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>
140 #include <linux/jump_label.h>
142 #include "net-sysfs.h"
144 /* Instead of increasing this, you should create a hash table. */
145 #define MAX_GRO_SKBS 8
147 /* This should be increased if a protocol with a bigger head is added. */
148 #define GRO_MAX_HEAD (MAX_HEADER + 128)
151 * The list of packet types we will receive (as opposed to discard)
152 * and the routines to invoke.
154 * Why 16. Because with 16 the only overlap we get on a hash of the
155 * low nibble of the protocol value is RARP/SNAP/X.25.
157 * NOTE: That is no longer true with the addition of VLAN tags. Not
158 * sure which should go first, but I bet it won't make much
159 * difference if we are running VLANs. The good news is that
160 * this protocol won't be in the list unless compiled in, so
161 * the average user (w/out VLANs) will not be adversely affected.
178 #define PTYPE_HASH_SIZE (16)
179 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
181 static DEFINE_SPINLOCK(ptype_lock);
182 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
183 static struct list_head ptype_all __read_mostly; /* Taps */
186 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
189 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
191 * Writers must hold the rtnl semaphore while they loop through the
192 * dev_base_head list, and hold dev_base_lock for writing when they do the
193 * actual updates. This allows pure readers to access the list even
194 * while a writer is preparing to update it.
196 * To put it another way, dev_base_lock is held for writing only to
197 * protect against pure readers; the rtnl semaphore provides the
198 * protection against other writers.
200 * See, for example usages, register_netdevice() and
201 * unregister_netdevice(), which must be called with the rtnl
204 DEFINE_RWLOCK(dev_base_lock);
205 EXPORT_SYMBOL(dev_base_lock);
207 static inline void dev_base_seq_inc(struct net *net)
209 while (++net->dev_base_seq == 0);
212 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
214 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
215 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
218 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
220 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
223 static inline void rps_lock(struct softnet_data *sd)
226 spin_lock(&sd->input_pkt_queue.lock);
230 static inline void rps_unlock(struct softnet_data *sd)
233 spin_unlock(&sd->input_pkt_queue.lock);
237 /* Device list insertion */
238 static int list_netdevice(struct net_device *dev)
240 struct net *net = dev_net(dev);
244 write_lock_bh(&dev_base_lock);
245 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
246 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
247 hlist_add_head_rcu(&dev->index_hlist,
248 dev_index_hash(net, dev->ifindex));
249 write_unlock_bh(&dev_base_lock);
251 dev_base_seq_inc(net);
256 /* Device list removal
257 * caller must respect a RCU grace period before freeing/reusing dev
259 static void unlist_netdevice(struct net_device *dev)
263 /* Unlink dev from the device chain */
264 write_lock_bh(&dev_base_lock);
265 list_del_rcu(&dev->dev_list);
266 hlist_del_rcu(&dev->name_hlist);
267 hlist_del_rcu(&dev->index_hlist);
268 write_unlock_bh(&dev_base_lock);
270 dev_base_seq_inc(dev_net(dev));
277 static RAW_NOTIFIER_HEAD(netdev_chain);
280 * Device drivers call our routines to queue packets here. We empty the
281 * queue in the local softnet handler.
284 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
285 EXPORT_PER_CPU_SYMBOL(softnet_data);
287 #ifdef CONFIG_LOCKDEP
289 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
290 * according to dev->type
292 static const unsigned short netdev_lock_type[] =
293 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
294 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
295 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
296 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
297 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
298 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
299 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
300 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
301 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
302 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
303 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
304 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
305 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
306 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
307 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
308 ARPHRD_VOID, ARPHRD_NONE};
310 static const char *const netdev_lock_name[] =
311 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
312 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
313 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
314 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
315 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
316 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
317 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
318 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
319 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
320 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
321 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
322 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
323 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
324 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
325 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
326 "_xmit_VOID", "_xmit_NONE"};
328 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
329 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
331 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
335 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
336 if (netdev_lock_type[i] == dev_type)
338 /* the last key is used by default */
339 return ARRAY_SIZE(netdev_lock_type) - 1;
342 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
343 unsigned short dev_type)
347 i = netdev_lock_pos(dev_type);
348 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
349 netdev_lock_name[i]);
352 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
356 i = netdev_lock_pos(dev->type);
357 lockdep_set_class_and_name(&dev->addr_list_lock,
358 &netdev_addr_lock_key[i],
359 netdev_lock_name[i]);
362 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
363 unsigned short dev_type)
366 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
371 /*******************************************************************************
373 Protocol management and registration routines
375 *******************************************************************************/
378 * Add a protocol ID to the list. Now that the input handler is
379 * smarter we can dispense with all the messy stuff that used to be
382 * BEWARE!!! Protocol handlers, mangling input packets,
383 * MUST BE last in hash buckets and checking protocol handlers
384 * MUST start from promiscuous ptype_all chain in net_bh.
385 * It is true now, do not change it.
386 * Explanation follows: if protocol handler, mangling packet, will
387 * be the first on list, it is not able to sense, that packet
388 * is cloned and should be copied-on-write, so that it will
389 * change it and subsequent readers will get broken packet.
393 static inline struct list_head *ptype_head(const struct packet_type *pt)
395 if (pt->type == htons(ETH_P_ALL))
398 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
402 * dev_add_pack - add packet handler
403 * @pt: packet type declaration
405 * Add a protocol handler to the networking stack. The passed &packet_type
406 * is linked into kernel lists and may not be freed until it has been
407 * removed from the kernel lists.
409 * This call does not sleep therefore it can not
410 * guarantee all CPU's that are in middle of receiving packets
411 * will see the new packet type (until the next received packet).
414 void dev_add_pack(struct packet_type *pt)
416 struct list_head *head = ptype_head(pt);
418 spin_lock(&ptype_lock);
419 list_add_rcu(&pt->list, head);
420 spin_unlock(&ptype_lock);
422 EXPORT_SYMBOL(dev_add_pack);
425 * __dev_remove_pack - remove packet handler
426 * @pt: packet type declaration
428 * Remove a protocol handler that was previously added to the kernel
429 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
430 * from the kernel lists and can be freed or reused once this function
433 * The packet type might still be in use by receivers
434 * and must not be freed until after all the CPU's have gone
435 * through a quiescent state.
437 void __dev_remove_pack(struct packet_type *pt)
439 struct list_head *head = ptype_head(pt);
440 struct packet_type *pt1;
442 spin_lock(&ptype_lock);
444 list_for_each_entry(pt1, head, list) {
446 list_del_rcu(&pt->list);
451 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
453 spin_unlock(&ptype_lock);
455 EXPORT_SYMBOL(__dev_remove_pack);
458 * dev_remove_pack - remove packet handler
459 * @pt: packet type declaration
461 * Remove a protocol handler that was previously added to the kernel
462 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
463 * from the kernel lists and can be freed or reused once this function
466 * This call sleeps to guarantee that no CPU is looking at the packet
469 void dev_remove_pack(struct packet_type *pt)
471 __dev_remove_pack(pt);
475 EXPORT_SYMBOL(dev_remove_pack);
477 /******************************************************************************
479 Device Boot-time Settings Routines
481 *******************************************************************************/
483 /* Boot time configuration table */
484 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
487 * netdev_boot_setup_add - add new setup entry
488 * @name: name of the device
489 * @map: configured settings for the device
491 * Adds new setup entry to the dev_boot_setup list. The function
492 * returns 0 on error and 1 on success. This is a generic routine to
495 static int netdev_boot_setup_add(char *name, struct ifmap *map)
497 struct netdev_boot_setup *s;
501 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
502 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
503 memset(s[i].name, 0, sizeof(s[i].name));
504 strlcpy(s[i].name, name, IFNAMSIZ);
505 memcpy(&s[i].map, map, sizeof(s[i].map));
510 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
514 * netdev_boot_setup_check - check boot time settings
515 * @dev: the netdevice
517 * Check boot time settings for the device.
518 * The found settings are set for the device to be used
519 * later in the device probing.
520 * Returns 0 if no settings found, 1 if they are.
522 int netdev_boot_setup_check(struct net_device *dev)
524 struct netdev_boot_setup *s = dev_boot_setup;
527 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
528 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
529 !strcmp(dev->name, s[i].name)) {
530 dev->irq = s[i].map.irq;
531 dev->base_addr = s[i].map.base_addr;
532 dev->mem_start = s[i].map.mem_start;
533 dev->mem_end = s[i].map.mem_end;
539 EXPORT_SYMBOL(netdev_boot_setup_check);
543 * netdev_boot_base - get address from boot time settings
544 * @prefix: prefix for network device
545 * @unit: id for network device
547 * Check boot time settings for the base address of device.
548 * The found settings are set for the device to be used
549 * later in the device probing.
550 * Returns 0 if no settings found.
552 unsigned long netdev_boot_base(const char *prefix, int unit)
554 const struct netdev_boot_setup *s = dev_boot_setup;
558 sprintf(name, "%s%d", prefix, unit);
561 * If device already registered then return base of 1
562 * to indicate not to probe for this interface
564 if (__dev_get_by_name(&init_net, name))
567 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
568 if (!strcmp(name, s[i].name))
569 return s[i].map.base_addr;
574 * Saves at boot time configured settings for any netdevice.
576 int __init netdev_boot_setup(char *str)
581 str = get_options(str, ARRAY_SIZE(ints), ints);
586 memset(&map, 0, sizeof(map));
590 map.base_addr = ints[2];
592 map.mem_start = ints[3];
594 map.mem_end = ints[4];
596 /* Add new entry to the list */
597 return netdev_boot_setup_add(str, &map);
600 __setup("netdev=", netdev_boot_setup);
602 /*******************************************************************************
604 Device Interface Subroutines
606 *******************************************************************************/
609 * __dev_get_by_name - find a device by its name
610 * @net: the applicable net namespace
611 * @name: name to find
613 * Find an interface by name. Must be called under RTNL semaphore
614 * or @dev_base_lock. If the name is found a pointer to the device
615 * is returned. If the name is not found then %NULL is returned. The
616 * reference counters are not incremented so the caller must be
617 * careful with locks.
620 struct net_device *__dev_get_by_name(struct net *net, const char *name)
622 struct hlist_node *p;
623 struct net_device *dev;
624 struct hlist_head *head = dev_name_hash(net, name);
626 hlist_for_each_entry(dev, p, head, name_hlist)
627 if (!strncmp(dev->name, name, IFNAMSIZ))
632 EXPORT_SYMBOL(__dev_get_by_name);
635 * dev_get_by_name_rcu - find a device by its name
636 * @net: the applicable net namespace
637 * @name: name to find
639 * Find an interface by name.
640 * If the name is found a pointer to the device is returned.
641 * If the name is not found then %NULL is returned.
642 * The reference counters are not incremented so the caller must be
643 * careful with locks. The caller must hold RCU lock.
646 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
648 struct hlist_node *p;
649 struct net_device *dev;
650 struct hlist_head *head = dev_name_hash(net, name);
652 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
653 if (!strncmp(dev->name, name, IFNAMSIZ))
658 EXPORT_SYMBOL(dev_get_by_name_rcu);
661 * dev_get_by_name - find a device by its name
662 * @net: the applicable net namespace
663 * @name: name to find
665 * Find an interface by name. This can be called from any
666 * context and does its own locking. The returned handle has
667 * the usage count incremented and the caller must use dev_put() to
668 * release it when it is no longer needed. %NULL is returned if no
669 * matching device is found.
672 struct net_device *dev_get_by_name(struct net *net, const char *name)
674 struct net_device *dev;
677 dev = dev_get_by_name_rcu(net, name);
683 EXPORT_SYMBOL(dev_get_by_name);
686 * __dev_get_by_index - find a device by its ifindex
687 * @net: the applicable net namespace
688 * @ifindex: index of device
690 * Search for an interface by index. Returns %NULL if the device
691 * is not found or a pointer to the device. The device has not
692 * had its reference counter increased so the caller must be careful
693 * about locking. The caller must hold either the RTNL semaphore
697 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
699 struct hlist_node *p;
700 struct net_device *dev;
701 struct hlist_head *head = dev_index_hash(net, ifindex);
703 hlist_for_each_entry(dev, p, head, index_hlist)
704 if (dev->ifindex == ifindex)
709 EXPORT_SYMBOL(__dev_get_by_index);
712 * dev_get_by_index_rcu - find a device by its ifindex
713 * @net: the applicable net namespace
714 * @ifindex: index of device
716 * Search for an interface by index. Returns %NULL if the device
717 * is not found or a pointer to the device. The device has not
718 * had its reference counter increased so the caller must be careful
719 * about locking. The caller must hold RCU lock.
722 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
724 struct hlist_node *p;
725 struct net_device *dev;
726 struct hlist_head *head = dev_index_hash(net, ifindex);
728 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
729 if (dev->ifindex == ifindex)
734 EXPORT_SYMBOL(dev_get_by_index_rcu);
738 * dev_get_by_index - find a device by its ifindex
739 * @net: the applicable net namespace
740 * @ifindex: index of device
742 * Search for an interface by index. Returns NULL if the device
743 * is not found or a pointer to the device. The device returned has
744 * had a reference added and the pointer is safe until the user calls
745 * dev_put to indicate they have finished with it.
748 struct net_device *dev_get_by_index(struct net *net, int ifindex)
750 struct net_device *dev;
753 dev = dev_get_by_index_rcu(net, ifindex);
759 EXPORT_SYMBOL(dev_get_by_index);
762 * dev_getbyhwaddr_rcu - find a device by its hardware address
763 * @net: the applicable net namespace
764 * @type: media type of device
765 * @ha: hardware address
767 * Search for an interface by MAC address. Returns NULL if the device
768 * is not found or a pointer to the device.
769 * The caller must hold RCU or RTNL.
770 * The returned device has not had its ref count increased
771 * and the caller must therefore be careful about locking
775 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
778 struct net_device *dev;
780 for_each_netdev_rcu(net, dev)
781 if (dev->type == type &&
782 !memcmp(dev->dev_addr, ha, dev->addr_len))
787 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
789 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
791 struct net_device *dev;
794 for_each_netdev(net, dev)
795 if (dev->type == type)
800 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
802 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
804 struct net_device *dev, *ret = NULL;
807 for_each_netdev_rcu(net, dev)
808 if (dev->type == type) {
816 EXPORT_SYMBOL(dev_getfirstbyhwtype);
819 * dev_get_by_flags_rcu - find any device with given flags
820 * @net: the applicable net namespace
821 * @if_flags: IFF_* values
822 * @mask: bitmask of bits in if_flags to check
824 * Search for any interface with the given flags. Returns NULL if a device
825 * is not found or a pointer to the device. Must be called inside
826 * rcu_read_lock(), and result refcount is unchanged.
829 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
832 struct net_device *dev, *ret;
835 for_each_netdev_rcu(net, dev) {
836 if (((dev->flags ^ if_flags) & mask) == 0) {
843 EXPORT_SYMBOL(dev_get_by_flags_rcu);
846 * dev_valid_name - check if name is okay for network device
849 * Network device names need to be valid file names to
850 * to allow sysfs to work. We also disallow any kind of
853 int dev_valid_name(const char *name)
857 if (strlen(name) >= IFNAMSIZ)
859 if (!strcmp(name, ".") || !strcmp(name, ".."))
863 if (*name == '/' || isspace(*name))
869 EXPORT_SYMBOL(dev_valid_name);
872 * __dev_alloc_name - allocate a name for a device
873 * @net: network namespace to allocate the device name in
874 * @name: name format string
875 * @buf: scratch buffer and result name string
877 * Passed a format string - eg "lt%d" it will try and find a suitable
878 * id. It scans list of devices to build up a free map, then chooses
879 * the first empty slot. The caller must hold the dev_base or rtnl lock
880 * while allocating the name and adding the device in order to avoid
882 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
883 * Returns the number of the unit assigned or a negative errno code.
886 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
890 const int max_netdevices = 8*PAGE_SIZE;
891 unsigned long *inuse;
892 struct net_device *d;
894 p = strnchr(name, IFNAMSIZ-1, '%');
897 * Verify the string as this thing may have come from
898 * the user. There must be either one "%d" and no other "%"
901 if (p[1] != 'd' || strchr(p + 2, '%'))
904 /* Use one page as a bit array of possible slots */
905 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
909 for_each_netdev(net, d) {
910 if (!sscanf(d->name, name, &i))
912 if (i < 0 || i >= max_netdevices)
915 /* avoid cases where sscanf is not exact inverse of printf */
916 snprintf(buf, IFNAMSIZ, name, i);
917 if (!strncmp(buf, d->name, IFNAMSIZ))
921 i = find_first_zero_bit(inuse, max_netdevices);
922 free_page((unsigned long) inuse);
926 snprintf(buf, IFNAMSIZ, name, i);
927 if (!__dev_get_by_name(net, buf))
930 /* It is possible to run out of possible slots
931 * when the name is long and there isn't enough space left
932 * for the digits, or if all bits are used.
938 * dev_alloc_name - allocate a name for a device
940 * @name: name format string
942 * Passed a format string - eg "lt%d" it will try and find a suitable
943 * id. It scans list of devices to build up a free map, then chooses
944 * the first empty slot. The caller must hold the dev_base or rtnl lock
945 * while allocating the name and adding the device in order to avoid
947 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
948 * Returns the number of the unit assigned or a negative errno code.
