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 */
1891 * Invalidate hardware checksum when packet is to be mangled, and
1892 * complete checksum manually on outgoing path.
1894 int skb_checksum_help(struct sk_buff *skb)
1897 int ret = 0, offset;
1899 if (skb->ip_summed == CHECKSUM_COMPLETE)
1900 goto out_set_summed;
1902 if (unlikely(skb_shinfo(skb)->gso_size)) {
1903 /* Let GSO fix up the checksum. */
1904 goto out_set_summed;
1907 offset = skb_checksum_start_offset(skb);
1908 BUG_ON(offset >= skb_headlen(skb));
1909 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1911 offset += skb->csum_offset;
1912 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1914 if (skb_cloned(skb) &&
1915 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1916 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1921 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1923 skb->ip_summed = CHECKSUM_NONE;
1927 EXPORT_SYMBOL(skb_checksum_help);
1930 * skb_gso_segment - Perform segmentation on skb.
1931 * @skb: buffer to segment
1932 * @features: features for the output path (see dev->features)
1934 * This function segments the given skb and returns a list of segments.
1936 * It may return NULL if the skb requires no segmentation. This is
1937 * only possible when GSO is used for verifying header integrity.
1939 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1940 netdev_features_t features)
1942 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1943 struct packet_type *ptype;
1944 __be16 type = skb->protocol;
1945 int vlan_depth = ETH_HLEN;
1948 while (type == htons(ETH_P_8021Q)) {
1949 struct vlan_hdr *vh;
1951 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1952 return ERR_PTR(-EINVAL);
1954 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1955 type = vh->h_vlan_encapsulated_proto;
1956 vlan_depth += VLAN_HLEN;
1959 skb_reset_mac_header(skb);
1960 skb->mac_len = skb->network_header - skb->mac_header;
1961 __skb_pull(skb, skb->mac_len);
1963 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1964 struct net_device *dev = skb->dev;
1965 struct ethtool_drvinfo info = {};
1967 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1968 dev->ethtool_ops->get_drvinfo(dev, &info);
1970 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d ip_summed=%d\n",
1971 info.driver, dev ? &dev->features : NULL,
1972 skb->sk ? &skb->sk->sk_route_caps : NULL,
1973 skb->len, skb->data_len, skb->ip_summed);
1975 if (skb_header_cloned(skb) &&
1976 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1977 return ERR_PTR(err);
1981 list_for_each_entry_rcu(ptype,
1982 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1983 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1984 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1985 err = ptype->gso_send_check(skb);
1986 segs = ERR_PTR(err);
1987 if (err || skb_gso_ok(skb, features))
1989 __skb_push(skb, (skb->data -
1990 skb_network_header(skb)));
1992 segs = ptype->gso_segment(skb, features);
1998 __skb_push(skb, skb->data - skb_mac_header(skb));
2002 EXPORT_SYMBOL(skb_gso_segment);
2004 /* Take action when hardware reception checksum errors are detected. */
2006 void netdev_rx_csum_fault(struct net_device *dev)
2008 if (net_ratelimit()) {
2009 printk(KERN_ERR "%s: hw csum failure.\n",
2010 dev ? dev->name : "<unknown>");
2014 EXPORT_SYMBOL(netdev_rx_csum_fault);
2017 /* Actually, we should eliminate this check as soon as we know, that:
2018 * 1. IOMMU is present and allows to map all the memory.
2019 * 2. No high memory really exists on this machine.
2022 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2024 #ifdef CONFIG_HIGHMEM
2026 if (!(dev->features & NETIF_F_HIGHDMA)) {
2027 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2028 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2029 if (PageHighMem(skb_frag_page(frag)))
2034 if (PCI_DMA_BUS_IS_PHYS) {
2035 struct device *pdev = dev->dev.parent;
2039 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2040 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2041 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2042 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2051 void (*destructor)(struct sk_buff *skb);
2054 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2056 static void dev_gso_skb_destructor(struct sk_buff *skb)
2058 struct dev_gso_cb *cb;
2061 struct sk_buff *nskb = skb->next;
2063 skb->next = nskb->next;
2066 } while (skb->next);
2068 cb = DEV_GSO_CB(skb);
2070 cb->destructor(skb);
2074 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2075 * @skb: buffer to segment
2076 * @features: device features as applicable to this skb
2078 * This function segments the given skb and stores the list of segments
2081 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2083 struct sk_buff *segs;
2085 segs = skb_gso_segment(skb, features);
2087 /* Verifying header integrity only. */
2092 return PTR_ERR(segs);
2095 DEV_GSO_CB(skb)->destructor = skb->destructor;
2096 skb->destructor = dev_gso_skb_destructor;
2102 * Try to orphan skb early, right before transmission by the device.
2103 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2104 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2106 static inline void skb_orphan_try(struct sk_buff *skb)
2108 struct sock *sk = skb->sk;
2110 if (sk && !skb_shinfo(skb)->tx_flags) {
2111 /* skb_tx_hash() wont be able to get sk.
2112 * We copy sk_hash into skb->rxhash
2115 skb->rxhash = sk->sk_hash;
2120 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2122 return ((features & NETIF_F_GEN_CSUM) ||
2123 ((features & NETIF_F_V4_CSUM) &&
2124 protocol == htons(ETH_P_IP)) ||
2125 ((features & NETIF_F_V6_CSUM) &&
2126 protocol == htons(ETH_P_IPV6)) ||
2127 ((features & NETIF_F_FCOE_CRC) &&
2128 protocol == htons(ETH_P_FCOE)));
2131 static netdev_features_t harmonize_features(struct sk_buff *skb,
2132 __be16 protocol, netdev_features_t features)
2134 if (!can_checksum_protocol(features, protocol)) {
2135 features &= ~NETIF_F_ALL_CSUM;
2136 features &= ~NETIF_F_SG;
2137 } else if (illegal_highdma(skb->dev, skb)) {
2138 features &= ~NETIF_F_SG;
2144 netdev_features_t netif_skb_features(struct sk_buff *skb)
2146 __be16 protocol = skb->protocol;
2147 netdev_features_t features = skb->dev->features;
2149 if (protocol == htons(ETH_P_8021Q)) {
2150 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2151 protocol = veh->h_vlan_encapsulated_proto;
2152 } else if (!vlan_tx_tag_present(skb)) {
2153 return harmonize_features(skb, protocol, features);
2156 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2158 if (protocol != htons(ETH_P_8021Q)) {
2159 return harmonize_features(skb, protocol, features);
2161 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2162 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2163 return harmonize_features(skb, protocol, features);
2166 EXPORT_SYMBOL(netif_skb_features);
2169 * Returns true if either:
2170 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2171 * 2. skb is fragmented and the device does not support SG, or if
2172 * at least one of fragments is in highmem and device does not
2173 * support DMA from it.
2175 static inline int skb_needs_linearize(struct sk_buff *skb,
2178 return skb_is_nonlinear(skb) &&
2179 ((skb_has_frag_list(skb) &&
2180 !(features & NETIF_F_FRAGLIST)) ||
2181 (skb_shinfo(skb)->nr_frags &&
2182 !(features & NETIF_F_SG)));
2185 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2186 struct netdev_queue *txq)
2188 const struct net_device_ops *ops = dev->netdev_ops;
2189 int rc = NETDEV_TX_OK;
2190 unsigned int skb_len;
2192 if (likely(!skb->next)) {
2193 netdev_features_t features;
2196 * If device doesn't need skb->dst, release it right now while
2197 * its hot in this cpu cache
2199 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2202 if (!list_empty(&ptype_all))
2203 dev_queue_xmit_nit(skb, dev);
2205 skb_orphan_try(skb);
2207 features = netif_skb_features(skb);
2209 if (vlan_tx_tag_present(skb) &&
2210 !(features & NETIF_F_HW_VLAN_TX)) {
2211 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2218 if (netif_needs_gso(skb, features)) {
2219 if (unlikely(dev_gso_segment(skb, features)))
2224 if (skb_needs_linearize(skb, features) &&
2225 __skb_linearize(skb))
2228 /* If packet is not checksummed and device does not
2229 * support checksumming for this protocol, complete
2230 * checksumming here.
2232 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2233 skb_set_transport_header(skb,
2234 skb_checksum_start_offset(skb));
2235 if (!(features & NETIF_F_ALL_CSUM) &&
2236 skb_checksum_help(skb))
2242 rc = ops->ndo_start_xmit(skb, dev);
2243 trace_net_dev_xmit(skb, rc, dev, skb_len);
2244 if (rc == NETDEV_TX_OK)
2245 txq_trans_update(txq);
2251 struct sk_buff *nskb = skb->next;
2253 skb->next = nskb->next;
2257 * If device doesn't need nskb->dst, release it right now while
2258 * its hot in this cpu cache
2260 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2263 skb_len = nskb->len;
2264 rc = ops->ndo_start_xmit(nskb, dev);
2265 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2266 if (unlikely(rc != NETDEV_TX_OK)) {
2267 if (rc & ~NETDEV_TX_MASK)
2268 goto out_kfree_gso_skb;
2269 nskb->next = skb->next;
2273 txq_trans_update(txq);
2274 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2275 return NETDEV_TX_BUSY;
2276 } while (skb->next);
2279 if (likely(skb->next == NULL))
2280 skb->destructor = DEV_GSO_CB(skb)->destructor;
2287 static u32 hashrnd __read_mostly;
2290 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2291 * to be used as a distribution range.
2293 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2294 unsigned int num_tx_queues)
2298 u16 qcount = num_tx_queues;
2300 if (skb_rx_queue_recorded(skb)) {
2301 hash = skb_get_rx_queue(skb);
2302 while (unlikely(hash >= num_tx_queues))
2303 hash -= num_tx_queues;
2308 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2309 qoffset = dev->tc_to_txq[tc].offset;
2310 qcount = dev->tc_to_txq[tc].count;
2313 if (skb->sk && skb->sk->sk_hash)
2314 hash = skb->sk->sk_hash;
2316 hash = (__force u16) skb->protocol ^ skb->rxhash;
2317 hash = jhash_1word(hash, hashrnd);
2319 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2321 EXPORT_SYMBOL(__skb_tx_hash);
2323 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2325 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2326 if (net_ratelimit()) {
2327 pr_warning("%s selects TX queue %d, but "
2328 "real number of TX queues is %d\n",
2329 dev->name, queue_index, dev->real_num_tx_queues);
2336 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2339 struct xps_dev_maps *dev_maps;
2340 struct xps_map *map;
2341 int queue_index = -1;
2344 dev_maps = rcu_dereference(dev->xps_maps);
2346 map = rcu_dereference(
2347 dev_maps->cpu_map[raw_smp_processor_id()]);
2350 queue_index = map->queues[0];
2353 if (skb->sk && skb->sk->sk_hash)
2354 hash = skb->sk->sk_hash;
2356 hash = (__force u16) skb->protocol ^
2358 hash = jhash_1word(hash, hashrnd);
2359 queue_index = map->queues[
2360 ((u64)hash * map->len) >> 32];
2362 if (unlikely(queue_index >= dev->real_num_tx_queues))
2374 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2375 struct sk_buff *skb)
2378 const struct net_device_ops *ops = dev->netdev_ops;
2380 if (dev->real_num_tx_queues == 1)
2382 else if (ops->ndo_select_queue) {
2383 queue_index = ops->ndo_select_queue(dev, skb);
2384 queue_index = dev_cap_txqueue(dev, queue_index);
2386 struct sock *sk = skb->sk;
2387 queue_index = sk_tx_queue_get(sk);
2389 if (queue_index < 0 || skb->ooo_okay ||
2390 queue_index >= dev->real_num_tx_queues) {
2391 int old_index = queue_index;
2393 queue_index = get_xps_queue(dev, skb);
2394 if (queue_index < 0)
2395 queue_index = skb_tx_hash(dev, skb);
2397 if (queue_index != old_index && sk) {
2398 struct dst_entry *dst =
2399 rcu_dereference_check(sk->sk_dst_cache, 1);
2401 if (dst && skb_dst(skb) == dst)
2402 sk_tx_queue_set(sk, queue_index);
2407 skb_set_queue_mapping(skb, queue_index);
2408 return netdev_get_tx_queue(dev, queue_index);
2411 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2412 struct net_device *dev,
2413 struct netdev_queue *txq)
2415 spinlock_t *root_lock = qdisc_lock(q);
2419 qdisc_skb_cb(skb)->pkt_len = skb->len;
2420 qdisc_calculate_pkt_len(skb, q);
2422 * Heuristic to force contended enqueues to serialize on a
2423 * separate lock before trying to get qdisc main lock.
