2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/if_tunnel.h>
137 #include <linux/if_pppox.h>
138 #include <linux/ppp_defs.h>
139 #include <linux/net_tstamp.h>
141 #include "net-sysfs.h"
143 /* Instead of increasing this, you should create a hash table. */
144 #define MAX_GRO_SKBS 8
146 /* This should be increased if a protocol with a bigger head is added. */
147 #define GRO_MAX_HEAD (MAX_HEADER + 128)
150 * The list of packet types we will receive (as opposed to discard)
151 * and the routines to invoke.
153 * Why 16. Because with 16 the only overlap we get on a hash of the
154 * low nibble of the protocol value is RARP/SNAP/X.25.
156 * NOTE: That is no longer true with the addition of VLAN tags. Not
157 * sure which should go first, but I bet it won't make much
158 * difference if we are running VLANs. The good news is that
159 * this protocol won't be in the list unless compiled in, so
160 * the average user (w/out VLANs) will not be adversely affected.
177 #define PTYPE_HASH_SIZE (16)
178 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
180 static DEFINE_SPINLOCK(ptype_lock);
181 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
182 static struct list_head ptype_all __read_mostly; /* Taps */
185 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
188 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
190 * Writers must hold the rtnl semaphore while they loop through the
191 * dev_base_head list, and hold dev_base_lock for writing when they do the
192 * actual updates. This allows pure readers to access the list even
193 * while a writer is preparing to update it.
195 * To put it another way, dev_base_lock is held for writing only to
196 * protect against pure readers; the rtnl semaphore provides the
197 * protection against other writers.
199 * See, for example usages, register_netdevice() and
200 * unregister_netdevice(), which must be called with the rtnl
203 DEFINE_RWLOCK(dev_base_lock);
204 EXPORT_SYMBOL(dev_base_lock);
206 static inline void dev_base_seq_inc(struct net *net)
208 while (++net->dev_base_seq == 0);
211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
213 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
214 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
217 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
219 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
222 static inline void rps_lock(struct softnet_data *sd)
225 spin_lock(&sd->input_pkt_queue.lock);
229 static inline void rps_unlock(struct softnet_data *sd)
232 spin_unlock(&sd->input_pkt_queue.lock);
236 /* Device list insertion */
237 static int list_netdevice(struct net_device *dev)
239 struct net *net = dev_net(dev);
243 write_lock_bh(&dev_base_lock);
244 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
245 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
246 hlist_add_head_rcu(&dev->index_hlist,
247 dev_index_hash(net, dev->ifindex));
248 write_unlock_bh(&dev_base_lock);
250 dev_base_seq_inc(net);
255 /* Device list removal
256 * caller must respect a RCU grace period before freeing/reusing dev
258 static void unlist_netdevice(struct net_device *dev)
262 /* Unlink dev from the device chain */
263 write_lock_bh(&dev_base_lock);
264 list_del_rcu(&dev->dev_list);
265 hlist_del_rcu(&dev->name_hlist);
266 hlist_del_rcu(&dev->index_hlist);
267 write_unlock_bh(&dev_base_lock);
269 dev_base_seq_inc(dev_net(dev));
276 static RAW_NOTIFIER_HEAD(netdev_chain);
279 * Device drivers call our routines to queue packets here. We empty the
280 * queue in the local softnet handler.
283 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
284 EXPORT_PER_CPU_SYMBOL(softnet_data);
286 #ifdef CONFIG_LOCKDEP
288 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
289 * according to dev->type
291 static const unsigned short netdev_lock_type[] =
292 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
293 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
294 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
295 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
296 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
297 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
298 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
299 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
300 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
301 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
302 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
303 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
304 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
305 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
306 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
307 ARPHRD_VOID, ARPHRD_NONE};
309 static const char *const netdev_lock_name[] =
310 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
311 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
312 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
313 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
314 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
315 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
316 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
317 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
318 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
319 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
320 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
321 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
322 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
323 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
324 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
325 "_xmit_VOID", "_xmit_NONE"};
327 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
328 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
330 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
334 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
335 if (netdev_lock_type[i] == dev_type)
337 /* the last key is used by default */
338 return ARRAY_SIZE(netdev_lock_type) - 1;
341 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
342 unsigned short dev_type)
346 i = netdev_lock_pos(dev_type);
347 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
348 netdev_lock_name[i]);
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
355 i = netdev_lock_pos(dev->type);
356 lockdep_set_class_and_name(&dev->addr_list_lock,
357 &netdev_addr_lock_key[i],
358 netdev_lock_name[i]);
361 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
362 unsigned short dev_type)
365 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
370 /*******************************************************************************
372 Protocol management and registration routines
374 *******************************************************************************/
377 * Add a protocol ID to the list. Now that the input handler is
378 * smarter we can dispense with all the messy stuff that used to be
381 * BEWARE!!! Protocol handlers, mangling input packets,
382 * MUST BE last in hash buckets and checking protocol handlers
383 * MUST start from promiscuous ptype_all chain in net_bh.
384 * It is true now, do not change it.
385 * Explanation follows: if protocol handler, mangling packet, will
386 * be the first on list, it is not able to sense, that packet
387 * is cloned and should be copied-on-write, so that it will
388 * change it and subsequent readers will get broken packet.
392 static inline struct list_head *ptype_head(const struct packet_type *pt)
394 if (pt->type == htons(ETH_P_ALL))
397 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
401 * dev_add_pack - add packet handler
402 * @pt: packet type declaration
404 * Add a protocol handler to the networking stack. The passed &packet_type
405 * is linked into kernel lists and may not be freed until it has been
406 * removed from the kernel lists.
408 * This call does not sleep therefore it can not
409 * guarantee all CPU's that are in middle of receiving packets
410 * will see the new packet type (until the next received packet).
413 void dev_add_pack(struct packet_type *pt)
415 struct list_head *head = ptype_head(pt);
417 spin_lock(&ptype_lock);
418 list_add_rcu(&pt->list, head);
419 spin_unlock(&ptype_lock);
421 EXPORT_SYMBOL(dev_add_pack);
424 * __dev_remove_pack - remove packet handler
425 * @pt: packet type declaration
427 * Remove a protocol handler that was previously added to the kernel
428 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
429 * from the kernel lists and can be freed or reused once this function
432 * The packet type might still be in use by receivers
433 * and must not be freed until after all the CPU's have gone
434 * through a quiescent state.
436 void __dev_remove_pack(struct packet_type *pt)
438 struct list_head *head = ptype_head(pt);
439 struct packet_type *pt1;
441 spin_lock(&ptype_lock);
443 list_for_each_entry(pt1, head, list) {
445 list_del_rcu(&pt->list);
450 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
452 spin_unlock(&ptype_lock);
454 EXPORT_SYMBOL(__dev_remove_pack);
457 * dev_remove_pack - remove packet handler
458 * @pt: packet type declaration
460 * Remove a protocol handler that was previously added to the kernel
461 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
462 * from the kernel lists and can be freed or reused once this function
465 * This call sleeps to guarantee that no CPU is looking at the packet
468 void dev_remove_pack(struct packet_type *pt)
470 __dev_remove_pack(pt);
474 EXPORT_SYMBOL(dev_remove_pack);
476 /******************************************************************************
478 Device Boot-time Settings Routines
480 *******************************************************************************/
482 /* Boot time configuration table */
483 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
486 * netdev_boot_setup_add - add new setup entry
487 * @name: name of the device
488 * @map: configured settings for the device
490 * Adds new setup entry to the dev_boot_setup list. The function
491 * returns 0 on error and 1 on success. This is a generic routine to
494 static int netdev_boot_setup_add(char *name, struct ifmap *map)
496 struct netdev_boot_setup *s;
500 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
501 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
502 memset(s[i].name, 0, sizeof(s[i].name));
503 strlcpy(s[i].name, name, IFNAMSIZ);
504 memcpy(&s[i].map, map, sizeof(s[i].map));
509 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
513 * netdev_boot_setup_check - check boot time settings
514 * @dev: the netdevice
516 * Check boot time settings for the device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found, 1 if they are.
521 int netdev_boot_setup_check(struct net_device *dev)
523 struct netdev_boot_setup *s = dev_boot_setup;
526 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
527 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
528 !strcmp(dev->name, s[i].name)) {
529 dev->irq = s[i].map.irq;
530 dev->base_addr = s[i].map.base_addr;
531 dev->mem_start = s[i].map.mem_start;
532 dev->mem_end = s[i].map.mem_end;
538 EXPORT_SYMBOL(netdev_boot_setup_check);
542 * netdev_boot_base - get address from boot time settings
543 * @prefix: prefix for network device
544 * @unit: id for network device
546 * Check boot time settings for the base address of device.
547 * The found settings are set for the device to be used
548 * later in the device probing.
549 * Returns 0 if no settings found.
551 unsigned long netdev_boot_base(const char *prefix, int unit)
553 const struct netdev_boot_setup *s = dev_boot_setup;
557 sprintf(name, "%s%d", prefix, unit);
560 * If device already registered then return base of 1
561 * to indicate not to probe for this interface
563 if (__dev_get_by_name(&init_net, name))
566 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
567 if (!strcmp(name, s[i].name))
568 return s[i].map.base_addr;
573 * Saves at boot time configured settings for any netdevice.
575 int __init netdev_boot_setup(char *str)
580 str = get_options(str, ARRAY_SIZE(ints), ints);
585 memset(&map, 0, sizeof(map));
589 map.base_addr = ints[2];
591 map.mem_start = ints[3];
593 map.mem_end = ints[4];
595 /* Add new entry to the list */
596 return netdev_boot_setup_add(str, &map);
599 __setup("netdev=", netdev_boot_setup);
601 /*******************************************************************************
603 Device Interface Subroutines
605 *******************************************************************************/
608 * __dev_get_by_name - find a device by its name
609 * @net: the applicable net namespace
610 * @name: name to find
612 * Find an interface by name. Must be called under RTNL semaphore
613 * or @dev_base_lock. If the name is found a pointer to the device
614 * is returned. If the name is not found then %NULL is returned. The
615 * reference counters are not incremented so the caller must be
616 * careful with locks.
619 struct net_device *__dev_get_by_name(struct net *net, const char *name)
621 struct hlist_node *p;
622 struct net_device *dev;
623 struct hlist_head *head = dev_name_hash(net, name);
625 hlist_for_each_entry(dev, p, head, name_hlist)
626 if (!strncmp(dev->name, name, IFNAMSIZ))
631 EXPORT_SYMBOL(__dev_get_by_name);
634 * dev_get_by_name_rcu - find a device by its name
635 * @net: the applicable net namespace
636 * @name: name to find
638 * Find an interface by name.
639 * If the name is found a pointer to the device is returned.
640 * If the name is not found then %NULL is returned.
641 * The reference counters are not incremented so the caller must be
642 * careful with locks. The caller must hold RCU lock.
645 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
647 struct hlist_node *p;
648 struct net_device *dev;
649 struct hlist_head *head = dev_name_hash(net, name);
651 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
652 if (!strncmp(dev->name, name, IFNAMSIZ))
657 EXPORT_SYMBOL(dev_get_by_name_rcu);
660 * dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. This can be called from any
665 * context and does its own locking. The returned handle has
666 * the usage count incremented and the caller must use dev_put() to
667 * release it when it is no longer needed. %NULL is returned if no
668 * matching device is found.
671 struct net_device *dev_get_by_name(struct net *net, const char *name)
673 struct net_device *dev;
676 dev = dev_get_by_name_rcu(net, name);
682 EXPORT_SYMBOL(dev_get_by_name);
685 * __dev_get_by_index - find a device by its ifindex
686 * @net: the applicable net namespace
687 * @ifindex: index of device
689 * Search for an interface by index. Returns %NULL if the device
690 * is not found or a pointer to the device. The device has not
691 * had its reference counter increased so the caller must be careful
692 * about locking. The caller must hold either the RTNL semaphore
696 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
698 struct hlist_node *p;
699 struct net_device *dev;
700 struct hlist_head *head = dev_index_hash(net, ifindex);
702 hlist_for_each_entry(dev, p, head, index_hlist)
703 if (dev->ifindex == ifindex)
708 EXPORT_SYMBOL(__dev_get_by_index);
711 * dev_get_by_index_rcu - find a device by its ifindex
712 * @net: the applicable net namespace
713 * @ifindex: index of device
715 * Search for an interface by index. Returns %NULL if the device
716 * is not found or a pointer to the device. The device has not
717 * had its reference counter increased so the caller must be careful
718 * about locking. The caller must hold RCU lock.
721 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
723 struct hlist_node *p;
724 struct net_device *dev;
725 struct hlist_head *head = dev_index_hash(net, ifindex);
727 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
728 if (dev->ifindex == ifindex)
733 EXPORT_SYMBOL(dev_get_by_index_rcu);
737 * dev_get_by_index - find a device by its ifindex
738 * @net: the applicable net namespace
739 * @ifindex: index of device
741 * Search for an interface by index. Returns NULL if the device
742 * is not found or a pointer to the device. The device returned has
743 * had a reference added and the pointer is safe until the user calls
744 * dev_put to indicate they have finished with it.
747 struct net_device *dev_get_by_index(struct net *net, int ifindex)
749 struct net_device *dev;
752 dev = dev_get_by_index_rcu(net, ifindex);
758 EXPORT_SYMBOL(dev_get_by_index);
761 * dev_getbyhwaddr_rcu - find a device by its hardware address
762 * @net: the applicable net namespace
763 * @type: media type of device
764 * @ha: hardware address
766 * Search for an interface by MAC address. Returns NULL if the device
767 * is not found or a pointer to the device.
768 * The caller must hold RCU or RTNL.
769 * The returned device has not had its ref count increased
770 * and the caller must therefore be careful about locking
774 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
777 struct net_device *dev;
779 for_each_netdev_rcu(net, dev)
780 if (dev->type == type &&
781 !memcmp(dev->dev_addr, ha, dev->addr_len))
786 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
788 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
790 struct net_device *dev;
793 for_each_netdev(net, dev)
794 if (dev->type == type)
799 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
801 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
803 struct net_device *dev, *ret = NULL;
806 for_each_netdev_rcu(net, dev)
807 if (dev->type == type) {
815 EXPORT_SYMBOL(dev_getfirstbyhwtype);
818 * dev_get_by_flags_rcu - find any device with given flags
819 * @net: the applicable net namespace
820 * @if_flags: IFF_* values
821 * @mask: bitmask of bits in if_flags to check
823 * Search for any interface with the given flags. Returns NULL if a device
824 * is not found or a pointer to the device. Must be called inside
825 * rcu_read_lock(), and result refcount is unchanged.
828 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
831 struct net_device *dev, *ret;
834 for_each_netdev_rcu(net, dev) {
835 if (((dev->flags ^ if_flags) & mask) == 0) {
842 EXPORT_SYMBOL(dev_get_by_flags_rcu);
845 * dev_valid_name - check if name is okay for network device
848 * Network device names need to be valid file names to
849 * to allow sysfs to work. We also disallow any kind of
852 int dev_valid_name(const char *name)
856 if (strlen(name) >= IFNAMSIZ)
858 if (!strcmp(name, ".") || !strcmp(name, ".."))
862 if (*name == '/' || isspace(*name))
868 EXPORT_SYMBOL(dev_valid_name);
871 * __dev_alloc_name - allocate a name for a device
872 * @net: network namespace to allocate the device name in
873 * @name: name format string
874 * @buf: scratch buffer and result name string
876 * Passed a format string - eg "lt%d" it will try and find a suitable
877 * id. It scans list of devices to build up a free map, then chooses
878 * the first empty slot. The caller must hold the dev_base or rtnl lock
879 * while allocating the name and adding the device in order to avoid
881 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
882 * Returns the number of the unit assigned or a negative errno code.
885 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
889 const int max_netdevices = 8*PAGE_SIZE;
890 unsigned long *inuse;
891 struct net_device *d;
893 p = strnchr(name, IFNAMSIZ-1, '%');
896 * Verify the string as this thing may have come from
897 * the user. There must be either one "%d" and no other "%"
900 if (p[1] != 'd' || strchr(p + 2, '%'))
903 /* Use one page as a bit array of possible slots */
904 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
908 for_each_netdev(net, d) {
909 if (!sscanf(d->name, name, &i))
911 if (i < 0 || i >= max_netdevices)
914 /* avoid cases where sscanf is not exact inverse of printf */
915 snprintf(buf, IFNAMSIZ, name, i);
916 if (!strncmp(buf, d->name, IFNAMSIZ))
920 i = find_first_zero_bit(inuse, max_netdevices);
921 free_page((unsigned long) inuse);
925 snprintf(buf, IFNAMSIZ, name, i);
926 if (!__dev_get_by_name(net, buf))
929 /* It is possible to run out of possible slots
930 * when the name is long and there isn't enough space left
931 * for the digits, or if all bits are used.
937 * dev_alloc_name - allocate a name for a device
939 * @name: name format string
941 * Passed a format string - eg "lt%d" it will try and find a suitable
942 * id. It scans list of devices to build up a free map, then chooses
943 * the first empty slot. The caller must hold the dev_base or rtnl lock
944 * while allocating the name and adding the device in order to avoid
946 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
947 * Returns the number of the unit assigned or a negative errno code.
