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/net_tstamp.h>
137 #include <linux/jump_label.h>
138 #include <net/flow_keys.h>
140 #include "net-sysfs.h"
142 /* Instead of increasing this, you should create a hash table. */
143 #define MAX_GRO_SKBS 8
145 /* This should be increased if a protocol with a bigger head is added. */
146 #define GRO_MAX_HEAD (MAX_HEADER + 128)
149 * The list of packet types we will receive (as opposed to discard)
150 * and the routines to invoke.
152 * Why 16. Because with 16 the only overlap we get on a hash of the
153 * low nibble of the protocol value is RARP/SNAP/X.25.
155 * NOTE: That is no longer true with the addition of VLAN tags. Not
156 * sure which should go first, but I bet it won't make much
157 * difference if we are running VLANs. The good news is that
158 * this protocol won't be in the list unless compiled in, so
159 * the average user (w/out VLANs) will not be adversely affected.
176 #define PTYPE_HASH_SIZE (16)
177 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
179 static DEFINE_SPINLOCK(ptype_lock);
180 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
181 static struct list_head ptype_all __read_mostly; /* Taps */
184 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
187 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
189 * Writers must hold the rtnl semaphore while they loop through the
190 * dev_base_head list, and hold dev_base_lock for writing when they do the
191 * actual updates. This allows pure readers to access the list even
192 * while a writer is preparing to update it.
194 * To put it another way, dev_base_lock is held for writing only to
195 * protect against pure readers; the rtnl semaphore provides the
196 * protection against other writers.
198 * See, for example usages, register_netdevice() and
199 * unregister_netdevice(), which must be called with the rtnl
202 DEFINE_RWLOCK(dev_base_lock);
203 EXPORT_SYMBOL(dev_base_lock);
205 static inline void dev_base_seq_inc(struct net *net)
207 while (++net->dev_base_seq == 0);
210 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
212 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
213 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
216 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
218 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
221 static inline void rps_lock(struct softnet_data *sd)
224 spin_lock(&sd->input_pkt_queue.lock);
228 static inline void rps_unlock(struct softnet_data *sd)
231 spin_unlock(&sd->input_pkt_queue.lock);
235 /* Device list insertion */
236 static int list_netdevice(struct net_device *dev)
238 struct net *net = dev_net(dev);
242 write_lock_bh(&dev_base_lock);
243 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
244 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
245 hlist_add_head_rcu(&dev->index_hlist,
246 dev_index_hash(net, dev->ifindex));
247 write_unlock_bh(&dev_base_lock);
249 dev_base_seq_inc(net);
254 /* Device list removal
255 * caller must respect a RCU grace period before freeing/reusing dev
257 static void unlist_netdevice(struct net_device *dev)
261 /* Unlink dev from the device chain */
262 write_lock_bh(&dev_base_lock);
263 list_del_rcu(&dev->dev_list);
264 hlist_del_rcu(&dev->name_hlist);
265 hlist_del_rcu(&dev->index_hlist);
266 write_unlock_bh(&dev_base_lock);
268 dev_base_seq_inc(dev_net(dev));
275 static RAW_NOTIFIER_HEAD(netdev_chain);
278 * Device drivers call our routines to queue packets here. We empty the
279 * queue in the local softnet handler.
282 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
283 EXPORT_PER_CPU_SYMBOL(softnet_data);
285 #ifdef CONFIG_LOCKDEP
287 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
288 * according to dev->type
290 static const unsigned short netdev_lock_type[] =
291 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
292 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
293 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
294 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
295 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
296 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
297 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
298 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
299 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
300 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
301 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
302 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
303 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
304 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
305 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
306 ARPHRD_VOID, ARPHRD_NONE};
308 static const char *const netdev_lock_name[] =
309 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
310 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
311 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
312 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
313 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
314 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
315 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
316 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
317 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
318 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
319 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
320 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
321 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
322 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
323 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
324 "_xmit_VOID", "_xmit_NONE"};
326 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
327 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
329 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
333 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
334 if (netdev_lock_type[i] == dev_type)
336 /* the last key is used by default */
337 return ARRAY_SIZE(netdev_lock_type) - 1;
340 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
341 unsigned short dev_type)
345 i = netdev_lock_pos(dev_type);
346 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
347 netdev_lock_name[i]);
350 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 i = netdev_lock_pos(dev->type);
355 lockdep_set_class_and_name(&dev->addr_list_lock,
356 &netdev_addr_lock_key[i],
357 netdev_lock_name[i]);
360 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
361 unsigned short dev_type)
364 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
369 /*******************************************************************************
371 Protocol management and registration routines
373 *******************************************************************************/
376 * Add a protocol ID to the list. Now that the input handler is
377 * smarter we can dispense with all the messy stuff that used to be
380 * BEWARE!!! Protocol handlers, mangling input packets,
381 * MUST BE last in hash buckets and checking protocol handlers
382 * MUST start from promiscuous ptype_all chain in net_bh.
383 * It is true now, do not change it.
384 * Explanation follows: if protocol handler, mangling packet, will
385 * be the first on list, it is not able to sense, that packet
386 * is cloned and should be copied-on-write, so that it will
387 * change it and subsequent readers will get broken packet.
391 static inline struct list_head *ptype_head(const struct packet_type *pt)
393 if (pt->type == htons(ETH_P_ALL))
396 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
400 * dev_add_pack - add packet handler
401 * @pt: packet type declaration
403 * Add a protocol handler to the networking stack. The passed &packet_type
404 * is linked into kernel lists and may not be freed until it has been
405 * removed from the kernel lists.
407 * This call does not sleep therefore it can not
408 * guarantee all CPU's that are in middle of receiving packets
409 * will see the new packet type (until the next received packet).
412 void dev_add_pack(struct packet_type *pt)
414 struct list_head *head = ptype_head(pt);
416 spin_lock(&ptype_lock);
417 list_add_rcu(&pt->list, head);
418 spin_unlock(&ptype_lock);
420 EXPORT_SYMBOL(dev_add_pack);
423 * __dev_remove_pack - remove packet handler
424 * @pt: packet type declaration
426 * Remove a protocol handler that was previously added to the kernel
427 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
428 * from the kernel lists and can be freed or reused once this function
431 * The packet type might still be in use by receivers
432 * and must not be freed until after all the CPU's have gone
433 * through a quiescent state.
435 void __dev_remove_pack(struct packet_type *pt)
437 struct list_head *head = ptype_head(pt);
438 struct packet_type *pt1;
440 spin_lock(&ptype_lock);
442 list_for_each_entry(pt1, head, list) {
444 list_del_rcu(&pt->list);
449 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
451 spin_unlock(&ptype_lock);
453 EXPORT_SYMBOL(__dev_remove_pack);
456 * dev_remove_pack - remove packet handler
457 * @pt: packet type declaration
459 * Remove a protocol handler that was previously added to the kernel
460 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
461 * from the kernel lists and can be freed or reused once this function
464 * This call sleeps to guarantee that no CPU is looking at the packet
467 void dev_remove_pack(struct packet_type *pt)
469 __dev_remove_pack(pt);
473 EXPORT_SYMBOL(dev_remove_pack);
475 /******************************************************************************
477 Device Boot-time Settings Routines
479 *******************************************************************************/
481 /* Boot time configuration table */
482 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
485 * netdev_boot_setup_add - add new setup entry
486 * @name: name of the device
487 * @map: configured settings for the device
489 * Adds new setup entry to the dev_boot_setup list. The function
490 * returns 0 on error and 1 on success. This is a generic routine to
493 static int netdev_boot_setup_add(char *name, struct ifmap *map)
495 struct netdev_boot_setup *s;
499 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
500 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
501 memset(s[i].name, 0, sizeof(s[i].name));
502 strlcpy(s[i].name, name, IFNAMSIZ);
503 memcpy(&s[i].map, map, sizeof(s[i].map));
508 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
512 * netdev_boot_setup_check - check boot time settings
513 * @dev: the netdevice
515 * Check boot time settings for the device.
516 * The found settings are set for the device to be used
517 * later in the device probing.
518 * Returns 0 if no settings found, 1 if they are.
520 int netdev_boot_setup_check(struct net_device *dev)
522 struct netdev_boot_setup *s = dev_boot_setup;
525 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
526 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
527 !strcmp(dev->name, s[i].name)) {
528 dev->irq = s[i].map.irq;
529 dev->base_addr = s[i].map.base_addr;
530 dev->mem_start = s[i].map.mem_start;
531 dev->mem_end = s[i].map.mem_end;
537 EXPORT_SYMBOL(netdev_boot_setup_check);
541 * netdev_boot_base - get address from boot time settings
542 * @prefix: prefix for network device
543 * @unit: id for network device
545 * Check boot time settings for the base address of device.
546 * The found settings are set for the device to be used
547 * later in the device probing.
548 * Returns 0 if no settings found.
550 unsigned long netdev_boot_base(const char *prefix, int unit)
552 const struct netdev_boot_setup *s = dev_boot_setup;
556 sprintf(name, "%s%d", prefix, unit);
559 * If device already registered then return base of 1
560 * to indicate not to probe for this interface
562 if (__dev_get_by_name(&init_net, name))
565 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
566 if (!strcmp(name, s[i].name))
567 return s[i].map.base_addr;
572 * Saves at boot time configured settings for any netdevice.
574 int __init netdev_boot_setup(char *str)
579 str = get_options(str, ARRAY_SIZE(ints), ints);
584 memset(&map, 0, sizeof(map));
588 map.base_addr = ints[2];
590 map.mem_start = ints[3];
592 map.mem_end = ints[4];
594 /* Add new entry to the list */
595 return netdev_boot_setup_add(str, &map);
598 __setup("netdev=", netdev_boot_setup);
600 /*******************************************************************************
602 Device Interface Subroutines
604 *******************************************************************************/
607 * __dev_get_by_name - find a device by its name
608 * @net: the applicable net namespace
609 * @name: name to find
611 * Find an interface by name. Must be called under RTNL semaphore
612 * or @dev_base_lock. If the name is found a pointer to the device
613 * is returned. If the name is not found then %NULL is returned. The
614 * reference counters are not incremented so the caller must be
615 * careful with locks.
618 struct net_device *__dev_get_by_name(struct net *net, const char *name)
620 struct hlist_node *p;
621 struct net_device *dev;
622 struct hlist_head *head = dev_name_hash(net, name);
624 hlist_for_each_entry(dev, p, head, name_hlist)
625 if (!strncmp(dev->name, name, IFNAMSIZ))
630 EXPORT_SYMBOL(__dev_get_by_name);
633 * dev_get_by_name_rcu - find a device by its name
634 * @net: the applicable net namespace
635 * @name: name to find
637 * Find an interface by name.
638 * If the name is found a pointer to the device is returned.
639 * If the name is not found then %NULL is returned.
640 * The reference counters are not incremented so the caller must be
641 * careful with locks. The caller must hold RCU lock.
644 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
646 struct hlist_node *p;
647 struct net_device *dev;
648 struct hlist_head *head = dev_name_hash(net, name);
650 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
651 if (!strncmp(dev->name, name, IFNAMSIZ))
656 EXPORT_SYMBOL(dev_get_by_name_rcu);
659 * dev_get_by_name - find a device by its name
660 * @net: the applicable net namespace
661 * @name: name to find
663 * Find an interface by name. This can be called from any
664 * context and does its own locking. The returned handle has
665 * the usage count incremented and the caller must use dev_put() to
666 * release it when it is no longer needed. %NULL is returned if no
667 * matching device is found.
670 struct net_device *dev_get_by_name(struct net *net, const char *name)
672 struct net_device *dev;
675 dev = dev_get_by_name_rcu(net, name);
681 EXPORT_SYMBOL(dev_get_by_name);
684 * __dev_get_by_index - find a device by its ifindex
685 * @net: the applicable net namespace
686 * @ifindex: index of device
688 * Search for an interface by index. Returns %NULL if the device
689 * is not found or a pointer to the device. The device has not
690 * had its reference counter increased so the caller must be careful
691 * about locking. The caller must hold either the RTNL semaphore
695 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
697 struct hlist_node *p;
698 struct net_device *dev;
699 struct hlist_head *head = dev_index_hash(net, ifindex);
701 hlist_for_each_entry(dev, p, head, index_hlist)
702 if (dev->ifindex == ifindex)
707 EXPORT_SYMBOL(__dev_get_by_index);
710 * dev_get_by_index_rcu - find a device by its ifindex
711 * @net: the applicable net namespace
712 * @ifindex: index of device
714 * Search for an interface by index. Returns %NULL if the device
715 * is not found or a pointer to the device. The device has not
716 * had its reference counter increased so the caller must be careful
717 * about locking. The caller must hold RCU lock.
720 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
722 struct hlist_node *p;
723 struct net_device *dev;
724 struct hlist_head *head = dev_index_hash(net, ifindex);
726 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
727 if (dev->ifindex == ifindex)
732 EXPORT_SYMBOL(dev_get_by_index_rcu);
736 * dev_get_by_index - find a device by its ifindex
737 * @net: the applicable net namespace
738 * @ifindex: index of device
740 * Search for an interface by index. Returns NULL if the device
741 * is not found or a pointer to the device. The device returned has
742 * had a reference added and the pointer is safe until the user calls
743 * dev_put to indicate they have finished with it.
746 struct net_device *dev_get_by_index(struct net *net, int ifindex)
748 struct net_device *dev;
751 dev = dev_get_by_index_rcu(net, ifindex);
757 EXPORT_SYMBOL(dev_get_by_index);
760 * dev_getbyhwaddr_rcu - find a device by its hardware address
761 * @net: the applicable net namespace
762 * @type: media type of device
763 * @ha: hardware address
765 * Search for an interface by MAC address. Returns NULL if the device
766 * is not found or a pointer to the device.
767 * The caller must hold RCU or RTNL.
768 * The returned device has not had its ref count increased
769 * and the caller must therefore be careful about locking
773 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
776 struct net_device *dev;
778 for_each_netdev_rcu(net, dev)
779 if (dev->type == type &&
780 !memcmp(dev->dev_addr, ha, dev->addr_len))
785 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
787 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
789 struct net_device *dev;
792 for_each_netdev(net, dev)
793 if (dev->type == type)
798 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
800 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
802 struct net_device *dev, *ret = NULL;
805 for_each_netdev_rcu(net, dev)
806 if (dev->type == type) {
814 EXPORT_SYMBOL(dev_getfirstbyhwtype);
817 * dev_get_by_flags_rcu - find any device with given flags
818 * @net: the applicable net namespace
819 * @if_flags: IFF_* values
820 * @mask: bitmask of bits in if_flags to check
822 * Search for any interface with the given flags. Returns NULL if a device
823 * is not found or a pointer to the device. Must be called inside
824 * rcu_read_lock(), and result refcount is unchanged.
827 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
830 struct net_device *dev, *ret;
833 for_each_netdev_rcu(net, dev) {
834 if (((dev->flags ^ if_flags) & mask) == 0) {
841 EXPORT_SYMBOL(dev_get_by_flags_rcu);
844 * dev_valid_name - check if name is okay for network device
847 * Network device names need to be valid file names to
848 * to allow sysfs to work. We also disallow any kind of
851 int dev_valid_name(const char *name)
855 if (strlen(name) >= IFNAMSIZ)
857 if (!strcmp(name, ".") || !strcmp(name, ".."))
861 if (*name == '/' || isspace(*name))
867 EXPORT_SYMBOL(dev_valid_name);
870 * __dev_alloc_name - allocate a name for a device
871 * @net: network namespace to allocate the device name in
872 * @name: name format string
873 * @buf: scratch buffer and result name string
875 * Passed a format string - eg "lt%d" it will try and find a suitable
876 * id. It scans list of devices to build up a free map, then chooses
877 * the first empty slot. The caller must hold the dev_base or rtnl lock
878 * while allocating the name and adding the device in order to avoid
880 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
881 * Returns the number of the unit assigned or a negative errno code.
884 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
888 const int max_netdevices = 8*PAGE_SIZE;
889 unsigned long *inuse;
890 struct net_device *d;
892 p = strnchr(name, IFNAMSIZ-1, '%');
895 * Verify the string as this thing may have come from
896 * the user. There must be either one "%d" and no other "%"
899 if (p[1] != 'd' || strchr(p + 2, '%'))
902 /* Use one page as a bit array of possible slots */
903 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
907 for_each_netdev(net, d) {
908 if (!sscanf(d->name, name, &i))
910 if (i < 0 || i >= max_netdevices)
913 /* avoid cases where sscanf is not exact inverse of printf */
914 snprintf(buf, IFNAMSIZ, name, i);
915 if (!strncmp(buf, d->name, IFNAMSIZ))
919 i = find_first_zero_bit(inuse, max_netdevices);
920 free_page((unsigned long) inuse);
924 snprintf(buf, IFNAMSIZ, name, i);
925 if (!__dev_get_by_name(net, buf))
928 /* It is possible to run out of possible slots
929 * when the name is long and there isn't enough space left
930 * for the digits, or if all bits are used.
936 * dev_alloc_name - allocate a name for a device
938 * @name: name format string
940 * Passed a format string - eg "lt%d" it will try and find a suitable
941 * id. It scans list of devices to build up a free map, then chooses
942 * the first empty slot. The caller must hold the dev_base or rtnl lock
943 * while allocating the name and adding the device in order to avoid
945 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
946 * Returns the number of the unit assigned or a negative errno code.
