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 nf_reset_trace(skb);
1623 return netif_rx(skb);
1625 EXPORT_SYMBOL_GPL(dev_forward_skb);
1627 static inline int deliver_skb(struct sk_buff *skb,
1628 struct packet_type *pt_prev,
1629 struct net_device *orig_dev)
1631 atomic_inc(&skb->users);
1632 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1635 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1637 if (!ptype->af_packet_priv || !skb->sk)
1640 if (ptype->id_match)
1641 return ptype->id_match(ptype, skb->sk);
1642 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1649 * Support routine. Sends outgoing frames to any network
1650 * taps currently in use.
1653 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1655 struct packet_type *ptype;
1656 struct sk_buff *skb2 = NULL;
1657 struct packet_type *pt_prev = NULL;
1660 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1661 /* Never send packets back to the socket
1662 * they originated from - MvS (miquels@drinkel.ow.org)
1664 if ((ptype->dev == dev || !ptype->dev) &&
1665 (!skb_loop_sk(ptype, skb))) {
1667 deliver_skb(skb2, pt_prev, skb->dev);
1672 skb2 = skb_clone(skb, GFP_ATOMIC);
1676 net_timestamp_set(skb2);
1678 /* skb->nh should be correctly
1679 set by sender, so that the second statement is
1680 just protection against buggy protocols.
1682 skb_reset_mac_header(skb2);
1684 if (skb_network_header(skb2) < skb2->data ||
1685 skb2->network_header > skb2->tail) {
1686 if (net_ratelimit())
1687 printk(KERN_CRIT "protocol %04x is "
1689 ntohs(skb2->protocol),
1691 skb_reset_network_header(skb2);
1694 skb2->transport_header = skb2->network_header;
1695 skb2->pkt_type = PACKET_OUTGOING;
1700 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1704 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1705 * @dev: Network device
1706 * @txq: number of queues available
1708 * If real_num_tx_queues is changed the tc mappings may no longer be
1709 * valid. To resolve this verify the tc mapping remains valid and if
1710 * not NULL the mapping. With no priorities mapping to this
1711 * offset/count pair it will no longer be used. In the worst case TC0
1712 * is invalid nothing can be done so disable priority mappings. If is
1713 * expected that drivers will fix this mapping if they can before
1714 * calling netif_set_real_num_tx_queues.
1716 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1719 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1721 /* If TC0 is invalidated disable TC mapping */
1722 if (tc->offset + tc->count > txq) {
1723 pr_warning("Number of in use tx queues changed "
1724 "invalidating tc mappings. Priority "
1725 "traffic classification disabled!\n");
1730 /* Invalidated prio to tc mappings set to TC0 */
1731 for (i = 1; i < TC_BITMASK + 1; i++) {
1732 int q = netdev_get_prio_tc_map(dev, i);
1734 tc = &dev->tc_to_txq[q];
1735 if (tc->offset + tc->count > txq) {
1736 pr_warning("Number of in use tx queues "
1737 "changed. Priority %i to tc "
1738 "mapping %i is no longer valid "
1739 "setting map to 0\n",
1741 netdev_set_prio_tc_map(dev, i, 0);
1747 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1748 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1750 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1754 if (txq < 1 || txq > dev->num_tx_queues)
1757 if (dev->reg_state == NETREG_REGISTERED ||
1758 dev->reg_state == NETREG_UNREGISTERING) {
1761 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1767 netif_setup_tc(dev, txq);
1769 if (txq < dev->real_num_tx_queues)
1770 qdisc_reset_all_tx_gt(dev, txq);
1773 dev->real_num_tx_queues = txq;
1776 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1780 * netif_set_real_num_rx_queues - set actual number of RX queues used
1781 * @dev: Network device
1782 * @rxq: Actual number of RX queues
1784 * This must be called either with the rtnl_lock held or before
1785 * registration of the net device. Returns 0 on success, or a
1786 * negative error code. If called before registration, it always
1789 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1793 if (rxq < 1 || rxq > dev->num_rx_queues)
1796 if (dev->reg_state == NETREG_REGISTERED) {
1799 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1805 dev->real_num_rx_queues = rxq;
1808 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1811 static inline void __netif_reschedule(struct Qdisc *q)
1813 struct softnet_data *sd;
1814 unsigned long flags;
1816 local_irq_save(flags);
1817 sd = &__get_cpu_var(softnet_data);
1818 q->next_sched = NULL;
1819 *sd->output_queue_tailp = q;
1820 sd->output_queue_tailp = &q->next_sched;
1821 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1822 local_irq_restore(flags);
1825 void __netif_schedule(struct Qdisc *q)
1827 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1828 __netif_reschedule(q);
1830 EXPORT_SYMBOL(__netif_schedule);
1832 void dev_kfree_skb_irq(struct sk_buff *skb)
1834 if (atomic_dec_and_test(&skb->users)) {
1835 struct softnet_data *sd;
1836 unsigned long flags;
1838 local_irq_save(flags);
1839 sd = &__get_cpu_var(softnet_data);
1840 skb->next = sd->completion_queue;
1841 sd->completion_queue = skb;
1842 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1843 local_irq_restore(flags);
1846 EXPORT_SYMBOL(dev_kfree_skb_irq);
1848 void dev_kfree_skb_any(struct sk_buff *skb)
1850 if (in_irq() || irqs_disabled())
1851 dev_kfree_skb_irq(skb);
1855 EXPORT_SYMBOL(dev_kfree_skb_any);
1859 * netif_device_detach - mark device as removed
1860 * @dev: network device
1862 * Mark device as removed from system and therefore no longer available.
1864 void netif_device_detach(struct net_device *dev)
1866 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1867 netif_running(dev)) {
1868 netif_tx_stop_all_queues(dev);
1871 EXPORT_SYMBOL(netif_device_detach);
1874 * netif_device_attach - mark device as attached
1875 * @dev: network device
1877 * Mark device as attached from system and restart if needed.
1879 void netif_device_attach(struct net_device *dev)
1881 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1882 netif_running(dev)) {
1883 netif_tx_wake_all_queues(dev);
1884 __netdev_watchdog_up(dev);
1887 EXPORT_SYMBOL(netif_device_attach);
1890 * Invalidate hardware checksum when packet is to be mangled, and
1891 * complete checksum manually on outgoing path.
1893 int skb_checksum_help(struct sk_buff *skb)
1896 int ret = 0, offset;
1898 if (skb->ip_summed == CHECKSUM_COMPLETE)
1899 goto out_set_summed;
1901 if (unlikely(skb_shinfo(skb)->gso_size)) {
1902 /* Let GSO fix up the checksum. */
1903 goto out_set_summed;
1906 offset = skb_checksum_start_offset(skb);
1907 BUG_ON(offset >= skb_headlen(skb));
1908 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1910 offset += skb->csum_offset;
1911 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1913 if (skb_cloned(skb) &&
1914 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1915 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1920 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1922 skb->ip_summed = CHECKSUM_NONE;
1926 EXPORT_SYMBOL(skb_checksum_help);
1929 * skb_gso_segment - Perform segmentation on skb.
1930 * @skb: buffer to segment
1931 * @features: features for the output path (see dev->features)
1933 * This function segments the given skb and returns a list of segments.
1935 * It may return NULL if the skb requires no segmentation. This is
1936 * only possible when GSO is used for verifying header integrity.
1938 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1940 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1941 struct packet_type *ptype;
1942 __be16 type = skb->protocol;
1943 int vlan_depth = ETH_HLEN;
1946 while (type == htons(ETH_P_8021Q)) {
1947 struct vlan_hdr *vh;
1949 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1950 return ERR_PTR(-EINVAL);
1952 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1953 type = vh->h_vlan_encapsulated_proto;
1954 vlan_depth += VLAN_HLEN;
1957 skb_reset_mac_header(skb);
1958 skb->mac_len = skb->network_header - skb->mac_header;
1959 __skb_pull(skb, skb->mac_len);
1961 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1962 struct net_device *dev = skb->dev;
1963 struct ethtool_drvinfo info = {};
1965 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1966 dev->ethtool_ops->get_drvinfo(dev, &info);
1968 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1969 info.driver, dev ? dev->features : 0L,
1970 skb->sk ? skb->sk->sk_route_caps : 0L,
1971 skb->len, skb->data_len, skb->ip_summed);
1973 if (skb_header_cloned(skb) &&
1974 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1975 return ERR_PTR(err);
1979 list_for_each_entry_rcu(ptype,
1980 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1981 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1982 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1983 err = ptype->gso_send_check(skb);
1984 segs = ERR_PTR(err);
1985 if (err || skb_gso_ok(skb, features))
1987 __skb_push(skb, (skb->data -
1988 skb_network_header(skb)));
1990 segs = ptype->gso_segment(skb, features);
1996 __skb_push(skb, skb->data - skb_mac_header(skb));
2000 EXPORT_SYMBOL(skb_gso_segment);
2002 /* Take action when hardware reception checksum errors are detected. */
2004 void netdev_rx_csum_fault(struct net_device *dev)
2006 if (net_ratelimit()) {
2007 printk(KERN_ERR "%s: hw csum failure.\n",
2008 dev ? dev->name : "<unknown>");
2012 EXPORT_SYMBOL(netdev_rx_csum_fault);
2015 /* Actually, we should eliminate this check as soon as we know, that:
2016 * 1. IOMMU is present and allows to map all the memory.
2017 * 2. No high memory really exists on this machine.
2020 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2022 #ifdef CONFIG_HIGHMEM
2024 if (!(dev->features & NETIF_F_HIGHDMA)) {
2025 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2026 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2027 if (PageHighMem(skb_frag_page(frag)))
2032 if (PCI_DMA_BUS_IS_PHYS) {
2033 struct device *pdev = dev->dev.parent;
2037 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2038 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2039 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2040 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2049 void (*destructor)(struct sk_buff *skb);
2052 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2054 static void dev_gso_skb_destructor(struct sk_buff *skb)
2056 struct dev_gso_cb *cb;
2059 struct sk_buff *nskb = skb->next;
2061 skb->next = nskb->next;
2064 } while (skb->next);
2066 cb = DEV_GSO_CB(skb);
2068 cb->destructor(skb);
2072 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2073 * @skb: buffer to segment
2074 * @features: device features as applicable to this skb
2076 * This function segments the given skb and stores the list of segments
2079 static int dev_gso_segment(struct sk_buff *skb, int features)
2081 struct sk_buff *segs;
2083 segs = skb_gso_segment(skb, features);
2085 /* Verifying header integrity only. */
2090 return PTR_ERR(segs);
2093 DEV_GSO_CB(skb)->destructor = skb->destructor;
2094 skb->destructor = dev_gso_skb_destructor;
2099 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2101 return ((features & NETIF_F_GEN_CSUM) ||
2102 ((features & NETIF_F_V4_CSUM) &&
2103 protocol == htons(ETH_P_IP)) ||
2104 ((features & NETIF_F_V6_CSUM) &&
2105 protocol == htons(ETH_P_IPV6)) ||
2106 ((features & NETIF_F_FCOE_CRC) &&
2107 protocol == htons(ETH_P_FCOE)));
2110 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2112 if (skb->ip_summed != CHECKSUM_NONE &&
2113 !can_checksum_protocol(features, protocol)) {
2114 features &= ~NETIF_F_ALL_CSUM;
2115 features &= ~NETIF_F_SG;
2116 } else if (illegal_highdma(skb->dev, skb)) {
2117 features &= ~NETIF_F_SG;
2123 u32 netif_skb_features(struct sk_buff *skb)
2125 __be16 protocol = skb->protocol;
2126 u32 features = skb->dev->features;
2128 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2129 features &= ~NETIF_F_GSO_MASK;
2131 if (!vlan_tx_tag_present(skb)) {
2132 if (unlikely(protocol == htons(ETH_P_8021Q))) {
2133 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2134 protocol = veh->h_vlan_encapsulated_proto;
2136 return harmonize_features(skb, protocol, features);
2140 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2142 if (protocol != htons(ETH_P_8021Q)) {
2143 return harmonize_features(skb, protocol, features);
2145 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2146 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2147 return harmonize_features(skb, protocol, features);
2150 EXPORT_SYMBOL(netif_skb_features);
2153 * Returns true if either:
2154 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2155 * 2. skb is fragmented and the device does not support SG, or if
2156 * at least one of fragments is in highmem and device does not
2157 * support DMA from it.
2159 static inline int skb_needs_linearize(struct sk_buff *skb,
2162 return skb_is_nonlinear(skb) &&
2163 ((skb_has_frag_list(skb) &&
2164 !(features & NETIF_F_FRAGLIST)) ||
2165 (skb_shinfo(skb)->nr_frags &&
2166 !(features & NETIF_F_SG)));
2169 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2170 struct netdev_queue *txq)
2172 const struct net_device_ops *ops = dev->netdev_ops;
2173 int rc = NETDEV_TX_OK;
2174 unsigned int skb_len;
2176 if (likely(!skb->next)) {
2180 * If device doesn't need skb->dst, release it right now while
2181 * its hot in this cpu cache
2183 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2186 if (!list_empty(&ptype_all))
2187 dev_queue_xmit_nit(skb, dev);
2189 features = netif_skb_features(skb);
2191 if (vlan_tx_tag_present(skb) &&
2192 !(features & NETIF_F_HW_VLAN_TX)) {
2193 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2200 if (netif_needs_gso(skb, features)) {
2201 if (unlikely(dev_gso_segment(skb, features)))
2206 if (skb_needs_linearize(skb, features) &&
2207 __skb_linearize(skb))
2210 /* If packet is not checksummed and device does not
2211 * support checksumming for this protocol, complete
2212 * checksumming here.