2424 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2425 * and dequeue packets faster.
2427 contended = qdisc_is_running(q);
2428 if (unlikely(contended))
2429 spin_lock(&q->busylock);
2431 spin_lock(root_lock);
2432 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2435 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2436 qdisc_run_begin(q)) {
2438 * This is a work-conserving queue; there are no old skbs
2439 * waiting to be sent out; and the qdisc is not running -
2440 * xmit the skb directly.
2442 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2445 qdisc_bstats_update(q, skb);
2447 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2448 if (unlikely(contended)) {
2449 spin_unlock(&q->busylock);
2456 rc = NET_XMIT_SUCCESS;
2459 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2460 if (qdisc_run_begin(q)) {
2461 if (unlikely(contended)) {
2462 spin_unlock(&q->busylock);
2468 spin_unlock(root_lock);
2469 if (unlikely(contended))
2470 spin_unlock(&q->busylock);
2474 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2475 static void skb_update_prio(struct sk_buff *skb)
2477 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2479 if ((!skb->priority) && (skb->sk) && map)
2480 skb->priority = map->priomap[skb->sk->sk_cgrp_prioidx];
2483 #define skb_update_prio(skb)
2486 static DEFINE_PER_CPU(int, xmit_recursion);
2487 #define RECURSION_LIMIT 10
2490 * dev_queue_xmit - transmit a buffer
2491 * @skb: buffer to transmit
2493 * Queue a buffer for transmission to a network device. The caller must
2494 * have set the device and priority and built the buffer before calling
2495 * this function. The function can be called from an interrupt.
2497 * A negative errno code is returned on a failure. A success does not
2498 * guarantee the frame will be transmitted as it may be dropped due
2499 * to congestion or traffic shaping.
2501 * -----------------------------------------------------------------------------------
2502 * I notice this method can also return errors from the queue disciplines,
2503 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2506 * Regardless of the return value, the skb is consumed, so it is currently
2507 * difficult to retry a send to this method. (You can bump the ref count
2508 * before sending to hold a reference for retry if you are careful.)
2510 * When calling this method, interrupts MUST be enabled. This is because
2511 * the BH enable code must have IRQs enabled so that it will not deadlock.
2514 int dev_queue_xmit(struct sk_buff *skb)
2516 struct net_device *dev = skb->dev;
2517 struct netdev_queue *txq;
2521 /* Disable soft irqs for various locks below. Also
2522 * stops preemption for RCU.
2526 skb_update_prio(skb);
2528 txq = dev_pick_tx(dev, skb);
2529 q = rcu_dereference_bh(txq->qdisc);
2531 #ifdef CONFIG_NET_CLS_ACT
2532 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2534 trace_net_dev_queue(skb);
2536 rc = __dev_xmit_skb(skb, q, dev, txq);
2540 /* The device has no queue. Common case for software devices:
2541 loopback, all the sorts of tunnels...
2543 Really, it is unlikely that netif_tx_lock protection is necessary
2544 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2546 However, it is possible, that they rely on protection
2549 Check this and shot the lock. It is not prone from deadlocks.
2550 Either shot noqueue qdisc, it is even simpler 8)
2552 if (dev->flags & IFF_UP) {
2553 int cpu = smp_processor_id(); /* ok because BHs are off */
2555 if (txq->xmit_lock_owner != cpu) {
2557 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2558 goto recursion_alert;
2560 HARD_TX_LOCK(dev, txq, cpu);
2562 if (!netif_xmit_stopped(txq)) {
2563 __this_cpu_inc(xmit_recursion);
2564 rc = dev_hard_start_xmit(skb, dev, txq);
2565 __this_cpu_dec(xmit_recursion);
2566 if (dev_xmit_complete(rc)) {
2567 HARD_TX_UNLOCK(dev, txq);
2571 HARD_TX_UNLOCK(dev, txq);
2572 if (net_ratelimit())
2573 printk(KERN_CRIT "Virtual device %s asks to "
2574 "queue packet!\n", dev->name);
2576 /* Recursion is detected! It is possible,
2580 if (net_ratelimit())
2581 printk(KERN_CRIT "Dead loop on virtual device "
2582 "%s, fix it urgently!\n", dev->name);
2587 rcu_read_unlock_bh();
2592 rcu_read_unlock_bh();
2595 EXPORT_SYMBOL(dev_queue_xmit);
2598 /*=======================================================================
2600 =======================================================================*/
2602 int netdev_max_backlog __read_mostly = 1000;
2603 int netdev_tstamp_prequeue __read_mostly = 1;
2604 int netdev_budget __read_mostly = 300;
2605 int weight_p __read_mostly = 64; /* old backlog weight */
2607 /* Called with irq disabled */
2608 static inline void ____napi_schedule(struct softnet_data *sd,
2609 struct napi_struct *napi)
2611 list_add_tail(&napi->poll_list, &sd->poll_list);
2612 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2616 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2617 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2618 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2619 * if hash is a canonical 4-tuple hash over transport ports.
2621 void __skb_get_rxhash(struct sk_buff *skb)
2623 struct flow_keys keys;
2626 if (!skb_flow_dissect(skb, &keys))
2630 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2631 swap(keys.port16[0], keys.port16[1]);
2635 /* get a consistent hash (same value on both flow directions) */
2636 if ((__force u32)keys.dst < (__force u32)keys.src)
2637 swap(keys.dst, keys.src);
2639 hash = jhash_3words((__force u32)keys.dst,
2640 (__force u32)keys.src,
2641 (__force u32)keys.ports, hashrnd);
2647 EXPORT_SYMBOL(__skb_get_rxhash);
2651 /* One global table that all flow-based protocols share. */
2652 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2653 EXPORT_SYMBOL(rps_sock_flow_table);
2655 struct jump_label_key rps_needed __read_mostly;
2657 static struct rps_dev_flow *
2658 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2659 struct rps_dev_flow *rflow, u16 next_cpu)
2661 if (next_cpu != RPS_NO_CPU) {
2662 #ifdef CONFIG_RFS_ACCEL
2663 struct netdev_rx_queue *rxqueue;
2664 struct rps_dev_flow_table *flow_table;
2665 struct rps_dev_flow *old_rflow;
2670 /* Should we steer this flow to a different hardware queue? */
2671 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2672 !(dev->features & NETIF_F_NTUPLE))
2674 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2675 if (rxq_index == skb_get_rx_queue(skb))
2678 rxqueue = dev->_rx + rxq_index;
2679 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2682 flow_id = skb->rxhash & flow_table->mask;
2683 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2684 rxq_index, flow_id);
2688 rflow = &flow_table->flows[flow_id];
2690 if (old_rflow->filter == rflow->filter)
2691 old_rflow->filter = RPS_NO_FILTER;
2695 per_cpu(softnet_data, next_cpu).input_queue_head;
2698 rflow->cpu = next_cpu;
2703 * get_rps_cpu is called from netif_receive_skb and returns the target
2704 * CPU from the RPS map of the receiving queue for a given skb.
2705 * rcu_read_lock must be held on entry.
2707 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2708 struct rps_dev_flow **rflowp)
2710 struct netdev_rx_queue *rxqueue;
2711 struct rps_map *map;
2712 struct rps_dev_flow_table *flow_table;
2713 struct rps_sock_flow_table *sock_flow_table;
2717 if (skb_rx_queue_recorded(skb)) {
2718 u16 index = skb_get_rx_queue(skb);
2719 if (unlikely(index >= dev->real_num_rx_queues)) {
2720 WARN_ONCE(dev->real_num_rx_queues > 1,
2721 "%s received packet on queue %u, but number "
2722 "of RX queues is %u\n",
2723 dev->name, index, dev->real_num_rx_queues);
2726 rxqueue = dev->_rx + index;
2730 map = rcu_dereference(rxqueue->rps_map);
2732 if (map->len == 1 &&
2733 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2734 tcpu = map->cpus[0];
2735 if (cpu_online(tcpu))
2739 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2743 skb_reset_network_header(skb);
2744 if (!skb_get_rxhash(skb))
2747 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2748 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2749 if (flow_table && sock_flow_table) {
2751 struct rps_dev_flow *rflow;
2753 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2756 next_cpu = sock_flow_table->ents[skb->rxhash &
2757 sock_flow_table->mask];
2760 * If the desired CPU (where last recvmsg was done) is
2761 * different from current CPU (one in the rx-queue flow
2762 * table entry), switch if one of the following holds:
2763 * - Current CPU is unset (equal to RPS_NO_CPU).
2764 * - Current CPU is offline.
2765 * - The current CPU's queue tail has advanced beyond the
2766 * last packet that was enqueued using this table entry.
2767 * This guarantees that all previous packets for the flow
2768 * have been dequeued, thus preserving in order delivery.
2770 if (unlikely(tcpu != next_cpu) &&
2771 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2772 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2773 rflow->last_qtail)) >= 0))
2774 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2776 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2784 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2786 if (cpu_online(tcpu)) {
2796 #ifdef CONFIG_RFS_ACCEL
2799 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2800 * @dev: Device on which the filter was set
2801 * @rxq_index: RX queue index
2802 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2803 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2805 * Drivers that implement ndo_rx_flow_steer() should periodically call
2806 * this function for each installed filter and remove the filters for
2807 * which it returns %true.
2809 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2810 u32 flow_id, u16 filter_id)
2812 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2813 struct rps_dev_flow_table *flow_table;
2814 struct rps_dev_flow *rflow;
2819 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2820 if (flow_table && flow_id <= flow_table->mask) {
2821 rflow = &flow_table->flows[flow_id];
2822 cpu = ACCESS_ONCE(rflow->cpu);
2823 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2824 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2825 rflow->last_qtail) <
2826 (int)(10 * flow_table->mask)))
2832 EXPORT_SYMBOL(rps_may_expire_flow);
2834 #endif /* CONFIG_RFS_ACCEL */
2836 /* Called from hardirq (IPI) context */
2837 static void rps_trigger_softirq(void *data)
2839 struct softnet_data *sd = data;
2841 ____napi_schedule(sd, &sd->backlog);
2845 #endif /* CONFIG_RPS */
2848 * Check if this softnet_data structure is another cpu one
2849 * If yes, queue it to our IPI list and return 1
2852 static int rps_ipi_queued(struct softnet_data *sd)
2855 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2858 sd->rps_ipi_next = mysd->rps_ipi_list;
2859 mysd->rps_ipi_list = sd;
2861 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2864 #endif /* CONFIG_RPS */
2869 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2870 * queue (may be a remote CPU queue).
2872 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2873 unsigned int *qtail)
2875 struct softnet_data *sd;
2876 unsigned long flags;
2878 sd = &per_cpu(softnet_data, cpu);
2880 local_irq_save(flags);
2883 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2884 if (skb_queue_len(&sd->input_pkt_queue)) {
2886 __skb_queue_tail(&sd->input_pkt_queue, skb);
2887 input_queue_tail_incr_save(sd, qtail);
2889 local_irq_restore(flags);
2890 return NET_RX_SUCCESS;
2893 /* Schedule NAPI for backlog device
2894 * We can use non atomic operation since we own the queue lock
2896 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2897 if (!rps_ipi_queued(sd))
2898 ____napi_schedule(sd, &sd->backlog);
2906 local_irq_restore(flags);
2908 atomic_long_inc(&skb->dev->rx_dropped);
2914 * netif_rx - post buffer to the network code
2915 * @skb: buffer to post
2917 * This function receives a packet from a device driver and queues it for
2918 * the upper (protocol) levels to process. It always succeeds. The buffer
2919 * may be dropped during processing for congestion control or by the
2923 * NET_RX_SUCCESS (no congestion)
2924 * NET_RX_DROP (packet was dropped)
2928 int netif_rx(struct sk_buff *skb)
2932 /* if netpoll wants it, pretend we never saw it */
2933 if (netpoll_rx(skb))
2936 net_timestamp_check(netdev_tstamp_prequeue, skb);
2938 trace_netif_rx(skb);
2940 if (static_branch(&rps_needed)) {
2941 struct rps_dev_flow voidflow, *rflow = &voidflow;
2947 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2949 cpu = smp_processor_id();
2951 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2959 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2964 EXPORT_SYMBOL(netif_rx);
2966 int netif_rx_ni(struct sk_buff *skb)
2971 err = netif_rx(skb);
2972 if (local_softirq_pending())
2978 EXPORT_SYMBOL(netif_rx_ni);
2980 static void net_tx_action(struct softirq_action *h)
2982 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2984 if (sd->completion_queue) {
2985 struct sk_buff *clist;
2987 local_irq_disable();
2988 clist = sd->completion_queue;
2989 sd->completion_queue = NULL;
2993 struct sk_buff *skb = clist;
2994 clist = clist->next;
2996 WARN_ON(atomic_read(&skb->users));
2997 trace_kfree_skb(skb, net_tx_action);
3002 if (sd->output_queue) {
3005 local_irq_disable();
3006 head = sd->output_queue;
3007 sd->output_queue = NULL;
3008 sd->output_queue_tailp = &sd->output_queue;
3012 struct Qdisc *q = head;
3013 spinlock_t *root_lock;
3015 head = head->next_sched;
3017 root_lock = qdisc_lock(q);
3018 if (spin_trylock(root_lock)) {
3019 smp_mb__before_clear_bit();
3020 clear_bit(__QDISC_STATE_SCHED,
3023 spin_unlock(root_lock);
3025 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3027 __netif_reschedule(q);
3029 smp_mb__before_clear_bit();
3030 clear_bit(__QDISC_STATE_SCHED,
3038 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3039 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3040 /* This hook is defined here for ATM LANE */
3041 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3042 unsigned char *addr) __read_mostly;
3043 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3046 #ifdef CONFIG_NET_CLS_ACT
3047 /* TODO: Maybe we should just force sch_ingress to be compiled in
3048 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3049 * a compare and 2 stores extra right now if we dont have it on
3050 * but have CONFIG_NET_CLS_ACT
3051 * NOTE: This doesn't stop any functionality; if you dont have
3052 * the ingress scheduler, you just can't add policies on ingress.