951 int dev_alloc_name(struct net_device *dev, const char *name)
957 BUG_ON(!dev_net(dev));
959 ret = __dev_alloc_name(net, name, buf);
961 strlcpy(dev->name, buf, IFNAMSIZ);
964 EXPORT_SYMBOL(dev_alloc_name);
966 static int dev_get_valid_name(struct net_device *dev, const char *name)
970 BUG_ON(!dev_net(dev));
973 if (!dev_valid_name(name))
976 if (strchr(name, '%'))
977 return dev_alloc_name(dev, name);
978 else if (__dev_get_by_name(net, name))
980 else if (dev->name != name)
981 strlcpy(dev->name, name, IFNAMSIZ);
987 * dev_change_name - change name of a device
989 * @newname: name (or format string) must be at least IFNAMSIZ
991 * Change name of a device, can pass format strings "eth%d".
994 int dev_change_name(struct net_device *dev, const char *newname)
996 char oldname[IFNAMSIZ];
1002 BUG_ON(!dev_net(dev));
1005 if (dev->flags & IFF_UP)
1008 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1011 memcpy(oldname, dev->name, IFNAMSIZ);
1013 err = dev_get_valid_name(dev, newname);
1018 ret = device_rename(&dev->dev, dev->name);
1020 memcpy(dev->name, oldname, IFNAMSIZ);
1024 write_lock_bh(&dev_base_lock);
1025 hlist_del_rcu(&dev->name_hlist);
1026 write_unlock_bh(&dev_base_lock);
1030 write_lock_bh(&dev_base_lock);
1031 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1032 write_unlock_bh(&dev_base_lock);
1034 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1035 ret = notifier_to_errno(ret);
1038 /* err >= 0 after dev_alloc_name() or stores the first errno */
1041 memcpy(dev->name, oldname, IFNAMSIZ);
1045 "%s: name change rollback failed: %d.\n",
1054 * dev_set_alias - change ifalias of a device
1056 * @alias: name up to IFALIASZ
1057 * @len: limit of bytes to copy from info
1059 * Set ifalias for a device,
1061 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1065 if (len >= IFALIASZ)
1070 kfree(dev->ifalias);
1071 dev->ifalias = NULL;
1076 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1080 strlcpy(dev->ifalias, alias, len+1);
1086 * netdev_features_change - device changes features
1087 * @dev: device to cause notification
1089 * Called to indicate a device has changed features.
1091 void netdev_features_change(struct net_device *dev)
1093 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1095 EXPORT_SYMBOL(netdev_features_change);
1098 * netdev_state_change - device changes state
1099 * @dev: device to cause notification
1101 * Called to indicate a device has changed state. This function calls
1102 * the notifier chains for netdev_chain and sends a NEWLINK message
1103 * to the routing socket.
1105 void netdev_state_change(struct net_device *dev)
1107 if (dev->flags & IFF_UP) {
1108 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1109 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1112 EXPORT_SYMBOL(netdev_state_change);
1114 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1116 return call_netdevice_notifiers(event, dev);
1118 EXPORT_SYMBOL(netdev_bonding_change);
1121 * dev_load - load a network module
1122 * @net: the applicable net namespace
1123 * @name: name of interface
1125 * If a network interface is not present and the process has suitable
1126 * privileges this function loads the module. If module loading is not
1127 * available in this kernel then it becomes a nop.
1130 void dev_load(struct net *net, const char *name)
1132 struct net_device *dev;
1136 dev = dev_get_by_name_rcu(net, name);
1140 if (no_module && capable(CAP_NET_ADMIN))
1141 no_module = request_module("netdev-%s", name);
1142 if (no_module && capable(CAP_SYS_MODULE)) {
1143 if (!request_module("%s", name))
1144 pr_err("Loading kernel module for a network device "
1145 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1149 EXPORT_SYMBOL(dev_load);
1151 static int __dev_open(struct net_device *dev)
1153 const struct net_device_ops *ops = dev->netdev_ops;
1158 if (!netif_device_present(dev))
1161 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1162 ret = notifier_to_errno(ret);
1166 set_bit(__LINK_STATE_START, &dev->state);
1168 if (ops->ndo_validate_addr)
1169 ret = ops->ndo_validate_addr(dev);
1171 if (!ret && ops->ndo_open)
1172 ret = ops->ndo_open(dev);
1175 clear_bit(__LINK_STATE_START, &dev->state);
1177 dev->flags |= IFF_UP;
1178 net_dmaengine_get();
1179 dev_set_rx_mode(dev);
1187 * dev_open - prepare an interface for use.
1188 * @dev: device to open
1190 * Takes a device from down to up state. The device's private open
1191 * function is invoked and then the multicast lists are loaded. Finally
1192 * the device is moved into the up state and a %NETDEV_UP message is
1193 * sent to the netdev notifier chain.
1195 * Calling this function on an active interface is a nop. On a failure
1196 * a negative errno code is returned.
1198 int dev_open(struct net_device *dev)
1202 if (dev->flags & IFF_UP)
1205 ret = __dev_open(dev);
1209 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1210 call_netdevice_notifiers(NETDEV_UP, dev);
1214 EXPORT_SYMBOL(dev_open);
1216 static int __dev_close_many(struct list_head *head)
1218 struct net_device *dev;
1223 list_for_each_entry(dev, head, unreg_list) {
1224 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1226 clear_bit(__LINK_STATE_START, &dev->state);
1228 /* Synchronize to scheduled poll. We cannot touch poll list, it
1229 * can be even on different cpu. So just clear netif_running().
1231 * dev->stop() will invoke napi_disable() on all of it's
1232 * napi_struct instances on this device.
1234 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1237 dev_deactivate_many(head);
1239 list_for_each_entry(dev, head, unreg_list) {
1240 const struct net_device_ops *ops = dev->netdev_ops;
1243 * Call the device specific close. This cannot fail.
1244 * Only if device is UP
1246 * We allow it to be called even after a DETACH hot-plug
1252 dev->flags &= ~IFF_UP;
1253 net_dmaengine_put();
1259 static int __dev_close(struct net_device *dev)
1264 list_add(&dev->unreg_list, &single);
1265 retval = __dev_close_many(&single);
1270 static int dev_close_many(struct list_head *head)
1272 struct net_device *dev, *tmp;
1273 LIST_HEAD(tmp_list);
1275 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1276 if (!(dev->flags & IFF_UP))
1277 list_move(&dev->unreg_list, &tmp_list);
1279 __dev_close_many(head);
1281 list_for_each_entry(dev, head, unreg_list) {
1282 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1283 call_netdevice_notifiers(NETDEV_DOWN, dev);
1286 /* rollback_registered_many needs the complete original list */
1287 list_splice(&tmp_list, head);
1292 * dev_close - shutdown an interface.
1293 * @dev: device to shutdown
1295 * This function moves an active device into down state. A
1296 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1297 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1300 int dev_close(struct net_device *dev)
1302 if (dev->flags & IFF_UP) {
1305 list_add(&dev->unreg_list, &single);
1306 dev_close_many(&single);
1311 EXPORT_SYMBOL(dev_close);
1315 * dev_disable_lro - disable Large Receive Offload on a device
1318 * Disable Large Receive Offload (LRO) on a net device. Must be
1319 * called under RTNL. This is needed if received packets may be
1320 * forwarded to another interface.
1322 void dev_disable_lro(struct net_device *dev)
1325 * If we're trying to disable lro on a vlan device
1326 * use the underlying physical device instead
1328 if (is_vlan_dev(dev))
1329 dev = vlan_dev_real_dev(dev);
1331 dev->wanted_features &= ~NETIF_F_LRO;
1332 netdev_update_features(dev);
1334 if (unlikely(dev->features & NETIF_F_LRO))
1335 netdev_WARN(dev, "failed to disable LRO!\n");
1337 EXPORT_SYMBOL(dev_disable_lro);
1340 static int dev_boot_phase = 1;
1343 * register_netdevice_notifier - register a network notifier block
1346 * Register a notifier to be called when network device events occur.
1347 * The notifier passed is linked into the kernel structures and must
1348 * not be reused until it has been unregistered. A negative errno code
1349 * is returned on a failure.
1351 * When registered all registration and up events are replayed
1352 * to the new notifier to allow device to have a race free
1353 * view of the network device list.
1356 int register_netdevice_notifier(struct notifier_block *nb)
1358 struct net_device *dev;
1359 struct net_device *last;
1364 err = raw_notifier_chain_register(&netdev_chain, nb);
1370 for_each_netdev(net, dev) {
1371 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1372 err = notifier_to_errno(err);
1376 if (!(dev->flags & IFF_UP))
1379 nb->notifier_call(nb, NETDEV_UP, dev);
1390 for_each_netdev(net, dev) {
1394 if (dev->flags & IFF_UP) {
1395 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1396 nb->notifier_call(nb, NETDEV_DOWN, dev);
1398 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1399 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1403 raw_notifier_chain_unregister(&netdev_chain, nb);
1406 EXPORT_SYMBOL(register_netdevice_notifier);
1409 * unregister_netdevice_notifier - unregister a network notifier block
1412 * Unregister a notifier previously registered by
1413 * register_netdevice_notifier(). The notifier is unlinked into the
1414 * kernel structures and may then be reused. A negative errno code
1415 * is returned on a failure.
1418 int unregister_netdevice_notifier(struct notifier_block *nb)
1423 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1427 EXPORT_SYMBOL(unregister_netdevice_notifier);
1430 * call_netdevice_notifiers - call all network notifier blocks
1431 * @val: value passed unmodified to notifier function
1432 * @dev: net_device pointer passed unmodified to notifier function
1434 * Call all network notifier blocks. Parameters and return value
1435 * are as for raw_notifier_call_chain().
1438 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1441 return raw_notifier_call_chain(&netdev_chain, val, dev);
1443 EXPORT_SYMBOL(call_netdevice_notifiers);
1445 static struct jump_label_key netstamp_needed __read_mostly;
1447 void net_enable_timestamp(void)
1449 jump_label_inc(&netstamp_needed);
1451 EXPORT_SYMBOL(net_enable_timestamp);
1453 void net_disable_timestamp(void)
1455 jump_label_dec(&netstamp_needed);
1457 EXPORT_SYMBOL(net_disable_timestamp);
1459 static inline void net_timestamp_set(struct sk_buff *skb)
1461 skb->tstamp.tv64 = 0;
1462 if (static_branch(&netstamp_needed))
1463 __net_timestamp(skb);
1466 #define net_timestamp_check(COND, SKB) \
1467 if (static_branch(&netstamp_needed)) { \
1468 if ((COND) && !(SKB)->tstamp.tv64) \
1469 __net_timestamp(SKB); \
1472 static int net_hwtstamp_validate(struct ifreq *ifr)
1474 struct hwtstamp_config cfg;
1475 enum hwtstamp_tx_types tx_type;
1476 enum hwtstamp_rx_filters rx_filter;
1477 int tx_type_valid = 0;
1478 int rx_filter_valid = 0;
1480 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1483 if (cfg.flags) /* reserved for future extensions */
1486 tx_type = cfg.tx_type;
1487 rx_filter = cfg.rx_filter;
1490 case HWTSTAMP_TX_OFF:
1491 case HWTSTAMP_TX_ON:
1492 case HWTSTAMP_TX_ONESTEP_SYNC:
1497 switch (rx_filter) {
1498 case HWTSTAMP_FILTER_NONE:
1499 case HWTSTAMP_FILTER_ALL:
1500 case HWTSTAMP_FILTER_SOME:
1501 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1502 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1503 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1504 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1505 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1506 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1507 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1508 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1509 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1510 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1511 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1512 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1513 rx_filter_valid = 1;
1517 if (!tx_type_valid || !rx_filter_valid)
1523 static inline bool is_skb_forwardable(struct net_device *dev,
1524 struct sk_buff *skb)
1528 if (!(dev->flags & IFF_UP))
1531 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1532 if (skb->len <= len)
1535 /* if TSO is enabled, we don't care about the length as the packet
1536 * could be forwarded without being segmented before
1538 if (skb_is_gso(skb))
1545 * dev_forward_skb - loopback an skb to another netif
1547 * @dev: destination network device
1548 * @skb: buffer to forward
1551 * NET_RX_SUCCESS (no congestion)
1552 * NET_RX_DROP (packet was dropped, but freed)
1554 * dev_forward_skb can be used for injecting an skb from the
1555 * start_xmit function of one device into the receive queue
1556 * of another device.
1558 * The receiving device may be in another namespace, so
1559 * we have to clear all information in the skb that could
1560 * impact namespace isolation.
1562 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1564 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1565 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1566 atomic_long_inc(&dev->rx_dropped);
1575 if (unlikely(!is_skb_forwardable(dev, skb))) {
1576 atomic_long_inc(&dev->rx_dropped);
1580 skb_set_dev(skb, dev);
1581 skb->tstamp.tv64 = 0;
1582 skb->pkt_type = PACKET_HOST;
1583 skb->protocol = eth_type_trans(skb, dev);
1584 return netif_rx(skb);
1586 EXPORT_SYMBOL_GPL(dev_forward_skb);
1588 static inline int deliver_skb(struct sk_buff *skb,
1589 struct packet_type *pt_prev,
1590 struct net_device *orig_dev)
1592 atomic_inc(&skb->users);
1593 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1597 * Support routine. Sends outgoing frames to any network
1598 * taps currently in use.
1601 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1603 struct packet_type *ptype;
1604 struct sk_buff *skb2 = NULL;
1605 struct packet_type *pt_prev = NULL;
1608 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1609 /* Never send packets back to the socket
1610 * they originated from - MvS (miquels@drinkel.ow.org)
1612 if ((ptype->dev == dev || !ptype->dev) &&
1613 (ptype->af_packet_priv == NULL ||
1614 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1616 deliver_skb(skb2, pt_prev, skb->dev);
1621 skb2 = skb_clone(skb, GFP_ATOMIC);
1625 net_timestamp_set(skb2);
1627 /* skb->nh should be correctly
1628 set by sender, so that the second statement is
1629 just protection against buggy protocols.
1631 skb_reset_mac_header(skb2);
1633 if (skb_network_header(skb2) < skb2->data ||
1634 skb2->network_header > skb2->tail) {
1635 if (net_ratelimit())
1636 printk(KERN_CRIT "protocol %04x is "
1638 ntohs(skb2->protocol),
1640 skb_reset_network_header(skb2);
1643 skb2->transport_header = skb2->network_header;
1644 skb2->pkt_type = PACKET_OUTGOING;
1649 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1653 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1654 * @dev: Network device
1655 * @txq: number of queues available
1657 * If real_num_tx_queues is changed the tc mappings may no longer be
1658 * valid. To resolve this verify the tc mapping remains valid and if
1659 * not NULL the mapping. With no priorities mapping to this
1660 * offset/count pair it will no longer be used. In the worst case TC0
1661 * is invalid nothing can be done so disable priority mappings. If is
1662 * expected that drivers will fix this mapping if they can before
1663 * calling netif_set_real_num_tx_queues.
1665 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1668 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1670 /* If TC0 is invalidated disable TC mapping */
1671 if (tc->offset + tc->count > txq) {
1672 pr_warning("Number of in use tx queues changed "
1673 "invalidating tc mappings. Priority "
1674 "traffic classification disabled!\n");
1679 /* Invalidated prio to tc mappings set to TC0 */
1680 for (i = 1; i < TC_BITMASK + 1; i++) {
1681 int q = netdev_get_prio_tc_map(dev, i);
1683 tc = &dev->tc_to_txq[q];
1684 if (tc->offset + tc->count > txq) {
1685 pr_warning("Number of in use tx queues "
1686 "changed. Priority %i to tc "
1687 "mapping %i is no longer valid "
1688 "setting map to 0\n",
1690 netdev_set_prio_tc_map(dev, i, 0);
1696 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1697 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1699 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1703 if (txq < 1 || txq > dev->num_tx_queues)
1706 if (dev->reg_state == NETREG_REGISTERED ||
1707 dev->reg_state == NETREG_UNREGISTERING) {
1710 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1716 netif_setup_tc(dev, txq);
1718 if (txq < dev->real_num_tx_queues)
1719 qdisc_reset_all_tx_gt(dev, txq);
1722 dev->real_num_tx_queues = txq;
1725 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1729 * netif_set_real_num_rx_queues - set actual number of RX queues used
1730 * @dev: Network device
1731 * @rxq: Actual number of RX queues
1733 * This must be called either with the rtnl_lock held or before
1734 * registration of the net device. Returns 0 on success, or a
1735 * negative error code. If called before registration, it always
1738 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1742 if (rxq < 1 || rxq > dev->num_rx_queues)
1745 if (dev->reg_state == NETREG_REGISTERED) {
1748 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1754 dev->real_num_rx_queues = rxq;
1757 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1760 static inline void __netif_reschedule(struct Qdisc *q)
1762 struct softnet_data *sd;
1763 unsigned long flags;
1765 local_irq_save(flags);
1766 sd = &__get_cpu_var(softnet_data);
1767 q->next_sched = NULL;
1768 *sd->output_queue_tailp = q;
1769 sd->output_queue_tailp = &q->next_sched;
1770 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1771 local_irq_restore(flags);
1774 void __netif_schedule(struct Qdisc *q)
1776 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1777 __netif_reschedule(q);
1779 EXPORT_SYMBOL(__netif_schedule);
1781 void dev_kfree_skb_irq(struct sk_buff *skb)
1783 if (atomic_dec_and_test(&skb->users)) {
1784 struct softnet_data *sd;
1785 unsigned long flags;
1787 local_irq_save(flags);
1788 sd = &__get_cpu_var(softnet_data);
1789 skb->next = sd->completion_queue;
1790 sd->completion_queue = skb;
1791 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1792 local_irq_restore(flags);
1795 EXPORT_SYMBOL(dev_kfree_skb_irq);
1797 void dev_kfree_skb_any(struct sk_buff *skb)
1799 if (in_irq() || irqs_disabled())
1800 dev_kfree_skb_irq(skb);
1804 EXPORT_SYMBOL(dev_kfree_skb_any);
1808 * netif_device_detach - mark device as removed
1809 * @dev: network device
1811 * Mark device as removed from system and therefore no longer available.