950 int dev_alloc_name(struct net_device *dev, const char *name)
956 BUG_ON(!dev_net(dev));
958 ret = __dev_alloc_name(net, name, buf);
960 strlcpy(dev->name, buf, IFNAMSIZ);
963 EXPORT_SYMBOL(dev_alloc_name);
965 static int dev_get_valid_name(struct net_device *dev, const char *name)
969 BUG_ON(!dev_net(dev));
972 if (!dev_valid_name(name))
975 if (strchr(name, '%'))
976 return dev_alloc_name(dev, name);
977 else if (__dev_get_by_name(net, name))
979 else if (dev->name != name)
980 strlcpy(dev->name, name, IFNAMSIZ);
986 * dev_change_name - change name of a device
988 * @newname: name (or format string) must be at least IFNAMSIZ
990 * Change name of a device, can pass format strings "eth%d".
993 int dev_change_name(struct net_device *dev, const char *newname)
995 char oldname[IFNAMSIZ];
1001 BUG_ON(!dev_net(dev));
1004 if (dev->flags & IFF_UP)
1007 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1010 memcpy(oldname, dev->name, IFNAMSIZ);
1012 err = dev_get_valid_name(dev, newname);
1017 ret = device_rename(&dev->dev, dev->name);
1019 memcpy(dev->name, oldname, IFNAMSIZ);
1023 write_lock_bh(&dev_base_lock);
1024 hlist_del_rcu(&dev->name_hlist);
1025 write_unlock_bh(&dev_base_lock);
1029 write_lock_bh(&dev_base_lock);
1030 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1031 write_unlock_bh(&dev_base_lock);
1033 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1034 ret = notifier_to_errno(ret);
1037 /* err >= 0 after dev_alloc_name() or stores the first errno */
1040 memcpy(dev->name, oldname, IFNAMSIZ);
1044 "%s: name change rollback failed: %d.\n",
1053 * dev_set_alias - change ifalias of a device
1055 * @alias: name up to IFALIASZ
1056 * @len: limit of bytes to copy from info
1058 * Set ifalias for a device,
1060 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1064 if (len >= IFALIASZ)
1069 kfree(dev->ifalias);
1070 dev->ifalias = NULL;
1075 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1079 strlcpy(dev->ifalias, alias, len+1);
1085 * netdev_features_change - device changes features
1086 * @dev: device to cause notification
1088 * Called to indicate a device has changed features.
1090 void netdev_features_change(struct net_device *dev)
1092 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1094 EXPORT_SYMBOL(netdev_features_change);
1097 * netdev_state_change - device changes state
1098 * @dev: device to cause notification
1100 * Called to indicate a device has changed state. This function calls
1101 * the notifier chains for netdev_chain and sends a NEWLINK message
1102 * to the routing socket.
1104 void netdev_state_change(struct net_device *dev)
1106 if (dev->flags & IFF_UP) {
1107 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1108 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1111 EXPORT_SYMBOL(netdev_state_change);
1113 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1115 return call_netdevice_notifiers(event, dev);
1117 EXPORT_SYMBOL(netdev_bonding_change);
1120 * dev_load - load a network module
1121 * @net: the applicable net namespace
1122 * @name: name of interface
1124 * If a network interface is not present and the process has suitable
1125 * privileges this function loads the module. If module loading is not
1126 * available in this kernel then it becomes a nop.
1129 void dev_load(struct net *net, const char *name)
1131 struct net_device *dev;
1135 dev = dev_get_by_name_rcu(net, name);
1139 if (no_module && capable(CAP_NET_ADMIN))
1140 no_module = request_module("netdev-%s", name);
1141 if (no_module && capable(CAP_SYS_MODULE)) {
1142 if (!request_module("%s", name))
1143 pr_err("Loading kernel module for a network device "
1144 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1148 EXPORT_SYMBOL(dev_load);
1150 static int __dev_open(struct net_device *dev)
1152 const struct net_device_ops *ops = dev->netdev_ops;
1157 if (!netif_device_present(dev))
1160 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1161 ret = notifier_to_errno(ret);
1165 set_bit(__LINK_STATE_START, &dev->state);
1167 if (ops->ndo_validate_addr)
1168 ret = ops->ndo_validate_addr(dev);
1170 if (!ret && ops->ndo_open)
1171 ret = ops->ndo_open(dev);
1174 clear_bit(__LINK_STATE_START, &dev->state);
1176 dev->flags |= IFF_UP;
1177 net_dmaengine_get();
1178 dev_set_rx_mode(dev);
1186 * dev_open - prepare an interface for use.
1187 * @dev: device to open
1189 * Takes a device from down to up state. The device's private open
1190 * function is invoked and then the multicast lists are loaded. Finally
1191 * the device is moved into the up state and a %NETDEV_UP message is
1192 * sent to the netdev notifier chain.
1194 * Calling this function on an active interface is a nop. On a failure
1195 * a negative errno code is returned.
1197 int dev_open(struct net_device *dev)
1201 if (dev->flags & IFF_UP)
1204 ret = __dev_open(dev);
1208 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1209 call_netdevice_notifiers(NETDEV_UP, dev);
1213 EXPORT_SYMBOL(dev_open);
1215 static int __dev_close_many(struct list_head *head)
1217 struct net_device *dev;
1222 list_for_each_entry(dev, head, unreg_list) {
1223 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1225 clear_bit(__LINK_STATE_START, &dev->state);
1227 /* Synchronize to scheduled poll. We cannot touch poll list, it
1228 * can be even on different cpu. So just clear netif_running().
1230 * dev->stop() will invoke napi_disable() on all of it's
1231 * napi_struct instances on this device.
1233 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1236 dev_deactivate_many(head);
1238 list_for_each_entry(dev, head, unreg_list) {
1239 const struct net_device_ops *ops = dev->netdev_ops;
1242 * Call the device specific close. This cannot fail.
1243 * Only if device is UP
1245 * We allow it to be called even after a DETACH hot-plug
1251 dev->flags &= ~IFF_UP;
1252 net_dmaengine_put();
1258 static int __dev_close(struct net_device *dev)
1263 list_add(&dev->unreg_list, &single);
1264 retval = __dev_close_many(&single);
1269 static int dev_close_many(struct list_head *head)
1271 struct net_device *dev, *tmp;
1272 LIST_HEAD(tmp_list);
1274 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1275 if (!(dev->flags & IFF_UP))
1276 list_move(&dev->unreg_list, &tmp_list);
1278 __dev_close_many(head);
1280 list_for_each_entry(dev, head, unreg_list) {
1281 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1282 call_netdevice_notifiers(NETDEV_DOWN, dev);
1285 /* rollback_registered_many needs the complete original list */
1286 list_splice(&tmp_list, head);
1291 * dev_close - shutdown an interface.
1292 * @dev: device to shutdown
1294 * This function moves an active device into down state. A
1295 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1296 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1299 int dev_close(struct net_device *dev)
1301 if (dev->flags & IFF_UP) {
1304 list_add(&dev->unreg_list, &single);
1305 dev_close_many(&single);
1310 EXPORT_SYMBOL(dev_close);
1314 * dev_disable_lro - disable Large Receive Offload on a device
1317 * Disable Large Receive Offload (LRO) on a net device. Must be
1318 * called under RTNL. This is needed if received packets may be
1319 * forwarded to another interface.
1321 void dev_disable_lro(struct net_device *dev)
1326 * If we're trying to disable lro on a vlan device
1327 * use the underlying physical device instead
1329 if (is_vlan_dev(dev))
1330 dev = vlan_dev_real_dev(dev);
1332 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1333 flags = dev->ethtool_ops->get_flags(dev);
1335 flags = ethtool_op_get_flags(dev);
1337 if (!(flags & ETH_FLAG_LRO))
1340 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1341 if (unlikely(dev->features & NETIF_F_LRO))
1342 netdev_WARN(dev, "failed to disable LRO!\n");
1344 EXPORT_SYMBOL(dev_disable_lro);
1347 static int dev_boot_phase = 1;
1350 * register_netdevice_notifier - register a network notifier block
1353 * Register a notifier to be called when network device events occur.
1354 * The notifier passed is linked into the kernel structures and must
1355 * not be reused until it has been unregistered. A negative errno code
1356 * is returned on a failure.
1358 * When registered all registration and up events are replayed
1359 * to the new notifier to allow device to have a race free
1360 * view of the network device list.
1363 int register_netdevice_notifier(struct notifier_block *nb)
1365 struct net_device *dev;
1366 struct net_device *last;
1371 err = raw_notifier_chain_register(&netdev_chain, nb);
1377 for_each_netdev(net, dev) {
1378 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1379 err = notifier_to_errno(err);
1383 if (!(dev->flags & IFF_UP))
1386 nb->notifier_call(nb, NETDEV_UP, dev);
1397 for_each_netdev(net, dev) {
1401 if (dev->flags & IFF_UP) {
1402 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1403 nb->notifier_call(nb, NETDEV_DOWN, dev);
1405 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1406 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1411 raw_notifier_chain_unregister(&netdev_chain, nb);
1414 EXPORT_SYMBOL(register_netdevice_notifier);
1417 * unregister_netdevice_notifier - unregister a network notifier block
1420 * Unregister a notifier previously registered by
1421 * register_netdevice_notifier(). The notifier is unlinked into the
1422 * kernel structures and may then be reused. A negative errno code
1423 * is returned on a failure.
1425 * After unregistering unregister and down device events are synthesized
1426 * for all devices on the device list to the removed notifier to remove
1427 * the need for special case cleanup code.
1430 int unregister_netdevice_notifier(struct notifier_block *nb)
1432 struct net_device *dev;
1437 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1442 for_each_netdev(net, dev) {
1443 if (dev->flags & IFF_UP) {
1444 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1445 nb->notifier_call(nb, NETDEV_DOWN, dev);
1447 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1448 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1455 EXPORT_SYMBOL(unregister_netdevice_notifier);
1458 * call_netdevice_notifiers - call all network notifier blocks
1459 * @val: value passed unmodified to notifier function
1460 * @dev: net_device pointer passed unmodified to notifier function
1462 * Call all network notifier blocks. Parameters and return value
1463 * are as for raw_notifier_call_chain().
1466 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1469 return raw_notifier_call_chain(&netdev_chain, val, dev);
1471 EXPORT_SYMBOL(call_netdevice_notifiers);
1473 /* When > 0 there are consumers of rx skb time stamps */
1474 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1476 void net_enable_timestamp(void)
1478 atomic_inc(&netstamp_needed);
1480 EXPORT_SYMBOL(net_enable_timestamp);
1482 void net_disable_timestamp(void)
1484 atomic_dec(&netstamp_needed);
1486 EXPORT_SYMBOL(net_disable_timestamp);
1488 static inline void net_timestamp_set(struct sk_buff *skb)
1490 if (atomic_read(&netstamp_needed))
1491 __net_timestamp(skb);
1493 skb->tstamp.tv64 = 0;
1496 static inline void net_timestamp_check(struct sk_buff *skb)
1498 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1499 __net_timestamp(skb);
1502 static int net_hwtstamp_validate(struct ifreq *ifr)
1504 struct hwtstamp_config cfg;
1505 enum hwtstamp_tx_types tx_type;
1506 enum hwtstamp_rx_filters rx_filter;
1507 int tx_type_valid = 0;
1508 int rx_filter_valid = 0;
1510 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1513 if (cfg.flags) /* reserved for future extensions */
1516 tx_type = cfg.tx_type;
1517 rx_filter = cfg.rx_filter;
1520 case HWTSTAMP_TX_OFF:
1521 case HWTSTAMP_TX_ON:
1522 case HWTSTAMP_TX_ONESTEP_SYNC:
1527 switch (rx_filter) {
1528 case HWTSTAMP_FILTER_NONE:
1529 case HWTSTAMP_FILTER_ALL:
1530 case HWTSTAMP_FILTER_SOME:
1531 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1532 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1533 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1534 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1535 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1536 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1537 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1538 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1539 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1540 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1541 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1542 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1543 rx_filter_valid = 1;
1547 if (!tx_type_valid || !rx_filter_valid)
1553 static inline bool is_skb_forwardable(struct net_device *dev,
1554 struct sk_buff *skb)
1558 if (!(dev->flags & IFF_UP))
1561 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1562 if (skb->len <= len)
1565 /* if TSO is enabled, we don't care about the length as the packet
1566 * could be forwarded without being segmented before
1568 if (skb_is_gso(skb))
1575 * dev_forward_skb - loopback an skb to another netif
1577 * @dev: destination network device
1578 * @skb: buffer to forward
1581 * NET_RX_SUCCESS (no congestion)
1582 * NET_RX_DROP (packet was dropped, but freed)
1584 * dev_forward_skb can be used for injecting an skb from the
1585 * start_xmit function of one device into the receive queue
1586 * of another device.
1588 * The receiving device may be in another namespace, so
1589 * we have to clear all information in the skb that could
1590 * impact namespace isolation.
1592 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1594 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1595 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1596 atomic_long_inc(&dev->rx_dropped);
1605 if (unlikely(!is_skb_forwardable(dev, skb))) {
1606 atomic_long_inc(&dev->rx_dropped);
1610 skb_set_dev(skb, dev);
1611 skb->tstamp.tv64 = 0;
1612 skb->pkt_type = PACKET_HOST;
1613 skb->protocol = eth_type_trans(skb, dev);
1614 return netif_rx(skb);
1616 EXPORT_SYMBOL_GPL(dev_forward_skb);
1618 static inline int deliver_skb(struct sk_buff *skb,
1619 struct packet_type *pt_prev,
1620 struct net_device *orig_dev)
1622 atomic_inc(&skb->users);
1623 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1627 * Support routine. Sends outgoing frames to any network
1628 * taps currently in use.
1631 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1633 struct packet_type *ptype;
1634 struct sk_buff *skb2 = NULL;
1635 struct packet_type *pt_prev = NULL;
1638 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1639 /* Never send packets back to the socket
1640 * they originated from - MvS (miquels@drinkel.ow.org)
1642 if ((ptype->dev == dev || !ptype->dev) &&
1643 (ptype->af_packet_priv == NULL ||
1644 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1646 deliver_skb(skb2, pt_prev, skb->dev);
1651 skb2 = skb_clone(skb, GFP_ATOMIC);
1655 net_timestamp_set(skb2);
1657 /* skb->nh should be correctly
1658 set by sender, so that the second statement is
1659 just protection against buggy protocols.
1661 skb_reset_mac_header(skb2);
1663 if (skb_network_header(skb2) < skb2->data ||
1664 skb2->network_header > skb2->tail) {
1665 if (net_ratelimit())
1666 printk(KERN_CRIT "protocol %04x is "
1668 ntohs(skb2->protocol),
1670 skb_reset_network_header(skb2);
1673 skb2->transport_header = skb2->network_header;
1674 skb2->pkt_type = PACKET_OUTGOING;
1679 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1683 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1684 * @dev: Network device
1685 * @txq: number of queues available
1687 * If real_num_tx_queues is changed the tc mappings may no longer be
1688 * valid. To resolve this verify the tc mapping remains valid and if
1689 * not NULL the mapping. With no priorities mapping to this
1690 * offset/count pair it will no longer be used. In the worst case TC0
1691 * is invalid nothing can be done so disable priority mappings. If is
1692 * expected that drivers will fix this mapping if they can before
1693 * calling netif_set_real_num_tx_queues.
1695 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1698 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1700 /* If TC0 is invalidated disable TC mapping */
1701 if (tc->offset + tc->count > txq) {
1702 pr_warning("Number of in use tx queues changed "
1703 "invalidating tc mappings. Priority "
1704 "traffic classification disabled!\n");
1709 /* Invalidated prio to tc mappings set to TC0 */
1710 for (i = 1; i < TC_BITMASK + 1; i++) {
1711 int q = netdev_get_prio_tc_map(dev, i);
1713 tc = &dev->tc_to_txq[q];
1714 if (tc->offset + tc->count > txq) {
1715 pr_warning("Number of in use tx queues "
1716 "changed. Priority %i to tc "
1717 "mapping %i is no longer valid "
1718 "setting map to 0\n",
1720 netdev_set_prio_tc_map(dev, i, 0);
1726 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1727 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1729 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1733 if (txq < 1 || txq > dev->num_tx_queues)
1736 if (dev->reg_state == NETREG_REGISTERED ||
1737 dev->reg_state == NETREG_UNREGISTERING) {
1740 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1746 netif_setup_tc(dev, txq);
1748 if (txq < dev->real_num_tx_queues)
1749 qdisc_reset_all_tx_gt(dev, txq);
1752 dev->real_num_tx_queues = txq;
1755 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1759 * netif_set_real_num_rx_queues - set actual number of RX queues used
1760 * @dev: Network device
1761 * @rxq: Actual number of RX queues
1763 * This must be called either with the rtnl_lock held or before
1764 * registration of the net device. Returns 0 on success, or a
1765 * negative error code. If called before registration, it always
1768 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1772 if (rxq < 1 || rxq > dev->num_rx_queues)
1775 if (dev->reg_state == NETREG_REGISTERED) {
1778 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1784 dev->real_num_rx_queues = rxq;
1787 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1790 static inline void __netif_reschedule(struct Qdisc *q)
1792 struct softnet_data *sd;
1793 unsigned long flags;
1795 local_irq_save(flags);
1796 sd = &__get_cpu_var(softnet_data);
1797 q->next_sched = NULL;
1798 *sd->output_queue_tailp = q;
1799 sd->output_queue_tailp = &q->next_sched;
1800 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1801 local_irq_restore(flags);
1804 void __netif_schedule(struct Qdisc *q)
1806 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1807 __netif_reschedule(q);
1809 EXPORT_SYMBOL(__netif_schedule);
1811 void dev_kfree_skb_irq(struct sk_buff *skb)
1813 if (atomic_dec_and_test(&skb->users)) {
1814 struct softnet_data *sd;
1815 unsigned long flags;
1817 local_irq_save(flags);
1818 sd = &__get_cpu_var(softnet_data);
1819 skb->next = sd->completion_queue;
1820 sd->completion_queue = skb;
1821 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1822 local_irq_restore(flags);
1825 EXPORT_SYMBOL(dev_kfree_skb_irq);
1827 void dev_kfree_skb_any(struct sk_buff *skb)
1829 if (in_irq() || irqs_disabled())
1830 dev_kfree_skb_irq(skb);
1834 EXPORT_SYMBOL(dev_kfree_skb_any);
1838 * netif_device_detach - mark device as removed
1839 * @dev: network device
1841 * Mark device as removed from system and therefore no longer available.