3055 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3057 struct net_device *dev = skb->dev;
3058 u32 ttl = G_TC_RTTL(skb->tc_verd);
3059 int result = TC_ACT_OK;
3062 if (unlikely(MAX_RED_LOOP < ttl++)) {
3063 if (net_ratelimit())
3064 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3065 skb->skb_iif, dev->ifindex);
3069 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3070 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3073 if (q != &noop_qdisc) {
3074 spin_lock(qdisc_lock(q));
3075 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3076 result = qdisc_enqueue_root(skb, q);
3077 spin_unlock(qdisc_lock(q));
3083 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3084 struct packet_type **pt_prev,
3085 int *ret, struct net_device *orig_dev)
3087 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3089 if (!rxq || rxq->qdisc == &noop_qdisc)
3093 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3097 switch (ing_filter(skb, rxq)) {
3111 * netdev_rx_handler_register - register receive handler
3112 * @dev: device to register a handler for
3113 * @rx_handler: receive handler to register
3114 * @rx_handler_data: data pointer that is used by rx handler
3116 * Register a receive hander for a device. This handler will then be
3117 * called from __netif_receive_skb. A negative errno code is returned
3120 * The caller must hold the rtnl_mutex.
3122 * For a general description of rx_handler, see enum rx_handler_result.
3124 int netdev_rx_handler_register(struct net_device *dev,
3125 rx_handler_func_t *rx_handler,
3126 void *rx_handler_data)
3130 if (dev->rx_handler)
3133 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3134 rcu_assign_pointer(dev->rx_handler, rx_handler);
3138 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3141 * netdev_rx_handler_unregister - unregister receive handler
3142 * @dev: device to unregister a handler from
3144 * Unregister a receive hander from a device.
3146 * The caller must hold the rtnl_mutex.
3148 void netdev_rx_handler_unregister(struct net_device *dev)
3152 RCU_INIT_POINTER(dev->rx_handler, NULL);
3153 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3155 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3157 static int __netif_receive_skb(struct sk_buff *skb)
3159 struct packet_type *ptype, *pt_prev;
3160 rx_handler_func_t *rx_handler;
3161 struct net_device *orig_dev;
3162 struct net_device *null_or_dev;
3163 bool deliver_exact = false;
3164 int ret = NET_RX_DROP;
3167 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3169 trace_netif_receive_skb(skb);
3171 /* if we've gotten here through NAPI, check netpoll */
3172 if (netpoll_receive_skb(skb))
3176 skb->skb_iif = skb->dev->ifindex;
3177 orig_dev = skb->dev;
3179 skb_reset_network_header(skb);
3180 skb_reset_transport_header(skb);
3181 skb_reset_mac_len(skb);
3189 __this_cpu_inc(softnet_data.processed);
3191 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3192 skb = vlan_untag(skb);
3197 #ifdef CONFIG_NET_CLS_ACT
3198 if (skb->tc_verd & TC_NCLS) {
3199 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3204 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3205 if (!ptype->dev || ptype->dev == skb->dev) {
3207 ret = deliver_skb(skb, pt_prev, orig_dev);
3212 #ifdef CONFIG_NET_CLS_ACT
3213 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3219 rx_handler = rcu_dereference(skb->dev->rx_handler);
3220 if (vlan_tx_tag_present(skb)) {
3222 ret = deliver_skb(skb, pt_prev, orig_dev);
3225 if (vlan_do_receive(&skb, !rx_handler))
3227 else if (unlikely(!skb))
3233 ret = deliver_skb(skb, pt_prev, orig_dev);
3236 switch (rx_handler(&skb)) {
3237 case RX_HANDLER_CONSUMED:
3239 case RX_HANDLER_ANOTHER:
3241 case RX_HANDLER_EXACT:
3242 deliver_exact = true;
3243 case RX_HANDLER_PASS:
3250 /* deliver only exact match when indicated */
3251 null_or_dev = deliver_exact ? skb->dev : NULL;
3253 type = skb->protocol;
3254 list_for_each_entry_rcu(ptype,
3255 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3256 if (ptype->type == type &&
3257 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3258 ptype->dev == orig_dev)) {
3260 ret = deliver_skb(skb, pt_prev, orig_dev);
3266 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3268 atomic_long_inc(&skb->dev->rx_dropped);
3270 /* Jamal, now you will not able to escape explaining
3271 * me how you were going to use this. :-)
3282 * netif_receive_skb - process receive buffer from network
3283 * @skb: buffer to process
3285 * netif_receive_skb() is the main receive data processing function.
3286 * It always succeeds. The buffer may be dropped during processing
3287 * for congestion control or by the protocol layers.
3289 * This function may only be called from softirq context and interrupts
3290 * should be enabled.
3292 * Return values (usually ignored):
3293 * NET_RX_SUCCESS: no congestion
3294 * NET_RX_DROP: packet was dropped
3296 int netif_receive_skb(struct sk_buff *skb)
3298 net_timestamp_check(netdev_tstamp_prequeue, skb);
3300 if (skb_defer_rx_timestamp(skb))
3301 return NET_RX_SUCCESS;
3304 if (static_branch(&rps_needed)) {
3305 struct rps_dev_flow voidflow, *rflow = &voidflow;
3310 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3313 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3320 return __netif_receive_skb(skb);
3322 EXPORT_SYMBOL(netif_receive_skb);
3324 /* Network device is going away, flush any packets still pending
3325 * Called with irqs disabled.
3327 static void flush_backlog(void *arg)
3329 struct net_device *dev = arg;
3330 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3331 struct sk_buff *skb, *tmp;
3334 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3335 if (skb->dev == dev) {
3336 __skb_unlink(skb, &sd->input_pkt_queue);
3338 input_queue_head_incr(sd);
3343 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3344 if (skb->dev == dev) {
3345 __skb_unlink(skb, &sd->process_queue);
3347 input_queue_head_incr(sd);
3352 static int napi_gro_complete(struct sk_buff *skb)
3354 struct packet_type *ptype;
3355 __be16 type = skb->protocol;
3356 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3359 if (NAPI_GRO_CB(skb)->count == 1) {
3360 skb_shinfo(skb)->gso_size = 0;
3365 list_for_each_entry_rcu(ptype, head, list) {
3366 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3369 err = ptype->gro_complete(skb);
3375 WARN_ON(&ptype->list == head);
3377 return NET_RX_SUCCESS;
3381 return netif_receive_skb(skb);
3384 inline void napi_gro_flush(struct napi_struct *napi)
3386 struct sk_buff *skb, *next;
3388 for (skb = napi->gro_list; skb; skb = next) {
3391 napi_gro_complete(skb);
3394 napi->gro_count = 0;
3395 napi->gro_list = NULL;
3397 EXPORT_SYMBOL(napi_gro_flush);
3399 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3401 struct sk_buff **pp = NULL;
3402 struct packet_type *ptype;
3403 __be16 type = skb->protocol;
3404 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3407 enum gro_result ret;
3409 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3412 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3416 list_for_each_entry_rcu(ptype, head, list) {
3417 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3420 skb_set_network_header(skb, skb_gro_offset(skb));
3421 mac_len = skb->network_header - skb->mac_header;
3422 skb->mac_len = mac_len;
3423 NAPI_GRO_CB(skb)->same_flow = 0;
3424 NAPI_GRO_CB(skb)->flush = 0;
3425 NAPI_GRO_CB(skb)->free = 0;
3427 pp = ptype->gro_receive(&napi->gro_list, skb);
3432 if (&ptype->list == head)
3435 same_flow = NAPI_GRO_CB(skb)->same_flow;
3436 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3439 struct sk_buff *nskb = *pp;
3443 napi_gro_complete(nskb);
3450 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3454 NAPI_GRO_CB(skb)->count = 1;
3455 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3456 skb->next = napi->gro_list;
3457 napi->gro_list = skb;
3461 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3462 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3464 BUG_ON(skb->end - skb->tail < grow);
3466 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3469 skb->data_len -= grow;
3471 skb_shinfo(skb)->frags[0].page_offset += grow;
3472 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3474 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3475 skb_frag_unref(skb, 0);
3476 memmove(skb_shinfo(skb)->frags,
3477 skb_shinfo(skb)->frags + 1,
3478 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3489 EXPORT_SYMBOL(dev_gro_receive);
3491 static inline gro_result_t
3492 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3496 for (p = napi->gro_list; p; p = p->next) {
3497 unsigned long diffs;
3499 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3500 diffs |= p->vlan_tci ^ skb->vlan_tci;
3501 diffs |= compare_ether_header(skb_mac_header(p),
3502 skb_gro_mac_header(skb));
3503 NAPI_GRO_CB(p)->same_flow = !diffs;
3504 NAPI_GRO_CB(p)->flush = 0;
3507 return dev_gro_receive(napi, skb);
3510 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3514 if (netif_receive_skb(skb))
3519 case GRO_MERGED_FREE:
3530 EXPORT_SYMBOL(napi_skb_finish);
3532 void skb_gro_reset_offset(struct sk_buff *skb)
3534 NAPI_GRO_CB(skb)->data_offset = 0;
3535 NAPI_GRO_CB(skb)->frag0 = NULL;
3536 NAPI_GRO_CB(skb)->frag0_len = 0;
3538 if (skb->mac_header == skb->tail &&
3539 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3540 NAPI_GRO_CB(skb)->frag0 =
3541 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3542 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3545 EXPORT_SYMBOL(skb_gro_reset_offset);
3547 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3549 skb_gro_reset_offset(skb);
3551 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3553 EXPORT_SYMBOL(napi_gro_receive);
3555 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3557 __skb_pull(skb, skb_headlen(skb));
3558 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3560 skb->dev = napi->dev;
3566 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3568 struct sk_buff *skb = napi->skb;
3571 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3577 EXPORT_SYMBOL(napi_get_frags);
3579 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3585 skb->protocol = eth_type_trans(skb, skb->dev);
3587 if (ret == GRO_HELD)
3588 skb_gro_pull(skb, -ETH_HLEN);
3589 else if (netif_receive_skb(skb))
3594 case GRO_MERGED_FREE:
3595 napi_reuse_skb(napi, skb);
3604 EXPORT_SYMBOL(napi_frags_finish);
3606 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3608 struct sk_buff *skb = napi->skb;
3615 skb_reset_mac_header(skb);
3616 skb_gro_reset_offset(skb);
3618 off = skb_gro_offset(skb);
3619 hlen = off + sizeof(*eth);
3620 eth = skb_gro_header_fast(skb, off);
3621 if (skb_gro_header_hard(skb, hlen)) {
3622 eth = skb_gro_header_slow(skb, hlen, off);
3623 if (unlikely(!eth)) {
3624 napi_reuse_skb(napi, skb);
3630 skb_gro_pull(skb, sizeof(*eth));
3633 * This works because the only protocols we care about don't require
3634 * special handling. We'll fix it up properly at the end.
3636 skb->protocol = eth->h_proto;
3641 EXPORT_SYMBOL(napi_frags_skb);
3643 gro_result_t napi_gro_frags(struct napi_struct *napi)
3645 struct sk_buff *skb = napi_frags_skb(napi);
3650 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3652 EXPORT_SYMBOL(napi_gro_frags);
3655 * net_rps_action sends any pending IPI's for rps.