2214 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2215 skb_set_transport_header(skb,
2216 skb_checksum_start_offset(skb));
2217 if (!(features & NETIF_F_ALL_CSUM) &&
2218 skb_checksum_help(skb))
2224 rc = ops->ndo_start_xmit(skb, dev);
2225 trace_net_dev_xmit(skb, rc, dev, skb_len);
2226 if (rc == NETDEV_TX_OK)
2227 txq_trans_update(txq);
2233 struct sk_buff *nskb = skb->next;
2235 skb->next = nskb->next;
2239 * If device doesn't need nskb->dst, release it right now while
2240 * its hot in this cpu cache
2242 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2245 skb_len = nskb->len;
2246 rc = ops->ndo_start_xmit(nskb, dev);
2247 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2248 if (unlikely(rc != NETDEV_TX_OK)) {
2249 if (rc & ~NETDEV_TX_MASK)
2250 goto out_kfree_gso_skb;
2251 nskb->next = skb->next;
2255 txq_trans_update(txq);
2256 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2257 return NETDEV_TX_BUSY;
2258 } while (skb->next);
2261 if (likely(skb->next == NULL))
2262 skb->destructor = DEV_GSO_CB(skb)->destructor;
2269 static u32 hashrnd __read_mostly;
2272 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2273 * to be used as a distribution range.
2275 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2276 unsigned int num_tx_queues)
2280 u16 qcount = num_tx_queues;
2282 if (skb_rx_queue_recorded(skb)) {
2283 hash = skb_get_rx_queue(skb);
2284 while (unlikely(hash >= num_tx_queues))
2285 hash -= num_tx_queues;
2290 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2291 qoffset = dev->tc_to_txq[tc].offset;
2292 qcount = dev->tc_to_txq[tc].count;
2295 if (skb->sk && skb->sk->sk_hash)
2296 hash = skb->sk->sk_hash;
2298 hash = (__force u16) skb->protocol;
2299 hash = jhash_1word(hash, hashrnd);
2301 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2303 EXPORT_SYMBOL(__skb_tx_hash);
2305 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2307 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2308 if (net_ratelimit()) {
2309 pr_warning("%s selects TX queue %d, but "
2310 "real number of TX queues is %d\n",
2311 dev->name, queue_index, dev->real_num_tx_queues);
2318 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2321 struct xps_dev_maps *dev_maps;
2322 struct xps_map *map;
2323 int queue_index = -1;
2326 dev_maps = rcu_dereference(dev->xps_maps);
2328 map = rcu_dereference(
2329 dev_maps->cpu_map[raw_smp_processor_id()]);
2332 queue_index = map->queues[0];
2335 if (skb->sk && skb->sk->sk_hash)
2336 hash = skb->sk->sk_hash;
2338 hash = (__force u16) skb->protocol ^
2340 hash = jhash_1word(hash, hashrnd);
2341 queue_index = map->queues[
2342 ((u64)hash * map->len) >> 32];
2344 if (unlikely(queue_index >= dev->real_num_tx_queues))
2356 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2357 struct sk_buff *skb)
2360 const struct net_device_ops *ops = dev->netdev_ops;
2362 if (dev->real_num_tx_queues == 1)
2364 else if (ops->ndo_select_queue) {
2365 queue_index = ops->ndo_select_queue(dev, skb);
2366 queue_index = dev_cap_txqueue(dev, queue_index);
2368 struct sock *sk = skb->sk;
2369 queue_index = sk_tx_queue_get(sk);
2371 if (queue_index < 0 || skb->ooo_okay ||
2372 queue_index >= dev->real_num_tx_queues) {
2373 int old_index = queue_index;
2375 queue_index = get_xps_queue(dev, skb);
2376 if (queue_index < 0)
2377 queue_index = skb_tx_hash(dev, skb);
2379 if (queue_index != old_index && sk) {
2380 struct dst_entry *dst =
2381 rcu_dereference_check(sk->sk_dst_cache, 1);
2383 if (dst && skb_dst(skb) == dst)
2384 sk_tx_queue_set(sk, queue_index);
2389 skb_set_queue_mapping(skb, queue_index);
2390 return netdev_get_tx_queue(dev, queue_index);
2393 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2394 struct net_device *dev,
2395 struct netdev_queue *txq)
2397 spinlock_t *root_lock = qdisc_lock(q);
2401 qdisc_skb_cb(skb)->pkt_len = skb->len;
2402 qdisc_calculate_pkt_len(skb, q);
2404 * Heuristic to force contended enqueues to serialize on a
2405 * separate lock before trying to get qdisc main lock.
2406 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2407 * and dequeue packets faster.
2409 contended = qdisc_is_running(q);
2410 if (unlikely(contended))
2411 spin_lock(&q->busylock);
2413 spin_lock(root_lock);
2414 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2417 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2418 qdisc_run_begin(q)) {
2420 * This is a work-conserving queue; there are no old skbs
2421 * waiting to be sent out; and the qdisc is not running -
2422 * xmit the skb directly.
2424 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2427 qdisc_bstats_update(q, skb);
2429 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2430 if (unlikely(contended)) {
2431 spin_unlock(&q->busylock);
2438 rc = NET_XMIT_SUCCESS;
2441 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2442 if (qdisc_run_begin(q)) {
2443 if (unlikely(contended)) {
2444 spin_unlock(&q->busylock);
2450 spin_unlock(root_lock);
2451 if (unlikely(contended))
2452 spin_unlock(&q->busylock);
2456 static DEFINE_PER_CPU(int, xmit_recursion);
2457 #define RECURSION_LIMIT 10
2460 * dev_queue_xmit - transmit a buffer
2461 * @skb: buffer to transmit
2463 * Queue a buffer for transmission to a network device. The caller must
2464 * have set the device and priority and built the buffer before calling
2465 * this function. The function can be called from an interrupt.
2467 * A negative errno code is returned on a failure. A success does not
2468 * guarantee the frame will be transmitted as it may be dropped due
2469 * to congestion or traffic shaping.
2471 * -----------------------------------------------------------------------------------
2472 * I notice this method can also return errors from the queue disciplines,
2473 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2476 * Regardless of the return value, the skb is consumed, so it is currently
2477 * difficult to retry a send to this method. (You can bump the ref count
2478 * before sending to hold a reference for retry if you are careful.)
2480 * When calling this method, interrupts MUST be enabled. This is because
2481 * the BH enable code must have IRQs enabled so that it will not deadlock.
2484 int dev_queue_xmit(struct sk_buff *skb)
2486 struct net_device *dev = skb->dev;
2487 struct netdev_queue *txq;
2491 /* Disable soft irqs for various locks below. Also
2492 * stops preemption for RCU.
2496 txq = dev_pick_tx(dev, skb);
2497 q = rcu_dereference_bh(txq->qdisc);
2499 #ifdef CONFIG_NET_CLS_ACT
2500 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2502 trace_net_dev_queue(skb);
2504 rc = __dev_xmit_skb(skb, q, dev, txq);
2508 /* The device has no queue. Common case for software devices:
2509 loopback, all the sorts of tunnels...
2511 Really, it is unlikely that netif_tx_lock protection is necessary
2512 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2514 However, it is possible, that they rely on protection
2517 Check this and shot the lock. It is not prone from deadlocks.
2518 Either shot noqueue qdisc, it is even simpler 8)
2520 if (dev->flags & IFF_UP) {
2521 int cpu = smp_processor_id(); /* ok because BHs are off */
2523 if (txq->xmit_lock_owner != cpu) {
2525 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2526 goto recursion_alert;
2528 HARD_TX_LOCK(dev, txq, cpu);
2530 if (!netif_tx_queue_stopped(txq)) {
2531 __this_cpu_inc(xmit_recursion);
2532 rc = dev_hard_start_xmit(skb, dev, txq);
2533 __this_cpu_dec(xmit_recursion);
2534 if (dev_xmit_complete(rc)) {
2535 HARD_TX_UNLOCK(dev, txq);
2539 HARD_TX_UNLOCK(dev, txq);
2540 if (net_ratelimit())
2541 printk(KERN_CRIT "Virtual device %s asks to "
2542 "queue packet!\n", dev->name);
2544 /* Recursion is detected! It is possible,
2548 if (net_ratelimit())
2549 printk(KERN_CRIT "Dead loop on virtual device "
2550 "%s, fix it urgently!\n", dev->name);
2555 rcu_read_unlock_bh();
2560 rcu_read_unlock_bh();
2563 EXPORT_SYMBOL(dev_queue_xmit);
2566 /*=======================================================================
2568 =======================================================================*/
2570 int netdev_max_backlog __read_mostly = 1000;
2571 int netdev_tstamp_prequeue __read_mostly = 1;
2572 int netdev_budget __read_mostly = 300;
2573 int weight_p __read_mostly = 64; /* old backlog weight */
2575 /* Called with irq disabled */
2576 static inline void ____napi_schedule(struct softnet_data *sd,
2577 struct napi_struct *napi)
2579 list_add_tail(&napi->poll_list, &sd->poll_list);
2580 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2584 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2585 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2586 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2587 * if hash is a canonical 4-tuple hash over transport ports.
2589 void __skb_get_rxhash(struct sk_buff *skb)
2591 int nhoff, hash = 0, poff;
2592 const struct ipv6hdr *ip6;
2593 const struct iphdr *ip;
2594 const struct vlan_hdr *vlan;
2603 nhoff = skb_network_offset(skb);
2604 proto = skb->protocol;
2608 case __constant_htons(ETH_P_IP):
2610 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2613 ip = (const struct iphdr *) (skb->data + nhoff);
2616 if (ip_is_fragment(ip))
2619 ip_proto = ip->protocol;
2620 addr1 = (__force u32) ip->saddr;
2621 addr2 = (__force u32) ip->daddr;
2622 nhoff += ip->ihl * 4;
2624 case __constant_htons(ETH_P_IPV6):
2626 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2629 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2630 ip_proto = ip6->nexthdr;
2631 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2632 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2635 case __constant_htons(ETH_P_8021Q):
2636 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2638 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2639 proto = vlan->h_vlan_encapsulated_proto;
2640 nhoff += sizeof(*vlan);
2642 case __constant_htons(ETH_P_PPP_SES):
2643 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2645 proto = *((__be16 *) (skb->data + nhoff +
2646 sizeof(struct pppoe_hdr)));
2647 nhoff += PPPOE_SES_HLEN;
2649 case __constant_htons(PPP_IP):
2651 case __constant_htons(PPP_IPV6):
2662 if (pskb_may_pull(skb, nhoff + 16)) {
2663 u8 *h = skb->data + nhoff;
2664 __be16 flags = *(__be16 *)h;
2667 * Only look inside GRE if version zero and no
2670 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2671 proto = *(__be16 *)(h + 2);
2673 if (flags & GRE_CSUM)
2675 if (flags & GRE_KEY)
2677 if (flags & GRE_SEQ)
2690 poff = proto_ports_offset(ip_proto);
2693 if (pskb_may_pull(skb, nhoff + 4)) {
2694 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2699 /* get a consistent hash (same value on both flow directions) */
2700 if (addr2 < addr1 ||
2702 ports.v16[1] < ports.v16[0])) {
2704 swap(ports.v16[0], ports.v16[1]);
2706 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2713 EXPORT_SYMBOL(__skb_get_rxhash);
2717 /* One global table that all flow-based protocols share. */
2718 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2719 EXPORT_SYMBOL(rps_sock_flow_table);
2721 static struct rps_dev_flow *
2722 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2723 struct rps_dev_flow *rflow, u16 next_cpu)
2725 if (next_cpu != RPS_NO_CPU) {
2726 #ifdef CONFIG_RFS_ACCEL
2727 struct netdev_rx_queue *rxqueue;
2728 struct rps_dev_flow_table *flow_table;
2729 struct rps_dev_flow *old_rflow;
2734 /* Should we steer this flow to a different hardware queue? */
2735 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2736 !(dev->features & NETIF_F_NTUPLE))
2738 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2739 if (rxq_index == skb_get_rx_queue(skb))
2742 rxqueue = dev->_rx + rxq_index;
2743 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2746 flow_id = skb->rxhash & flow_table->mask;
2747 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2748 rxq_index, flow_id);
2752 rflow = &flow_table->flows[flow_id];
2754 if (old_rflow->filter == rflow->filter)
2755 old_rflow->filter = RPS_NO_FILTER;
2759 per_cpu(softnet_data, next_cpu).input_queue_head;
2762 rflow->cpu = next_cpu;
2767 * get_rps_cpu is called from netif_receive_skb and returns the target
2768 * CPU from the RPS map of the receiving queue for a given skb.
2769 * rcu_read_lock must be held on entry.