949 int dev_alloc_name(struct net_device *dev, const char *name)
955 BUG_ON(!dev_net(dev));
957 ret = __dev_alloc_name(net, name, buf);
959 strlcpy(dev->name, buf, IFNAMSIZ);
962 EXPORT_SYMBOL(dev_alloc_name);
964 static int dev_get_valid_name(struct net_device *dev, const char *name)
968 BUG_ON(!dev_net(dev));
971 if (!dev_valid_name(name))
974 if (strchr(name, '%'))
975 return dev_alloc_name(dev, name);
976 else if (__dev_get_by_name(net, name))
978 else if (dev->name != name)
979 strlcpy(dev->name, name, IFNAMSIZ);
985 * dev_change_name - change name of a device
987 * @newname: name (or format string) must be at least IFNAMSIZ
989 * Change name of a device, can pass format strings "eth%d".
992 int dev_change_name(struct net_device *dev, const char *newname)
994 char oldname[IFNAMSIZ];
1000 BUG_ON(!dev_net(dev));
1003 if (dev->flags & IFF_UP)
1006 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1009 memcpy(oldname, dev->name, IFNAMSIZ);
1011 err = dev_get_valid_name(dev, newname);
1016 ret = device_rename(&dev->dev, dev->name);
1018 memcpy(dev->name, oldname, IFNAMSIZ);
1022 write_lock_bh(&dev_base_lock);
1023 hlist_del_rcu(&dev->name_hlist);
1024 write_unlock_bh(&dev_base_lock);
1028 write_lock_bh(&dev_base_lock);
1029 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1030 write_unlock_bh(&dev_base_lock);
1032 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1033 ret = notifier_to_errno(ret);
1036 /* err >= 0 after dev_alloc_name() or stores the first errno */
1039 memcpy(dev->name, oldname, IFNAMSIZ);
1043 "%s: name change rollback failed: %d.\n",
1052 * dev_set_alias - change ifalias of a device
1054 * @alias: name up to IFALIASZ
1055 * @len: limit of bytes to copy from info
1057 * Set ifalias for a device,
1059 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1063 if (len >= IFALIASZ)
1068 kfree(dev->ifalias);
1069 dev->ifalias = NULL;
1074 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1078 strlcpy(dev->ifalias, alias, len+1);
1084 * netdev_features_change - device changes features
1085 * @dev: device to cause notification
1087 * Called to indicate a device has changed features.
1089 void netdev_features_change(struct net_device *dev)
1091 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1093 EXPORT_SYMBOL(netdev_features_change);
1096 * netdev_state_change - device changes state
1097 * @dev: device to cause notification
1099 * Called to indicate a device has changed state. This function calls
1100 * the notifier chains for netdev_chain and sends a NEWLINK message
1101 * to the routing socket.
1103 void netdev_state_change(struct net_device *dev)
1105 if (dev->flags & IFF_UP) {
1106 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1107 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1110 EXPORT_SYMBOL(netdev_state_change);
1112 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1114 return call_netdevice_notifiers(event, dev);
1116 EXPORT_SYMBOL(netdev_bonding_change);
1119 * dev_load - load a network module
1120 * @net: the applicable net namespace
1121 * @name: name of interface
1123 * If a network interface is not present and the process has suitable
1124 * privileges this function loads the module. If module loading is not
1125 * available in this kernel then it becomes a nop.
1128 void dev_load(struct net *net, const char *name)
1130 struct net_device *dev;
1134 dev = dev_get_by_name_rcu(net, name);
1138 if (no_module && capable(CAP_NET_ADMIN))
1139 no_module = request_module("netdev-%s", name);
1140 if (no_module && capable(CAP_SYS_MODULE)) {
1141 if (!request_module("%s", name))
1142 pr_err("Loading kernel module for a network device "
1143 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1147 EXPORT_SYMBOL(dev_load);
1149 static int __dev_open(struct net_device *dev)
1151 const struct net_device_ops *ops = dev->netdev_ops;
1156 if (!netif_device_present(dev))
1159 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1160 ret = notifier_to_errno(ret);
1164 set_bit(__LINK_STATE_START, &dev->state);
1166 if (ops->ndo_validate_addr)
1167 ret = ops->ndo_validate_addr(dev);
1169 if (!ret && ops->ndo_open)
1170 ret = ops->ndo_open(dev);
1173 clear_bit(__LINK_STATE_START, &dev->state);
1175 dev->flags |= IFF_UP;
1176 net_dmaengine_get();
1177 dev_set_rx_mode(dev);
1185 * dev_open - prepare an interface for use.
1186 * @dev: device to open
1188 * Takes a device from down to up state. The device's private open
1189 * function is invoked and then the multicast lists are loaded. Finally
1190 * the device is moved into the up state and a %NETDEV_UP message is
1191 * sent to the netdev notifier chain.
1193 * Calling this function on an active interface is a nop. On a failure
1194 * a negative errno code is returned.
1196 int dev_open(struct net_device *dev)
1200 if (dev->flags & IFF_UP)
1203 ret = __dev_open(dev);
1207 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1208 call_netdevice_notifiers(NETDEV_UP, dev);
1212 EXPORT_SYMBOL(dev_open);
1214 static int __dev_close_many(struct list_head *head)
1216 struct net_device *dev;
1221 list_for_each_entry(dev, head, unreg_list) {
1222 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1224 clear_bit(__LINK_STATE_START, &dev->state);
1226 /* Synchronize to scheduled poll. We cannot touch poll list, it
1227 * can be even on different cpu. So just clear netif_running().
1229 * dev->stop() will invoke napi_disable() on all of it's
1230 * napi_struct instances on this device.
1232 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1235 dev_deactivate_many(head);
1237 list_for_each_entry(dev, head, unreg_list) {
1238 const struct net_device_ops *ops = dev->netdev_ops;
1241 * Call the device specific close. This cannot fail.
1242 * Only if device is UP
1244 * We allow it to be called even after a DETACH hot-plug
1250 dev->flags &= ~IFF_UP;
1251 net_dmaengine_put();
1257 static int __dev_close(struct net_device *dev)
1262 list_add(&dev->unreg_list, &single);
1263 retval = __dev_close_many(&single);
1268 static int dev_close_many(struct list_head *head)
1270 struct net_device *dev, *tmp;
1271 LIST_HEAD(tmp_list);
1273 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1274 if (!(dev->flags & IFF_UP))
1275 list_move(&dev->unreg_list, &tmp_list);
1277 __dev_close_many(head);
1279 list_for_each_entry(dev, head, unreg_list) {
1280 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1281 call_netdevice_notifiers(NETDEV_DOWN, dev);
1284 /* rollback_registered_many needs the complete original list */
1285 list_splice(&tmp_list, head);
1290 * dev_close - shutdown an interface.
1291 * @dev: device to shutdown
1293 * This function moves an active device into down state. A
1294 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1295 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1298 int dev_close(struct net_device *dev)
1300 if (dev->flags & IFF_UP) {
1303 list_add(&dev->unreg_list, &single);
1304 dev_close_many(&single);
1309 EXPORT_SYMBOL(dev_close);
1313 * dev_disable_lro - disable Large Receive Offload on a device
1316 * Disable Large Receive Offload (LRO) on a net device. Must be
1317 * called under RTNL. This is needed if received packets may be
1318 * forwarded to another interface.
1320 void dev_disable_lro(struct net_device *dev)
1323 * If we're trying to disable lro on a vlan device
1324 * use the underlying physical device instead
1326 if (is_vlan_dev(dev))
1327 dev = vlan_dev_real_dev(dev);
1329 dev->wanted_features &= ~NETIF_F_LRO;
1330 netdev_update_features(dev);
1332 if (unlikely(dev->features & NETIF_F_LRO))
1333 netdev_WARN(dev, "failed to disable LRO!\n");
1335 EXPORT_SYMBOL(dev_disable_lro);
1338 static int dev_boot_phase = 1;
1341 * register_netdevice_notifier - register a network notifier block
1344 * Register a notifier to be called when network device events occur.
1345 * The notifier passed is linked into the kernel structures and must
1346 * not be reused until it has been unregistered. A negative errno code
1347 * is returned on a failure.
1349 * When registered all registration and up events are replayed
1350 * to the new notifier to allow device to have a race free
1351 * view of the network device list.
1354 int register_netdevice_notifier(struct notifier_block *nb)
1356 struct net_device *dev;
1357 struct net_device *last;
1362 err = raw_notifier_chain_register(&netdev_chain, nb);
1368 for_each_netdev(net, dev) {
1369 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1370 err = notifier_to_errno(err);
1374 if (!(dev->flags & IFF_UP))
1377 nb->notifier_call(nb, NETDEV_UP, dev);
1388 for_each_netdev(net, dev) {
1392 if (dev->flags & IFF_UP) {
1393 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1394 nb->notifier_call(nb, NETDEV_DOWN, dev);
1396 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1397 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1402 raw_notifier_chain_unregister(&netdev_chain, nb);
1405 EXPORT_SYMBOL(register_netdevice_notifier);
1408 * unregister_netdevice_notifier - unregister a network notifier block
1411 * Unregister a notifier previously registered by
1412 * register_netdevice_notifier(). The notifier is unlinked into the
1413 * kernel structures and may then be reused. A negative errno code
1414 * is returned on a failure.
1417 int unregister_netdevice_notifier(struct notifier_block *nb)
1422 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1426 EXPORT_SYMBOL(unregister_netdevice_notifier);
1429 * call_netdevice_notifiers - call all network notifier blocks
1430 * @val: value passed unmodified to notifier function
1431 * @dev: net_device pointer passed unmodified to notifier function
1433 * Call all network notifier blocks. Parameters and return value
1434 * are as for raw_notifier_call_chain().
1437 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1440 return raw_notifier_call_chain(&netdev_chain, val, dev);
1442 EXPORT_SYMBOL(call_netdevice_notifiers);
1444 static struct jump_label_key netstamp_needed __read_mostly;
1445 #ifdef HAVE_JUMP_LABEL
1446 /* We are not allowed to call jump_label_dec() from irq context
1447 * If net_disable_timestamp() is called from irq context, defer the
1448 * jump_label_dec() calls.
1450 static atomic_t netstamp_needed_deferred;
1453 void net_enable_timestamp(void)
1455 #ifdef HAVE_JUMP_LABEL
1456 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1460 jump_label_dec(&netstamp_needed);
1464 WARN_ON(in_interrupt());
1465 jump_label_inc(&netstamp_needed);
1467 EXPORT_SYMBOL(net_enable_timestamp);
1469 void net_disable_timestamp(void)
1471 #ifdef HAVE_JUMP_LABEL
1472 if (in_interrupt()) {
1473 atomic_inc(&netstamp_needed_deferred);
1477 jump_label_dec(&netstamp_needed);
1479 EXPORT_SYMBOL(net_disable_timestamp);
1481 static inline void net_timestamp_set(struct sk_buff *skb)
1483 skb->tstamp.tv64 = 0;
1484 if (static_branch(&netstamp_needed))
1485 __net_timestamp(skb);
1488 #define net_timestamp_check(COND, SKB) \
1489 if (static_branch(&netstamp_needed)) { \
1490 if ((COND) && !(SKB)->tstamp.tv64) \
1491 __net_timestamp(SKB); \
1494 static int net_hwtstamp_validate(struct ifreq *ifr)
1496 struct hwtstamp_config cfg;
1497 enum hwtstamp_tx_types tx_type;
1498 enum hwtstamp_rx_filters rx_filter;
1499 int tx_type_valid = 0;
1500 int rx_filter_valid = 0;
1502 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1505 if (cfg.flags) /* reserved for future extensions */
1508 tx_type = cfg.tx_type;
1509 rx_filter = cfg.rx_filter;
1512 case HWTSTAMP_TX_OFF:
1513 case HWTSTAMP_TX_ON:
1514 case HWTSTAMP_TX_ONESTEP_SYNC:
1519 switch (rx_filter) {
1520 case HWTSTAMP_FILTER_NONE:
1521 case HWTSTAMP_FILTER_ALL:
1522 case HWTSTAMP_FILTER_SOME:
1523 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1524 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1525 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1526 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1527 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1528 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1529 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1530 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1531 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1532 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1533 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1534 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1535 rx_filter_valid = 1;
1539 if (!tx_type_valid || !rx_filter_valid)
1545 static inline bool is_skb_forwardable(struct net_device *dev,
1546 struct sk_buff *skb)
1550 if (!(dev->flags & IFF_UP))
1553 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1554 if (skb->len <= len)
1557 /* if TSO is enabled, we don't care about the length as the packet
1558 * could be forwarded without being segmented before
1560 if (skb_is_gso(skb))
1567 * dev_forward_skb - loopback an skb to another netif
1569 * @dev: destination network device
1570 * @skb: buffer to forward
1573 * NET_RX_SUCCESS (no congestion)
1574 * NET_RX_DROP (packet was dropped, but freed)
1576 * dev_forward_skb can be used for injecting an skb from the
1577 * start_xmit function of one device into the receive queue
1578 * of another device.
1580 * The receiving device may be in another namespace, so
1581 * we have to clear all information in the skb that could
1582 * impact namespace isolation.
1584 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1586 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1587 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1588 atomic_long_inc(&dev->rx_dropped);
1597 if (unlikely(!is_skb_forwardable(dev, skb))) {
1598 atomic_long_inc(&dev->rx_dropped);
1602 skb_set_dev(skb, dev);
1603 skb->tstamp.tv64 = 0;
1604 skb->pkt_type = PACKET_HOST;
1605 skb->protocol = eth_type_trans(skb, dev);
1606 return netif_rx(skb);
1608 EXPORT_SYMBOL_GPL(dev_forward_skb);
1610 static inline int deliver_skb(struct sk_buff *skb,
1611 struct packet_type *pt_prev,
1612 struct net_device *orig_dev)
1614 atomic_inc(&skb->users);
1615 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1619 * Support routine. Sends outgoing frames to any network
1620 * taps currently in use.
1623 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1625 struct packet_type *ptype;
1626 struct sk_buff *skb2 = NULL;
1627 struct packet_type *pt_prev = NULL;
1630 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1631 /* Never send packets back to the socket
1632 * they originated from - MvS (miquels@drinkel.ow.org)
1634 if ((ptype->dev == dev || !ptype->dev) &&
1635 (ptype->af_packet_priv == NULL ||
1636 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1638 deliver_skb(skb2, pt_prev, skb->dev);
1643 skb2 = skb_clone(skb, GFP_ATOMIC);
1647 net_timestamp_set(skb2);
1649 /* skb->nh should be correctly
1650 set by sender, so that the second statement is
1651 just protection against buggy protocols.
1653 skb_reset_mac_header(skb2);
1655 if (skb_network_header(skb2) < skb2->data ||
1656 skb2->network_header > skb2->tail) {
1657 if (net_ratelimit())
1658 printk(KERN_CRIT "protocol %04x is "
1660 ntohs(skb2->protocol),
1662 skb_reset_network_header(skb2);
1665 skb2->transport_header = skb2->network_header;
1666 skb2->pkt_type = PACKET_OUTGOING;
1671 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1675 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1676 * @dev: Network device
1677 * @txq: number of queues available
1679 * If real_num_tx_queues is changed the tc mappings may no longer be
1680 * valid. To resolve this verify the tc mapping remains valid and if
1681 * not NULL the mapping. With no priorities mapping to this
1682 * offset/count pair it will no longer be used. In the worst case TC0
1683 * is invalid nothing can be done so disable priority mappings. If is
1684 * expected that drivers will fix this mapping if they can before
1685 * calling netif_set_real_num_tx_queues.