1813 void netif_device_detach(struct net_device *dev)
1815 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1816 netif_running(dev)) {
1817 netif_tx_stop_all_queues(dev);
1820 EXPORT_SYMBOL(netif_device_detach);
1823 * netif_device_attach - mark device as attached
1824 * @dev: network device
1826 * Mark device as attached from system and restart if needed.
1828 void netif_device_attach(struct net_device *dev)
1830 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1831 netif_running(dev)) {
1832 netif_tx_wake_all_queues(dev);
1833 __netdev_watchdog_up(dev);
1836 EXPORT_SYMBOL(netif_device_attach);
1839 * skb_dev_set -- assign a new device to a buffer
1840 * @skb: buffer for the new device
1841 * @dev: network device
1843 * If an skb is owned by a device already, we have to reset
1844 * all data private to the namespace a device belongs to
1845 * before assigning it a new device.
1847 #ifdef CONFIG_NET_NS
1848 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1851 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1854 skb_init_secmark(skb);
1858 skb->ipvs_property = 0;
1859 #ifdef CONFIG_NET_SCHED
1865 EXPORT_SYMBOL(skb_set_dev);
1866 #endif /* CONFIG_NET_NS */
1869 * Invalidate hardware checksum when packet is to be mangled, and
1870 * complete checksum manually on outgoing path.
1872 int skb_checksum_help(struct sk_buff *skb)
1875 int ret = 0, offset;
1877 if (skb->ip_summed == CHECKSUM_COMPLETE)
1878 goto out_set_summed;
1880 if (unlikely(skb_shinfo(skb)->gso_size)) {
1881 /* Let GSO fix up the checksum. */
1882 goto out_set_summed;
1885 offset = skb_checksum_start_offset(skb);
1886 BUG_ON(offset >= skb_headlen(skb));
1887 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1889 offset += skb->csum_offset;
1890 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1892 if (skb_cloned(skb) &&
1893 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1894 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1899 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1901 skb->ip_summed = CHECKSUM_NONE;
1905 EXPORT_SYMBOL(skb_checksum_help);
1908 * skb_gso_segment - Perform segmentation on skb.
1909 * @skb: buffer to segment
1910 * @features: features for the output path (see dev->features)
1912 * This function segments the given skb and returns a list of segments.
1914 * It may return NULL if the skb requires no segmentation. This is
1915 * only possible when GSO is used for verifying header integrity.
1917 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1919 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1920 struct packet_type *ptype;
1921 __be16 type = skb->protocol;
1922 int vlan_depth = ETH_HLEN;
1925 while (type == htons(ETH_P_8021Q)) {
1926 struct vlan_hdr *vh;
1928 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1929 return ERR_PTR(-EINVAL);
1931 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1932 type = vh->h_vlan_encapsulated_proto;
1933 vlan_depth += VLAN_HLEN;
1936 skb_reset_mac_header(skb);
1937 skb->mac_len = skb->network_header - skb->mac_header;
1938 __skb_pull(skb, skb->mac_len);
1940 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1941 struct net_device *dev = skb->dev;
1942 struct ethtool_drvinfo info = {};
1944 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1945 dev->ethtool_ops->get_drvinfo(dev, &info);
1947 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1948 info.driver, dev ? dev->features : 0L,
1949 skb->sk ? skb->sk->sk_route_caps : 0L,
1950 skb->len, skb->data_len, skb->ip_summed);
1952 if (skb_header_cloned(skb) &&
1953 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1954 return ERR_PTR(err);
1958 list_for_each_entry_rcu(ptype,
1959 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1960 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1961 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1962 err = ptype->gso_send_check(skb);
1963 segs = ERR_PTR(err);
1964 if (err || skb_gso_ok(skb, features))
1966 __skb_push(skb, (skb->data -
1967 skb_network_header(skb)));
1969 segs = ptype->gso_segment(skb, features);
1975 __skb_push(skb, skb->data - skb_mac_header(skb));
1979 EXPORT_SYMBOL(skb_gso_segment);
1981 /* Take action when hardware reception checksum errors are detected. */
1983 void netdev_rx_csum_fault(struct net_device *dev)
1985 if (net_ratelimit()) {
1986 printk(KERN_ERR "%s: hw csum failure.\n",
1987 dev ? dev->name : "<unknown>");
1991 EXPORT_SYMBOL(netdev_rx_csum_fault);
1994 /* Actually, we should eliminate this check as soon as we know, that:
1995 * 1. IOMMU is present and allows to map all the memory.
1996 * 2. No high memory really exists on this machine.
1999 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2001 #ifdef CONFIG_HIGHMEM
2003 if (!(dev->features & NETIF_F_HIGHDMA)) {
2004 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2005 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2006 if (PageHighMem(skb_frag_page(frag)))
2011 if (PCI_DMA_BUS_IS_PHYS) {
2012 struct device *pdev = dev->dev.parent;
2016 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2017 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2018 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2019 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2028 void (*destructor)(struct sk_buff *skb);
2031 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2033 static void dev_gso_skb_destructor(struct sk_buff *skb)
2035 struct dev_gso_cb *cb;
2038 struct sk_buff *nskb = skb->next;
2040 skb->next = nskb->next;
2043 } while (skb->next);
2045 cb = DEV_GSO_CB(skb);
2047 cb->destructor(skb);
2051 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2052 * @skb: buffer to segment
2053 * @features: device features as applicable to this skb
2055 * This function segments the given skb and stores the list of segments
2058 static int dev_gso_segment(struct sk_buff *skb, int features)
2060 struct sk_buff *segs;
2062 segs = skb_gso_segment(skb, features);
2064 /* Verifying header integrity only. */
2069 return PTR_ERR(segs);
2072 DEV_GSO_CB(skb)->destructor = skb->destructor;
2073 skb->destructor = dev_gso_skb_destructor;
2079 * Try to orphan skb early, right before transmission by the device.
2080 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2081 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2083 static inline void skb_orphan_try(struct sk_buff *skb)
2085 struct sock *sk = skb->sk;
2087 if (sk && !skb_shinfo(skb)->tx_flags) {
2088 /* skb_tx_hash() wont be able to get sk.
2089 * We copy sk_hash into skb->rxhash
2092 skb->rxhash = sk->sk_hash;
2097 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2099 return ((features & NETIF_F_GEN_CSUM) ||
2100 ((features & NETIF_F_V4_CSUM) &&
2101 protocol == htons(ETH_P_IP)) ||
2102 ((features & NETIF_F_V6_CSUM) &&
2103 protocol == htons(ETH_P_IPV6)) ||
2104 ((features & NETIF_F_FCOE_CRC) &&
2105 protocol == htons(ETH_P_FCOE)));
2108 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2110 if (!can_checksum_protocol(features, protocol)) {
2111 features &= ~NETIF_F_ALL_CSUM;
2112 features &= ~NETIF_F_SG;
2113 } else if (illegal_highdma(skb->dev, skb)) {
2114 features &= ~NETIF_F_SG;
2120 u32 netif_skb_features(struct sk_buff *skb)
2122 __be16 protocol = skb->protocol;
2123 u32 features = skb->dev->features;
2125 if (protocol == htons(ETH_P_8021Q)) {
2126 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2127 protocol = veh->h_vlan_encapsulated_proto;
2128 } else if (!vlan_tx_tag_present(skb)) {
2129 return harmonize_features(skb, protocol, features);
2132 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2134 if (protocol != htons(ETH_P_8021Q)) {
2135 return harmonize_features(skb, protocol, features);
2137 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2138 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2139 return harmonize_features(skb, protocol, features);
2142 EXPORT_SYMBOL(netif_skb_features);
2145 * Returns true if either:
2146 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2147 * 2. skb is fragmented and the device does not support SG, or if
2148 * at least one of fragments is in highmem and device does not
2149 * support DMA from it.
2151 static inline int skb_needs_linearize(struct sk_buff *skb,
2154 return skb_is_nonlinear(skb) &&
2155 ((skb_has_frag_list(skb) &&
2156 !(features & NETIF_F_FRAGLIST)) ||
2157 (skb_shinfo(skb)->nr_frags &&
2158 !(features & NETIF_F_SG)));
2161 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2162 struct netdev_queue *txq)
2164 const struct net_device_ops *ops = dev->netdev_ops;
2165 int rc = NETDEV_TX_OK;
2166 unsigned int skb_len;
2168 if (likely(!skb->next)) {
2172 * If device doesn't need skb->dst, release it right now while
2173 * its hot in this cpu cache
2175 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2178 if (!list_empty(&ptype_all))
2179 dev_queue_xmit_nit(skb, dev);
2181 skb_orphan_try(skb);
2183 features = netif_skb_features(skb);
2185 if (vlan_tx_tag_present(skb) &&
2186 !(features & NETIF_F_HW_VLAN_TX)) {
2187 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2194 if (netif_needs_gso(skb, features)) {
2195 if (unlikely(dev_gso_segment(skb, features)))
2200 if (skb_needs_linearize(skb, features) &&
2201 __skb_linearize(skb))
2204 /* If packet is not checksummed and device does not
2205 * support checksumming for this protocol, complete
2206 * checksumming here.
2208 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2209 skb_set_transport_header(skb,
2210 skb_checksum_start_offset(skb));
2211 if (!(features & NETIF_F_ALL_CSUM) &&
2212 skb_checksum_help(skb))
2218 rc = ops->ndo_start_xmit(skb, dev);
2219 trace_net_dev_xmit(skb, rc, dev, skb_len);
2220 if (rc == NETDEV_TX_OK)
2221 txq_trans_update(txq);
2227 struct sk_buff *nskb = skb->next;
2229 skb->next = nskb->next;
2233 * If device doesn't need nskb->dst, release it right now while
2234 * its hot in this cpu cache
2236 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2239 skb_len = nskb->len;
2240 rc = ops->ndo_start_xmit(nskb, dev);
2241 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2242 if (unlikely(rc != NETDEV_TX_OK)) {
2243 if (rc & ~NETDEV_TX_MASK)
2244 goto out_kfree_gso_skb;
2245 nskb->next = skb->next;
2249 txq_trans_update(txq);
2250 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2251 return NETDEV_TX_BUSY;
2252 } while (skb->next);
2255 if (likely(skb->next == NULL))
2256 skb->destructor = DEV_GSO_CB(skb)->destructor;
2263 static u32 hashrnd __read_mostly;
2266 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2267 * to be used as a distribution range.
2269 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2270 unsigned int num_tx_queues)
2274 u16 qcount = num_tx_queues;
2276 if (skb_rx_queue_recorded(skb)) {
2277 hash = skb_get_rx_queue(skb);
2278 while (unlikely(hash >= num_tx_queues))
2279 hash -= num_tx_queues;
2284 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2285 qoffset = dev->tc_to_txq[tc].offset;
2286 qcount = dev->tc_to_txq[tc].count;
2289 if (skb->sk && skb->sk->sk_hash)
2290 hash = skb->sk->sk_hash;
2292 hash = (__force u16) skb->protocol ^ skb->rxhash;
2293 hash = jhash_1word(hash, hashrnd);
2295 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2297 EXPORT_SYMBOL(__skb_tx_hash);
2299 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2301 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2302 if (net_ratelimit()) {
2303 pr_warning("%s selects TX queue %d, but "
2304 "real number of TX queues is %d\n",
2305 dev->name, queue_index, dev->real_num_tx_queues);
2312 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2315 struct xps_dev_maps *dev_maps;
2316 struct xps_map *map;
2317 int queue_index = -1;
2320 dev_maps = rcu_dereference(dev->xps_maps);
2322 map = rcu_dereference(
2323 dev_maps->cpu_map[raw_smp_processor_id()]);
2326 queue_index = map->queues[0];
2329 if (skb->sk && skb->sk->sk_hash)
2330 hash = skb->sk->sk_hash;
2332 hash = (__force u16) skb->protocol ^
2334 hash = jhash_1word(hash, hashrnd);
2335 queue_index = map->queues[
2336 ((u64)hash * map->len) >> 32];
2338 if (unlikely(queue_index >= dev->real_num_tx_queues))
2350 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2351 struct sk_buff *skb)
2354 const struct net_device_ops *ops = dev->netdev_ops;
2356 if (dev->real_num_tx_queues == 1)
2358 else if (ops->ndo_select_queue) {
2359 queue_index = ops->ndo_select_queue(dev, skb);
2360 queue_index = dev_cap_txqueue(dev, queue_index);
2362 struct sock *sk = skb->sk;
2363 queue_index = sk_tx_queue_get(sk);
2365 if (queue_index < 0 || skb->ooo_okay ||
2366 queue_index >= dev->real_num_tx_queues) {
2367 int old_index = queue_index;
2369 queue_index = get_xps_queue(dev, skb);
2370 if (queue_index < 0)
2371 queue_index = skb_tx_hash(dev, skb);
2373 if (queue_index != old_index && sk) {
2374 struct dst_entry *dst =
2375 rcu_dereference_check(sk->sk_dst_cache, 1);
2377 if (dst && skb_dst(skb) == dst)
2378 sk_tx_queue_set(sk, queue_index);
2383 skb_set_queue_mapping(skb, queue_index);
2384 return netdev_get_tx_queue(dev, queue_index);
2387 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2388 struct net_device *dev,
2389 struct netdev_queue *txq)
2391 spinlock_t *root_lock = qdisc_lock(q);
2395 qdisc_skb_cb(skb)->pkt_len = skb->len;
2396 qdisc_calculate_pkt_len(skb, q);
2398 * Heuristic to force contended enqueues to serialize on a
2399 * separate lock before trying to get qdisc main lock.
2400 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2401 * and dequeue packets faster.
2403 contended = qdisc_is_running(q);
2404 if (unlikely(contended))
2405 spin_lock(&q->busylock);
2407 spin_lock(root_lock);
2408 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2411 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2412 qdisc_run_begin(q)) {
2414 * This is a work-conserving queue; there are no old skbs
2415 * waiting to be sent out; and the qdisc is not running -
2416 * xmit the skb directly.
2418 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2421 qdisc_bstats_update(q, skb);
2423 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2424 if (unlikely(contended)) {
2425 spin_unlock(&q->busylock);
2432 rc = NET_XMIT_SUCCESS;
2435 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2436 if (qdisc_run_begin(q)) {
2437 if (unlikely(contended)) {
2438 spin_unlock(&q->busylock);
2444 spin_unlock(root_lock);
2445 if (unlikely(contended))
2446 spin_unlock(&q->busylock);
2450 static DEFINE_PER_CPU(int, xmit_recursion);
2451 #define RECURSION_LIMIT 10
2454 * dev_queue_xmit - transmit a buffer
2455 * @skb: buffer to transmit
2457 * Queue a buffer for transmission to a network device. The caller must
2458 * have set the device and priority and built the buffer before calling
2459 * this function. The function can be called from an interrupt.
2461 * A negative errno code is returned on a failure. A success does not
2462 * guarantee the frame will be transmitted as it may be dropped due
2463 * to congestion or traffic shaping.
2465 * -----------------------------------------------------------------------------------
2466 * I notice this method can also return errors from the queue disciplines,
2467 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2470 * Regardless of the return value, the skb is consumed, so it is currently
2471 * difficult to retry a send to this method. (You can bump the ref count
2472 * before sending to hold a reference for retry if you are careful.)
2474 * When calling this method, interrupts MUST be enabled. This is because
2475 * the BH enable code must have IRQs enabled so that it will not deadlock.
2478 int dev_queue_xmit(struct sk_buff *skb)
2480 struct net_device *dev = skb->dev;
2481 struct netdev_queue *txq;
2485 /* Disable soft irqs for various locks below. Also
2486 * stops preemption for RCU.
2490 txq = dev_pick_tx(dev, skb);
2491 q = rcu_dereference_bh(txq->qdisc);
2493 #ifdef CONFIG_NET_CLS_ACT
2494 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2496 trace_net_dev_queue(skb);
2498 rc = __dev_xmit_skb(skb, q, dev, txq);
2502 /* The device has no queue. Common case for software devices:
2503 loopback, all the sorts of tunnels...