2771 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2772 struct rps_dev_flow **rflowp)
2774 struct netdev_rx_queue *rxqueue;
2775 struct rps_map *map;
2776 struct rps_dev_flow_table *flow_table;
2777 struct rps_sock_flow_table *sock_flow_table;
2781 if (skb_rx_queue_recorded(skb)) {
2782 u16 index = skb_get_rx_queue(skb);
2783 if (unlikely(index >= dev->real_num_rx_queues)) {
2784 WARN_ONCE(dev->real_num_rx_queues > 1,
2785 "%s received packet on queue %u, but number "
2786 "of RX queues is %u\n",
2787 dev->name, index, dev->real_num_rx_queues);
2790 rxqueue = dev->_rx + index;
2794 map = rcu_dereference(rxqueue->rps_map);
2796 if (map->len == 1 &&
2797 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2798 tcpu = map->cpus[0];
2799 if (cpu_online(tcpu))
2803 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2807 skb_reset_network_header(skb);
2808 if (!skb_get_rxhash(skb))
2811 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2812 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2813 if (flow_table && sock_flow_table) {
2815 struct rps_dev_flow *rflow;
2817 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2820 next_cpu = sock_flow_table->ents[skb->rxhash &
2821 sock_flow_table->mask];
2824 * If the desired CPU (where last recvmsg was done) is
2825 * different from current CPU (one in the rx-queue flow
2826 * table entry), switch if one of the following holds:
2827 * - Current CPU is unset (equal to RPS_NO_CPU).
2828 * - Current CPU is offline.
2829 * - The current CPU's queue tail has advanced beyond the
2830 * last packet that was enqueued using this table entry.
2831 * This guarantees that all previous packets for the flow
2832 * have been dequeued, thus preserving in order delivery.
2834 if (unlikely(tcpu != next_cpu) &&
2835 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2836 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2837 rflow->last_qtail)) >= 0)) {
2839 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2842 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2850 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2852 if (cpu_online(tcpu)) {
2862 #ifdef CONFIG_RFS_ACCEL
2865 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2866 * @dev: Device on which the filter was set
2867 * @rxq_index: RX queue index
2868 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2869 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2871 * Drivers that implement ndo_rx_flow_steer() should periodically call
2872 * this function for each installed filter and remove the filters for
2873 * which it returns %true.
2875 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2876 u32 flow_id, u16 filter_id)
2878 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2879 struct rps_dev_flow_table *flow_table;
2880 struct rps_dev_flow *rflow;
2885 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2886 if (flow_table && flow_id <= flow_table->mask) {
2887 rflow = &flow_table->flows[flow_id];
2888 cpu = ACCESS_ONCE(rflow->cpu);
2889 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2890 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2891 rflow->last_qtail) <
2892 (int)(10 * flow_table->mask)))
2898 EXPORT_SYMBOL(rps_may_expire_flow);
2900 #endif /* CONFIG_RFS_ACCEL */
2902 /* Called from hardirq (IPI) context */
2903 static void rps_trigger_softirq(void *data)
2905 struct softnet_data *sd = data;
2907 ____napi_schedule(sd, &sd->backlog);
2911 #endif /* CONFIG_RPS */
2914 * Check if this softnet_data structure is another cpu one
2915 * If yes, queue it to our IPI list and return 1
2918 static int rps_ipi_queued(struct softnet_data *sd)
2921 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2924 sd->rps_ipi_next = mysd->rps_ipi_list;
2925 mysd->rps_ipi_list = sd;
2927 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2930 #endif /* CONFIG_RPS */
2935 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2936 * queue (may be a remote CPU queue).
2938 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2939 unsigned int *qtail)
2941 struct softnet_data *sd;
2942 unsigned long flags;
2944 sd = &per_cpu(softnet_data, cpu);
2946 local_irq_save(flags);
2949 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2950 if (skb_queue_len(&sd->input_pkt_queue)) {
2952 __skb_queue_tail(&sd->input_pkt_queue, skb);
2953 input_queue_tail_incr_save(sd, qtail);
2955 local_irq_restore(flags);
2956 return NET_RX_SUCCESS;
2959 /* Schedule NAPI for backlog device
2960 * We can use non atomic operation since we own the queue lock
2962 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2963 if (!rps_ipi_queued(sd))
2964 ____napi_schedule(sd, &sd->backlog);
2972 local_irq_restore(flags);
2974 atomic_long_inc(&skb->dev->rx_dropped);
2980 * netif_rx - post buffer to the network code
2981 * @skb: buffer to post
2983 * This function receives a packet from a device driver and queues it for
2984 * the upper (protocol) levels to process. It always succeeds. The buffer
2985 * may be dropped during processing for congestion control or by the
2989 * NET_RX_SUCCESS (no congestion)
2990 * NET_RX_DROP (packet was dropped)
2994 int netif_rx(struct sk_buff *skb)
2998 /* if netpoll wants it, pretend we never saw it */
2999 if (netpoll_rx(skb))
3002 if (netdev_tstamp_prequeue)
3003 net_timestamp_check(skb);
3005 trace_netif_rx(skb);
3008 struct rps_dev_flow voidflow, *rflow = &voidflow;
3014 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3016 cpu = smp_processor_id();
3018 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3026 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3032 EXPORT_SYMBOL(netif_rx);
3034 int netif_rx_ni(struct sk_buff *skb)
3039 err = netif_rx(skb);
3040 if (local_softirq_pending())
3046 EXPORT_SYMBOL(netif_rx_ni);
3048 static void net_tx_action(struct softirq_action *h)
3050 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3052 if (sd->completion_queue) {
3053 struct sk_buff *clist;
3055 local_irq_disable();
3056 clist = sd->completion_queue;
3057 sd->completion_queue = NULL;
3061 struct sk_buff *skb = clist;
3062 clist = clist->next;
3064 WARN_ON(atomic_read(&skb->users));
3065 trace_kfree_skb(skb, net_tx_action);
3070 if (sd->output_queue) {
3073 local_irq_disable();
3074 head = sd->output_queue;
3075 sd->output_queue = NULL;
3076 sd->output_queue_tailp = &sd->output_queue;
3080 struct Qdisc *q = head;
3081 spinlock_t *root_lock;
3083 head = head->next_sched;
3085 root_lock = qdisc_lock(q);
3086 if (spin_trylock(root_lock)) {
3087 smp_mb__before_clear_bit();
3088 clear_bit(__QDISC_STATE_SCHED,
3091 spin_unlock(root_lock);
3093 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3095 __netif_reschedule(q);
3097 smp_mb__before_clear_bit();
3098 clear_bit(__QDISC_STATE_SCHED,
3106 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3107 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3108 /* This hook is defined here for ATM LANE */
3109 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3110 unsigned char *addr) __read_mostly;
3111 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3114 #ifdef CONFIG_NET_CLS_ACT
3115 /* TODO: Maybe we should just force sch_ingress to be compiled in
3116 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3117 * a compare and 2 stores extra right now if we dont have it on
3118 * but have CONFIG_NET_CLS_ACT
3119 * NOTE: This doesn't stop any functionality; if you dont have
3120 * the ingress scheduler, you just can't add policies on ingress.
3123 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3125 struct net_device *dev = skb->dev;
3126 u32 ttl = G_TC_RTTL(skb->tc_verd);
3127 int result = TC_ACT_OK;
3130 if (unlikely(MAX_RED_LOOP < ttl++)) {
3131 if (net_ratelimit())
3132 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3133 skb->skb_iif, dev->ifindex);
3137 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3138 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3141 if (q != &noop_qdisc) {
3142 spin_lock(qdisc_lock(q));
3143 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3144 result = qdisc_enqueue_root(skb, q);
3145 spin_unlock(qdisc_lock(q));
3151 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3152 struct packet_type **pt_prev,
3153 int *ret, struct net_device *orig_dev)
3155 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3157 if (!rxq || rxq->qdisc == &noop_qdisc)
3161 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3165 switch (ing_filter(skb, rxq)) {
3179 * netdev_rx_handler_register - register receive handler
3180 * @dev: device to register a handler for
3181 * @rx_handler: receive handler to register
3182 * @rx_handler_data: data pointer that is used by rx handler
3184 * Register a receive hander for a device. This handler will then be
3185 * called from __netif_receive_skb. A negative errno code is returned
3188 * The caller must hold the rtnl_mutex.
3190 * For a general description of rx_handler, see enum rx_handler_result.
3192 int netdev_rx_handler_register(struct net_device *dev,
3193 rx_handler_func_t *rx_handler,
3194 void *rx_handler_data)
3198 if (dev->rx_handler)
3201 /* Note: rx_handler_data must be set before rx_handler */
3202 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3203 rcu_assign_pointer(dev->rx_handler, rx_handler);
3207 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3210 * netdev_rx_handler_unregister - unregister receive handler
3211 * @dev: device to unregister a handler from
3213 * Unregister a receive hander from a device.
3215 * The caller must hold the rtnl_mutex.
3217 void netdev_rx_handler_unregister(struct net_device *dev)
3221 RCU_INIT_POINTER(dev->rx_handler, NULL);
3222 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3223 * section has a guarantee to see a non NULL rx_handler_data
3227 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3229 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3231 static int __netif_receive_skb(struct sk_buff *skb)
3233 struct packet_type *ptype, *pt_prev;
3234 rx_handler_func_t *rx_handler;
3235 struct net_device *orig_dev;
3236 struct net_device *null_or_dev;
3237 bool deliver_exact = false;
3238 int ret = NET_RX_DROP;
3241 if (!netdev_tstamp_prequeue)
3242 net_timestamp_check(skb);
3244 trace_netif_receive_skb(skb);
3246 /* if we've gotten here through NAPI, check netpoll */
3247 if (netpoll_receive_skb(skb))
3251 skb->skb_iif = skb->dev->ifindex;
3252 orig_dev = skb->dev;
3254 skb_reset_network_header(skb);
3255 skb_reset_transport_header(skb);
3256 skb_reset_mac_len(skb);
3264 __this_cpu_inc(softnet_data.processed);
3266 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3267 skb = vlan_untag(skb);
3272 #ifdef CONFIG_NET_CLS_ACT
3273 if (skb->tc_verd & TC_NCLS) {
3274 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3279 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3280 if (!ptype->dev || ptype->dev == skb->dev) {
3282 ret = deliver_skb(skb, pt_prev, orig_dev);
3287 #ifdef CONFIG_NET_CLS_ACT
3288 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3294 if (vlan_tx_tag_present(skb)) {
3296 ret = deliver_skb(skb, pt_prev, orig_dev);
3299 if (vlan_do_receive(&skb))
3301 else if (unlikely(!skb))
3305 rx_handler = rcu_dereference(skb->dev->rx_handler);
3308 ret = deliver_skb(skb, pt_prev, orig_dev);
3311 switch (rx_handler(&skb)) {
3312 case RX_HANDLER_CONSUMED:
3313 ret = NET_RX_SUCCESS;
3315 case RX_HANDLER_ANOTHER:
3317 case RX_HANDLER_EXACT:
3318 deliver_exact = true;
3319 case RX_HANDLER_PASS:
3326 if (vlan_tx_nonzero_tag_present(skb))
3327 skb->pkt_type = PACKET_OTHERHOST;
3329 /* deliver only exact match when indicated */
3330 null_or_dev = deliver_exact ? skb->dev : NULL;
3332 type = skb->protocol;
3333 list_for_each_entry_rcu(ptype,
3334 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3335 if (ptype->type == type &&
3336 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3337 ptype->dev == orig_dev)) {
3339 ret = deliver_skb(skb, pt_prev, orig_dev);
3345 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3347 atomic_long_inc(&skb->dev->rx_dropped);
3349 /* Jamal, now you will not able to escape explaining
3350 * me how you were going to use this. :-)
3361 * netif_receive_skb - process receive buffer from network
3362 * @skb: buffer to process
3364 * netif_receive_skb() is the main receive data processing function.
3365 * It always succeeds. The buffer may be dropped during processing
3366 * for congestion control or by the protocol layers.
3368 * This function may only be called from softirq context and interrupts
3369 * should be enabled.
3371 * Return values (usually ignored):
3372 * NET_RX_SUCCESS: no congestion
3373 * NET_RX_DROP: packet was dropped
3375 int netif_receive_skb(struct sk_buff *skb)
3377 if (netdev_tstamp_prequeue)
3378 net_timestamp_check(skb);
3380 if (skb_defer_rx_timestamp(skb))
3381 return NET_RX_SUCCESS;
3385 struct rps_dev_flow voidflow, *rflow = &voidflow;
3390 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3393 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3397 ret = __netif_receive_skb(skb);
3403 return __netif_receive_skb(skb);
3406 EXPORT_SYMBOL(netif_receive_skb);
3408 /* Network device is going away, flush any packets still pending
3409 * Called with irqs disabled.