1687 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1690 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1692 /* If TC0 is invalidated disable TC mapping */
1693 if (tc->offset + tc->count > txq) {
1694 pr_warning("Number of in use tx queues changed "
1695 "invalidating tc mappings. Priority "
1696 "traffic classification disabled!\n");
1701 /* Invalidated prio to tc mappings set to TC0 */
1702 for (i = 1; i < TC_BITMASK + 1; i++) {
1703 int q = netdev_get_prio_tc_map(dev, i);
1705 tc = &dev->tc_to_txq[q];
1706 if (tc->offset + tc->count > txq) {
1707 pr_warning("Number of in use tx queues "
1708 "changed. Priority %i to tc "
1709 "mapping %i is no longer valid "
1710 "setting map to 0\n",
1712 netdev_set_prio_tc_map(dev, i, 0);
1718 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1719 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1721 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1725 if (txq < 1 || txq > dev->num_tx_queues)
1728 if (dev->reg_state == NETREG_REGISTERED ||
1729 dev->reg_state == NETREG_UNREGISTERING) {
1732 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1738 netif_setup_tc(dev, txq);
1740 if (txq < dev->real_num_tx_queues)
1741 qdisc_reset_all_tx_gt(dev, txq);
1744 dev->real_num_tx_queues = txq;
1747 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1751 * netif_set_real_num_rx_queues - set actual number of RX queues used
1752 * @dev: Network device
1753 * @rxq: Actual number of RX queues
1755 * This must be called either with the rtnl_lock held or before
1756 * registration of the net device. Returns 0 on success, or a
1757 * negative error code. If called before registration, it always
1760 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1764 if (rxq < 1 || rxq > dev->num_rx_queues)
1767 if (dev->reg_state == NETREG_REGISTERED) {
1770 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1776 dev->real_num_rx_queues = rxq;
1779 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1782 static inline void __netif_reschedule(struct Qdisc *q)
1784 struct softnet_data *sd;
1785 unsigned long flags;
1787 local_irq_save(flags);
1788 sd = &__get_cpu_var(softnet_data);
1789 q->next_sched = NULL;
1790 *sd->output_queue_tailp = q;
1791 sd->output_queue_tailp = &q->next_sched;
1792 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1793 local_irq_restore(flags);
1796 void __netif_schedule(struct Qdisc *q)
1798 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1799 __netif_reschedule(q);
1801 EXPORT_SYMBOL(__netif_schedule);
1803 void dev_kfree_skb_irq(struct sk_buff *skb)
1805 if (atomic_dec_and_test(&skb->users)) {
1806 struct softnet_data *sd;
1807 unsigned long flags;
1809 local_irq_save(flags);
1810 sd = &__get_cpu_var(softnet_data);
1811 skb->next = sd->completion_queue;
1812 sd->completion_queue = skb;
1813 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1814 local_irq_restore(flags);
1817 EXPORT_SYMBOL(dev_kfree_skb_irq);
1819 void dev_kfree_skb_any(struct sk_buff *skb)
1821 if (in_irq() || irqs_disabled())
1822 dev_kfree_skb_irq(skb);
1826 EXPORT_SYMBOL(dev_kfree_skb_any);
1830 * netif_device_detach - mark device as removed
1831 * @dev: network device
1833 * Mark device as removed from system and therefore no longer available.
1835 void netif_device_detach(struct net_device *dev)
1837 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1838 netif_running(dev)) {
1839 netif_tx_stop_all_queues(dev);
1842 EXPORT_SYMBOL(netif_device_detach);
1845 * netif_device_attach - mark device as attached
1846 * @dev: network device
1848 * Mark device as attached from system and restart if needed.
1850 void netif_device_attach(struct net_device *dev)
1852 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1853 netif_running(dev)) {
1854 netif_tx_wake_all_queues(dev);
1855 __netdev_watchdog_up(dev);
1858 EXPORT_SYMBOL(netif_device_attach);
1861 * skb_dev_set -- assign a new device to a buffer
1862 * @skb: buffer for the new device
1863 * @dev: network device
1865 * If an skb is owned by a device already, we have to reset
1866 * all data private to the namespace a device belongs to
1867 * before assigning it a new device.
1869 #ifdef CONFIG_NET_NS
1870 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1873 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1876 skb_init_secmark(skb);
1880 skb->ipvs_property = 0;
1881 #ifdef CONFIG_NET_SCHED
1887 EXPORT_SYMBOL(skb_set_dev);
1888 #endif /* CONFIG_NET_NS */
1890 static void skb_warn_bad_offload(const struct sk_buff *skb)
1892 static const netdev_features_t null_features = 0;
1893 struct net_device *dev = skb->dev;
1894 const char *driver = "";
1896 if (dev && dev->dev.parent)
1897 driver = dev_driver_string(dev->dev.parent);
1899 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1900 "gso_type=%d ip_summed=%d\n",
1901 driver, dev ? &dev->features : &null_features,
1902 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1903 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1904 skb_shinfo(skb)->gso_type, skb->ip_summed);
1908 * Invalidate hardware checksum when packet is to be mangled, and
1909 * complete checksum manually on outgoing path.
1911 int skb_checksum_help(struct sk_buff *skb)
1914 int ret = 0, offset;
1916 if (skb->ip_summed == CHECKSUM_COMPLETE)
1917 goto out_set_summed;
1919 if (unlikely(skb_shinfo(skb)->gso_size)) {
1920 skb_warn_bad_offload(skb);
1924 offset = skb_checksum_start_offset(skb);
1925 BUG_ON(offset >= skb_headlen(skb));
1926 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1928 offset += skb->csum_offset;
1929 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1931 if (skb_cloned(skb) &&
1932 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1933 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1938 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1940 skb->ip_summed = CHECKSUM_NONE;
1944 EXPORT_SYMBOL(skb_checksum_help);
1947 * skb_gso_segment - Perform segmentation on skb.
1948 * @skb: buffer to segment
1949 * @features: features for the output path (see dev->features)
1951 * This function segments the given skb and returns a list of segments.
1953 * It may return NULL if the skb requires no segmentation. This is
1954 * only possible when GSO is used for verifying header integrity.
1956 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1957 netdev_features_t features)
1959 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1960 struct packet_type *ptype;
1961 __be16 type = skb->protocol;
1962 int vlan_depth = ETH_HLEN;
1965 while (type == htons(ETH_P_8021Q)) {
1966 struct vlan_hdr *vh;
1968 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1969 return ERR_PTR(-EINVAL);
1971 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1972 type = vh->h_vlan_encapsulated_proto;
1973 vlan_depth += VLAN_HLEN;
1976 skb_reset_mac_header(skb);
1977 skb->mac_len = skb->network_header - skb->mac_header;
1978 __skb_pull(skb, skb->mac_len);
1980 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1981 skb_warn_bad_offload(skb);
1983 if (skb_header_cloned(skb) &&
1984 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1985 return ERR_PTR(err);
1989 list_for_each_entry_rcu(ptype,
1990 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1991 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1992 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1993 err = ptype->gso_send_check(skb);
1994 segs = ERR_PTR(err);
1995 if (err || skb_gso_ok(skb, features))
1997 __skb_push(skb, (skb->data -
1998 skb_network_header(skb)));
2000 segs = ptype->gso_segment(skb, features);
2006 __skb_push(skb, skb->data - skb_mac_header(skb));
2010 EXPORT_SYMBOL(skb_gso_segment);
2012 /* Take action when hardware reception checksum errors are detected. */
2014 void netdev_rx_csum_fault(struct net_device *dev)
2016 if (net_ratelimit()) {
2017 printk(KERN_ERR "%s: hw csum failure.\n",
2018 dev ? dev->name : "<unknown>");
2022 EXPORT_SYMBOL(netdev_rx_csum_fault);
2025 /* Actually, we should eliminate this check as soon as we know, that:
2026 * 1. IOMMU is present and allows to map all the memory.
2027 * 2. No high memory really exists on this machine.
2030 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2032 #ifdef CONFIG_HIGHMEM
2034 if (!(dev->features & NETIF_F_HIGHDMA)) {
2035 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2036 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2037 if (PageHighMem(skb_frag_page(frag)))
2042 if (PCI_DMA_BUS_IS_PHYS) {
2043 struct device *pdev = dev->dev.parent;
2047 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2048 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2049 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2050 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2059 void (*destructor)(struct sk_buff *skb);
2062 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2064 static void dev_gso_skb_destructor(struct sk_buff *skb)
2066 struct dev_gso_cb *cb;
2069 struct sk_buff *nskb = skb->next;
2071 skb->next = nskb->next;
2074 } while (skb->next);
2076 cb = DEV_GSO_CB(skb);
2078 cb->destructor(skb);
2082 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2083 * @skb: buffer to segment
2084 * @features: device features as applicable to this skb
2086 * This function segments the given skb and stores the list of segments
2089 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2091 struct sk_buff *segs;
2093 segs = skb_gso_segment(skb, features);
2095 /* Verifying header integrity only. */
2100 return PTR_ERR(segs);
2103 DEV_GSO_CB(skb)->destructor = skb->destructor;
2104 skb->destructor = dev_gso_skb_destructor;
2110 * Try to orphan skb early, right before transmission by the device.
2111 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2112 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2114 static inline void skb_orphan_try(struct sk_buff *skb)
2116 struct sock *sk = skb->sk;
2118 if (sk && !skb_shinfo(skb)->tx_flags) {
2119 /* skb_tx_hash() wont be able to get sk.
2120 * We copy sk_hash into skb->rxhash
2123 skb->rxhash = sk->sk_hash;
2128 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2130 return ((features & NETIF_F_GEN_CSUM) ||
2131 ((features & NETIF_F_V4_CSUM) &&
2132 protocol == htons(ETH_P_IP)) ||
2133 ((features & NETIF_F_V6_CSUM) &&
2134 protocol == htons(ETH_P_IPV6)) ||
2135 ((features & NETIF_F_FCOE_CRC) &&
2136 protocol == htons(ETH_P_FCOE)));
2139 static netdev_features_t harmonize_features(struct sk_buff *skb,
2140 __be16 protocol, netdev_features_t features)
2142 if (!can_checksum_protocol(features, protocol)) {
2143 features &= ~NETIF_F_ALL_CSUM;
2144 features &= ~NETIF_F_SG;
2145 } else if (illegal_highdma(skb->dev, skb)) {
2146 features &= ~NETIF_F_SG;
2152 netdev_features_t netif_skb_features(struct sk_buff *skb)
2154 __be16 protocol = skb->protocol;
2155 netdev_features_t features = skb->dev->features;
2157 if (protocol == htons(ETH_P_8021Q)) {
2158 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2159 protocol = veh->h_vlan_encapsulated_proto;
2160 } else if (!vlan_tx_tag_present(skb)) {
2161 return harmonize_features(skb, protocol, features);
2164 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2166 if (protocol != htons(ETH_P_8021Q)) {
2167 return harmonize_features(skb, protocol, features);
2169 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2170 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2171 return harmonize_features(skb, protocol, features);
2174 EXPORT_SYMBOL(netif_skb_features);
2177 * Returns true if either:
2178 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2179 * 2. skb is fragmented and the device does not support SG, or if
2180 * at least one of fragments is in highmem and device does not
2181 * support DMA from it.
2183 static inline int skb_needs_linearize(struct sk_buff *skb,
2186 return skb_is_nonlinear(skb) &&
2187 ((skb_has_frag_list(skb) &&
2188 !(features & NETIF_F_FRAGLIST)) ||
2189 (skb_shinfo(skb)->nr_frags &&
2190 !(features & NETIF_F_SG)));
2193 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2194 struct netdev_queue *txq)
2196 const struct net_device_ops *ops = dev->netdev_ops;
2197 int rc = NETDEV_TX_OK;
2198 unsigned int skb_len;
2200 if (likely(!skb->next)) {
2201 netdev_features_t features;
2204 * If device doesn't need skb->dst, release it right now while
2205 * its hot in this cpu cache
2207 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2210 if (!list_empty(&ptype_all))
2211 dev_queue_xmit_nit(skb, dev);
2213 skb_orphan_try(skb);
2215 features = netif_skb_features(skb);
2217 if (vlan_tx_tag_present(skb) &&
2218 !(features & NETIF_F_HW_VLAN_TX)) {
2219 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2226 if (netif_needs_gso(skb, features)) {
2227 if (unlikely(dev_gso_segment(skb, features)))
2232 if (skb_needs_linearize(skb, features) &&
2233 __skb_linearize(skb))
2236 /* If packet is not checksummed and device does not
2237 * support checksumming for this protocol, complete
2238 * checksumming here.
2240 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2241 skb_set_transport_header(skb,
2242 skb_checksum_start_offset(skb));
2243 if (!(features & NETIF_F_ALL_CSUM) &&
2244 skb_checksum_help(skb))
2250 rc = ops->ndo_start_xmit(skb, dev);
2251 trace_net_dev_xmit(skb, rc, dev, skb_len);
2252 if (rc == NETDEV_TX_OK)
2253 txq_trans_update(txq);
2259 struct sk_buff *nskb = skb->next;
2261 skb->next = nskb->next;
2265 * If device doesn't need nskb->dst, release it right now while
2266 * its hot in this cpu cache
2268 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2271 skb_len = nskb->len;
2272 rc = ops->ndo_start_xmit(nskb, dev);
2273 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2274 if (unlikely(rc != NETDEV_TX_OK)) {
2275 if (rc & ~NETDEV_TX_MASK)
2276 goto out_kfree_gso_skb;
2277 nskb->next = skb->next;
2281 txq_trans_update(txq);
2282 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2283 return NETDEV_TX_BUSY;
2284 } while (skb->next);
2287 if (likely(skb->next == NULL))
2288 skb->destructor = DEV_GSO_CB(skb)->destructor;
2295 static u32 hashrnd __read_mostly;
2298 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2299 * to be used as a distribution range.
2301 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2302 unsigned int num_tx_queues)
2306 u16 qcount = num_tx_queues;
2308 if (skb_rx_queue_recorded(skb)) {
2309 hash = skb_get_rx_queue(skb);
2310 while (unlikely(hash >= num_tx_queues))
2311 hash -= num_tx_queues;
2316 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2317 qoffset = dev->tc_to_txq[tc].offset;
2318 qcount = dev->tc_to_txq[tc].count;
2321 if (skb->sk && skb->sk->sk_hash)
2322 hash = skb->sk->sk_hash;
2324 hash = (__force u16) skb->protocol ^ skb->rxhash;
2325 hash = jhash_1word(hash, hashrnd);
2327 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2329 EXPORT_SYMBOL(__skb_tx_hash);
2331 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2333 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2334 if (net_ratelimit()) {
2335 pr_warning("%s selects TX queue %d, but "
2336 "real number of TX queues is %d\n",
2337 dev->name, queue_index, dev->real_num_tx_queues);
2344 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2347 struct xps_dev_maps *dev_maps;
2348 struct xps_map *map;
2349 int queue_index = -1;
2352 dev_maps = rcu_dereference(dev->xps_maps);
2354 map = rcu_dereference(
2355 dev_maps->cpu_map[raw_smp_processor_id()]);
2358 queue_index = map->queues[0];
2361 if (skb->sk && skb->sk->sk_hash)
2362 hash = skb->sk->sk_hash;
2364 hash = (__force u16) skb->protocol ^
2366 hash = jhash_1word(hash, hashrnd);
2367 queue_index = map->queues[
2368 ((u64)hash * map->len) >> 32];
2370 if (unlikely(queue_index >= dev->real_num_tx_queues))
2382 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2383 struct sk_buff *skb)
2386 const struct net_device_ops *ops = dev->netdev_ops;
2388 if (dev->real_num_tx_queues == 1)
2390 else if (ops->ndo_select_queue) {
2391 queue_index = ops->ndo_select_queue(dev, skb);
2392 queue_index = dev_cap_txqueue(dev, queue_index);
2394 struct sock *sk = skb->sk;
2395 queue_index = sk_tx_queue_get(sk);
2397 if (queue_index < 0 || skb->ooo_okay ||
2398 queue_index >= dev->real_num_tx_queues) {
2399 int old_index = queue_index;
2401 queue_index = get_xps_queue(dev, skb);
2402 if (queue_index < 0)
2403 queue_index = skb_tx_hash(dev, skb);
2405 if (queue_index != old_index && sk) {
2406 struct dst_entry *dst =
2407 rcu_dereference_check(sk->sk_dst_cache, 1);
2409 if (dst && skb_dst(skb) == dst)
2410 sk_tx_queue_set(sk, queue_index);
2415 skb_set_queue_mapping(skb, queue_index);
2416 return netdev_get_tx_queue(dev, queue_index);
2419 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2420 struct net_device *dev,
2421 struct netdev_queue *txq)
2423 spinlock_t *root_lock = qdisc_lock(q);
2427 qdisc_skb_cb(skb)->pkt_len = skb->len;
2428 qdisc_calculate_pkt_len(skb, q);
2430 * Heuristic to force contended enqueues to serialize on a
2431 * separate lock before trying to get qdisc main lock.
2432 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2433 * and dequeue packets faster.
2435 contended = qdisc_is_running(q);
2436 if (unlikely(contended))
2437 spin_lock(&q->busylock);
2439 spin_lock(root_lock);
2440 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2443 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2444 qdisc_run_begin(q)) {
2446 * This is a work-conserving queue; there are no old skbs
2447 * waiting to be sent out; and the qdisc is not running -
2448 * xmit the skb directly.
2450 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2453 qdisc_bstats_update(q, skb);
2455 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2456 if (unlikely(contended)) {
2457 spin_unlock(&q->busylock);
2464 rc = NET_XMIT_SUCCESS;
2467 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2468 if (qdisc_run_begin(q)) {
2469 if (unlikely(contended)) {
2470 spin_unlock(&q->busylock);
2476 spin_unlock(root_lock);
2477 if (unlikely(contended))
2478 spin_unlock(&q->busylock);
2482 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2483 static void skb_update_prio(struct sk_buff *skb)
2485 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2487 if ((!skb->priority) && (skb->sk) && map)
2488 skb->priority = map->priomap[skb->sk->sk_cgrp_prioidx];
2491 #define skb_update_prio(skb)
2494 static DEFINE_PER_CPU(int, xmit_recursion);
2495 #define RECURSION_LIMIT 10
2498 * dev_queue_xmit - transmit a buffer
2499 * @skb: buffer to transmit
2501 * Queue a buffer for transmission to a network device. The caller must
2502 * have set the device and priority and built the buffer before calling
2503 * this function. The function can be called from an interrupt.
2505 * A negative errno code is returned on a failure. A success does not
2506 * guarantee the frame will be transmitted as it may be dropped due
2507 * to congestion or traffic shaping.