3656 * Note: called with local irq disabled, but exits with local irq enabled.
3658 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3661 struct softnet_data *remsd = sd->rps_ipi_list;
3664 sd->rps_ipi_list = NULL;
3668 /* Send pending IPI's to kick RPS processing on remote cpus. */
3670 struct softnet_data *next = remsd->rps_ipi_next;
3672 if (cpu_online(remsd->cpu))
3673 __smp_call_function_single(remsd->cpu,
3682 static int process_backlog(struct napi_struct *napi, int quota)
3685 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3688 /* Check if we have pending ipi, its better to send them now,
3689 * not waiting net_rx_action() end.
3691 if (sd->rps_ipi_list) {
3692 local_irq_disable();
3693 net_rps_action_and_irq_enable(sd);
3696 napi->weight = weight_p;
3697 local_irq_disable();
3698 while (work < quota) {
3699 struct sk_buff *skb;
3702 while ((skb = __skb_dequeue(&sd->process_queue))) {
3704 __netif_receive_skb(skb);
3705 local_irq_disable();
3706 input_queue_head_incr(sd);
3707 if (++work >= quota) {
3714 qlen = skb_queue_len(&sd->input_pkt_queue);
3716 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3717 &sd->process_queue);
3719 if (qlen < quota - work) {
3721 * Inline a custom version of __napi_complete().
3722 * only current cpu owns and manipulates this napi,
3723 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3724 * we can use a plain write instead of clear_bit(),
3725 * and we dont need an smp_mb() memory barrier.
3727 list_del(&napi->poll_list);
3730 quota = work + qlen;
3740 * __napi_schedule - schedule for receive
3741 * @n: entry to schedule
3743 * The entry's receive function will be scheduled to run
3745 void __napi_schedule(struct napi_struct *n)
3747 unsigned long flags;
3749 local_irq_save(flags);
3750 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3751 local_irq_restore(flags);
3753 EXPORT_SYMBOL(__napi_schedule);
3755 void __napi_complete(struct napi_struct *n)
3757 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3758 BUG_ON(n->gro_list);
3760 list_del(&n->poll_list);
3761 smp_mb__before_clear_bit();
3762 clear_bit(NAPI_STATE_SCHED, &n->state);
3764 EXPORT_SYMBOL(__napi_complete);
3766 void napi_complete(struct napi_struct *n)
3768 unsigned long flags;
3771 * don't let napi dequeue from the cpu poll list
3772 * just in case its running on a different cpu
3774 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3778 local_irq_save(flags);
3780 local_irq_restore(flags);
3782 EXPORT_SYMBOL(napi_complete);
3784 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3785 int (*poll)(struct napi_struct *, int), int weight)
3787 INIT_LIST_HEAD(&napi->poll_list);
3788 napi->gro_count = 0;
3789 napi->gro_list = NULL;
3792 napi->weight = weight;
3793 list_add(&napi->dev_list, &dev->napi_list);
3795 #ifdef CONFIG_NETPOLL
3796 spin_lock_init(&napi->poll_lock);
3797 napi->poll_owner = -1;
3799 set_bit(NAPI_STATE_SCHED, &napi->state);
3801 EXPORT_SYMBOL(netif_napi_add);
3803 void netif_napi_del(struct napi_struct *napi)
3805 struct sk_buff *skb, *next;
3807 list_del_init(&napi->dev_list);
3808 napi_free_frags(napi);
3810 for (skb = napi->gro_list; skb; skb = next) {
3816 napi->gro_list = NULL;
3817 napi->gro_count = 0;
3819 EXPORT_SYMBOL(netif_napi_del);
3821 static void net_rx_action(struct softirq_action *h)
3823 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3824 unsigned long time_limit = jiffies + 2;
3825 int budget = netdev_budget;
3828 local_irq_disable();
3830 while (!list_empty(&sd->poll_list)) {
3831 struct napi_struct *n;
3834 /* If softirq window is exhuasted then punt.
3835 * Allow this to run for 2 jiffies since which will allow
3836 * an average latency of 1.5/HZ.
3838 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3843 /* Even though interrupts have been re-enabled, this
3844 * access is safe because interrupts can only add new
3845 * entries to the tail of this list, and only ->poll()
3846 * calls can remove this head entry from the list.
3848 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3850 have = netpoll_poll_lock(n);
3854 /* This NAPI_STATE_SCHED test is for avoiding a race
3855 * with netpoll's poll_napi(). Only the entity which
3856 * obtains the lock and sees NAPI_STATE_SCHED set will
3857 * actually make the ->poll() call. Therefore we avoid
3858 * accidentally calling ->poll() when NAPI is not scheduled.
3861 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3862 work = n->poll(n, weight);
3866 WARN_ON_ONCE(work > weight);
3870 local_irq_disable();
3872 /* Drivers must not modify the NAPI state if they
3873 * consume the entire weight. In such cases this code
3874 * still "owns" the NAPI instance and therefore can
3875 * move the instance around on the list at-will.
3877 if (unlikely(work == weight)) {
3878 if (unlikely(napi_disable_pending(n))) {
3881 local_irq_disable();
3883 list_move_tail(&n->poll_list, &sd->poll_list);
3886 netpoll_poll_unlock(have);
3889 net_rps_action_and_irq_enable(sd);
3891 #ifdef CONFIG_NET_DMA
3893 * There may not be any more sk_buffs coming right now, so push
3894 * any pending DMA copies to hardware
3896 dma_issue_pending_all();
3903 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3907 static gifconf_func_t *gifconf_list[NPROTO];
3910 * register_gifconf - register a SIOCGIF handler
3911 * @family: Address family
3912 * @gifconf: Function handler
3914 * Register protocol dependent address dumping routines. The handler
3915 * that is passed must not be freed or reused until it has been replaced
3916 * by another handler.
3918 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3920 if (family >= NPROTO)
3922 gifconf_list[family] = gifconf;
3925 EXPORT_SYMBOL(register_gifconf);
3929 * Map an interface index to its name (SIOCGIFNAME)
3933 * We need this ioctl for efficient implementation of the
3934 * if_indextoname() function required by the IPv6 API. Without
3935 * it, we would have to search all the interfaces to find a
3939 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3941 struct net_device *dev;
3945 * Fetch the caller's info block.
3948 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3952 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3958 strcpy(ifr.ifr_name, dev->name);
3961 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3967 * Perform a SIOCGIFCONF call. This structure will change
3968 * size eventually, and there is nothing I can do about it.
3969 * Thus we will need a 'compatibility mode'.
3972 static int dev_ifconf(struct net *net, char __user *arg)
3975 struct net_device *dev;
3982 * Fetch the caller's info block.
3985 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3992 * Loop over the interfaces, and write an info block for each.
3996 for_each_netdev(net, dev) {
3997 for (i = 0; i < NPROTO; i++) {
3998 if (gifconf_list[i]) {
4001 done = gifconf_list[i](dev, NULL, 0);
4003 done = gifconf_list[i](dev, pos + total,
4013 * All done. Write the updated control block back to the caller.
4015 ifc.ifc_len = total;
4018 * Both BSD and Solaris return 0 here, so we do too.
4020 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4023 #ifdef CONFIG_PROC_FS
4025 #define BUCKET_SPACE (32 - NETDEV_HASHBITS)
4027 struct dev_iter_state {
4028 struct seq_net_private p;
4029 unsigned int pos; /* bucket << BUCKET_SPACE + offset */
4032 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4033 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4034 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4036 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq)
4038 struct dev_iter_state *state = seq->private;
4039 struct net *net = seq_file_net(seq);
4040 struct net_device *dev;
4041 struct hlist_node *p;
4042 struct hlist_head *h;
4043 unsigned int count, bucket, offset;
4045 bucket = get_bucket(state->pos);
4046 offset = get_offset(state->pos);
4047 h = &net->dev_name_head[bucket];
4049 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4050 if (count++ == offset) {
4051 state->pos = set_bucket_offset(bucket, count);
4059 static inline struct net_device *dev_from_new_bucket(struct seq_file *seq)
4061 struct dev_iter_state *state = seq->private;
4062 struct net_device *dev;
4063 unsigned int bucket;
4065 bucket = get_bucket(state->pos);
4067 dev = dev_from_same_bucket(seq);
4072 state->pos = set_bucket_offset(bucket, 0);
4073 } while (bucket < NETDEV_HASHENTRIES);
4079 * This is invoked by the /proc filesystem handler to display a device
4082 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4085 struct dev_iter_state *state = seq->private;
4089 return SEQ_START_TOKEN;
4091 /* check for end of the hash */
4092 if (state->pos == 0 && *pos > 1)
4095 return dev_from_new_bucket(seq);
4098 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4100 struct net_device *dev;
4104 if (v == SEQ_START_TOKEN)
4105 return dev_from_new_bucket(seq);
4107 dev = dev_from_same_bucket(seq);
4111 return dev_from_new_bucket(seq);
4114 void dev_seq_stop(struct seq_file *seq, void *v)
4120 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4122 struct rtnl_link_stats64 temp;
4123 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4125 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4126 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4127 dev->name, stats->rx_bytes, stats->rx_packets,
4129 stats->rx_dropped + stats->rx_missed_errors,
4130 stats->rx_fifo_errors,
4131 stats->rx_length_errors + stats->rx_over_errors +
4132 stats->rx_crc_errors + stats->rx_frame_errors,
4133 stats->rx_compressed, stats->multicast,
4134 stats->tx_bytes, stats->tx_packets,
4135 stats->tx_errors, stats->tx_dropped,
4136 stats->tx_fifo_errors, stats->collisions,
4137 stats->tx_carrier_errors +
4138 stats->tx_aborted_errors +
4139 stats->tx_window_errors +
4140 stats->tx_heartbeat_errors,
4141 stats->tx_compressed);
4145 * Called from the PROCfs module. This now uses the new arbitrary sized
4146 * /proc/net interface to create /proc/net/dev
4148 static int dev_seq_show(struct seq_file *seq, void *v)
4150 if (v == SEQ_START_TOKEN)
4151 seq_puts(seq, "Inter-| Receive "
4153 " face |bytes packets errs drop fifo frame "
4154 "compressed multicast|bytes packets errs "
4155 "drop fifo colls carrier compressed\n");
4157 dev_seq_printf_stats(seq, v);
4161 static struct softnet_data *softnet_get_online(loff_t *pos)
4163 struct softnet_data *sd = NULL;
4165 while (*pos < nr_cpu_ids)
4166 if (cpu_online(*pos)) {
4167 sd = &per_cpu(softnet_data, *pos);
4174 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4176 return softnet_get_online(pos);
4179 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4182 return softnet_get_online(pos);
4185 static void softnet_seq_stop(struct seq_file *seq, void *v)
4189 static int softnet_seq_show(struct seq_file *seq, void *v)
4191 struct softnet_data *sd = v;
4193 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4194 sd->processed, sd->dropped, sd->time_squeeze, 0,
4195 0, 0, 0, 0, /* was fastroute */
4196 sd->cpu_collision, sd->received_rps);
4200 static const struct seq_operations dev_seq_ops = {
4201 .start = dev_seq_start,
4202 .next = dev_seq_next,
4203 .stop = dev_seq_stop,
4204 .show = dev_seq_show,
4207 static int dev_seq_open(struct inode *inode, struct file *file)
4209 return seq_open_net(inode, file, &dev_seq_ops,
4210 sizeof(struct dev_iter_state));
4213 int dev_seq_open_ops(struct inode *inode, struct file *file,
4214 const struct seq_operations *ops)
4216 return seq_open_net(inode, file, ops, sizeof(struct dev_iter_state));
4219 static const struct file_operations dev_seq_fops = {
4220 .owner = THIS_MODULE,
4221 .open = dev_seq_open,
4223 .llseek = seq_lseek,
4224 .release = seq_release_net,
4227 static const struct seq_operations softnet_seq_ops = {
4228 .start = softnet_seq_start,
4229 .next = softnet_seq_next,
4230 .stop = softnet_seq_stop,
4231 .show = softnet_seq_show,
4234 static int softnet_seq_open(struct inode *inode, struct file *file)
4236 return seq_open(file, &softnet_seq_ops);
4239 static const struct file_operations softnet_seq_fops = {
4240 .owner = THIS_MODULE,
4241 .open = softnet_seq_open,
4243 .llseek = seq_lseek,
4244 .release = seq_release,
4247 static void *ptype_get_idx(loff_t pos)
4249 struct packet_type *pt = NULL;
4253 list_for_each_entry_rcu(pt, &ptype_all, list) {
4259 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4260 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4269 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4273 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4276 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4278 struct packet_type *pt;
4279 struct list_head *nxt;
4283 if (v == SEQ_START_TOKEN)
4284 return ptype_get_idx(0);
4287 nxt = pt->list.next;
4288 if (pt->type == htons(ETH_P_ALL)) {
4289 if (nxt != &ptype_all)
4292 nxt = ptype_base[0].next;
4294 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4296 while (nxt == &ptype_base[hash]) {
4297 if (++hash >= PTYPE_HASH_SIZE)
4299 nxt = ptype_base[hash].next;
4302 return list_entry(nxt, struct packet_type, list);
4305 static void ptype_seq_stop(struct seq_file *seq, void *v)
4311 static int ptype_seq_show(struct seq_file *seq, void *v)
4313 struct packet_type *pt = v;
4315 if (v == SEQ_START_TOKEN)
4316 seq_puts(seq, "Type Device Function\n");
4317 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4318 if (pt->type == htons(ETH_P_ALL))
4319 seq_puts(seq, "ALL ");
4321 seq_printf(seq, "%04x", ntohs(pt->type));
4323 seq_printf(seq, " %-8s %pF\n",
4324 pt->dev ? pt->dev->name : "", pt->func);
4330 static const struct seq_operations ptype_seq_ops = {
4331 .start = ptype_seq_start,
4332 .next = ptype_seq_next,
4333 .stop = ptype_seq_stop,
4334 .show = ptype_seq_show,
4337 static int ptype_seq_open(struct inode *inode, struct file *file)
4339 return seq_open_net(inode, file, &ptype_seq_ops,
4340 sizeof(struct seq_net_private));
4343 static const struct file_operations ptype_seq_fops = {
4344 .owner = THIS_MODULE,
4345 .open = ptype_seq_open,
4347 .llseek = seq_lseek,
4348 .release = seq_release_net,
4352 static int __net_init dev_proc_net_init(struct net *net)
4356 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4358 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4360 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4363 if (wext_proc_init(net))
4369 proc_net_remove(net, "ptype");
4371 proc_net_remove(net, "softnet_stat");
4373 proc_net_remove(net, "dev");
4377 static void __net_exit dev_proc_net_exit(struct net *net)
4379 wext_proc_exit(net);
4381 proc_net_remove(net, "ptype");
4382 proc_net_remove(net, "softnet_stat");
4383 proc_net_remove(net, "dev");
4386 static struct pernet_operations __net_initdata dev_proc_ops = {
4387 .init = dev_proc_net_init,
4388 .exit = dev_proc_net_exit,
4391 static int __init dev_proc_init(void)
4393 return register_pernet_subsys(&dev_proc_ops);
4396 #define dev_proc_init() 0
4397 #endif /* CONFIG_PROC_FS */
4401 * netdev_set_master - set up master pointer
4402 * @slave: slave device
4403 * @master: new master device
4405 * Changes the master device of the slave. Pass %NULL to break the
4406 * bonding. The caller must hold the RTNL semaphore. On a failure
4407 * a negative errno code is returned. On success the reference counts
4408 * are adjusted and the function returns zero.