3411 static void flush_backlog(void *arg)
3413 struct net_device *dev = arg;
3414 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3415 struct sk_buff *skb, *tmp;
3418 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3419 if (skb->dev == dev) {
3420 __skb_unlink(skb, &sd->input_pkt_queue);
3422 input_queue_head_incr(sd);
3427 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3428 if (skb->dev == dev) {
3429 __skb_unlink(skb, &sd->process_queue);
3431 input_queue_head_incr(sd);
3436 static int napi_gro_complete(struct sk_buff *skb)
3438 struct packet_type *ptype;
3439 __be16 type = skb->protocol;
3440 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3443 if (NAPI_GRO_CB(skb)->count == 1) {
3444 skb_shinfo(skb)->gso_size = 0;
3449 list_for_each_entry_rcu(ptype, head, list) {
3450 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3453 err = ptype->gro_complete(skb);
3459 WARN_ON(&ptype->list == head);
3461 return NET_RX_SUCCESS;
3465 return netif_receive_skb(skb);
3468 inline void napi_gro_flush(struct napi_struct *napi)
3470 struct sk_buff *skb, *next;
3472 for (skb = napi->gro_list; skb; skb = next) {
3475 napi_gro_complete(skb);
3478 napi->gro_count = 0;
3479 napi->gro_list = NULL;
3481 EXPORT_SYMBOL(napi_gro_flush);
3483 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3485 struct sk_buff **pp = NULL;
3486 struct packet_type *ptype;
3487 __be16 type = skb->protocol;
3488 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3491 enum gro_result ret;
3493 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3496 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3500 list_for_each_entry_rcu(ptype, head, list) {
3501 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3504 skb_set_network_header(skb, skb_gro_offset(skb));
3505 mac_len = skb->network_header - skb->mac_header;
3506 skb->mac_len = mac_len;
3507 NAPI_GRO_CB(skb)->same_flow = 0;
3508 NAPI_GRO_CB(skb)->flush = 0;
3509 NAPI_GRO_CB(skb)->free = 0;
3511 pp = ptype->gro_receive(&napi->gro_list, skb);
3516 if (&ptype->list == head)
3519 same_flow = NAPI_GRO_CB(skb)->same_flow;
3520 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3523 struct sk_buff *nskb = *pp;
3527 napi_gro_complete(nskb);
3534 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3538 NAPI_GRO_CB(skb)->count = 1;
3539 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3540 skb->next = napi->gro_list;
3541 napi->gro_list = skb;
3545 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3546 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3548 BUG_ON(skb->end - skb->tail < grow);
3550 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3553 skb->data_len -= grow;
3555 skb_shinfo(skb)->frags[0].page_offset += grow;
3556 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3558 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3559 skb_frag_unref(skb, 0);
3560 memmove(skb_shinfo(skb)->frags,
3561 skb_shinfo(skb)->frags + 1,
3562 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3573 EXPORT_SYMBOL(dev_gro_receive);
3575 static inline gro_result_t
3576 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3579 unsigned int maclen = skb->dev->hard_header_len;
3581 for (p = napi->gro_list; p; p = p->next) {
3582 unsigned long diffs;
3584 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3585 diffs |= p->vlan_tci ^ skb->vlan_tci;
3586 if (maclen == ETH_HLEN)
3587 diffs |= compare_ether_header(skb_mac_header(p),
3588 skb_gro_mac_header(skb));
3590 diffs = memcmp(skb_mac_header(p),
3591 skb_gro_mac_header(skb),
3593 NAPI_GRO_CB(p)->same_flow = !diffs;
3594 NAPI_GRO_CB(p)->flush = 0;
3597 return dev_gro_receive(napi, skb);
3600 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3604 if (netif_receive_skb(skb))
3609 case GRO_MERGED_FREE:
3620 EXPORT_SYMBOL(napi_skb_finish);
3622 void skb_gro_reset_offset(struct sk_buff *skb)
3624 NAPI_GRO_CB(skb)->data_offset = 0;
3625 NAPI_GRO_CB(skb)->frag0 = NULL;
3626 NAPI_GRO_CB(skb)->frag0_len = 0;
3628 if (skb->mac_header == skb->tail &&
3629 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3630 NAPI_GRO_CB(skb)->frag0 =
3631 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3632 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3635 EXPORT_SYMBOL(skb_gro_reset_offset);
3637 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3639 skb_gro_reset_offset(skb);
3641 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3643 EXPORT_SYMBOL(napi_gro_receive);
3645 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3647 __skb_pull(skb, skb_headlen(skb));
3648 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3649 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3651 skb->dev = napi->dev;
3653 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
3658 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3660 struct sk_buff *skb = napi->skb;
3663 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3669 EXPORT_SYMBOL(napi_get_frags);
3671 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3677 skb->protocol = eth_type_trans(skb, skb->dev);
3679 if (ret == GRO_HELD)
3680 skb_gro_pull(skb, -ETH_HLEN);
3681 else if (netif_receive_skb(skb))
3686 case GRO_MERGED_FREE:
3687 napi_reuse_skb(napi, skb);
3696 EXPORT_SYMBOL(napi_frags_finish);
3698 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3700 struct sk_buff *skb = napi->skb;
3707 skb_reset_mac_header(skb);
3708 skb_gro_reset_offset(skb);
3710 off = skb_gro_offset(skb);
3711 hlen = off + sizeof(*eth);
3712 eth = skb_gro_header_fast(skb, off);
3713 if (skb_gro_header_hard(skb, hlen)) {
3714 eth = skb_gro_header_slow(skb, hlen, off);
3715 if (unlikely(!eth)) {
3716 napi_reuse_skb(napi, skb);
3722 skb_gro_pull(skb, sizeof(*eth));
3725 * This works because the only protocols we care about don't require
3726 * special handling. We'll fix it up properly at the end.
3728 skb->protocol = eth->h_proto;
3733 EXPORT_SYMBOL(napi_frags_skb);
3735 gro_result_t napi_gro_frags(struct napi_struct *napi)
3737 struct sk_buff *skb = napi_frags_skb(napi);
3742 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3744 EXPORT_SYMBOL(napi_gro_frags);
3747 * net_rps_action sends any pending IPI's for rps.
3748 * Note: called with local irq disabled, but exits with local irq enabled.
3750 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3753 struct softnet_data *remsd = sd->rps_ipi_list;
3756 sd->rps_ipi_list = NULL;
3760 /* Send pending IPI's to kick RPS processing on remote cpus. */
3762 struct softnet_data *next = remsd->rps_ipi_next;
3764 if (cpu_online(remsd->cpu))
3765 __smp_call_function_single(remsd->cpu,
3774 static int process_backlog(struct napi_struct *napi, int quota)
3777 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3780 /* Check if we have pending ipi, its better to send them now,
3781 * not waiting net_rx_action() end.
3783 if (sd->rps_ipi_list) {
3784 local_irq_disable();
3785 net_rps_action_and_irq_enable(sd);
3788 napi->weight = weight_p;
3789 local_irq_disable();
3790 while (work < quota) {
3791 struct sk_buff *skb;
3794 while ((skb = __skb_dequeue(&sd->process_queue))) {
3796 __netif_receive_skb(skb);
3797 local_irq_disable();
3798 input_queue_head_incr(sd);
3799 if (++work >= quota) {
3806 qlen = skb_queue_len(&sd->input_pkt_queue);
3808 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3809 &sd->process_queue);
3811 if (qlen < quota - work) {
3813 * Inline a custom version of __napi_complete().
3814 * only current cpu owns and manipulates this napi,
3815 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3816 * we can use a plain write instead of clear_bit(),
3817 * and we dont need an smp_mb() memory barrier.
3819 list_del(&napi->poll_list);
3822 quota = work + qlen;
3832 * __napi_schedule - schedule for receive
3833 * @n: entry to schedule
3835 * The entry's receive function will be scheduled to run
3837 void __napi_schedule(struct napi_struct *n)
3839 unsigned long flags;
3841 local_irq_save(flags);
3842 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3843 local_irq_restore(flags);
3845 EXPORT_SYMBOL(__napi_schedule);
3847 void __napi_complete(struct napi_struct *n)
3849 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3850 BUG_ON(n->gro_list);
3852 list_del(&n->poll_list);
3853 smp_mb__before_clear_bit();
3854 clear_bit(NAPI_STATE_SCHED, &n->state);
3856 EXPORT_SYMBOL(__napi_complete);
3858 void napi_complete(struct napi_struct *n)
3860 unsigned long flags;
3863 * don't let napi dequeue from the cpu poll list
3864 * just in case its running on a different cpu
3866 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3870 local_irq_save(flags);
3872 local_irq_restore(flags);
3874 EXPORT_SYMBOL(napi_complete);
3876 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3877 int (*poll)(struct napi_struct *, int), int weight)
3879 INIT_LIST_HEAD(&napi->poll_list);
3880 napi->gro_count = 0;
3881 napi->gro_list = NULL;
3884 napi->weight = weight;
3885 list_add(&napi->dev_list, &dev->napi_list);
3887 #ifdef CONFIG_NETPOLL
3888 spin_lock_init(&napi->poll_lock);
3889 napi->poll_owner = -1;
3891 set_bit(NAPI_STATE_SCHED, &napi->state);
3893 EXPORT_SYMBOL(netif_napi_add);
3895 void netif_napi_del(struct napi_struct *napi)
3897 struct sk_buff *skb, *next;
3899 list_del_init(&napi->dev_list);
3900 napi_free_frags(napi);
3902 for (skb = napi->gro_list; skb; skb = next) {
3908 napi->gro_list = NULL;
3909 napi->gro_count = 0;
3911 EXPORT_SYMBOL(netif_napi_del);
3913 static void net_rx_action(struct softirq_action *h)
3915 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3916 unsigned long time_limit = jiffies + 2;
3917 int budget = netdev_budget;
3920 local_irq_disable();
3922 while (!list_empty(&sd->poll_list)) {
3923 struct napi_struct *n;
3926 /* If softirq window is exhuasted then punt.
3927 * Allow this to run for 2 jiffies since which will allow
3928 * an average latency of 1.5/HZ.
3930 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3935 /* Even though interrupts have been re-enabled, this
3936 * access is safe because interrupts can only add new
3937 * entries to the tail of this list, and only ->poll()
3938 * calls can remove this head entry from the list.
3940 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3942 have = netpoll_poll_lock(n);
3946 /* This NAPI_STATE_SCHED test is for avoiding a race
3947 * with netpoll's poll_napi(). Only the entity which
3948 * obtains the lock and sees NAPI_STATE_SCHED set will
3949 * actually make the ->poll() call. Therefore we avoid
3950 * accidentally calling ->poll() when NAPI is not scheduled.
3953 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3954 work = n->poll(n, weight);
3958 WARN_ON_ONCE(work > weight);
3962 local_irq_disable();
3964 /* Drivers must not modify the NAPI state if they
3965 * consume the entire weight. In such cases this code
3966 * still "owns" the NAPI instance and therefore can
3967 * move the instance around on the list at-will.
3969 if (unlikely(work == weight)) {
3970 if (unlikely(napi_disable_pending(n))) {
3973 local_irq_disable();
3975 list_move_tail(&n->poll_list, &sd->poll_list);
3978 netpoll_poll_unlock(have);
3981 net_rps_action_and_irq_enable(sd);
3983 #ifdef CONFIG_NET_DMA
3985 * There may not be any more sk_buffs coming right now, so push
3986 * any pending DMA copies to hardware
3988 dma_issue_pending_all();
3995 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3999 static gifconf_func_t *gifconf_list[NPROTO];
4002 * register_gifconf - register a SIOCGIF handler
4003 * @family: Address family
4004 * @gifconf: Function handler
4006 * Register protocol dependent address dumping routines. The handler
4007 * that is passed must not be freed or reused until it has been replaced
4008 * by another handler.
4010 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4012 if (family >= NPROTO)
4014 gifconf_list[family] = gifconf;
4017 EXPORT_SYMBOL(register_gifconf);
4021 * Map an interface index to its name (SIOCGIFNAME)
4025 * We need this ioctl for efficient implementation of the
4026 * if_indextoname() function required by the IPv6 API. Without
4027 * it, we would have to search all the interfaces to find a
4031 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4033 struct net_device *dev;
4037 * Fetch the caller's info block.
4040 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4044 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4050 strcpy(ifr.ifr_name, dev->name);
4053 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4059 * Perform a SIOCGIFCONF call. This structure will change
4060 * size eventually, and there is nothing I can do about it.
4061 * Thus we will need a 'compatibility mode'.
4064 static int dev_ifconf(struct net *net, char __user *arg)
4067 struct net_device *dev;
4074 * Fetch the caller's info block.
4077 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4084 * Loop over the interfaces, and write an info block for each.
4088 for_each_netdev(net, dev) {
4089 for (i = 0; i < NPROTO; i++) {
4090 if (gifconf_list[i]) {
4093 done = gifconf_list[i](dev, NULL, 0);
4095 done = gifconf_list[i](dev, pos + total,
4105 * All done. Write the updated control block back to the caller.
4107 ifc.ifc_len = total;
4110 * Both BSD and Solaris return 0 here, so we do too.