2509 * -----------------------------------------------------------------------------------
2510 * I notice this method can also return errors from the queue disciplines,
2511 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2514 * Regardless of the return value, the skb is consumed, so it is currently
2515 * difficult to retry a send to this method. (You can bump the ref count
2516 * before sending to hold a reference for retry if you are careful.)
2518 * When calling this method, interrupts MUST be enabled. This is because
2519 * the BH enable code must have IRQs enabled so that it will not deadlock.
2522 int dev_queue_xmit(struct sk_buff *skb)
2524 struct net_device *dev = skb->dev;
2525 struct netdev_queue *txq;
2529 /* Disable soft irqs for various locks below. Also
2530 * stops preemption for RCU.
2534 skb_update_prio(skb);
2536 txq = dev_pick_tx(dev, skb);
2537 q = rcu_dereference_bh(txq->qdisc);
2539 #ifdef CONFIG_NET_CLS_ACT
2540 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2542 trace_net_dev_queue(skb);
2544 rc = __dev_xmit_skb(skb, q, dev, txq);
2548 /* The device has no queue. Common case for software devices:
2549 loopback, all the sorts of tunnels...
2551 Really, it is unlikely that netif_tx_lock protection is necessary
2552 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2554 However, it is possible, that they rely on protection
2557 Check this and shot the lock. It is not prone from deadlocks.
2558 Either shot noqueue qdisc, it is even simpler 8)
2560 if (dev->flags & IFF_UP) {
2561 int cpu = smp_processor_id(); /* ok because BHs are off */
2563 if (txq->xmit_lock_owner != cpu) {
2565 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2566 goto recursion_alert;
2568 HARD_TX_LOCK(dev, txq, cpu);
2570 if (!netif_xmit_stopped(txq)) {
2571 __this_cpu_inc(xmit_recursion);
2572 rc = dev_hard_start_xmit(skb, dev, txq);
2573 __this_cpu_dec(xmit_recursion);
2574 if (dev_xmit_complete(rc)) {
2575 HARD_TX_UNLOCK(dev, txq);
2579 HARD_TX_UNLOCK(dev, txq);
2580 if (net_ratelimit())
2581 printk(KERN_CRIT "Virtual device %s asks to "
2582 "queue packet!\n", dev->name);
2584 /* Recursion is detected! It is possible,
2588 if (net_ratelimit())
2589 printk(KERN_CRIT "Dead loop on virtual device "
2590 "%s, fix it urgently!\n", dev->name);
2595 rcu_read_unlock_bh();
2600 rcu_read_unlock_bh();
2603 EXPORT_SYMBOL(dev_queue_xmit);
2606 /*=======================================================================
2608 =======================================================================*/
2610 int netdev_max_backlog __read_mostly = 1000;
2611 int netdev_tstamp_prequeue __read_mostly = 1;
2612 int netdev_budget __read_mostly = 300;
2613 int weight_p __read_mostly = 64; /* old backlog weight */
2615 /* Called with irq disabled */
2616 static inline void ____napi_schedule(struct softnet_data *sd,
2617 struct napi_struct *napi)
2619 list_add_tail(&napi->poll_list, &sd->poll_list);
2620 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2624 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2625 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2626 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2627 * if hash is a canonical 4-tuple hash over transport ports.
2629 void __skb_get_rxhash(struct sk_buff *skb)
2631 struct flow_keys keys;
2634 if (!skb_flow_dissect(skb, &keys))
2638 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2639 swap(keys.port16[0], keys.port16[1]);
2643 /* get a consistent hash (same value on both flow directions) */
2644 if ((__force u32)keys.dst < (__force u32)keys.src)
2645 swap(keys.dst, keys.src);
2647 hash = jhash_3words((__force u32)keys.dst,
2648 (__force u32)keys.src,
2649 (__force u32)keys.ports, hashrnd);
2655 EXPORT_SYMBOL(__skb_get_rxhash);
2659 /* One global table that all flow-based protocols share. */
2660 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2661 EXPORT_SYMBOL(rps_sock_flow_table);
2663 struct jump_label_key rps_needed __read_mostly;
2665 static struct rps_dev_flow *
2666 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2667 struct rps_dev_flow *rflow, u16 next_cpu)
2669 if (next_cpu != RPS_NO_CPU) {
2670 #ifdef CONFIG_RFS_ACCEL
2671 struct netdev_rx_queue *rxqueue;
2672 struct rps_dev_flow_table *flow_table;
2673 struct rps_dev_flow *old_rflow;
2678 /* Should we steer this flow to a different hardware queue? */
2679 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2680 !(dev->features & NETIF_F_NTUPLE))
2682 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2683 if (rxq_index == skb_get_rx_queue(skb))
2686 rxqueue = dev->_rx + rxq_index;
2687 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2690 flow_id = skb->rxhash & flow_table->mask;
2691 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2692 rxq_index, flow_id);
2696 rflow = &flow_table->flows[flow_id];
2698 if (old_rflow->filter == rflow->filter)
2699 old_rflow->filter = RPS_NO_FILTER;
2703 per_cpu(softnet_data, next_cpu).input_queue_head;
2706 rflow->cpu = next_cpu;
2711 * get_rps_cpu is called from netif_receive_skb and returns the target
2712 * CPU from the RPS map of the receiving queue for a given skb.
2713 * rcu_read_lock must be held on entry.
2715 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2716 struct rps_dev_flow **rflowp)
2718 struct netdev_rx_queue *rxqueue;
2719 struct rps_map *map;
2720 struct rps_dev_flow_table *flow_table;
2721 struct rps_sock_flow_table *sock_flow_table;
2725 if (skb_rx_queue_recorded(skb)) {
2726 u16 index = skb_get_rx_queue(skb);
2727 if (unlikely(index >= dev->real_num_rx_queues)) {
2728 WARN_ONCE(dev->real_num_rx_queues > 1,
2729 "%s received packet on queue %u, but number "
2730 "of RX queues is %u\n",
2731 dev->name, index, dev->real_num_rx_queues);
2734 rxqueue = dev->_rx + index;
2738 map = rcu_dereference(rxqueue->rps_map);
2740 if (map->len == 1 &&
2741 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2742 tcpu = map->cpus[0];
2743 if (cpu_online(tcpu))
2747 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2751 skb_reset_network_header(skb);
2752 if (!skb_get_rxhash(skb))
2755 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2756 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2757 if (flow_table && sock_flow_table) {
2759 struct rps_dev_flow *rflow;
2761 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2764 next_cpu = sock_flow_table->ents[skb->rxhash &
2765 sock_flow_table->mask];
2768 * If the desired CPU (where last recvmsg was done) is
2769 * different from current CPU (one in the rx-queue flow
2770 * table entry), switch if one of the following holds:
2771 * - Current CPU is unset (equal to RPS_NO_CPU).
2772 * - Current CPU is offline.
2773 * - The current CPU's queue tail has advanced beyond the
2774 * last packet that was enqueued using this table entry.
2775 * This guarantees that all previous packets for the flow
2776 * have been dequeued, thus preserving in order delivery.
2778 if (unlikely(tcpu != next_cpu) &&
2779 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2780 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2781 rflow->last_qtail)) >= 0))
2782 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2784 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2792 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2794 if (cpu_online(tcpu)) {
2804 #ifdef CONFIG_RFS_ACCEL
2807 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2808 * @dev: Device on which the filter was set
2809 * @rxq_index: RX queue index
2810 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2811 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2813 * Drivers that implement ndo_rx_flow_steer() should periodically call
2814 * this function for each installed filter and remove the filters for
2815 * which it returns %true.
2817 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2818 u32 flow_id, u16 filter_id)
2820 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2821 struct rps_dev_flow_table *flow_table;
2822 struct rps_dev_flow *rflow;
2827 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2828 if (flow_table && flow_id <= flow_table->mask) {
2829 rflow = &flow_table->flows[flow_id];
2830 cpu = ACCESS_ONCE(rflow->cpu);
2831 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2832 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2833 rflow->last_qtail) <
2834 (int)(10 * flow_table->mask)))
2840 EXPORT_SYMBOL(rps_may_expire_flow);
2842 #endif /* CONFIG_RFS_ACCEL */
2844 /* Called from hardirq (IPI) context */
2845 static void rps_trigger_softirq(void *data)
2847 struct softnet_data *sd = data;
2849 ____napi_schedule(sd, &sd->backlog);
2853 #endif /* CONFIG_RPS */
2856 * Check if this softnet_data structure is another cpu one
2857 * If yes, queue it to our IPI list and return 1
2860 static int rps_ipi_queued(struct softnet_data *sd)
2863 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2866 sd->rps_ipi_next = mysd->rps_ipi_list;
2867 mysd->rps_ipi_list = sd;
2869 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2872 #endif /* CONFIG_RPS */
2877 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2878 * queue (may be a remote CPU queue).
2880 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2881 unsigned int *qtail)
2883 struct softnet_data *sd;
2884 unsigned long flags;
2886 sd = &per_cpu(softnet_data, cpu);
2888 local_irq_save(flags);
2891 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2892 if (skb_queue_len(&sd->input_pkt_queue)) {
2894 __skb_queue_tail(&sd->input_pkt_queue, skb);
2895 input_queue_tail_incr_save(sd, qtail);
2897 local_irq_restore(flags);
2898 return NET_RX_SUCCESS;
2901 /* Schedule NAPI for backlog device
2902 * We can use non atomic operation since we own the queue lock
2904 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2905 if (!rps_ipi_queued(sd))
2906 ____napi_schedule(sd, &sd->backlog);
2914 local_irq_restore(flags);
2916 atomic_long_inc(&skb->dev->rx_dropped);
2922 * netif_rx - post buffer to the network code
2923 * @skb: buffer to post
2925 * This function receives a packet from a device driver and queues it for
2926 * the upper (protocol) levels to process. It always succeeds. The buffer
2927 * may be dropped during processing for congestion control or by the
2931 * NET_RX_SUCCESS (no congestion)
2932 * NET_RX_DROP (packet was dropped)
2936 int netif_rx(struct sk_buff *skb)
2940 /* if netpoll wants it, pretend we never saw it */
2941 if (netpoll_rx(skb))
2944 net_timestamp_check(netdev_tstamp_prequeue, skb);
2946 trace_netif_rx(skb);
2948 if (static_branch(&rps_needed)) {
2949 struct rps_dev_flow voidflow, *rflow = &voidflow;
2955 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2957 cpu = smp_processor_id();
2959 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2967 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2972 EXPORT_SYMBOL(netif_rx);
2974 int netif_rx_ni(struct sk_buff *skb)
2979 err = netif_rx(skb);
2980 if (local_softirq_pending())
2986 EXPORT_SYMBOL(netif_rx_ni);
2988 static void net_tx_action(struct softirq_action *h)
2990 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2992 if (sd->completion_queue) {
2993 struct sk_buff *clist;
2995 local_irq_disable();
2996 clist = sd->completion_queue;
2997 sd->completion_queue = NULL;
3001 struct sk_buff *skb = clist;
3002 clist = clist->next;
3004 WARN_ON(atomic_read(&skb->users));
3005 trace_kfree_skb(skb, net_tx_action);
3010 if (sd->output_queue) {
3013 local_irq_disable();
3014 head = sd->output_queue;
3015 sd->output_queue = NULL;
3016 sd->output_queue_tailp = &sd->output_queue;
3020 struct Qdisc *q = head;
3021 spinlock_t *root_lock;
3023 head = head->next_sched;
3025 root_lock = qdisc_lock(q);
3026 if (spin_trylock(root_lock)) {
3027 smp_mb__before_clear_bit();
3028 clear_bit(__QDISC_STATE_SCHED,
3031 spin_unlock(root_lock);
3033 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3035 __netif_reschedule(q);
3037 smp_mb__before_clear_bit();
3038 clear_bit(__QDISC_STATE_SCHED,
3046 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3047 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3048 /* This hook is defined here for ATM LANE */
3049 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3050 unsigned char *addr) __read_mostly;
3051 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3054 #ifdef CONFIG_NET_CLS_ACT
3055 /* TODO: Maybe we should just force sch_ingress to be compiled in
3056 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3057 * a compare and 2 stores extra right now if we dont have it on
3058 * but have CONFIG_NET_CLS_ACT
3059 * NOTE: This doesn't stop any functionality; if you dont have
3060 * the ingress scheduler, you just can't add policies on ingress.
3063 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3065 struct net_device *dev = skb->dev;
3066 u32 ttl = G_TC_RTTL(skb->tc_verd);
3067 int result = TC_ACT_OK;
3070 if (unlikely(MAX_RED_LOOP < ttl++)) {
3071 if (net_ratelimit())
3072 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3073 skb->skb_iif, dev->ifindex);
3077 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3078 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3081 if (q != &noop_qdisc) {
3082 spin_lock(qdisc_lock(q));
3083 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3084 result = qdisc_enqueue_root(skb, q);
3085 spin_unlock(qdisc_lock(q));
3091 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3092 struct packet_type **pt_prev,
3093 int *ret, struct net_device *orig_dev)
3095 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3097 if (!rxq || rxq->qdisc == &noop_qdisc)
3101 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3105 switch (ing_filter(skb, rxq)) {
3119 * netdev_rx_handler_register - register receive handler
3120 * @dev: device to register a handler for
3121 * @rx_handler: receive handler to register
3122 * @rx_handler_data: data pointer that is used by rx handler
3124 * Register a receive hander for a device. This handler will then be
3125 * called from __netif_receive_skb. A negative errno code is returned
3128 * The caller must hold the rtnl_mutex.
3130 * For a general description of rx_handler, see enum rx_handler_result.
3132 int netdev_rx_handler_register(struct net_device *dev,
3133 rx_handler_func_t *rx_handler,
3134 void *rx_handler_data)
3138 if (dev->rx_handler)
3141 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3142 rcu_assign_pointer(dev->rx_handler, rx_handler);
3146 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3149 * netdev_rx_handler_unregister - unregister receive handler
3150 * @dev: device to unregister a handler from
3152 * Unregister a receive hander from a device.
3154 * The caller must hold the rtnl_mutex.
3156 void netdev_rx_handler_unregister(struct net_device *dev)
3160 RCU_INIT_POINTER(dev->rx_handler, NULL);
3161 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3163 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3165 static int __netif_receive_skb(struct sk_buff *skb)
3167 struct packet_type *ptype, *pt_prev;
3168 rx_handler_func_t *rx_handler;
3169 struct net_device *orig_dev;
3170 struct net_device *null_or_dev;
3171 bool deliver_exact = false;
3172 int ret = NET_RX_DROP;
3175 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3177 trace_netif_receive_skb(skb);
3179 /* if we've gotten here through NAPI, check netpoll */
3180 if (netpoll_receive_skb(skb))
3184 skb->skb_iif = skb->dev->ifindex;
3185 orig_dev = skb->dev;
3187 skb_reset_network_header(skb);
3188 skb_reset_transport_header(skb);
3189 skb_reset_mac_len(skb);
3197 __this_cpu_inc(softnet_data.processed);
3199 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3200 skb = vlan_untag(skb);
3205 #ifdef CONFIG_NET_CLS_ACT
3206 if (skb->tc_verd & TC_NCLS) {
3207 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3212 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3213 if (!ptype->dev || ptype->dev == skb->dev) {
3215 ret = deliver_skb(skb, pt_prev, orig_dev);
3220 #ifdef CONFIG_NET_CLS_ACT
3221 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3227 rx_handler = rcu_dereference(skb->dev->rx_handler);
3228 if (vlan_tx_tag_present(skb)) {
3230 ret = deliver_skb(skb, pt_prev, orig_dev);
3233 if (vlan_do_receive(&skb, !rx_handler))
3235 else if (unlikely(!skb))
3241 ret = deliver_skb(skb, pt_prev, orig_dev);
3244 switch (rx_handler(&skb)) {
3245 case RX_HANDLER_CONSUMED:
3247 case RX_HANDLER_ANOTHER:
3249 case RX_HANDLER_EXACT:
3250 deliver_exact = true;
3251 case RX_HANDLER_PASS:
3258 /* deliver only exact match when indicated */
3259 null_or_dev = deliver_exact ? skb->dev : NULL;
3261 type = skb->protocol;
3262 list_for_each_entry_rcu(ptype,
3263 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3264 if (ptype->type == type &&
3265 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3266 ptype->dev == orig_dev)) {
3268 ret = deliver_skb(skb, pt_prev, orig_dev);
3274 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3276 atomic_long_inc(&skb->dev->rx_dropped);
3278 /* Jamal, now you will not able to escape explaining
3279 * me how you were going to use this. :-)
3290 * netif_receive_skb - process receive buffer from network
3291 * @skb: buffer to process
3293 * netif_receive_skb() is the main receive data processing function.
3294 * It always succeeds. The buffer may be dropped during processing
3295 * for congestion control or by the protocol layers.
3297 * This function may only be called from softirq context and interrupts
3298 * should be enabled.
3300 * Return values (usually ignored):
3301 * NET_RX_SUCCESS: no congestion
3302 * NET_RX_DROP: packet was dropped
3304 int netif_receive_skb(struct sk_buff *skb)
3306 net_timestamp_check(netdev_tstamp_prequeue, skb);
3308 if (skb_defer_rx_timestamp(skb))
3309 return NET_RX_SUCCESS;
3312 if (static_branch(&rps_needed)) {
3313 struct rps_dev_flow voidflow, *rflow = &voidflow;
3318 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3321 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3328 return __netif_receive_skb(skb);
3330 EXPORT_SYMBOL(netif_receive_skb);
3332 /* Network device is going away, flush any packets still pending
3333 * Called with irqs disabled.