4410 int netdev_set_master(struct net_device *slave, struct net_device *master)
4412 struct net_device *old = slave->master;
4422 slave->master = master;
4428 EXPORT_SYMBOL(netdev_set_master);
4431 * netdev_set_bond_master - set up bonding master/slave pair
4432 * @slave: slave device
4433 * @master: new master device
4435 * Changes the master device of the slave. Pass %NULL to break the
4436 * bonding. The caller must hold the RTNL semaphore. On a failure
4437 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4438 * to the routing socket and the function returns zero.
4440 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4446 err = netdev_set_master(slave, master);
4450 slave->flags |= IFF_SLAVE;
4452 slave->flags &= ~IFF_SLAVE;
4454 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4457 EXPORT_SYMBOL(netdev_set_bond_master);
4459 static void dev_change_rx_flags(struct net_device *dev, int flags)
4461 const struct net_device_ops *ops = dev->netdev_ops;
4463 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4464 ops->ndo_change_rx_flags(dev, flags);
4467 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4469 unsigned int old_flags = dev->flags;
4475 dev->flags |= IFF_PROMISC;
4476 dev->promiscuity += inc;
4477 if (dev->promiscuity == 0) {
4480 * If inc causes overflow, untouch promisc and return error.
4483 dev->flags &= ~IFF_PROMISC;
4485 dev->promiscuity -= inc;
4486 printk(KERN_WARNING "%s: promiscuity touches roof, "
4487 "set promiscuity failed, promiscuity feature "
4488 "of device might be broken.\n", dev->name);
4492 if (dev->flags != old_flags) {
4493 printk(KERN_INFO "device %s %s promiscuous mode\n",
4494 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4496 if (audit_enabled) {
4497 current_uid_gid(&uid, &gid);
4498 audit_log(current->audit_context, GFP_ATOMIC,
4499 AUDIT_ANOM_PROMISCUOUS,
4500 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4501 dev->name, (dev->flags & IFF_PROMISC),
4502 (old_flags & IFF_PROMISC),
4503 audit_get_loginuid(current),
4505 audit_get_sessionid(current));
4508 dev_change_rx_flags(dev, IFF_PROMISC);
4514 * dev_set_promiscuity - update promiscuity count on a device
4518 * Add or remove promiscuity from a device. While the count in the device
4519 * remains above zero the interface remains promiscuous. Once it hits zero
4520 * the device reverts back to normal filtering operation. A negative inc
4521 * value is used to drop promiscuity on the device.
4522 * Return 0 if successful or a negative errno code on error.
4524 int dev_set_promiscuity(struct net_device *dev, int inc)
4526 unsigned int old_flags = dev->flags;
4529 err = __dev_set_promiscuity(dev, inc);
4532 if (dev->flags != old_flags)
4533 dev_set_rx_mode(dev);
4536 EXPORT_SYMBOL(dev_set_promiscuity);
4539 * dev_set_allmulti - update allmulti count on a device
4543 * Add or remove reception of all multicast frames to a device. While the
4544 * count in the device remains above zero the interface remains listening
4545 * to all interfaces. Once it hits zero the device reverts back to normal
4546 * filtering operation. A negative @inc value is used to drop the counter
4547 * when releasing a resource needing all multicasts.
4548 * Return 0 if successful or a negative errno code on error.
4551 int dev_set_allmulti(struct net_device *dev, int inc)
4553 unsigned int old_flags = dev->flags;
4557 dev->flags |= IFF_ALLMULTI;
4558 dev->allmulti += inc;
4559 if (dev->allmulti == 0) {
4562 * If inc causes overflow, untouch allmulti and return error.
4565 dev->flags &= ~IFF_ALLMULTI;
4567 dev->allmulti -= inc;
4568 printk(KERN_WARNING "%s: allmulti touches roof, "
4569 "set allmulti failed, allmulti feature of "
4570 "device might be broken.\n", dev->name);
4574 if (dev->flags ^ old_flags) {
4575 dev_change_rx_flags(dev, IFF_ALLMULTI);
4576 dev_set_rx_mode(dev);
4580 EXPORT_SYMBOL(dev_set_allmulti);
4583 * Upload unicast and multicast address lists to device and
4584 * configure RX filtering. When the device doesn't support unicast
4585 * filtering it is put in promiscuous mode while unicast addresses
4588 void __dev_set_rx_mode(struct net_device *dev)
4590 const struct net_device_ops *ops = dev->netdev_ops;
4592 /* dev_open will call this function so the list will stay sane. */
4593 if (!(dev->flags&IFF_UP))
4596 if (!netif_device_present(dev))
4599 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4600 /* Unicast addresses changes may only happen under the rtnl,
4601 * therefore calling __dev_set_promiscuity here is safe.
4603 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4604 __dev_set_promiscuity(dev, 1);
4605 dev->uc_promisc = true;
4606 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4607 __dev_set_promiscuity(dev, -1);
4608 dev->uc_promisc = false;
4612 if (ops->ndo_set_rx_mode)
4613 ops->ndo_set_rx_mode(dev);
4616 void dev_set_rx_mode(struct net_device *dev)
4618 netif_addr_lock_bh(dev);
4619 __dev_set_rx_mode(dev);
4620 netif_addr_unlock_bh(dev);
4624 * dev_get_flags - get flags reported to userspace
4627 * Get the combination of flag bits exported through APIs to userspace.
4629 unsigned dev_get_flags(const struct net_device *dev)
4633 flags = (dev->flags & ~(IFF_PROMISC |
4638 (dev->gflags & (IFF_PROMISC |
4641 if (netif_running(dev)) {
4642 if (netif_oper_up(dev))
4643 flags |= IFF_RUNNING;
4644 if (netif_carrier_ok(dev))
4645 flags |= IFF_LOWER_UP;
4646 if (netif_dormant(dev))
4647 flags |= IFF_DORMANT;
4652 EXPORT_SYMBOL(dev_get_flags);
4654 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4656 unsigned int old_flags = dev->flags;
4662 * Set the flags on our device.
4665 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4666 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4668 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4672 * Load in the correct multicast list now the flags have changed.
4675 if ((old_flags ^ flags) & IFF_MULTICAST)
4676 dev_change_rx_flags(dev, IFF_MULTICAST);
4678 dev_set_rx_mode(dev);
4681 * Have we downed the interface. We handle IFF_UP ourselves
4682 * according to user attempts to set it, rather than blindly
4687 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4688 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4691 dev_set_rx_mode(dev);
4694 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4695 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4697 dev->gflags ^= IFF_PROMISC;
4698 dev_set_promiscuity(dev, inc);
4701 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4702 is important. Some (broken) drivers set IFF_PROMISC, when
4703 IFF_ALLMULTI is requested not asking us and not reporting.
4705 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4706 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4708 dev->gflags ^= IFF_ALLMULTI;
4709 dev_set_allmulti(dev, inc);
4715 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4717 unsigned int changes = dev->flags ^ old_flags;
4719 if (changes & IFF_UP) {
4720 if (dev->flags & IFF_UP)
4721 call_netdevice_notifiers(NETDEV_UP, dev);
4723 call_netdevice_notifiers(NETDEV_DOWN, dev);
4726 if (dev->flags & IFF_UP &&
4727 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4728 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4732 * dev_change_flags - change device settings
4734 * @flags: device state flags
4736 * Change settings on device based state flags. The flags are
4737 * in the userspace exported format.
4739 int dev_change_flags(struct net_device *dev, unsigned int flags)
4742 unsigned int changes, old_flags = dev->flags;
4744 ret = __dev_change_flags(dev, flags);
4748 changes = old_flags ^ dev->flags;
4750 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4752 __dev_notify_flags(dev, old_flags);
4755 EXPORT_SYMBOL(dev_change_flags);
4758 * dev_set_mtu - Change maximum transfer unit
4760 * @new_mtu: new transfer unit
4762 * Change the maximum transfer size of the network device.