2505 Really, it is unlikely that netif_tx_lock protection is necessary
2506 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2508 However, it is possible, that they rely on protection
2511 Check this and shot the lock. It is not prone from deadlocks.
2512 Either shot noqueue qdisc, it is even simpler 8)
2514 if (dev->flags & IFF_UP) {
2515 int cpu = smp_processor_id(); /* ok because BHs are off */
2517 if (txq->xmit_lock_owner != cpu) {
2519 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2520 goto recursion_alert;
2522 HARD_TX_LOCK(dev, txq, cpu);
2524 if (!netif_tx_queue_stopped(txq)) {
2525 __this_cpu_inc(xmit_recursion);
2526 rc = dev_hard_start_xmit(skb, dev, txq);
2527 __this_cpu_dec(xmit_recursion);
2528 if (dev_xmit_complete(rc)) {
2529 HARD_TX_UNLOCK(dev, txq);
2533 HARD_TX_UNLOCK(dev, txq);
2534 if (net_ratelimit())
2535 printk(KERN_CRIT "Virtual device %s asks to "
2536 "queue packet!\n", dev->name);
2538 /* Recursion is detected! It is possible,
2542 if (net_ratelimit())
2543 printk(KERN_CRIT "Dead loop on virtual device "
2544 "%s, fix it urgently!\n", dev->name);
2549 rcu_read_unlock_bh();
2554 rcu_read_unlock_bh();
2557 EXPORT_SYMBOL(dev_queue_xmit);
2560 /*=======================================================================
2562 =======================================================================*/
2564 int netdev_max_backlog __read_mostly = 1000;
2565 int netdev_tstamp_prequeue __read_mostly = 1;
2566 int netdev_budget __read_mostly = 300;
2567 int weight_p __read_mostly = 64; /* old backlog weight */
2569 /* Called with irq disabled */
2570 static inline void ____napi_schedule(struct softnet_data *sd,
2571 struct napi_struct *napi)
2573 list_add_tail(&napi->poll_list, &sd->poll_list);
2574 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2578 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2579 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2580 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2581 * if hash is a canonical 4-tuple hash over transport ports.
2583 void __skb_get_rxhash(struct sk_buff *skb)
2585 int nhoff, hash = 0, poff;
2586 const struct ipv6hdr *ip6;
2587 const struct iphdr *ip;
2588 const struct vlan_hdr *vlan;
2597 nhoff = skb_network_offset(skb);
2598 proto = skb->protocol;
2602 case __constant_htons(ETH_P_IP):
2604 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2607 ip = (const struct iphdr *) (skb->data + nhoff);
2608 if (ip_is_fragment(ip))
2611 ip_proto = ip->protocol;
2612 addr1 = (__force u32) ip->saddr;
2613 addr2 = (__force u32) ip->daddr;
2614 nhoff += ip->ihl * 4;
2616 case __constant_htons(ETH_P_IPV6):
2618 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2621 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2622 ip_proto = ip6->nexthdr;
2623 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2624 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2627 case __constant_htons(ETH_P_8021Q):
2628 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2630 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2631 proto = vlan->h_vlan_encapsulated_proto;
2632 nhoff += sizeof(*vlan);
2634 case __constant_htons(ETH_P_PPP_SES):
2635 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2637 proto = *((__be16 *) (skb->data + nhoff +
2638 sizeof(struct pppoe_hdr)));
2639 nhoff += PPPOE_SES_HLEN;
2641 case __constant_htons(PPP_IP):
2643 case __constant_htons(PPP_IPV6):
2654 if (pskb_may_pull(skb, nhoff + 16)) {
2655 u8 *h = skb->data + nhoff;
2656 __be16 flags = *(__be16 *)h;
2659 * Only look inside GRE if version zero and no
2662 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2663 proto = *(__be16 *)(h + 2);
2665 if (flags & GRE_CSUM)
2667 if (flags & GRE_KEY)
2669 if (flags & GRE_SEQ)
2682 poff = proto_ports_offset(ip_proto);
2685 if (pskb_may_pull(skb, nhoff + 4)) {
2686 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2687 if (ports.v16[1] < ports.v16[0])
2688 swap(ports.v16[0], ports.v16[1]);
2693 /* get a consistent hash (same value on both flow directions) */
2697 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2704 EXPORT_SYMBOL(__skb_get_rxhash);
2708 /* One global table that all flow-based protocols share. */
2709 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2710 EXPORT_SYMBOL(rps_sock_flow_table);
2712 static struct rps_dev_flow *
2713 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2714 struct rps_dev_flow *rflow, u16 next_cpu)
2716 if (next_cpu != RPS_NO_CPU) {
2717 #ifdef CONFIG_RFS_ACCEL
2718 struct netdev_rx_queue *rxqueue;
2719 struct rps_dev_flow_table *flow_table;
2720 struct rps_dev_flow *old_rflow;
2725 /* Should we steer this flow to a different hardware queue? */
2726 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2727 !(dev->features & NETIF_F_NTUPLE))
2729 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2730 if (rxq_index == skb_get_rx_queue(skb))
2733 rxqueue = dev->_rx + rxq_index;
2734 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2737 flow_id = skb->rxhash & flow_table->mask;
2738 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2739 rxq_index, flow_id);
2743 rflow = &flow_table->flows[flow_id];
2745 if (old_rflow->filter == rflow->filter)
2746 old_rflow->filter = RPS_NO_FILTER;
2750 per_cpu(softnet_data, next_cpu).input_queue_head;
2753 rflow->cpu = next_cpu;
2758 * get_rps_cpu is called from netif_receive_skb and returns the target
2759 * CPU from the RPS map of the receiving queue for a given skb.
2760 * rcu_read_lock must be held on entry.
2762 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2763 struct rps_dev_flow **rflowp)
2765 struct netdev_rx_queue *rxqueue;
2766 struct rps_map *map;
2767 struct rps_dev_flow_table *flow_table;
2768 struct rps_sock_flow_table *sock_flow_table;
2772 if (skb_rx_queue_recorded(skb)) {
2773 u16 index = skb_get_rx_queue(skb);
2774 if (unlikely(index >= dev->real_num_rx_queues)) {
2775 WARN_ONCE(dev->real_num_rx_queues > 1,
2776 "%s received packet on queue %u, but number "
2777 "of RX queues is %u\n",
2778 dev->name, index, dev->real_num_rx_queues);
2781 rxqueue = dev->_rx + index;
2785 map = rcu_dereference(rxqueue->rps_map);
2787 if (map->len == 1 &&
2788 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2789 tcpu = map->cpus[0];
2790 if (cpu_online(tcpu))
2794 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2798 skb_reset_network_header(skb);
2799 if (!skb_get_rxhash(skb))
2802 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2803 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2804 if (flow_table && sock_flow_table) {
2806 struct rps_dev_flow *rflow;
2808 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2811 next_cpu = sock_flow_table->ents[skb->rxhash &
2812 sock_flow_table->mask];
2815 * If the desired CPU (where last recvmsg was done) is
2816 * different from current CPU (one in the rx-queue flow
2817 * table entry), switch if one of the following holds:
2818 * - Current CPU is unset (equal to RPS_NO_CPU).
2819 * - Current CPU is offline.
2820 * - The current CPU's queue tail has advanced beyond the
2821 * last packet that was enqueued using this table entry.
2822 * This guarantees that all previous packets for the flow
2823 * have been dequeued, thus preserving in order delivery.
2825 if (unlikely(tcpu != next_cpu) &&
2826 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2827 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2828 rflow->last_qtail)) >= 0))
2829 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2831 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2839 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2841 if (cpu_online(tcpu)) {
2851 #ifdef CONFIG_RFS_ACCEL
2854 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2855 * @dev: Device on which the filter was set
2856 * @rxq_index: RX queue index
2857 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2858 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2860 * Drivers that implement ndo_rx_flow_steer() should periodically call
2861 * this function for each installed filter and remove the filters for
2862 * which it returns %true.
2864 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2865 u32 flow_id, u16 filter_id)
2867 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2868 struct rps_dev_flow_table *flow_table;
2869 struct rps_dev_flow *rflow;
2874 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2875 if (flow_table && flow_id <= flow_table->mask) {
2876 rflow = &flow_table->flows[flow_id];
2877 cpu = ACCESS_ONCE(rflow->cpu);
2878 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2879 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2880 rflow->last_qtail) <
2881 (int)(10 * flow_table->mask)))
2887 EXPORT_SYMBOL(rps_may_expire_flow);
2889 #endif /* CONFIG_RFS_ACCEL */
2891 /* Called from hardirq (IPI) context */
2892 static void rps_trigger_softirq(void *data)
2894 struct softnet_data *sd = data;
2896 ____napi_schedule(sd, &sd->backlog);
2900 #endif /* CONFIG_RPS */
2903 * Check if this softnet_data structure is another cpu one
2904 * If yes, queue it to our IPI list and return 1
2907 static int rps_ipi_queued(struct softnet_data *sd)
2910 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2913 sd->rps_ipi_next = mysd->rps_ipi_list;
2914 mysd->rps_ipi_list = sd;
2916 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2919 #endif /* CONFIG_RPS */
2924 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2925 * queue (may be a remote CPU queue).
2927 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2928 unsigned int *qtail)
2930 struct softnet_data *sd;
2931 unsigned long flags;
2933 sd = &per_cpu(softnet_data, cpu);
2935 local_irq_save(flags);
2938 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2939 if (skb_queue_len(&sd->input_pkt_queue)) {
2941 __skb_queue_tail(&sd->input_pkt_queue, skb);
2942 input_queue_tail_incr_save(sd, qtail);
2944 local_irq_restore(flags);
2945 return NET_RX_SUCCESS;
2948 /* Schedule NAPI for backlog device
2949 * We can use non atomic operation since we own the queue lock
2951 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2952 if (!rps_ipi_queued(sd))
2953 ____napi_schedule(sd, &sd->backlog);
2961 local_irq_restore(flags);
2963 atomic_long_inc(&skb->dev->rx_dropped);
2969 * netif_rx - post buffer to the network code
2970 * @skb: buffer to post
2972 * This function receives a packet from a device driver and queues it for
2973 * the upper (protocol) levels to process. It always succeeds. The buffer
2974 * may be dropped during processing for congestion control or by the
2978 * NET_RX_SUCCESS (no congestion)
2979 * NET_RX_DROP (packet was dropped)
2983 int netif_rx(struct sk_buff *skb)
2987 /* if netpoll wants it, pretend we never saw it */
2988 if (netpoll_rx(skb))
2991 net_timestamp_check(netdev_tstamp_prequeue, skb);
2993 trace_netif_rx(skb);
2996 struct rps_dev_flow voidflow, *rflow = &voidflow;
3002 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3004 cpu = smp_processor_id();
3006 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3014 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3020 EXPORT_SYMBOL(netif_rx);
3022 int netif_rx_ni(struct sk_buff *skb)
3027 err = netif_rx(skb);
3028 if (local_softirq_pending())
3034 EXPORT_SYMBOL(netif_rx_ni);
3036 static void net_tx_action(struct softirq_action *h)
3038 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3040 if (sd->completion_queue) {
3041 struct sk_buff *clist;
3043 local_irq_disable();
3044 clist = sd->completion_queue;
3045 sd->completion_queue = NULL;
3049 struct sk_buff *skb = clist;
3050 clist = clist->next;
3052 WARN_ON(atomic_read(&skb->users));
3053 trace_kfree_skb(skb, net_tx_action);
3058 if (sd->output_queue) {
3061 local_irq_disable();
3062 head = sd->output_queue;
3063 sd->output_queue = NULL;
3064 sd->output_queue_tailp = &sd->output_queue;
3068 struct Qdisc *q = head;
3069 spinlock_t *root_lock;
3071 head = head->next_sched;
3073 root_lock = qdisc_lock(q);
3074 if (spin_trylock(root_lock)) {
3075 smp_mb__before_clear_bit();
3076 clear_bit(__QDISC_STATE_SCHED,
3079 spin_unlock(root_lock);
3081 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3083 __netif_reschedule(q);
3085 smp_mb__before_clear_bit();
3086 clear_bit(__QDISC_STATE_SCHED,
3094 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3095 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3096 /* This hook is defined here for ATM LANE */
3097 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3098 unsigned char *addr) __read_mostly;
3099 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3102 #ifdef CONFIG_NET_CLS_ACT
3103 /* TODO: Maybe we should just force sch_ingress to be compiled in
3104 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3105 * a compare and 2 stores extra right now if we dont have it on
3106 * but have CONFIG_NET_CLS_ACT
3107 * NOTE: This doesn't stop any functionality; if you dont have
3108 * the ingress scheduler, you just can't add policies on ingress.
3111 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3113 struct net_device *dev = skb->dev;
3114 u32 ttl = G_TC_RTTL(skb->tc_verd);
3115 int result = TC_ACT_OK;
3118 if (unlikely(MAX_RED_LOOP < ttl++)) {
3119 if (net_ratelimit())
3120 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3121 skb->skb_iif, dev->ifindex);
3125 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3126 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3129 if (q != &noop_qdisc) {
3130 spin_lock(qdisc_lock(q));
3131 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3132 result = qdisc_enqueue_root(skb, q);
3133 spin_unlock(qdisc_lock(q));
3139 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3140 struct packet_type **pt_prev,
3141 int *ret, struct net_device *orig_dev)
3143 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3145 if (!rxq || rxq->qdisc == &noop_qdisc)
3149 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3153 switch (ing_filter(skb, rxq)) {
3167 * netdev_rx_handler_register - register receive handler
3168 * @dev: device to register a handler for
3169 * @rx_handler: receive handler to register
3170 * @rx_handler_data: data pointer that is used by rx handler
3172 * Register a receive hander for a device. This handler will then be
3173 * called from __netif_receive_skb. A negative errno code is returned
3176 * The caller must hold the rtnl_mutex.
3178 * For a general description of rx_handler, see enum rx_handler_result.
3180 int netdev_rx_handler_register(struct net_device *dev,
3181 rx_handler_func_t *rx_handler,
3182 void *rx_handler_data)
3186 if (dev->rx_handler)
3189 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3190 rcu_assign_pointer(dev->rx_handler, rx_handler);
3194 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3197 * netdev_rx_handler_unregister - unregister receive handler
3198 * @dev: device to unregister a handler from
3200 * Unregister a receive hander from a device.
3202 * The caller must hold the rtnl_mutex.
3204 void netdev_rx_handler_unregister(struct net_device *dev)
3208 RCU_INIT_POINTER(dev->rx_handler, NULL);
3209 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3211 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3213 static int __netif_receive_skb(struct sk_buff *skb)
3215 struct packet_type *ptype, *pt_prev;
3216 rx_handler_func_t *rx_handler;
3217 struct net_device *orig_dev;
3218 struct net_device *null_or_dev;
3219 bool deliver_exact = false;
3220 int ret = NET_RX_DROP;
3223 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3225 trace_netif_receive_skb(skb);
3227 /* if we've gotten here through NAPI, check netpoll */
3228 if (netpoll_receive_skb(skb))
3232 skb->skb_iif = skb->dev->ifindex;
3233 orig_dev = skb->dev;
3235 skb_reset_network_header(skb);
3236 skb_reset_transport_header(skb);
3237 skb_reset_mac_len(skb);
3245 __this_cpu_inc(softnet_data.processed);
3247 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3248 skb = vlan_untag(skb);
3253 #ifdef CONFIG_NET_CLS_ACT
3254 if (skb->tc_verd & TC_NCLS) {
3255 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3260 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3261 if (!ptype->dev || ptype->dev == skb->dev) {
3263 ret = deliver_skb(skb, pt_prev, orig_dev);
3268 #ifdef CONFIG_NET_CLS_ACT
3269 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3275 rx_handler = rcu_dereference(skb->dev->rx_handler);
3276 if (vlan_tx_tag_present(skb)) {
3278 ret = deliver_skb(skb, pt_prev, orig_dev);
3281 if (vlan_do_receive(&skb, !rx_handler))
3283 else if (unlikely(!skb))
3289 ret = deliver_skb(skb, pt_prev, orig_dev);
3292 switch (rx_handler(&skb)) {
3293 case RX_HANDLER_CONSUMED:
3295 case RX_HANDLER_ANOTHER:
3297 case RX_HANDLER_EXACT:
3298 deliver_exact = true;
3299 case RX_HANDLER_PASS:
3306 /* deliver only exact match when indicated */
3307 null_or_dev = deliver_exact ? skb->dev : NULL;
3309 type = skb->protocol;
3310 list_for_each_entry_rcu(ptype,
3311 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3312 if (ptype->type == type &&
3313 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3314 ptype->dev == orig_dev)) {
3316 ret = deliver_skb(skb, pt_prev, orig_dev);
3322 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3324 atomic_long_inc(&skb->dev->rx_dropped);
3326 /* Jamal, now you will not able to escape explaining
3327 * me how you were going to use this. :-)
3338 * netif_receive_skb - process receive buffer from network
3339 * @skb: buffer to process
3341 * netif_receive_skb() is the main receive data processing function.
3342 * It always succeeds. The buffer may be dropped during processing
3343 * for congestion control or by the protocol layers.
3345 * This function may only be called from softirq context and interrupts
3346 * should be enabled.
