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/static_key.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 pr_warn("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 bool 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);
1042 pr_err("%s: name change rollback failed: %d\n",
1051 * dev_set_alias - change ifalias of a device
1053 * @alias: name up to IFALIASZ
1054 * @len: limit of bytes to copy from info
1056 * Set ifalias for a device,
1058 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1062 if (len >= IFALIASZ)
1067 kfree(dev->ifalias);
1068 dev->ifalias = NULL;
1073 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1077 strlcpy(dev->ifalias, alias, len+1);
1083 * netdev_features_change - device changes features
1084 * @dev: device to cause notification
1086 * Called to indicate a device has changed features.
1088 void netdev_features_change(struct net_device *dev)
1090 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1092 EXPORT_SYMBOL(netdev_features_change);
1095 * netdev_state_change - device changes state
1096 * @dev: device to cause notification
1098 * Called to indicate a device has changed state. This function calls
1099 * the notifier chains for netdev_chain and sends a NEWLINK message
1100 * to the routing socket.
1102 void netdev_state_change(struct net_device *dev)
1104 if (dev->flags & IFF_UP) {
1105 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1106 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1109 EXPORT_SYMBOL(netdev_state_change);
1111 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1113 return call_netdevice_notifiers(event, dev);
1115 EXPORT_SYMBOL(netdev_bonding_change);
1118 * dev_load - load a network module
1119 * @net: the applicable net namespace
1120 * @name: name of interface
1122 * If a network interface is not present and the process has suitable
1123 * privileges this function loads the module. If module loading is not
1124 * available in this kernel then it becomes a nop.
1127 void dev_load(struct net *net, const char *name)
1129 struct net_device *dev;
1133 dev = dev_get_by_name_rcu(net, name);
1137 if (no_module && capable(CAP_NET_ADMIN))
1138 no_module = request_module("netdev-%s", name);
1139 if (no_module && capable(CAP_SYS_MODULE)) {
1140 if (!request_module("%s", name))
1141 pr_err("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1145 EXPORT_SYMBOL(dev_load);
1147 static int __dev_open(struct net_device *dev)
1149 const struct net_device_ops *ops = dev->netdev_ops;
1154 if (!netif_device_present(dev))
1157 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1158 ret = notifier_to_errno(ret);
1162 set_bit(__LINK_STATE_START, &dev->state);
1164 if (ops->ndo_validate_addr)
1165 ret = ops->ndo_validate_addr(dev);
1167 if (!ret && ops->ndo_open)
1168 ret = ops->ndo_open(dev);
1171 clear_bit(__LINK_STATE_START, &dev->state);
1173 dev->flags |= IFF_UP;
1174 net_dmaengine_get();
1175 dev_set_rx_mode(dev);
1183 * dev_open - prepare an interface for use.
1184 * @dev: device to open
1186 * Takes a device from down to up state. The device's private open
1187 * function is invoked and then the multicast lists are loaded. Finally
1188 * the device is moved into the up state and a %NETDEV_UP message is
1189 * sent to the netdev notifier chain.
1191 * Calling this function on an active interface is a nop. On a failure
1192 * a negative errno code is returned.
1194 int dev_open(struct net_device *dev)
1198 if (dev->flags & IFF_UP)
1201 ret = __dev_open(dev);
1205 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1206 call_netdevice_notifiers(NETDEV_UP, dev);
1210 EXPORT_SYMBOL(dev_open);
1212 static int __dev_close_many(struct list_head *head)
1214 struct net_device *dev;
1219 list_for_each_entry(dev, head, unreg_list) {
1220 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1222 clear_bit(__LINK_STATE_START, &dev->state);
1224 /* Synchronize to scheduled poll. We cannot touch poll list, it
1225 * can be even on different cpu. So just clear netif_running().
1227 * dev->stop() will invoke napi_disable() on all of it's
1228 * napi_struct instances on this device.
1230 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1233 dev_deactivate_many(head);
1235 list_for_each_entry(dev, head, unreg_list) {
1236 const struct net_device_ops *ops = dev->netdev_ops;
1239 * Call the device specific close. This cannot fail.
1240 * Only if device is UP
1242 * We allow it to be called even after a DETACH hot-plug
1248 dev->flags &= ~IFF_UP;
1249 net_dmaengine_put();
1255 static int __dev_close(struct net_device *dev)
1260 list_add(&dev->unreg_list, &single);
1261 retval = __dev_close_many(&single);
1266 static int dev_close_many(struct list_head *head)
1268 struct net_device *dev, *tmp;
1269 LIST_HEAD(tmp_list);
1271 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1272 if (!(dev->flags & IFF_UP))
1273 list_move(&dev->unreg_list, &tmp_list);
1275 __dev_close_many(head);
1277 list_for_each_entry(dev, head, unreg_list) {
1278 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1279 call_netdevice_notifiers(NETDEV_DOWN, dev);
1282 /* rollback_registered_many needs the complete original list */
1283 list_splice(&tmp_list, head);
1288 * dev_close - shutdown an interface.