1843 void netif_device_detach(struct net_device *dev)
1845 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1846 netif_running(dev)) {
1847 netif_tx_stop_all_queues(dev);
1850 EXPORT_SYMBOL(netif_device_detach);
1853 * netif_device_attach - mark device as attached
1854 * @dev: network device
1856 * Mark device as attached from system and restart if needed.
1858 void netif_device_attach(struct net_device *dev)
1860 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1861 netif_running(dev)) {
1862 netif_tx_wake_all_queues(dev);
1863 __netdev_watchdog_up(dev);
1866 EXPORT_SYMBOL(netif_device_attach);
1869 * skb_dev_set -- assign a new device to a buffer
1870 * @skb: buffer for the new device
1871 * @dev: network device
1873 * If an skb is owned by a device already, we have to reset
1874 * all data private to the namespace a device belongs to
1875 * before assigning it a new device.
1877 #ifdef CONFIG_NET_NS
1878 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1881 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1884 skb_init_secmark(skb);
1888 skb->ipvs_property = 0;
1889 #ifdef CONFIG_NET_SCHED
1895 EXPORT_SYMBOL(skb_set_dev);
1896 #endif /* CONFIG_NET_NS */
1899 * Invalidate hardware checksum when packet is to be mangled, and
1900 * complete checksum manually on outgoing path.
1902 int skb_checksum_help(struct sk_buff *skb)
1905 int ret = 0, offset;
1907 if (skb->ip_summed == CHECKSUM_COMPLETE)
1908 goto out_set_summed;
1910 if (unlikely(skb_shinfo(skb)->gso_size)) {
1911 /* Let GSO fix up the checksum. */
1912 goto out_set_summed;
1915 offset = skb_checksum_start_offset(skb);
1916 BUG_ON(offset >= skb_headlen(skb));
1917 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1919 offset += skb->csum_offset;
1920 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1922 if (skb_cloned(skb) &&
1923 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1924 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1929 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1931 skb->ip_summed = CHECKSUM_NONE;
1935 EXPORT_SYMBOL(skb_checksum_help);
1938 * skb_gso_segment - Perform segmentation on skb.
1939 * @skb: buffer to segment
1940 * @features: features for the output path (see dev->features)
1942 * This function segments the given skb and returns a list of segments.
1944 * It may return NULL if the skb requires no segmentation. This is
1945 * only possible when GSO is used for verifying header integrity.
1947 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1949 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1950 struct packet_type *ptype;
1951 __be16 type = skb->protocol;
1952 int vlan_depth = ETH_HLEN;
1955 while (type == htons(ETH_P_8021Q)) {
1956 struct vlan_hdr *vh;
1958 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1959 return ERR_PTR(-EINVAL);
1961 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1962 type = vh->h_vlan_encapsulated_proto;
1963 vlan_depth += VLAN_HLEN;
1966 skb_reset_mac_header(skb);
1967 skb->mac_len = skb->network_header - skb->mac_header;
1968 __skb_pull(skb, skb->mac_len);
1970 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1971 struct net_device *dev = skb->dev;
1972 struct ethtool_drvinfo info = {};
1974 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1975 dev->ethtool_ops->get_drvinfo(dev, &info);
1977 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1978 info.driver, dev ? dev->features : 0L,
1979 skb->sk ? skb->sk->sk_route_caps : 0L,
1980 skb->len, skb->data_len, skb->ip_summed);
1982 if (skb_header_cloned(skb) &&
1983 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1984 return ERR_PTR(err);
1988 list_for_each_entry_rcu(ptype,
1989 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1990 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1991 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1992 err = ptype->gso_send_check(skb);
1993 segs = ERR_PTR(err);
1994 if (err || skb_gso_ok(skb, features))
1996 __skb_push(skb, (skb->data -
1997 skb_network_header(skb)));
1999 segs = ptype->gso_segment(skb, features);
2005 __skb_push(skb, skb->data - skb_mac_header(skb));
2009 EXPORT_SYMBOL(skb_gso_segment);
2011 /* Take action when hardware reception checksum errors are detected. */
2013 void netdev_rx_csum_fault(struct net_device *dev)
2015 if (net_ratelimit()) {
2016 printk(KERN_ERR "%s: hw csum failure.\n",
2017 dev ? dev->name : "<unknown>");
2021 EXPORT_SYMBOL(netdev_rx_csum_fault);
2024 /* Actually, we should eliminate this check as soon as we know, that:
2025 * 1. IOMMU is present and allows to map all the memory.
2026 * 2. No high memory really exists on this machine.
2029 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2031 #ifdef CONFIG_HIGHMEM
2033 if (!(dev->features & NETIF_F_HIGHDMA)) {
2034 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2035 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2036 if (PageHighMem(skb_frag_page(frag)))
2041 if (PCI_DMA_BUS_IS_PHYS) {
2042 struct device *pdev = dev->dev.parent;
2046 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2047 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2048 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2049 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2058 void (*destructor)(struct sk_buff *skb);
2061 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2063 static void dev_gso_skb_destructor(struct sk_buff *skb)
2065 struct dev_gso_cb *cb;
2068 struct sk_buff *nskb = skb->next;
2070 skb->next = nskb->next;
2073 } while (skb->next);
2075 cb = DEV_GSO_CB(skb);
2077 cb->destructor(skb);
2081 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2082 * @skb: buffer to segment
2083 * @features: device features as applicable to this skb
2085 * This function segments the given skb and stores the list of segments
2088 static int dev_gso_segment(struct sk_buff *skb, int features)
2090 struct sk_buff *segs;
2092 segs = skb_gso_segment(skb, features);
2094 /* Verifying header integrity only. */
2099 return PTR_ERR(segs);
2102 DEV_GSO_CB(skb)->destructor = skb->destructor;
2103 skb->destructor = dev_gso_skb_destructor;
2109 * Try to orphan skb early, right before transmission by the device.
2110 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2111 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2113 static inline void skb_orphan_try(struct sk_buff *skb)
2115 struct sock *sk = skb->sk;
2117 if (sk && !skb_shinfo(skb)->tx_flags) {
2118 /* skb_tx_hash() wont be able to get sk.
2119 * We copy sk_hash into skb->rxhash
2122 skb->rxhash = sk->sk_hash;
2127 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2129 return ((features & NETIF_F_GEN_CSUM) ||
2130 ((features & NETIF_F_V4_CSUM) &&
2131 protocol == htons(ETH_P_IP)) ||
2132 ((features & NETIF_F_V6_CSUM) &&
2133 protocol == htons(ETH_P_IPV6)) ||
2134 ((features & NETIF_F_FCOE_CRC) &&
2135 protocol == htons(ETH_P_FCOE)));
2138 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2140 if (!can_checksum_protocol(features, protocol)) {
2141 features &= ~NETIF_F_ALL_CSUM;
2142 features &= ~NETIF_F_SG;
2143 } else if (illegal_highdma(skb->dev, skb)) {
2144 features &= ~NETIF_F_SG;
2150 u32 netif_skb_features(struct sk_buff *skb)
2152 __be16 protocol = skb->protocol;
2153 u32 features = skb->dev->features;
2155 if (protocol == htons(ETH_P_8021Q)) {
2156 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2157 protocol = veh->h_vlan_encapsulated_proto;
2158 } else if (!vlan_tx_tag_present(skb)) {
2159 return harmonize_features(skb, protocol, features);
2162 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2164 if (protocol != htons(ETH_P_8021Q)) {
2165 return harmonize_features(skb, protocol, features);
2167 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2168 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2169 return harmonize_features(skb, protocol, features);
2172 EXPORT_SYMBOL(netif_skb_features);
2175 * Returns true if either:
2176 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2177 * 2. skb is fragmented and the device does not support SG, or if
2178 * at least one of fragments is in highmem and device does not
2179 * support DMA from it.
2181 static inline int skb_needs_linearize(struct sk_buff *skb,
2184 return skb_is_nonlinear(skb) &&
2185 ((skb_has_frag_list(skb) &&
2186 !(features & NETIF_F_FRAGLIST)) ||
2187 (skb_shinfo(skb)->nr_frags &&
2188 !(features & NETIF_F_SG)));
2191 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2192 struct netdev_queue *txq)
2194 const struct net_device_ops *ops = dev->netdev_ops;
2195 int rc = NETDEV_TX_OK;
2196 unsigned int skb_len;
2198 if (likely(!skb->next)) {
2202 * If device doesn't need skb->dst, release it right now while
2203 * its hot in this cpu cache
2205 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2208 if (!list_empty(&ptype_all))
2209 dev_queue_xmit_nit(skb, dev);
2211 skb_orphan_try(skb);
2213 features = netif_skb_features(skb);
2215 if (vlan_tx_tag_present(skb) &&
2216 !(features & NETIF_F_HW_VLAN_TX)) {
2217 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2224 if (netif_needs_gso(skb, features)) {
2225 if (unlikely(dev_gso_segment(skb, features)))
2230 if (skb_needs_linearize(skb, features) &&
2231 __skb_linearize(skb))
2234 /* If packet is not checksummed and device does not
2235 * support checksumming for this protocol, complete
2236 * checksumming here.
2238 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2239 skb_set_transport_header(skb,
2240 skb_checksum_start_offset(skb));
2241 if (!(features & NETIF_F_ALL_CSUM) &&
2242 skb_checksum_help(skb))
2248 rc = ops->ndo_start_xmit(skb, dev);
2249 trace_net_dev_xmit(skb, rc, dev, skb_len);
2250 if (rc == NETDEV_TX_OK)
2251 txq_trans_update(txq);
2257 struct sk_buff *nskb = skb->next;
2259 skb->next = nskb->next;
2263 * If device doesn't need nskb->dst, release it right now while
2264 * its hot in this cpu cache
2266 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2269 skb_len = nskb->len;
2270 rc = ops->ndo_start_xmit(nskb, dev);
2271 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2272 if (unlikely(rc != NETDEV_TX_OK)) {
2273 if (rc & ~NETDEV_TX_MASK)
2274 goto out_kfree_gso_skb;
2275 nskb->next = skb->next;
2279 txq_trans_update(txq);
2280 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2281 return NETDEV_TX_BUSY;
2282 } while (skb->next);
2285 if (likely(skb->next == NULL))
2286 skb->destructor = DEV_GSO_CB(skb)->destructor;
2293 static u32 hashrnd __read_mostly;
2296 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2297 * to be used as a distribution range.
2299 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2300 unsigned int num_tx_queues)
2304 u16 qcount = num_tx_queues;
2306 if (skb_rx_queue_recorded(skb)) {
2307 hash = skb_get_rx_queue(skb);
2308 while (unlikely(hash >= num_tx_queues))
2309 hash -= num_tx_queues;
2314 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2315 qoffset = dev->tc_to_txq[tc].offset;
2316 qcount = dev->tc_to_txq[tc].count;
2319 if (skb->sk && skb->sk->sk_hash)
2320 hash = skb->sk->sk_hash;
2322 hash = (__force u16) skb->protocol ^ skb->rxhash;
2323 hash = jhash_1word(hash, hashrnd);
2325 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2327 EXPORT_SYMBOL(__skb_tx_hash);
2329 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2331 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2332 if (net_ratelimit()) {
2333 pr_warning("%s selects TX queue %d, but "
2334 "real number of TX queues is %d\n",
2335 dev->name, queue_index, dev->real_num_tx_queues);
2342 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2345 struct xps_dev_maps *dev_maps;
2346 struct xps_map *map;
2347 int queue_index = -1;
2350 dev_maps = rcu_dereference(dev->xps_maps);
2352 map = rcu_dereference(
2353 dev_maps->cpu_map[raw_smp_processor_id()]);
2356 queue_index = map->queues[0];
2359 if (skb->sk && skb->sk->sk_hash)
2360 hash = skb->sk->sk_hash;
2362 hash = (__force u16) skb->protocol ^
2364 hash = jhash_1word(hash, hashrnd);
2365 queue_index = map->queues[
2366 ((u64)hash * map->len) >> 32];
2368 if (unlikely(queue_index >= dev->real_num_tx_queues))
2380 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2381 struct sk_buff *skb)
2384 const struct net_device_ops *ops = dev->netdev_ops;
2386 if (dev->real_num_tx_queues == 1)
2388 else if (ops->ndo_select_queue) {
2389 queue_index = ops->ndo_select_queue(dev, skb);
2390 queue_index = dev_cap_txqueue(dev, queue_index);
2392 struct sock *sk = skb->sk;
2393 queue_index = sk_tx_queue_get(sk);
2395 if (queue_index < 0 || skb->ooo_okay ||
2396 queue_index >= dev->real_num_tx_queues) {
2397 int old_index = queue_index;
2399 queue_index = get_xps_queue(dev, skb);
2400 if (queue_index < 0)
2401 queue_index = skb_tx_hash(dev, skb);
2403 if (queue_index != old_index && sk) {
2404 struct dst_entry *dst =
2405 rcu_dereference_check(sk->sk_dst_cache, 1);
2407 if (dst && skb_dst(skb) == dst)
2408 sk_tx_queue_set(sk, queue_index);
2413 skb_set_queue_mapping(skb, queue_index);
2414 return netdev_get_tx_queue(dev, queue_index);
2417 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2418 struct net_device *dev,
2419 struct netdev_queue *txq)
2421 spinlock_t *root_lock = qdisc_lock(q);
2425 qdisc_skb_cb(skb)->pkt_len = skb->len;
2426 qdisc_calculate_pkt_len(skb, q);
2428 * Heuristic to force contended enqueues to serialize on a
2429 * separate lock before trying to get qdisc main lock.
2430 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2431 * and dequeue packets faster.
2433 contended = qdisc_is_running(q);
2434 if (unlikely(contended))
2435 spin_lock(&q->busylock);
2437 spin_lock(root_lock);
2438 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2441 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2442 qdisc_run_begin(q)) {
2444 * This is a work-conserving queue; there are no old skbs
2445 * waiting to be sent out; and the qdisc is not running -
2446 * xmit the skb directly.
2448 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2451 qdisc_bstats_update(q, skb);
2453 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2454 if (unlikely(contended)) {
2455 spin_unlock(&q->busylock);
2462 rc = NET_XMIT_SUCCESS;
2465 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2466 if (qdisc_run_begin(q)) {
2467 if (unlikely(contended)) {
2468 spin_unlock(&q->busylock);
2474 spin_unlock(root_lock);
2475 if (unlikely(contended))
2476 spin_unlock(&q->busylock);
2480 static DEFINE_PER_CPU(int, xmit_recursion);
2481 #define RECURSION_LIMIT 10
2484 * dev_queue_xmit - transmit a buffer
2485 * @skb: buffer to transmit
2487 * Queue a buffer for transmission to a network device. The caller must
2488 * have set the device and priority and built the buffer before calling
2489 * this function. The function can be called from an interrupt.
2491 * A negative errno code is returned on a failure. A success does not
2492 * guarantee the frame will be transmitted as it may be dropped due
2493 * to congestion or traffic shaping.
2495 * -----------------------------------------------------------------------------------
2496 * I notice this method can also return errors from the queue disciplines,
2497 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2500 * Regardless of the return value, the skb is consumed, so it is currently
2501 * difficult to retry a send to this method. (You can bump the ref count
2502 * before sending to hold a reference for retry if you are careful.)
2504 * When calling this method, interrupts MUST be enabled. This is because
2505 * the BH enable code must have IRQs enabled so that it will not deadlock.
2508 int dev_queue_xmit(struct sk_buff *skb)
2510 struct net_device *dev = skb->dev;
2511 struct netdev_queue *txq;
2515 /* Disable soft irqs for various locks below. Also
2516 * stops preemption for RCU.
