependent address dumping routines. The handler
4001 * that is passed must not be freed or reused until it has been replaced
4002 * by another handler.
4004 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4006 if (family >= NPROTO)
4008 gifconf_list[family] = gifconf;
4011 EXPORT_SYMBOL(register_gifconf);
4015 * Map an interface index to its name (SIOCGIFNAME)
4019 * We need this ioctl for efficient implementation of the
4020 * if_indextoname() function required by the IPv6 API. Without
4021 * it, we would have to search all the interfaces to find a
4025 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4027 struct net_device *dev;
4031 * Fetch the caller's info block.
4034 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4038 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4044 strcpy(ifr.ifr_name, dev->name);
4047 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4053 * Perform a SIOCGIFCONF call. This structure will change
4054 * size eventually, and there is nothing I can do about it.
4055 * Thus we will need a 'compatibility mode'.
4058 static int dev_ifconf(struct net *net, char __user *arg)
4061 struct net_device *dev;
4068 * Fetch the caller's info block.
4071 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4078 * Loop over the interfaces, and write an info block for each.
4082 for_each_netdev(net, dev) {
4083 for (i = 0; i < NPROTO; i++) {
4084 if (gifconf_list[i]) {
4087 done = gifconf_list[i](dev, NULL, 0);
4089 done = gifconf_list[i](dev, pos + total,
4099 * All done. Write the updated control block back to the caller.
4101 ifc.ifc_len = total;
4104 * Both BSD and Solaris return 0 here, so we do too.
4106 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4109 #ifdef CONFIG_PROC_FS
4111 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4113 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4114 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4115 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4117 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4119 struct net *net = seq_file_net(seq);
4120 struct net_device *dev;
4121 struct hlist_node *p;
4122 struct hlist_head *h;
4123 unsigned int count = 0, offset = get_offset(*pos);
4125 h = &net->dev_name_head[get_bucket(*pos)];
4126 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4127 if (++count == offset)
4134 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4136 struct net_device *dev;
4137 unsigned int bucket;
4140 dev = dev_from_same_bucket(seq, pos);
4144 bucket = get_bucket(*pos) + 1;
4145 *pos = set_bucket_offset(bucket, 1);
4146 } while (bucket < NETDEV_HASHENTRIES);
4152 * This is invoked by the /proc filesystem handler to display a device
4155 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4160 return SEQ_START_TOKEN;
4162 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4165 return dev_from_bucket(seq, pos);
4168 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4171 return dev_from_bucket(seq, pos);
4174 void dev_seq_stop(struct seq_file *seq, void *v)
4180 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4182 struct rtnl_link_stats64 temp;
4183 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4185 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4186 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4187 dev->name, stats->rx_bytes, stats->rx_packets,
4189 stats->rx_dropped + stats->rx_missed_errors,
4190 stats->rx_fifo_errors,
4191 stats->rx_length_errors + stats->rx_over_errors +
4192 stats->rx_crc_errors + stats->rx_frame_errors,
4193 stats->rx_compressed, stats->multicast,
4194 stats->tx_bytes, stats->tx_packets,
4195 stats->tx_errors, stats->tx_dropped,
4196 stats->tx_fifo_errors, stats->collisions,
4197 stats->tx_carrier_errors +
4198 stats->tx_aborted_errors +
4199 stats->tx_window_errors +
4200 stats->tx_heartbeat_errors,
4201 stats->tx_compressed);
4205 * Called from the PROCfs module. This now uses the new arbitrary sized
4206 * /proc/net interface to create /proc/net/dev
4208 static int dev_seq_show(struct seq_file *seq, void *v)
4210 if (v == SEQ_START_TOKEN)
4211 seq_puts(seq, "Inter-| Receive "
4213 " face |bytes packets errs drop fifo frame "
4214 "compressed multicast|bytes packets errs "
4215 "drop fifo colls carrier compressed\n");
4217 dev_seq_printf_stats(seq, v);
4221 static struct softnet_data *softnet_get_online(loff_t *pos)
4223 struct softnet_data *sd = NULL;
4225 while (*pos < nr_cpu_ids)
4226 if (cpu_online(*pos)) {
4227 sd = &per_cpu(softnet_data, *pos);
4234 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4236 return softnet_get_online(pos);
4239 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4242 return softnet_get_online(pos);
4245 static void softnet_seq_stop(struct seq_file *seq, void *v)
4249 static int softnet_seq_show(struct seq_file *seq, void *v)
4251 struct softnet_data *sd = v;
4253 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4254 sd->processed, sd->dropped, sd->time_squeeze, 0,
4255 0, 0, 0, 0, /* was fastroute */
4256 sd->cpu_collision, sd->received_rps);
4260 static const struct seq_operations dev_seq_ops = {
4261 .start = dev_seq_start,
4262 .next = dev_seq_next,
4263 .stop = dev_seq_stop,
4264 .show = dev_seq_show,
4267 static int dev_seq_open(struct inode *inode, struct file *file)
4269 return seq_open_net(inode, file, &dev_seq_ops,
4270 sizeof(struct seq_net_private));
4273 static const struct file_operations dev_seq_fops = {
4274 .owner = THIS_MODULE,
4275 .open = dev_seq_open,
4277 .llseek = seq_lseek,
4278 .release = seq_release_net,
4281 static const struct seq_operations softnet_seq_ops = {
4282 .start = softnet_seq_start,
4283 .next = softnet_seq_next,
4284 .stop = softnet_seq_stop,
4285 .show = softnet_seq_show,
4288 static int softnet_seq_open(struct inode *inode, struct file *file)
4290 return seq_open(file, &softnet_seq_ops);
4293 static const struct file_operations softnet_seq_fops = {
4294 .owner = THIS_MODULE,
4295 .open = softnet_seq_open,
4297 .llseek = seq_lseek,
4298 .release = seq_release,
4301 static void *ptype_get_idx(loff_t pos)
4303 struct packet_type *pt = NULL;
4307 list_for_each_entry_rcu(pt, &ptype_all, list) {
4313 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4314 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4323 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4327 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4330 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4332 struct packet_type *pt;
4333 struct list_head *nxt;
4337 if (v == SEQ_START_TOKEN)
4338 return ptype_get_idx(0);
4341 nxt = pt->list.next;
4342 if (pt->type == htons(ETH_P_ALL)) {
4343 if (nxt != &ptype_all)
4346 nxt = ptype_base[0].next;
4348 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4350 while (nxt == &ptype_base[hash]) {
4351 if (++hash >= PTYPE_HASH_SIZE)
4353 nxt = ptype_base[hash].next;
4356 return list_entry(nxt, struct packet_type, list);
4359 static void ptype_seq_stop(struct seq_file *seq, void *v)
4365 static int ptype_seq_show(struct seq_file *seq, void *v)
4367 struct packet_type *pt = v;
4369 if (v == SEQ_START_TOKEN)
4370 seq_puts(seq, "Type Device Function\n");
4371 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4372 if (pt->type == htons(ETH_P_ALL))
4373 seq_puts(seq, "ALL ");
4375 seq_printf(seq, "%04x", ntohs(pt->type));
4377 seq_printf(seq, " %-8s %pF\n",
4378 pt->dev ? pt->dev->name : "", pt->func);
4384 static const struct seq_operations ptype_seq_ops = {
4385 .start = ptype_seq_start,
4386 .next = ptype_seq_next,
4387 .stop = ptype_seq_stop,
4388 .show = ptype_seq_show,
4391 static int ptype_seq_open(struct inode *inode, struct file *file)
4393 return seq_open_net(inode, file, &ptype_seq_ops,
4394 sizeof(struct seq_net_private));
4397 static const struct file_operations ptype_seq_fops = {
4398 .owner = THIS_MODULE,
4399 .open = ptype_seq_open,
4401 .llseek = seq_lseek,
4402 .release = seq_release_net,
4406 static int __net_init dev_proc_net_init(struct net *net)
4410 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4412 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4414 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4417 if (wext_proc_init(net))
4423 proc_net_remove(net, "ptype");
4425 proc_net_remove(net, "softnet_stat");
4427 proc_net_remove(net, "dev");
4431 static void __net_exit dev_proc_net_exit(struct net *net)
4433 wext_proc_exit(net);
4435 proc_net_remove(net, "ptype");
4436 proc_net_remove(net, "softnet_stat");
4437 proc_net_remove(net, "dev");
4440 static struct pernet_operations __net_initdata dev_proc_ops = {
4441 .init = dev_proc_net_init,
4442 .exit = dev_proc_net_exit,
4445 static int __init dev_proc_init(void)
4447 return register_pernet_subsys(&dev_proc_ops);
4450 #define dev_proc_init() 0
4451 #endif /* CONFIG_PROC_FS */
4455 * netdev_set_master - set up master pointer
4456 * @slave: slave device
4457 * @master: new master device
4459 * Changes the master device of the slave. Pass %NULL to break the
4460 * bonding. The caller must hold the RTNL semaphore. On a failure
4461 * a negative errno code is returned. On success the reference counts
4462 * are adjusted and the function returns zero.