1289 * @dev: device to shutdown
1291 * This function moves an active device into down state. A
1292 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1293 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1296 int dev_close(struct net_device *dev)
1298 if (dev->flags & IFF_UP) {
1301 list_add(&dev->unreg_list, &single);
1302 dev_close_many(&single);
1307 EXPORT_SYMBOL(dev_close);
1311 * dev_disable_lro - disable Large Receive Offload on a device
1314 * Disable Large Receive Offload (LRO) on a net device. Must be
1315 * called under RTNL. This is needed if received packets may be
1316 * forwarded to another interface.
1318 void dev_disable_lro(struct net_device *dev)
1321 * If we're trying to disable lro on a vlan device
1322 * use the underlying physical device instead
1324 if (is_vlan_dev(dev))
1325 dev = vlan_dev_real_dev(dev);
1327 dev->wanted_features &= ~NETIF_F_LRO;
1328 netdev_update_features(dev);
1330 if (unlikely(dev->features & NETIF_F_LRO))
1331 netdev_WARN(dev, "failed to disable LRO!\n");
1333 EXPORT_SYMBOL(dev_disable_lro);
1336 static int dev_boot_phase = 1;
1339 * register_netdevice_notifier - register a network notifier block
1342 * Register a notifier to be called when network device events occur.
1343 * The notifier passed is linked into the kernel structures and must
1344 * not be reused until it has been unregistered. A negative errno code
1345 * is returned on a failure.
1347 * When registered all registration and up events are replayed
1348 * to the new notifier to allow device to have a race free
1349 * view of the network device list.
1352 int register_netdevice_notifier(struct notifier_block *nb)
1354 struct net_device *dev;
1355 struct net_device *last;
1360 err = raw_notifier_chain_register(&netdev_chain, nb);
1366 for_each_netdev(net, dev) {
1367 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1368 err = notifier_to_errno(err);
1372 if (!(dev->flags & IFF_UP))
1375 nb->notifier_call(nb, NETDEV_UP, dev);
1386 for_each_netdev(net, dev) {
1390 if (dev->flags & IFF_UP) {
1391 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1392 nb->notifier_call(nb, NETDEV_DOWN, dev);
1394 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1395 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1400 raw_notifier_chain_unregister(&netdev_chain, nb);
1403 EXPORT_SYMBOL(register_netdevice_notifier);
1406 * unregister_netdevice_notifier - unregister a network notifier block
1409 * Unregister a notifier previously registered by
1410 * register_netdevice_notifier(). The notifier is unlinked into the
1411 * kernel structures and may then be reused. A negative errno code
1412 * is returned on a failure.
1415 int unregister_netdevice_notifier(struct notifier_block *nb)
1420 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1424 EXPORT_SYMBOL(unregister_netdevice_notifier);
1427 * call_netdevice_notifiers - call all network notifier blocks
1428 * @val: value passed unmodified to notifier function
1429 * @dev: net_device pointer passed unmodified to notifier function
1431 * Call all network notifier blocks. Parameters and return value
1432 * are as for raw_notifier_call_chain().
1435 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1438 return raw_notifier_call_chain(&netdev_chain, val, dev);
1440 EXPORT_SYMBOL(call_netdevice_notifiers);
1442 static struct static_key netstamp_needed __read_mostly;
1443 #ifdef HAVE_JUMP_LABEL
1444 /* We are not allowed to call static_key_slow_dec() from irq context
1445 * If net_disable_timestamp() is called from irq context, defer the
1446 * static_key_slow_dec() calls.
1448 static atomic_t netstamp_needed_deferred;
1451 void net_enable_timestamp(void)
1453 #ifdef HAVE_JUMP_LABEL
1454 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1458 static_key_slow_dec(&netstamp_needed);
1462 WARN_ON(in_interrupt());
1463 static_key_slow_inc(&netstamp_needed);
1465 EXPORT_SYMBOL(net_enable_timestamp);
1467 void net_disable_timestamp(void)
1469 #ifdef HAVE_JUMP_LABEL
1470 if (in_interrupt()) {
1471 atomic_inc(&netstamp_needed_deferred);
1475 static_key_slow_dec(&netstamp_needed);
1477 EXPORT_SYMBOL(net_disable_timestamp);
1479 static inline void net_timestamp_set(struct sk_buff *skb)
1481 skb->tstamp.tv64 = 0;
1482 if (static_key_false(&netstamp_needed))
1483 __net_timestamp(skb);
1486 #define net_timestamp_check(COND, SKB) \
1487 if (static_key_false(&netstamp_needed)) { \
1488 if ((COND) && !(SKB)->tstamp.tv64) \
1489 __net_timestamp(SKB); \
1492 static int net_hwtstamp_validate(struct ifreq *ifr)
1494 struct hwtstamp_config cfg;
1495 enum hwtstamp_tx_types tx_type;
1496 enum hwtstamp_rx_filters rx_filter;
1497 int tx_type_valid = 0;
1498 int rx_filter_valid = 0;
1500 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1503 if (cfg.flags) /* reserved for future extensions */
1506 tx_type = cfg.tx_type;
1507 rx_filter = cfg.rx_filter;
1510 case HWTSTAMP_TX_OFF:
1511 case HWTSTAMP_TX_ON:
1512 case HWTSTAMP_TX_ONESTEP_SYNC:
1517 switch (rx_filter) {
1518 case HWTSTAMP_FILTER_NONE:
1519 case HWTSTAMP_FILTER_ALL:
1520 case HWTSTAMP_FILTER_SOME:
1521 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1522 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1523 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1524 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1525 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1526 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1527 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1528 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1529 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1530 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1531 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1532 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1533 rx_filter_valid = 1;
1537 if (!tx_type_valid || !rx_filter_valid)
1543 static inline bool is_skb_forwardable(struct net_device *dev,
1544 struct sk_buff *skb)
1548 if (!(dev->flags & IFF_UP))
1551 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1552 if (skb->len <= len)
1555 /* if TSO is enabled, we don't care about the length as the packet
1556 * could be forwarded without being segmented before
1558 if (skb_is_gso(skb))
1565 * dev_forward_skb - loopback an skb to another netif
1567 * @dev: destination network device
1568 * @skb: buffer to forward
1571 * NET_RX_SUCCESS (no congestion)
1572 * NET_RX_DROP (packet was dropped, but freed)
1574 * dev_forward_skb can be used for injecting an skb from the
1575 * start_xmit function of one device into the receive queue
1576 * of another device.