2520 txq = dev_pick_tx(dev, skb);
2521 q = rcu_dereference_bh(txq->qdisc);
2523 #ifdef CONFIG_NET_CLS_ACT
2524 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2526 trace_net_dev_queue(skb);
2528 rc = __dev_xmit_skb(skb, q, dev, txq);
2532 /* The device has no queue. Common case for software devices:
2533 loopback, all the sorts of tunnels...
2535 Really, it is unlikely that netif_tx_lock protection is necessary
2536 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2538 However, it is possible, that they rely on protection
2541 Check this and shot the lock. It is not prone from deadlocks.
2542 Either shot noqueue qdisc, it is even simpler 8)
2544 if (dev->flags & IFF_UP) {
2545 int cpu = smp_processor_id(); /* ok because BHs are off */
2547 if (txq->xmit_lock_owner != cpu) {
2549 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2550 goto recursion_alert;
2552 HARD_TX_LOCK(dev, txq, cpu);
2554 if (!netif_tx_queue_stopped(txq)) {
2555 __this_cpu_inc(xmit_recursion);
2556 rc = dev_hard_start_xmit(skb, dev, txq);
2557 __this_cpu_dec(xmit_recursion);
2558 if (dev_xmit_complete(rc)) {
2559 HARD_TX_UNLOCK(dev, txq);
2563 HARD_TX_UNLOCK(dev, txq);
2564 if (net_ratelimit())
2565 printk(KERN_CRIT "Virtual device %s asks to "
2566 "queue packet!\n", dev->name);
2568 /* Recursion is detected! It is possible,
2572 if (net_ratelimit())
2573 printk(KERN_CRIT "Dead loop on virtual device "
2574 "%s, fix it urgently!\n", dev->name);
2579 rcu_read_unlock_bh();
2584 rcu_read_unlock_bh();
2587 EXPORT_SYMBOL(dev_queue_xmit);
2590 /*=======================================================================
2592 =======================================================================*/
2594 int netdev_max_backlog __read_mostly = 1000;
2595 int netdev_tstamp_prequeue __read_mostly = 1;
2596 int netdev_budget __read_mostly = 300;
2597 int weight_p __read_mostly = 64; /* old backlog weight */
2599 /* Called with irq disabled */
2600 static inline void ____napi_schedule(struct softnet_data *sd,
2601 struct napi_struct *napi)
2603 list_add_tail(&napi->poll_list, &sd->poll_list);
2604 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2608 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2609 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2610 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2611 * if hash is a canonical 4-tuple hash over transport ports.
2613 void __skb_get_rxhash(struct sk_buff *skb)
2615 int nhoff, hash = 0, poff;
2616 const struct ipv6hdr *ip6;
2617 const struct iphdr *ip;
2618 const struct vlan_hdr *vlan;
2627 nhoff = skb_network_offset(skb);
2628 proto = skb->protocol;
2632 case __constant_htons(ETH_P_IP):
2634 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2637 ip = (const struct iphdr *) (skb->data + nhoff);
2638 if (ip_is_fragment(ip))
2641 ip_proto = ip->protocol;
2642 addr1 = (__force u32) ip->saddr;
2643 addr2 = (__force u32) ip->daddr;
2644 nhoff += ip->ihl * 4;
2646 case __constant_htons(ETH_P_IPV6):
2648 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2651 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2652 ip_proto = ip6->nexthdr;
2653 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2654 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2657 case __constant_htons(ETH_P_8021Q):
2658 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2660 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2661 proto = vlan->h_vlan_encapsulated_proto;
2662 nhoff += sizeof(*vlan);
2664 case __constant_htons(ETH_P_PPP_SES):
2665 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2667 proto = *((__be16 *) (skb->data + nhoff +
2668 sizeof(struct pppoe_hdr)));
2669 nhoff += PPPOE_SES_HLEN;
2671 case __constant_htons(PPP_IP):
2673 case __constant_htons(PPP_IPV6):
2684 if (pskb_may_pull(skb, nhoff + 16)) {
2685 u8 *h = skb->data + nhoff;
2686 __be16 flags = *(__be16 *)h;
2689 * Only look inside GRE if version zero and no
2692 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2693 proto = *(__be16 *)(h + 2);
2695 if (flags & GRE_CSUM)
2697 if (flags & GRE_KEY)
2699 if (flags & GRE_SEQ)
2712 poff = proto_ports_offset(ip_proto);
2715 if (pskb_may_pull(skb, nhoff + 4)) {
2716 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2717 if (ports.v16[1] < ports.v16[0])
2718 swap(ports.v16[0], ports.v16[1]);
2723 /* get a consistent hash (same value on both flow directions) */
2727 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2734 EXPORT_SYMBOL(__skb_get_rxhash);
2738 /* One global table that all flow-based protocols share. */
2739 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2740 EXPORT_SYMBOL(rps_sock_flow_table);
2742 static struct rps_dev_flow *
2743 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2744 struct rps_dev_flow *rflow, u16 next_cpu)
2746 if (next_cpu != RPS_NO_CPU) {
2747 #ifdef CONFIG_RFS_ACCEL
2748 struct netdev_rx_queue *rxqueue;
2749 struct rps_dev_flow_table *flow_table;
2750 struct rps_dev_flow *old_rflow;
2755 /* Should we steer this flow to a different hardware queue? */
2756 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2757 !(dev->features & NETIF_F_NTUPLE))
2759 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2760 if (rxq_index == skb_get_rx_queue(skb))
2763 rxqueue = dev->_rx + rxq_index;
2764 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2767 flow_id = skb->rxhash & flow_table->mask;
2768 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2769 rxq_index, flow_id);
2773 rflow = &flow_table->flows[flow_id];
2775 if (old_rflow->filter == rflow->filter)
2776 old_rflow->filter = RPS_NO_FILTER;
2780 per_cpu(softnet_data, next_cpu).input_queue_head;
2783 rflow->cpu = next_cpu;
2788 * get_rps_cpu is called from netif_receive_skb and returns the target
2789 * CPU from the RPS map of the receiving queue for a given skb.
2790 * rcu_read_lock must be held on entry.
2792 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2793 struct rps_dev_flow **rflowp)
2795 struct netdev_rx_queue *rxqueue;
2796 struct rps_map *map;
2797 struct rps_dev_flow_table *flow_table;
2798 struct rps_sock_flow_table *sock_flow_table;
2802 if (skb_rx_queue_recorded(skb)) {
2803 u16 index = skb_get_rx_queue(skb);
2804 if (unlikely(index >= dev->real_num_rx_queues)) {
2805 WARN_ONCE(dev->real_num_rx_queues > 1,
2806 "%s received packet on queue %u, but number "
2807 "of RX queues is %u\n",
2808 dev->name, index, dev->real_num_rx_queues);
2811 rxqueue = dev->_rx + index;
2815 map = rcu_dereference(rxqueue->rps_map);
2817 if (map->len == 1 &&
2818 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2819 tcpu = map->cpus[0];
2820 if (cpu_online(tcpu))
2824 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2828 skb_reset_network_header(skb);
2829 if (!skb_get_rxhash(skb))
2832 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2833 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2834 if (flow_table && sock_flow_table) {
2836 struct rps_dev_flow *rflow;
2838 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2841 next_cpu = sock_flow_table->ents[skb->rxhash &
2842 sock_flow_table->mask];
2845 * If the desired CPU (where last recvmsg was done) is
2846 * different from current CPU (one in the rx-queue flow
2847 * table entry), switch if one of the following holds:
2848 * - Current CPU is unset (equal to RPS_NO_CPU).
2849 * - Current CPU is offline.
2850 * - The current CPU's queue tail has advanced beyond the
2851 * last packet that was enqueued using this table entry.
2852 * This guarantees that all previous packets for the flow
2853 * have been dequeued, thus preserving in order delivery.
2855 if (unlikely(tcpu != next_cpu) &&
2856 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2857 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2858 rflow->last_qtail)) >= 0))
2859 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2861 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2869 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2871 if (cpu_online(tcpu)) {
2881 #ifdef CONFIG_RFS_ACCEL
2884 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2885 * @dev: Device on which the filter was set
2886 * @rxq_index: RX queue index
2887 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2888 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2890 * Drivers that implement ndo_rx_flow_steer() should periodically call
2891 * this function for each installed filter and remove the filters for
2892 * which it returns %true.
2894 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2895 u32 flow_id, u16 filter_id)
2897 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2898 struct rps_dev_flow_table *flow_table;
2899 struct rps_dev_flow *rflow;
2904 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2905 if (flow_table && flow_id <= flow_table->mask) {
2906 rflow = &flow_table->flows[flow_id];
2907 cpu = ACCESS_ONCE(rflow->cpu);
2908 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2909 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2910 rflow->last_qtail) <
2911 (int)(10 * flow_table->mask)))
2917 EXPORT_SYMBOL(rps_may_expire_flow);
2919 #endif /* CONFIG_RFS_ACCEL */
2921 /* Called from hardirq (IPI) context */
2922 static void rps_trigger_softirq(void *data)
2924 struct softnet_data *sd = data;
2926 ____napi_schedule(sd, &sd->backlog);
2930 #endif /* CONFIG_RPS */
2933 * Check if this softnet_data structure is another cpu one
2934 * If yes, queue it to our IPI list and return 1
2937 static int rps_ipi_queued(struct softnet_data *sd)
2940 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2943 sd->rps_ipi_next = mysd->rps_ipi_list;
2944 mysd->rps_ipi_list = sd;
2946 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2949 #endif /* CONFIG_RPS */
2954 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2955 * queue (may be a remote CPU queue).
2957 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2958 unsigned int *qtail)
2960 struct softnet_data *sd;
2961 unsigned long flags;
2963 sd = &per_cpu(softnet_data, cpu);
2965 local_irq_save(flags);
2968 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2969 if (skb_queue_len(&sd->input_pkt_queue)) {
2971 __skb_queue_tail(&sd->input_pkt_queue, skb);
2972 input_queue_tail_incr_save(sd, qtail);
2974 local_irq_restore(flags);
2975 return NET_RX_SUCCESS;
2978 /* Schedule NAPI for backlog device
2979 * We can use non atomic operation since we own the queue lock
2981 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2982 if (!rps_ipi_queued(sd))
2983 ____napi_schedule(sd, &sd->backlog);
2991 local_irq_restore(flags);
2993 atomic_long_inc(&skb->dev->rx_dropped);
2999 * netif_rx - post buffer to the network code
3000 * @skb: buffer to post
3002 * This function receives a packet from a device driver and queues it for
3003 * the upper (protocol) levels to process. It always succeeds. The buffer
3004 * may be dropped during processing for congestion control or by the
3008 * NET_RX_SUCCESS (no congestion)
3009 * NET_RX_DROP (packet was dropped)
3013 int netif_rx(struct sk_buff *skb)
3017 /* if netpoll wants it, pretend we never saw it */
3018 if (netpoll_rx(skb))
3021 if (netdev_tstamp_prequeue)
3022 net_timestamp_check(skb);
3024 trace_netif_rx(skb);
3027 struct rps_dev_flow voidflow, *rflow = &voidflow;
3033 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3035 cpu = smp_processor_id();
3037 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3045 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3051 EXPORT_SYMBOL(netif_rx);
3053 int netif_rx_ni(struct sk_buff *skb)
3058 err = netif_rx(skb);
3059 if (local_softirq_pending())
3065 EXPORT_SYMBOL(netif_rx_ni);
3067 static void net_tx_action(struct softirq_action *h)
3069 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3071 if (sd->completion_queue) {
3072 struct sk_buff *clist;
3074 local_irq_disable();
3075 clist = sd->completion_queue;
3076 sd->completion_queue = NULL;
3080 struct sk_buff *skb = clist;
3081 clist = clist->next;
3083 WARN_ON(atomic_read(&skb->users));
3084 trace_kfree_skb(skb, net_tx_action);
3089 if (sd->output_queue) {
3092 local_irq_disable();
3093 head = sd->output_queue;
3094 sd->output_queue = NULL;
3095 sd->output_queue_tailp = &sd->output_queue;
3099 struct Qdisc *q = head;
3100 spinlock_t *root_lock;
3102 head = head->next_sched;
3104 root_lock = qdisc_lock(q);
3105 if (spin_trylock(root_lock)) {
3106 smp_mb__before_clear_bit();
3107 clear_bit(__QDISC_STATE_SCHED,
3110 spin_unlock(root_lock);
3112 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3114 __netif_reschedule(q);
3116 smp_mb__before_clear_bit();
3117 clear_bit(__QDISC_STATE_SCHED,
3125 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3126 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3127 /* This hook is defined here for ATM LANE */
3128 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3129 unsigned char *addr) __read_mostly;
3130 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3133 #ifdef CONFIG_NET_CLS_ACT
3134 /* TODO: Maybe we should just force sch_ingress to be compiled in
3135 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3136 * a compare and 2 stores extra right now if we dont have it on
3137 * but have CONFIG_NET_CLS_ACT
3138 * NOTE: This doesn't stop any functionality; if you dont have
3139 * the ingress scheduler, you just can't add policies on ingress.
3142 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3144 struct net_device *dev = skb->dev;
3145 u32 ttl = G_TC_RTTL(skb->tc_verd);
3146 int result = TC_ACT_OK;
3149 if (unlikely(MAX_RED_LOOP < ttl++)) {
3150 if (net_ratelimit())
3151 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3152 skb->skb_iif, dev->ifindex);
3156 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3157 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3160 if (q != &noop_qdisc) {
3161 spin_lock(qdisc_lock(q));
3162 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3163 result = qdisc_enqueue_root(skb, q);
3164 spin_unlock(qdisc_lock(q));
3170 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3171 struct packet_type **pt_prev,
3172 int *ret, struct net_device *orig_dev)
3174 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3176 if (!rxq || rxq->qdisc == &noop_qdisc)
3180 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3184 switch (ing_filter(skb, rxq)) {
3198 * netdev_rx_handler_register - register receive handler
3199 * @dev: device to register a handler for
3200 * @rx_handler: receive handler to register
3201 * @rx_handler_data: data pointer that is used by rx handler
3203 * Register a receive hander for a device. This handler will then be
3204 * called from __netif_receive_skb. A negative errno code is returned
3207 * The caller must hold the rtnl_mutex.
3209 * For a general description of rx_handler, see enum rx_handler_result.
3211 int netdev_rx_handler_register(struct net_device *dev,
3212 rx_handler_func_t *rx_handler,
3213 void *rx_handler_data)
3217 if (dev->rx_handler)
3220 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3221 rcu_assign_pointer(dev->rx_handler, rx_handler);
3225 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3228 * netdev_rx_handler_unregister - unregister receive handler
3229 * @dev: device to unregister a handler from
3231 * Unregister a receive hander from a device.
3233 * The caller must hold the rtnl_mutex.
3235 void netdev_rx_handler_unregister(struct net_device *dev)
3239 RCU_INIT_POINTER(dev->rx_handler, NULL);
3240 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3242 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3244 static int __netif_receive_skb(struct sk_buff *skb)
3246 struct packet_type *ptype, *pt_prev;
3247 rx_handler_func_t *rx_handler;
3248 struct net_device *orig_dev;
3249 struct net_device *null_or_dev;
3250 bool deliver_exact = false;
3251 int ret = NET_RX_DROP;
3254 if (!netdev_tstamp_prequeue)
3255 net_timestamp_check(skb);
3257 trace_netif_receive_skb(skb);
3259 /* if we've gotten here through NAPI, check netpoll */
3260 if (netpoll_receive_skb(skb))
3264 skb->skb_iif = skb->dev->ifindex;
3265 orig_dev = skb->dev;
3267 skb_reset_network_header(skb);
3268 skb_reset_transport_header(skb);
3269 skb_reset_mac_len(skb);
3277 __this_cpu_inc(softnet_data.processed);
3279 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3280 skb = vlan_untag(skb);
3285 #ifdef CONFIG_NET_CLS_ACT
3286 if (skb->tc_verd & TC_NCLS) {
3287 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3292 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3293 if (!ptype->dev || ptype->dev == skb->dev) {
3295 ret = deliver_skb(skb, pt_prev, orig_dev);
3300 #ifdef CONFIG_NET_CLS_ACT
3301 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3307 rx_handler = rcu_dereference(skb->dev->rx_handler);
3308 if (vlan_tx_tag_present(skb)) {
3310 ret = deliver_skb(skb, pt_prev, orig_dev);
3313 if (vlan_do_receive(&skb, !rx_handler))
3315 else if (unlikely(!skb))
3321 ret = deliver_skb(skb, pt_prev, orig_dev);
3324 switch (rx_handler(&skb)) {
3325 case RX_HANDLER_CONSUMED:
3327 case RX_HANDLER_ANOTHER:
3329 case RX_HANDLER_EXACT:
3330 deliver_exact = true;
3331 case RX_HANDLER_PASS:
3338 /* deliver only exact match when indicated */
3339 null_or_dev = deliver_exact ? skb->dev : NULL;
3341 type = skb->protocol;
3342 list_for_each_entry_rcu(ptype,
3343 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3344 if (ptype->type == type &&
3345 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3346 ptype->dev == orig_dev)) {
3348 ret = deliver_skb(skb, pt_prev, orig_dev);
3354 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3356 atomic_long_inc(&skb->dev->rx_dropped);
3358 /* Jamal, now you will not able to escape explaining
3359 * me how you were going to use this. :-)
3370 * netif_receive_skb - process receive buffer from network
3371 * @skb: buffer to process
3373 * netif_receive_skb() is the main receive data processing function.