4112 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4115 #ifdef CONFIG_PROC_FS
4117 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4119 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4120 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4121 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4123 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4125 struct net *net = seq_file_net(seq);
4126 struct net_device *dev;
4127 struct hlist_node *p;
4128 struct hlist_head *h;
4129 unsigned int count = 0, offset = get_offset(*pos);
4131 h = &net->dev_name_head[get_bucket(*pos)];
4132 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4133 if (++count == offset)
4140 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4142 struct net_device *dev;
4143 unsigned int bucket;
4146 dev = dev_from_same_bucket(seq, pos);
4150 bucket = get_bucket(*pos) + 1;
4151 *pos = set_bucket_offset(bucket, 1);
4152 } while (bucket < NETDEV_HASHENTRIES);
4158 * This is invoked by the /proc filesystem handler to display a device
4161 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4166 return SEQ_START_TOKEN;
4168 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4171 return dev_from_bucket(seq, pos);
4174 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4177 return dev_from_bucket(seq, pos);
4180 void dev_seq_stop(struct seq_file *seq, void *v)
4186 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4188 struct rtnl_link_stats64 temp;
4189 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4191 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4192 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4193 dev->name, stats->rx_bytes, stats->rx_packets,
4195 stats->rx_dropped + stats->rx_missed_errors,
4196 stats->rx_fifo_errors,
4197 stats->rx_length_errors + stats->rx_over_errors +
4198 stats->rx_crc_errors + stats->rx_frame_errors,
4199 stats->rx_compressed, stats->multicast,
4200 stats->tx_bytes, stats->tx_packets,
4201 stats->tx_errors, stats->tx_dropped,
4202 stats->tx_fifo_errors, stats->collisions,
4203 stats->tx_carrier_errors +
4204 stats->tx_aborted_errors +
4205 stats->tx_window_errors +
4206 stats->tx_heartbeat_errors,
4207 stats->tx_compressed);
4211 * Called from the PROCfs module. This now uses the new arbitrary sized
4212 * /proc/net interface to create /proc/net/dev
4214 static int dev_seq_show(struct seq_file *seq, void *v)
4216 if (v == SEQ_START_TOKEN)
4217 seq_puts(seq, "Inter-| Receive "
4219 " face |bytes packets errs drop fifo frame "
4220 "compressed multicast|bytes packets errs "
4221 "drop fifo colls carrier compressed\n");
4223 dev_seq_printf_stats(seq, v);
4227 static struct softnet_data *softnet_get_online(loff_t *pos)
4229 struct softnet_data *sd = NULL;
4231 while (*pos < nr_cpu_ids)
4232 if (cpu_online(*pos)) {
4233 sd = &per_cpu(softnet_data, *pos);
4240 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4242 return softnet_get_online(pos);
4245 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4248 return softnet_get_online(pos);
4251 static void softnet_seq_stop(struct seq_file *seq, void *v)
4255 static int softnet_seq_show(struct seq_file *seq, void *v)
4257 struct softnet_data *sd = v;
4259 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4260 sd->processed, sd->dropped, sd->time_squeeze, 0,
4261 0, 0, 0, 0, /* was fastroute */
4262 sd->cpu_collision, sd->received_rps);
4266 static const struct seq_operations dev_seq_ops = {
4267 .start = dev_seq_start,
4268 .next = dev_seq_next,
4269 .stop = dev_seq_stop,
4270 .show = dev_seq_show,
4273 static int dev_seq_open(struct inode *inode, struct file *file)
4275 return seq_open_net(inode, file, &dev_seq_ops,
4276 sizeof(struct seq_net_private));
4279 static const struct file_operations dev_seq_fops = {
4280 .owner = THIS_MODULE,
4281 .open = dev_seq_open,
4283 .llseek = seq_lseek,
4284 .release = seq_release_net,
4287 static const struct seq_operations softnet_seq_ops = {
4288 .start = softnet_seq_start,
4289 .next = softnet_seq_next,
4290 .stop = softnet_seq_stop,
4291 .show = softnet_seq_show,
4294 static int softnet_seq_open(struct inode *inode, struct file *file)
4296 return seq_open(file, &softnet_seq_ops);
4299 static const struct file_operations softnet_seq_fops = {
4300 .owner = THIS_MODULE,
4301 .open = softnet_seq_open,
4303 .llseek = seq_lseek,
4304 .release = seq_release,
4307 static void *ptype_get_idx(loff_t pos)
4309 struct packet_type *pt = NULL;
4313 list_for_each_entry_rcu(pt, &ptype_all, list) {
4319 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4320 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4329 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4333 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4336 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4338 struct packet_type *pt;
4339 struct list_head *nxt;
4343 if (v == SEQ_START_TOKEN)
4344 return ptype_get_idx(0);
4347 nxt = pt->list.next;
4348 if (pt->type == htons(ETH_P_ALL)) {
4349 if (nxt != &ptype_all)
4352 nxt = ptype_base[0].next;
4354 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4356 while (nxt == &ptype_base[hash]) {
4357 if (++hash >= PTYPE_HASH_SIZE)
4359 nxt = ptype_base[hash].next;
4362 return list_entry(nxt, struct packet_type, list);
4365 static void ptype_seq_stop(struct seq_file *seq, void *v)
4371 static int ptype_seq_show(struct seq_file *seq, void *v)
4373 struct packet_type *pt = v;
4375 if (v == SEQ_START_TOKEN)
4376 seq_puts(seq, "Type Device Function\n");
4377 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4378 if (pt->type == htons(ETH_P_ALL))
4379 seq_puts(seq, "ALL ");
4381 seq_printf(seq, "%04x", ntohs(pt->type));
4383 seq_printf(seq, " %-8s %pF\n",
4384 pt->dev ? pt->dev->name : "", pt->func);
4390 static const struct seq_operations ptype_seq_ops = {
4391 .start = ptype_seq_start,
4392 .next = ptype_seq_next,
4393 .stop = ptype_seq_stop,
4394 .show = ptype_seq_show,
4397 static int ptype_seq_open(struct inode *inode, struct file *file)
4399 return seq_open_net(inode, file, &ptype_seq_ops,
4400 sizeof(struct seq_net_private));
4403 static const struct file_operations ptype_seq_fops = {
4404 .owner = THIS_MODULE,
4405 .open = ptype_seq_open,
4407 .llseek = seq_lseek,
4408 .release = seq_release_net,
4412 static int __net_init dev_proc_net_init(struct net *net)
4416 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4418 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4420 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4423 if (wext_proc_init(net))
4429 proc_net_remove(net, "ptype");
4431 proc_net_remove(net, "softnet_stat");
4433 proc_net_remove(net, "dev");
4437 static void __net_exit dev_proc_net_exit(struct net *net)
4439 wext_proc_exit(net);
4441 proc_net_remove(net, "ptype");
4442 proc_net_remove(net, "softnet_stat");
4443 proc_net_remove(net, "dev");
4446 static struct pernet_operations __net_initdata dev_proc_ops = {
4447 .init = dev_proc_net_init,
4448 .exit = dev_proc_net_exit,
4451 static int __init dev_proc_init(void)
4453 return register_pernet_subsys(&dev_proc_ops);
4456 #define dev_proc_init() 0
4457 #endif /* CONFIG_PROC_FS */
4461 * netdev_set_master - set up master pointer
4462 * @slave: slave device
4463 * @master: new master device
4465 * Changes the master device of the slave. Pass %NULL to break the
4466 * bonding. The caller must hold the RTNL semaphore. On a failure
4467 * a negative errno code is returned. On success the reference counts
4468 * are adjusted and the function returns zero.
4470 int netdev_set_master(struct net_device *slave, struct net_device *master)
4472 struct net_device *old = slave->master;
4482 slave->master = master;
4488 EXPORT_SYMBOL(netdev_set_master);
4491 * netdev_set_bond_master - set up bonding master/slave pair
4492 * @slave: slave device
4493 * @master: new master device
4495 * Changes the master device of the slave. Pass %NULL to break the
4496 * bonding. The caller must hold the RTNL semaphore. On a failure
4497 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4498 * to the routing socket and the function returns zero.
4500 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4506 err = netdev_set_master(slave, master);
4510 slave->flags |= IFF_SLAVE;
4512 slave->flags &= ~IFF_SLAVE;
4514 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4517 EXPORT_SYMBOL(netdev_set_bond_master);
4519 static void dev_change_rx_flags(struct net_device *dev, int flags)
4521 const struct net_device_ops *ops = dev->netdev_ops;
4523 if (ops->ndo_change_rx_flags)
4524 ops->ndo_change_rx_flags(dev, flags);
4527 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4529 unsigned short old_flags = dev->flags;
4535 dev->flags |= IFF_PROMISC;
4536 dev->promiscuity += inc;
4537 if (dev->promiscuity == 0) {
4540 * If inc causes overflow, untouch promisc and return error.
4543 dev->flags &= ~IFF_PROMISC;
4545 dev->promiscuity -= inc;
4546 printk(KERN_WARNING "%s: promiscuity touches roof, "
4547 "set promiscuity failed, promiscuity feature "
4548 "of device might be broken.\n", dev->name);
4552 if (dev->flags != old_flags) {
4553 printk(KERN_INFO "device %s %s promiscuous mode\n",
4554 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4556 if (audit_enabled) {
4557 current_uid_gid(&uid, &gid);
4558 audit_log(current->audit_context, GFP_ATOMIC,
4559 AUDIT_ANOM_PROMISCUOUS,
4560 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4561 dev->name, (dev->flags & IFF_PROMISC),
4562 (old_flags & IFF_PROMISC),
4563 audit_get_loginuid(current),
4565 audit_get_sessionid(current));
4568 dev_change_rx_flags(dev, IFF_PROMISC);
4574 * dev_set_promiscuity - update promiscuity count on a device
4578 * Add or remove promiscuity from a device. While the count in the device
4579 * remains above zero the interface remains promiscuous. Once it hits zero
4580 * the device reverts back to normal filtering operation. A negative inc
4581 * value is used to drop promiscuity on the device.
4582 * Return 0 if successful or a negative errno code on error.
4584 int dev_set_promiscuity(struct net_device *dev, int inc)
4586 unsigned short old_flags = dev->flags;
4589 err = __dev_set_promiscuity(dev, inc);
4592 if (dev->flags != old_flags)
4593 dev_set_rx_mode(dev);
4596 EXPORT_SYMBOL(dev_set_promiscuity);
4599 * dev_set_allmulti - update allmulti count on a device
4603 * Add or remove reception of all multicast frames to a device. While the
4604 * count in the device remains above zero the interface remains listening
4605 * to all interfaces. Once it hits zero the device reverts back to normal
4606 * filtering operation. A negative @inc value is used to drop the counter
4607 * when releasing a resource needing all multicasts.
4608 * Return 0 if successful or a negative errno code on error.
4611 int dev_set_allmulti(struct net_device *dev, int inc)
4613 unsigned short old_flags = dev->flags;
4617 dev->flags |= IFF_ALLMULTI;
4618 dev->allmulti += inc;
4619 if (dev->allmulti == 0) {
4622 * If inc causes overflow, untouch allmulti and return error.
4625 dev->flags &= ~IFF_ALLMULTI;
4627 dev->allmulti -= inc;
4628 printk(KERN_WARNING "%s: allmulti touches roof, "
4629 "set allmulti failed, allmulti feature of "
4630 "device might be broken.\n", dev->name);
4634 if (dev->flags ^ old_flags) {
4635 dev_change_rx_flags(dev, IFF_ALLMULTI);
4636 dev_set_rx_mode(dev);
4640 EXPORT_SYMBOL(dev_set_allmulti);
4643 * Upload unicast and multicast address lists to device and
4644 * configure RX filtering. When the device doesn't support unicast
4645 * filtering it is put in promiscuous mode while unicast addresses
4648 void __dev_set_rx_mode(struct net_device *dev)
4650 const struct net_device_ops *ops = dev->netdev_ops;
4652 /* dev_open will call this function so the list will stay sane. */
4653 if (!(dev->flags&IFF_UP))
4656 if (!netif_device_present(dev))
4659 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4660 /* Unicast addresses changes may only happen under the rtnl,
4661 * therefore calling __dev_set_promiscuity here is safe.
4663 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4664 __dev_set_promiscuity(dev, 1);
4665 dev->uc_promisc = true;
4666 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4667 __dev_set_promiscuity(dev, -1);
4668 dev->uc_promisc = false;
4672 if (ops->ndo_set_rx_mode)
4673 ops->ndo_set_rx_mode(dev);
4676 void dev_set_rx_mode(struct net_device *dev)
4678 netif_addr_lock_bh(dev);
4679 __dev_set_rx_mode(dev);
4680 netif_addr_unlock_bh(dev);
4684 * dev_get_flags - get flags reported to userspace
4687 * Get the combination of flag bits exported through APIs to userspace.
4689 unsigned dev_get_flags(const struct net_device *dev)
4693 flags = (dev->flags & ~(IFF_PROMISC |
4698 (dev->gflags & (IFF_PROMISC |
4701 if (netif_running(dev)) {
4702 if (netif_oper_up(dev))
4703 flags |= IFF_RUNNING;
4704 if (netif_carrier_ok(dev))
4705 flags |= IFF_LOWER_UP;
4706 if (netif_dormant(dev))
4707 flags |= IFF_DORMANT;
4712 EXPORT_SYMBOL(dev_get_flags);
4714 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4716 int old_flags = dev->flags;
4722 * Set the flags on our device.
4725 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4726 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4728 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4732 * Load in the correct multicast list now the flags have changed.
4735 if ((old_flags ^ flags) & IFF_MULTICAST)
4736 dev_change_rx_flags(dev, IFF_MULTICAST);
4738 dev_set_rx_mode(dev);
4741 * Have we downed the interface. We handle IFF_UP ourselves
4742 * according to user attempts to set it, rather than blindly
4747 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4748 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4751 dev_set_rx_mode(dev);
4754 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4755 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4757 dev->gflags ^= IFF_PROMISC;
4758 dev_set_promiscuity(dev, inc);
4761 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4762 is important. Some (broken) drivers set IFF_PROMISC, when
4763 IFF_ALLMULTI is requested not asking us and not reporting.
4765 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4766 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4768 dev->gflags ^= IFF_ALLMULTI;
4769 dev_set_allmulti(dev, inc);
4775 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4777 unsigned int changes = dev->flags ^ old_flags;
4779 if (changes & IFF_UP) {
4780 if (dev->flags & IFF_UP)
4781 call_netdevice_notifiers(NETDEV_UP, dev);
4783 call_netdevice_notifiers(NETDEV_DOWN, dev);
4786 if (dev->flags & IFF_UP &&
4787 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4788 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4792 * dev_change_flags - change device settings
4794 * @flags: device state flags
4796 * Change settings on device based state flags. The flags are
4797 * in the userspace exported format.