4764 int dev_set_mtu(struct net_device *dev, int new_mtu)
4766 const struct net_device_ops *ops = dev->netdev_ops;
4769 if (new_mtu == dev->mtu)
4772 /* MTU must be positive. */
4776 if (!netif_device_present(dev))
4780 if (ops->ndo_change_mtu)
4781 err = ops->ndo_change_mtu(dev, new_mtu);
4785 if (!err && dev->flags & IFF_UP)
4786 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4789 EXPORT_SYMBOL(dev_set_mtu);
4792 * dev_set_group - Change group this device belongs to
4794 * @new_group: group this device should belong to
4796 void dev_set_group(struct net_device *dev, int new_group)
4798 dev->group = new_group;
4800 EXPORT_SYMBOL(dev_set_group);
4803 * dev_set_mac_address - Change Media Access Control Address
4807 * Change the hardware (MAC) address of the device
4809 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4811 const struct net_device_ops *ops = dev->netdev_ops;
4814 if (!ops->ndo_set_mac_address)
4816 if (sa->sa_family != dev->type)
4818 if (!netif_device_present(dev))
4820 err = ops->ndo_set_mac_address(dev, sa);
4822 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4825 EXPORT_SYMBOL(dev_set_mac_address);
4828 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4830 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4833 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4839 case SIOCGIFFLAGS: /* Get interface flags */
4840 ifr->ifr_flags = (short) dev_get_flags(dev);
4843 case SIOCGIFMETRIC: /* Get the metric on the interface
4844 (currently unused) */
4845 ifr->ifr_metric = 0;
4848 case SIOCGIFMTU: /* Get the MTU of a device */
4849 ifr->ifr_mtu = dev->mtu;
4854 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4856 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4857 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4858 ifr->ifr_hwaddr.sa_family = dev->type;
4866 ifr->ifr_map.mem_start = dev->mem_start;
4867 ifr->ifr_map.mem_end = dev->mem_end;
4868 ifr->ifr_map.base_addr = dev->base_addr;
4869 ifr->ifr_map.irq = dev->irq;
4870 ifr->ifr_map.dma = dev->dma;
4871 ifr->ifr_map.port = dev->if_port;
4875 ifr->ifr_ifindex = dev->ifindex;
4879 ifr->ifr_qlen = dev->tx_queue_len;
4883 /* dev_ioctl() should ensure this case
4895 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4897 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4900 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4901 const struct net_device_ops *ops;
4906 ops = dev->netdev_ops;
4909 case SIOCSIFFLAGS: /* Set interface flags */
4910 return dev_change_flags(dev, ifr->ifr_flags);
4912 case SIOCSIFMETRIC: /* Set the metric on the interface
4913 (currently unused) */
4916 case SIOCSIFMTU: /* Set the MTU of a device */
4917 return dev_set_mtu(dev, ifr->ifr_mtu);
4920 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4922 case SIOCSIFHWBROADCAST:
4923 if (ifr->ifr_hwaddr.sa_family != dev->type)
4925 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4926 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4927 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4931 if (ops->ndo_set_config) {
4932 if (!netif_device_present(dev))
4934 return ops->ndo_set_config(dev, &ifr->ifr_map);
4939 if (!ops->ndo_set_rx_mode ||
4940 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4942 if (!netif_device_present(dev))
4944 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4947 if (!ops->ndo_set_rx_mode ||
4948 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4950 if (!netif_device_present(dev))
4952 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4955 if (ifr->ifr_qlen < 0)
4957 dev->tx_queue_len = ifr->ifr_qlen;
4961 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4962 return dev_change_name(dev, ifr->ifr_newname);
4965 err = net_hwtstamp_validate(ifr);
4971 * Unknown or private ioctl
4974 if ((cmd >= SIOCDEVPRIVATE &&
4975 cmd <= SIOCDEVPRIVATE + 15) ||
4976 cmd == SIOCBONDENSLAVE ||
4977 cmd == SIOCBONDRELEASE ||
4978 cmd == SIOCBONDSETHWADDR ||
4979 cmd == SIOCBONDSLAVEINFOQUERY ||
4980 cmd == SIOCBONDINFOQUERY ||
4981 cmd == SIOCBONDCHANGEACTIVE ||
4982 cmd == SIOCGMIIPHY ||
4983 cmd == SIOCGMIIREG ||
4984 cmd == SIOCSMIIREG ||
4985 cmd == SIOCBRADDIF ||
4986 cmd == SIOCBRDELIF ||
4987 cmd == SIOCSHWTSTAMP ||
4988 cmd == SIOCWANDEV) {
4990 if (ops->ndo_do_ioctl) {
4991 if (netif_device_present(dev))
4992 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5004 * This function handles all "interface"-type I/O control requests. The actual
5005 * 'doing' part of this is dev_ifsioc above.
5009 * dev_ioctl - network device ioctl
5010 * @net: the applicable net namespace
5011 * @cmd: command to issue
5012 * @arg: pointer to a struct ifreq in user space
5014 * Issue ioctl functions to devices. This is normally called by the
5015 * user space syscall interfaces but can sometimes be useful for
5016 * other purposes. The return value is the return from the syscall if
5017 * positive or a negative errno code on error.
5020 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5026 /* One special case: SIOCGIFCONF takes ifconf argument
5027 and requires shared lock, because it sleeps writing
5031 if (cmd == SIOCGIFCONF) {
5033 ret = dev_ifconf(net, (char __user *) arg);
5037 if (cmd == SIOCGIFNAME)
5038 return dev_ifname(net, (struct ifreq __user *)arg);
5040 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5043 ifr.ifr_name[IFNAMSIZ-1] = 0;
5045 colon = strchr(ifr.ifr_name, ':');
5050 * See which interface the caller is talking about.
5055 * These ioctl calls:
5056 * - can be done by all.
5057 * - atomic and do not require locking.
5068 dev_load(net, ifr.ifr_name);
5070 ret = dev_ifsioc_locked(net, &ifr, cmd);
5075 if (copy_to_user(arg, &ifr,
5076 sizeof(struct ifreq)))
5082 dev_load(net, ifr.ifr_name);
5084 ret = dev_ethtool(net, &ifr);
5089 if (copy_to_user(arg, &ifr,
5090 sizeof(struct ifreq)))
5096 * These ioctl calls:
5097 * - require superuser power.
5098 * - require strict serialization.
5104 if (!capable(CAP_NET_ADMIN))
5106 dev_load(net, ifr.ifr_name);
5108 ret = dev_ifsioc(net, &ifr, cmd);
5113 if (copy_to_user(arg, &ifr,
5114 sizeof(struct ifreq)))
5120 * These ioctl calls:
5121 * - require superuser power.
5122 * - require strict serialization.
5123 * - do not return a value
5133 case SIOCSIFHWBROADCAST:
5136 case SIOCBONDENSLAVE:
5137 case SIOCBONDRELEASE:
5138 case SIOCBONDSETHWADDR:
5139 case SIOCBONDCHANGEACTIVE:
5143 if (!capable(CAP_NET_ADMIN))
5146 case SIOCBONDSLAVEINFOQUERY:
5147 case SIOCBONDINFOQUERY:
5148 dev_load(net, ifr.ifr_name);
5150 ret = dev_ifsioc(net, &ifr, cmd);
5155 /* Get the per device memory space. We can add this but
5156 * currently do not support it */
5158 /* Set the per device memory buffer space.
5159 * Not applicable in our case */
5164 * Unknown or private ioctl.
5167 if (cmd == SIOCWANDEV ||
5168 (cmd >= SIOCDEVPRIVATE &&
5169 cmd <= SIOCDEVPRIVATE + 15)) {
5170 dev_load(net, ifr.ifr_name);
5172 ret = dev_ifsioc(net, &ifr, cmd);
5174 if (!ret && copy_to_user(arg, &ifr,
5175 sizeof(struct ifreq)))
5179 /* Take care of Wireless Extensions */
5180 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5181 return wext_handle_ioctl(net, &ifr, cmd, arg);
5188 * dev_new_index - allocate an ifindex
5189 * @net: the applicable net namespace
5191 * Returns a suitable unique value for a new device interface
5192 * number. The caller must hold the rtnl semaphore or the
5193 * dev_base_lock to be sure it remains unique.
5195 static int dev_new_index(struct net *net)
5201 if (!__dev_get_by_index(net, ifindex))
5206 /* Delayed registration/unregisteration */
5207 static LIST_HEAD(net_todo_list);
5209 static void net_set_todo(struct net_device *dev)
5211 list_add_tail(&dev->todo_list, &net_todo_list);
5214 static void rollback_registered_many(struct list_head *head)
5216 struct net_device *dev, *tmp;
5218 BUG_ON(dev_boot_phase);
5221 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5222 /* Some devices call without registering
5223 * for initialization unwind. Remove those
5224 * devices and proceed with the remaining.
5226 if (dev->reg_state == NETREG_UNINITIALIZED) {
5227 pr_debug("unregister_netdevice: device %s/%p never "
5228 "was registered\n", dev->name, dev);
5231 list_del(&dev->unreg_list);
5234 dev->dismantle = true;
5235 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5238 /* If device is running, close it first. */
5239 dev_close_many(head);
5241 list_for_each_entry(dev, head, unreg_list) {
5242 /* And unlink it from device chain. */
5243 unlist_netdevice(dev);
5245 dev->reg_state = NETREG_UNREGISTERING;
5250 list_for_each_entry(dev, head, unreg_list) {
5251 /* Shutdown queueing discipline. */
5255 /* Notify protocols, that we are about to destroy
5256 this device. They should clean all the things.
5258 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5260 if (!dev->rtnl_link_ops ||
5261 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5262 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5265 * Flush the unicast and multicast chains
5270 if (dev->netdev_ops->ndo_uninit)
5271 dev->netdev_ops->ndo_uninit(dev);
5273 /* Notifier chain MUST detach us from master device. */
5274 WARN_ON(dev->master);
5276 /* Remove entries from kobject tree */
5277 netdev_unregister_kobject(dev);
5280 /* Process any work delayed until the end of the batch */
5281 dev = list_first_entry(head, struct net_device, unreg_list);
5282 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5286 list_for_each_entry(dev, head, unreg_list)
5290 static void rollback_registered(struct net_device *dev)
5294 list_add(&dev->unreg_list, &single);
5295 rollback_registered_many(&single);
5299 static netdev_features_t netdev_fix_features(struct net_device *dev,
5300 netdev_features_t features)
5302 /* Fix illegal checksum combinations */
5303 if ((features & NETIF_F_HW_CSUM) &&
5304 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5305 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5306 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5309 /* Fix illegal SG+CSUM combinations. */
5310 if ((features & NETIF_F_SG) &&
5311 !(features & NETIF_F_ALL_CSUM)) {
5313 "Dropping NETIF_F_SG since no checksum feature.\n");
5314 features &= ~NETIF_F_SG;
5317 /* TSO requires that SG is present as well. */
5318 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5319 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5320 features &= ~NETIF_F_ALL_TSO;
5323 /* TSO ECN requires that TSO is present as well. */
5324 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5325 features &= ~NETIF_F_TSO_ECN;
5327 /* Software GSO depends on SG. */
5328 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5329 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5330 features &= ~NETIF_F_GSO;
5333 /* UFO needs SG and checksumming */
5334 if (features & NETIF_F_UFO) {
5335 /* maybe split UFO into V4 and V6? */
5336 if (!((features & NETIF_F_GEN_CSUM) ||
5337 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5338 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5340 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5341 features &= ~NETIF_F_UFO;
5344 if (!(features & NETIF_F_SG)) {
5346 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5347 features &= ~NETIF_F_UFO;
5354 int __netdev_update_features(struct net_device *dev)
5356 netdev_features_t features;
5361 features = netdev_get_wanted_features(dev);
5363 if (dev->netdev_ops->ndo_fix_features)
5364 features = dev->netdev_ops->ndo_fix_features(dev, features);
5366 /* driver might be less strict about feature dependencies */
5367 features = netdev_fix_features(dev, features);
5369 if (dev->features == features)
5372 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5373 &dev->features, &features);
5375 if (dev->netdev_ops->ndo_set_features)
5376 err = dev->netdev_ops->ndo_set_features(dev, features);
5378 if (unlikely(err < 0)) {
5380 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5381 err, &features, &dev->features);
5386 dev->features = features;
5392 * netdev_update_features - recalculate device features
5393 * @dev: the device to check
5395 * Recalculate dev->features set and send notifications if it
5396 * has changed. Should be called after driver or hardware dependent
5397 * conditions might have changed that influence the features.
5399 void netdev_update_features(struct net_device *dev)
5401 if (__netdev_update_features(dev))
5402 netdev_features_change(dev);
5404 EXPORT_SYMBOL(netdev_update_features);
5407 * netdev_change_features - recalculate device features
5408 * @dev: the device to check
5410 * Recalculate dev->features set and send notifications even
5411 * if they have not changed. Should be called instead of
5412 * netdev_update_features() if also dev->vlan_features might
5413 * have changed to allow the changes to be propagated to stacked
5416 void netdev_change_features(struct net_device *dev)
5418 __netdev_update_features(dev);
5419 netdev_features_change(dev);
5421 EXPORT_SYMBOL(netdev_change_features);
5424 * netif_stacked_transfer_operstate - transfer operstate
5425 * @rootdev: the root or lower level device to transfer state from
5426 * @dev: the device to transfer operstate to
5428 * Transfer operational state from root to device. This is normally
5429 * called when a stacking relationship exists between the root
5430 * device and the device(a leaf device).
5432 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5433 struct net_device *dev)
5435 if (rootdev->operstate == IF_OPER_DORMANT)
5436 netif_dormant_on(dev);
5438 netif_dormant_off(dev);
5440 if (netif_carrier_ok(rootdev)) {
5441 if (!netif_carrier_ok(dev))
5442 netif_carrier_on(dev);
5444 if (netif_carrier_ok(dev))
5445 netif_carrier_off(dev);
5448 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5451 static int netif_alloc_rx_queues(struct net_device *dev)
5453 unsigned int i, count = dev->num_rx_queues;
5454 struct netdev_rx_queue *rx;
5458 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5460 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5465 for (i = 0; i < count; i++)
5471 static void netdev_init_one_queue(struct net_device *dev,
5472 struct netdev_queue *queue, void *_unused)
5474 /* Initialize queue lock */
5475 spin_lock_init(&queue->_xmit_lock);
5476 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5477 queue->xmit_lock_owner = -1;
5478 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5481 dql_init(&queue->dql, HZ);
5485 static int netif_alloc_netdev_queues(struct net_device *dev)
5487 unsigned int count = dev->num_tx_queues;
5488 struct netdev_queue *tx;
5492 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5494 pr_err("netdev: Unable to allocate %u tx queues.\n",
5500 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5501 spin_lock_init(&dev->tx_global_lock);
5507 * register_netdevice - register a network device
5508 * @dev: device to register
5510 * Take a completed network device structure and add it to the kernel
5511 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5512 * chain. 0 is returned on success. A negative errno code is returned
5513 * on a failure to set up the device, or if the name is a duplicate.