1578 * The receiving device may be in another namespace, so
1579 * we have to clear all information in the skb that could
1580 * impact namespace isolation.
1582 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1584 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1585 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1586 atomic_long_inc(&dev->rx_dropped);
1595 if (unlikely(!is_skb_forwardable(dev, skb))) {
1596 atomic_long_inc(&dev->rx_dropped);
1600 skb_set_dev(skb, dev);
1601 skb->tstamp.tv64 = 0;
1602 skb->pkt_type = PACKET_HOST;
1603 skb->protocol = eth_type_trans(skb, dev);
1604 return netif_rx(skb);
1606 EXPORT_SYMBOL_GPL(dev_forward_skb);
1608 static inline int deliver_skb(struct sk_buff *skb,
1609 struct packet_type *pt_prev,
1610 struct net_device *orig_dev)
1612 atomic_inc(&skb->users);
1613 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1617 * Support routine. Sends outgoing frames to any network
1618 * taps currently in use.
1621 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1623 struct packet_type *ptype;
1624 struct sk_buff *skb2 = NULL;
1625 struct packet_type *pt_prev = NULL;
1628 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1629 /* Never send packets back to the socket
1630 * they originated from - MvS (miquels@drinkel.ow.org)
1632 if ((ptype->dev == dev || !ptype->dev) &&
1633 (ptype->af_packet_priv == NULL ||
1634 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1636 deliver_skb(skb2, pt_prev, skb->dev);
1641 skb2 = skb_clone(skb, GFP_ATOMIC);
1645 net_timestamp_set(skb2);
1647 /* skb->nh should be correctly
1648 set by sender, so that the second statement is
1649 just protection against buggy protocols.
1651 skb_reset_mac_header(skb2);
1653 if (skb_network_header(skb2) < skb2->data ||
1654 skb2->network_header > skb2->tail) {
1655 if (net_ratelimit())
1656 pr_crit("protocol %04x is buggy, dev %s\n",
1657 ntohs(skb2->protocol),
1659 skb_reset_network_header(skb2);
1662 skb2->transport_header = skb2->network_header;
1663 skb2->pkt_type = PACKET_OUTGOING;
1668 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1672 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1673 * @dev: Network device
1674 * @txq: number of queues available
1676 * If real_num_tx_queues is changed the tc mappings may no longer be
1677 * valid. To resolve this verify the tc mapping remains valid and if
1678 * not NULL the mapping. With no priorities mapping to this
1679 * offset/count pair it will no longer be used. In the worst case TC0
1680 * is invalid nothing can be done so disable priority mappings. If is
1681 * expected that drivers will fix this mapping if they can before
1682 * calling netif_set_real_num_tx_queues.
1684 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1687 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1689 /* If TC0 is invalidated disable TC mapping */
1690 if (tc->offset + tc->count > txq) {
1691 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1696 /* Invalidated prio to tc mappings set to TC0 */
1697 for (i = 1; i < TC_BITMASK + 1; i++) {
1698 int q = netdev_get_prio_tc_map(dev, i);
1700 tc = &dev->tc_to_txq[q];
1701 if (tc->offset + tc->count > txq) {
1702 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1704 netdev_set_prio_tc_map(dev, i, 0);
1710 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1711 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1713 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1717 if (txq < 1 || txq > dev->num_tx_queues)
1720 if (dev->reg_state == NETREG_REGISTERED ||
1721 dev->reg_state == NETREG_UNREGISTERING) {
1724 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1730 netif_setup_tc(dev, txq);
1732 if (txq < dev->real_num_tx_queues)
1733 qdisc_reset_all_tx_gt(dev, txq);
1736 dev->real_num_tx_queues = txq;
1739 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1743 * netif_set_real_num_rx_queues - set actual number of RX queues used
1744 * @dev: Network device
1745 * @rxq: Actual number of RX queues
1747 * This must be called either with the rtnl_lock held or before
1748 * registration of the net device. Returns 0 on success, or a
1749 * negative error code. If called before registration, it always
1752 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1756 if (rxq < 1 || rxq > dev->num_rx_queues)
1759 if (dev->reg_state == NETREG_REGISTERED) {
1762 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1768 dev->real_num_rx_queues = rxq;
1771 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1774 static inline void __netif_reschedule(struct Qdisc *q)
1776 struct softnet_data *sd;
1777 unsigned long flags;
1779 local_irq_save(flags);
1780 sd = &__get_cpu_var(softnet_data);
1781 q->next_sched = NULL;
1782 *sd->output_queue_tailp = q;
1783 sd->output_queue_tailp = &q->next_sched;
1784 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1785 local_irq_restore(flags);
1788 void __netif_schedule(struct Qdisc *q)
1790 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1791 __netif_reschedule(q);
1793 EXPORT_SYMBOL(__netif_schedule);
1795 void dev_kfree_skb_irq(struct sk_buff *skb)
1797 if (atomic_dec_and_test(&skb->users)) {
1798 struct softnet_data *sd;
1799 unsigned long flags;
1801 local_irq_save(flags);
1802 sd = &__get_cpu_var(softnet_data);
1803 skb->next = sd->completion_queue;
1804 sd->completion_queue = skb;
1805 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1806 local_irq_restore(flags);
1809 EXPORT_SYMBOL(dev_kfree_skb_irq);
1811 void dev_kfree_skb_any(struct sk_buff *skb)
1813 if (in_irq() || irqs_disabled())
1814 dev_kfree_skb_irq(skb);
1818 EXPORT_SYMBOL(dev_kfree_skb_any);
1822 * netif_device_detach - mark device as removed
1823 * @dev: network device
1825 * Mark device as removed from system and therefore no longer available.