3348 * Return values (usually ignored):
3349 * NET_RX_SUCCESS: no congestion
3350 * NET_RX_DROP: packet was dropped
3352 int netif_receive_skb(struct sk_buff *skb)
3354 net_timestamp_check(netdev_tstamp_prequeue, skb);
3356 if (skb_defer_rx_timestamp(skb))
3357 return NET_RX_SUCCESS;
3361 struct rps_dev_flow voidflow, *rflow = &voidflow;
3366 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3369 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3373 ret = __netif_receive_skb(skb);
3379 return __netif_receive_skb(skb);
3382 EXPORT_SYMBOL(netif_receive_skb);
3384 /* Network device is going away, flush any packets still pending
3385 * Called with irqs disabled.
3387 static void flush_backlog(void *arg)
3389 struct net_device *dev = arg;
3390 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3391 struct sk_buff *skb, *tmp;
3394 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3395 if (skb->dev == dev) {
3396 __skb_unlink(skb, &sd->input_pkt_queue);
3398 input_queue_head_incr(sd);
3403 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3404 if (skb->dev == dev) {
3405 __skb_unlink(skb, &sd->process_queue);
3407 input_queue_head_incr(sd);
3412 static int napi_gro_complete(struct sk_buff *skb)
3414 struct packet_type *ptype;
3415 __be16 type = skb->protocol;
3416 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3419 if (NAPI_GRO_CB(skb)->count == 1) {
3420 skb_shinfo(skb)->gso_size = 0;
3425 list_for_each_entry_rcu(ptype, head, list) {
3426 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3429 err = ptype->gro_complete(skb);
3435 WARN_ON(&ptype->list == head);
3437 return NET_RX_SUCCESS;
3441 return netif_receive_skb(skb);
3444 inline void napi_gro_flush(struct napi_struct *napi)
3446 struct sk_buff *skb, *next;
3448 for (skb = napi->gro_list; skb; skb = next) {
3451 napi_gro_complete(skb);
3454 napi->gro_count = 0;
3455 napi->gro_list = NULL;
3457 EXPORT_SYMBOL(napi_gro_flush);
3459 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3461 struct sk_buff **pp = NULL;
3462 struct packet_type *ptype;
3463 __be16 type = skb->protocol;
3464 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3467 enum gro_result ret;
3469 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3472 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3476 list_for_each_entry_rcu(ptype, head, list) {
3477 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3480 skb_set_network_header(skb, skb_gro_offset(skb));
3481 mac_len = skb->network_header - skb->mac_header;
3482 skb->mac_len = mac_len;
3483 NAPI_GRO_CB(skb)->same_flow = 0;
3484 NAPI_GRO_CB(skb)->flush = 0;
3485 NAPI_GRO_CB(skb)->free = 0;
3487 pp = ptype->gro_receive(&napi->gro_list, skb);
3492 if (&ptype->list == head)
3495 same_flow = NAPI_GRO_CB(skb)->same_flow;
3496 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3499 struct sk_buff *nskb = *pp;
3503 napi_gro_complete(nskb);
3510 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3514 NAPI_GRO_CB(skb)->count = 1;
3515 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3516 skb->next = napi->gro_list;
3517 napi->gro_list = skb;
3521 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3522 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3524 BUG_ON(skb->end - skb->tail < grow);
3526 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3529 skb->data_len -= grow;
3531 skb_shinfo(skb)->frags[0].page_offset += grow;
3532 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3534 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3535 skb_frag_unref(skb, 0);
3536 memmove(skb_shinfo(skb)->frags,
3537 skb_shinfo(skb)->frags + 1,
3538 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3549 EXPORT_SYMBOL(dev_gro_receive);
3551 static inline gro_result_t
3552 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3556 for (p = napi->gro_list; p; p = p->next) {
3557 unsigned long diffs;
3559 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3560 diffs |= p->vlan_tci ^ skb->vlan_tci;
3561 diffs |= compare_ether_header(skb_mac_header(p),
3562 skb_gro_mac_header(skb));
3563 NAPI_GRO_CB(p)->same_flow = !diffs;
3564 NAPI_GRO_CB(p)->flush = 0;
3567 return dev_gro_receive(napi, skb);
3570 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3574 if (netif_receive_skb(skb))
3579 case GRO_MERGED_FREE:
3590 EXPORT_SYMBOL(napi_skb_finish);
3592 void skb_gro_reset_offset(struct sk_buff *skb)
3594 NAPI_GRO_CB(skb)->data_offset = 0;
3595 NAPI_GRO_CB(skb)->frag0 = NULL;
3596 NAPI_GRO_CB(skb)->frag0_len = 0;
3598 if (skb->mac_header == skb->tail &&
3599 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3600 NAPI_GRO_CB(skb)->frag0 =
3601 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3602 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3605 EXPORT_SYMBOL(skb_gro_reset_offset);
3607 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3609 skb_gro_reset_offset(skb);
3611 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3613 EXPORT_SYMBOL(napi_gro_receive);
3615 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3617 __skb_pull(skb, skb_headlen(skb));
3618 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3620 skb->dev = napi->dev;
3626 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3628 struct sk_buff *skb = napi->skb;
3631 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3637 EXPORT_SYMBOL(napi_get_frags);
3639 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3645 skb->protocol = eth_type_trans(skb, skb->dev);
3647 if (ret == GRO_HELD)
3648 skb_gro_pull(skb, -ETH_HLEN);
3649 else if (netif_receive_skb(skb))
3654 case GRO_MERGED_FREE:
3655 napi_reuse_skb(napi, skb);
3664 EXPORT_SYMBOL(napi_frags_finish);
3666 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3668 struct sk_buff *skb = napi->skb;
3675 skb_reset_mac_header(skb);
3676 skb_gro_reset_offset(skb);
3678 off = skb_gro_offset(skb);
3679 hlen = off + sizeof(*eth);
3680 eth = skb_gro_header_fast(skb, off);
3681 if (skb_gro_header_hard(skb, hlen)) {
3682 eth = skb_gro_header_slow(skb, hlen, off);
3683 if (unlikely(!eth)) {
3684 napi_reuse_skb(napi, skb);
3690 skb_gro_pull(skb, sizeof(*eth));
3693 * This works because the only protocols we care about don't require
3694 * special handling. We'll fix it up properly at the end.
3696 skb->protocol = eth->h_proto;
3701 EXPORT_SYMBOL(napi_frags_skb);
3703 gro_result_t napi_gro_frags(struct napi_struct *napi)
3705 struct sk_buff *skb = napi_frags_skb(napi);
3710 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3712 EXPORT_SYMBOL(napi_gro_frags);
3715 * net_rps_action sends any pending IPI's for rps.
3716 * Note: called with local irq disabled, but exits with local irq enabled.
3718 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3721 struct softnet_data *remsd = sd->rps_ipi_list;
3724 sd->rps_ipi_list = NULL;
3728 /* Send pending IPI's to kick RPS processing on remote cpus. */
3730 struct softnet_data *next = remsd->rps_ipi_next;
3732 if (cpu_online(remsd->cpu))
3733 __smp_call_function_single(remsd->cpu,
3742 static int process_backlog(struct napi_struct *napi, int quota)
3745 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3748 /* Check if we have pending ipi, its better to send them now,
3749 * not waiting net_rx_action() end.
3751 if (sd->rps_ipi_list) {
3752 local_irq_disable();
3753 net_rps_action_and_irq_enable(sd);
3756 napi->weight = weight_p;
3757 local_irq_disable();
3758 while (work < quota) {
3759 struct sk_buff *skb;
3762 while ((skb = __skb_dequeue(&sd->process_queue))) {
3764 __netif_receive_skb(skb);
3765 local_irq_disable();
3766 input_queue_head_incr(sd);
3767 if (++work >= quota) {
3774 qlen = skb_queue_len(&sd->input_pkt_queue);
3776 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3777 &sd->process_queue);
3779 if (qlen < quota - work) {
3781 * Inline a custom version of __napi_complete().
3782 * only current cpu owns and manipulates this napi,
3783 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3784 * we can use a plain write instead of clear_bit(),
3785 * and we dont need an smp_mb() memory barrier.
3787 list_del(&napi->poll_list);
3790 quota = work + qlen;
3800 * __napi_schedule - schedule for receive
3801 * @n: entry to schedule
3803 * The entry's receive function will be scheduled to run
3805 void __napi_schedule(struct napi_struct *n)
3807 unsigned long flags;
3809 local_irq_save(flags);
3810 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3811 local_irq_restore(flags);
3813 EXPORT_SYMBOL(__napi_schedule);
3815 void __napi_complete(struct napi_struct *n)
3817 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3818 BUG_ON(n->gro_list);
3820 list_del(&n->poll_list);
3821 smp_mb__before_clear_bit();
3822 clear_bit(NAPI_STATE_SCHED, &n->state);
3824 EXPORT_SYMBOL(__napi_complete);
3826 void napi_complete(struct napi_struct *n)
3828 unsigned long flags;
3831 * don't let napi dequeue from the cpu poll list
3832 * just in case its running on a different cpu
3834 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3838 local_irq_save(flags);
3840 local_irq_restore(flags);
3842 EXPORT_SYMBOL(napi_complete);
3844 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3845 int (*poll)(struct napi_struct *, int), int weight)
3847 INIT_LIST_HEAD(&napi->poll_list);
3848 napi->gro_count = 0;
3849 napi->gro_list = NULL;
3852 napi->weight = weight;
3853 list_add(&napi->dev_list, &dev->napi_list);
3855 #ifdef CONFIG_NETPOLL
3856 spin_lock_init(&napi->poll_lock);
3857 napi->poll_owner = -1;
3859 set_bit(NAPI_STATE_SCHED, &napi->state);
3861 EXPORT_SYMBOL(netif_napi_add);
3863 void netif_napi_del(struct napi_struct *napi)
3865 struct sk_buff *skb, *next;
3867 list_del_init(&napi->dev_list);
3868 napi_free_frags(napi);
3870 for (skb = napi->gro_list; skb; skb = next) {
3876 napi->gro_list = NULL;
3877 napi->gro_count = 0;
3879 EXPORT_SYMBOL(netif_napi_del);
3881 static void net_rx_action(struct softirq_action *h)
3883 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3884 unsigned long time_limit = jiffies + 2;
3885 int budget = netdev_budget;
3888 local_irq_disable();
3890 while (!list_empty(&sd->poll_list)) {
3891 struct napi_struct *n;
3894 /* If softirq window is exhuasted then punt.
3895 * Allow this to run for 2 jiffies since which will allow
3896 * an average latency of 1.5/HZ.
3898 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3903 /* Even though interrupts have been re-enabled, this
3904 * access is safe because interrupts can only add new
3905 * entries to the tail of this list, and only ->poll()
3906 * calls can remove this head entry from the list.
3908 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3910 have = netpoll_poll_lock(n);
3914 /* This NAPI_STATE_SCHED test is for avoiding a race
3915 * with netpoll's poll_napi(). Only the entity which
3916 * obtains the lock and sees NAPI_STATE_SCHED set will
3917 * actually make the ->poll() call. Therefore we avoid
3918 * accidentally calling ->poll() when NAPI is not scheduled.
3921 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3922 work = n->poll(n, weight);
3926 WARN_ON_ONCE(work > weight);
3930 local_irq_disable();
3932 /* Drivers must not modify the NAPI state if they
3933 * consume the entire weight. In such cases this code
3934 * still "owns" the NAPI instance and therefore can
3935 * move the instance around on the list at-will.
3937 if (unlikely(work == weight)) {
3938 if (unlikely(napi_disable_pending(n))) {
3941 local_irq_disable();
3943 list_move_tail(&n->poll_list, &sd->poll_list);
3946 netpoll_poll_unlock(have);
3949 net_rps_action_and_irq_enable(sd);
3951 #ifdef CONFIG_NET_DMA
3953 * There may not be any more sk_buffs coming right now, so push
3954 * any pending DMA copies to hardware
3956 dma_issue_pending_all();
3963 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3967 static gifconf_func_t *gifconf_list[NPROTO];
3970 * register_gifconf - register a SIOCGIF handler
3971 * @family: Address family
3972 * @gifconf: Function handler
3974 * Register protocol dependent address dumping routines. The handler
3975 * that is passed must not be freed or reused until it has been replaced
3976 * by another handler.
3978 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3980 if (family >= NPROTO)
3982 gifconf_list[family] = gifconf;
3985 EXPORT_SYMBOL(register_gifconf);
3989 * Map an interface index to its name (SIOCGIFNAME)
3993 * We need this ioctl for efficient implementation of the
3994 * if_indextoname() function required by the IPv6 API. Without
3995 * it, we would have to search all the interfaces to find a
3999 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4001 struct net_device *dev;
4005 * Fetch the caller's info block.
4008 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4012 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4018 strcpy(ifr.ifr_name, dev->name);
4021 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4027 * Perform a SIOCGIFCONF call. This structure will change
4028 * size eventually, and there is nothing I can do about it.
4029 * Thus we will need a 'compatibility mode'.
4032 static int dev_ifconf(struct net *net, char __user *arg)
4035 struct net_device *dev;
4042 * Fetch the caller's info block.
4045 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4052 * Loop over the interfaces, and write an info block for each.
4056 for_each_netdev(net, dev) {
4057 for (i = 0; i < NPROTO; i++) {
4058 if (gifconf_list[i]) {
4061 done = gifconf_list[i](dev, NULL, 0);
4063 done = gifconf_list[i](dev, pos + total,
4073 * All done. Write the updated control block back to the caller.
4075 ifc.ifc_len = total;
4078 * Both BSD and Solaris return 0 here, so we do too.