4464 int netdev_set_master(struct net_device *slave, struct net_device *master)
4466 struct net_device *old = slave->master;
4476 slave->master = master;
4482 EXPORT_SYMBOL(netdev_set_master);
4485 * netdev_set_bond_master - set up bonding master/slave pair
4486 * @slave: slave device
4487 * @master: new master device
4489 * Changes the master device of the slave. Pass %NULL to break the
4490 * bonding. The caller must hold the RTNL semaphore. On a failure
4491 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4492 * to the routing socket and the function returns zero.
4494 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4500 err = netdev_set_master(slave, master);
4504 slave->flags |= IFF_SLAVE;
4506 slave->flags &= ~IFF_SLAVE;
4508 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4511 EXPORT_SYMBOL(netdev_set_bond_master);
4513 static void dev_change_rx_flags(struct net_device *dev, int flags)
4515 const struct net_device_ops *ops = dev->netdev_ops;
4517 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4518 ops->ndo_change_rx_flags(dev, flags);
4521 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4523 unsigned short old_flags = dev->flags;
4529 dev->flags |= IFF_PROMISC;
4530 dev->promiscuity += inc;
4531 if (dev->promiscuity == 0) {
4534 * If inc causes overflow, untouch promisc and return error.
4537 dev->flags &= ~IFF_PROMISC;
4539 dev->promiscuity -= inc;
4540 printk(KERN_WARNING "%s: promiscuity touches roof, "
4541 "set promiscuity failed, promiscuity feature "
4542 "of device might be broken.\n", dev->name);
4546 if (dev->flags != old_flags) {
4547 printk(KERN_INFO "device %s %s promiscuous mode\n",
4548 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4550 if (audit_enabled) {
4551 current_uid_gid(&uid, &gid);
4552 audit_log(current->audit_context, GFP_ATOMIC,
4553 AUDIT_ANOM_PROMISCUOUS,
4554 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4555 dev->name, (dev->flags & IFF_PROMISC),
4556 (old_flags & IFF_PROMISC),
4557 audit_get_loginuid(current),
4559 audit_get_sessionid(current));
4562 dev_change_rx_flags(dev, IFF_PROMISC);
4568 * dev_set_promiscuity - update promiscuity count on a device
4572 * Add or remove promiscuity from a device. While the count in the device
4573 * remains above zero the interface remains promiscuous. Once it hits zero
4574 * the device reverts back to normal filtering operation. A negative inc
4575 * value is used to drop promiscuity on the device.
4576 * Return 0 if successful or a negative errno code on error.
4578 int dev_set_promiscuity(struct net_device *dev, int inc)
4580 unsigned short old_flags = dev->flags;
4583 err = __dev_set_promiscuity(dev, inc);
4586 if (dev->flags != old_flags)
4587 dev_set_rx_mode(dev);
4590 EXPORT_SYMBOL(dev_set_promiscuity);
4593 * dev_set_allmulti - update allmulti count on a device
4597 * Add or remove reception of all multicast frames to a device. While the
4598 * count in the device remains above zero the interface remains listening
4599 * to all interfaces. Once it hits zero the device reverts back to normal
4600 * filtering operation. A negative @inc value is used to drop the counter
4601 * when releasing a resource needing all multicasts.
4602 * Return 0 if successful or a negative errno code on error.
4605 int dev_set_allmulti(struct net_device *dev, int inc)
4607 unsigned short old_flags = dev->flags;
4611 dev->flags |= IFF_ALLMULTI;
4612 dev->allmulti += inc;
4613 if (dev->allmulti == 0) {
4616 * If inc causes overflow, untouch allmulti and return error.
4619 dev->flags &= ~IFF_ALLMULTI;
4621 dev->allmulti -= inc;
4622 printk(KERN_WARNING "%s: allmulti touches roof, "
4623 "set allmulti failed, allmulti feature of "
4624 "device might be broken.\n", dev->name);
4628 if (dev->flags ^ old_flags) {
4629 dev_change_rx_flags(dev, IFF_ALLMULTI);
4630 dev_set_rx_mode(dev);
4634 EXPORT_SYMBOL(dev_set_allmulti);
4637 * Upload unicast and multicast address lists to device and
4638 * configure RX filtering. When the device doesn't support unicast
4639 * filtering it is put in promiscuous mode while unicast addresses
4642 void __dev_set_rx_mode(struct net_device *dev)
4644 const struct net_device_ops *ops = dev->netdev_ops;
4646 /* dev_open will call this function so the list will stay sane. */
4647 if (!(dev->flags&IFF_UP))
4650 if (!netif_device_present(dev))
4653 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4654 /* Unicast addresses changes may only happen under the rtnl,
4655 * therefore calling __dev_set_promiscuity here is safe.
4657 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4658 __dev_set_promiscuity(dev, 1);
4659 dev->uc_promisc = true;
4660 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4661 __dev_set_promiscuity(dev, -1);
4662 dev->uc_promisc = false;
4666 if (ops->ndo_set_rx_mode)
4667 ops->ndo_set_rx_mode(dev);
4670 void dev_set_rx_mode(struct net_device *dev)
4672 netif_addr_lock_bh(dev);
4673 __dev_set_rx_mode(dev);
4674 netif_addr_unlock_bh(dev);
4678 * dev_get_flags - get flags reported to userspace
4681 * Get the combination of flag bits exported through APIs to userspace.