3374 * It always succeeds. The buffer may be dropped during processing
3375 * for congestion control or by the protocol layers.
3377 * This function may only be called from softirq context and interrupts
3378 * should be enabled.
3380 * Return values (usually ignored):
3381 * NET_RX_SUCCESS: no congestion
3382 * NET_RX_DROP: packet was dropped
3384 int netif_receive_skb(struct sk_buff *skb)
3386 if (netdev_tstamp_prequeue)
3387 net_timestamp_check(skb);
3389 if (skb_defer_rx_timestamp(skb))
3390 return NET_RX_SUCCESS;
3394 struct rps_dev_flow voidflow, *rflow = &voidflow;
3399 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3402 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3406 ret = __netif_receive_skb(skb);
3412 return __netif_receive_skb(skb);
3415 EXPORT_SYMBOL(netif_receive_skb);
3417 /* Network device is going away, flush any packets still pending
3418 * Called with irqs disabled.
3420 static void flush_backlog(void *arg)
3422 struct net_device *dev = arg;
3423 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3424 struct sk_buff *skb, *tmp;
3427 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3428 if (skb->dev == dev) {
3429 __skb_unlink(skb, &sd->input_pkt_queue);
3431 input_queue_head_incr(sd);
3436 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3437 if (skb->dev == dev) {
3438 __skb_unlink(skb, &sd->process_queue);
3440 input_queue_head_incr(sd);
3445 static int napi_gro_complete(struct sk_buff *skb)
3447 struct packet_type *ptype;
3448 __be16 type = skb->protocol;
3449 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3452 if (NAPI_GRO_CB(skb)->count == 1) {
3453 skb_shinfo(skb)->gso_size = 0;
3458 list_for_each_entry_rcu(ptype, head, list) {
3459 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3462 err = ptype->gro_complete(skb);
3468 WARN_ON(&ptype->list == head);
3470 return NET_RX_SUCCESS;
3474 return netif_receive_skb(skb);
3477 inline void napi_gro_flush(struct napi_struct *napi)
3479 struct sk_buff *skb, *next;
3481 for (skb = napi->gro_list; skb; skb = next) {
3484 napi_gro_complete(skb);
3487 napi->gro_count = 0;
3488 napi->gro_list = NULL;
3490 EXPORT_SYMBOL(napi_gro_flush);
3492 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3494 struct sk_buff **pp = NULL;
3495 struct packet_type *ptype;
3496 __be16 type = skb->protocol;
3497 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3500 enum gro_result ret;
3502 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3505 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3509 list_for_each_entry_rcu(ptype, head, list) {
3510 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3513 skb_set_network_header(skb, skb_gro_offset(skb));
3514 mac_len = skb->network_header - skb->mac_header;
3515 skb->mac_len = mac_len;
3516 NAPI_GRO_CB(skb)->same_flow = 0;
3517 NAPI_GRO_CB(skb)->flush = 0;
3518 NAPI_GRO_CB(skb)->free = 0;
3520 pp = ptype->gro_receive(&napi->gro_list, skb);
3525 if (&ptype->list == head)
3528 same_flow = NAPI_GRO_CB(skb)->same_flow;
3529 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3532 struct sk_buff *nskb = *pp;
3536 napi_gro_complete(nskb);
3543 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3547 NAPI_GRO_CB(skb)->count = 1;
3548 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3549 skb->next = napi->gro_list;
3550 napi->gro_list = skb;
3554 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3555 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3557 BUG_ON(skb->end - skb->tail < grow);
3559 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3562 skb->data_len -= grow;
3564 skb_shinfo(skb)->frags[0].page_offset += grow;
3565 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3567 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3568 skb_frag_unref(skb, 0);
3569 memmove(skb_shinfo(skb)->frags,
3570 skb_shinfo(skb)->frags + 1,
3571 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3582 EXPORT_SYMBOL(dev_gro_receive);
3584 static inline gro_result_t
3585 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3588 unsigned int maclen = skb->dev->hard_header_len;
3590 for (p = napi->gro_list; p; p = p->next) {
3591 unsigned long diffs;
3593 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3594 diffs |= p->vlan_tci ^ skb->vlan_tci;
3595 if (maclen == ETH_HLEN)
3596 diffs |= compare_ether_header(skb_mac_header(p),
3597 skb_gro_mac_header(skb));
3599 diffs = memcmp(skb_mac_header(p),
3600 skb_gro_mac_header(skb),
3602 NAPI_GRO_CB(p)->same_flow = !diffs;
3603 NAPI_GRO_CB(p)->flush = 0;
3606 return dev_gro_receive(napi, skb);
3609 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3613 if (netif_receive_skb(skb))
3618 case GRO_MERGED_FREE:
3629 EXPORT_SYMBOL(napi_skb_finish);
3631 void skb_gro_reset_offset(struct sk_buff *skb)
3633 NAPI_GRO_CB(skb)->data_offset = 0;
3634 NAPI_GRO_CB(skb)->frag0 = NULL;
3635 NAPI_GRO_CB(skb)->frag0_len = 0;
3637 if (skb->mac_header == skb->tail &&
3638 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3639 NAPI_GRO_CB(skb)->frag0 =
3640 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3641 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3644 EXPORT_SYMBOL(skb_gro_reset_offset);
3646 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3648 skb_gro_reset_offset(skb);
3650 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3652 EXPORT_SYMBOL(napi_gro_receive);
3654 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3656 __skb_pull(skb, skb_headlen(skb));
3657 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3658 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3660 skb->dev = napi->dev;
3666 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3668 struct sk_buff *skb = napi->skb;
3671 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3677 EXPORT_SYMBOL(napi_get_frags);
3679 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3685 skb->protocol = eth_type_trans(skb, skb->dev);
3687 if (ret == GRO_HELD)
3688 skb_gro_pull(skb, -ETH_HLEN);
3689 else if (netif_receive_skb(skb))
3694 case GRO_MERGED_FREE:
3695 napi_reuse_skb(napi, skb);
3704 EXPORT_SYMBOL(napi_frags_finish);
3706 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3708 struct sk_buff *skb = napi->skb;
3715 skb_reset_mac_header(skb);
3716 skb_gro_reset_offset(skb);
3718 off = skb_gro_offset(skb);
3719 hlen = off + sizeof(*eth);
3720 eth = skb_gro_header_fast(skb, off);
3721 if (skb_gro_header_hard(skb, hlen)) {
3722 eth = skb_gro_header_slow(skb, hlen, off);
3723 if (unlikely(!eth)) {
3724 napi_reuse_skb(napi, skb);
3730 skb_gro_pull(skb, sizeof(*eth));
3733 * This works because the only protocols we care about don't require
3734 * special handling. We'll fix it up properly at the end.
3736 skb->protocol = eth->h_proto;
3741 EXPORT_SYMBOL(napi_frags_skb);
3743 gro_result_t napi_gro_frags(struct napi_struct *napi)
3745 struct sk_buff *skb = napi_frags_skb(napi);
3750 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3752 EXPORT_SYMBOL(napi_gro_frags);
3755 * net_rps_action sends any pending IPI's for rps.
3756 * Note: called with local irq disabled, but exits with local irq enabled.
3758 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3761 struct softnet_data *remsd = sd->rps_ipi_list;
3764 sd->rps_ipi_list = NULL;
3768 /* Send pending IPI's to kick RPS processing on remote cpus. */
3770 struct softnet_data *next = remsd->rps_ipi_next;
3772 if (cpu_online(remsd->cpu))
3773 __smp_call_function_single(remsd->cpu,
3782 static int process_backlog(struct napi_struct *napi, int quota)
3785 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3788 /* Check if we have pending ipi, its better to send them now,
3789 * not waiting net_rx_action() end.
3791 if (sd->rps_ipi_list) {
3792 local_irq_disable();
3793 net_rps_action_and_irq_enable(sd);
3796 napi->weight = weight_p;
3797 local_irq_disable();
3798 while (work < quota) {
3799 struct sk_buff *skb;
3802 while ((skb = __skb_dequeue(&sd->process_queue))) {
3804 __netif_receive_skb(skb);
3805 local_irq_disable();
3806 input_queue_head_incr(sd);
3807 if (++work >= quota) {
3814 qlen = skb_queue_len(&sd->input_pkt_queue);
3816 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3817 &sd->process_queue);
3819 if (qlen < quota - work) {
3821 * Inline a custom version of __napi_complete().
3822 * only current cpu owns and manipulates this napi,
3823 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3824 * we can use a plain write instead of clear_bit(),
3825 * and we dont need an smp_mb() memory barrier.
3827 list_del(&napi->poll_list);
3830 quota = work + qlen;
3840 * __napi_schedule - schedule for receive
3841 * @n: entry to schedule
3843 * The entry's receive function will be scheduled to run
3845 void __napi_schedule(struct napi_struct *n)
3847 unsigned long flags;
3849 local_irq_save(flags);
3850 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3851 local_irq_restore(flags);
3853 EXPORT_SYMBOL(__napi_schedule);
3855 void __napi_complete(struct napi_struct *n)
3857 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3858 BUG_ON(n->gro_list);
3860 list_del(&n->poll_list);
3861 smp_mb__before_clear_bit();
3862 clear_bit(NAPI_STATE_SCHED, &n->state);
3864 EXPORT_SYMBOL(__napi_complete);
3866 void napi_complete(struct napi_struct *n)
3868 unsigned long flags;
3871 * don't let napi dequeue from the cpu poll list
3872 * just in case its running on a different cpu
3874 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3878 local_irq_save(flags);
3880 local_irq_restore(flags);
3882 EXPORT_SYMBOL(napi_complete);
3884 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3885 int (*poll)(struct napi_struct *, int), int weight)
3887 INIT_LIST_HEAD(&napi->poll_list);
3888 napi->gro_count = 0;
3889 napi->gro_list = NULL;
3892 napi->weight = weight;
3893 list_add(&napi->dev_list, &dev->napi_list);
3895 #ifdef CONFIG_NETPOLL
3896 spin_lock_init(&napi->poll_lock);
3897 napi->poll_owner = -1;
3899 set_bit(NAPI_STATE_SCHED, &napi->state);
3901 EXPORT_SYMBOL(netif_napi_add);
3903 void netif_napi_del(struct napi_struct *napi)
3905 struct sk_buff *skb, *next;
3907 list_del_init(&napi->dev_list);
3908 napi_free_frags(napi);
3910 for (skb = napi->gro_list; skb; skb = next) {
3916 napi->gro_list = NULL;
3917 napi->gro_count = 0;
3919 EXPORT_SYMBOL(netif_napi_del);
3921 static void net_rx_action(struct softirq_action *h)
3923 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3924 unsigned long time_limit = jiffies + 2;
3925 int budget = netdev_budget;
3928 local_irq_disable();
3930 while (!list_empty(&sd->poll_list)) {
3931 struct napi_struct *n;
3934 /* If softirq window is exhuasted then punt.
3935 * Allow this to run for 2 jiffies since which will allow
3936 * an average latency of 1.5/HZ.
3938 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3943 /* Even though interrupts have been re-enabled, this
3944 * access is safe because interrupts can only add new
3945 * entries to the tail of this list, and only ->poll()
3946 * calls can remove this head entry from the list.
3948 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3950 have = netpoll_poll_lock(n);
3954 /* This NAPI_STATE_SCHED test is for avoiding a race
3955 * with netpoll's poll_napi(). Only the entity which
3956 * obtains the lock and sees NAPI_STATE_SCHED set will
3957 * actually make the ->poll() call. Therefore we avoid
3958 * accidentally calling ->poll() when NAPI is not scheduled.
3961 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3962 work = n->poll(n, weight);
3966 WARN_ON_ONCE(work > weight);
3970 local_irq_disable();
3972 /* Drivers must not modify the NAPI state if they
3973 * consume the entire weight. In such cases this code
3974 * still "owns" the NAPI instance and therefore can
3975 * move the instance around on the list at-will.
3977 if (unlikely(work == weight)) {
3978 if (unlikely(napi_disable_pending(n))) {
3981 local_irq_disable();
3983 list_move_tail(&n->poll_list, &sd->poll_list);
3986 netpoll_poll_unlock(have);
3989 net_rps_action_and_irq_enable(sd);
3991 #ifdef CONFIG_NET_DMA
3993 * There may not be any more sk_buffs coming right now, so push
3994 * any pending DMA copies to hardware
3996 dma_issue_pending_all();
4003 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4007 static gifconf_func_t *gifconf_list[NPROTO];
4010 * register_gifconf - register a SIOCGIF handler
4011 * @family: Address family
4012 * @gifconf: Function handler
4014 * Register protocol dependent address dumping routines. The handler
4015 * that is passed must not be freed or reused until it has been replaced
4016 * by another handler.
4018 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4020 if (family >= NPROTO)
4022 gifconf_list[family] = gifconf;
4025 EXPORT_SYMBOL(register_gifconf);
4029 * Map an interface index to its name (SIOCGIFNAME)
4033 * We need this ioctl for efficient implementation of the
4034 * if_indextoname() function required by the IPv6 API. Without
4035 * it, we would have to search all the interfaces to find a
4039 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4041 struct net_device *dev;
4045 * Fetch the caller's info block.
4048 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4052 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4058 strcpy(ifr.ifr_name, dev->name);
4061 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4067 * Perform a SIOCGIFCONF call. This structure will change
4068 * size eventually, and there is nothing I can do about it.
4069 * Thus we will need a 'compatibility mode'.
4072 static int dev_ifconf(struct net *net, char __user *arg)
4075 struct net_device *dev;
4082 * Fetch the caller's info block.
4085 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4092 * Loop over the interfaces, and write an info block for each.
4096 for_each_netdev(net, dev) {
4097 for (i = 0; i < NPROTO; i++) {
4098 if (gifconf_list[i]) {
4101 done = gifconf_list[i](dev, NULL, 0);
4103 done = gifconf_list[i](dev, pos + total,
4113 * All done. Write the updated control block back to the caller.
4115 ifc.ifc_len = total;
4118 * Both BSD and Solaris return 0 here, so we do too.