4080 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4083 #ifdef CONFIG_PROC_FS
4085 #define BUCKET_SPACE (32 - NETDEV_HASHBITS)
4087 struct dev_iter_state {
4088 struct seq_net_private p;
4089 unsigned int pos; /* bucket << BUCKET_SPACE + offset */
4092 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4093 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4094 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4096 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq)
4098 struct dev_iter_state *state = seq->private;
4099 struct net *net = seq_file_net(seq);
4100 struct net_device *dev;
4101 struct hlist_node *p;
4102 struct hlist_head *h;
4103 unsigned int count, bucket, offset;
4105 bucket = get_bucket(state->pos);
4106 offset = get_offset(state->pos);
4107 h = &net->dev_name_head[bucket];
4109 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4110 if (count++ == offset) {
4111 state->pos = set_bucket_offset(bucket, count);
4119 static inline struct net_device *dev_from_new_bucket(struct seq_file *seq)
4121 struct dev_iter_state *state = seq->private;
4122 struct net_device *dev;
4123 unsigned int bucket;
4125 bucket = get_bucket(state->pos);
4127 dev = dev_from_same_bucket(seq);
4132 state->pos = set_bucket_offset(bucket, 0);
4133 } while (bucket < NETDEV_HASHENTRIES);
4139 * This is invoked by the /proc filesystem handler to display a device
4142 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4145 struct dev_iter_state *state = seq->private;
4149 return SEQ_START_TOKEN;
4151 /* check for end of the hash */
4152 if (state->pos == 0 && *pos > 1)
4155 return dev_from_new_bucket(seq);
4158 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4160 struct net_device *dev;
4164 if (v == SEQ_START_TOKEN)
4165 return dev_from_new_bucket(seq);
4167 dev = dev_from_same_bucket(seq);
4171 return dev_from_new_bucket(seq);
4174 void dev_seq_stop(struct seq_file *seq, void *v)
4180 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4182 struct rtnl_link_stats64 temp;
4183 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4185 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4186 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4187 dev->name, stats->rx_bytes, stats->rx_packets,
4189 stats->rx_dropped + stats->rx_missed_errors,
4190 stats->rx_fifo_errors,
4191 stats->rx_length_errors + stats->rx_over_errors +
4192 stats->rx_crc_errors + stats->rx_frame_errors,
4193 stats->rx_compressed, stats->multicast,
4194 stats->tx_bytes, stats->tx_packets,
4195 stats->tx_errors, stats->tx_dropped,
4196 stats->tx_fifo_errors, stats->collisions,
4197 stats->tx_carrier_errors +
4198 stats->tx_aborted_errors +
4199 stats->tx_window_errors +
4200 stats->tx_heartbeat_errors,
4201 stats->tx_compressed);
4205 * Called from the PROCfs module. This now uses the new arbitrary sized
4206 * /proc/net interface to create /proc/net/dev
4208 static int dev_seq_show(struct seq_file *seq, void *v)
4210 if (v == SEQ_START_TOKEN)
4211 seq_puts(seq, "Inter-| Receive "
4213 " face |bytes packets errs drop fifo frame "
4214 "compressed multicast|bytes packets errs "
4215 "drop fifo colls carrier compressed\n");
4217 dev_seq_printf_stats(seq, v);
4221 static struct softnet_data *softnet_get_online(loff_t *pos)
4223 struct softnet_data *sd = NULL;
4225 while (*pos < nr_cpu_ids)
4226 if (cpu_online(*pos)) {
4227 sd = &per_cpu(softnet_data, *pos);
4234 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4236 return softnet_get_online(pos);
4239 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4242 return softnet_get_online(pos);
4245 static void softnet_seq_stop(struct seq_file *seq, void *v)
4249 static int softnet_seq_show(struct seq_file *seq, void *v)
4251 struct softnet_data *sd = v;
4253 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4254 sd->processed, sd->dropped, sd->time_squeeze, 0,
4255 0, 0, 0, 0, /* was fastroute */
4256 sd->cpu_collision, sd->received_rps);
4260 static const struct seq_operations dev_seq_ops = {
4261 .start = dev_seq_start,
4262 .next = dev_seq_next,
4263 .stop = dev_seq_stop,
4264 .show = dev_seq_show,
4267 static int dev_seq_open(struct inode *inode, struct file *file)
4269 return seq_open_net(inode, file, &dev_seq_ops,
4270 sizeof(struct dev_iter_state));
4273 static const struct file_operations dev_seq_fops = {
4274 .owner = THIS_MODULE,
4275 .open = dev_seq_open,
4277 .llseek = seq_lseek,
4278 .release = seq_release_net,
4281 static const struct seq_operations softnet_seq_ops = {
4282 .start = softnet_seq_start,
4283 .next = softnet_seq_next,
4284 .stop = softnet_seq_stop,
4285 .show = softnet_seq_show,
4288 static int softnet_seq_open(struct inode *inode, struct file *file)
4290 return seq_open(file, &softnet_seq_ops);
4293 static const struct file_operations softnet_seq_fops = {
4294 .owner = THIS_MODULE,
4295 .open = softnet_seq_open,
4297 .llseek = seq_lseek,
4298 .release = seq_release,
4301 static void *ptype_get_idx(loff_t pos)
4303 struct packet_type *pt = NULL;
4307 list_for_each_entry_rcu(pt, &ptype_all, list) {
4313 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4314 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4323 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4327 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4330 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4332 struct packet_type *pt;
4333 struct list_head *nxt;
4337 if (v == SEQ_START_TOKEN)
4338 return ptype_get_idx(0);
4341 nxt = pt->list.next;
4342 if (pt->type == htons(ETH_P_ALL)) {
4343 if (nxt != &ptype_all)
4346 nxt = ptype_base[0].next;
4348 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4350 while (nxt == &ptype_base[hash]) {
4351 if (++hash >= PTYPE_HASH_SIZE)
4353 nxt = ptype_base[hash].next;
4356 return list_entry(nxt, struct packet_type, list);
4359 static void ptype_seq_stop(struct seq_file *seq, void *v)
4365 static int ptype_seq_show(struct seq_file *seq, void *v)
4367 struct packet_type *pt = v;
4369 if (v == SEQ_START_TOKEN)
4370 seq_puts(seq, "Type Device Function\n");
4371 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4372 if (pt->type == htons(ETH_P_ALL))
4373 seq_puts(seq, "ALL ");
4375 seq_printf(seq, "%04x", ntohs(pt->type));
4377 seq_printf(seq, " %-8s %pF\n",
4378 pt->dev ? pt->dev->name : "", pt->func);
4384 static const struct seq_operations ptype_seq_ops = {
4385 .start = ptype_seq_start,
4386 .next = ptype_seq_next,
4387 .stop = ptype_seq_stop,
4388 .show = ptype_seq_show,
4391 static int ptype_seq_open(struct inode *inode, struct file *file)
4393 return seq_open_net(inode, file, &ptype_seq_ops,
4394 sizeof(struct seq_net_private));
4397 static const struct file_operations ptype_seq_fops = {
4398 .owner = THIS_MODULE,
4399 .open = ptype_seq_open,
4401 .llseek = seq_lseek,
4402 .release = seq_release_net,
4406 static int __net_init dev_proc_net_init(struct net *net)
4410 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4412 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4414 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4417 if (wext_proc_init(net))
4423 proc_net_remove(net, "ptype");
4425 proc_net_remove(net, "softnet_stat");
4427 proc_net_remove(net, "dev");
4431 static void __net_exit dev_proc_net_exit(struct net *net)
4433 wext_proc_exit(net);
4435 proc_net_remove(net, "ptype");
4436 proc_net_remove(net, "softnet_stat");
4437 proc_net_remove(net, "dev");
4440 static struct pernet_operations __net_initdata dev_proc_ops = {
4441 .init = dev_proc_net_init,
4442 .exit = dev_proc_net_exit,
4445 static int __init dev_proc_init(void)
4447 return register_pernet_subsys(&dev_proc_ops);
4450 #define dev_proc_init() 0
4451 #endif /* CONFIG_PROC_FS */
4455 * netdev_set_master - set up master pointer
4456 * @slave: slave device
4457 * @master: new master device
4459 * Changes the master device of the slave. Pass %NULL to break the
4460 * bonding. The caller must hold the RTNL semaphore. On a failure
4461 * a negative errno code is returned. On success the reference counts
4462 * are adjusted and the function returns zero.
4464 int netdev_set_master(struct net_device *slave, struct net_device *master)
4466 struct net_device *old = slave->master;
4476 slave->master = master;
4482 EXPORT_SYMBOL(netdev_set_master);
4485 * netdev_set_bond_master - set up bonding master/slave pair
4486 * @slave: slave device
4487 * @master: new master device
4489 * Changes the master device of the slave. Pass %NULL to break the
4490 * bonding. The caller must hold the RTNL semaphore. On a failure
4491 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4492 * to the routing socket and the function returns zero.
4494 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4500 err = netdev_set_master(slave, master);
4504 slave->flags |= IFF_SLAVE;
4506 slave->flags &= ~IFF_SLAVE;
4508 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4511 EXPORT_SYMBOL(netdev_set_bond_master);
4513 static void dev_change_rx_flags(struct net_device *dev, int flags)
4515 const struct net_device_ops *ops = dev->netdev_ops;
4517 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4518 ops->ndo_change_rx_flags(dev, flags);
4521 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4523 unsigned short old_flags = dev->flags;
4529 dev->flags |= IFF_PROMISC;
4530 dev->promiscuity += inc;
4531 if (dev->promiscuity == 0) {
4534 * If inc causes overflow, untouch promisc and return error.
4537 dev->flags &= ~IFF_PROMISC;
4539 dev->promiscuity -= inc;
4540 printk(KERN_WARNING "%s: promiscuity touches roof, "
4541 "set promiscuity failed, promiscuity feature "
4542 "of device might be broken.\n", dev->name);
4546 if (dev->flags != old_flags) {
4547 printk(KERN_INFO "device %s %s promiscuous mode\n",
4548 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4550 if (audit_enabled) {
4551 current_uid_gid(&uid, &gid);
4552 audit_log(current->audit_context, GFP_ATOMIC,
4553 AUDIT_ANOM_PROMISCUOUS,
4554 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4555 dev->name, (dev->flags & IFF_PROMISC),
4556 (old_flags & IFF_PROMISC),
4557 audit_get_loginuid(current),
4559 audit_get_sessionid(current));
4562 dev_change_rx_flags(dev, IFF_PROMISC);
4568 * dev_set_promiscuity - update promiscuity count on a device
4572 * Add or remove promiscuity from a device. While the count in the device
4573 * remains above zero the interface remains promiscuous. Once it hits zero
4574 * the device reverts back to normal filtering operation. A negative inc
4575 * value is used to drop promiscuity on the device.
4576 * Return 0 if successful or a negative errno code on error.
4578 int dev_set_promiscuity(struct net_device *dev, int inc)
4580 unsigned short old_flags = dev->flags;
4583 err = __dev_set_promiscuity(dev, inc);
4586 if (dev->flags != old_flags)
4587 dev_set_rx_mode(dev);
4590 EXPORT_SYMBOL(dev_set_promiscuity);
4593 * dev_set_allmulti - update allmulti count on a device
4597 * Add or remove reception of all multicast frames to a device. While the
4598 * count in the device remains above zero the interface remains listening
4599 * to all interfaces. Once it hits zero the device reverts back to normal
4600 * filtering operation. A negative @inc value is used to drop the counter
4601 * when releasing a resource needing all multicasts.
4602 * Return 0 if successful or a negative errno code on error.
4605 int dev_set_allmulti(struct net_device *dev, int inc)
4607 unsigned short old_flags = dev->flags;
4611 dev->flags |= IFF_ALLMULTI;
4612 dev->allmulti += inc;
4613 if (dev->allmulti == 0) {
4616 * If inc causes overflow, untouch allmulti and return error.
4619 dev->flags &= ~IFF_ALLMULTI;
4621 dev->allmulti -= inc;
4622 printk(KERN_WARNING "%s: allmulti touches roof, "
4623 "set allmulti failed, allmulti feature of "
4624 "device might be broken.\n", dev->name);
4628 if (dev->flags ^ old_flags) {
4629 dev_change_rx_flags(dev, IFF_ALLMULTI);
4630 dev_set_rx_mode(dev);
4634 EXPORT_SYMBOL(dev_set_allmulti);
4637 * Upload unicast and multicast address lists to device and
4638 * configure RX filtering. When the device doesn't support unicast
4639 * filtering it is put in promiscuous mode while unicast addresses
4642 void __dev_set_rx_mode(struct net_device *dev)
4644 const struct net_device_ops *ops = dev->netdev_ops;
4646 /* dev_open will call this function so the list will stay sane. */
4647 if (!(dev->flags&IFF_UP))
4650 if (!netif_device_present(dev))
4653 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4654 /* Unicast addresses changes may only happen under the rtnl,
4655 * therefore calling __dev_set_promiscuity here is safe.
4657 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4658 __dev_set_promiscuity(dev, 1);
4659 dev->uc_promisc = true;
4660 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4661 __dev_set_promiscuity(dev, -1);
4662 dev->uc_promisc = false;
4666 if (ops->ndo_set_rx_mode)
4667 ops->ndo_set_rx_mode(dev);
4670 void dev_set_rx_mode(struct net_device *dev)
4672 netif_addr_lock_bh(dev);
4673 __dev_set_rx_mode(dev);
4674 netif_addr_unlock_bh(dev);
4678 * dev_get_flags - get flags reported to userspace
4681 * Get the combination of flag bits exported through APIs to userspace.
4683 unsigned dev_get_flags(const struct net_device *dev)
4687 flags = (dev->flags & ~(IFF_PROMISC |
4692 (dev->gflags & (IFF_PROMISC |
4695 if (netif_running(dev)) {
4696 if (netif_oper_up(dev))
4697 flags |= IFF_RUNNING;
4698 if (netif_carrier_ok(dev))
4699 flags |= IFF_LOWER_UP;
4700 if (netif_dormant(dev))
4701 flags |= IFF_DORMANT;
4706 EXPORT_SYMBOL(dev_get_flags);
4708 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4710 int old_flags = dev->flags;
4716 * Set the flags on our device.
4719 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4720 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4722 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4726 * Load in the correct multicast list now the flags have changed.
4729 if ((old_flags ^ flags) & IFF_MULTICAST)
4730 dev_change_rx_flags(dev, IFF_MULTICAST);
4732 dev_set_rx_mode(dev);
4735 * Have we downed the interface. We handle IFF_UP ourselves
4736 * according to user attempts to set it, rather than blindly
4741 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4742 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4745 dev_set_rx_mode(dev);
4748 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4749 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4751 dev->gflags ^= IFF_PROMISC;
4752 dev_set_promiscuity(dev, inc);
4755 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4756 is important. Some (broken) drivers set IFF_PROMISC, when
4757 IFF_ALLMULTI is requested not asking us and not reporting.
4759 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4760 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4762 dev->gflags ^= IFF_ALLMULTI;
4763 dev_set_allmulti(dev, inc);
4769 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4771 unsigned int changes = dev->flags ^ old_flags;
4773 if (changes & IFF_UP) {
4774 if (dev->flags & IFF_UP)
4775 call_netdevice_notifiers(NETDEV_UP, dev);
4777 call_netdevice_notifiers(NETDEV_DOWN, dev);
4780 if (dev->flags & IFF_UP &&
4781 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4782 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4786 * dev_change_flags - change device settings
4788 * @flags: device state flags
4790 * Change settings on device based state flags. The flags are
4791 * in the userspace exported format.
4793 int dev_change_flags(struct net_device *dev, unsigned flags)
4796 int old_flags = dev->flags;
4798 ret = __dev_change_flags(dev, flags);
4802 changes = old_flags ^ dev->flags;
4804 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4806 __dev_notify_flags(dev, old_flags);
4809 EXPORT_SYMBOL(dev_change_flags);
4812 * dev_set_mtu - Change maximum transfer unit
4814 * @new_mtu: new transfer unit
4816 * Change the maximum transfer size of the network device.
4818 int dev_set_mtu(struct net_device *dev, int new_mtu)
4820 const struct net_device_ops *ops = dev->netdev_ops;
4823 if (new_mtu == dev->mtu)
4826 /* MTU must be positive. */
4830 if (!netif_device_present(dev))
4834 if (ops->ndo_change_mtu)
4835 err = ops->ndo_change_mtu(dev, new_mtu);
4839 if (!err && dev->flags & IFF_UP)
4840 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4843 EXPORT_SYMBOL(dev_set_mtu);
4846 * dev_set_group - Change group this device belongs to
4848 * @new_group: group this device should belong to
4850 void dev_set_group(struct net_device *dev, int new_group)
4852 dev->group = new_group;
4854 EXPORT_SYMBOL(dev_set_group);
4857 * dev_set_mac_address - Change Media Access Control Address
4861 * Change the hardware (MAC) address of the device
4863 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4865 const struct net_device_ops *ops = dev->netdev_ops;
4868 if (!ops->ndo_set_mac_address)
4870 if (sa->sa_family != dev->type)
4872 if (!netif_device_present(dev))
4874 err = ops->ndo_set_mac_address(dev, sa);
4876 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4879 EXPORT_SYMBOL(dev_set_mac_address);
4882 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4884 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4887 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4893 case SIOCGIFFLAGS: /* Get interface flags */
4894 ifr->ifr_flags = (short) dev_get_flags(dev);
4897 case SIOCGIFMETRIC: /* Get the metric on the interface
4898 (currently unused) */
4899 ifr->ifr_metric = 0;
4902 case SIOCGIFMTU: /* Get the MTU of a device */
4903 ifr->ifr_mtu = dev->mtu;
4908 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4910 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4911 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4912 ifr->ifr_hwaddr.sa_family = dev->type;
4920 ifr->ifr_map.mem_start = dev->mem_start;
4921 ifr->ifr_map.mem_end = dev->mem_end;
4922 ifr->ifr_map.base_addr = dev->base_addr;
4923 ifr->ifr_map.irq = dev->irq;
4924 ifr->ifr_map.dma = dev->dma;
4925 ifr->ifr_map.port = dev->if_port;
4929 ifr->ifr_ifindex = dev->ifindex;
4933 ifr->ifr_qlen = dev->tx_queue_len;
4937 /* dev_ioctl() should ensure this case
4949 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4951 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4954 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4955 const struct net_device_ops *ops;
4960 ops = dev->netdev_ops;
4963 case SIOCSIFFLAGS: /* Set interface flags */
4964 return dev_change_flags(dev, ifr->ifr_flags);
4966 case SIOCSIFMETRIC: /* Set the metric on the interface
4967 (currently unused) */
4970 case SIOCSIFMTU: /* Set the MTU of a device */
4971 return dev_set_mtu(dev, ifr->ifr_mtu);
4974 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4976 case SIOCSIFHWBROADCAST:
4977 if (ifr->ifr_hwaddr.sa_family != dev->type)
4979 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4980 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4981 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4985 if (ops->ndo_set_config) {
4986 if (!netif_device_present(dev))
4988 return ops->ndo_set_config(dev, &ifr->ifr_map);
4993 if (!ops->ndo_set_rx_mode ||
4994 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4996 if (!netif_device_present(dev))
4998 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5001 if (!ops->ndo_set_rx_mode ||
5002 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5004 if (!netif_device_present(dev))
5006 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5009 if (ifr->ifr_qlen < 0)
5011 dev->tx_queue_len = ifr->ifr_qlen;
5015 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5016 return dev_change_name(dev, ifr->ifr_newname);
5019 err = net_hwtstamp_validate(ifr);
5025 * Unknown or private ioctl
5028 if ((cmd >= SIOCDEVPRIVATE &&
5029 cmd <= SIOCDEVPRIVATE + 15) ||
5030 cmd == SIOCBONDENSLAVE ||
5031 cmd == SIOCBONDRELEASE ||
5032 cmd == SIOCBONDSETHWADDR ||
5033 cmd == SIOCBONDSLAVEINFOQUERY ||
5034 cmd == SIOCBONDINFOQUERY ||
5035 cmd == SIOCBONDCHANGEACTIVE ||
5036 cmd == SIOCGMIIPHY ||
5037 cmd == SIOCGMIIREG ||
5038 cmd == SIOCSMIIREG ||
5039 cmd == SIOCBRADDIF ||
5040 cmd == SIOCBRDELIF ||
5041 cmd == SIOCSHWTSTAMP ||
5042 cmd == SIOCWANDEV) {
5044 if (ops->ndo_do_ioctl) {
5045 if (netif_device_present(dev))
5046 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5058 * This function handles all "interface"-type I/O control requests. The actual
5059 * 'doing' part of this is dev_ifsioc above.
5063 * dev_ioctl - network device ioctl
5064 * @net: the applicable net namespace
5065 * @cmd: command to issue
5066 * @arg: pointer to a struct ifreq in user space
5068 * Issue ioctl functions to devices. This is normally called by the
5069 * user space syscall interfaces but can sometimes be useful for
5070 * other purposes. The return value is the return from the syscall if
5071 * positive or a negative errno code on error.