4683 unsigned dev_get_flags(const struct net_device *dev)
4687 flags = (dev->flags & ~(IFF_PROMISC |
4692 (dev->gflags & (IFF_PROMISC |
4695 if (netif_running(dev)) {
4696 if (netif_oper_up(dev))
4697 flags |= IFF_RUNNING;
4698 if (netif_carrier_ok(dev))
4699 flags |= IFF_LOWER_UP;
4700 if (netif_dormant(dev))
4701 flags |= IFF_DORMANT;
4706 EXPORT_SYMBOL(dev_get_flags);
4708 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4710 int old_flags = dev->flags;
4716 * Set the flags on our device.
4719 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4720 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4722 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4726 * Load in the correct multicast list now the flags have changed.
4729 if ((old_flags ^ flags) & IFF_MULTICAST)
4730 dev_change_rx_flags(dev, IFF_MULTICAST);
4732 dev_set_rx_mode(dev);
4735 * Have we downed the interface. We handle IFF_UP ourselves
4736 * according to user attempts to set it, rather than blindly
4741 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4742 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4745 dev_set_rx_mode(dev);
4748 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4749 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4751 dev->gflags ^= IFF_PROMISC;
4752 dev_set_promiscuity(dev, inc);
4755 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4756 is important. Some (broken) drivers set IFF_PROMISC, when
4757 IFF_ALLMULTI is requested not asking us and not reporting.
4759 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4760 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4762 dev->gflags ^= IFF_ALLMULTI;
4763 dev_set_allmulti(dev, inc);
4769 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4771 unsigned int changes = dev->flags ^ old_flags;
4773 if (changes & IFF_UP) {
4774 if (dev->flags & IFF_UP)
4775 call_netdevice_notifiers(NETDEV_UP, dev);
4777 call_netdevice_notifiers(NETDEV_DOWN, dev);
4780 if (dev->flags & IFF_UP &&
4781 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4782 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4786 * dev_change_flags - change device settings
4788 * @flags: device state flags
4790 * Change settings on device based state flags. The flags are
4791 * in the userspace exported format.
4793 int dev_change_flags(struct net_device *dev, unsigned flags)
4796 int old_flags = dev->flags;
4798 ret = __dev_change_flags(dev, flags);
4802 changes = old_flags ^ dev->flags;
4804 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4806 __dev_notify_flags(dev, old_flags);
4809 EXPORT_SYMBOL(dev_change_flags);
4812 * dev_set_mtu - Change maximum transfer unit
4814 * @new_mtu: new transfer unit
4816 * Change the maximum transfer size of the network device.
4818 int dev_set_mtu(struct net_device *dev, int new_mtu)
4820 const struct net_device_ops *ops = dev->netdev_ops;
4823 if (new_mtu == dev->mtu)
4826 /* MTU must be positive. */
4830 if (!netif_device_present(dev))
4834 if (ops->ndo_change_mtu)
4835 err = ops->ndo_change_mtu(dev, new_mtu);
4839 if (!err && dev->flags & IFF_UP)
4840 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4843 EXPORT_SYMBOL(dev_set_mtu);
4846 * dev_set_group - Change group this device belongs to
4848 * @new_group: group this device should belong to
4850 void dev_set_group(struct net_device *dev, int new_group)
4852 dev->group = new_group;
4854 EXPORT_SYMBOL(dev_set_group);
4857 * dev_set_mac_address - Change Media Access Control Address
4861 * Change the hardware (MAC) address of the device
4863 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4865 const struct net_device_ops *ops = dev->netdev_ops;
4868 if (!ops->ndo_set_mac_address)
4870 if (sa->sa_family != dev->type)
4872 if (!netif_device_present(dev))
4874 err = ops->ndo_set_mac_address(dev, sa);
4876 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4877 add_device_randomness(dev->dev_addr, dev->addr_len);
4880 EXPORT_SYMBOL(dev_set_mac_address);
4883 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4885 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4888 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4894 case SIOCGIFFLAGS: /* Get interface flags */
4895 ifr->ifr_flags = (short) dev_get_flags(dev);
4898 case SIOCGIFMETRIC: /* Get the metric on the interface
4899 (currently unused) */
4900 ifr->ifr_metric = 0;
4903 case SIOCGIFMTU: /* Get the MTU of a device */
4904 ifr->ifr_mtu = dev->mtu;
4909 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4911 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4912 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4913 ifr->ifr_hwaddr.sa_family = dev->type;
4921 ifr->ifr_map.mem_start = dev->mem_start;
4922 ifr->ifr_map.mem_end = dev->mem_end;
4923 ifr->ifr_map.base_addr = dev->base_addr;
4924 ifr->ifr_map.irq = dev->irq;
4925 ifr->ifr_map.dma = dev->dma;
4926 ifr->ifr_map.port = dev->if_port;
4930 ifr->ifr_ifindex = dev->ifindex;
4934 ifr->ifr_qlen = dev->tx_queue_len;
4938 /* dev_ioctl() should ensure this case
4950 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4952 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4955 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4956 const struct net_device_ops *ops;
4961 ops = dev->netdev_ops;
4964 case SIOCSIFFLAGS: /* Set interface flags */
4965 return dev_change_flags(dev, ifr->ifr_flags);
4967 case SIOCSIFMETRIC: /* Set the metric on the interface
4968 (currently unused) */
4971 case SIOCSIFMTU: /* Set the MTU of a device */
4972 return dev_set_mtu(dev, ifr->ifr_mtu);
4975 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4977 case SIOCSIFHWBROADCAST:
4978 if (ifr->ifr_hwaddr.sa_family != dev->type)
4980 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4981 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4982 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4986 if (ops->ndo_set_config) {
4987 if (!netif_device_present(dev))
4989 return ops->ndo_set_config(dev, &ifr->ifr_map);
4994 if (!ops->ndo_set_rx_mode ||
4995 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4997 if (!netif_device_present(dev))
4999 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5002 if (!ops->ndo_set_rx_mode ||
5003 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5005 if (!netif_device_present(dev))
5007 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5010 if (ifr->ifr_qlen < 0)
5012 dev->tx_queue_len = ifr->ifr_qlen;
5016 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5017 return dev_change_name(dev, ifr->ifr_newname);
5020 err = net_hwtstamp_validate(ifr);
5026 * Unknown or private ioctl
5029 if ((cmd >= SIOCDEVPRIVATE &&
5030 cmd <= SIOCDEVPRIVATE + 15) ||
5031 cmd == SIOCBONDENSLAVE ||
5032 cmd == SIOCBONDRELEASE ||
5033 cmd == SIOCBONDSETHWADDR ||
5034 cmd == SIOCBONDSLAVEINFOQUERY ||
5035 cmd == SIOCBONDINFOQUERY ||
5036 cmd == SIOCBONDCHANGEACTIVE ||
5037 cmd == SIOCGMIIPHY ||
5038 cmd == SIOCGMIIREG ||
5039 cmd == SIOCSMIIREG ||
5040 cmd == SIOCBRADDIF ||
5041 cmd == SIOCBRDELIF ||
5042 cmd == SIOCSHWTSTAMP ||
5043 cmd == SIOCWANDEV) {
5045 if (ops->ndo_do_ioctl) {
5046 if (netif_device_present(dev))
5047 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5059 * This function handles all "interface"-type I/O control requests. The actual
5060 * 'doing' part of this is dev_ifsioc above.