3335 static void flush_backlog(void *arg)
3337 struct net_device *dev = arg;
3338 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3339 struct sk_buff *skb, *tmp;
3342 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3343 if (skb->dev == dev) {
3344 __skb_unlink(skb, &sd->input_pkt_queue);
3346 input_queue_head_incr(sd);
3351 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3352 if (skb->dev == dev) {
3353 __skb_unlink(skb, &sd->process_queue);
3355 input_queue_head_incr(sd);
3360 static int napi_gro_complete(struct sk_buff *skb)
3362 struct packet_type *ptype;
3363 __be16 type = skb->protocol;
3364 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3367 if (NAPI_GRO_CB(skb)->count == 1) {
3368 skb_shinfo(skb)->gso_size = 0;
3373 list_for_each_entry_rcu(ptype, head, list) {
3374 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3377 err = ptype->gro_complete(skb);
3383 WARN_ON(&ptype->list == head);
3385 return NET_RX_SUCCESS;
3389 return netif_receive_skb(skb);
3392 inline void napi_gro_flush(struct napi_struct *napi)
3394 struct sk_buff *skb, *next;
3396 for (skb = napi->gro_list; skb; skb = next) {
3399 napi_gro_complete(skb);
3402 napi->gro_count = 0;
3403 napi->gro_list = NULL;
3405 EXPORT_SYMBOL(napi_gro_flush);
3407 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3409 struct sk_buff **pp = NULL;
3410 struct packet_type *ptype;
3411 __be16 type = skb->protocol;
3412 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3415 enum gro_result ret;
3417 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3420 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3424 list_for_each_entry_rcu(ptype, head, list) {
3425 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3428 skb_set_network_header(skb, skb_gro_offset(skb));
3429 mac_len = skb->network_header - skb->mac_header;
3430 skb->mac_len = mac_len;
3431 NAPI_GRO_CB(skb)->same_flow = 0;
3432 NAPI_GRO_CB(skb)->flush = 0;
3433 NAPI_GRO_CB(skb)->free = 0;
3435 pp = ptype->gro_receive(&napi->gro_list, skb);
3440 if (&ptype->list == head)
3443 same_flow = NAPI_GRO_CB(skb)->same_flow;
3444 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3447 struct sk_buff *nskb = *pp;
3451 napi_gro_complete(nskb);
3458 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3462 NAPI_GRO_CB(skb)->count = 1;
3463 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3464 skb->next = napi->gro_list;
3465 napi->gro_list = skb;
3469 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3470 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3472 BUG_ON(skb->end - skb->tail < grow);
3474 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3477 skb->data_len -= grow;
3479 skb_shinfo(skb)->frags[0].page_offset += grow;
3480 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3482 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3483 skb_frag_unref(skb, 0);
3484 memmove(skb_shinfo(skb)->frags,
3485 skb_shinfo(skb)->frags + 1,
3486 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3497 EXPORT_SYMBOL(dev_gro_receive);
3499 static inline gro_result_t
3500 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3503 unsigned int maclen = skb->dev->hard_header_len;
3505 for (p = napi->gro_list; p; p = p->next) {
3506 unsigned long diffs;
3508 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3509 diffs |= p->vlan_tci ^ skb->vlan_tci;
3510 if (maclen == ETH_HLEN)
3511 diffs |= compare_ether_header(skb_mac_header(p),
3512 skb_gro_mac_header(skb));
3514 diffs = memcmp(skb_mac_header(p),
3515 skb_gro_mac_header(skb),
3517 NAPI_GRO_CB(p)->same_flow = !diffs;
3518 NAPI_GRO_CB(p)->flush = 0;
3521 return dev_gro_receive(napi, skb);
3524 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3528 if (netif_receive_skb(skb))
3533 case GRO_MERGED_FREE:
3544 EXPORT_SYMBOL(napi_skb_finish);
3546 void skb_gro_reset_offset(struct sk_buff *skb)
3548 NAPI_GRO_CB(skb)->data_offset = 0;
3549 NAPI_GRO_CB(skb)->frag0 = NULL;
3550 NAPI_GRO_CB(skb)->frag0_len = 0;
3552 if (skb->mac_header == skb->tail &&
3553 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3554 NAPI_GRO_CB(skb)->frag0 =
3555 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3556 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3559 EXPORT_SYMBOL(skb_gro_reset_offset);
3561 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3563 skb_gro_reset_offset(skb);
3565 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3567 EXPORT_SYMBOL(napi_gro_receive);
3569 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3571 __skb_pull(skb, skb_headlen(skb));
3572 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3574 skb->dev = napi->dev;
3580 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3582 struct sk_buff *skb = napi->skb;
3585 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3591 EXPORT_SYMBOL(napi_get_frags);
3593 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3599 skb->protocol = eth_type_trans(skb, skb->dev);
3601 if (ret == GRO_HELD)
3602 skb_gro_pull(skb, -ETH_HLEN);
3603 else if (netif_receive_skb(skb))
3608 case GRO_MERGED_FREE:
3609 napi_reuse_skb(napi, skb);
3618 EXPORT_SYMBOL(napi_frags_finish);
3620 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3622 struct sk_buff *skb = napi->skb;
3629 skb_reset_mac_header(skb);
3630 skb_gro_reset_offset(skb);
3632 off = skb_gro_offset(skb);
3633 hlen = off + sizeof(*eth);
3634 eth = skb_gro_header_fast(skb, off);
3635 if (skb_gro_header_hard(skb, hlen)) {
3636 eth = skb_gro_header_slow(skb, hlen, off);
3637 if (unlikely(!eth)) {
3638 napi_reuse_skb(napi, skb);
3644 skb_gro_pull(skb, sizeof(*eth));
3647 * This works because the only protocols we care about don't require
3648 * special handling. We'll fix it up properly at the end.
3650 skb->protocol = eth->h_proto;
3655 EXPORT_SYMBOL(napi_frags_skb);
3657 gro_result_t napi_gro_frags(struct napi_struct *napi)
3659 struct sk_buff *skb = napi_frags_skb(napi);
3664 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3666 EXPORT_SYMBOL(napi_gro_frags);
3669 * net_rps_action sends any pending IPI's for rps.
3670 * Note: called with local irq disabled, but exits with local irq enabled.
3672 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3675 struct softnet_data *remsd = sd->rps_ipi_list;
3678 sd->rps_ipi_list = NULL;
3682 /* Send pending IPI's to kick RPS processing on remote cpus. */
3684 struct softnet_data *next = remsd->rps_ipi_next;
3686 if (cpu_online(remsd->cpu))
3687 __smp_call_function_single(remsd->cpu,
3696 static int process_backlog(struct napi_struct *napi, int quota)
3699 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3702 /* Check if we have pending ipi, its better to send them now,
3703 * not waiting net_rx_action() end.
3705 if (sd->rps_ipi_list) {
3706 local_irq_disable();
3707 net_rps_action_and_irq_enable(sd);
3710 napi->weight = weight_p;
3711 local_irq_disable();
3712 while (work < quota) {
3713 struct sk_buff *skb;
3716 while ((skb = __skb_dequeue(&sd->process_queue))) {
3718 __netif_receive_skb(skb);
3719 local_irq_disable();
3720 input_queue_head_incr(sd);
3721 if (++work >= quota) {
3728 qlen = skb_queue_len(&sd->input_pkt_queue);
3730 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3731 &sd->process_queue);
3733 if (qlen < quota - work) {
3735 * Inline a custom version of __napi_complete().
3736 * only current cpu owns and manipulates this napi,
3737 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3738 * we can use a plain write instead of clear_bit(),
3739 * and we dont need an smp_mb() memory barrier.
3741 list_del(&napi->poll_list);
3744 quota = work + qlen;
3754 * __napi_schedule - schedule for receive
3755 * @n: entry to schedule
3757 * The entry's receive function will be scheduled to run
3759 void __napi_schedule(struct napi_struct *n)
3761 unsigned long flags;
3763 local_irq_save(flags);
3764 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3765 local_irq_restore(flags);
3767 EXPORT_SYMBOL(__napi_schedule);
3769 void __napi_complete(struct napi_struct *n)
3771 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3772 BUG_ON(n->gro_list);
3774 list_del(&n->poll_list);
3775 smp_mb__before_clear_bit();
3776 clear_bit(NAPI_STATE_SCHED, &n->state);
3778 EXPORT_SYMBOL(__napi_complete);
3780 void napi_complete(struct napi_struct *n)
3782 unsigned long flags;
3785 * don't let napi dequeue from the cpu poll list
3786 * just in case its running on a different cpu
3788 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3792 local_irq_save(flags);
3794 local_irq_restore(flags);
3796 EXPORT_SYMBOL(napi_complete);
3798 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3799 int (*poll)(struct napi_struct *, int), int weight)
3801 INIT_LIST_HEAD(&napi->poll_list);
3802 napi->gro_count = 0;
3803 napi->gro_list = NULL;
3806 napi->weight = weight;
3807 list_add(&napi->dev_list, &dev->napi_list);
3809 #ifdef CONFIG_NETPOLL
3810 spin_lock_init(&napi->poll_lock);
3811 napi->poll_owner = -1;
3813 set_bit(NAPI_STATE_SCHED, &napi->state);
3815 EXPORT_SYMBOL(netif_napi_add);
3817 void netif_napi_del(struct napi_struct *napi)
3819 struct sk_buff *skb, *next;
3821 list_del_init(&napi->dev_list);
3822 napi_free_frags(napi);
3824 for (skb = napi->gro_list; skb; skb = next) {
3830 napi->gro_list = NULL;
3831 napi->gro_count = 0;
3833 EXPORT_SYMBOL(netif_napi_del);
3835 static void net_rx_action(struct softirq_action *h)
3837 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3838 unsigned long time_limit = jiffies + 2;
3839 int budget = netdev_budget;
3842 local_irq_disable();
3844 while (!list_empty(&sd->poll_list)) {
3845 struct napi_struct *n;
3848 /* If softirq window is exhuasted then punt.
3849 * Allow this to run for 2 jiffies since which will allow
3850 * an average latency of 1.5/HZ.
3852 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3857 /* Even though interrupts have been re-enabled, this
3858 * access is safe because interrupts can only add new
3859 * entries to the tail of this list, and only ->poll()
3860 * calls can remove this head entry from the list.
3862 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3864 have = netpoll_poll_lock(n);
3868 /* This NAPI_STATE_SCHED test is for avoiding a race
3869 * with netpoll's poll_napi(). Only the entity which
3870 * obtains the lock and sees NAPI_STATE_SCHED set will
3871 * actually make the ->poll() call. Therefore we avoid
3872 * accidentally calling ->poll() when NAPI is not scheduled.
3875 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3876 work = n->poll(n, weight);
3880 WARN_ON_ONCE(work > weight);
3884 local_irq_disable();
3886 /* Drivers must not modify the NAPI state if they
3887 * consume the entire weight. In such cases this code
3888 * still "owns" the NAPI instance and therefore can
3889 * move the instance around on the list at-will.
3891 if (unlikely(work == weight)) {
3892 if (unlikely(napi_disable_pending(n))) {
3895 local_irq_disable();
3897 list_move_tail(&n->poll_list, &sd->poll_list);
3900 netpoll_poll_unlock(have);
3903 net_rps_action_and_irq_enable(sd);
3905 #ifdef CONFIG_NET_DMA
3907 * There may not be any more sk_buffs coming right now, so push
3908 * any pending DMA copies to hardware
3910 dma_issue_pending_all();
3917 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3921 static gifconf_func_t *gifconf_list[NPROTO];
3924 * register_gifconf - register a SIOCGIF handler
3925 * @family: Address family
3926 * @gifconf: Function handler
3928 * Register protocol dependent address dumping routines. The handler
3929 * that is passed must not be freed or reused until it has been replaced
3930 * by another handler.
3932 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3934 if (family >= NPROTO)
3936 gifconf_list[family] = gifconf;
3939 EXPORT_SYMBOL(register_gifconf);
3943 * Map an interface index to its name (SIOCGIFNAME)
3947 * We need this ioctl for efficient implementation of the
3948 * if_indextoname() function required by the IPv6 API. Without
3949 * it, we would have to search all the interfaces to find a
3953 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3955 struct net_device *dev;
3959 * Fetch the caller's info block.
3962 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3966 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3972 strcpy(ifr.ifr_name, dev->name);
3975 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3981 * Perform a SIOCGIFCONF call. This structure will change
3982 * size eventually, and there is nothing I can do about it.
3983 * Thus we will need a 'compatibility mode'.
3986 static int dev_ifconf(struct net *net, char __user *arg)
3989 struct net_device *dev;
3996 * Fetch the caller's info block.
3999 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4006 * Loop over the interfaces, and write an info block for each.
4010 for_each_netdev(net, dev) {
4011 for (i = 0; i < NPROTO; i++) {
4012 if (gifconf_list[i]) {
4015 done = gifconf_list[i](dev, NULL, 0);
4017 done = gifconf_list[i](dev, pos + total,
4027 * All done. Write the updated control block back to the caller.
4029 ifc.ifc_len = total;
4032 * Both BSD and Solaris return 0 here, so we do too.