1827 void netif_device_detach(struct net_device *dev)
1829 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1830 netif_running(dev)) {
1831 netif_tx_stop_all_queues(dev);
1834 EXPORT_SYMBOL(netif_device_detach);
1837 * netif_device_attach - mark device as attached
1838 * @dev: network device
1840 * Mark device as attached from system and restart if needed.
1842 void netif_device_attach(struct net_device *dev)
1844 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1845 netif_running(dev)) {
1846 netif_tx_wake_all_queues(dev);
1847 __netdev_watchdog_up(dev);
1850 EXPORT_SYMBOL(netif_device_attach);
1853 * skb_dev_set -- assign a new device to a buffer
1854 * @skb: buffer for the new device
1855 * @dev: network device
1857 * If an skb is owned by a device already, we have to reset
1858 * all data private to the namespace a device belongs to
1859 * before assigning it a new device.
1861 #ifdef CONFIG_NET_NS
1862 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1865 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1868 skb_init_secmark(skb);
1872 skb->ipvs_property = 0;
1873 #ifdef CONFIG_NET_SCHED
1879 EXPORT_SYMBOL(skb_set_dev);
1880 #endif /* CONFIG_NET_NS */
1882 static void skb_warn_bad_offload(const struct sk_buff *skb)
1884 static const netdev_features_t null_features = 0;
1885 struct net_device *dev = skb->dev;
1886 const char *driver = "";
1888 if (dev && dev->dev.parent)
1889 driver = dev_driver_string(dev->dev.parent);
1891 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1892 "gso_type=%d ip_summed=%d\n",
1893 driver, dev ? &dev->features : &null_features,
1894 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1895 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1896 skb_shinfo(skb)->gso_type, skb->ip_summed);
1900 * Invalidate hardware checksum when packet is to be mangled, and
1901 * complete checksum manually on outgoing path.
1903 int skb_checksum_help(struct sk_buff *skb)
1906 int ret = 0, offset;
1908 if (skb->ip_summed == CHECKSUM_COMPLETE)
1909 goto out_set_summed;
1911 if (unlikely(skb_shinfo(skb)->gso_size)) {
1912 skb_warn_bad_offload(skb);
1916 offset = skb_checksum_start_offset(skb);
1917 BUG_ON(offset >= skb_headlen(skb));
1918 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1920 offset += skb->csum_offset;
1921 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1923 if (skb_cloned(skb) &&
1924 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1925 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1930 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1932 skb->ip_summed = CHECKSUM_NONE;
1936 EXPORT_SYMBOL(skb_checksum_help);
1939 * skb_gso_segment - Perform segmentation on skb.
1940 * @skb: buffer to segment
1941 * @features: features for the output path (see dev->features)
1943 * This function segments the given skb and returns a list of segments.
1945 * It may return NULL if the skb requires no segmentation. This is
1946 * only possible when GSO is used for verifying header integrity.
1948 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1949 netdev_features_t features)
1951 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1952 struct packet_type *ptype;
1953 __be16 type = skb->protocol;
1954 int vlan_depth = ETH_HLEN;
1957 while (type == htons(ETH_P_8021Q)) {
1958 struct vlan_hdr *vh;
1960 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1961 return ERR_PTR(-EINVAL);
1963 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1964 type = vh->h_vlan_encapsulated_proto;
1965 vlan_depth += VLAN_HLEN;
1968 skb_reset_mac_header(skb);
1969 skb->mac_len = skb->network_header - skb->mac_header;
1970 __skb_pull(skb, skb->mac_len);
1972 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1973 skb_warn_bad_offload(skb);
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 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2013 EXPORT_SYMBOL(netdev_rx_csum_fault);
2016 /* Actually, we should eliminate this check as soon as we know, that:
2017 * 1. IOMMU is present and allows to map all the memory.