4120 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4123 #ifdef CONFIG_PROC_FS
4125 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4127 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4128 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4129 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4131 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4133 struct net *net = seq_file_net(seq);
4134 struct net_device *dev;
4135 struct hlist_node *p;
4136 struct hlist_head *h;
4137 unsigned int count = 0, offset = get_offset(*pos);
4139 h = &net->dev_name_head[get_bucket(*pos)];
4140 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4141 if (++count == offset)
4148 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4150 struct net_device *dev;
4151 unsigned int bucket;
4154 dev = dev_from_same_bucket(seq, pos);
4158 bucket = get_bucket(*pos) + 1;
4159 *pos = set_bucket_offset(bucket, 1);
4160 } while (bucket < NETDEV_HASHENTRIES);
4166 * This is invoked by the /proc filesystem handler to display a device
4169 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4174 return SEQ_START_TOKEN;
4176 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4179 return dev_from_bucket(seq, pos);
4182 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4185 return dev_from_bucket(seq, pos);
4188 void dev_seq_stop(struct seq_file *seq, void *v)
4194 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4196 struct rtnl_link_stats64 temp;
4197 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4199 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4200 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4201 dev->name, stats->rx_bytes, stats->rx_packets,
4203 stats->rx_dropped + stats->rx_missed_errors,
4204 stats->rx_fifo_errors,
4205 stats->rx_length_errors + stats->rx_over_errors +
4206 stats->rx_crc_errors + stats->rx_frame_errors,
4207 stats->rx_compressed, stats->multicast,
4208 stats->tx_bytes, stats->tx_packets,
4209 stats->tx_errors, stats->tx_dropped,
4210 stats->tx_fifo_errors, stats->collisions,
4211 stats->tx_carrier_errors +
4212 stats->tx_aborted_errors +
4213 stats->tx_window_errors +
4214 stats->tx_heartbeat_errors,
4215 stats->tx_compressed);
4219 * Called from the PROCfs module. This now uses the new arbitrary sized
4220 * /proc/net interface to create /proc/net/dev
4222 static int dev_seq_show(struct seq_file *seq, void *v)
4224 if (v == SEQ_START_TOKEN)
4225 seq_puts(seq, "Inter-| Receive "
4227 " face |bytes packets errs drop fifo frame "
4228 "compressed multicast|bytes packets errs "
4229 "drop fifo colls carrier compressed\n");
4231 dev_seq_printf_stats(seq, v);
4235 static struct softnet_data *softnet_get_online(loff_t *pos)
4237 struct softnet_data *sd = NULL;
4239 while (*pos < nr_cpu_ids)
4240 if (cpu_online(*pos)) {
4241 sd = &per_cpu(softnet_data, *pos);
4248 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4250 return softnet_get_online(pos);
4253 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4256 return softnet_get_online(pos);
4259 static void softnet_seq_stop(struct seq_file *seq, void *v)
4263 static int softnet_seq_show(struct seq_file *seq, void *v)
4265 struct softnet_data *sd = v;
4267 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4268 sd->processed, sd->dropped, sd->time_squeeze, 0,
4269 0, 0, 0, 0, /* was fastroute */
4270 sd->cpu_collision, sd->received_rps);
4274 static const struct seq_operations dev_seq_ops = {
4275 .start = dev_seq_start,
4276 .next = dev_seq_next,
4277 .stop = dev_seq_stop,
4278 .show = dev_seq_show,
4281 static int dev_seq_open(struct inode *inode, struct file *file)
4283 return seq_open_net(inode, file, &dev_seq_ops,
4284 sizeof(struct seq_net_private));
4287 static const struct file_operations dev_seq_fops = {
4288 .owner = THIS_MODULE,
4289 .open = dev_seq_open,
4291 .llseek = seq_lseek,
4292 .release = seq_release_net,
4295 static const struct seq_operations softnet_seq_ops = {
4296 .start = softnet_seq_start,
4297 .next = softnet_seq_next,
4298 .stop = softnet_seq_stop,
4299 .show = softnet_seq_show,
4302 static int softnet_seq_open(struct inode *inode, struct file *file)
4304 return seq_open(file, &softnet_seq_ops);
4307 static const struct file_operations softnet_seq_fops = {
4308 .owner = THIS_MODULE,
4309 .open = softnet_seq_open,
4311 .llseek = seq_lseek,
4312 .release = seq_release,
4315 static void *ptype_get_idx(loff_t pos)
4317 struct packet_type *pt = NULL;
4321 list_for_each_entry_rcu(pt, &ptype_all, list) {
4327 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4328 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4337 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4341 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4344 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4346 struct packet_type *pt;
4347 struct list_head *nxt;
4351 if (v == SEQ_START_TOKEN)
4352 return ptype_get_idx(0);
4355 nxt = pt->list.next;
4356 if (pt->type == htons(ETH_P_ALL)) {
4357 if (nxt != &ptype_all)
4360 nxt = ptype_base[0].next;
4362 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4364 while (nxt == &ptype_base[hash]) {
4365 if (++hash >= PTYPE_HASH_SIZE)
4367 nxt = ptype_base[hash].next;
4370 return list_entry(nxt, struct packet_type, list);
4373 static void ptype_seq_stop(struct seq_file *seq, void *v)
4379 static int ptype_seq_show(struct seq_file *seq, void *v)
4381 struct packet_type *pt = v;
4383 if (v == SEQ_START_TOKEN)
4384 seq_puts(seq, "Type Device Function\n");
4385 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4386 if (pt->type == htons(ETH_P_ALL))
4387 seq_puts(seq, "ALL ");
4389 seq_printf(seq, "%04x", ntohs(pt->type));
4391 seq_printf(seq, " %-8s %pF\n",
4392 pt->dev ? pt->dev->name : "", pt->func);
4398 static const struct seq_operations ptype_seq_ops = {
4399 .start = ptype_seq_start,
4400 .next = ptype_seq_next,
4401 .stop = ptype_seq_stop,
4402 .show = ptype_seq_show,
4405 static int ptype_seq_open(struct inode *inode, struct file *file)
4407 return seq_open_net(inode, file, &ptype_seq_ops,
4408 sizeof(struct seq_net_private));
4411 static const struct file_operations ptype_seq_fops = {
4412 .owner = THIS_MODULE,
4413 .open = ptype_seq_open,
4415 .llseek = seq_lseek,
4416 .release = seq_release_net,
4420 static int __net_init dev_proc_net_init(struct net *net)
4424 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4426 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4428 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4431 if (wext_proc_init(net))
4437 proc_net_remove(net, "ptype");
4439 proc_net_remove(net, "softnet_stat");
4441 proc_net_remove(net, "dev");
4445 static void __net_exit dev_proc_net_exit(struct net *net)
4447 wext_proc_exit(net);
4449 proc_net_remove(net, "ptype");
4450 proc_net_remove(net, "softnet_stat");
4451 proc_net_remove(net, "dev");
4454 static struct pernet_operations __net_initdata dev_proc_ops = {
4455 .init = dev_proc_net_init,
4456 .exit = dev_proc_net_exit,
4459 static int __init dev_proc_init(void)
4461 return register_pernet_subsys(&dev_proc_ops);
4464 #define dev_proc_init() 0
4465 #endif /* CONFIG_PROC_FS */
4469 * netdev_set_master - set up master pointer
4470 * @slave: slave device
4471 * @master: new master device
4473 * Changes the master device of the slave. Pass %NULL to break the
4474 * bonding. The caller must hold the RTNL semaphore. On a failure
4475 * a negative errno code is returned. On success the reference counts
4476 * are adjusted and the function returns zero.
4478 int netdev_set_master(struct net_device *slave, struct net_device *master)
4480 struct net_device *old = slave->master;
4490 slave->master = master;
4496 EXPORT_SYMBOL(netdev_set_master);
4499 * netdev_set_bond_master - set up bonding master/slave pair
4500 * @slave: slave device
4501 * @master: new master device
4503 * Changes the master device of the slave. Pass %NULL to break the
4504 * bonding. The caller must hold the RTNL semaphore. On a failure
4505 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4506 * to the routing socket and the function returns zero.
4508 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4514 err = netdev_set_master(slave, master);
4518 slave->flags |= IFF_SLAVE;
4520 slave->flags &= ~IFF_SLAVE;
4522 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4525 EXPORT_SYMBOL(netdev_set_bond_master);
4527 static void dev_change_rx_flags(struct net_device *dev, int flags)
4529 const struct net_device_ops *ops = dev->netdev_ops;
4531 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4532 ops->ndo_change_rx_flags(dev, flags);
4535 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4537 unsigned short old_flags = dev->flags;
4543 dev->flags |= IFF_PROMISC;
4544 dev->promiscuity += inc;
4545 if (dev->promiscuity == 0) {
4548 * If inc causes overflow, untouch promisc and return error.
4551 dev->flags &= ~IFF_PROMISC;
4553 dev->promiscuity -= inc;
4554 printk(KERN_WARNING "%s: promiscuity touches roof, "
4555 "set promiscuity failed, promiscuity feature "
4556 "of device might be broken.\n", dev->name);
4560 if (dev->flags != old_flags) {
4561 printk(KERN_INFO "device %s %s promiscuous mode\n",
4562 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4564 if (audit_enabled) {
4565 current_uid_gid(&uid, &gid);
4566 audit_log(current->audit_context, GFP_ATOMIC,
4567 AUDIT_ANOM_PROMISCUOUS,
4568 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4569 dev->name, (dev->flags & IFF_PROMISC),
4570 (old_flags & IFF_PROMISC),
4571 audit_get_loginuid(current),
4573 audit_get_sessionid(current));
4576 dev_change_rx_flags(dev, IFF_PROMISC);
4582 * dev_set_promiscuity - update promiscuity count on a device
4586 * Add or remove promiscuity from a device. While the count in the device
4587 * remains above zero the interface remains promiscuous. Once it hits zero
4588 * the device reverts back to normal filtering operation. A negative inc
4589 * value is used to drop promiscuity on the device.
4590 * Return 0 if successful or a negative errno code on error.
4592 int dev_set_promiscuity(struct net_device *dev, int inc)
4594 unsigned short old_flags = dev->flags;
4597 err = __dev_set_promiscuity(dev, inc);
4600 if (dev->flags != old_flags)
4601 dev_set_rx_mode(dev);
4604 EXPORT_SYMBOL(dev_set_promiscuity);
4607 * dev_set_allmulti - update allmulti count on a device
4611 * Add or remove reception of all multicast frames to a device. While the
4612 * count in the device remains above zero the interface remains listening
4613 * to all interfaces. Once it hits zero the device reverts back to normal
4614 * filtering operation. A negative @inc value is used to drop the counter
4615 * when releasing a resource needing all multicasts.
4616 * Return 0 if successful or a negative errno code on error.
4619 int dev_set_allmulti(struct net_device *dev, int inc)
4621 unsigned short old_flags = dev->flags;
4625 dev->flags |= IFF_ALLMULTI;
4626 dev->allmulti += inc;
4627 if (dev->allmulti == 0) {
4630 * If inc causes overflow, untouch allmulti and return error.
4633 dev->flags &= ~IFF_ALLMULTI;
4635 dev->allmulti -= inc;
4636 printk(KERN_WARNING "%s: allmulti touches roof, "
4637 "set allmulti failed, allmulti feature of "
4638 "device might be broken.\n", dev->name);
4642 if (dev->flags ^ old_flags) {
4643 dev_change_rx_flags(dev, IFF_ALLMULTI);
4644 dev_set_rx_mode(dev);
4648 EXPORT_SYMBOL(dev_set_allmulti);
4651 * Upload unicast and multicast address lists to device and
4652 * configure RX filtering. When the device doesn't support unicast
4653 * filtering it is put in promiscuous mode while unicast addresses
4656 void __dev_set_rx_mode(struct net_device *dev)
4658 const struct net_device_ops *ops = dev->netdev_ops;
4660 /* dev_open will call this function so the list will stay sane. */
4661 if (!(dev->flags&IFF_UP))
4664 if (!netif_device_present(dev))
4667 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4668 /* Unicast addresses changes may only happen under the rtnl,
4669 * therefore calling __dev_set_promiscuity here is safe.
4671 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4672 __dev_set_promiscuity(dev, 1);
4673 dev->uc_promisc = true;
4674 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4675 __dev_set_promiscuity(dev, -1);
4676 dev->uc_promisc = false;
4680 if (ops->ndo_set_rx_mode)
4681 ops->ndo_set_rx_mode(dev);
4684 void dev_set_rx_mode(struct net_device *dev)
4686 netif_addr_lock_bh(dev);
4687 __dev_set_rx_mode(dev);
4688 netif_addr_unlock_bh(dev);
4692 * dev_get_flags - get flags reported to userspace
4695 * Get the combination of flag bits exported through APIs to userspace.
4697 unsigned dev_get_flags(const struct net_device *dev)
4701 flags = (dev->flags & ~(IFF_PROMISC |
4706 (dev->gflags & (IFF_PROMISC |
4709 if (netif_running(dev)) {
4710 if (netif_oper_up(dev))
4711 flags |= IFF_RUNNING;
4712 if (netif_carrier_ok(dev))
4713 flags |= IFF_LOWER_UP;
4714 if (netif_dormant(dev))
4715 flags |= IFF_DORMANT;
4720 EXPORT_SYMBOL(dev_get_flags);
4722 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4724 int old_flags = dev->flags;
4730 * Set the flags on our device.
4733 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4734 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4736 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4740 * Load in the correct multicast list now the flags have changed.
4743 if ((old_flags ^ flags) & IFF_MULTICAST)
4744 dev_change_rx_flags(dev, IFF_MULTICAST);
4746 dev_set_rx_mode(dev);
4749 * Have we downed the interface. We handle IFF_UP ourselves
4750 * according to user attempts to set it, rather than blindly
4755 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4756 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4759 dev_set_rx_mode(dev);
4762 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4763 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4765 dev->gflags ^= IFF_PROMISC;
4766 dev_set_promiscuity(dev, inc);
4769 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4770 is important. Some (broken) drivers set IFF_PROMISC, when
4771 IFF_ALLMULTI is requested not asking us and not reporting.
4773 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4774 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4776 dev->gflags ^= IFF_ALLMULTI;
4777 dev_set_allmulti(dev, inc);
4783 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4785 unsigned int changes = dev->flags ^ old_flags;
4787 if (changes & IFF_UP) {
4788 if (dev->flags & IFF_UP)
4789 call_netdevice_notifiers(NETDEV_UP, dev);
4791 call_netdevice_notifiers(NETDEV_DOWN, dev);
4794 if (dev->flags & IFF_UP &&
4795 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4796 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4800 * dev_change_flags - change device settings
4802 * @flags: device state flags
4804 * Change settings on device based state flags. The flags are
4805 * in the userspace exported format.
4807 int dev_change_flags(struct net_device *dev, unsigned flags)
4810 int old_flags = dev->flags;
4812 ret = __dev_change_flags(dev, flags);
4816 changes = old_flags ^ dev->flags;
4818 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4820 __dev_notify_flags(dev, old_flags);
4823 EXPORT_SYMBOL(dev_change_flags);
4826 * dev_set_mtu - Change maximum transfer unit
4828 * @new_mtu: new transfer unit
4830 * Change the maximum transfer size of the network device.
4832 int dev_set_mtu(struct net_device *dev, int new_mtu)
4834 const struct net_device_ops *ops = dev->netdev_ops;
4837 if (new_mtu == dev->mtu)
4840 /* MTU must be positive. */
4844 if (!netif_device_present(dev))
4848 if (ops->ndo_change_mtu)
4849 err = ops->ndo_change_mtu(dev, new_mtu);
4853 if (!err && dev->flags & IFF_UP)
4854 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4857 EXPORT_SYMBOL(dev_set_mtu);
4860 * dev_set_group - Change group this device belongs to
4862 * @new_group: group this device should belong to
4864 void dev_set_group(struct net_device *dev, int new_group)
4866 dev->group = new_group;
4868 EXPORT_SYMBOL(dev_set_group);
4871 * dev_set_mac_address - Change Media Access Control Address
4875 * Change the hardware (MAC) address of the device
4877 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4879 const struct net_device_ops *ops = dev->netdev_ops;
4882 if (!ops->ndo_set_mac_address)
4884 if (sa->sa_family != dev->type)
4886 if (!netif_device_present(dev))
4888 err = ops->ndo_set_mac_address(dev, sa);
4890 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4893 EXPORT_SYMBOL(dev_set_mac_address);
4896 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4898 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4901 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4907 case SIOCGIFFLAGS: /* Get interface flags */
4908 ifr->ifr_flags = (short) dev_get_flags(dev);
4911 case SIOCGIFMETRIC: /* Get the metric on the interface
4912 (currently unused) */
4913 ifr->ifr_metric = 0;
4916 case SIOCGIFMTU: /* Get the MTU of a device */
4917 ifr->ifr_mtu = dev->mtu;
4922 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4924 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4925 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4926 ifr->ifr_hwaddr.sa_family = dev->type;
4934 ifr->ifr_map.mem_start = dev->mem_start;
4935 ifr->ifr_map.mem_end = dev->mem_end;
4936 ifr->ifr_map.base_addr = dev->base_addr;
4937 ifr->ifr_map.irq = dev->irq;
4938 ifr->ifr_map.dma = dev->dma;
4939 ifr->ifr_map.port = dev->if_port;
4943 ifr->ifr_ifindex = dev->ifindex;
4947 ifr->ifr_qlen = dev->tx_queue_len;
4951 /* dev_ioctl() should ensure this case
4963 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4965 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4968 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4969 const struct net_device_ops *ops;
4974 ops = dev->netdev_ops;
4977 case SIOCSIFFLAGS: /* Set interface flags */
4978 return dev_change_flags(dev, ifr->ifr_flags);
4980 case SIOCSIFMETRIC: /* Set the metric on the interface
4981 (currently unused) */
4984 case SIOCSIFMTU: /* Set the MTU of a device */
4985 return dev_set_mtu(dev, ifr->ifr_mtu);
4988 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4990 case SIOCSIFHWBROADCAST:
4991 if (ifr->ifr_hwaddr.sa_family != dev->type)
4993 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4994 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4995 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4999 if (ops->ndo_set_config) {
5000 if (!netif_device_present(dev))
5002 return ops->ndo_set_config(dev, &ifr->ifr_map);
5007 if (!ops->ndo_set_rx_mode ||
5008 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5010 if (!netif_device_present(dev))
5012 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5015 if (!ops->ndo_set_rx_mode ||
5016 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5018 if (!netif_device_present(dev))
5020 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5023 if (ifr->ifr_qlen < 0)
5025 dev->tx_queue_len = ifr->ifr_qlen;
5029 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5030 return dev_change_name(dev, ifr->ifr_newname);
5033 err = net_hwtstamp_validate(ifr);
5039 * Unknown or private ioctl
5042 if ((cmd >= SIOCDEVPRIVATE &&
5043 cmd <= SIOCDEVPRIVATE + 15) ||
5044 cmd == SIOCBONDENSLAVE ||
5045 cmd == SIOCBONDRELEASE ||
5046 cmd == SIOCBONDSETHWADDR ||
5047 cmd == SIOCBONDSLAVEINFOQUERY ||
5048 cmd == SIOCBONDINFOQUERY ||
5049 cmd == SIOCBONDCHANGEACTIVE ||
5050 cmd == SIOCGMIIPHY ||
5051 cmd == SIOCGMIIREG ||
5052 cmd == SIOCSMIIREG ||
5053 cmd == SIOCBRADDIF ||
5054 cmd == SIOCBRDELIF ||
5055 cmd == SIOCSHWTSTAMP ||
5056 cmd == SIOCWANDEV) {
5058 if (ops->ndo_do_ioctl) {
5059 if (netif_device_present(dev))
5060 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5072 * This function handles all "interface"-type I/O control requests. The actual
5073 * 'doing' part of this is dev_ifsioc above.
5077 * dev_ioctl - network device ioctl
5078 * @net: the applicable net namespace
5079 * @cmd: command to issue
5080 * @arg: pointer to a struct ifreq in user space
5082 * Issue ioctl functions to devices. This is normally called by the
5083 * user space syscall interfaces but can sometimes be useful for
5084 * other purposes. The return value is the return from the syscall if
5085 * positive or a negative errno code on error.