4034 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4037 #ifdef CONFIG_PROC_FS
4039 #define BUCKET_SPACE (32 - NETDEV_HASHBITS)
4041 struct dev_iter_state {
4042 struct seq_net_private p;
4043 unsigned int pos; /* bucket << BUCKET_SPACE + offset */
4046 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4047 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4048 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4050 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq)
4052 struct dev_iter_state *state = seq->private;
4053 struct net *net = seq_file_net(seq);
4054 struct net_device *dev;
4055 struct hlist_node *p;
4056 struct hlist_head *h;
4057 unsigned int count, bucket, offset;
4059 bucket = get_bucket(state->pos);
4060 offset = get_offset(state->pos);
4061 h = &net->dev_name_head[bucket];
4063 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4064 if (count++ == offset) {
4065 state->pos = set_bucket_offset(bucket, count);
4073 static inline struct net_device *dev_from_new_bucket(struct seq_file *seq)
4075 struct dev_iter_state *state = seq->private;
4076 struct net_device *dev;
4077 unsigned int bucket;
4079 bucket = get_bucket(state->pos);
4081 dev = dev_from_same_bucket(seq);
4086 state->pos = set_bucket_offset(bucket, 0);
4087 } while (bucket < NETDEV_HASHENTRIES);
4093 * This is invoked by the /proc filesystem handler to display a device
4096 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4099 struct dev_iter_state *state = seq->private;
4103 return SEQ_START_TOKEN;
4105 /* check for end of the hash */
4106 if (state->pos == 0 && *pos > 1)
4109 return dev_from_new_bucket(seq);
4112 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4114 struct net_device *dev;
4118 if (v == SEQ_START_TOKEN)
4119 return dev_from_new_bucket(seq);
4121 dev = dev_from_same_bucket(seq);
4125 return dev_from_new_bucket(seq);
4128 void dev_seq_stop(struct seq_file *seq, void *v)
4134 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4136 struct rtnl_link_stats64 temp;
4137 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4139 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4140 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4141 dev->name, stats->rx_bytes, stats->rx_packets,
4143 stats->rx_dropped + stats->rx_missed_errors,
4144 stats->rx_fifo_errors,
4145 stats->rx_length_errors + stats->rx_over_errors +
4146 stats->rx_crc_errors + stats->rx_frame_errors,
4147 stats->rx_compressed, stats->multicast,
4148 stats->tx_bytes, stats->tx_packets,
4149 stats->tx_errors, stats->tx_dropped,
4150 stats->tx_fifo_errors, stats->collisions,
4151 stats->tx_carrier_errors +
4152 stats->tx_aborted_errors +
4153 stats->tx_window_errors +
4154 stats->tx_heartbeat_errors,
4155 stats->tx_compressed);
4159 * Called from the PROCfs module. This now uses the new arbitrary sized
4160 * /proc/net interface to create /proc/net/dev
4162 static int dev_seq_show(struct seq_file *seq, void *v)
4164 if (v == SEQ_START_TOKEN)
4165 seq_puts(seq, "Inter-| Receive "
4167 " face |bytes packets errs drop fifo frame "
4168 "compressed multicast|bytes packets errs "
4169 "drop fifo colls carrier compressed\n");
4171 dev_seq_printf_stats(seq, v);
4175 static struct softnet_data *softnet_get_online(loff_t *pos)
4177 struct softnet_data *sd = NULL;
4179 while (*pos < nr_cpu_ids)
4180 if (cpu_online(*pos)) {
4181 sd = &per_cpu(softnet_data, *pos);
4188 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4190 return softnet_get_online(pos);
4193 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4196 return softnet_get_online(pos);
4199 static void softnet_seq_stop(struct seq_file *seq, void *v)
4203 static int softnet_seq_show(struct seq_file *seq, void *v)
4205 struct softnet_data *sd = v;
4207 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4208 sd->processed, sd->dropped, sd->time_squeeze, 0,
4209 0, 0, 0, 0, /* was fastroute */
4210 sd->cpu_collision, sd->received_rps);
4214 static const struct seq_operations dev_seq_ops = {
4215 .start = dev_seq_start,
4216 .next = dev_seq_next,
4217 .stop = dev_seq_stop,
4218 .show = dev_seq_show,
4221 static int dev_seq_open(struct inode *inode, struct file *file)
4223 return seq_open_net(inode, file, &dev_seq_ops,
4224 sizeof(struct dev_iter_state));
4227 int dev_seq_open_ops(struct inode *inode, struct file *file,
4228 const struct seq_operations *ops)
4230 return seq_open_net(inode, file, ops, sizeof(struct dev_iter_state));
4233 static const struct file_operations dev_seq_fops = {
4234 .owner = THIS_MODULE,
4235 .open = dev_seq_open,
4237 .llseek = seq_lseek,
4238 .release = seq_release_net,
4241 static const struct seq_operations softnet_seq_ops = {
4242 .start = softnet_seq_start,
4243 .next = softnet_seq_next,
4244 .stop = softnet_seq_stop,
4245 .show = softnet_seq_show,
4248 static int softnet_seq_open(struct inode *inode, struct file *file)
4250 return seq_open(file, &softnet_seq_ops);
4253 static const struct file_operations softnet_seq_fops = {
4254 .owner = THIS_MODULE,
4255 .open = softnet_seq_open,
4257 .llseek = seq_lseek,
4258 .release = seq_release,
4261 static void *ptype_get_idx(loff_t pos)
4263 struct packet_type *pt = NULL;
4267 list_for_each_entry_rcu(pt, &ptype_all, list) {
4273 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4274 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4283 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4287 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4290 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4292 struct packet_type *pt;
4293 struct list_head *nxt;
4297 if (v == SEQ_START_TOKEN)
4298 return ptype_get_idx(0);
4301 nxt = pt->list.next;
4302 if (pt->type == htons(ETH_P_ALL)) {
4303 if (nxt != &ptype_all)
4306 nxt = ptype_base[0].next;
4308 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4310 while (nxt == &ptype_base[hash]) {
4311 if (++hash >= PTYPE_HASH_SIZE)
4313 nxt = ptype_base[hash].next;
4316 return list_entry(nxt, struct packet_type, list);
4319 static void ptype_seq_stop(struct seq_file *seq, void *v)
4325 static int ptype_seq_show(struct seq_file *seq, void *v)
4327 struct packet_type *pt = v;
4329 if (v == SEQ_START_TOKEN)
4330 seq_puts(seq, "Type Device Function\n");
4331 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4332 if (pt->type == htons(ETH_P_ALL))
4333 seq_puts(seq, "ALL ");
4335 seq_printf(seq, "%04x", ntohs(pt->type));
4337 seq_printf(seq, " %-8s %pF\n",
4338 pt->dev ? pt->dev->name : "", pt->func);
4344 static const struct seq_operations ptype_seq_ops = {
4345 .start = ptype_seq_start,
4346 .next = ptype_seq_next,
4347 .stop = ptype_seq_stop,
4348 .show = ptype_seq_show,
4351 static int ptype_seq_open(struct inode *inode, struct file *file)
4353 return seq_open_net(inode, file, &ptype_seq_ops,
4354 sizeof(struct seq_net_private));
4357 static const struct file_operations ptype_seq_fops = {
4358 .owner = THIS_MODULE,
4359 .open = ptype_seq_open,
4361 .llseek = seq_lseek,
4362 .release = seq_release_net,
4366 static int __net_init dev_proc_net_init(struct net *net)
4370 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4372 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4374 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4377 if (wext_proc_init(net))
4383 proc_net_remove(net, "ptype");
4385 proc_net_remove(net, "softnet_stat");
4387 proc_net_remove(net, "dev");
4391 static void __net_exit dev_proc_net_exit(struct net *net)
4393 wext_proc_exit(net);
4395 proc_net_remove(net, "ptype");
4396 proc_net_remove(net, "softnet_stat");
4397 proc_net_remove(net, "dev");
4400 static struct pernet_operations __net_initdata dev_proc_ops = {
4401 .init = dev_proc_net_init,
4402 .exit = dev_proc_net_exit,
4405 static int __init dev_proc_init(void)
4407 return register_pernet_subsys(&dev_proc_ops);
4410 #define dev_proc_init() 0
4411 #endif /* CONFIG_PROC_FS */
4415 * netdev_set_master - set up master pointer
4416 * @slave: slave device
4417 * @master: new master device
4419 * Changes the master device of the slave. Pass %NULL to break the
4420 * bonding. The caller must hold the RTNL semaphore. On a failure
4421 * a negative errno code is returned. On success the reference counts
4422 * are adjusted and the function returns zero.
4424 int netdev_set_master(struct net_device *slave, struct net_device *master)
4426 struct net_device *old = slave->master;
4436 slave->master = master;
4442 EXPORT_SYMBOL(netdev_set_master);
4445 * netdev_set_bond_master - set up bonding master/slave pair
4446 * @slave: slave device
4447 * @master: new master device
4449 * Changes the master device of the slave. Pass %NULL to break the
4450 * bonding. The caller must hold the RTNL semaphore. On a failure
4451 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4452 * to the routing socket and the function returns zero.
4454 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4460 err = netdev_set_master(slave, master);
4464 slave->flags |= IFF_SLAVE;
4466 slave->flags &= ~IFF_SLAVE;
4468 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4471 EXPORT_SYMBOL(netdev_set_bond_master);
4473 static void dev_change_rx_flags(struct net_device *dev, int flags)
4475 const struct net_device_ops *ops = dev->netdev_ops;
4477 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4478 ops->ndo_change_rx_flags(dev, flags);
4481 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4483 unsigned int old_flags = dev->flags;
4489 dev->flags |= IFF_PROMISC;
4490 dev->promiscuity += inc;
4491 if (dev->promiscuity == 0) {
4494 * If inc causes overflow, untouch promisc and return error.
4497 dev->flags &= ~IFF_PROMISC;
4499 dev->promiscuity -= inc;
4500 printk(KERN_WARNING "%s: promiscuity touches roof, "
4501 "set promiscuity failed, promiscuity feature "
4502 "of device might be broken.\n", dev->name);
4506 if (dev->flags != old_flags) {
4507 printk(KERN_INFO "device %s %s promiscuous mode\n",
4508 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4510 if (audit_enabled) {
4511 current_uid_gid(&uid, &gid);
4512 audit_log(current->audit_context, GFP_ATOMIC,
4513 AUDIT_ANOM_PROMISCUOUS,
4514 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4515 dev->name, (dev->flags & IFF_PROMISC),
4516 (old_flags & IFF_PROMISC),
4517 audit_get_loginuid(current),
4519 audit_get_sessionid(current));
4522 dev_change_rx_flags(dev, IFF_PROMISC);
4528 * dev_set_promiscuity - update promiscuity count on a device
4532 * Add or remove promiscuity from a device. While the count in the device
4533 * remains above zero the interface remains promiscuous. Once it hits zero
4534 * the device reverts back to normal filtering operation. A negative inc
4535 * value is used to drop promiscuity on the device.
4536 * Return 0 if successful or a negative errno code on error.
4538 int dev_set_promiscuity(struct net_device *dev, int inc)
4540 unsigned int old_flags = dev->flags;
4543 err = __dev_set_promiscuity(dev, inc);
4546 if (dev->flags != old_flags)
4547 dev_set_rx_mode(dev);
4550 EXPORT_SYMBOL(dev_set_promiscuity);
4553 * dev_set_allmulti - update allmulti count on a device
4557 * Add or remove reception of all multicast frames to a device. While the
4558 * count in the device remains above zero the interface remains listening
4559 * to all interfaces. Once it hits zero the device reverts back to normal
4560 * filtering operation. A negative @inc value is used to drop the counter
4561 * when releasing a resource needing all multicasts.
4562 * Return 0 if successful or a negative errno code on error.
4565 int dev_set_allmulti(struct net_device *dev, int inc)
4567 unsigned int old_flags = dev->flags;
4571 dev->flags |= IFF_ALLMULTI;
4572 dev->allmulti += inc;
4573 if (dev->allmulti == 0) {
4576 * If inc causes overflow, untouch allmulti and return error.
4579 dev->flags &= ~IFF_ALLMULTI;
4581 dev->allmulti -= inc;
4582 printk(KERN_WARNING "%s: allmulti touches roof, "
4583 "set allmulti failed, allmulti feature of "
4584 "device might be broken.\n", dev->name);
4588 if (dev->flags ^ old_flags) {
4589 dev_change_rx_flags(dev, IFF_ALLMULTI);
4590 dev_set_rx_mode(dev);
4594 EXPORT_SYMBOL(dev_set_allmulti);
4597 * Upload unicast and multicast address lists to device and
4598 * configure RX filtering. When the device doesn't support unicast
4599 * filtering it is put in promiscuous mode while unicast addresses
4602 void __dev_set_rx_mode(struct net_device *dev)
4604 const struct net_device_ops *ops = dev->netdev_ops;
4606 /* dev_open will call this function so the list will stay sane. */
4607 if (!(dev->flags&IFF_UP))
4610 if (!netif_device_present(dev))
4613 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4614 /* Unicast addresses changes may only happen under the rtnl,
4615 * therefore calling __dev_set_promiscuity here is safe.
4617 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4618 __dev_set_promiscuity(dev, 1);
4619 dev->uc_promisc = true;
4620 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4621 __dev_set_promiscuity(dev, -1);
4622 dev->uc_promisc = false;
4626 if (ops->ndo_set_rx_mode)
4627 ops->ndo_set_rx_mode(dev);
4630 void dev_set_rx_mode(struct net_device *dev)
4632 netif_addr_lock_bh(dev);
4633 __dev_set_rx_mode(dev);
4634 netif_addr_unlock_bh(dev);
4638 * dev_get_flags - get flags reported to userspace
4641 * Get the combination of flag bits exported through APIs to userspace.
4643 unsigned dev_get_flags(const struct net_device *dev)
4647 flags = (dev->flags & ~(IFF_PROMISC |
4652 (dev->gflags & (IFF_PROMISC |
4655 if (netif_running(dev)) {
4656 if (netif_oper_up(dev))
4657 flags |= IFF_RUNNING;
4658 if (netif_carrier_ok(dev))
4659 flags |= IFF_LOWER_UP;
4660 if (netif_dormant(dev))
4661 flags |= IFF_DORMANT;
4666 EXPORT_SYMBOL(dev_get_flags);
4668 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4670 unsigned int old_flags = dev->flags;
4676 * Set the flags on our device.
4679 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4680 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4682 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4686 * Load in the correct multicast list now the flags have changed.
4689 if ((old_flags ^ flags) & IFF_MULTICAST)
4690 dev_change_rx_flags(dev, IFF_MULTICAST);
4692 dev_set_rx_mode(dev);
4695 * Have we downed the interface. We handle IFF_UP ourselves
4696 * according to user attempts to set it, rather than blindly
4701 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4702 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4705 dev_set_rx_mode(dev);
4708 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4709 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4711 dev->gflags ^= IFF_PROMISC;
4712 dev_set_promiscuity(dev, inc);
4715 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4716 is important. Some (broken) drivers set IFF_PROMISC, when
4717 IFF_ALLMULTI is requested not asking us and not reporting.
4719 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4720 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4722 dev->gflags ^= IFF_ALLMULTI;
4723 dev_set_allmulti(dev, inc);
4729 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4731 unsigned int changes = dev->flags ^ old_flags;
4733 if (changes & IFF_UP) {
4734 if (dev->flags & IFF_UP)
4735 call_netdevice_notifiers(NETDEV_UP, dev);
4737 call_netdevice_notifiers(NETDEV_DOWN, dev);
4740 if (dev->flags & IFF_UP &&
4741 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4742 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4746 * dev_change_flags - change device settings
4748 * @flags: device state flags
4750 * Change settings on device based state flags. The flags are
4751 * in the userspace exported format.
4753 int dev_change_flags(struct net_device *dev, unsigned int flags)
4756 unsigned int changes, old_flags = dev->flags;
4758 ret = __dev_change_flags(dev, flags);
4762 changes = old_flags ^ dev->flags;
4764 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4766 __dev_notify_flags(dev, old_flags);
4769 EXPORT_SYMBOL(dev_change_flags);
4772 * dev_set_mtu - Change maximum transfer unit
4774 * @new_mtu: new transfer unit
4776 * Change the maximum transfer size of the network device.
4778 int dev_set_mtu(struct net_device *dev, int new_mtu)
4780 const struct net_device_ops *ops = dev->netdev_ops;
4783 if (new_mtu == dev->mtu)
4786 /* MTU must be positive. */
4790 if (!netif_device_present(dev))
4794 if (ops->ndo_change_mtu)
4795 err = ops->ndo_change_mtu(dev, new_mtu);
4799 if (!err && dev->flags & IFF_UP)
4800 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4803 EXPORT_SYMBOL(dev_set_mtu);
4806 * dev_set_group - Change group this device belongs to
4808 * @new_group: group this device should belong to
4810 void dev_set_group(struct net_device *dev, int new_group)
4812 dev->group = new_group;
4814 EXPORT_SYMBOL(dev_set_group);
4817 * dev_set_mac_address - Change Media Access Control Address
4821 * Change the hardware (MAC) address of the device
4823 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4825 const struct net_device_ops *ops = dev->netdev_ops;
4828 if (!ops->ndo_set_mac_address)
4830 if (sa->sa_family != dev->type)
4832 if (!netif_device_present(dev))
4834 err = ops->ndo_set_mac_address(dev, sa);
4836 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4839 EXPORT_SYMBOL(dev_set_mac_address);
4842 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4844 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4847 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4853 case SIOCGIFFLAGS: /* Get interface flags */
4854 ifr->ifr_flags = (short) dev_get_flags(dev);
4857 case SIOCGIFMETRIC: /* Get the metric on the interface
4858 (currently unused) */
4859 ifr->ifr_metric = 0;
4862 case SIOCGIFMTU: /* Get the MTU of a device */
4863 ifr->ifr_mtu = dev->mtu;
4868 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4870 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4871 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4872 ifr->ifr_hwaddr.sa_family = dev->type;
4880 ifr->ifr_map.mem_start = dev->mem_start;
4881 ifr->ifr_map.mem_end = dev->mem_end;
4882 ifr->ifr_map.base_addr = dev->base_addr;
4883 ifr->ifr_map.irq = dev->irq;
4884 ifr->ifr_map.dma = dev->dma;
4885 ifr->ifr_map.port = dev->if_port;
4889 ifr->ifr_ifindex = dev->ifindex;
4893 ifr->ifr_qlen = dev->tx_queue_len;
4897 /* dev_ioctl() should ensure this case
4909 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4911 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4914 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4915 const struct net_device_ops *ops;
4920 ops = dev->netdev_ops;
4923 case SIOCSIFFLAGS: /* Set interface flags */
4924 return dev_change_flags(dev, ifr->ifr_flags);
4926 case SIOCSIFMETRIC: /* Set the metric on the interface
4927 (currently unused) */
4930 case SIOCSIFMTU: /* Set the MTU of a device */
4931 return dev_set_mtu(dev, ifr->ifr_mtu);
4934 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4936 case SIOCSIFHWBROADCAST:
4937 if (ifr->ifr_hwaddr.sa_family != dev->type)
4939 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4940 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4941 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4945 if (ops->ndo_set_config) {
4946 if (!netif_device_present(dev))
4948 return ops->ndo_set_config(dev, &ifr->ifr_map);
4953 if (!ops->ndo_set_rx_mode ||
4954 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4956 if (!netif_device_present(dev))
4958 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4961 if (!ops->ndo_set_rx_mode ||
4962 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4964 if (!netif_device_present(dev))
4966 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4969 if (ifr->ifr_qlen < 0)
4971 dev->tx_queue_len = ifr->ifr_qlen;
4975 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4976 return dev_change_name(dev, ifr->ifr_newname);
4979 err = net_hwtstamp_validate(ifr);
4985 * Unknown or private ioctl
4988 if ((cmd >= SIOCDEVPRIVATE &&
4989 cmd <= SIOCDEVPRIVATE + 15) ||
4990 cmd == SIOCBONDENSLAVE ||
4991 cmd == SIOCBONDRELEASE ||
4992 cmd == SIOCBONDSETHWADDR ||
4993 cmd == SIOCBONDSLAVEINFOQUERY ||
4994 cmd == SIOCBONDINFOQUERY ||
4995 cmd == SIOCBONDCHANGEACTIVE ||
4996 cmd == SIOCGMIIPHY ||
4997 cmd == SIOCGMIIREG ||
4998 cmd == SIOCSMIIREG ||
4999 cmd == SIOCBRADDIF ||
5000 cmd == SIOCBRDELIF ||
5001 cmd == SIOCSHWTSTAMP ||
5002 cmd == SIOCWANDEV) {
5004 if (ops->ndo_do_ioctl) {
5005 if (netif_device_present(dev))
5006 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5018 * This function handles all "interface"-type I/O control requests. The actual
5019 * 'doing' part of this is dev_ifsioc above.
5023 * dev_ioctl - network device ioctl
5024 * @net: the applicable net namespace
5025 * @cmd: command to issue
5026 * @arg: pointer to a struct ifreq in user space
5028 * Issue ioctl functions to devices. This is normally called by the
5029 * user space syscall interfaces but can sometimes be useful for
5030 * other purposes. The return value is the return from the syscall if
5031 * positive or a negative errno code on error.