2018 * 2. No high memory really exists on this machine.
2021 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2023 #ifdef CONFIG_HIGHMEM
2025 if (!(dev->features & NETIF_F_HIGHDMA)) {
2026 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2027 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2028 if (PageHighMem(skb_frag_page(frag)))
2033 if (PCI_DMA_BUS_IS_PHYS) {
2034 struct device *pdev = dev->dev.parent;
2038 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2039 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2040 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2041 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2050 void (*destructor)(struct sk_buff *skb);
2053 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2055 static void dev_gso_skb_destructor(struct sk_buff *skb)
2057 struct dev_gso_cb *cb;
2060 struct sk_buff *nskb = skb->next;
2062 skb->next = nskb->next;
2065 } while (skb->next);
2067 cb = DEV_GSO_CB(skb);
2069 cb->destructor(skb);
2073 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2074 * @skb: buffer to segment
2075 * @features: device features as applicable to this skb
2077 * This function segments the given skb and stores the list of segments
2080 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2082 struct sk_buff *segs;
2084 segs = skb_gso_segment(skb, features);
2086 /* Verifying header integrity only. */
2091 return PTR_ERR(segs);
2094 DEV_GSO_CB(skb)->destructor = skb->destructor;
2095 skb->destructor = dev_gso_skb_destructor;
2101 * Try to orphan skb early, right before transmission by the device.
2102 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2103 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2105 static inline void skb_orphan_try(struct sk_buff *skb)
2107 struct sock *sk = skb->sk;
2109 if (sk && !skb_shinfo(skb)->tx_flags) {
2110 /* skb_tx_hash() wont be able to get sk.
2111 * We copy sk_hash into skb->rxhash
2114 skb->rxhash = sk->sk_hash;
2119 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2121 return ((features & NETIF_F_GEN_CSUM) ||
2122 ((features & NETIF_F_V4_CSUM) &&
2123 protocol == htons(ETH_P_IP)) ||
2124 ((features & NETIF_F_V6_CSUM) &&
2125 protocol == htons(ETH_P_IPV6)) ||
2126 ((features & NETIF_F_FCOE_CRC) &&
2127 protocol == htons(ETH_P_FCOE)));
2130 static netdev_features_t harmonize_features(struct sk_buff *skb,
2131 __be16 protocol, netdev_features_t features)
2133 if (!can_checksum_protocol(features, protocol)) {
2134 features &= ~NETIF_F_ALL_CSUM;
2135 features &= ~NETIF_F_SG;
2136 } else if (illegal_highdma(skb->dev, skb)) {
2137 features &= ~NETIF_F_SG;
2143 netdev_features_t netif_skb_features(struct sk_buff *skb)
2145 __be16 protocol = skb->protocol;
2146 netdev_features_t features = skb->dev->features;
2148 if (protocol == htons(ETH_P_8021Q)) {
2149 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2150 protocol = veh->h_vlan_encapsulated_proto;
2151 } else if (!vlan_tx_tag_present(skb)) {
2152 return harmonize_features(skb, protocol, features);
2155 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2157 if (protocol != htons(ETH_P_8021Q)) {
2158 return harmonize_features(skb, protocol, features);
2160 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2161 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2162 return harmonize_features(skb, protocol, features);
2165 EXPORT_SYMBOL(netif_skb_features);
2168 * Returns true if either:
2169 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2170 * 2. skb is fragmented and the device does not support SG, or if
2171 * at least one of fragments is in highmem and device does not
2172 * support DMA from it.
2174 static inline int skb_needs_linearize(struct sk_buff *skb,
2177 return skb_is_nonlinear(skb) &&
2178 ((skb_has_frag_list(skb) &&
2179 !(features & NETIF_F_FRAGLIST)) ||
2180 (skb_shinfo(skb)->nr_frags &&
2181 !(features & NETIF_F_SG)));
2184 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2185 struct netdev_queue *txq)
2187 const struct net_device_ops *ops = dev->netdev_ops;
2188 int rc = NETDEV_TX_OK;
2189 unsigned int skb_len;
2191 if (likely(!skb->next)) {
2192 netdev_features_t features;
2195 * If device doesn't need skb->dst, release it right now while
2196 * its hot in this cpu cache
2198 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2201 if (!list_empty(&ptype_all))
2202 dev_queue_xmit_nit(skb, dev);
2204 skb_orphan_try(skb);
2206 features = netif_skb_features(skb);
2208 if (vlan_tx_tag_present(skb) &&
2209 !(features & NETIF_F_HW_VLAN_TX)) {
2210 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2217 if (netif_needs_gso(skb, features)) {
2218 if (unlikely(dev_gso_segment(skb, features)))
2223 if (skb_needs_linearize(skb, features) &&
2224 __skb_linearize(skb))
2227 /* If packet is not checksummed and device does not
2228 * support checksumming for this protocol, complete
2229 * checksumming here.