5064 * dev_ioctl - network device ioctl
5065 * @net: the applicable net namespace
5066 * @cmd: command to issue
5067 * @arg: pointer to a struct ifreq in user space
5069 * Issue ioctl functions to devices. This is normally called by the
5070 * user space syscall interfaces but can sometimes be useful for
5071 * other purposes. The return value is the return from the syscall if
5072 * positive or a negative errno code on error.
5075 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5081 /* One special case: SIOCGIFCONF takes ifconf argument
5082 and requires shared lock, because it sleeps writing
5086 if (cmd == SIOCGIFCONF) {
5088 ret = dev_ifconf(net, (char __user *) arg);
5092 if (cmd == SIOCGIFNAME)
5093 return dev_ifname(net, (struct ifreq __user *)arg);
5095 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5098 ifr.ifr_name[IFNAMSIZ-1] = 0;
5100 colon = strchr(ifr.ifr_name, ':');
5105 * See which interface the caller is talking about.
5110 * These ioctl calls:
5111 * - can be done by all.
5112 * - atomic and do not require locking.
5123 dev_load(net, ifr.ifr_name);
5125 ret = dev_ifsioc_locked(net, &ifr, cmd);
5130 if (copy_to_user(arg, &ifr,
5131 sizeof(struct ifreq)))
5137 dev_load(net, ifr.ifr_name);
5139 ret = dev_ethtool(net, &ifr);
5144 if (copy_to_user(arg, &ifr,
5145 sizeof(struct ifreq)))
5151 * These ioctl calls:
5152 * - require superuser power.
5153 * - require strict serialization.
5159 if (!capable(CAP_NET_ADMIN))
5161 dev_load(net, ifr.ifr_name);
5163 ret = dev_ifsioc(net, &ifr, cmd);
5168 if (copy_to_user(arg, &ifr,
5169 sizeof(struct ifreq)))
5175 * These ioctl calls:
5176 * - require superuser power.
5177 * - require strict serialization.
5178 * - do not return a value
5188 case SIOCSIFHWBROADCAST:
5191 case SIOCBONDENSLAVE:
5192 case SIOCBONDRELEASE:
5193 case SIOCBONDSETHWADDR:
5194 case SIOCBONDCHANGEACTIVE:
5198 if (!capable(CAP_NET_ADMIN))
5201 case SIOCBONDSLAVEINFOQUERY:
5202 case SIOCBONDINFOQUERY:
5203 dev_load(net, ifr.ifr_name);
5205 ret = dev_ifsioc(net, &ifr, cmd);
5210 /* Get the per device memory space. We can add this but
5211 * currently do not support it */
5213 /* Set the per device memory buffer space.
5214 * Not applicable in our case */
5219 * Unknown or private ioctl.
5222 if (cmd == SIOCWANDEV ||
5223 (cmd >= SIOCDEVPRIVATE &&
5224 cmd <= SIOCDEVPRIVATE + 15)) {
5225 dev_load(net, ifr.ifr_name);
5227 ret = dev_ifsioc(net, &ifr, cmd);
5229 if (!ret && copy_to_user(arg, &ifr,
5230 sizeof(struct ifreq)))
5234 /* Take care of Wireless Extensions */
5235 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5236 return wext_handle_ioctl(net, &ifr, cmd, arg);
5243 * dev_new_index - allocate an ifindex
5244 * @net: the applicable net namespace
5246 * Returns a suitable unique value for a new device interface
5247 * number. The caller must hold the rtnl semaphore or the
5248 * dev_base_lock to be sure it remains unique.
5250 static int dev_new_index(struct net *net)
5256 if (!__dev_get_by_index(net, ifindex))
5261 /* Delayed registration/unregisteration */
5262 static LIST_HEAD(net_todo_list);
5264 static void net_set_todo(struct net_device *dev)
5266 list_add_tail(&dev->todo_list, &net_todo_list);
5269 static void rollback_registered_many(struct list_head *head)
5271 struct net_device *dev, *tmp;
5273 BUG_ON(dev_boot_phase);
5276 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5277 /* Some devices call without registering
5278 * for initialization unwind. Remove those
5279 * devices and proceed with the remaining.
5281 if (dev->reg_state == NETREG_UNINITIALIZED) {
5282 pr_debug("unregister_netdevice: device %s/%p never "
5283 "was registered\n", dev->name, dev);
5286 list_del(&dev->unreg_list);
5289 dev->dismantle = true;
5290 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5293 /* If device is running, close it first. */
5294 dev_close_many(head);
5296 list_for_each_entry(dev, head, unreg_list) {
5297 /* And unlink it from device chain. */
5298 unlist_netdevice(dev);
5300 dev->reg_state = NETREG_UNREGISTERING;
5305 list_for_each_entry(dev, head, unreg_list) {
5306 /* Shutdown queueing discipline. */
5310 /* Notify protocols, that we are about to destroy
5311 this device. They should clean all the things.
5313 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5315 if (!dev->rtnl_link_ops ||
5316 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5317 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5320 * Flush the unicast and multicast chains
5325 if (dev->netdev_ops->ndo_uninit)
5326 dev->netdev_ops->ndo_uninit(dev);
5328 /* Notifier chain MUST detach us from master device. */
5329 WARN_ON(dev->master);
5331 /* Remove entries from kobject tree */
5332 netdev_unregister_kobject(dev);
5335 /* Process any work delayed until the end of the batch */
5336 dev = list_first_entry(head, struct net_device, unreg_list);
5337 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5341 list_for_each_entry(dev, head, unreg_list)
5345 static void rollback_registered(struct net_device *dev)
5349 list_add(&dev->unreg_list, &single);
5350 rollback_registered_many(&single);
5354 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5356 /* Fix illegal checksum combinations */
5357 if ((features & NETIF_F_HW_CSUM) &&
5358 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5359 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5360 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5363 if ((features & NETIF_F_NO_CSUM) &&
5364 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5365 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5366 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5369 /* Fix illegal SG+CSUM combinations. */
5370 if ((features & NETIF_F_SG) &&
5371 !(features & NETIF_F_ALL_CSUM)) {
5373 "Dropping NETIF_F_SG since no checksum feature.\n");
5374 features &= ~NETIF_F_SG;
5377 /* TSO requires that SG is present as well. */
5378 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5379 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5380 features &= ~NETIF_F_ALL_TSO;
5383 /* TSO ECN requires that TSO is present as well. */
5384 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5385 features &= ~NETIF_F_TSO_ECN;
5387 /* Software GSO depends on SG. */
5388 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5389 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5390 features &= ~NETIF_F_GSO;
5393 /* UFO needs SG and checksumming */
5394 if (features & NETIF_F_UFO) {
5395 /* maybe split UFO into V4 and V6? */
5396 if (!((features & NETIF_F_GEN_CSUM) ||
5397 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5398 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5400 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5401 features &= ~NETIF_F_UFO;
5404 if (!(features & NETIF_F_SG)) {
5406 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5407 features &= ~NETIF_F_UFO;
5414 int __netdev_update_features(struct net_device *dev)
5421 features = netdev_get_wanted_features(dev);
5423 if (dev->netdev_ops->ndo_fix_features)
5424 features = dev->netdev_ops->ndo_fix_features(dev, features);
5426 /* driver might be less strict about feature dependencies */
5427 features = netdev_fix_features(dev, features);
5429 if (dev->features == features)
5432 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5433 dev->features, features);
5435 if (dev->netdev_ops->ndo_set_features)
5436 err = dev->netdev_ops->ndo_set_features(dev, features);
5438 if (unlikely(err < 0)) {
5440 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5441 err, features, dev->features);
5446 dev->features = features;
5452 * netdev_update_features - recalculate device features
5453 * @dev: the device to check
5455 * Recalculate dev->features set and send notifications if it
5456 * has changed. Should be called after driver or hardware dependent
5457 * conditions might have changed that influence the features.