5088 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5094 /* One special case: SIOCGIFCONF takes ifconf argument
5095 and requires shared lock, because it sleeps writing
5099 if (cmd == SIOCGIFCONF) {
5101 ret = dev_ifconf(net, (char __user *) arg);
5105 if (cmd == SIOCGIFNAME)
5106 return dev_ifname(net, (struct ifreq __user *)arg);
5108 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5111 ifr.ifr_name[IFNAMSIZ-1] = 0;
5113 colon = strchr(ifr.ifr_name, ':');
5118 * See which interface the caller is talking about.
5123 * These ioctl calls:
5124 * - can be done by all.
5125 * - atomic and do not require locking.
5136 dev_load(net, ifr.ifr_name);
5138 ret = dev_ifsioc_locked(net, &ifr, cmd);
5143 if (copy_to_user(arg, &ifr,
5144 sizeof(struct ifreq)))
5150 dev_load(net, ifr.ifr_name);
5152 ret = dev_ethtool(net, &ifr);
5157 if (copy_to_user(arg, &ifr,
5158 sizeof(struct ifreq)))
5164 * These ioctl calls:
5165 * - require superuser power.
5166 * - require strict serialization.
5172 if (!capable(CAP_NET_ADMIN))
5174 dev_load(net, ifr.ifr_name);
5176 ret = dev_ifsioc(net, &ifr, cmd);
5181 if (copy_to_user(arg, &ifr,
5182 sizeof(struct ifreq)))
5188 * These ioctl calls:
5189 * - require superuser power.
5190 * - require strict serialization.
5191 * - do not return a value
5201 case SIOCSIFHWBROADCAST:
5204 case SIOCBONDENSLAVE:
5205 case SIOCBONDRELEASE:
5206 case SIOCBONDSETHWADDR:
5207 case SIOCBONDCHANGEACTIVE:
5211 if (!capable(CAP_NET_ADMIN))
5214 case SIOCBONDSLAVEINFOQUERY:
5215 case SIOCBONDINFOQUERY:
5216 dev_load(net, ifr.ifr_name);
5218 ret = dev_ifsioc(net, &ifr, cmd);
5223 /* Get the per device memory space. We can add this but
5224 * currently do not support it */
5226 /* Set the per device memory buffer space.
5227 * Not applicable in our case */
5232 * Unknown or private ioctl.
5235 if (cmd == SIOCWANDEV ||
5236 (cmd >= SIOCDEVPRIVATE &&
5237 cmd <= SIOCDEVPRIVATE + 15)) {
5238 dev_load(net, ifr.ifr_name);
5240 ret = dev_ifsioc(net, &ifr, cmd);
5242 if (!ret && copy_to_user(arg, &ifr,
5243 sizeof(struct ifreq)))
5247 /* Take care of Wireless Extensions */
5248 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5249 return wext_handle_ioctl(net, &ifr, cmd, arg);
5256 * dev_new_index - allocate an ifindex
5257 * @net: the applicable net namespace
5259 * Returns a suitable unique value for a new device interface
5260 * number. The caller must hold the rtnl semaphore or the
5261 * dev_base_lock to be sure it remains unique.
5263 static int dev_new_index(struct net *net)
5269 if (!__dev_get_by_index(net, ifindex))
5274 /* Delayed registration/unregisteration */
5275 static LIST_HEAD(net_todo_list);
5277 static void net_set_todo(struct net_device *dev)
5279 list_add_tail(&dev->todo_list, &net_todo_list);
5282 static void rollback_registered_many(struct list_head *head)
5284 struct net_device *dev, *tmp;
5286 BUG_ON(dev_boot_phase);
5289 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5290 /* Some devices call without registering
5291 * for initialization unwind. Remove those
5292 * devices and proceed with the remaining.
5294 if (dev->reg_state == NETREG_UNINITIALIZED) {
5295 pr_debug("unregister_netdevice: device %s/%p never "
5296 "was registered\n", dev->name, dev);
5299 list_del(&dev->unreg_list);
5302 dev->dismantle = true;
5303 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5306 /* If device is running, close it first. */
5307 dev_close_many(head);
5309 list_for_each_entry(dev, head, unreg_list) {
5310 /* And unlink it from device chain. */
5311 unlist_netdevice(dev);
5313 dev->reg_state = NETREG_UNREGISTERING;
5318 list_for_each_entry(dev, head, unreg_list) {
5319 /* Shutdown queueing discipline. */
5323 /* Notify protocols, that we are about to destroy
5324 this device. They should clean all the things.
5326 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5328 if (!dev->rtnl_link_ops ||
5329 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5330 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5333 * Flush the unicast and multicast chains
5338 if (dev->netdev_ops->ndo_uninit)
5339 dev->netdev_ops->ndo_uninit(dev);
5341 /* Notifier chain MUST detach us from master device. */
5342 WARN_ON(dev->master);
5344 /* Remove entries from kobject tree */
5345 netdev_unregister_kobject(dev);
5348 /* Process any work delayed until the end of the batch */
5349 dev = list_first_entry(head, struct net_device, unreg_list);
5350 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5354 list_for_each_entry(dev, head, unreg_list)
5358 static void rollback_registered(struct net_device *dev)
5362 list_add(&dev->unreg_list, &single);
5363 rollback_registered_many(&single);
5367 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5369 /* Fix illegal checksum combinations */
5370 if ((features & NETIF_F_HW_CSUM) &&
5371 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5372 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5373 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5376 if ((features & NETIF_F_NO_CSUM) &&
5377 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5378 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5379 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5382 /* Fix illegal SG+CSUM combinations. */
5383 if ((features & NETIF_F_SG) &&
5384 !(features & NETIF_F_ALL_CSUM)) {
5386 "Dropping NETIF_F_SG since no checksum feature.\n");
5387 features &= ~NETIF_F_SG;
5390 /* TSO requires that SG is present as well. */
5391 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5392 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5393 features &= ~NETIF_F_ALL_TSO;
5396 /* TSO ECN requires that TSO is present as well. */
5397 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5398 features &= ~NETIF_F_TSO_ECN;
5400 /* Software GSO depends on SG. */
5401 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5402 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5403 features &= ~NETIF_F_GSO;
5406 /* UFO needs SG and checksumming */
5407 if (features & NETIF_F_UFO) {
5408 /* maybe split UFO into V4 and V6? */
5409 if (!((features & NETIF_F_GEN_CSUM) ||
5410 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5411 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5413 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5414 features &= ~NETIF_F_UFO;
5417 if (!(features & NETIF_F_SG)) {
5419 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5420 features &= ~NETIF_F_UFO;
5427 int __netdev_update_features(struct net_device *dev)
5434 features = netdev_get_wanted_features(dev);
5436 if (dev->netdev_ops->ndo_fix_features)
5437 features = dev->netdev_ops->ndo_fix_features(dev, features);
5439 /* driver might be less strict about feature dependencies */
5440 features = netdev_fix_features(dev, features);
5442 if (dev->features == features)
5445 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5446 dev->features, features);
5448 if (dev->netdev_ops->ndo_set_features)
5449 err = dev->netdev_ops->ndo_set_features(dev, features);
5451 if (unlikely(err < 0)) {
5453 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5454 err, features, dev->features);
5459 dev->features = features;
5465 * netdev_update_features - recalculate device features
5466 * @dev: the device to check
5468 * Recalculate dev->features set and send notifications if it
5469 * has changed. Should be called after driver or hardware dependent
5470 * conditions might have changed that influence the features.
5472 void netdev_update_features(struct net_device *dev)
5474 if (__netdev_update_features(dev))
5475 netdev_features_change(dev);
5477 EXPORT_SYMBOL(netdev_update_features);
5480 * netdev_change_features - recalculate device features
5481 * @dev: the device to check
5483 * Recalculate dev->features set and send notifications even
5484 * if they have not changed. Should be called instead of
5485 * netdev_update_features() if also dev->vlan_features might
5486 * have changed to allow the changes to be propagated to stacked
5489 void netdev_change_features(struct net_device *dev)
5491 __netdev_update_features(dev);
5492 netdev_features_change(dev);
5494 EXPORT_SYMBOL(netdev_change_features);
5497 * netif_stacked_transfer_operstate - transfer operstate
5498 * @rootdev: the root or lower level device to transfer state from
5499 * @dev: the device to transfer operstate to
5501 * Transfer operational state from root to device. This is normally
5502 * called when a stacking relationship exists between the root
5503 * device and the device(a leaf device).
5505 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5506 struct net_device *dev)
5508 if (rootdev->operstate == IF_OPER_DORMANT)
5509 netif_dormant_on(dev);
5511 netif_dormant_off(dev);
5513 if (netif_carrier_ok(rootdev)) {
5514 if (!netif_carrier_ok(dev))
5515 netif_carrier_on(dev);
5517 if (netif_carrier_ok(dev))
5518 netif_carrier_off(dev);
5521 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5524 static int netif_alloc_rx_queues(struct net_device *dev)
5526 unsigned int i, count = dev->num_rx_queues;
5527 struct netdev_rx_queue *rx;
5531 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5533 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5538 for (i = 0; i < count; i++)
5544 static void netdev_init_one_queue(struct net_device *dev,
5545 struct netdev_queue *queue, void *_unused)
5547 /* Initialize queue lock */
5548 spin_lock_init(&queue->_xmit_lock);
5549 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5550 queue->xmit_lock_owner = -1;
5551 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5555 static int netif_alloc_netdev_queues(struct net_device *dev)
5557 unsigned int count = dev->num_tx_queues;
5558 struct netdev_queue *tx;
5562 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5564 pr_err("netdev: Unable to allocate %u tx queues.\n",
5570 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5571 spin_lock_init(&dev->tx_global_lock);
5577 * register_netdevice - register a network device
5578 * @dev: device to register
5580 * Take a completed network device structure and add it to the kernel
5581 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5582 * chain. 0 is returned on success. A negative errno code is returned
5583 * on a failure to set up the device, or if the name is a duplicate.
5585 * Callers must hold the rtnl semaphore. You may want
5586 * register_netdev() instead of this.
5589 * The locking appears insufficient to guarantee two parallel registers
5590 * will not get the same name.
5593 int register_netdevice(struct net_device *dev)
5596 struct net *net = dev_net(dev);
5598 BUG_ON(dev_boot_phase);
5603 /* When net_device's are persistent, this will be fatal. */
5604 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5607 spin_lock_init(&dev->addr_list_lock);
5608 netdev_set_addr_lockdep_class(dev);
5612 ret = dev_get_valid_name(dev, dev->name);
5616 /* Init, if this function is available */
5617 if (dev->netdev_ops->ndo_init) {
5618 ret = dev->netdev_ops->ndo_init(dev);
5626 dev->ifindex = dev_new_index(net);
5627 if (dev->iflink == -1)
5628 dev->iflink = dev->ifindex;
5630 /* Transfer changeable features to wanted_features and enable
5631 * software offloads (GSO and GRO).
5633 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5634 dev->features |= NETIF_F_SOFT_FEATURES;
5635 dev->wanted_features = dev->features & dev->hw_features;
5637 /* Turn on no cache copy if HW is doing checksum */
5638 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5639 if ((dev->features & NETIF_F_ALL_CSUM) &&
5640 !(dev->features & NETIF_F_NO_CSUM)) {
5641 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5642 dev->features |= NETIF_F_NOCACHE_COPY;
5645 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5647 dev->vlan_features |= NETIF_F_HIGHDMA;
5649 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5650 ret = notifier_to_errno(ret);
5654 ret = netdev_register_kobject(dev);
5657 dev->reg_state = NETREG_REGISTERED;
5659 __netdev_update_features(dev);
5662 * Default initial state at registry is that the
5663 * device is present.
5666 set_bit(__LINK_STATE_PRESENT, &dev->state);
5668 dev_init_scheduler(dev);
5670 list_netdevice(dev);
5672 /* Notify protocols, that a new device appeared. */
5673 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5674 ret = notifier_to_errno(ret);
5676 rollback_registered(dev);
5677 dev->reg_state = NETREG_UNREGISTERED;
5680 * Prevent userspace races by waiting until the network
5681 * device is fully setup before sending notifications.
5683 if (!dev->rtnl_link_ops ||
5684 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5685 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5691 if (dev->netdev_ops->ndo_uninit)
5692 dev->netdev_ops->ndo_uninit(dev);
5695 EXPORT_SYMBOL(register_netdevice);
5698 * init_dummy_netdev - init a dummy network device for NAPI
5699 * @dev: device to init
5701 * This takes a network device structure and initialize the minimum
5702 * amount of fields so it can be used to schedule NAPI polls without
5703 * registering a full blown interface. This is to be used by drivers
5704 * that need to tie several hardware interfaces to a single NAPI
5705 * poll scheduler due to HW limitations.
5707 int init_dummy_netdev(struct net_device *dev)
5709 /* Clear everything. Note we don't initialize spinlocks
5710 * are they aren't supposed to be taken by any of the
5711 * NAPI code and this dummy netdev is supposed to be
5712 * only ever used for NAPI polls
5714 memset(dev, 0, sizeof(struct net_device));
5716 /* make sure we BUG if trying to hit standard
5717 * register/unregister code path
5719 dev->reg_state = NETREG_DUMMY;
5721 /* NAPI wants this */
5722 INIT_LIST_HEAD(&dev->napi_list);
5724 /* a dummy interface is started by default */
5725 set_bit(__LINK_STATE_PRESENT, &dev->state);
5726 set_bit(__LINK_STATE_START, &dev->state);
5728 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5729 * because users of this 'device' dont need to change
5735 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5739 * register_netdev - register a network device
5740 * @dev: device to register
5742 * Take a completed network device structure and add it to the kernel
5743 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5744 * chain. 0 is returned on success. A negative errno code is returned
5745 * on a failure to set up the device, or if the name is a duplicate.
5747 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5748 * and expands the device name if you passed a format string to
5751 int register_netdev(struct net_device *dev)
5756 err = register_netdevice(dev);
5760 EXPORT_SYMBOL(register_netdev);
5762 int netdev_refcnt_read(const struct net_device *dev)
5766 for_each_possible_cpu(i)
5767 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5770 EXPORT_SYMBOL(netdev_refcnt_read);
5773 * netdev_wait_allrefs - wait until all references are gone.
5775 * This is called when unregistering network devices.
5777 * Any protocol or device that holds a reference should register
5778 * for netdevice notification, and cleanup and put back the
5779 * reference if they receive an UNREGISTER event.
5780 * We can get stuck here if buggy protocols don't correctly
5783 static void netdev_wait_allrefs(struct net_device *dev)
5785 unsigned long rebroadcast_time, warning_time;
5788 linkwatch_forget_dev(dev);
5790 rebroadcast_time = warning_time = jiffies;
5791 refcnt = netdev_refcnt_read(dev);
5793 while (refcnt != 0) {
5794 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5797 /* Rebroadcast unregister notification */
5798 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5799 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5800 * should have already handle it the first time */
5802 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5804 /* We must not have linkwatch events
5805 * pending on unregister. If this
5806 * happens, we simply run the queue
5807 * unscheduled, resulting in a noop
5810 linkwatch_run_queue();
5815 rebroadcast_time = jiffies;
5820 refcnt = netdev_refcnt_read(dev);
5822 if (time_after(jiffies, warning_time + 10 * HZ)) {
5823 printk(KERN_EMERG "unregister_netdevice: "
5824 "waiting for %s to become free. Usage "
5827 warning_time = jiffies;
5836 * register_netdevice(x1);
5837 * register_netdevice(x2);
5839 * unregister_netdevice(y1);
5840 * unregister_netdevice(y2);
5846 * We are invoked by rtnl_unlock().
5847 * This allows us to deal with problems:
5848 * 1) We can delete sysfs objects which invoke hotplug
5849 * without deadlocking with linkwatch via keventd.
5850 * 2) Since we run with the RTNL semaphore not held, we can sleep
5851 * safely in order to wait for the netdev refcnt to drop to zero.
5853 * We must not return until all unregister events added during
5854 * the interval the lock was held have been completed.
5856 void netdev_run_todo(void)
5858 struct list_head list;
5860 /* Snapshot list, allow later requests */
5861 list_replace_init(&net_todo_list, &list);
5865 /* Wait for rcu callbacks to finish before attempting to drain
5866 * the device list. This usually avoids a 250ms wait.
5868 if (!list_empty(&list))
5871 while (!list_empty(&list)) {
5872 struct net_device *dev
5873 = list_first_entry(&list, struct net_device, todo_list);
5874 list_del(&dev->todo_list);
5876 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5877 printk(KERN_ERR "network todo '%s' but state %d\n",
5878 dev->name, dev->reg_state);
5883 dev->reg_state = NETREG_UNREGISTERED;
5885 on_each_cpu(flush_backlog, dev, 1);
5887 netdev_wait_allrefs(dev);
5890 BUG_ON(netdev_refcnt_read(dev));
5891 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5892 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5893 WARN_ON(dev->dn_ptr);
5895 if (dev->destructor)
5896 dev->destructor(dev);
5898 /* Free network device */
5899 kobject_put(&dev->dev.kobj);
5903 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5904 * fields in the same order, with only the type differing.