5034 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5040 /* One special case: SIOCGIFCONF takes ifconf argument
5041 and requires shared lock, because it sleeps writing
5045 if (cmd == SIOCGIFCONF) {
5047 ret = dev_ifconf(net, (char __user *) arg);
5051 if (cmd == SIOCGIFNAME)
5052 return dev_ifname(net, (struct ifreq __user *)arg);
5054 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5057 ifr.ifr_name[IFNAMSIZ-1] = 0;
5059 colon = strchr(ifr.ifr_name, ':');
5064 * See which interface the caller is talking about.
5069 * These ioctl calls:
5070 * - can be done by all.
5071 * - atomic and do not require locking.
5082 dev_load(net, ifr.ifr_name);
5084 ret = dev_ifsioc_locked(net, &ifr, cmd);
5089 if (copy_to_user(arg, &ifr,
5090 sizeof(struct ifreq)))
5096 dev_load(net, ifr.ifr_name);
5098 ret = dev_ethtool(net, &ifr);
5103 if (copy_to_user(arg, &ifr,
5104 sizeof(struct ifreq)))
5110 * These ioctl calls:
5111 * - require superuser power.
5112 * - require strict serialization.
5118 if (!capable(CAP_NET_ADMIN))
5120 dev_load(net, ifr.ifr_name);
5122 ret = dev_ifsioc(net, &ifr, cmd);
5127 if (copy_to_user(arg, &ifr,
5128 sizeof(struct ifreq)))
5134 * These ioctl calls:
5135 * - require superuser power.
5136 * - require strict serialization.
5137 * - do not return a value
5147 case SIOCSIFHWBROADCAST:
5150 case SIOCBONDENSLAVE:
5151 case SIOCBONDRELEASE:
5152 case SIOCBONDSETHWADDR:
5153 case SIOCBONDCHANGEACTIVE:
5157 if (!capable(CAP_NET_ADMIN))
5160 case SIOCBONDSLAVEINFOQUERY:
5161 case SIOCBONDINFOQUERY:
5162 dev_load(net, ifr.ifr_name);
5164 ret = dev_ifsioc(net, &ifr, cmd);
5169 /* Get the per device memory space. We can add this but
5170 * currently do not support it */
5172 /* Set the per device memory buffer space.
5173 * Not applicable in our case */
5178 * Unknown or private ioctl.
5181 if (cmd == SIOCWANDEV ||
5182 (cmd >= SIOCDEVPRIVATE &&
5183 cmd <= SIOCDEVPRIVATE + 15)) {
5184 dev_load(net, ifr.ifr_name);
5186 ret = dev_ifsioc(net, &ifr, cmd);
5188 if (!ret && copy_to_user(arg, &ifr,
5189 sizeof(struct ifreq)))
5193 /* Take care of Wireless Extensions */
5194 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5195 return wext_handle_ioctl(net, &ifr, cmd, arg);
5202 * dev_new_index - allocate an ifindex
5203 * @net: the applicable net namespace
5205 * Returns a suitable unique value for a new device interface
5206 * number. The caller must hold the rtnl semaphore or the
5207 * dev_base_lock to be sure it remains unique.
5209 static int dev_new_index(struct net *net)
5215 if (!__dev_get_by_index(net, ifindex))
5220 /* Delayed registration/unregisteration */
5221 static LIST_HEAD(net_todo_list);
5223 static void net_set_todo(struct net_device *dev)
5225 list_add_tail(&dev->todo_list, &net_todo_list);
5228 static void rollback_registered_many(struct list_head *head)
5230 struct net_device *dev, *tmp;
5232 BUG_ON(dev_boot_phase);
5235 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5236 /* Some devices call without registering
5237 * for initialization unwind. Remove those
5238 * devices and proceed with the remaining.
5240 if (dev->reg_state == NETREG_UNINITIALIZED) {
5241 pr_debug("unregister_netdevice: device %s/%p never "
5242 "was registered\n", dev->name, dev);
5245 list_del(&dev->unreg_list);
5248 dev->dismantle = true;
5249 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5252 /* If device is running, close it first. */
5253 dev_close_many(head);
5255 list_for_each_entry(dev, head, unreg_list) {
5256 /* And unlink it from device chain. */
5257 unlist_netdevice(dev);
5259 dev->reg_state = NETREG_UNREGISTERING;
5264 list_for_each_entry(dev, head, unreg_list) {
5265 /* Shutdown queueing discipline. */
5269 /* Notify protocols, that we are about to destroy
5270 this device. They should clean all the things.
5272 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5274 if (!dev->rtnl_link_ops ||
5275 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5276 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5279 * Flush the unicast and multicast chains
5284 if (dev->netdev_ops->ndo_uninit)
5285 dev->netdev_ops->ndo_uninit(dev);
5287 /* Notifier chain MUST detach us from master device. */
5288 WARN_ON(dev->master);
5290 /* Remove entries from kobject tree */
5291 netdev_unregister_kobject(dev);
5294 /* Process any work delayed until the end of the batch */
5295 dev = list_first_entry(head, struct net_device, unreg_list);
5296 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5300 list_for_each_entry(dev, head, unreg_list)
5304 static void rollback_registered(struct net_device *dev)
5308 list_add(&dev->unreg_list, &single);
5309 rollback_registered_many(&single);
5313 static netdev_features_t netdev_fix_features(struct net_device *dev,
5314 netdev_features_t features)
5316 /* Fix illegal checksum combinations */
5317 if ((features & NETIF_F_HW_CSUM) &&
5318 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5319 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5320 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5323 /* Fix illegal SG+CSUM combinations. */
5324 if ((features & NETIF_F_SG) &&
5325 !(features & NETIF_F_ALL_CSUM)) {
5327 "Dropping NETIF_F_SG since no checksum feature.\n");
5328 features &= ~NETIF_F_SG;
5331 /* TSO requires that SG is present as well. */
5332 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5333 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5334 features &= ~NETIF_F_ALL_TSO;
5337 /* TSO ECN requires that TSO is present as well. */
5338 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5339 features &= ~NETIF_F_TSO_ECN;
5341 /* Software GSO depends on SG. */
5342 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5343 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5344 features &= ~NETIF_F_GSO;
5347 /* UFO needs SG and checksumming */
5348 if (features & NETIF_F_UFO) {
5349 /* maybe split UFO into V4 and V6? */
5350 if (!((features & NETIF_F_GEN_CSUM) ||
5351 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5352 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5354 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5355 features &= ~NETIF_F_UFO;
5358 if (!(features & NETIF_F_SG)) {
5360 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5361 features &= ~NETIF_F_UFO;
5368 int __netdev_update_features(struct net_device *dev)
5370 netdev_features_t features;
5375 features = netdev_get_wanted_features(dev);
5377 if (dev->netdev_ops->ndo_fix_features)
5378 features = dev->netdev_ops->ndo_fix_features(dev, features);
5380 /* driver might be less strict about feature dependencies */
5381 features = netdev_fix_features(dev, features);
5383 if (dev->features == features)
5386 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5387 &dev->features, &features);
5389 if (dev->netdev_ops->ndo_set_features)
5390 err = dev->netdev_ops->ndo_set_features(dev, features);
5392 if (unlikely(err < 0)) {
5394 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5395 err, &features, &dev->features);
5400 dev->features = features;
5406 * netdev_update_features - recalculate device features
5407 * @dev: the device to check
5409 * Recalculate dev->features set and send notifications if it
5410 * has changed. Should be called after driver or hardware dependent
5411 * conditions might have changed that influence the features.
5413 void netdev_update_features(struct net_device *dev)
5415 if (__netdev_update_features(dev))
5416 netdev_features_change(dev);
5418 EXPORT_SYMBOL(netdev_update_features);
5421 * netdev_change_features - recalculate device features
5422 * @dev: the device to check
5424 * Recalculate dev->features set and send notifications even
5425 * if they have not changed. Should be called instead of
5426 * netdev_update_features() if also dev->vlan_features might
5427 * have changed to allow the changes to be propagated to stacked
5430 void netdev_change_features(struct net_device *dev)
5432 __netdev_update_features(dev);
5433 netdev_features_change(dev);
5435 EXPORT_SYMBOL(netdev_change_features);
5438 * netif_stacked_transfer_operstate - transfer operstate
5439 * @rootdev: the root or lower level device to transfer state from
5440 * @dev: the device to transfer operstate to
5442 * Transfer operational state from root to device. This is normally
5443 * called when a stacking relationship exists between the root
5444 * device and the device(a leaf device).
5446 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5447 struct net_device *dev)
5449 if (rootdev->operstate == IF_OPER_DORMANT)
5450 netif_dormant_on(dev);
5452 netif_dormant_off(dev);
5454 if (netif_carrier_ok(rootdev)) {
5455 if (!netif_carrier_ok(dev))
5456 netif_carrier_on(dev);
5458 if (netif_carrier_ok(dev))
5459 netif_carrier_off(dev);
5462 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5465 static int netif_alloc_rx_queues(struct net_device *dev)
5467 unsigned int i, count = dev->num_rx_queues;
5468 struct netdev_rx_queue *rx;
5472 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5474 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5479 for (i = 0; i < count; i++)
5485 static void netdev_init_one_queue(struct net_device *dev,
5486 struct netdev_queue *queue, void *_unused)
5488 /* Initialize queue lock */
5489 spin_lock_init(&queue->_xmit_lock);
5490 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5491 queue->xmit_lock_owner = -1;
5492 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5495 dql_init(&queue->dql, HZ);
5499 static int netif_alloc_netdev_queues(struct net_device *dev)
5501 unsigned int count = dev->num_tx_queues;
5502 struct netdev_queue *tx;
5506 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5508 pr_err("netdev: Unable to allocate %u tx queues.\n",
5514 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5515 spin_lock_init(&dev->tx_global_lock);
5521 * register_netdevice - register a network device
5522 * @dev: device to register
5524 * Take a completed network device structure and add it to the kernel
5525 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5526 * chain. 0 is returned on success. A negative errno code is returned
5527 * on a failure to set up the device, or if the name is a duplicate.
5529 * Callers must hold the rtnl semaphore. You may want
5530 * register_netdev() instead of this.
5533 * The locking appears insufficient to guarantee two parallel registers
5534 * will not get the same name.
5537 int register_netdevice(struct net_device *dev)
5540 struct net *net = dev_net(dev);
5542 BUG_ON(dev_boot_phase);
5547 /* When net_device's are persistent, this will be fatal. */
5548 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5551 spin_lock_init(&dev->addr_list_lock);
5552 netdev_set_addr_lockdep_class(dev);
5556 ret = dev_get_valid_name(dev, dev->name);
5560 /* Init, if this function is available */
5561 if (dev->netdev_ops->ndo_init) {
5562 ret = dev->netdev_ops->ndo_init(dev);
5570 dev->ifindex = dev_new_index(net);
5571 if (dev->iflink == -1)
5572 dev->iflink = dev->ifindex;
5574 /* Transfer changeable features to wanted_features and enable
5575 * software offloads (GSO and GRO).
5577 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5578 dev->features |= NETIF_F_SOFT_FEATURES;
5579 dev->wanted_features = dev->features & dev->hw_features;
5581 /* Turn on no cache copy if HW is doing checksum */
5582 if (!(dev->flags & IFF_LOOPBACK)) {
5583 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5584 if (dev->features & NETIF_F_ALL_CSUM) {
5585 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5586 dev->features |= NETIF_F_NOCACHE_COPY;
5590 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5592 dev->vlan_features |= NETIF_F_HIGHDMA;
5594 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5595 ret = notifier_to_errno(ret);
5599 ret = netdev_register_kobject(dev);
5602 dev->reg_state = NETREG_REGISTERED;
5604 __netdev_update_features(dev);
5607 * Default initial state at registry is that the
5608 * device is present.
5611 set_bit(__LINK_STATE_PRESENT, &dev->state);
5613 dev_init_scheduler(dev);
5615 list_netdevice(dev);
5617 /* Notify protocols, that a new device appeared. */
5618 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5619 ret = notifier_to_errno(ret);
5621 rollback_registered(dev);
5622 dev->reg_state = NETREG_UNREGISTERED;
5625 * Prevent userspace races by waiting until the network
5626 * device is fully setup before sending notifications.
5628 if (!dev->rtnl_link_ops ||
5629 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5630 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5636 if (dev->netdev_ops->ndo_uninit)
5637 dev->netdev_ops->ndo_uninit(dev);
5640 EXPORT_SYMBOL(register_netdevice);
5643 * init_dummy_netdev - init a dummy network device for NAPI
5644 * @dev: device to init
5646 * This takes a network device structure and initialize the minimum
5647 * amount of fields so it can be used to schedule NAPI polls without
5648 * registering a full blown interface. This is to be used by drivers
5649 * that need to tie several hardware interfaces to a single NAPI
5650 * poll scheduler due to HW limitations.
5652 int init_dummy_netdev(struct net_device *dev)
5654 /* Clear everything. Note we don't initialize spinlocks
5655 * are they aren't supposed to be taken by any of the
5656 * NAPI code and this dummy netdev is supposed to be
5657 * only ever used for NAPI polls
5659 memset(dev, 0, sizeof(struct net_device));
5661 /* make sure we BUG if trying to hit standard
5662 * register/unregister code path
5664 dev->reg_state = NETREG_DUMMY;
5666 /* NAPI wants this */
5667 INIT_LIST_HEAD(&dev->napi_list);
5669 /* a dummy interface is started by default */
5670 set_bit(__LINK_STATE_PRESENT, &dev->state);
5671 set_bit(__LINK_STATE_START, &dev->state);
5673 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5674 * because users of this 'device' dont need to change
5680 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5684 * register_netdev - register a network device
5685 * @dev: device to register
5687 * Take a completed network device structure and add it to the kernel
5688 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5689 * chain. 0 is returned on success. A negative errno code is returned
5690 * on a failure to set up the device, or if the name is a duplicate.
5692 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5693 * and expands the device name if you passed a format string to
5696 int register_netdev(struct net_device *dev)
5701 err = register_netdevice(dev);
5705 EXPORT_SYMBOL(register_netdev);
5707 int netdev_refcnt_read(const struct net_device *dev)
5711 for_each_possible_cpu(i)
5712 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5715 EXPORT_SYMBOL(netdev_refcnt_read);
5718 * netdev_wait_allrefs - wait until all references are gone.
5720 * This is called when unregistering network devices.
5722 * Any protocol or device that holds a reference should register
5723 * for netdevice notification, and cleanup and put back the
5724 * reference if they receive an UNREGISTER event.
5725 * We can get stuck here if buggy protocols don't correctly
5728 static void netdev_wait_allrefs(struct net_device *dev)
5730 unsigned long rebroadcast_time, warning_time;
5733 linkwatch_forget_dev(dev);
5735 rebroadcast_time = warning_time = jiffies;
5736 refcnt = netdev_refcnt_read(dev);
5738 while (refcnt != 0) {
5739 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5742 /* Rebroadcast unregister notification */
5743 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5744 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5745 * should have already handle it the first time */
5747 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5749 /* We must not have linkwatch events
5750 * pending on unregister. If this
5751 * happens, we simply run the queue
5752 * unscheduled, resulting in a noop
5755 linkwatch_run_queue();
5760 rebroadcast_time = jiffies;
5765 refcnt = netdev_refcnt_read(dev);
5767 if (time_after(jiffies, warning_time + 10 * HZ)) {
5768 printk(KERN_EMERG "unregister_netdevice: "
5769 "waiting for %s to become free. Usage "
5772 warning_time = jiffies;
5781 * register_netdevice(x1);
5782 * register_netdevice(x2);
5784 * unregister_netdevice(y1);
5785 * unregister_netdevice(y2);
5791 * We are invoked by rtnl_unlock().
5792 * This allows us to deal with problems:
5793 * 1) We can delete sysfs objects which invoke hotplug
5794 * without deadlocking with linkwatch via keventd.
5795 * 2) Since we run with the RTNL semaphore not held, we can sleep
5796 * safely in order to wait for the netdev refcnt to drop to zero.
5798 * We must not return until all unregister events added during
5799 * the interval the lock was held have been completed.
5801 void netdev_run_todo(void)
5803 struct list_head list;
5805 /* Snapshot list, allow later requests */
5806 list_replace_init(&net_todo_list, &list);
5810 /* Wait for rcu callbacks to finish before attempting to drain
5811 * the device list. This usually avoids a 250ms wait.
5813 if (!list_empty(&list))
5816 while (!list_empty(&list)) {
5817 struct net_device *dev
5818 = list_first_entry(&list, struct net_device, todo_list);
5819 list_del(&dev->todo_list);
5821 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5822 printk(KERN_ERR "network todo '%s' but state %d\n",
5823 dev->name, dev->reg_state);
5828 dev->reg_state = NETREG_UNREGISTERED;
5830 on_each_cpu(flush_backlog, dev, 1);
5832 netdev_wait_allrefs(dev);
5835 BUG_ON(netdev_refcnt_read(dev));
5836 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5837 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5838 WARN_ON(dev->dn_ptr);
5840 if (dev->destructor)
5841 dev->destructor(dev);
5843 /* Free network device */
5844 kobject_put(&dev->dev.kobj);
5848 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5849 * fields in the same order, with only the type differing.