5459 void netdev_update_features(struct net_device *dev)
5461 if (__netdev_update_features(dev))
5462 netdev_features_change(dev);
5464 EXPORT_SYMBOL(netdev_update_features);
5467 * netdev_change_features - recalculate device features
5468 * @dev: the device to check
5470 * Recalculate dev->features set and send notifications even
5471 * if they have not changed. Should be called instead of
5472 * netdev_update_features() if also dev->vlan_features might
5473 * have changed to allow the changes to be propagated to stacked
5476 void netdev_change_features(struct net_device *dev)
5478 __netdev_update_features(dev);
5479 netdev_features_change(dev);
5481 EXPORT_SYMBOL(netdev_change_features);
5484 * netif_stacked_transfer_operstate - transfer operstate
5485 * @rootdev: the root or lower level device to transfer state from
5486 * @dev: the device to transfer operstate to
5488 * Transfer operational state from root to device. This is normally
5489 * called when a stacking relationship exists between the root
5490 * device and the device(a leaf device).
5492 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5493 struct net_device *dev)
5495 if (rootdev->operstate == IF_OPER_DORMANT)
5496 netif_dormant_on(dev);
5498 netif_dormant_off(dev);
5500 if (netif_carrier_ok(rootdev)) {
5501 if (!netif_carrier_ok(dev))
5502 netif_carrier_on(dev);
5504 if (netif_carrier_ok(dev))
5505 netif_carrier_off(dev);
5508 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5511 static int netif_alloc_rx_queues(struct net_device *dev)
5513 unsigned int i, count = dev->num_rx_queues;
5514 struct netdev_rx_queue *rx;
5518 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5520 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5525 for (i = 0; i < count; i++)
5531 static void netdev_init_one_queue(struct net_device *dev,
5532 struct netdev_queue *queue, void *_unused)
5534 /* Initialize queue lock */
5535 spin_lock_init(&queue->_xmit_lock);
5536 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5537 queue->xmit_lock_owner = -1;
5538 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5542 static int netif_alloc_netdev_queues(struct net_device *dev)
5544 unsigned int count = dev->num_tx_queues;
5545 struct netdev_queue *tx;
5549 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5551 pr_err("netdev: Unable to allocate %u tx queues.\n",
5557 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5558 spin_lock_init(&dev->tx_global_lock);
5564 * register_netdevice - register a network device
5565 * @dev: device to register
5567 * Take a completed network device structure and add it to the kernel
5568 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5569 * chain. 0 is returned on success. A negative errno code is returned
5570 * on a failure to set up the device, or if the name is a duplicate.
5572 * Callers must hold the rtnl semaphore. You may want
5573 * register_netdev() instead of this.
5576 * The locking appears insufficient to guarantee two parallel registers
5577 * will not get the same name.
5580 int register_netdevice(struct net_device *dev)
5583 struct net *net = dev_net(dev);
5585 BUG_ON(dev_boot_phase);
5590 /* When net_device's are persistent, this will be fatal. */
5591 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5594 spin_lock_init(&dev->addr_list_lock);
5595 netdev_set_addr_lockdep_class(dev);
5599 ret = dev_get_valid_name(dev, dev->name);
5603 /* Init, if this function is available */
5604 if (dev->netdev_ops->ndo_init) {
5605 ret = dev->netdev_ops->ndo_init(dev);
5613 dev->ifindex = dev_new_index(net);
5614 if (dev->iflink == -1)
5615 dev->iflink = dev->ifindex;
5617 /* Transfer changeable features to wanted_features and enable
5618 * software offloads (GSO and GRO).
5620 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5621 dev->features |= NETIF_F_SOFT_FEATURES;
5622 dev->wanted_features = dev->features & dev->hw_features;
5624 /* Turn on no cache copy if HW is doing checksum */
5625 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5626 if ((dev->features & NETIF_F_ALL_CSUM) &&
5627 !(dev->features & NETIF_F_NO_CSUM)) {
5628 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5629 dev->features |= NETIF_F_NOCACHE_COPY;
5632 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5634 dev->vlan_features |= NETIF_F_HIGHDMA;
5636 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5637 ret = notifier_to_errno(ret);
5641 ret = netdev_register_kobject(dev);
5644 dev->reg_state = NETREG_REGISTERED;
5646 __netdev_update_features(dev);
5649 * Default initial state at registry is that the
5650 * device is present.
5653 set_bit(__LINK_STATE_PRESENT, &dev->state);
5655 dev_init_scheduler(dev);
5657 list_netdevice(dev);
5658 add_device_randomness(dev->dev_addr, dev->addr_len);
5660 /* Notify protocols, that a new device appeared. */
5661 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5662 ret = notifier_to_errno(ret);
5664 rollback_registered(dev);
5665 dev->reg_state = NETREG_UNREGISTERED;
5668 * Prevent userspace races by waiting until the network
5669 * device is fully setup before sending notifications.
5671 if (!dev->rtnl_link_ops ||
5672 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5673 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5679 if (dev->netdev_ops->ndo_uninit)
5680 dev->netdev_ops->ndo_uninit(dev);
5683 EXPORT_SYMBOL(register_netdevice);
5686 * init_dummy_netdev - init a dummy network device for NAPI
5687 * @dev: device to init
5689 * This takes a network device structure and initialize the minimum
5690 * amount of fields so it can be used to schedule NAPI polls without
5691 * registering a full blown interface. This is to be used by drivers
5692 * that need to tie several hardware interfaces to a single NAPI
5693 * poll scheduler due to HW limitations.
5695 int init_dummy_netdev(struct net_device *dev)
5697 /* Clear everything. Note we don't initialize spinlocks
5698 * are they aren't supposed to be taken by any of the
5699 * NAPI code and this dummy netdev is supposed to be
5700 * only ever used for NAPI polls
5702 memset(dev, 0, sizeof(struct net_device));
5704 /* make sure we BUG if trying to hit standard
5705 * register/unregister code path
5707 dev->reg_state = NETREG_DUMMY;
5709 /* NAPI wants this */
5710 INIT_LIST_HEAD(&dev->napi_list);
5712 /* a dummy interface is started by default */
5713 set_bit(__LINK_STATE_PRESENT, &dev->state);
5714 set_bit(__LINK_STATE_START, &dev->state);
5716 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5717 * because users of this 'device' dont need to change
5723 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5727 * register_netdev - register a network device
5728 * @dev: device to register
5730 * Take a completed network device structure and add it to the kernel
5731 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5732 * chain. 0 is returned on success. A negative errno code is returned
5733 * on a failure to set up the device, or if the name is a duplicate.