5074 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5080 /* One special case: SIOCGIFCONF takes ifconf argument
5081 and requires shared lock, because it sleeps writing
5085 if (cmd == SIOCGIFCONF) {
5087 ret = dev_ifconf(net, (char __user *) arg);
5091 if (cmd == SIOCGIFNAME)
5092 return dev_ifname(net, (struct ifreq __user *)arg);
5094 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5097 ifr.ifr_name[IFNAMSIZ-1] = 0;
5099 colon = strchr(ifr.ifr_name, ':');
5104 * See which interface the caller is talking about.
5109 * These ioctl calls:
5110 * - can be done by all.
5111 * - atomic and do not require locking.
5122 dev_load(net, ifr.ifr_name);
5124 ret = dev_ifsioc_locked(net, &ifr, cmd);
5129 if (copy_to_user(arg, &ifr,
5130 sizeof(struct ifreq)))
5136 dev_load(net, ifr.ifr_name);
5138 ret = dev_ethtool(net, &ifr);
5143 if (copy_to_user(arg, &ifr,
5144 sizeof(struct ifreq)))
5150 * These ioctl calls:
5151 * - require superuser power.
5152 * - require strict serialization.
5158 if (!capable(CAP_NET_ADMIN))
5160 dev_load(net, ifr.ifr_name);
5162 ret = dev_ifsioc(net, &ifr, cmd);
5167 if (copy_to_user(arg, &ifr,
5168 sizeof(struct ifreq)))
5174 * These ioctl calls:
5175 * - require superuser power.
5176 * - require strict serialization.
5177 * - do not return a value
5187 case SIOCSIFHWBROADCAST:
5190 case SIOCBONDENSLAVE:
5191 case SIOCBONDRELEASE:
5192 case SIOCBONDSETHWADDR:
5193 case SIOCBONDCHANGEACTIVE:
5197 if (!capable(CAP_NET_ADMIN))
5200 case SIOCBONDSLAVEINFOQUERY:
5201 case SIOCBONDINFOQUERY:
5202 dev_load(net, ifr.ifr_name);
5204 ret = dev_ifsioc(net, &ifr, cmd);
5209 /* Get the per device memory space. We can add this but
5210 * currently do not support it */
5212 /* Set the per device memory buffer space.
5213 * Not applicable in our case */
5218 * Unknown or private ioctl.
5221 if (cmd == SIOCWANDEV ||
5222 (cmd >= SIOCDEVPRIVATE &&
5223 cmd <= SIOCDEVPRIVATE + 15)) {
5224 dev_load(net, ifr.ifr_name);
5226 ret = dev_ifsioc(net, &ifr, cmd);
5228 if (!ret && copy_to_user(arg, &ifr,
5229 sizeof(struct ifreq)))
5233 /* Take care of Wireless Extensions */
5234 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5235 return wext_handle_ioctl(net, &ifr, cmd, arg);
5242 * dev_new_index - allocate an ifindex
5243 * @net: the applicable net namespace
5245 * Returns a suitable unique value for a new device interface
5246 * number. The caller must hold the rtnl semaphore or the
5247 * dev_base_lock to be sure it remains unique.
5249 static int dev_new_index(struct net *net)
5255 if (!__dev_get_by_index(net, ifindex))
5260 /* Delayed registration/unregisteration */
5261 static LIST_HEAD(net_todo_list);
5263 static void net_set_todo(struct net_device *dev)
5265 list_add_tail(&dev->todo_list, &net_todo_list);
5268 static void rollback_registered_many(struct list_head *head)
5270 struct net_device *dev, *tmp;
5272 BUG_ON(dev_boot_phase);
5275 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5276 /* Some devices call without registering
5277 * for initialization unwind. Remove those
5278 * devices and proceed with the remaining.
5280 if (dev->reg_state == NETREG_UNINITIALIZED) {
5281 pr_debug("unregister_netdevice: device %s/%p never "
5282 "was registered\n", dev->name, dev);
5285 list_del(&dev->unreg_list);
5288 dev->dismantle = true;
5289 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5292 /* If device is running, close it first. */
5293 dev_close_many(head);
5295 list_for_each_entry(dev, head, unreg_list) {
5296 /* And unlink it from device chain. */
5297 unlist_netdevice(dev);
5299 dev->reg_state = NETREG_UNREGISTERING;
5304 list_for_each_entry(dev, head, unreg_list) {
5305 /* Shutdown queueing discipline. */
5309 /* Notify protocols, that we are about to destroy
5310 this device. They should clean all the things.
5312 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5314 if (!dev->rtnl_link_ops ||
5315 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5316 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5319 * Flush the unicast and multicast chains
5324 if (dev->netdev_ops->ndo_uninit)
5325 dev->netdev_ops->ndo_uninit(dev);
5327 /* Notifier chain MUST detach us from master device. */
5328 WARN_ON(dev->master);
5330 /* Remove entries from kobject tree */
5331 netdev_unregister_kobject(dev);
5334 /* Process any work delayed until the end of the batch */
5335 dev = list_first_entry(head, struct net_device, unreg_list);
5336 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5340 list_for_each_entry(dev, head, unreg_list)
5344 static void rollback_registered(struct net_device *dev)
5348 list_add(&dev->unreg_list, &single);
5349 rollback_registered_many(&single);
5353 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5355 /* Fix illegal checksum combinations */
5356 if ((features & NETIF_F_HW_CSUM) &&
5357 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5358 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5359 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5362 if ((features & NETIF_F_NO_CSUM) &&
5363 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5364 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5365 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5368 /* Fix illegal SG+CSUM combinations. */
5369 if ((features & NETIF_F_SG) &&
5370 !(features & NETIF_F_ALL_CSUM)) {
5372 "Dropping NETIF_F_SG since no checksum feature.\n");
5373 features &= ~NETIF_F_SG;
5376 /* TSO requires that SG is present as well. */
5377 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5378 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5379 features &= ~NETIF_F_ALL_TSO;
5382 /* TSO ECN requires that TSO is present as well. */
5383 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5384 features &= ~NETIF_F_TSO_ECN;
5386 /* Software GSO depends on SG. */
5387 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5388 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5389 features &= ~NETIF_F_GSO;
5392 /* UFO needs SG and checksumming */
5393 if (features & NETIF_F_UFO) {
5394 /* maybe split UFO into V4 and V6? */
5395 if (!((features & NETIF_F_GEN_CSUM) ||
5396 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5397 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5399 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5400 features &= ~NETIF_F_UFO;
5403 if (!(features & NETIF_F_SG)) {
5405 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5406 features &= ~NETIF_F_UFO;
5413 int __netdev_update_features(struct net_device *dev)
5420 features = netdev_get_wanted_features(dev);
5422 if (dev->netdev_ops->ndo_fix_features)
5423 features = dev->netdev_ops->ndo_fix_features(dev, features);
5425 /* driver might be less strict about feature dependencies */
5426 features = netdev_fix_features(dev, features);
5428 if (dev->features == features)
5431 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5432 dev->features, features);
5434 if (dev->netdev_ops->ndo_set_features)
5435 err = dev->netdev_ops->ndo_set_features(dev, features);
5437 if (unlikely(err < 0)) {
5439 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5440 err, features, dev->features);
5445 dev->features = features;
5451 * netdev_update_features - recalculate device features
5452 * @dev: the device to check
5454 * Recalculate dev->features set and send notifications if it
5455 * has changed. Should be called after driver or hardware dependent
5456 * conditions might have changed that influence the features.
5458 void netdev_update_features(struct net_device *dev)
5460 if (__netdev_update_features(dev))
5461 netdev_features_change(dev);
5463 EXPORT_SYMBOL(netdev_update_features);
5466 * netdev_change_features - recalculate device features
5467 * @dev: the device to check
5469 * Recalculate dev->features set and send notifications even
5470 * if they have not changed. Should be called instead of
5471 * netdev_update_features() if also dev->vlan_features might
5472 * have changed to allow the changes to be propagated to stacked
5475 void netdev_change_features(struct net_device *dev)
5477 __netdev_update_features(dev);
5478 netdev_features_change(dev);
5480 EXPORT_SYMBOL(netdev_change_features);
5483 * netif_stacked_transfer_operstate - transfer operstate
5484 * @rootdev: the root or lower level device to transfer state from
5485 * @dev: the device to transfer operstate to
5487 * Transfer operational state from root to device. This is normally
5488 * called when a stacking relationship exists between the root
5489 * device and the device(a leaf device).
5491 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5492 struct net_device *dev)
5494 if (rootdev->operstate == IF_OPER_DORMANT)
5495 netif_dormant_on(dev);
5497 netif_dormant_off(dev);
5499 if (netif_carrier_ok(rootdev)) {
5500 if (!netif_carrier_ok(dev))
5501 netif_carrier_on(dev);
5503 if (netif_carrier_ok(dev))
5504 netif_carrier_off(dev);
5507 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5510 static int netif_alloc_rx_queues(struct net_device *dev)
5512 unsigned int i, count = dev->num_rx_queues;
5513 struct netdev_rx_queue *rx;
5517 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5519 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5524 for (i = 0; i < count; i++)
5530 static void netdev_init_one_queue(struct net_device *dev,
5531 struct netdev_queue *queue, void *_unused)
5533 /* Initialize queue lock */
5534 spin_lock_init(&queue->_xmit_lock);
5535 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5536 queue->xmit_lock_owner = -1;
5537 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5541 static int netif_alloc_netdev_queues(struct net_device *dev)
5543 unsigned int count = dev->num_tx_queues;
5544 struct netdev_queue *tx;
5548 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5550 pr_err("netdev: Unable to allocate %u tx queues.\n",
5556 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5557 spin_lock_init(&dev->tx_global_lock);
5563 * register_netdevice - register a network device
5564 * @dev: device to register
5566 * Take a completed network device structure and add it to the kernel
5567 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5568 * chain. 0 is returned on success. A negative errno code is returned
5569 * on a failure to set up the device, or if the name is a duplicate.
5571 * Callers must hold the rtnl semaphore. You may want
5572 * register_netdev() instead of this.
5575 * The locking appears insufficient to guarantee two parallel registers
5576 * will not get the same name.
5579 int register_netdevice(struct net_device *dev)
5582 struct net *net = dev_net(dev);
5584 BUG_ON(dev_boot_phase);
5589 /* When net_device's are persistent, this will be fatal. */
5590 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5593 spin_lock_init(&dev->addr_list_lock);
5594 netdev_set_addr_lockdep_class(dev);
5598 ret = dev_get_valid_name(dev, dev->name);
5602 /* Init, if this function is available */
5603 if (dev->netdev_ops->ndo_init) {
5604 ret = dev->netdev_ops->ndo_init(dev);
5612 dev->ifindex = dev_new_index(net);
5613 if (dev->iflink == -1)
5614 dev->iflink = dev->ifindex;
5616 /* Transfer changeable features to wanted_features and enable
5617 * software offloads (GSO and GRO).
5619 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5620 dev->features |= NETIF_F_SOFT_FEATURES;
5621 dev->wanted_features = dev->features & dev->hw_features;
5623 /* Turn on no cache copy if HW is doing checksum */
5624 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5625 if ((dev->features & NETIF_F_ALL_CSUM) &&
5626 !(dev->features & NETIF_F_NO_CSUM)) {
5627 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5628 dev->features |= NETIF_F_NOCACHE_COPY;
5631 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5633 dev->vlan_features |= NETIF_F_HIGHDMA;
5635 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5636 ret = notifier_to_errno(ret);
5640 ret = netdev_register_kobject(dev);
5643 dev->reg_state = NETREG_REGISTERED;
5645 __netdev_update_features(dev);
5648 * Default initial state at registry is that the
5649 * device is present.
5652 set_bit(__LINK_STATE_PRESENT, &dev->state);
5654 dev_init_scheduler(dev);
5656 list_netdevice(dev);
5658 /* Notify protocols, that a new device appeared. */
5659 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5660 ret = notifier_to_errno(ret);
5662 rollback_registered(dev);
5663 dev->reg_state = NETREG_UNREGISTERED;
5666 * Prevent userspace races by waiting until the network
5667 * device is fully setup before sending notifications.
5669 if (!dev->rtnl_link_ops ||
5670 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5671 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5677 if (dev->netdev_ops->ndo_uninit)
5678 dev->netdev_ops->ndo_uninit(dev);
5681 EXPORT_SYMBOL(register_netdevice);
5684 * init_dummy_netdev - init a dummy network device for NAPI
5685 * @dev: device to init
5687 * This takes a network device structure and initialize the minimum
5688 * amount of fields so it can be used to schedule NAPI polls without
5689 * registering a full blown interface. This is to be used by drivers
5690 * that need to tie several hardware interfaces to a single NAPI
5691 * poll scheduler due to HW limitations.
5693 int init_dummy_netdev(struct net_device *dev)
5695 /* Clear everything. Note we don't initialize spinlocks
5696 * are they aren't supposed to be taken by any of the
5697 * NAPI code and this dummy netdev is supposed to be
5698 * only ever used for NAPI polls
5700 memset(dev, 0, sizeof(struct net_device));
5702 /* make sure we BUG if trying to hit standard
5703 * register/unregister code path
5705 dev->reg_state = NETREG_DUMMY;
5707 /* NAPI wants this */
5708 INIT_LIST_HEAD(&dev->napi_list);
5710 /* a dummy interface is started by default */
5711 set_bit(__LINK_STATE_PRESENT, &dev->state);
5712 set_bit(__LINK_STATE_START, &dev->state);
5714 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5715 * because users of this 'device' dont need to change
5721 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5725 * register_netdev - register a network device
5726 * @dev: device to register
5728 * Take a completed network device structure and add it to the kernel
5729 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5730 * chain. 0 is returned on success. A negative errno code is returned
5731 * on a failure to set up the device, or if the name is a duplicate.
5733 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5734 * and expands the device name if you passed a format string to
5737 int register_netdev(struct net_device *dev)
5742 err = register_netdevice(dev);
5746 EXPORT_SYMBOL(register_netdev);
5748 int netdev_refcnt_read(const struct net_device *dev)
5752 for_each_possible_cpu(i)
5753 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5756 EXPORT_SYMBOL(netdev_refcnt_read);
5759 * netdev_wait_allrefs - wait until all references are gone.
5761 * This is called when unregistering network devices.
5763 * Any protocol or device that holds a reference should register
5764 * for netdevice notification, and cleanup and put back the
5765 * reference if they receive an UNREGISTER event.
5766 * We can get stuck here if buggy protocols don't correctly
5769 static void netdev_wait_allrefs(struct net_device *dev)
5771 unsigned long rebroadcast_time, warning_time;
5774 linkwatch_forget_dev(dev);
5776 rebroadcast_time = warning_time = jiffies;
5777 refcnt = netdev_refcnt_read(dev);
5779 while (refcnt != 0) {
5780 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5783 /* Rebroadcast unregister notification */
5784 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5785 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5786 * should have already handle it the first time */
5788 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5790 /* We must not have linkwatch events
5791 * pending on unregister. If this
5792 * happens, we simply run the queue
5793 * unscheduled, resulting in a noop
5796 linkwatch_run_queue();
5801 rebroadcast_time = jiffies;
5806 refcnt = netdev_refcnt_read(dev);
5808 if (time_after(jiffies, warning_time + 10 * HZ)) {
5809 printk(KERN_EMERG "unregister_netdevice: "
5810 "waiting for %s to become free. Usage "
5813 warning_time = jiffies;
5822 * register_netdevice(x1);
5823 * register_netdevice(x2);
5825 * unregister_netdevice(y1);
5826 * unregister_netdevice(y2);
5832 * We are invoked by rtnl_unlock().
5833 * This allows us to deal with problems:
5834 * 1) We can delete sysfs objects which invoke hotplug
5835 * without deadlocking with linkwatch via keventd.
5836 * 2) Since we run with the RTNL semaphore not held, we can sleep
5837 * safely in order to wait for the netdev refcnt to drop to zero.
5839 * We must not return until all unregister events added during
5840 * the interval the lock was held have been completed.
5842 void netdev_run_todo(void)
5844 struct list_head list;
5846 /* Snapshot list, allow later requests */
5847 list_replace_init(&net_todo_list, &list);
5851 /* Wait for rcu callbacks to finish before attempting to drain
5852 * the device list. This usually avoids a 250ms wait.
5854 if (!list_empty(&list))
5857 while (!list_empty(&list)) {
5858 struct net_device *dev
5859 = list_first_entry(&list, struct net_device, todo_list);
5860 list_del(&dev->todo_list);
5862 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5863 printk(KERN_ERR "network todo '%s' but state %d\n",
5864 dev->name, dev->reg_state);
5869 dev->reg_state = NETREG_UNREGISTERED;
5871 on_each_cpu(flush_backlog, dev, 1);
5873 netdev_wait_allrefs(dev);
5876 BUG_ON(netdev_refcnt_read(dev));
5877 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5878 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5879 WARN_ON(dev->dn_ptr);
5881 if (dev->destructor)
5882 dev->destructor(dev);
5884 /* Free network device */
5885 kobject_put(&dev->dev.kobj);
5889 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5890 * fields in the same order, with only the type differing.