5851 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5852 const struct net_device_stats *netdev_stats)
5854 #if BITS_PER_LONG == 64
5855 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5856 memcpy(stats64, netdev_stats, sizeof(*stats64));
5858 size_t i, n = sizeof(*stats64) / sizeof(u64);
5859 const unsigned long *src = (const unsigned long *)netdev_stats;
5860 u64 *dst = (u64 *)stats64;
5862 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5863 sizeof(*stats64) / sizeof(u64));
5864 for (i = 0; i < n; i++)
5870 * dev_get_stats - get network device statistics
5871 * @dev: device to get statistics from
5872 * @storage: place to store stats
5874 * Get network statistics from device. Return @storage.
5875 * The device driver may provide its own method by setting
5876 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5877 * otherwise the internal statistics structure is used.
5879 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5880 struct rtnl_link_stats64 *storage)
5882 const struct net_device_ops *ops = dev->netdev_ops;
5884 if (ops->ndo_get_stats64) {
5885 memset(storage, 0, sizeof(*storage));
5886 ops->ndo_get_stats64(dev, storage);
5887 } else if (ops->ndo_get_stats) {
5888 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5890 netdev_stats_to_stats64(storage, &dev->stats);
5892 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5895 EXPORT_SYMBOL(dev_get_stats);
5897 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5899 struct netdev_queue *queue = dev_ingress_queue(dev);
5901 #ifdef CONFIG_NET_CLS_ACT
5904 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5907 netdev_init_one_queue(dev, queue, NULL);
5908 queue->qdisc = &noop_qdisc;
5909 queue->qdisc_sleeping = &noop_qdisc;
5910 rcu_assign_pointer(dev->ingress_queue, queue);
5916 * alloc_netdev_mqs - allocate network device
5917 * @sizeof_priv: size of private data to allocate space for
5918 * @name: device name format string
5919 * @setup: callback to initialize device
5920 * @txqs: the number of TX subqueues to allocate
5921 * @rxqs: the number of RX subqueues to allocate
5923 * Allocates a struct net_device with private data area for driver use
5924 * and performs basic initialization. Also allocates subquue structs
5925 * for each queue on the device.
5927 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5928 void (*setup)(struct net_device *),
5929 unsigned int txqs, unsigned int rxqs)
5931 struct net_device *dev;
5933 struct net_device *p;
5935 BUG_ON(strlen(name) >= sizeof(dev->name));
5938 pr_err("alloc_netdev: Unable to allocate device "
5939 "with zero queues.\n");
5945 pr_err("alloc_netdev: Unable to allocate device "
5946 "with zero RX queues.\n");
5951 alloc_size = sizeof(struct net_device);
5953 /* ensure 32-byte alignment of private area */
5954 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5955 alloc_size += sizeof_priv;
5957 /* ensure 32-byte alignment of whole construct */
5958 alloc_size += NETDEV_ALIGN - 1;
5960 p = kzalloc(alloc_size, GFP_KERNEL);
5962 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5966 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5967 dev->padded = (char *)dev - (char *)p;
5969 dev->pcpu_refcnt = alloc_percpu(int);
5970 if (!dev->pcpu_refcnt)
5973 if (dev_addr_init(dev))
5979 dev_net_set(dev, &init_net);
5981 dev->gso_max_size = GSO_MAX_SIZE;
5983 INIT_LIST_HEAD(&dev->napi_list);
5984 INIT_LIST_HEAD(&dev->unreg_list);
5985 INIT_LIST_HEAD(&dev->link_watch_list);
5986 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5989 dev->num_tx_queues = txqs;
5990 dev->real_num_tx_queues = txqs;
5991 if (netif_alloc_netdev_queues(dev))
5995 dev->num_rx_queues = rxqs;
5996 dev->real_num_rx_queues = rxqs;
5997 if (netif_alloc_rx_queues(dev))
6001 strcpy(dev->name, name);
6002 dev->group = INIT_NETDEV_GROUP;
6010 free_percpu(dev->pcpu_refcnt);
6020 EXPORT_SYMBOL(alloc_netdev_mqs);
6023 * free_netdev - free network device
6026 * This function does the last stage of destroying an allocated device
6027 * interface. The reference to the device object is released.
6028 * If this is the last reference then it will be freed.
6030 void free_netdev(struct net_device *dev)
6032 struct napi_struct *p, *n;
6034 release_net(dev_net(dev));
6041 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6043 /* Flush device addresses */
6044 dev_addr_flush(dev);
6046 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6049 free_percpu(dev->pcpu_refcnt);
6050 dev->pcpu_refcnt = NULL;
6052 /* Compatibility with error handling in drivers */
6053 if (dev->reg_state == NETREG_UNINITIALIZED) {
6054 kfree((char *)dev - dev->padded);
6058 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6059 dev->reg_state = NETREG_RELEASED;
6061 /* will free via device release */
6062 put_device(&dev->dev);
6064 EXPORT_SYMBOL(free_netdev);
6067 * synchronize_net - Synchronize with packet receive processing
6069 * Wait for packets currently being received to be done.
6070 * Does not block later packets from starting.
6072 void synchronize_net(void)
6075 if (rtnl_is_locked())
6076 synchronize_rcu_expedited();
6080 EXPORT_SYMBOL(synchronize_net);
6083 * unregister_netdevice_queue - remove device from the kernel
6087 * This function shuts down a device interface and removes it
6088 * from the kernel tables.
6089 * If head not NULL, device is queued to be unregistered later.
6091 * Callers must hold the rtnl semaphore. You may want
6092 * unregister_netdev() instead of this.
6095 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6100 list_move_tail(&dev->unreg_list, head);
6102 rollback_registered(dev);
6103 /* Finish processing unregister after unlock */
6107 EXPORT_SYMBOL(unregister_netdevice_queue);
6110 * unregister_netdevice_many - unregister many devices
6111 * @head: list of devices
6113 void unregister_netdevice_many(struct list_head *head)
6115 struct net_device *dev;
6117 if (!list_empty(head)) {
6118 rollback_registered_many(head);
6119 list_for_each_entry(dev, head, unreg_list)
6123 EXPORT_SYMBOL(unregister_netdevice_many);
6126 * unregister_netdev - remove device from the kernel
6129 * This function shuts down a device interface and removes it
6130 * from the kernel tables.
6132 * This is just a wrapper for unregister_netdevice that takes
6133 * the rtnl semaphore. In general you want to use this and not
6134 * unregister_netdevice.
6136 void unregister_netdev(struct net_device *dev)
6139 unregister_netdevice(dev);
6142 EXPORT_SYMBOL(unregister_netdev);
6145 * dev_change_net_namespace - move device to different nethost namespace
6147 * @net: network namespace
6148 * @pat: If not NULL name pattern to try if the current device name
6149 * is already taken in the destination network namespace.
6151 * This function shuts down a device interface and moves it
6152 * to a new network namespace. On success 0 is returned, on
6153 * a failure a netagive errno code is returned.
6155 * Callers must hold the rtnl semaphore.
6158 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6164 /* Don't allow namespace local devices to be moved. */
6166 if (dev->features & NETIF_F_NETNS_LOCAL)
6169 /* Ensure the device has been registrered */
6171 if (dev->reg_state != NETREG_REGISTERED)
6174 /* Get out if there is nothing todo */
6176 if (net_eq(dev_net(dev), net))
6179 /* Pick the destination device name, and ensure
6180 * we can use it in the destination network namespace.
6183 if (__dev_get_by_name(net, dev->name)) {
6184 /* We get here if we can't use the current device name */
6187 if (dev_get_valid_name(dev, pat) < 0)
6192 * And now a mini version of register_netdevice unregister_netdevice.
6195 /* If device is running close it first. */
6198 /* And unlink it from device chain */
6200 unlist_netdevice(dev);
6204 /* Shutdown queueing discipline. */
6207 /* Notify protocols, that we are about to destroy
6208 this device. They should clean all the things.
6210 Note that dev->reg_state stays at NETREG_REGISTERED.
6211 This is wanted because this way 8021q and macvlan know
6212 the device is just moving and can keep their slaves up.
6214 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6215 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6216 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6219 * Flush the unicast and multicast chains
6224 /* Actually switch the network namespace */
6225 dev_net_set(dev, net);
6227 /* If there is an ifindex conflict assign a new one */
6228 if (__dev_get_by_index(net, dev->ifindex)) {
6229 int iflink = (dev->iflink == dev->ifindex);
6230 dev->ifindex = dev_new_index(net);
6232 dev->iflink = dev->ifindex;
6235 /* Fixup kobjects */
6236 err = device_rename(&dev->dev, dev->name);
6239 /* Add the device back in the hashes */
6240 list_netdevice(dev);
6242 /* Notify protocols, that a new device appeared. */
6243 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6246 * Prevent userspace races by waiting until the network
6247 * device is fully setup before sending notifications.
6249 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6256 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6258 static int dev_cpu_callback(struct notifier_block *nfb,
6259 unsigned long action,
6262 struct sk_buff **list_skb;
6263 struct sk_buff *skb;
6264 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6265 struct softnet_data *sd, *oldsd;
6267 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6270 local_irq_disable();
6271 cpu = smp_processor_id();
6272 sd = &per_cpu(softnet_data, cpu);
6273 oldsd = &per_cpu(softnet_data, oldcpu);
6275 /* Find end of our completion_queue. */
6276 list_skb = &sd->completion_queue;
6278 list_skb = &(*list_skb)->next;
6279 /* Append completion queue from offline CPU. */
6280 *list_skb = oldsd->completion_queue;
6281 oldsd->completion_queue = NULL;
6283 /* Append output queue from offline CPU. */
6284 if (oldsd->output_queue) {
6285 *sd->output_queue_tailp = oldsd->output_queue;
6286 sd->output_queue_tailp = oldsd->output_queue_tailp;
6287 oldsd->output_queue = NULL;
6288 oldsd->output_queue_tailp = &oldsd->output_queue;
6290 /* Append NAPI poll list from offline CPU. */
6291 if (!list_empty(&oldsd->poll_list)) {
6292 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6293 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6296 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6299 /* Process offline CPU's input_pkt_queue */
6300 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6302 input_queue_head_incr(oldsd);
6304 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6306 input_queue_head_incr(oldsd);
6314 * netdev_increment_features - increment feature set by one
6315 * @all: current feature set
6316 * @one: new feature set
6317 * @mask: mask feature set
6319 * Computes a new feature set after adding a device with feature set
6320 * @one to the master device with current feature set @all. Will not
6321 * enable anything that is off in @mask. Returns the new feature set.
6323 netdev_features_t netdev_increment_features(netdev_features_t all,
6324 netdev_features_t one, netdev_features_t mask)
6326 if (mask & NETIF_F_GEN_CSUM)
6327 mask |= NETIF_F_ALL_CSUM;
6328 mask |= NETIF_F_VLAN_CHALLENGED;
6330 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6331 all &= one | ~NETIF_F_ALL_FOR_ALL;
6333 /* If one device supports hw checksumming, set for all. */
6334 if (all & NETIF_F_GEN_CSUM)
6335 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6339 EXPORT_SYMBOL(netdev_increment_features);
6341 static struct hlist_head *netdev_create_hash(void)
6344 struct hlist_head *hash;
6346 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6348 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6349 INIT_HLIST_HEAD(&hash[i]);
6354 /* Initialize per network namespace state */
6355 static int __net_init netdev_init(struct net *net)
6357 INIT_LIST_HEAD(&net->dev_base_head);
6359 net->dev_name_head = netdev_create_hash();
6360 if (net->dev_name_head == NULL)
6363 net->dev_index_head = netdev_create_hash();
6364 if (net->dev_index_head == NULL)
6370 kfree(net->dev_name_head);
6376 * netdev_drivername - network driver for the device
6377 * @dev: network device
6379 * Determine network driver for device.
6381 const char *netdev_drivername(const struct net_device *dev)
6383 const struct device_driver *driver;
6384 const struct device *parent;
6385 const char *empty = "";
6387 parent = dev->dev.parent;
6391 driver = parent->driver;
6392 if (driver && driver->name)
6393 return driver->name;
6397 int __netdev_printk(const char *level, const struct net_device *dev,
6398 struct va_format *vaf)
6402 if (dev && dev->dev.parent)
6403 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6404 netdev_name(dev), vaf);
6406 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6408 r = printk("%s(NULL net_device): %pV", level, vaf);
6412 EXPORT_SYMBOL(__netdev_printk);
6414 int netdev_printk(const char *level, const struct net_device *dev,
6415 const char *format, ...)
6417 struct va_format vaf;
6421 va_start(args, format);
6426 r = __netdev_printk(level, dev, &vaf);
6431 EXPORT_SYMBOL(netdev_printk);
6433 #define define_netdev_printk_level(func, level) \
6434 int func(const struct net_device *dev, const char *fmt, ...) \
6437 struct va_format vaf; \
6440 va_start(args, fmt); \
6445 r = __netdev_printk(level, dev, &vaf); \
6450 EXPORT_SYMBOL(func);
6452 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6453 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6454 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6455 define_netdev_printk_level(netdev_err, KERN_ERR);
6456 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6457 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6458 define_netdev_printk_level(netdev_info, KERN_INFO);
6460 static void __net_exit netdev_exit(struct net *net)
6462 kfree(net->dev_name_head);
6463 kfree(net->dev_index_head);
6466 static struct pernet_operations __net_initdata netdev_net_ops = {
6467 .init = netdev_init,
6468 .exit = netdev_exit,
6471 static void __net_exit default_device_exit(struct net *net)
6473 struct net_device *dev, *aux;
6475 * Push all migratable network devices back to the
6476 * initial network namespace
6479 for_each_netdev_safe(net, dev, aux) {
6481 char fb_name[IFNAMSIZ];
6483 /* Ignore unmoveable devices (i.e. loopback) */
6484 if (dev->features & NETIF_F_NETNS_LOCAL)
6487 /* Leave virtual devices for the generic cleanup */
6488 if (dev->rtnl_link_ops)
6491 /* Push remaining network devices to init_net */
6492 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6493 err = dev_change_net_namespace(dev, &init_net, fb_name);
6495 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6496 __func__, dev->name, err);
6503 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6505 /* At exit all network devices most be removed from a network
6506 * namespace. Do this in the reverse order of registration.
6507 * Do this across as many network namespaces as possible to
6508 * improve batching efficiency.
6510 struct net_device *dev;
6512 LIST_HEAD(dev_kill_list);
6515 list_for_each_entry(net, net_list, exit_list) {
6516 for_each_netdev_reverse(net, dev) {
6517 if (dev->rtnl_link_ops)
6518 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6520 unregister_netdevice_queue(dev, &dev_kill_list);
6523 unregister_netdevice_many(&dev_kill_list);
6524 list_del(&dev_kill_list);
6528 static struct pernet_operations __net_initdata default_device_ops = {
6529 .exit = default_device_exit,
6530 .exit_batch = default_device_exit_batch,
6534 * Initialize the DEV module. At boot time this walks the device list and
6535 * unhooks any devices that fail to initialise (normally hardware not
6536 * present) and leaves us with a valid list of present and active devices.
6541 * This is called single threaded during boot, so no need
6542 * to take the rtnl semaphore.
6544 static int __init net_dev_init(void)
6546 int i, rc = -ENOMEM;
6548 BUG_ON(!dev_boot_phase);
6550 if (dev_proc_init())
6553 if (netdev_kobject_init())
6556 INIT_LIST_HEAD(&ptype_all);
6557 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6558 INIT_LIST_HEAD(&ptype_base[i]);
6560 if (register_pernet_subsys(&netdev_net_ops))
6564 * Initialise the packet receive queues.
6567 for_each_possible_cpu(i) {
6568 struct softnet_data *sd = &per_cpu(softnet_data, i);
6570 memset(sd, 0, sizeof(*sd));
6571 skb_queue_head_init(&sd->input_pkt_queue);
6572 skb_queue_head_init(&sd->process_queue);
6573 sd->completion_queue = NULL;
6574 INIT_LIST_HEAD(&sd->poll_list);
6575 sd->output_queue = NULL;
6576 sd->output_queue_tailp = &sd->output_queue;
6578 sd->csd.func = rps_trigger_softirq;
6584 sd->backlog.poll = process_backlog;
6585 sd->backlog.weight = weight_p;
6586 sd->backlog.gro_list = NULL;
6587 sd->backlog.gro_count = 0;
6592 /* The loopback device is special if any other network devices
6593 * is present in a network namespace the loopback device must
6594 * be present. Since we now dynamically allocate and free the
6595 * loopback device ensure this invariant is maintained by
6596 * keeping the loopback device as the first device on the
6597 * list of network devices. Ensuring the loopback devices
6598 * is the first device that appears and the last network device
6601 if (register_pernet_device(&loopback_net_ops))
6604 if (register_pernet_device(&default_device_ops))
6607 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6608 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6610 hotcpu_notifier(dev_cpu_callback, 0);
6618 subsys_initcall(net_dev_init);
6620 static int __init initialize_hashrnd(void)
6622 get_random_bytes(&hashrnd, sizeof(hashrnd));
6626 late_initcall_sync(initialize_hashrnd);