5735 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5736 * and expands the device name if you passed a format string to
5739 int register_netdev(struct net_device *dev)
5744 err = register_netdevice(dev);
5748 EXPORT_SYMBOL(register_netdev);
5750 int netdev_refcnt_read(const struct net_device *dev)
5754 for_each_possible_cpu(i)
5755 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5758 EXPORT_SYMBOL(netdev_refcnt_read);
5761 * netdev_wait_allrefs - wait until all references are gone.
5763 * This is called when unregistering network devices.
5765 * Any protocol or device that holds a reference should register
5766 * for netdevice notification, and cleanup and put back the
5767 * reference if they receive an UNREGISTER event.
5768 * We can get stuck here if buggy protocols don't correctly
5771 static void netdev_wait_allrefs(struct net_device *dev)
5773 unsigned long rebroadcast_time, warning_time;
5776 linkwatch_forget_dev(dev);
5778 rebroadcast_time = warning_time = jiffies;
5779 refcnt = netdev_refcnt_read(dev);
5781 while (refcnt != 0) {
5782 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5785 /* Rebroadcast unregister notification */
5786 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5787 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5788 * should have already handle it the first time */
5790 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5792 /* We must not have linkwatch events
5793 * pending on unregister. If this
5794 * happens, we simply run the queue
5795 * unscheduled, resulting in a noop
5798 linkwatch_run_queue();
5803 rebroadcast_time = jiffies;
5808 refcnt = netdev_refcnt_read(dev);
5810 if (time_after(jiffies, warning_time + 10 * HZ)) {
5811 printk(KERN_EMERG "unregister_netdevice: "
5812 "waiting for %s to become free. Usage "
5815 warning_time = jiffies;
5824 * register_netdevice(x1);
5825 * register_netdevice(x2);
5827 * unregister_netdevice(y1);
5828 * unregister_netdevice(y2);
5834 * We are invoked by rtnl_unlock().
5835 * This allows us to deal with problems:
5836 * 1) We can delete sysfs objects which invoke hotplug
5837 * without deadlocking with linkwatch via keventd.
5838 * 2) Since we run with the RTNL semaphore not held, we can sleep
5839 * safely in order to wait for the netdev refcnt to drop to zero.
5841 * We must not return until all unregister events added during
5842 * the interval the lock was held have been completed.
5844 void netdev_run_todo(void)
5846 struct list_head list;
5848 /* Snapshot list, allow later requests */
5849 list_replace_init(&net_todo_list, &list);
5853 /* Wait for rcu callbacks to finish before attempting to drain
5854 * the device list. This usually avoids a 250ms wait.
5856 if (!list_empty(&list))
5859 while (!list_empty(&list)) {
5860 struct net_device *dev
5861 = list_first_entry(&list, struct net_device, todo_list);
5862 list_del(&dev->todo_list);
5864 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5865 printk(KERN_ERR "network todo '%s' but state %d\n",
5866 dev->name, dev->reg_state);
5871 dev->reg_state = NETREG_UNREGISTERED;
5873 on_each_cpu(flush_backlog, dev, 1);
5875 netdev_wait_allrefs(dev);
5878 BUG_ON(netdev_refcnt_read(dev));
5879 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5880 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5881 WARN_ON(dev->dn_ptr);
5883 if (dev->destructor)
5884 dev->destructor(dev);
5886 /* Free network device */
5887 kobject_put(&dev->dev.kobj);
5891 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5892 * fields in the same order, with only the type differing.
5894 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5895 const struct net_device_stats *netdev_stats)
5897 #if BITS_PER_LONG == 64
5898 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5899 memcpy(stats64, netdev_stats, sizeof(*stats64));
5901 size_t i, n = sizeof(*stats64) / sizeof(u64);
5902 const unsigned long *src = (const unsigned long *)netdev_stats;
5903 u64 *dst = (u64 *)stats64;
5905 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5906 sizeof(*stats64) / sizeof(u64));
5907 for (i = 0; i < n; i++)
5913 * dev_get_stats - get network device statistics
5914 * @dev: device to get statistics from
5915 * @storage: place to store stats
5917 * Get network statistics from device. Return @storage.
5918 * The device driver may provide its own method by setting
5919 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5920 * otherwise the internal statistics structure is used.
5922 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5923 struct rtnl_link_stats64 *storage)
5925 const struct net_device_ops *ops = dev->netdev_ops;
5927 if (ops->ndo_get_stats64) {
5928 memset(storage, 0, sizeof(*storage));
5929 ops->ndo_get_stats64(dev, storage);
5930 } else if (ops->ndo_get_stats) {
5931 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5933 netdev_stats_to_stats64(storage, &dev->stats);
5935 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5938 EXPORT_SYMBOL(dev_get_stats);
5940 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5942 struct netdev_queue *queue = dev_ingress_queue(dev);
5944 #ifdef CONFIG_NET_CLS_ACT
5947 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5950 netdev_init_one_queue(dev, queue, NULL);
5951 queue->qdisc = &noop_qdisc;
5952 queue->qdisc_sleeping = &noop_qdisc;
5953 rcu_assign_pointer(dev->ingress_queue, queue);
5959 * alloc_netdev_mqs - allocate network device
5960 * @sizeof_priv: size of private data to allocate space for
5961 * @name: device name format string
5962 * @setup: callback to initialize device
5963 * @txqs: the number of TX subqueues to allocate
5964 * @rxqs: the number of RX subqueues to allocate
5966 * Allocates a struct net_device with private data area for driver use
5967 * and performs basic initialization. Also allocates subquue structs
5968 * for each queue on the device.
5970 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5971 void (*setup)(struct net_device *),
5972 unsigned int txqs, unsigned int rxqs)
5974 struct net_device *dev;
5976 struct net_device *p;
5978 BUG_ON(strlen(name) >= sizeof(dev->name));
5981 pr_err("alloc_netdev: Unable to allocate device "
5982 "with zero queues.\n");
5988 pr_err("alloc_netdev: Unable to allocate device "
5989 "with zero RX queues.\n");
5994 alloc_size = sizeof(struct net_device);
5996 /* ensure 32-byte alignment of private area */
5997 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5998 alloc_size += sizeof_priv;
6000 /* ensure 32-byte alignment of whole construct */
6001 alloc_size += NETDEV_ALIGN - 1;
6003 p = kzalloc(alloc_size, GFP_KERNEL);
6005 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
6009 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6010 dev->padded = (char *)dev - (char *)p;
6012 dev->pcpu_refcnt = alloc_percpu(int);
6013 if (!dev->pcpu_refcnt)
6016 if (dev_addr_init(dev))
6022 dev_net_set(dev, &init_net);
6024 dev->gso_max_size = GSO_MAX_SIZE;
6025 dev->gso_max_segs = GSO_MAX_SEGS;
6027 INIT_LIST_HEAD(&dev->napi_list);
6028 INIT_LIST_HEAD(&dev->unreg_list);
6029 INIT_LIST_HEAD(&dev->link_watch_list);
6030 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6033 dev->num_tx_queues = txqs;
6034 dev->real_num_tx_queues = txqs;
6035 if (netif_alloc_netdev_queues(dev))
6039 dev->num_rx_queues = rxqs;
6040 dev->real_num_rx_queues = rxqs;
6041 if (netif_alloc_rx_queues(dev))
6045 strcpy(dev->name, name);
6046 dev->group = INIT_NETDEV_GROUP;
6054 free_percpu(dev->pcpu_refcnt);
6064 EXPORT_SYMBOL(alloc_netdev_mqs);
6067 * free_netdev - free network device
6070 * This function does the last stage of destroying an allocated device
6071 * interface. The reference to the device object is released.