5515 * Callers must hold the rtnl semaphore. You may want
5516 * register_netdev() instead of this.
5519 * The locking appears insufficient to guarantee two parallel registers
5520 * will not get the same name.
5523 int register_netdevice(struct net_device *dev)
5526 struct net *net = dev_net(dev);
5528 BUG_ON(dev_boot_phase);
5533 /* When net_device's are persistent, this will be fatal. */
5534 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5537 spin_lock_init(&dev->addr_list_lock);
5538 netdev_set_addr_lockdep_class(dev);
5542 ret = dev_get_valid_name(dev, dev->name);
5546 /* Init, if this function is available */
5547 if (dev->netdev_ops->ndo_init) {
5548 ret = dev->netdev_ops->ndo_init(dev);
5556 dev->ifindex = dev_new_index(net);
5557 if (dev->iflink == -1)
5558 dev->iflink = dev->ifindex;
5560 /* Transfer changeable features to wanted_features and enable
5561 * software offloads (GSO and GRO).
5563 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5564 dev->features |= NETIF_F_SOFT_FEATURES;
5565 dev->wanted_features = dev->features & dev->hw_features;
5567 /* Turn on no cache copy if HW is doing checksum */
5568 if (!(dev->flags & IFF_LOOPBACK)) {
5569 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5570 if (dev->features & NETIF_F_ALL_CSUM) {
5571 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5572 dev->features |= NETIF_F_NOCACHE_COPY;
5576 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5578 dev->vlan_features |= NETIF_F_HIGHDMA;
5580 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5581 ret = notifier_to_errno(ret);
5585 ret = netdev_register_kobject(dev);
5588 dev->reg_state = NETREG_REGISTERED;
5590 __netdev_update_features(dev);
5593 * Default initial state at registry is that the
5594 * device is present.
5597 set_bit(__LINK_STATE_PRESENT, &dev->state);
5599 dev_init_scheduler(dev);
5601 list_netdevice(dev);
5603 /* Notify protocols, that a new device appeared. */
5604 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5605 ret = notifier_to_errno(ret);
5607 rollback_registered(dev);
5608 dev->reg_state = NETREG_UNREGISTERED;
5611 * Prevent userspace races by waiting until the network
5612 * device is fully setup before sending notifications.
5614 if (!dev->rtnl_link_ops ||
5615 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5616 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5622 if (dev->netdev_ops->ndo_uninit)
5623 dev->netdev_ops->ndo_uninit(dev);
5626 EXPORT_SYMBOL(register_netdevice);
5629 * init_dummy_netdev - init a dummy network device for NAPI
5630 * @dev: device to init
5632 * This takes a network device structure and initialize the minimum
5633 * amount of fields so it can be used to schedule NAPI polls without
5634 * registering a full blown interface. This is to be used by drivers
5635 * that need to tie several hardware interfaces to a single NAPI
5636 * poll scheduler due to HW limitations.
5638 int init_dummy_netdev(struct net_device *dev)
5640 /* Clear everything. Note we don't initialize spinlocks
5641 * are they aren't supposed to be taken by any of the
5642 * NAPI code and this dummy netdev is supposed to be
5643 * only ever used for NAPI polls
5645 memset(dev, 0, sizeof(struct net_device));
5647 /* make sure we BUG if trying to hit standard
5648 * register/unregister code path
5650 dev->reg_state = NETREG_DUMMY;
5652 /* NAPI wants this */
5653 INIT_LIST_HEAD(&dev->napi_list);
5655 /* a dummy interface is started by default */
5656 set_bit(__LINK_STATE_PRESENT, &dev->state);
5657 set_bit(__LINK_STATE_START, &dev->state);
5659 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5660 * because users of this 'device' dont need to change
5666 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5670 * register_netdev - register a network device
5671 * @dev: device to register
5673 * Take a completed network device structure and add it to the kernel
5674 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5675 * chain. 0 is returned on success. A negative errno code is returned
5676 * on a failure to set up the device, or if the name is a duplicate.
5678 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5679 * and expands the device name if you passed a format string to
5682 int register_netdev(struct net_device *dev)
5687 err = register_netdevice(dev);
5691 EXPORT_SYMBOL(register_netdev);
5693 int netdev_refcnt_read(const struct net_device *dev)
5697 for_each_possible_cpu(i)
5698 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5701 EXPORT_SYMBOL(netdev_refcnt_read);
5704 * netdev_wait_allrefs - wait until all references are gone.
5706 * This is called when unregistering network devices.
5708 * Any protocol or device that holds a reference should register
5709 * for netdevice notification, and cleanup and put back the
5710 * reference if they receive an UNREGISTER event.
5711 * We can get stuck here if buggy protocols don't correctly
5714 static void netdev_wait_allrefs(struct net_device *dev)
5716 unsigned long rebroadcast_time, warning_time;
5719 linkwatch_forget_dev(dev);
5721 rebroadcast_time = warning_time = jiffies;
5722 refcnt = netdev_refcnt_read(dev);
5724 while (refcnt != 0) {
5725 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5728 /* Rebroadcast unregister notification */
5729 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5730 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5731 * should have already handle it the first time */
5733 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5735 /* We must not have linkwatch events
5736 * pending on unregister. If this
5737 * happens, we simply run the queue
5738 * unscheduled, resulting in a noop
5741 linkwatch_run_queue();
5746 rebroadcast_time = jiffies;
5751 refcnt = netdev_refcnt_read(dev);
5753 if (time_after(jiffies, warning_time + 10 * HZ)) {
5754 printk(KERN_EMERG "unregister_netdevice: "
5755 "waiting for %s to become free. Usage "
5758 warning_time = jiffies;
5767 * register_netdevice(x1);
5768 * register_netdevice(x2);
5770 * unregister_netdevice(y1);
5771 * unregister_netdevice(y2);
5777 * We are invoked by rtnl_unlock().
5778 * This allows us to deal with problems:
5779 * 1) We can delete sysfs objects which invoke hotplug
5780 * without deadlocking with linkwatch via keventd.
5781 * 2) Since we run with the RTNL semaphore not held, we can sleep
5782 * safely in order to wait for the netdev refcnt to drop to zero.
5784 * We must not return until all unregister events added during
5785 * the interval the lock was held have been completed.
5787 void netdev_run_todo(void)
5789 struct list_head list;
5791 /* Snapshot list, allow later requests */
5792 list_replace_init(&net_todo_list, &list);
5796 /* Wait for rcu callbacks to finish before attempting to drain
5797 * the device list. This usually avoids a 250ms wait.
5799 if (!list_empty(&list))
5802 while (!list_empty(&list)) {
5803 struct net_device *dev
5804 = list_first_entry(&list, struct net_device, todo_list);
5805 list_del(&dev->todo_list);
5807 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5808 printk(KERN_ERR "network todo '%s' but state %d\n",
5809 dev->name, dev->reg_state);
5814 dev->reg_state = NETREG_UNREGISTERED;
5816 on_each_cpu(flush_backlog, dev, 1);
5818 netdev_wait_allrefs(dev);
5821 BUG_ON(netdev_refcnt_read(dev));
5822 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5823 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5824 WARN_ON(dev->dn_ptr);
5826 if (dev->destructor)
5827 dev->destructor(dev);
5829 /* Free network device */
5830 kobject_put(&dev->dev.kobj);
5834 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5835 * fields in the same order, with only the type differing.
5837 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5838 const struct net_device_stats *netdev_stats)
5840 #if BITS_PER_LONG == 64
5841 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5842 memcpy(stats64, netdev_stats, sizeof(*stats64));
5844 size_t i, n = sizeof(*stats64) / sizeof(u64);
5845 const unsigned long *src = (const unsigned long *)netdev_stats;
5846 u64 *dst = (u64 *)stats64;
5848 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5849 sizeof(*stats64) / sizeof(u64));
5850 for (i = 0; i < n; i++)
5856 * dev_get_stats - get network device statistics
5857 * @dev: device to get statistics from
5858 * @storage: place to store stats
5860 * Get network statistics from device. Return @storage.
5861 * The device driver may provide its own method by setting
5862 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5863 * otherwise the internal statistics structure is used.
5865 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5866 struct rtnl_link_stats64 *storage)
5868 const struct net_device_ops *ops = dev->netdev_ops;
5870 if (ops->ndo_get_stats64) {
5871 memset(storage, 0, sizeof(*storage));
5872 ops->ndo_get_stats64(dev, storage);
5873 } else if (ops->ndo_get_stats) {
5874 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5876 netdev_stats_to_stats64(storage, &dev->stats);
5878 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5881 EXPORT_SYMBOL(dev_get_stats);
5883 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5885 struct netdev_queue *queue = dev_ingress_queue(dev);
5887 #ifdef CONFIG_NET_CLS_ACT
5890 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5893 netdev_init_one_queue(dev, queue, NULL);
5894 queue->qdisc = &noop_qdisc;
5895 queue->qdisc_sleeping = &noop_qdisc;
5896 rcu_assign_pointer(dev->ingress_queue, queue);
5902 * alloc_netdev_mqs - allocate network device
5903 * @sizeof_priv: size of private data to allocate space for
5904 * @name: device name format string
5905 * @setup: callback to initialize device
5906 * @txqs: the number of TX subqueues to allocate
5907 * @rxqs: the number of RX subqueues to allocate
5909 * Allocates a struct net_device with private data area for driver use
5910 * and performs basic initialization. Also allocates subquue structs
5911 * for each queue on the device.
5913 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5914 void (*setup)(struct net_device *),
5915 unsigned int txqs, unsigned int rxqs)
5917 struct net_device *dev;
5919 struct net_device *p;
5921 BUG_ON(strlen(name) >= sizeof(dev->name));
5924 pr_err("alloc_netdev: Unable to allocate device "
5925 "with zero queues.\n");
5931 pr_err("alloc_netdev: Unable to allocate device "
5932 "with zero RX queues.\n");
5937 alloc_size = sizeof(struct net_device);
5939 /* ensure 32-byte alignment of private area */
5940 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5941 alloc_size += sizeof_priv;
5943 /* ensure 32-byte alignment of whole construct */
5944 alloc_size += NETDEV_ALIGN - 1;
5946 p = kzalloc(alloc_size, GFP_KERNEL);
5948 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5952 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5953 dev->padded = (char *)dev - (char *)p;
5955 dev->pcpu_refcnt = alloc_percpu(int);
5956 if (!dev->pcpu_refcnt)
5959 if (dev_addr_init(dev))
5965 dev_net_set(dev, &init_net);
5967 dev->gso_max_size = GSO_MAX_SIZE;
5969 INIT_LIST_HEAD(&dev->napi_list);
5970 INIT_LIST_HEAD(&dev->unreg_list);
5971 INIT_LIST_HEAD(&dev->link_watch_list);
5972 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5975 dev->num_tx_queues = txqs;
5976 dev->real_num_tx_queues = txqs;
5977 if (netif_alloc_netdev_queues(dev))
5981 dev->num_rx_queues = rxqs;
5982 dev->real_num_rx_queues = rxqs;
5983 if (netif_alloc_rx_queues(dev))
5987 strcpy(dev->name, name);
5988 dev->group = INIT_NETDEV_GROUP;
5996 free_percpu(dev->pcpu_refcnt);
6006 EXPORT_SYMBOL(alloc_netdev_mqs);
6009 * free_netdev - free network device
6012 * This function does the last stage of destroying an allocated device
6013 * interface. The reference to the device object is released.
6014 * If this is the last reference then it will be freed.
6016 void free_netdev(struct net_device *dev)
6018 struct napi_struct *p, *n;
6020 release_net(dev_net(dev));
6027 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6029 /* Flush device addresses */
6030 dev_addr_flush(dev);
6032 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6035 free_percpu(dev->pcpu_refcnt);
6036 dev->pcpu_refcnt = NULL;
6038 /* Compatibility with error handling in drivers */
6039 if (dev->reg_state == NETREG_UNINITIALIZED) {
6040 kfree((char *)dev - dev->padded);
6044 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6045 dev->reg_state = NETREG_RELEASED;
6047 /* will free via device release */
6048 put_device(&dev->dev);
6050 EXPORT_SYMBOL(free_netdev);
6053 * synchronize_net - Synchronize with packet receive processing
6055 * Wait for packets currently being received to be done.