4799 int dev_change_flags(struct net_device *dev, unsigned flags)
4802 int old_flags = dev->flags;
4804 ret = __dev_change_flags(dev, flags);
4808 changes = old_flags ^ dev->flags;
4810 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4812 __dev_notify_flags(dev, old_flags);
4815 EXPORT_SYMBOL(dev_change_flags);
4818 * dev_set_mtu - Change maximum transfer unit
4820 * @new_mtu: new transfer unit
4822 * Change the maximum transfer size of the network device.
4824 int dev_set_mtu(struct net_device *dev, int new_mtu)
4826 const struct net_device_ops *ops = dev->netdev_ops;
4829 if (new_mtu == dev->mtu)
4832 /* MTU must be positive. */
4836 if (!netif_device_present(dev))
4840 if (ops->ndo_change_mtu)
4841 err = ops->ndo_change_mtu(dev, new_mtu);
4845 if (!err && dev->flags & IFF_UP)
4846 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4849 EXPORT_SYMBOL(dev_set_mtu);
4852 * dev_set_group - Change group this device belongs to
4854 * @new_group: group this device should belong to
4856 void dev_set_group(struct net_device *dev, int new_group)
4858 dev->group = new_group;
4860 EXPORT_SYMBOL(dev_set_group);
4863 * dev_set_mac_address - Change Media Access Control Address
4867 * Change the hardware (MAC) address of the device
4869 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4871 const struct net_device_ops *ops = dev->netdev_ops;
4874 if (!ops->ndo_set_mac_address)
4876 if (sa->sa_family != dev->type)
4878 if (!netif_device_present(dev))
4880 err = ops->ndo_set_mac_address(dev, sa);
4882 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4883 add_device_randomness(dev->dev_addr, dev->addr_len);
4886 EXPORT_SYMBOL(dev_set_mac_address);
4889 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4891 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4894 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4900 case SIOCGIFFLAGS: /* Get interface flags */
4901 ifr->ifr_flags = (short) dev_get_flags(dev);
4904 case SIOCGIFMETRIC: /* Get the metric on the interface
4905 (currently unused) */
4906 ifr->ifr_metric = 0;
4909 case SIOCGIFMTU: /* Get the MTU of a device */
4910 ifr->ifr_mtu = dev->mtu;
4915 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4917 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4918 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4919 ifr->ifr_hwaddr.sa_family = dev->type;
4927 ifr->ifr_map.mem_start = dev->mem_start;
4928 ifr->ifr_map.mem_end = dev->mem_end;
4929 ifr->ifr_map.base_addr = dev->base_addr;
4930 ifr->ifr_map.irq = dev->irq;
4931 ifr->ifr_map.dma = dev->dma;
4932 ifr->ifr_map.port = dev->if_port;
4936 ifr->ifr_ifindex = dev->ifindex;
4940 ifr->ifr_qlen = dev->tx_queue_len;
4944 /* dev_ioctl() should ensure this case
4956 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4958 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4961 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4962 const struct net_device_ops *ops;
4967 ops = dev->netdev_ops;
4970 case SIOCSIFFLAGS: /* Set interface flags */
4971 return dev_change_flags(dev, ifr->ifr_flags);
4973 case SIOCSIFMETRIC: /* Set the metric on the interface
4974 (currently unused) */
4977 case SIOCSIFMTU: /* Set the MTU of a device */
4978 return dev_set_mtu(dev, ifr->ifr_mtu);
4981 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4983 case SIOCSIFHWBROADCAST:
4984 if (ifr->ifr_hwaddr.sa_family != dev->type)
4986 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4987 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4988 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4992 if (ops->ndo_set_config) {
4993 if (!netif_device_present(dev))
4995 return ops->ndo_set_config(dev, &ifr->ifr_map);
5000 if (!ops->ndo_set_rx_mode ||
5001 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5003 if (!netif_device_present(dev))
5005 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5008 if (!ops->ndo_set_rx_mode ||
5009 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5011 if (!netif_device_present(dev))
5013 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5016 if (ifr->ifr_qlen < 0)
5018 dev->tx_queue_len = ifr->ifr_qlen;
5022 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5023 return dev_change_name(dev, ifr->ifr_newname);
5026 err = net_hwtstamp_validate(ifr);
5032 * Unknown or private ioctl
5035 if ((cmd >= SIOCDEVPRIVATE &&
5036 cmd <= SIOCDEVPRIVATE + 15) ||
5037 cmd == SIOCBONDENSLAVE ||
5038 cmd == SIOCBONDRELEASE ||
5039 cmd == SIOCBONDSETHWADDR ||
5040 cmd == SIOCBONDSLAVEINFOQUERY ||
5041 cmd == SIOCBONDINFOQUERY ||
5042 cmd == SIOCBONDCHANGEACTIVE ||
5043 cmd == SIOCGMIIPHY ||
5044 cmd == SIOCGMIIREG ||
5045 cmd == SIOCSMIIREG ||
5046 cmd == SIOCBRADDIF ||
5047 cmd == SIOCBRDELIF ||
5048 cmd == SIOCSHWTSTAMP ||
5049 cmd == SIOCWANDEV) {
5051 if (ops->ndo_do_ioctl) {
5052 if (netif_device_present(dev))
5053 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5065 * This function handles all "interface"-type I/O control requests. The actual
5066 * 'doing' part of this is dev_ifsioc above.
5070 * dev_ioctl - network device ioctl
5071 * @net: the applicable net namespace
5072 * @cmd: command to issue
5073 * @arg: pointer to a struct ifreq in user space
5075 * Issue ioctl functions to devices. This is normally called by the
5076 * user space syscall interfaces but can sometimes be useful for
5077 * other purposes. The return value is the return from the syscall if
5078 * positive or a negative errno code on error.
5081 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5087 /* One special case: SIOCGIFCONF takes ifconf argument
5088 and requires shared lock, because it sleeps writing
5092 if (cmd == SIOCGIFCONF) {
5094 ret = dev_ifconf(net, (char __user *) arg);
5098 if (cmd == SIOCGIFNAME)
5099 return dev_ifname(net, (struct ifreq __user *)arg);
5101 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5104 ifr.ifr_name[IFNAMSIZ-1] = 0;
5106 colon = strchr(ifr.ifr_name, ':');
5111 * See which interface the caller is talking about.
5116 * These ioctl calls:
5117 * - can be done by all.
5118 * - atomic and do not require locking.
5129 dev_load(net, ifr.ifr_name);
5131 ret = dev_ifsioc_locked(net, &ifr, cmd);
5136 if (copy_to_user(arg, &ifr,
5137 sizeof(struct ifreq)))
5143 dev_load(net, ifr.ifr_name);
5145 ret = dev_ethtool(net, &ifr);
5150 if (copy_to_user(arg, &ifr,
5151 sizeof(struct ifreq)))
5157 * These ioctl calls:
5158 * - require superuser power.
5159 * - require strict serialization.
5165 if (!capable(CAP_NET_ADMIN))
5167 dev_load(net, ifr.ifr_name);
5169 ret = dev_ifsioc(net, &ifr, cmd);
5174 if (copy_to_user(arg, &ifr,
5175 sizeof(struct ifreq)))
5181 * These ioctl calls:
5182 * - require superuser power.
5183 * - require strict serialization.
5184 * - do not return a value
5194 case SIOCSIFHWBROADCAST:
5197 case SIOCBONDENSLAVE:
5198 case SIOCBONDRELEASE:
5199 case SIOCBONDSETHWADDR:
5200 case SIOCBONDCHANGEACTIVE:
5204 if (!capable(CAP_NET_ADMIN))
5207 case SIOCBONDSLAVEINFOQUERY:
5208 case SIOCBONDINFOQUERY:
5209 dev_load(net, ifr.ifr_name);
5211 ret = dev_ifsioc(net, &ifr, cmd);
5216 /* Get the per device memory space. We can add this but
5217 * currently do not support it */
5219 /* Set the per device memory buffer space.
5220 * Not applicable in our case */
5225 * Unknown or private ioctl.
5228 if (cmd == SIOCWANDEV ||
5229 (cmd >= SIOCDEVPRIVATE &&
5230 cmd <= SIOCDEVPRIVATE + 15)) {
5231 dev_load(net, ifr.ifr_name);
5233 ret = dev_ifsioc(net, &ifr, cmd);
5235 if (!ret && copy_to_user(arg, &ifr,
5236 sizeof(struct ifreq)))
5240 /* Take care of Wireless Extensions */
5241 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5242 return wext_handle_ioctl(net, &ifr, cmd, arg);
5249 * dev_new_index - allocate an ifindex
5250 * @net: the applicable net namespace
5252 * Returns a suitable unique value for a new device interface
5253 * number. The caller must hold the rtnl semaphore or the
5254 * dev_base_lock to be sure it remains unique.
5256 static int dev_new_index(struct net *net)
5262 if (!__dev_get_by_index(net, ifindex))
5267 /* Delayed registration/unregisteration */
5268 static LIST_HEAD(net_todo_list);
5270 static void net_set_todo(struct net_device *dev)
5272 list_add_tail(&dev->todo_list, &net_todo_list);
5275 static void rollback_registered_many(struct list_head *head)
5277 struct net_device *dev, *tmp;
5279 BUG_ON(dev_boot_phase);
5282 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5283 /* Some devices call without registering
5284 * for initialization unwind. Remove those
5285 * devices and proceed with the remaining.
5287 if (dev->reg_state == NETREG_UNINITIALIZED) {
5288 pr_debug("unregister_netdevice: device %s/%p never "
5289 "was registered\n", dev->name, dev);
5292 list_del(&dev->unreg_list);
5295 dev->dismantle = true;
5296 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5299 /* If device is running, close it first. */
5300 dev_close_many(head);
5302 list_for_each_entry(dev, head, unreg_list) {
5303 /* And unlink it from device chain. */
5304 unlist_netdevice(dev);
5306 dev->reg_state = NETREG_UNREGISTERING;
5311 list_for_each_entry(dev, head, unreg_list) {
5312 /* Shutdown queueing discipline. */
5316 /* Notify protocols, that we are about to destroy
5317 this device. They should clean all the things.
5319 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5321 if (!dev->rtnl_link_ops ||
5322 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5323 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5326 * Flush the unicast and multicast chains
5331 if (dev->netdev_ops->ndo_uninit)
5332 dev->netdev_ops->ndo_uninit(dev);
5334 /* Notifier chain MUST detach us from master device. */
5335 WARN_ON(dev->master);
5337 /* Remove entries from kobject tree */
5338 netdev_unregister_kobject(dev);
5341 /* Process any work delayed until the end of the batch */
5342 dev = list_first_entry(head, struct net_device, unreg_list);
5343 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5347 list_for_each_entry(dev, head, unreg_list)
5351 static void rollback_registered(struct net_device *dev)
5355 list_add(&dev->unreg_list, &single);
5356 rollback_registered_many(&single);
5360 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5362 /* Fix illegal checksum combinations */
5363 if ((features & NETIF_F_HW_CSUM) &&
5364 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5365 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5366 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5369 if ((features & NETIF_F_NO_CSUM) &&
5370 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5371 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5372 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5375 /* Fix illegal SG+CSUM combinations. */
5376 if ((features & NETIF_F_SG) &&
5377 !(features & NETIF_F_ALL_CSUM)) {
5379 "Dropping NETIF_F_SG since no checksum feature.\n");
5380 features &= ~NETIF_F_SG;
5383 /* TSO requires that SG is present as well. */
5384 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5385 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5386 features &= ~NETIF_F_ALL_TSO;
5389 /* TSO ECN requires that TSO is present as well. */
5390 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5391 features &= ~NETIF_F_TSO_ECN;
5393 /* Software GSO depends on SG. */
5394 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5395 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5396 features &= ~NETIF_F_GSO;
5399 /* UFO needs SG and checksumming */
5400 if (features & NETIF_F_UFO) {
5401 /* maybe split UFO into V4 and V6? */
5402 if (!((features & NETIF_F_GEN_CSUM) ||
5403 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5404 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5406 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5407 features &= ~NETIF_F_UFO;
5410 if (!(features & NETIF_F_SG)) {
5412 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5413 features &= ~NETIF_F_UFO;
5420 int __netdev_update_features(struct net_device *dev)
5427 features = netdev_get_wanted_features(dev);
5429 if (dev->netdev_ops->ndo_fix_features)
5430 features = dev->netdev_ops->ndo_fix_features(dev, features);
5432 /* driver might be less strict about feature dependencies */
5433 features = netdev_fix_features(dev, features);
5435 if (dev->features == features)
5438 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5439 dev->features, features);
5441 if (dev->netdev_ops->ndo_set_features)
5442 err = dev->netdev_ops->ndo_set_features(dev, features);
5444 if (unlikely(err < 0)) {
5446 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5447 err, features, dev->features);
5452 dev->features = features;
5458 * netdev_update_features - recalculate device features
5459 * @dev: the device to check
5461 * Recalculate dev->features set and send notifications if it
5462 * has changed. Should be called after driver or hardware dependent
5463 * conditions might have changed that influence the features.