5892 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5893 const struct net_device_stats *netdev_stats)
5895 #if BITS_PER_LONG == 64
5896 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5897 memcpy(stats64, netdev_stats, sizeof(*stats64));
5899 size_t i, n = sizeof(*stats64) / sizeof(u64);
5900 const unsigned long *src = (const unsigned long *)netdev_stats;
5901 u64 *dst = (u64 *)stats64;
5903 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5904 sizeof(*stats64) / sizeof(u64));
5905 for (i = 0; i < n; i++)
5911 * dev_get_stats - get network device statistics
5912 * @dev: device to get statistics from
5913 * @storage: place to store stats
5915 * Get network statistics from device. Return @storage.
5916 * The device driver may provide its own method by setting
5917 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5918 * otherwise the internal statistics structure is used.
5920 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5921 struct rtnl_link_stats64 *storage)
5923 const struct net_device_ops *ops = dev->netdev_ops;
5925 if (ops->ndo_get_stats64) {
5926 memset(storage, 0, sizeof(*storage));
5927 ops->ndo_get_stats64(dev, storage);
5928 } else if (ops->ndo_get_stats) {
5929 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5931 netdev_stats_to_stats64(storage, &dev->stats);
5933 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5936 EXPORT_SYMBOL(dev_get_stats);
5938 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5940 struct netdev_queue *queue = dev_ingress_queue(dev);
5942 #ifdef CONFIG_NET_CLS_ACT
5945 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5948 netdev_init_one_queue(dev, queue, NULL);
5949 queue->qdisc = &noop_qdisc;
5950 queue->qdisc_sleeping = &noop_qdisc;
5951 rcu_assign_pointer(dev->ingress_queue, queue);
5957 * alloc_netdev_mqs - allocate network device
5958 * @sizeof_priv: size of private data to allocate space for
5959 * @name: device name format string
5960 * @setup: callback to initialize device
5961 * @txqs: the number of TX subqueues to allocate
5962 * @rxqs: the number of RX subqueues to allocate
5964 * Allocates a struct net_device with private data area for driver use
5965 * and performs basic initialization. Also allocates subquue structs
5966 * for each queue on the device.
5968 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5969 void (*setup)(struct net_device *),
5970 unsigned int txqs, unsigned int rxqs)
5972 struct net_device *dev;
5974 struct net_device *p;
5976 BUG_ON(strlen(name) >= sizeof(dev->name));
5979 pr_err("alloc_netdev: Unable to allocate device "
5980 "with zero queues.\n");
5986 pr_err("alloc_netdev: Unable to allocate device "
5987 "with zero RX queues.\n");
5992 alloc_size = sizeof(struct net_device);
5994 /* ensure 32-byte alignment of private area */
5995 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5996 alloc_size += sizeof_priv;
5998 /* ensure 32-byte alignment of whole construct */
5999 alloc_size += NETDEV_ALIGN - 1;
6001 p = kzalloc(alloc_size, GFP_KERNEL);
6003 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
6007 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6008 dev->padded = (char *)dev - (char *)p;
6010 dev->pcpu_refcnt = alloc_percpu(int);
6011 if (!dev->pcpu_refcnt)
6014 if (dev_addr_init(dev))
6020 dev_net_set(dev, &init_net);
6022 dev->gso_max_size = GSO_MAX_SIZE;
6024 INIT_LIST_HEAD(&dev->napi_list);
6025 INIT_LIST_HEAD(&dev->unreg_list);
6026 INIT_LIST_HEAD(&dev->link_watch_list);
6027 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6030 dev->num_tx_queues = txqs;
6031 dev->real_num_tx_queues = txqs;
6032 if (netif_alloc_netdev_queues(dev))
6036 dev->num_rx_queues = rxqs;
6037 dev->real_num_rx_queues = rxqs;
6038 if (netif_alloc_rx_queues(dev))
6042 strcpy(dev->name, name);
6043 dev->group = INIT_NETDEV_GROUP;
6051 free_percpu(dev->pcpu_refcnt);
6061 EXPORT_SYMBOL(alloc_netdev_mqs);
6064 * free_netdev - free network device
6067 * This function does the last stage of destroying an allocated device
6068 * interface. The reference to the device object is released.
6069 * If this is the last reference then it will be freed.
6071 void free_netdev(struct net_device *dev)
6073 struct napi_struct *p, *n;
6075 release_net(dev_net(dev));
6082 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6084 /* Flush device addresses */
6085 dev_addr_flush(dev);
6087 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6090 free_percpu(dev->pcpu_refcnt);
6091 dev->pcpu_refcnt = NULL;
6093 /* Compatibility with error handling in drivers */
6094 if (dev->reg_state == NETREG_UNINITIALIZED) {
6095 kfree((char *)dev - dev->padded);
6099 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6100 dev->reg_state = NETREG_RELEASED;
6102 /* will free via device release */
6103 put_device(&dev->dev);
6105 EXPORT_SYMBOL(free_netdev);
6108 * synchronize_net - Synchronize with packet receive processing
6110 * Wait for packets currently being received to be done.
6111 * Does not block later packets from starting.
6113 void synchronize_net(void)
6116 if (rtnl_is_locked())
6117 synchronize_rcu_expedited();
6121 EXPORT_SYMBOL(synchronize_net);
6124 * unregister_netdevice_queue - remove device from the kernel
6128 * This function shuts down a device interface and removes it
6129 * from the kernel tables.
6130 * If head not NULL, device is queued to be unregistered later.
6132 * Callers must hold the rtnl semaphore. You may want
6133 * unregister_netdev() instead of this.
6136 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6141 list_move_tail(&dev->unreg_list, head);
6143 rollback_registered(dev);
6144 /* Finish processing unregister after unlock */
6148 EXPORT_SYMBOL(unregister_netdevice_queue);
6151 * unregister_netdevice_many - unregister many devices
6152 * @head: list of devices
6154 void unregister_netdevice_many(struct list_head *head)
6156 struct net_device *dev;
6158 if (!list_empty(head)) {
6159 rollback_registered_many(head);
6160 list_for_each_entry(dev, head, unreg_list)
6164 EXPORT_SYMBOL(unregister_netdevice_many);
6167 * unregister_netdev - remove device from the kernel
6170 * This function shuts down a device interface and removes it
6171 * from the kernel tables.
6173 * This is just a wrapper for unregister_netdevice that takes
6174 * the rtnl semaphore. In general you want to use this and not
6175 * unregister_netdevice.
6177 void unregister_netdev(struct net_device *dev)
6180 unregister_netdevice(dev);
6183 EXPORT_SYMBOL(unregister_netdev);
6186 * dev_change_net_namespace - move device to different nethost namespace
6188 * @net: network namespace
6189 * @pat: If not NULL name pattern to try if the current device name
6190 * is already taken in the destination network namespace.
6192 * This function shuts down a device interface and moves it
6193 * to a new network namespace. On success 0 is returned, on
6194 * a failure a netagive errno code is returned.
6196 * Callers must hold the rtnl semaphore.
6199 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6205 /* Don't allow namespace local devices to be moved. */
6207 if (dev->features & NETIF_F_NETNS_LOCAL)
6210 /* Ensure the device has been registrered */
6212 if (dev->reg_state != NETREG_REGISTERED)
6215 /* Get out if there is nothing todo */
6217 if (net_eq(dev_net(dev), net))
6220 /* Pick the destination device name, and ensure
6221 * we can use it in the destination network namespace.
6224 if (__dev_get_by_name(net, dev->name)) {
6225 /* We get here if we can't use the current device name */
6228 if (dev_get_valid_name(dev, pat) < 0)
6233 * And now a mini version of register_netdevice unregister_netdevice.
6236 /* If device is running close it first. */
6239 /* And unlink it from device chain */
6241 unlist_netdevice(dev);
6245 /* Shutdown queueing discipline. */
6248 /* Notify protocols, that we are about to destroy
6249 this device. They should clean all the things.
6251 Note that dev->reg_state stays at NETREG_REGISTERED.
6252 This is wanted because this way 8021q and macvlan know
6253 the device is just moving and can keep their slaves up.
6255 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6256 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6257 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6260 * Flush the unicast and multicast chains
6265 /* Actually switch the network namespace */
6266 dev_net_set(dev, net);
6268 /* If there is an ifindex conflict assign a new one */
6269 if (__dev_get_by_index(net, dev->ifindex)) {
6270 int iflink = (dev->iflink == dev->ifindex);
6271 dev->ifindex = dev_new_index(net);
6273 dev->iflink = dev->ifindex;
6276 /* Fixup kobjects */
6277 err = device_rename(&dev->dev, dev->name);
6280 /* Add the device back in the hashes */
6281 list_netdevice(dev);
6283 /* Notify protocols, that a new device appeared. */
6284 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6287 * Prevent userspace races by waiting until the network
6288 * device is fully setup before sending notifications.
6290 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6297 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6299 static int dev_cpu_callback(struct notifier_block *nfb,
6300 unsigned long action,
6303 struct sk_buff **list_skb;
6304 struct sk_buff *skb;
6305 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6306 struct softnet_data *sd, *oldsd;
6308 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6311 local_irq_disable();
6312 cpu = smp_processor_id();
6313 sd = &per_cpu(softnet_data, cpu);
6314 oldsd = &per_cpu(softnet_data, oldcpu);
6316 /* Find end of our completion_queue. */
6317 list_skb = &sd->completion_queue;
6319 list_skb = &(*list_skb)->next;
6320 /* Append completion queue from offline CPU. */
6321 *list_skb = oldsd->completion_queue;
6322 oldsd->completion_queue = NULL;
6324 /* Append output queue from offline CPU. */
6325 if (oldsd->output_queue) {
6326 *sd->output_queue_tailp = oldsd->output_queue;
6327 sd->output_queue_tailp = oldsd->output_queue_tailp;
6328 oldsd->output_queue = NULL;
6329 oldsd->output_queue_tailp = &oldsd->output_queue;
6331 /* Append NAPI poll list from offline CPU. */
6332 if (!list_empty(&oldsd->poll_list)) {
6333 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6334 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6337 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6340 /* Process offline CPU's input_pkt_queue */
6341 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6343 input_queue_head_incr(oldsd);
6345 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6347 input_queue_head_incr(oldsd);
6355 * netdev_increment_features - increment feature set by one
6356 * @all: current feature set
6357 * @one: new feature set
6358 * @mask: mask feature set
6360 * Computes a new feature set after adding a device with feature set
6361 * @one to the master device with current feature set @all. Will not
6362 * enable anything that is off in @mask. Returns the new feature set.
6364 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6366 if (mask & NETIF_F_GEN_CSUM)
6367 mask |= NETIF_F_ALL_CSUM;
6368 mask |= NETIF_F_VLAN_CHALLENGED;
6370 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6371 all &= one | ~NETIF_F_ALL_FOR_ALL;
6373 /* If device needs checksumming, downgrade to it. */
6374 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6375 all &= ~NETIF_F_NO_CSUM;
6377 /* If one device supports hw checksumming, set for all. */
6378 if (all & NETIF_F_GEN_CSUM)
6379 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6383 EXPORT_SYMBOL(netdev_increment_features);
6385 static struct hlist_head *netdev_create_hash(void)
6388 struct hlist_head *hash;
6390 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6392 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6393 INIT_HLIST_HEAD(&hash[i]);
6398 /* Initialize per network namespace state */
6399 static int __net_init netdev_init(struct net *net)
6401 INIT_LIST_HEAD(&net->dev_base_head);
6403 net->dev_name_head = netdev_create_hash();
6404 if (net->dev_name_head == NULL)
6407 net->dev_index_head = netdev_create_hash();
6408 if (net->dev_index_head == NULL)
6414 kfree(net->dev_name_head);
6420 * netdev_drivername - network driver for the device
6421 * @dev: network device
6423 * Determine network driver for device.
6425 const char *netdev_drivername(const struct net_device *dev)
6427 const struct device_driver *driver;
6428 const struct device *parent;
6429 const char *empty = "";
6431 parent = dev->dev.parent;
6435 driver = parent->driver;
6436 if (driver && driver->name)
6437 return driver->name;
6441 int __netdev_printk(const char *level, const struct net_device *dev,
6442 struct va_format *vaf)
6446 if (dev && dev->dev.parent)
6447 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6448 netdev_name(dev), vaf);
6450 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6452 r = printk("%s(NULL net_device): %pV", level, vaf);
6456 EXPORT_SYMBOL(__netdev_printk);
6458 int netdev_printk(const char *level, const struct net_device *dev,
6459 const char *format, ...)
6461 struct va_format vaf;
6465 va_start(args, format);
6470 r = __netdev_printk(level, dev, &vaf);
6475 EXPORT_SYMBOL(netdev_printk);
6477 #define define_netdev_printk_level(func, level) \
6478 int func(const struct net_device *dev, const char *fmt, ...) \
6481 struct va_format vaf; \
6484 va_start(args, fmt); \
6489 r = __netdev_printk(level, dev, &vaf); \
6494 EXPORT_SYMBOL(func);
6496 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6497 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6498 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6499 define_netdev_printk_level(netdev_err, KERN_ERR);
6500 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6501 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6502 define_netdev_printk_level(netdev_info, KERN_INFO);
6504 static void __net_exit netdev_exit(struct net *net)
6506 kfree(net->dev_name_head);
6507 kfree(net->dev_index_head);
6510 static struct pernet_operations __net_initdata netdev_net_ops = {
6511 .init = netdev_init,
6512 .exit = netdev_exit,
6515 static void __net_exit default_device_exit(struct net *net)
6517 struct net_device *dev, *aux;
6519 * Push all migratable network devices back to the
6520 * initial network namespace
6523 for_each_netdev_safe(net, dev, aux) {
6525 char fb_name[IFNAMSIZ];
6527 /* Ignore unmoveable devices (i.e. loopback) */
6528 if (dev->features & NETIF_F_NETNS_LOCAL)
6531 /* Leave virtual devices for the generic cleanup */
6532 if (dev->rtnl_link_ops)
6535 /* Push remaining network devices to init_net */
6536 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6537 err = dev_change_net_namespace(dev, &init_net, fb_name);
6539 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6540 __func__, dev->name, err);
6547 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6549 /* At exit all network devices most be removed from a network
6550 * namespace. Do this in the reverse order of registration.
6551 * Do this across as many network namespaces as possible to
6552 * improve batching efficiency.
6554 struct net_device *dev;
6556 LIST_HEAD(dev_kill_list);
6559 list_for_each_entry(net, net_list, exit_list) {
6560 for_each_netdev_reverse(net, dev) {
6561 if (dev->rtnl_link_ops)
6562 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6564 unregister_netdevice_queue(dev, &dev_kill_list);
6567 unregister_netdevice_many(&dev_kill_list);
6568 list_del(&dev_kill_list);
6572 static struct pernet_operations __net_initdata default_device_ops = {
6573 .exit = default_device_exit,
6574 .exit_batch = default_device_exit_batch,
6578 * Initialize the DEV module. At boot time this walks the device list and
6579 * unhooks any devices that fail to initialise (normally hardware not
6580 * present) and leaves us with a valid list of present and active devices.
6585 * This is called single threaded during boot, so no need
6586 * to take the rtnl semaphore.
6588 static int __init net_dev_init(void)
6590 int i, rc = -ENOMEM;
6592 BUG_ON(!dev_boot_phase);
6594 if (dev_proc_init())
6597 if (netdev_kobject_init())
6600 INIT_LIST_HEAD(&ptype_all);
6601 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6602 INIT_LIST_HEAD(&ptype_base[i]);
6604 if (register_pernet_subsys(&netdev_net_ops))
6608 * Initialise the packet receive queues.
6611 for_each_possible_cpu(i) {
6612 struct softnet_data *sd = &per_cpu(softnet_data, i);
6614 memset(sd, 0, sizeof(*sd));
6615 skb_queue_head_init(&sd->input_pkt_queue);
6616 skb_queue_head_init(&sd->process_queue);
6617 sd->completion_queue = NULL;
6618 INIT_LIST_HEAD(&sd->poll_list);
6619 sd->output_queue = NULL;
6620 sd->output_queue_tailp = &sd->output_queue;
6622 sd->csd.func = rps_trigger_softirq;
6628 sd->backlog.poll = process_backlog;
6629 sd->backlog.weight = weight_p;
6630 sd->backlog.gro_list = NULL;
6631 sd->backlog.gro_count = 0;
6636 /* The loopback device is special if any other network devices
6637 * is present in a network namespace the loopback device must
6638 * be present. Since we now dynamically allocate and free the
6639 * loopback device ensure this invariant is maintained by
6640 * keeping the loopback device as the first device on the
6641 * list of network devices. Ensuring the loopback devices
6642 * is the first device that appears and the last network device
6645 if (register_pernet_device(&loopback_net_ops))
6648 if (register_pernet_device(&default_device_ops))
6651 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6652 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6654 hotcpu_notifier(dev_cpu_callback, 0);
6662 subsys_initcall(net_dev_init);
6664 static int __init initialize_hashrnd(void)
6666 get_random_bytes(&hashrnd, sizeof(hashrnd));
6670 late_initcall_sync(initialize_hashrnd);