6072 * If this is the last reference then it will be freed.
6074 void free_netdev(struct net_device *dev)
6076 struct napi_struct *p, *n;
6078 release_net(dev_net(dev));
6085 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6087 /* Flush device addresses */
6088 dev_addr_flush(dev);
6090 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6093 free_percpu(dev->pcpu_refcnt);
6094 dev->pcpu_refcnt = NULL;
6096 /* Compatibility with error handling in drivers */
6097 if (dev->reg_state == NETREG_UNINITIALIZED) {
6098 kfree((char *)dev - dev->padded);
6102 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6103 dev->reg_state = NETREG_RELEASED;
6105 /* will free via device release */
6106 put_device(&dev->dev);
6108 EXPORT_SYMBOL(free_netdev);
6111 * synchronize_net - Synchronize with packet receive processing
6113 * Wait for packets currently being received to be done.
6114 * Does not block later packets from starting.
6116 void synchronize_net(void)
6119 if (rtnl_is_locked())
6120 synchronize_rcu_expedited();
6124 EXPORT_SYMBOL(synchronize_net);
6127 * unregister_netdevice_queue - remove device from the kernel
6131 * This function shuts down a device interface and removes it
6132 * from the kernel tables.
6133 * If head not NULL, device is queued to be unregistered later.
6135 * Callers must hold the rtnl semaphore. You may want
6136 * unregister_netdev() instead of this.
6139 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6144 list_move_tail(&dev->unreg_list, head);
6146 rollback_registered(dev);
6147 /* Finish processing unregister after unlock */
6151 EXPORT_SYMBOL(unregister_netdevice_queue);
6154 * unregister_netdevice_many - unregister many devices
6155 * @head: list of devices
6157 void unregister_netdevice_many(struct list_head *head)
6159 struct net_device *dev;
6161 if (!list_empty(head)) {
6162 rollback_registered_many(head);
6163 list_for_each_entry(dev, head, unreg_list)
6167 EXPORT_SYMBOL(unregister_netdevice_many);
6170 * unregister_netdev - remove device from the kernel
6173 * This function shuts down a device interface and removes it
6174 * from the kernel tables.
6176 * This is just a wrapper for unregister_netdevice that takes
6177 * the rtnl semaphore. In general you want to use this and not
6178 * unregister_netdevice.
6180 void unregister_netdev(struct net_device *dev)
6183 unregister_netdevice(dev);
6186 EXPORT_SYMBOL(unregister_netdev);
6189 * dev_change_net_namespace - move device to different nethost namespace
6191 * @net: network namespace
6192 * @pat: If not NULL name pattern to try if the current device name
6193 * is already taken in the destination network namespace.
6195 * This function shuts down a device interface and moves it
6196 * to a new network namespace. On success 0 is returned, on
6197 * a failure a netagive errno code is returned.
6199 * Callers must hold the rtnl semaphore.
6202 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6208 /* Don't allow namespace local devices to be moved. */
6210 if (dev->features & NETIF_F_NETNS_LOCAL)
6213 /* Ensure the device has been registrered */
6215 if (dev->reg_state != NETREG_REGISTERED)
6218 /* Get out if there is nothing todo */
6220 if (net_eq(dev_net(dev), net))
6223 /* Pick the destination device name, and ensure
6224 * we can use it in the destination network namespace.
6227 if (__dev_get_by_name(net, dev->name)) {
6228 /* We get here if we can't use the current device name */
6231 if (dev_get_valid_name(dev, pat) < 0)
6236 * And now a mini version of register_netdevice unregister_netdevice.
6239 /* If device is running close it first. */
6242 /* And unlink it from device chain */
6244 unlist_netdevice(dev);
6248 /* Shutdown queueing discipline. */
6251 /* Notify protocols, that we are about to destroy
6252 this device. They should clean all the things.
6254 Note that dev->reg_state stays at NETREG_REGISTERED.
6255 This is wanted because this way 8021q and macvlan know
6256 the device is just moving and can keep their slaves up.
6258 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6259 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6260 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6263 * Flush the unicast and multicast chains
6268 /* Actually switch the network namespace */
6269 dev_net_set(dev, net);
6271 /* If there is an ifindex conflict assign a new one */
6272 if (__dev_get_by_index(net, dev->ifindex)) {
6273 int iflink = (dev->iflink == dev->ifindex);
6274 dev->ifindex = dev_new_index(net);
6276 dev->iflink = dev->ifindex;
6279 /* Fixup kobjects */
6280 err = device_rename(&dev->dev, dev->name);
6283 /* Add the device back in the hashes */
6284 list_netdevice(dev);
6286 /* Notify protocols, that a new device appeared. */
6287 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6290 * Prevent userspace races by waiting until the network
6291 * device is fully setup before sending notifications.
6293 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6300 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6302 static int dev_cpu_callback(struct notifier_block *nfb,
6303 unsigned long action,
6306 struct sk_buff **list_skb;
6307 struct sk_buff *skb;
6308 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6309 struct softnet_data *sd, *oldsd;
6311 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6314 local_irq_disable();
6315 cpu = smp_processor_id();
6316 sd = &per_cpu(softnet_data, cpu);
6317 oldsd = &per_cpu(softnet_data, oldcpu);
6319 /* Find end of our completion_queue. */
6320 list_skb = &sd->completion_queue;
6322 list_skb = &(*list_skb)->next;
6323 /* Append completion queue from offline CPU. */
6324 *list_skb = oldsd->completion_queue;
6325 oldsd->completion_queue = NULL;
6327 /* Append output queue from offline CPU. */
6328 if (oldsd->output_queue) {
6329 *sd->output_queue_tailp = oldsd->output_queue;
6330 sd->output_queue_tailp = oldsd->output_queue_tailp;
6331 oldsd->output_queue = NULL;
6332 oldsd->output_queue_tailp = &oldsd->output_queue;
6334 /* Append NAPI poll list from offline CPU. */
6335 if (!list_empty(&oldsd->poll_list)) {
6336 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6337 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6340 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6343 /* Process offline CPU's input_pkt_queue */
6344 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6346 input_queue_head_incr(oldsd);
6348 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6350 input_queue_head_incr(oldsd);
6358 * netdev_increment_features - increment feature set by one
6359 * @all: current feature set
6360 * @one: new feature set
6361 * @mask: mask feature set
6363 * Computes a new feature set after adding a device with feature set
6364 * @one to the master device with current feature set @all. Will not
6365 * enable anything that is off in @mask. Returns the new feature set.