6056 * Does not block later packets from starting.
6058 void synchronize_net(void)
6061 if (rtnl_is_locked())
6062 synchronize_rcu_expedited();
6066 EXPORT_SYMBOL(synchronize_net);
6069 * unregister_netdevice_queue - remove device from the kernel
6073 * This function shuts down a device interface and removes it
6074 * from the kernel tables.
6075 * If head not NULL, device is queued to be unregistered later.
6077 * Callers must hold the rtnl semaphore. You may want
6078 * unregister_netdev() instead of this.
6081 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6086 list_move_tail(&dev->unreg_list, head);
6088 rollback_registered(dev);
6089 /* Finish processing unregister after unlock */
6093 EXPORT_SYMBOL(unregister_netdevice_queue);
6096 * unregister_netdevice_many - unregister many devices
6097 * @head: list of devices
6099 void unregister_netdevice_many(struct list_head *head)
6101 struct net_device *dev;
6103 if (!list_empty(head)) {
6104 rollback_registered_many(head);
6105 list_for_each_entry(dev, head, unreg_list)
6109 EXPORT_SYMBOL(unregister_netdevice_many);
6112 * unregister_netdev - remove device from the kernel
6115 * This function shuts down a device interface and removes it
6116 * from the kernel tables.
6118 * This is just a wrapper for unregister_netdevice that takes
6119 * the rtnl semaphore. In general you want to use this and not
6120 * unregister_netdevice.
6122 void unregister_netdev(struct net_device *dev)
6125 unregister_netdevice(dev);
6128 EXPORT_SYMBOL(unregister_netdev);
6131 * dev_change_net_namespace - move device to different nethost namespace
6133 * @net: network namespace
6134 * @pat: If not NULL name pattern to try if the current device name
6135 * is already taken in the destination network namespace.
6137 * This function shuts down a device interface and moves it
6138 * to a new network namespace. On success 0 is returned, on
6139 * a failure a netagive errno code is returned.
6141 * Callers must hold the rtnl semaphore.
6144 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6150 /* Don't allow namespace local devices to be moved. */
6152 if (dev->features & NETIF_F_NETNS_LOCAL)
6155 /* Ensure the device has been registrered */
6157 if (dev->reg_state != NETREG_REGISTERED)
6160 /* Get out if there is nothing todo */
6162 if (net_eq(dev_net(dev), net))
6165 /* Pick the destination device name, and ensure
6166 * we can use it in the destination network namespace.
6169 if (__dev_get_by_name(net, dev->name)) {
6170 /* We get here if we can't use the current device name */
6173 if (dev_get_valid_name(dev, pat) < 0)
6178 * And now a mini version of register_netdevice unregister_netdevice.
6181 /* If device is running close it first. */
6184 /* And unlink it from device chain */
6186 unlist_netdevice(dev);
6190 /* Shutdown queueing discipline. */
6193 /* Notify protocols, that we are about to destroy
6194 this device. They should clean all the things.
6196 Note that dev->reg_state stays at NETREG_REGISTERED.
6197 This is wanted because this way 8021q and macvlan know
6198 the device is just moving and can keep their slaves up.
6200 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6201 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6202 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6205 * Flush the unicast and multicast chains
6210 /* Actually switch the network namespace */
6211 dev_net_set(dev, net);
6213 /* If there is an ifindex conflict assign a new one */
6214 if (__dev_get_by_index(net, dev->ifindex)) {
6215 int iflink = (dev->iflink == dev->ifindex);
6216 dev->ifindex = dev_new_index(net);
6218 dev->iflink = dev->ifindex;
6221 /* Fixup kobjects */
6222 err = device_rename(&dev->dev, dev->name);
6225 /* Add the device back in the hashes */
6226 list_netdevice(dev);
6228 /* Notify protocols, that a new device appeared. */
6229 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6232 * Prevent userspace races by waiting until the network
6233 * device is fully setup before sending notifications.
6235 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6242 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6244 static int dev_cpu_callback(struct notifier_block *nfb,
6245 unsigned long action,
6248 struct sk_buff **list_skb;
6249 struct sk_buff *skb;
6250 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6251 struct softnet_data *sd, *oldsd;
6253 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6256 local_irq_disable();
6257 cpu = smp_processor_id();
6258 sd = &per_cpu(softnet_data, cpu);
6259 oldsd = &per_cpu(softnet_data, oldcpu);
6261 /* Find end of our completion_queue. */
6262 list_skb = &sd->completion_queue;
6264 list_skb = &(*list_skb)->next;
6265 /* Append completion queue from offline CPU. */
6266 *list_skb = oldsd->completion_queue;
6267 oldsd->completion_queue = NULL;
6269 /* Append output queue from offline CPU. */
6270 if (oldsd->output_queue) {
6271 *sd->output_queue_tailp = oldsd->output_queue;
6272 sd->output_queue_tailp = oldsd->output_queue_tailp;
6273 oldsd->output_queue = NULL;
6274 oldsd->output_queue_tailp = &oldsd->output_queue;
6276 /* Append NAPI poll list from offline CPU. */
6277 if (!list_empty(&oldsd->poll_list)) {
6278 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6279 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6282 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6285 /* Process offline CPU's input_pkt_queue */
6286 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6288 input_queue_head_incr(oldsd);
6290 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6292 input_queue_head_incr(oldsd);
6300 * netdev_increment_features - increment feature set by one
6301 * @all: current feature set
6302 * @one: new feature set
6303 * @mask: mask feature set
6305 * Computes a new feature set after adding a device with feature set
6306 * @one to the master device with current feature set @all. Will not
6307 * enable anything that is off in @mask. Returns the new feature set.
6309 netdev_features_t netdev_increment_features(netdev_features_t all,
6310 netdev_features_t one, netdev_features_t mask)
6312 if (mask & NETIF_F_GEN_CSUM)
6313 mask |= NETIF_F_ALL_CSUM;
6314 mask |= NETIF_F_VLAN_CHALLENGED;
6316 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6317 all &= one | ~NETIF_F_ALL_FOR_ALL;
6319 /* If one device supports hw checksumming, set for all. */
6320 if (all & NETIF_F_GEN_CSUM)
6321 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6325 EXPORT_SYMBOL(netdev_increment_features);
6327 static struct hlist_head *netdev_create_hash(void)
6330 struct hlist_head *hash;
6332 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6334 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6335 INIT_HLIST_HEAD(&hash[i]);
6340 /* Initialize per network namespace state */
6341 static int __net_init netdev_init(struct net *net)
6343 INIT_LIST_HEAD(&net->dev_base_head);
6345 net->dev_name_head = netdev_create_hash();
6346 if (net->dev_name_head == NULL)
6349 net->dev_index_head = netdev_create_hash();
6350 if (net->dev_index_head == NULL)
6356 kfree(net->dev_name_head);
6362 * netdev_drivername - network driver for the device
6363 * @dev: network device
6365 * Determine network driver for device.
6367 const char *netdev_drivername(const struct net_device *dev)
6369 const struct device_driver *driver;
6370 const struct device *parent;
6371 const char *empty = "";
6373 parent = dev->dev.parent;
6377 driver = parent->driver;
6378 if (driver && driver->name)
6379 return driver->name;
6383 int __netdev_printk(const char *level, const struct net_device *dev,
6384 struct va_format *vaf)
6388 if (dev && dev->dev.parent)
6389 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6390 netdev_name(dev), vaf);
6392 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6394 r = printk("%s(NULL net_device): %pV", level, vaf);
6398 EXPORT_SYMBOL(__netdev_printk);
6400 int netdev_printk(const char *level, const struct net_device *dev,
6401 const char *format, ...)
6403 struct va_format vaf;
6407 va_start(args, format);
6412 r = __netdev_printk(level, dev, &vaf);
6417 EXPORT_SYMBOL(netdev_printk);
6419 #define define_netdev_printk_level(func, level) \
6420 int func(const struct net_device *dev, const char *fmt, ...) \
6423 struct va_format vaf; \
6426 va_start(args, fmt); \
6431 r = __netdev_printk(level, dev, &vaf); \
6436 EXPORT_SYMBOL(func);
6438 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6439 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6440 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6441 define_netdev_printk_level(netdev_err, KERN_ERR);
6442 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6443 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6444 define_netdev_printk_level(netdev_info, KERN_INFO);
6446 static void __net_exit netdev_exit(struct net *net)
6448 kfree(net->dev_name_head);
6449 kfree(net->dev_index_head);
6452 static struct pernet_operations __net_initdata netdev_net_ops = {
6453 .init = netdev_init,
6454 .exit = netdev_exit,
6457 static void __net_exit default_device_exit(struct net *net)
6459 struct net_device *dev, *aux;
6461 * Push all migratable network devices back to the
6462 * initial network namespace
6465 for_each_netdev_safe(net, dev, aux) {
6467 char fb_name[IFNAMSIZ];
6469 /* Ignore unmoveable devices (i.e. loopback) */
6470 if (dev->features & NETIF_F_NETNS_LOCAL)
6473 /* Leave virtual devices for the generic cleanup */
6474 if (dev->rtnl_link_ops)
6477 /* Push remaining network devices to init_net */
6478 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6479 err = dev_change_net_namespace(dev, &init_net, fb_name);
6481 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6482 __func__, dev->name, err);
6489 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6491 /* At exit all network devices most be removed from a network
6492 * namespace. Do this in the reverse order of registration.
6493 * Do this across as many network namespaces as possible to
6494 * improve batching efficiency.
6496 struct net_device *dev;
6498 LIST_HEAD(dev_kill_list);
6501 list_for_each_entry(net, net_list, exit_list) {
6502 for_each_netdev_reverse(net, dev) {
6503 if (dev->rtnl_link_ops)
6504 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6506 unregister_netdevice_queue(dev, &dev_kill_list);
6509 unregister_netdevice_many(&dev_kill_list);
6510 list_del(&dev_kill_list);
6514 static struct pernet_operations __net_initdata default_device_ops = {
6515 .exit = default_device_exit,
6516 .exit_batch = default_device_exit_batch,
6520 * Initialize the DEV module. At boot time this walks the device list and
6521 * unhooks any devices that fail to initialise (normally hardware not
6522 * present) and leaves us with a valid list of present and active devices.
6527 * This is called single threaded during boot, so no need
6528 * to take the rtnl semaphore.
6530 static int __init net_dev_init(void)
6532 int i, rc = -ENOMEM;
6534 BUG_ON(!dev_boot_phase);
6536 if (dev_proc_init())
6539 if (netdev_kobject_init())
6542 INIT_LIST_HEAD(&ptype_all);
6543 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6544 INIT_LIST_HEAD(&ptype_base[i]);
6546 if (register_pernet_subsys(&netdev_net_ops))
6550 * Initialise the packet receive queues.
6553 for_each_possible_cpu(i) {
6554 struct softnet_data *sd = &per_cpu(softnet_data, i);
6556 memset(sd, 0, sizeof(*sd));
6557 skb_queue_head_init(&sd->input_pkt_queue);
6558 skb_queue_head_init(&sd->process_queue);
6559 sd->completion_queue = NULL;
6560 INIT_LIST_HEAD(&sd->poll_list);
6561 sd->output_queue = NULL;
6562 sd->output_queue_tailp = &sd->output_queue;
6564 sd->csd.func = rps_trigger_softirq;
6570 sd->backlog.poll = process_backlog;
6571 sd->backlog.weight = weight_p;
6572 sd->backlog.gro_list = NULL;
6573 sd->backlog.gro_count = 0;
6578 /* The loopback device is special if any other network devices
6579 * is present in a network namespace the loopback device must
6580 * be present. Since we now dynamically allocate and free the
6581 * loopback device ensure this invariant is maintained by
6582 * keeping the loopback device as the first device on the
6583 * list of network devices. Ensuring the loopback devices
6584 * is the first device that appears and the last network device
6587 if (register_pernet_device(&loopback_net_ops))
6590 if (register_pernet_device(&default_device_ops))
6593 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6594 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6596 hotcpu_notifier(dev_cpu_callback, 0);
6604 subsys_initcall(net_dev_init);
6606 static int __init initialize_hashrnd(void)
6608 get_random_bytes(&hashrnd, sizeof(hashrnd));
6612 late_initcall_sync(initialize_hashrnd);