5906 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5907 const struct net_device_stats *netdev_stats)
5909 #if BITS_PER_LONG == 64
5910 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5911 memcpy(stats64, netdev_stats, sizeof(*stats64));
5913 size_t i, n = sizeof(*stats64) / sizeof(u64);
5914 const unsigned long *src = (const unsigned long *)netdev_stats;
5915 u64 *dst = (u64 *)stats64;
5917 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5918 sizeof(*stats64) / sizeof(u64));
5919 for (i = 0; i < n; i++)
5925 * dev_get_stats - get network device statistics
5926 * @dev: device to get statistics from
5927 * @storage: place to store stats
5929 * Get network statistics from device. Return @storage.
5930 * The device driver may provide its own method by setting
5931 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5932 * otherwise the internal statistics structure is used.
5934 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5935 struct rtnl_link_stats64 *storage)
5937 const struct net_device_ops *ops = dev->netdev_ops;
5939 if (ops->ndo_get_stats64) {
5940 memset(storage, 0, sizeof(*storage));
5941 ops->ndo_get_stats64(dev, storage);
5942 } else if (ops->ndo_get_stats) {
5943 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5945 netdev_stats_to_stats64(storage, &dev->stats);
5947 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5950 EXPORT_SYMBOL(dev_get_stats);
5952 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5954 struct netdev_queue *queue = dev_ingress_queue(dev);
5956 #ifdef CONFIG_NET_CLS_ACT
5959 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5962 netdev_init_one_queue(dev, queue, NULL);
5963 queue->qdisc = &noop_qdisc;
5964 queue->qdisc_sleeping = &noop_qdisc;
5965 rcu_assign_pointer(dev->ingress_queue, queue);
5971 * alloc_netdev_mqs - allocate network device
5972 * @sizeof_priv: size of private data to allocate space for
5973 * @name: device name format string
5974 * @setup: callback to initialize device
5975 * @txqs: the number of TX subqueues to allocate
5976 * @rxqs: the number of RX subqueues to allocate
5978 * Allocates a struct net_device with private data area for driver use
5979 * and performs basic initialization. Also allocates subquue structs
5980 * for each queue on the device.
5982 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5983 void (*setup)(struct net_device *),
5984 unsigned int txqs, unsigned int rxqs)
5986 struct net_device *dev;
5988 struct net_device *p;
5990 BUG_ON(strlen(name) >= sizeof(dev->name));
5993 pr_err("alloc_netdev: Unable to allocate device "
5994 "with zero queues.\n");
6000 pr_err("alloc_netdev: Unable to allocate device "
6001 "with zero RX queues.\n");
6006 alloc_size = sizeof(struct net_device);
6008 /* ensure 32-byte alignment of private area */
6009 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6010 alloc_size += sizeof_priv;
6012 /* ensure 32-byte alignment of whole construct */
6013 alloc_size += NETDEV_ALIGN - 1;
6015 p = kzalloc(alloc_size, GFP_KERNEL);
6017 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
6021 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6022 dev->padded = (char *)dev - (char *)p;
6024 dev->pcpu_refcnt = alloc_percpu(int);
6025 if (!dev->pcpu_refcnt)
6028 if (dev_addr_init(dev))
6034 dev_net_set(dev, &init_net);
6036 dev->gso_max_size = GSO_MAX_SIZE;
6038 INIT_LIST_HEAD(&dev->napi_list);
6039 INIT_LIST_HEAD(&dev->unreg_list);
6040 INIT_LIST_HEAD(&dev->link_watch_list);
6041 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6044 dev->num_tx_queues = txqs;
6045 dev->real_num_tx_queues = txqs;
6046 if (netif_alloc_netdev_queues(dev))
6050 dev->num_rx_queues = rxqs;
6051 dev->real_num_rx_queues = rxqs;
6052 if (netif_alloc_rx_queues(dev))
6056 strcpy(dev->name, name);
6057 dev->group = INIT_NETDEV_GROUP;
6065 free_percpu(dev->pcpu_refcnt);
6075 EXPORT_SYMBOL(alloc_netdev_mqs);
6078 * free_netdev - free network device
6081 * This function does the last stage of destroying an allocated device
6082 * interface. The reference to the device object is released.
6083 * If this is the last reference then it will be freed.
6085 void free_netdev(struct net_device *dev)
6087 struct napi_struct *p, *n;
6089 release_net(dev_net(dev));
6096 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6098 /* Flush device addresses */
6099 dev_addr_flush(dev);
6101 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6104 free_percpu(dev->pcpu_refcnt);
6105 dev->pcpu_refcnt = NULL;
6107 /* Compatibility with error handling in drivers */
6108 if (dev->reg_state == NETREG_UNINITIALIZED) {
6109 kfree((char *)dev - dev->padded);
6113 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6114 dev->reg_state = NETREG_RELEASED;
6116 /* will free via device release */
6117 put_device(&dev->dev);
6119 EXPORT_SYMBOL(free_netdev);
6122 * synchronize_net - Synchronize with packet receive processing
6124 * Wait for packets currently being received to be done.
6125 * Does not block later packets from starting.
6127 void synchronize_net(void)
6130 if (rtnl_is_locked())
6131 synchronize_rcu_expedited();
6135 EXPORT_SYMBOL(synchronize_net);
6138 * unregister_netdevice_queue - remove device from the kernel
6142 * This function shuts down a device interface and removes it
6143 * from the kernel tables.
6144 * If head not NULL, device is queued to be unregistered later.
6146 * Callers must hold the rtnl semaphore. You may want
6147 * unregister_netdev() instead of this.
6150 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6155 list_move_tail(&dev->unreg_list, head);
6157 rollback_registered(dev);
6158 /* Finish processing unregister after unlock */
6162 EXPORT_SYMBOL(unregister_netdevice_queue);
6165 * unregister_netdevice_many - unregister many devices
6166 * @head: list of devices
6168 void unregister_netdevice_many(struct list_head *head)
6170 struct net_device *dev;
6172 if (!list_empty(head)) {
6173 rollback_registered_many(head);
6174 list_for_each_entry(dev, head, unreg_list)
6178 EXPORT_SYMBOL(unregister_netdevice_many);
6181 * unregister_netdev - remove device from the kernel
6184 * This function shuts down a device interface and removes it
6185 * from the kernel tables.
6187 * This is just a wrapper for unregister_netdevice that takes
6188 * the rtnl semaphore. In general you want to use this and not
6189 * unregister_netdevice.
6191 void unregister_netdev(struct net_device *dev)
6194 unregister_netdevice(dev);
6197 EXPORT_SYMBOL(unregister_netdev);
6200 * dev_change_net_namespace - move device to different nethost namespace
6202 * @net: network namespace
6203 * @pat: If not NULL name pattern to try if the current device name
6204 * is already taken in the destination network namespace.
6206 * This function shuts down a device interface and moves it
6207 * to a new network namespace. On success 0 is returned, on
6208 * a failure a netagive errno code is returned.
6210 * Callers must hold the rtnl semaphore.
6213 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6219 /* Don't allow namespace local devices to be moved. */
6221 if (dev->features & NETIF_F_NETNS_LOCAL)
6224 /* Ensure the device has been registrered */
6226 if (dev->reg_state != NETREG_REGISTERED)
6229 /* Get out if there is nothing todo */
6231 if (net_eq(dev_net(dev), net))
6234 /* Pick the destination device name, and ensure
6235 * we can use it in the destination network namespace.
6238 if (__dev_get_by_name(net, dev->name)) {
6239 /* We get here if we can't use the current device name */
6242 if (dev_get_valid_name(dev, pat) < 0)
6247 * And now a mini version of register_netdevice unregister_netdevice.
6250 /* If device is running close it first. */
6253 /* And unlink it from device chain */
6255 unlist_netdevice(dev);
6259 /* Shutdown queueing discipline. */
6262 /* Notify protocols, that we are about to destroy
6263 this device. They should clean all the things.
6265 Note that dev->reg_state stays at NETREG_REGISTERED.
6266 This is wanted because this way 8021q and macvlan know
6267 the device is just moving and can keep their slaves up.
6269 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6270 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6271 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6274 * Flush the unicast and multicast chains
6279 /* Actually switch the network namespace */
6280 dev_net_set(dev, net);
6282 /* If there is an ifindex conflict assign a new one */
6283 if (__dev_get_by_index(net, dev->ifindex)) {
6284 int iflink = (dev->iflink == dev->ifindex);
6285 dev->ifindex = dev_new_index(net);
6287 dev->iflink = dev->ifindex;
6290 /* Fixup kobjects */
6291 err = device_rename(&dev->dev, dev->name);
6294 /* Add the device back in the hashes */
6295 list_netdevice(dev);
6297 /* Notify protocols, that a new device appeared. */
6298 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6301 * Prevent userspace races by waiting until the network
6302 * device is fully setup before sending notifications.
6304 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6311 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6313 static int dev_cpu_callback(struct notifier_block *nfb,
6314 unsigned long action,
6317 struct sk_buff **list_skb;
6318 struct sk_buff *skb;
6319 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6320 struct softnet_data *sd, *oldsd;
6322 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6325 local_irq_disable();
6326 cpu = smp_processor_id();
6327 sd = &per_cpu(softnet_data, cpu);
6328 oldsd = &per_cpu(softnet_data, oldcpu);
6330 /* Find end of our completion_queue. */
6331 list_skb = &sd->completion_queue;
6333 list_skb = &(*list_skb)->next;
6334 /* Append completion queue from offline CPU. */
6335 *list_skb = oldsd->completion_queue;
6336 oldsd->completion_queue = NULL;
6338 /* Append output queue from offline CPU. */
6339 if (oldsd->output_queue) {
6340 *sd->output_queue_tailp = oldsd->output_queue;
6341 sd->output_queue_tailp = oldsd->output_queue_tailp;
6342 oldsd->output_queue = NULL;
6343 oldsd->output_queue_tailp = &oldsd->output_queue;
6345 /* Append NAPI poll list from offline CPU. */
6346 if (!list_empty(&oldsd->poll_list)) {
6347 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6348 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6351 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6354 /* Process offline CPU's input_pkt_queue */
6355 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6357 input_queue_head_incr(oldsd);
6359 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6361 input_queue_head_incr(oldsd);
6369 * netdev_increment_features - increment feature set by one
6370 * @all: current feature set
6371 * @one: new feature set
6372 * @mask: mask feature set
6374 * Computes a new feature set after adding a device with feature set
6375 * @one to the master device with current feature set @all. Will not
6376 * enable anything that is off in @mask. Returns the new feature set.
6378 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6380 if (mask & NETIF_F_GEN_CSUM)
6381 mask |= NETIF_F_ALL_CSUM;
6382 mask |= NETIF_F_VLAN_CHALLENGED;
6384 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6385 all &= one | ~NETIF_F_ALL_FOR_ALL;
6387 /* If device needs checksumming, downgrade to it. */
6388 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6389 all &= ~NETIF_F_NO_CSUM;
6391 /* If one device supports hw checksumming, set for all. */
6392 if (all & NETIF_F_GEN_CSUM)
6393 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6397 EXPORT_SYMBOL(netdev_increment_features);
6399 static struct hlist_head *netdev_create_hash(void)
6402 struct hlist_head *hash;
6404 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6406 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6407 INIT_HLIST_HEAD(&hash[i]);
6412 /* Initialize per network namespace state */
6413 static int __net_init netdev_init(struct net *net)
6415 INIT_LIST_HEAD(&net->dev_base_head);
6417 net->dev_name_head = netdev_create_hash();
6418 if (net->dev_name_head == NULL)
6421 net->dev_index_head = netdev_create_hash();
6422 if (net->dev_index_head == NULL)
6428 kfree(net->dev_name_head);
6434 * netdev_drivername - network driver for the device
6435 * @dev: network device
6437 * Determine network driver for device.
6439 const char *netdev_drivername(const struct net_device *dev)
6441 const struct device_driver *driver;
6442 const struct device *parent;
6443 const char *empty = "";
6445 parent = dev->dev.parent;
6449 driver = parent->driver;
6450 if (driver && driver->name)
6451 return driver->name;
6455 int __netdev_printk(const char *level, const struct net_device *dev,
6456 struct va_format *vaf)
6460 if (dev && dev->dev.parent)
6461 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6462 netdev_name(dev), vaf);
6464 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6466 r = printk("%s(NULL net_device): %pV", level, vaf);
6470 EXPORT_SYMBOL(__netdev_printk);
6472 int netdev_printk(const char *level, const struct net_device *dev,
6473 const char *format, ...)
6475 struct va_format vaf;
6479 va_start(args, format);
6484 r = __netdev_printk(level, dev, &vaf);
6489 EXPORT_SYMBOL(netdev_printk);
6491 #define define_netdev_printk_level(func, level) \
6492 int func(const struct net_device *dev, const char *fmt, ...) \
6495 struct va_format vaf; \
6498 va_start(args, fmt); \
6503 r = __netdev_printk(level, dev, &vaf); \
6508 EXPORT_SYMBOL(func);
6510 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6511 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6512 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6513 define_netdev_printk_level(netdev_err, KERN_ERR);
6514 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6515 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6516 define_netdev_printk_level(netdev_info, KERN_INFO);
6518 static void __net_exit netdev_exit(struct net *net)
6520 kfree(net->dev_name_head);
6521 kfree(net->dev_index_head);
6524 static struct pernet_operations __net_initdata netdev_net_ops = {
6525 .init = netdev_init,
6526 .exit = netdev_exit,
6529 static void __net_exit default_device_exit(struct net *net)
6531 struct net_device *dev, *aux;
6533 * Push all migratable network devices back to the
6534 * initial network namespace
6537 for_each_netdev_safe(net, dev, aux) {
6539 char fb_name[IFNAMSIZ];
6541 /* Ignore unmoveable devices (i.e. loopback) */
6542 if (dev->features & NETIF_F_NETNS_LOCAL)
6545 /* Leave virtual devices for the generic cleanup */
6546 if (dev->rtnl_link_ops)
6549 /* Push remaining network devices to init_net */
6550 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6551 err = dev_change_net_namespace(dev, &init_net, fb_name);
6553 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6554 __func__, dev->name, err);
6561 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6563 /* At exit all network devices most be removed from a network
6564 * namespace. Do this in the reverse order of registration.
6565 * Do this across as many network namespaces as possible to
6566 * improve batching efficiency.
6568 struct net_device *dev;
6570 LIST_HEAD(dev_kill_list);
6573 list_for_each_entry(net, net_list, exit_list) {
6574 for_each_netdev_reverse(net, dev) {
6575 if (dev->rtnl_link_ops)
6576 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6578 unregister_netdevice_queue(dev, &dev_kill_list);
6581 unregister_netdevice_many(&dev_kill_list);
6582 list_del(&dev_kill_list);
6586 static struct pernet_operations __net_initdata default_device_ops = {
6587 .exit = default_device_exit,
6588 .exit_batch = default_device_exit_batch,
6592 * Initialize the DEV module. At boot time this walks the device list and
6593 * unhooks any devices that fail to initialise (normally hardware not
6594 * present) and leaves us with a valid list of present and active devices.
6599 * This is called single threaded during boot, so no need
6600 * to take the rtnl semaphore.
6602 static int __init net_dev_init(void)
6604 int i, rc = -ENOMEM;
6606 BUG_ON(!dev_boot_phase);
6608 if (dev_proc_init())
6611 if (netdev_kobject_init())
6614 INIT_LIST_HEAD(&ptype_all);
6615 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6616 INIT_LIST_HEAD(&ptype_base[i]);
6618 if (register_pernet_subsys(&netdev_net_ops))
6622 * Initialise the packet receive queues.
6625 for_each_possible_cpu(i) {
6626 struct softnet_data *sd = &per_cpu(softnet_data, i);
6628 memset(sd, 0, sizeof(*sd));
6629 skb_queue_head_init(&sd->input_pkt_queue);
6630 skb_queue_head_init(&sd->process_queue);
6631 sd->completion_queue = NULL;
6632 INIT_LIST_HEAD(&sd->poll_list);
6633 sd->output_queue = NULL;
6634 sd->output_queue_tailp = &sd->output_queue;
6636 sd->csd.func = rps_trigger_softirq;
6642 sd->backlog.poll = process_backlog;
6643 sd->backlog.weight = weight_p;
6644 sd->backlog.gro_list = NULL;
6645 sd->backlog.gro_count = 0;
6650 /* The loopback device is special if any other network devices
6651 * is present in a network namespace the loopback device must
6652 * be present. Since we now dynamically allocate and free the
6653 * loopback device ensure this invariant is maintained by
6654 * keeping the loopback device as the first device on the
6655 * list of network devices. Ensuring the loopback devices
6656 * is the first device that appears and the last network device
6659 if (register_pernet_device(&loopback_net_ops))
6662 if (register_pernet_device(&default_device_ops))
6665 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6666 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6668 hotcpu_notifier(dev_cpu_callback, 0);
6676 subsys_initcall(net_dev_init);
6678 static int __init initialize_hashrnd(void)
6680 get_random_bytes(&hashrnd, sizeof(hashrnd));
6684 late_initcall_sync(initialize_hashrnd);