5465 void netdev_update_features(struct net_device *dev)
5467 if (__netdev_update_features(dev))
5468 netdev_features_change(dev);
5470 EXPORT_SYMBOL(netdev_update_features);
5473 * netdev_change_features - recalculate device features
5474 * @dev: the device to check
5476 * Recalculate dev->features set and send notifications even
5477 * if they have not changed. Should be called instead of
5478 * netdev_update_features() if also dev->vlan_features might
5479 * have changed to allow the changes to be propagated to stacked
5482 void netdev_change_features(struct net_device *dev)
5484 __netdev_update_features(dev);
5485 netdev_features_change(dev);
5487 EXPORT_SYMBOL(netdev_change_features);
5490 * netif_stacked_transfer_operstate - transfer operstate
5491 * @rootdev: the root or lower level device to transfer state from
5492 * @dev: the device to transfer operstate to
5494 * Transfer operational state from root to device. This is normally
5495 * called when a stacking relationship exists between the root
5496 * device and the device(a leaf device).
5498 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5499 struct net_device *dev)
5501 if (rootdev->operstate == IF_OPER_DORMANT)
5502 netif_dormant_on(dev);
5504 netif_dormant_off(dev);
5506 if (netif_carrier_ok(rootdev)) {
5507 if (!netif_carrier_ok(dev))
5508 netif_carrier_on(dev);
5510 if (netif_carrier_ok(dev))
5511 netif_carrier_off(dev);
5514 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5517 static int netif_alloc_rx_queues(struct net_device *dev)
5519 unsigned int i, count = dev->num_rx_queues;
5520 struct netdev_rx_queue *rx;
5524 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5526 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5531 for (i = 0; i < count; i++)
5537 static void netdev_init_one_queue(struct net_device *dev,
5538 struct netdev_queue *queue, void *_unused)
5540 /* Initialize queue lock */
5541 spin_lock_init(&queue->_xmit_lock);
5542 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5543 queue->xmit_lock_owner = -1;
5544 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5548 static int netif_alloc_netdev_queues(struct net_device *dev)
5550 unsigned int count = dev->num_tx_queues;
5551 struct netdev_queue *tx;
5555 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5557 pr_err("netdev: Unable to allocate %u tx queues.\n",
5563 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5564 spin_lock_init(&dev->tx_global_lock);
5570 * register_netdevice - register a network device
5571 * @dev: device to register
5573 * Take a completed network device structure and add it to the kernel
5574 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5575 * chain. 0 is returned on success. A negative errno code is returned
5576 * on a failure to set up the device, or if the name is a duplicate.
5578 * Callers must hold the rtnl semaphore. You may want
5579 * register_netdev() instead of this.
5582 * The locking appears insufficient to guarantee two parallel registers
5583 * will not get the same name.
5586 int register_netdevice(struct net_device *dev)
5589 struct net *net = dev_net(dev);
5591 BUG_ON(dev_boot_phase);
5596 /* When net_device's are persistent, this will be fatal. */
5597 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5600 spin_lock_init(&dev->addr_list_lock);
5601 netdev_set_addr_lockdep_class(dev);
5605 ret = dev_get_valid_name(dev, dev->name);
5609 /* Init, if this function is available */
5610 if (dev->netdev_ops->ndo_init) {
5611 ret = dev->netdev_ops->ndo_init(dev);
5619 dev->ifindex = dev_new_index(net);
5620 if (dev->iflink == -1)
5621 dev->iflink = dev->ifindex;
5623 /* Transfer changeable features to wanted_features and enable
5624 * software offloads (GSO and GRO).
5626 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5627 dev->features |= NETIF_F_SOFT_FEATURES;
5628 dev->wanted_features = dev->features & dev->hw_features;
5630 /* Turn on no cache copy if HW is doing checksum */
5631 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5632 if ((dev->features & NETIF_F_ALL_CSUM) &&
5633 !(dev->features & NETIF_F_NO_CSUM)) {
5634 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5635 dev->features |= NETIF_F_NOCACHE_COPY;
5638 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5640 dev->vlan_features |= NETIF_F_HIGHDMA;
5642 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5643 ret = notifier_to_errno(ret);
5647 ret = netdev_register_kobject(dev);
5650 dev->reg_state = NETREG_REGISTERED;
5652 __netdev_update_features(dev);
5655 * Default initial state at registry is that the
5656 * device is present.
5659 set_bit(__LINK_STATE_PRESENT, &dev->state);
5661 dev_init_scheduler(dev);
5663 list_netdevice(dev);
5664 add_device_randomness(dev->dev_addr, dev->addr_len);
5666 /* Notify protocols, that a new device appeared. */
5667 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5668 ret = notifier_to_errno(ret);
5670 rollback_registered(dev);
5671 dev->reg_state = NETREG_UNREGISTERED;
5674 * Prevent userspace races by waiting until the network
5675 * device is fully setup before sending notifications.
5677 if (!dev->rtnl_link_ops ||
5678 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5679 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5685 if (dev->netdev_ops->ndo_uninit)
5686 dev->netdev_ops->ndo_uninit(dev);
5689 EXPORT_SYMBOL(register_netdevice);
5692 * init_dummy_netdev - init a dummy network device for NAPI
5693 * @dev: device to init
5695 * This takes a network device structure and initialize the minimum
5696 * amount of fields so it can be used to schedule NAPI polls without
5697 * registering a full blown interface. This is to be used by drivers
5698 * that need to tie several hardware interfaces to a single NAPI
5699 * poll scheduler due to HW limitations.
5701 int init_dummy_netdev(struct net_device *dev)
5703 /* Clear everything. Note we don't initialize spinlocks
5704 * are they aren't supposed to be taken by any of the
5705 * NAPI code and this dummy netdev is supposed to be
5706 * only ever used for NAPI polls
5708 memset(dev, 0, sizeof(struct net_device));
5710 /* make sure we BUG if trying to hit standard
5711 * register/unregister code path
5713 dev->reg_state = NETREG_DUMMY;
5715 /* NAPI wants this */
5716 INIT_LIST_HEAD(&dev->napi_list);
5718 /* a dummy interface is started by default */
5719 set_bit(__LINK_STATE_PRESENT, &dev->state);
5720 set_bit(__LINK_STATE_START, &dev->state);
5722 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5723 * because users of this 'device' dont need to change
5729 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5733 * register_netdev - register a network device
5734 * @dev: device to register
5736 * Take a completed network device structure and add it to the kernel
5737 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5738 * chain. 0 is returned on success. A negative errno code is returned
5739 * on a failure to set up the device, or if the name is a duplicate.
5741 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5742 * and expands the device name if you passed a format string to
5745 int register_netdev(struct net_device *dev)
5750 err = register_netdevice(dev);
5754 EXPORT_SYMBOL(register_netdev);
5756 int netdev_refcnt_read(const struct net_device *dev)
5760 for_each_possible_cpu(i)
5761 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5764 EXPORT_SYMBOL(netdev_refcnt_read);
5767 * netdev_wait_allrefs - wait until all references are gone.
5769 * This is called when unregistering network devices.
5771 * Any protocol or device that holds a reference should register
5772 * for netdevice notification, and cleanup and put back the
5773 * reference if they receive an UNREGISTER event.
5774 * We can get stuck here if buggy protocols don't correctly
5777 static void netdev_wait_allrefs(struct net_device *dev)
5779 unsigned long rebroadcast_time, warning_time;
5782 linkwatch_forget_dev(dev);
5784 rebroadcast_time = warning_time = jiffies;
5785 refcnt = netdev_refcnt_read(dev);
5787 while (refcnt != 0) {
5788 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5791 /* Rebroadcast unregister notification */
5792 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5793 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5794 * should have already handle it the first time */
5796 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5798 /* We must not have linkwatch events
5799 * pending on unregister. If this
5800 * happens, we simply run the queue
5801 * unscheduled, resulting in a noop
5804 linkwatch_run_queue();
5809 rebroadcast_time = jiffies;
5814 refcnt = netdev_refcnt_read(dev);
5816 if (time_after(jiffies, warning_time + 10 * HZ)) {
5817 printk(KERN_EMERG "unregister_netdevice: "
5818 "waiting for %s to become free. Usage "
5821 warning_time = jiffies;
5830 * register_netdevice(x1);
5831 * register_netdevice(x2);
5833 * unregister_netdevice(y1);
5834 * unregister_netdevice(y2);
5840 * We are invoked by rtnl_unlock().
5841 * This allows us to deal with problems:
5842 * 1) We can delete sysfs objects which invoke hotplug
5843 * without deadlocking with linkwatch via keventd.
5844 * 2) Since we run with the RTNL semaphore not held, we can sleep
5845 * safely in order to wait for the netdev refcnt to drop to zero.
5847 * We must not return until all unregister events added during
5848 * the interval the lock was held have been completed.
5850 void netdev_run_todo(void)
5852 struct list_head list;
5854 /* Snapshot list, allow later requests */
5855 list_replace_init(&net_todo_list, &list);
5859 /* Wait for rcu callbacks to finish before attempting to drain
5860 * the device list. This usually avoids a 250ms wait.
5862 if (!list_empty(&list))
5865 while (!list_empty(&list)) {
5866 struct net_device *dev
5867 = list_first_entry(&list, struct net_device, todo_list);
5868 list_del(&dev->todo_list);
5870 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5871 printk(KERN_ERR "network todo '%s' but state %d\n",
5872 dev->name, dev->reg_state);
5877 dev->reg_state = NETREG_UNREGISTERED;
5879 on_each_cpu(flush_backlog, dev, 1);
5881 netdev_wait_allrefs(dev);
5884 BUG_ON(netdev_refcnt_read(dev));
5885 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5886 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5887 WARN_ON(dev->dn_ptr);
5889 if (dev->destructor)
5890 dev->destructor(dev);
5892 /* Free network device */
5893 kobject_put(&dev->dev.kobj);
5897 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5898 * fields in the same order, with only the type differing.
5900 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5901 const struct net_device_stats *netdev_stats)
5903 #if BITS_PER_LONG == 64
5904 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5905 memcpy(stats64, netdev_stats, sizeof(*stats64));
5907 size_t i, n = sizeof(*stats64) / sizeof(u64);
5908 const unsigned long *src = (const unsigned long *)netdev_stats;
5909 u64 *dst = (u64 *)stats64;
5911 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5912 sizeof(*stats64) / sizeof(u64));
5913 for (i = 0; i < n; i++)
5919 * dev_get_stats - get network device statistics
5920 * @dev: device to get statistics from
5921 * @storage: place to store stats
5923 * Get network statistics from device. Return @storage.
5924 * The device driver may provide its own method by setting
5925 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5926 * otherwise the internal statistics structure is used.
5928 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5929 struct rtnl_link_stats64 *storage)
5931 const struct net_device_ops *ops = dev->netdev_ops;
5933 if (ops->ndo_get_stats64) {
5934 memset(storage, 0, sizeof(*storage));
5935 ops->ndo_get_stats64(dev, storage);
5936 } else if (ops->ndo_get_stats) {
5937 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5939 netdev_stats_to_stats64(storage, &dev->stats);
5941 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5944 EXPORT_SYMBOL(dev_get_stats);
5946 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5948 struct netdev_queue *queue = dev_ingress_queue(dev);
5950 #ifdef CONFIG_NET_CLS_ACT
5953 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5956 netdev_init_one_queue(dev, queue, NULL);
5957 queue->qdisc = &noop_qdisc;
5958 queue->qdisc_sleeping = &noop_qdisc;
5959 rcu_assign_pointer(dev->ingress_queue, queue);
5965 * alloc_netdev_mqs - allocate network device
5966 * @sizeof_priv: size of private data to allocate space for
5967 * @name: device name format string
5968 * @setup: callback to initialize device
5969 * @txqs: the number of TX subqueues to allocate
5970 * @rxqs: the number of RX subqueues to allocate
5972 * Allocates a struct net_device with private data area for driver use
5973 * and performs basic initialization. Also allocates subquue structs
5974 * for each queue on the device.
5976 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5977 void (*setup)(struct net_device *),
5978 unsigned int txqs, unsigned int rxqs)
5980 struct net_device *dev;
5982 struct net_device *p;
5984 BUG_ON(strlen(name) >= sizeof(dev->name));
5987 pr_err("alloc_netdev: Unable to allocate device "
5988 "with zero queues.\n");
5994 pr_err("alloc_netdev: Unable to allocate device "
5995 "with zero RX queues.\n");
6000 alloc_size = sizeof(struct net_device);
6002 /* ensure 32-byte alignment of private area */
6003 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6004 alloc_size += sizeof_priv;
6006 /* ensure 32-byte alignment of whole construct */
6007 alloc_size += NETDEV_ALIGN - 1;
6009 p = kzalloc(alloc_size, GFP_KERNEL);
6011 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
6015 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6016 dev->padded = (char *)dev - (char *)p;
6018 dev->pcpu_refcnt = alloc_percpu(int);
6019 if (!dev->pcpu_refcnt)
6022 if (dev_addr_init(dev))
6028 dev_net_set(dev, &init_net);
6030 dev->gso_max_size = GSO_MAX_SIZE;
6031 dev->gso_max_segs = GSO_MAX_SEGS;
6033 INIT_LIST_HEAD(&dev->napi_list);
6034 INIT_LIST_HEAD(&dev->unreg_list);
6035 INIT_LIST_HEAD(&dev->link_watch_list);
6036 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6039 dev->num_tx_queues = txqs;
6040 dev->real_num_tx_queues = txqs;
6041 if (netif_alloc_netdev_queues(dev))
6045 dev->num_rx_queues = rxqs;
6046 dev->real_num_rx_queues = rxqs;
6047 if (netif_alloc_rx_queues(dev))
6051 strcpy(dev->name, name);
6052 dev->group = INIT_NETDEV_GROUP;
6060 free_percpu(dev->pcpu_refcnt);
6070 EXPORT_SYMBOL(alloc_netdev_mqs);
6073 * free_netdev - free network device
6076 * This function does the last stage of destroying an allocated device
6077 * interface. The reference to the device object is released.