2231 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2232 skb_set_transport_header(skb,
2233 skb_checksum_start_offset(skb));
2234 if (!(features & NETIF_F_ALL_CSUM) &&
2235 skb_checksum_help(skb))
2241 rc = ops->ndo_start_xmit(skb, dev);
2242 trace_net_dev_xmit(skb, rc, dev, skb_len);
2243 if (rc == NETDEV_TX_OK)
2244 txq_trans_update(txq);
2250 struct sk_buff *nskb = skb->next;
2252 skb->next = nskb->next;
2256 * If device doesn't need nskb->dst, release it right now while
2257 * its hot in this cpu cache
2259 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2262 skb_len = nskb->len;
2263 rc = ops->ndo_start_xmit(nskb, dev);
2264 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2265 if (unlikely(rc != NETDEV_TX_OK)) {
2266 if (rc & ~NETDEV_TX_MASK)
2267 goto out_kfree_gso_skb;
2268 nskb->next = skb->next;
2272 txq_trans_update(txq);
2273 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2274 return NETDEV_TX_BUSY;
2275 } while (skb->next);
2278 if (likely(skb->next == NULL))
2279 skb->destructor = DEV_GSO_CB(skb)->destructor;
2286 static u32 hashrnd __read_mostly;
2289 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2290 * to be used as a distribution range.
2292 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2293 unsigned int num_tx_queues)
2297 u16 qcount = num_tx_queues;
2299 if (skb_rx_queue_recorded(skb)) {
2300 hash = skb_get_rx_queue(skb);
2301 while (unlikely(hash >= num_tx_queues))
2302 hash -= num_tx_queues;
2307 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2308 qoffset = dev->tc_to_txq[tc].offset;
2309 qcount = dev->tc_to_txq[tc].count;
2312 if (skb->sk && skb->sk->sk_hash)
2313 hash = skb->sk->sk_hash;
2315 hash = (__force u16) skb->protocol ^ skb->rxhash;
2316 hash = jhash_1word(hash, hashrnd);
2318 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2320 EXPORT_SYMBOL(__skb_tx_hash);
2322 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2324 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2325 if (net_ratelimit()) {
2326 pr_warn("%s selects TX queue %d, but real number of TX queues is %d\n",
2327 dev->name, queue_index,
2328 dev->real_num_tx_queues);
2335 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2338 struct xps_dev_maps *dev_maps;
2339 struct xps_map *map;
2340 int queue_index = -1;
2343 dev_maps = rcu_dereference(dev->xps_maps);
2345 map = rcu_dereference(
2346 dev_maps->cpu_map[raw_smp_processor_id()]);
2349 queue_index = map->queues[0];
2352 if (skb->sk && skb->sk->sk_hash)
2353 hash = skb->sk->sk_hash;
2355 hash = (__force u16) skb->protocol ^
2357 hash = jhash_1word(hash, hashrnd);
2358 queue_index = map->queues[
2359 ((u64)hash * map->len) >> 32];
2361 if (unlikely(queue_index >= dev->real_num_tx_queues))
2373 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2374 struct sk_buff *skb)
2377 const struct net_device_ops *ops = dev->netdev_ops;
2379 if (dev->real_num_tx_queues == 1)
2381 else if (ops->ndo_select_queue) {
2382 queue_index = ops->ndo_select_queue(dev, skb);
2383 queue_index = dev_cap_txqueue(dev, queue_index);
2385 struct sock *sk = skb->sk;
2386 queue_index = sk_tx_queue_get(sk);
2388 if (queue_index < 0 || skb->ooo_okay ||
2389 queue_index >= dev->real_num_tx_queues) {
2390 int old_index = queue_index;
2392 queue_index = get_xps_queue(dev, skb);
2393 if (queue_index < 0)
2394 queue_index = skb_tx_hash(dev, skb);
2396 if (queue_index != old_index && sk) {
2397 struct dst_entry *dst =
2398 rcu_dereference_check(sk->sk_dst_cache, 1);
2400 if (dst && skb_dst(skb) == dst)
2401 sk_tx_queue_set(sk, queue_index);
2406 skb_set_queue_mapping(skb, queue_index);
2407 return netdev_get_tx_queue(dev, queue_index);
2410 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2411 struct net_device *dev,
2412 struct netdev_queue *txq)
2414 spinlock_t *root_lock = qdisc_lock(q);
2418 qdisc_skb_cb(skb)->pkt_len = skb->len;
2419 qdisc_calculate_pkt_len(skb, q);
2421 * Heuristic to force contended enqueues to serialize on a
2422 * separate lock before trying to get qdisc main lock.
2423 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2424 * and dequeue packets faster.
2426 contended = qdisc_is_running(q);
2427 if (unlikely(contended))
2428 spin_lock(&q->busylock);
2430 spin_lock(root_lock);
2431 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2434 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2435 qdisc_run_begin(q)) {
2437 * This is a work-conserving queue; there are no old skbs
2438 * waiting to be sent out; and the qdisc is not running -
2439 * xmit the skb directly.
2441 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2444 qdisc_bstats_update(q, skb);
2446 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2447 if (unlikely(contended)) {
2448 spin_unlock(&q->busylock);
2455 rc = NET_XMIT_SUCCESS;
2458 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2459 if (qdisc_run_begin(q)) {
2460 if (unlikely(contended)) {
2461 spin_unlock(&q->busylock);
2467 spin_unlock(root_lock);
2468 if (unlikely(contended))
2469 spin_unlock(&q->busylock);
2473 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2474 static void skb_update_prio(struct sk_buff *skb)
2476 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2478 if ((!skb->priority) && (skb->sk) && map)
2479 skb->priority = map->priomap[skb->sk->sk_cgrp_prioidx];
2482 #define skb_update_prio(skb)
2485 static DEFINE_PER_CPU(int, xmit_recursion);
2486 #define RECURSION_LIMIT 10
2489 * dev_queue_xmit - transmit a buffer
2490 * @skb: buffer to transmit
2492 * Queue a buffer for transmission to a network device. The caller must
2493 * have set the device and priority and built the buffer before calling
2494 * this function. The function can be called from an interrupt.