6367 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6369 if (mask & NETIF_F_GEN_CSUM)
6370 mask |= NETIF_F_ALL_CSUM;
6371 mask |= NETIF_F_VLAN_CHALLENGED;
6373 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6374 all &= one | ~NETIF_F_ALL_FOR_ALL;
6376 /* If device needs checksumming, downgrade to it. */
6377 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6378 all &= ~NETIF_F_NO_CSUM;
6380 /* If one device supports hw checksumming, set for all. */
6381 if (all & NETIF_F_GEN_CSUM)
6382 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6386 EXPORT_SYMBOL(netdev_increment_features);
6388 static struct hlist_head *netdev_create_hash(void)
6391 struct hlist_head *hash;
6393 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6395 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6396 INIT_HLIST_HEAD(&hash[i]);
6401 /* Initialize per network namespace state */
6402 static int __net_init netdev_init(struct net *net)
6404 if (net != &init_net)
6405 INIT_LIST_HEAD(&net->dev_base_head);
6407 net->dev_name_head = netdev_create_hash();
6408 if (net->dev_name_head == NULL)
6411 net->dev_index_head = netdev_create_hash();
6412 if (net->dev_index_head == NULL)
6418 kfree(net->dev_name_head);
6424 * netdev_drivername - network driver for the device
6425 * @dev: network device
6427 * Determine network driver for device.
6429 const char *netdev_drivername(const struct net_device *dev)
6431 const struct device_driver *driver;
6432 const struct device *parent;
6433 const char *empty = "";
6435 parent = dev->dev.parent;
6439 driver = parent->driver;
6440 if (driver && driver->name)
6441 return driver->name;
6445 int __netdev_printk(const char *level, const struct net_device *dev,
6446 struct va_format *vaf)
6450 if (dev && dev->dev.parent)
6451 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6452 netdev_name(dev), vaf);
6454 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6456 r = printk("%s(NULL net_device): %pV", level, vaf);
6460 EXPORT_SYMBOL(__netdev_printk);
6462 int netdev_printk(const char *level, const struct net_device *dev,
6463 const char *format, ...)
6465 struct va_format vaf;
6469 va_start(args, format);
6474 r = __netdev_printk(level, dev, &vaf);
6479 EXPORT_SYMBOL(netdev_printk);
6481 #define define_netdev_printk_level(func, level) \
6482 int func(const struct net_device *dev, const char *fmt, ...) \
6485 struct va_format vaf; \
6488 va_start(args, fmt); \
6493 r = __netdev_printk(level, dev, &vaf); \
6498 EXPORT_SYMBOL(func);
6500 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6501 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6502 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6503 define_netdev_printk_level(netdev_err, KERN_ERR);
6504 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6505 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6506 define_netdev_printk_level(netdev_info, KERN_INFO);
6508 static void __net_exit netdev_exit(struct net *net)
6510 kfree(net->dev_name_head);
6511 kfree(net->dev_index_head);
6514 static struct pernet_operations __net_initdata netdev_net_ops = {
6515 .init = netdev_init,
6516 .exit = netdev_exit,
6519 static void __net_exit default_device_exit(struct net *net)
6521 struct net_device *dev, *aux;
6523 * Push all migratable network devices back to the
6524 * initial network namespace
6527 for_each_netdev_safe(net, dev, aux) {
6529 char fb_name[IFNAMSIZ];
6531 /* Ignore unmoveable devices (i.e. loopback) */
6532 if (dev->features & NETIF_F_NETNS_LOCAL)
6535 /* Leave virtual devices for the generic cleanup */
6536 if (dev->rtnl_link_ops)
6539 /* Push remaining network devices to init_net */
6540 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6541 err = dev_change_net_namespace(dev, &init_net, fb_name);
6543 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6544 __func__, dev->name, err);
6551 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6553 /* At exit all network devices most be removed from a network
6554 * namespace. Do this in the reverse order of registration.
6555 * Do this across as many network namespaces as possible to
6556 * improve batching efficiency.
6558 struct net_device *dev;
6560 LIST_HEAD(dev_kill_list);
6563 list_for_each_entry(net, net_list, exit_list) {
6564 for_each_netdev_reverse(net, dev) {
6565 if (dev->rtnl_link_ops)
6566 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6568 unregister_netdevice_queue(dev, &dev_kill_list);
6571 unregister_netdevice_many(&dev_kill_list);
6572 list_del(&dev_kill_list);
6576 static struct pernet_operations __net_initdata default_device_ops = {
6577 .exit = default_device_exit,
6578 .exit_batch = default_device_exit_batch,
6582 * Initialize the DEV module. At boot time this walks the device list and
6583 * unhooks any devices that fail to initialise (normally hardware not
6584 * present) and leaves us with a valid list of present and active devices.
6589 * This is called single threaded during boot, so no need
6590 * to take the rtnl semaphore.
6592 static int __init net_dev_init(void)
6594 int i, rc = -ENOMEM;
6596 BUG_ON(!dev_boot_phase);
6598 if (dev_proc_init())
6601 if (netdev_kobject_init())
6604 INIT_LIST_HEAD(&ptype_all);
6605 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6606 INIT_LIST_HEAD(&ptype_base[i]);
6608 if (register_pernet_subsys(&netdev_net_ops))
6612 * Initialise the packet receive queues.
6615 for_each_possible_cpu(i) {
6616 struct softnet_data *sd = &per_cpu(softnet_data, i);
6618 memset(sd, 0, sizeof(*sd));
6619 skb_queue_head_init(&sd->input_pkt_queue);
6620 skb_queue_head_init(&sd->process_queue);
6621 sd->completion_queue = NULL;
6622 INIT_LIST_HEAD(&sd->poll_list);
6623 sd->output_queue = NULL;
6624 sd->output_queue_tailp = &sd->output_queue;
6626 sd->csd.func = rps_trigger_softirq;
6632 sd->backlog.poll = process_backlog;
6633 sd->backlog.weight = weight_p;
6634 sd->backlog.gro_list = NULL;
6635 sd->backlog.gro_count = 0;
6640 /* The loopback device is special if any other network devices
6641 * is present in a network namespace the loopback device must
6642 * be present. Since we now dynamically allocate and free the
6643 * loopback device ensure this invariant is maintained by
6644 * keeping the loopback device as the first device on the
6645 * list of network devices. Ensuring the loopback devices
6646 * is the first device that appears and the last network device
6649 if (register_pernet_device(&loopback_net_ops))
6652 if (register_pernet_device(&default_device_ops))
6655 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6656 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6658 hotcpu_notifier(dev_cpu_callback, 0);
6666 subsys_initcall(net_dev_init);
6668 static int __init initialize_hashrnd(void)
6670 get_random_bytes(&hashrnd, sizeof(hashrnd));
6674 late_initcall_sync(initialize_hashrnd);