6078 * If this is the last reference then it will be freed.
6080 void free_netdev(struct net_device *dev)
6082 struct napi_struct *p, *n;
6084 release_net(dev_net(dev));
6091 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6093 /* Flush device addresses */
6094 dev_addr_flush(dev);
6096 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6099 free_percpu(dev->pcpu_refcnt);
6100 dev->pcpu_refcnt = NULL;
6102 /* Compatibility with error handling in drivers */
6103 if (dev->reg_state == NETREG_UNINITIALIZED) {
6104 kfree((char *)dev - dev->padded);
6108 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6109 dev->reg_state = NETREG_RELEASED;
6111 /* will free via device release */
6112 put_device(&dev->dev);
6114 EXPORT_SYMBOL(free_netdev);
6117 * synchronize_net - Synchronize with packet receive processing
6119 * Wait for packets currently being received to be done.
6120 * Does not block later packets from starting.
6122 void synchronize_net(void)
6125 if (rtnl_is_locked())
6126 synchronize_rcu_expedited();
6130 EXPORT_SYMBOL(synchronize_net);
6133 * unregister_netdevice_queue - remove device from the kernel
6137 * This function shuts down a device interface and removes it
6138 * from the kernel tables.
6139 * If head not NULL, device is queued to be unregistered later.
6141 * Callers must hold the rtnl semaphore. You may want
6142 * unregister_netdev() instead of this.
6145 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6150 list_move_tail(&dev->unreg_list, head);
6152 rollback_registered(dev);
6153 /* Finish processing unregister after unlock */
6157 EXPORT_SYMBOL(unregister_netdevice_queue);
6160 * unregister_netdevice_many - unregister many devices
6161 * @head: list of devices
6163 void unregister_netdevice_many(struct list_head *head)
6165 struct net_device *dev;
6167 if (!list_empty(head)) {
6168 rollback_registered_many(head);
6169 list_for_each_entry(dev, head, unreg_list)
6173 EXPORT_SYMBOL(unregister_netdevice_many);
6176 * unregister_netdev - remove device from the kernel
6179 * This function shuts down a device interface and removes it
6180 * from the kernel tables.
6182 * This is just a wrapper for unregister_netdevice that takes
6183 * the rtnl semaphore. In general you want to use this and not
6184 * unregister_netdevice.
6186 void unregister_netdev(struct net_device *dev)
6189 unregister_netdevice(dev);
6192 EXPORT_SYMBOL(unregister_netdev);
6195 * dev_change_net_namespace - move device to different nethost namespace
6197 * @net: network namespace
6198 * @pat: If not NULL name pattern to try if the current device name
6199 * is already taken in the destination network namespace.
6201 * This function shuts down a device interface and moves it
6202 * to a new network namespace. On success 0 is returned, on
6203 * a failure a netagive errno code is returned.
6205 * Callers must hold the rtnl semaphore.
6208 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6214 /* Don't allow namespace local devices to be moved. */
6216 if (dev->features & NETIF_F_NETNS_LOCAL)
6219 /* Ensure the device has been registrered */
6221 if (dev->reg_state != NETREG_REGISTERED)
6224 /* Get out if there is nothing todo */
6226 if (net_eq(dev_net(dev), net))
6229 /* Pick the destination device name, and ensure
6230 * we can use it in the destination network namespace.
6233 if (__dev_get_by_name(net, dev->name)) {
6234 /* We get here if we can't use the current device name */
6237 if (dev_get_valid_name(dev, pat) < 0)
6242 * And now a mini version of register_netdevice unregister_netdevice.
6245 /* If device is running close it first. */
6248 /* And unlink it from device chain */
6250 unlist_netdevice(dev);
6254 /* Shutdown queueing discipline. */
6257 /* Notify protocols, that we are about to destroy
6258 this device. They should clean all the things.
6260 Note that dev->reg_state stays at NETREG_REGISTERED.
6261 This is wanted because this way 8021q and macvlan know
6262 the device is just moving and can keep their slaves up.
6264 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6265 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6266 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6269 * Flush the unicast and multicast chains
6274 /* Actually switch the network namespace */
6275 dev_net_set(dev, net);
6277 /* If there is an ifindex conflict assign a new one */
6278 if (__dev_get_by_index(net, dev->ifindex)) {
6279 int iflink = (dev->iflink == dev->ifindex);
6280 dev->ifindex = dev_new_index(net);
6282 dev->iflink = dev->ifindex;
6285 /* Fixup kobjects */
6286 err = device_rename(&dev->dev, dev->name);
6289 /* Add the device back in the hashes */
6290 list_netdevice(dev);
6292 /* Notify protocols, that a new device appeared. */
6293 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6296 * Prevent userspace races by waiting until the network
6297 * device is fully setup before sending notifications.
6299 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6306 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6308 static int dev_cpu_callback(struct notifier_block *nfb,
6309 unsigned long action,
6312 struct sk_buff **list_skb;
6313 struct sk_buff *skb;
6314 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6315 struct softnet_data *sd, *oldsd;
6317 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6320 local_irq_disable();
6321 cpu = smp_processor_id();
6322 sd = &per_cpu(softnet_data, cpu);
6323 oldsd = &per_cpu(softnet_data, oldcpu);
6325 /* Find end of our completion_queue. */
6326 list_skb = &sd->completion_queue;
6328 list_skb = &(*list_skb)->next;
6329 /* Append completion queue from offline CPU. */
6330 *list_skb = oldsd->completion_queue;
6331 oldsd->completion_queue = NULL;
6333 /* Append output queue from offline CPU. */
6334 if (oldsd->output_queue) {
6335 *sd->output_queue_tailp = oldsd->output_queue;
6336 sd->output_queue_tailp = oldsd->output_queue_tailp;
6337 oldsd->output_queue = NULL;
6338 oldsd->output_queue_tailp = &oldsd->output_queue;
6340 /* Append NAPI poll list from offline CPU. */
6341 if (!list_empty(&oldsd->poll_list)) {
6342 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6343 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6346 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6349 /* Process offline CPU's input_pkt_queue */
6350 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6352 input_queue_head_incr(oldsd);
6354 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6356 input_queue_head_incr(oldsd);
6364 * netdev_increment_features - increment feature set by one
6365 * @all: current feature set
6366 * @one: new feature set
6367 * @mask: mask feature set
6369 * Computes a new feature set after adding a device with feature set
6370 * @one to the master device with current feature set @all. Will not
6371 * enable anything that is off in @mask. Returns the new feature set.
6373 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6375 if (mask & NETIF_F_GEN_CSUM)
6376 mask |= NETIF_F_ALL_CSUM;
6377 mask |= NETIF_F_VLAN_CHALLENGED;
6379 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6380 all &= one | ~NETIF_F_ALL_FOR_ALL;
6382 /* If device needs checksumming, downgrade to it. */
6383 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6384 all &= ~NETIF_F_NO_CSUM;
6386 /* If one device supports hw checksumming, set for all. */
6387 if (all & NETIF_F_GEN_CSUM)
6388 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6392 EXPORT_SYMBOL(netdev_increment_features);
6394 static struct hlist_head *netdev_create_hash(void)
6397 struct hlist_head *hash;
6399 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6401 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6402 INIT_HLIST_HEAD(&hash[i]);
6407 /* Initialize per network namespace state */
6408 static int __net_init netdev_init(struct net *net)
6410 if (net != &init_net)
6411 INIT_LIST_HEAD(&net->dev_base_head);
6413 net->dev_name_head = netdev_create_hash();
6414 if (net->dev_name_head == NULL)
6417 net->dev_index_head = netdev_create_hash();
6418 if (net->dev_index_head == NULL)
6424 kfree(net->dev_name_head);
6430 * netdev_drivername - network driver for the device
6431 * @dev: network device
6433 * Determine network driver for device.
6435 const char *netdev_drivername(const struct net_device *dev)
6437 const struct device_driver *driver;
6438 const struct device *parent;
6439 const char *empty = "";
6441 parent = dev->dev.parent;
6445 driver = parent->driver;
6446 if (driver && driver->name)
6447 return driver->name;
6451 int __netdev_printk(const char *level, const struct net_device *dev,
6452 struct va_format *vaf)
6456 if (dev && dev->dev.parent)
6457 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6458 netdev_name(dev), vaf);
6460 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6462 r = printk("%s(NULL net_device): %pV", level, vaf);
6466 EXPORT_SYMBOL(__netdev_printk);
6468 int netdev_printk(const char *level, const struct net_device *dev,
6469 const char *format, ...)
6471 struct va_format vaf;
6475 va_start(args, format);
6480 r = __netdev_printk(level, dev, &vaf);
6485 EXPORT_SYMBOL(netdev_printk);
6487 #define define_netdev_printk_level(func, level) \
6488 int func(const struct net_device *dev, const char *fmt, ...) \
6491 struct va_format vaf; \
6494 va_start(args, fmt); \
6499 r = __netdev_printk(level, dev, &vaf); \
6504 EXPORT_SYMBOL(func);
6506 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6507 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6508 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6509 define_netdev_printk_level(netdev_err, KERN_ERR);
6510 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6511 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6512 define_netdev_printk_level(netdev_info, KERN_INFO);
6514 static void __net_exit netdev_exit(struct net *net)
6516 kfree(net->dev_name_head);
6517 kfree(net->dev_index_head);
6520 static struct pernet_operations __net_initdata netdev_net_ops = {
6521 .init = netdev_init,
6522 .exit = netdev_exit,
6525 static void __net_exit default_device_exit(struct net *net)
6527 struct net_device *dev, *aux;
6529 * Push all migratable network devices back to the
6530 * initial network namespace
6533 for_each_netdev_safe(net, dev, aux) {
6535 char fb_name[IFNAMSIZ];
6537 /* Ignore unmoveable devices (i.e. loopback) */
6538 if (dev->features & NETIF_F_NETNS_LOCAL)
6541 /* Leave virtual devices for the generic cleanup */
6542 if (dev->rtnl_link_ops)
6545 /* Push remaining network devices to init_net */
6546 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6547 err = dev_change_net_namespace(dev, &init_net, fb_name);
6549 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6550 __func__, dev->name, err);
6557 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6559 /* At exit all network devices most be removed from a network
6560 * namespace. Do this in the reverse order of registration.
6561 * Do this across as many network namespaces as possible to
6562 * improve batching efficiency.
6564 struct net_device *dev;
6566 LIST_HEAD(dev_kill_list);
6569 list_for_each_entry(net, net_list, exit_list) {
6570 for_each_netdev_reverse(net, dev) {
6571 if (dev->rtnl_link_ops)
6572 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6574 unregister_netdevice_queue(dev, &dev_kill_list);
6577 unregister_netdevice_many(&dev_kill_list);
6578 list_del(&dev_kill_list);
6582 static struct pernet_operations __net_initdata default_device_ops = {
6583 .exit = default_device_exit,
6584 .exit_batch = default_device_exit_batch,
6588 * Initialize the DEV module. At boot time this walks the device list and
6589 * unhooks any devices that fail to initialise (normally hardware not
6590 * present) and leaves us with a valid list of present and active devices.
6595 * This is called single threaded during boot, so no need
6596 * to take the rtnl semaphore.
6598 static int __init net_dev_init(void)
6600 int i, rc = -ENOMEM;
6602 BUG_ON(!dev_boot_phase);
6604 if (dev_proc_init())
6607 if (netdev_kobject_init())
6610 INIT_LIST_HEAD(&ptype_all);
6611 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6612 INIT_LIST_HEAD(&ptype_base[i]);
6614 if (register_pernet_subsys(&netdev_net_ops))
6618 * Initialise the packet receive queues.
6621 for_each_possible_cpu(i) {
6622 struct softnet_data *sd = &per_cpu(softnet_data, i);
6624 memset(sd, 0, sizeof(*sd));
6625 skb_queue_head_init(&sd->input_pkt_queue);
6626 skb_queue_head_init(&sd->process_queue);
6627 sd->completion_queue = NULL;
6628 INIT_LIST_HEAD(&sd->poll_list);
6629 sd->output_queue = NULL;
6630 sd->output_queue_tailp = &sd->output_queue;
6632 sd->csd.func = rps_trigger_softirq;
6638 sd->backlog.poll = process_backlog;
6639 sd->backlog.weight = weight_p;
6640 sd->backlog.gro_list = NULL;
6641 sd->backlog.gro_count = 0;
6646 /* The loopback device is special if any other network devices
6647 * is present in a network namespace the loopback device must
6648 * be present. Since we now dynamically allocate and free the
6649 * loopback device ensure this invariant is maintained by
6650 * keeping the loopback device as the first device on the
6651 * list of network devices. Ensuring the loopback devices
6652 * is the first device that appears and the last network device
6655 if (register_pernet_device(&loopback_net_ops))
6658 if (register_pernet_device(&default_device_ops))
6661 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6662 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6664 hotcpu_notifier(dev_cpu_callback, 0);
6672 subsys_initcall(net_dev_init);
6674 static int __init initialize_hashrnd(void)
6676 get_random_bytes(&hashrnd, sizeof(hashrnd));
6680 late_initcall_sync(initialize_hashrnd);