2496 * A negative errno code is returned on a failure. A success does not
2497 * guarantee the frame will be transmitted as it may be dropped due
2498 * to congestion or traffic shaping.
2500 * -----------------------------------------------------------------------------------
2501 * I notice this method can also return errors from the queue disciplines,
2502 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2505 * Regardless of the return value, the skb is consumed, so it is currently
2506 * difficult to retry a send to this method. (You can bump the ref count
2507 * before sending to hold a reference for retry if you are careful.)
2509 * When calling this method, interrupts MUST be enabled. This is because
2510 * the BH enable code must have IRQs enabled so that it will not deadlock.
2513 int dev_queue_xmit(struct sk_buff *skb)
2515 struct net_device *dev = skb->dev;
2516 struct netdev_queue *txq;
2520 /* Disable soft irqs for various locks below. Also
2521 * stops preemption for RCU.
2525 skb_update_prio(skb);
2527 txq = dev_pick_tx(dev, skb);
2528 q = rcu_dereference_bh(txq->qdisc);
2530 #ifdef CONFIG_NET_CLS_ACT
2531 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2533 trace_net_dev_queue(skb);
2535 rc = __dev_xmit_skb(skb, q, dev, txq);
2539 /* The device has no queue. Common case for software devices:
2540 loopback, all the sorts of tunnels...
2542 Really, it is unlikely that netif_tx_lock protection is necessary
2543 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2545 However, it is possible, that they rely on protection
2548 Check this and shot the lock. It is not prone from deadlocks.
2549 Either shot noqueue qdisc, it is even simpler 8)
2551 if (dev->flags & IFF_UP) {
2552 int cpu = smp_processor_id(); /* ok because BHs are off */
2554 if (txq->xmit_lock_owner != cpu) {
2556 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2557 goto recursion_alert;
2559 HARD_TX_LOCK(dev, txq, cpu);
2561 if (!netif_xmit_stopped(txq)) {
2562 __this_cpu_inc(xmit_recursion);
2563 rc = dev_hard_start_xmit(skb, dev, txq);
2564 __this_cpu_dec(xmit_recursion);
2565 if (dev_xmit_complete(rc)) {
2566 HARD_TX_UNLOCK(dev, txq);
2570 HARD_TX_UNLOCK(dev, txq);
2571 if (net_ratelimit())
2572 pr_crit("Virtual device %s asks to queue packet!\n",
2575 /* Recursion is detected! It is possible,
2579 if (net_ratelimit())
2580 pr_crit("Dead loop on virtual device %s, fix it urgently!\n",
2586 rcu_read_unlock_bh();
2591 rcu_read_unlock_bh();
2594 EXPORT_SYMBOL(dev_queue_xmit);
2597 /*=======================================================================
2599 =======================================================================*/
2601 int netdev_max_backlog __read_mostly = 1000;
2602 int netdev_tstamp_prequeue __read_mostly = 1;
2603 int netdev_budget __read_mostly = 300;
2604 int weight_p __read_mostly = 64; /* old backlog weight */
2606 /* Called with irq disabled */
2607 static inline void ____napi_schedule(struct softnet_data *sd,
2608 struct napi_struct *napi)
2610 list_add_tail(&napi->poll_list, &sd->poll_list);
2611 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2615 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2616 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2617 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2618 * if hash is a canonical 4-tuple hash over transport ports.
2620 void __skb_get_rxhash(struct sk_buff *skb)
2622 struct flow_keys keys;
2625 if (!skb_flow_dissect(skb, &keys))
2629 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2630 swap(keys.port16[0], keys.port16[1]);
2634 /* get a consistent hash (same value on both flow directions) */
2635 if ((__force u32)keys.dst < (__force u32)keys.src)
2636 swap(keys.dst, keys.src);
2638 hash = jhash_3words((__force u32)keys.dst,
2639 (__force u32)keys.src,
2640 (__force u32)keys.ports, hashrnd);
2646 EXPORT_SYMBOL(__skb_get_rxhash);
2650 /* One global table that all flow-based protocols share. */
2651 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2652 EXPORT_SYMBOL(rps_sock_flow_table);
2654 struct static_key rps_needed __read_mostly;
2656 static struct rps_dev_flow *
2657 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2658 struct rps_dev_flow *rflow, u16 next_cpu)
2660 if (next_cpu != RPS_NO_CPU) {
2661 #ifdef CONFIG_RFS_ACCEL
2662 struct netdev_rx_queue *rxqueue;
2663 struct rps_dev_flow_table *flow_table;
2664 struct rps_dev_flow *old_rflow;
2669 /* Should we steer this flow to a different hardware queue? */
2670 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2671 !(dev->features & NETIF_F_NTUPLE))
2673 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2674 if (rxq_index == skb_get_rx_queue(skb))
2677 rxqueue = dev->_rx + rxq_index;
2678 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2681 flow_id = skb->rxhash & flow_table->mask;
2682 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2683 rxq_index, flow_id);
2687 rflow = &flow_table->flows[flow_id];
2689 if (old_rflow->filter == rflow->filter)
2690 old_rflow->filter = RPS_NO_FILTER;
2694 per_cpu(softnet_data, next_cpu).input_queue_head;
2697 rflow->cpu = next_cpu;
2702 * get_rps_cpu is called from netif_receive_skb and returns the target
2703 * CPU from the RPS map of the receiving queue for a given skb.
2704 * rcu_read_lock must be held on entry.
2706 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2707 struct rps_dev_flow **rflowp)
2709 struct netdev_rx_queue *rxqueue;
2710 struct rps_map *map;
2711 struct rps_dev_flow_table *flow_table;
2712 struct rps_sock_flow_table *sock_flow_table;
2716 if (skb_rx_queue_recorded(skb)) {
2717 u16 index = skb_get_rx_queue(skb);
2718 if (unlikely(index >= dev->real_num_rx_queues)) {
2719 WARN_ONCE(dev->real_num_rx_queues > 1,
2720 "%s received packet on queue %u, but number "
2721 "of RX queues is %u\n",
2722 dev->name, index, dev->real_num_rx_queues);
2725 rxqueue = dev->_rx + index;
2729 map = rcu_dereference(rxqueue->rps_map);
2731 if (map->len == 1 &&
2732 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2733 tcpu = map->cpus[0];
2734 if (cpu_online(tcpu))
2738 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2742 skb_reset_network_header(skb);
2743 if (!skb_get_rxhash(skb))
2746 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2747 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2748 if (flow_table && sock_flow_table) {
2750 struct rps_dev_flow *rflow;
2752 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2755 next_cpu = sock_flow_table->ents[skb->rxhash &
2756 sock_flow_table->mask];
2759 * If the desired CPU (where last recvmsg was done) is
2760 * different from current CPU (one in the rx-queue flow
2761 * table entry), switch if one of the following holds:
2762 * - Current CPU is unset (equal to RPS_NO_CPU).
2763 * - Current CPU is offline.
2764 * - The current CPU's queue tail has advanced beyond the
2765 * last packet that was enqueued using this table entry.
2766 * This guarantees that all previous packets for the flow
2767 * have been dequeued, thus preserving in order delivery.
2769 if (unlikely(tcpu != next_cpu) &&
2770 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2771 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2772 rflow->last_qtail)) >= 0))
2773 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2775 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2783 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2785 if (cpu_online(tcpu)) {
2795 #ifdef CONFIG_RFS_ACCEL
2798 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2799 * @dev: Device on which the filter was set
2800 * @rxq_index: RX queue index
2801 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2802 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2804 * Drivers that implement ndo_rx_flow_steer() should periodically call
2805 * this function for each installed filter and remove the filters for
2806 * which it returns %true.
2808 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2809 u32 flow_id, u16 filter_id)
2811 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2812 struct rps_dev_flow_table *flow_table;
2813 struct rps_dev_flow *rflow;
2818 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2819 if (flow_table && flow_id <= flow_table->mask) {
2820 rflow = &flow_table->flows[flow_id];
2821 cpu = ACCESS_ONCE(rflow->cpu);
2822 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2823 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2824 rflow->last_qtail) <
2825 (int)(10 * flow_table->mask)))
2831 EXPORT_SYMBOL(rps_may_expire_flow);
2833 #endif /* CONFIG_RFS_ACCEL */
2835 /* Called from hardirq (IPI) context */
2836 static void rps_trigger_softirq(void *data)
2838 struct softnet_data *sd = data;
2840 ____napi_schedule(sd, &sd->backlog);
2844 #endif /* CONFIG_RPS */
2847 * Check if this softnet_data structure is another cpu one
2848 * If yes, queue it to our IPI list and return 1
2851 static int rps_ipi_queued(struct softnet_data *sd)
2854 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2857 sd->rps_ipi_next = mysd->rps_ipi_list;
2858 mysd->rps_ipi_list = sd;
2860 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2863 #endif /* CONFIG_RPS */
2868 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2869 * queue (may be a remote CPU queue).
2871 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2872 unsigned int *qtail)
2874 struct softnet_data *sd;
2875 unsigned long flags;
2877 sd = &per_cpu(softnet_data, cpu);
2879 local_irq_save(flags);
2882 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2883 if (skb_queue_len(&sd->input_pkt_queue)) {
2885 __skb_queue_tail(&sd->input_pkt_queue, skb);
2886 input_queue_tail_incr_save(sd, qtail);
2888 local_irq_restore(flags);
2889 return NET_RX_SUCCESS;
2892 /* Schedule NAPI for backlog device
2893 * We can use non atomic operation since we own the queue lock
2895 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2896 if (!rps_ipi_queued(sd))
2897 ____napi_schedule(sd, &sd->backlog);
2905 local_irq_restore(flags);
2907 atomic_long_inc(&skb->dev->rx_dropped);
2913 * netif_rx - post buffer to the network code
2914 * @skb: buffer to post
2916 * This function receives a packet from a device driver and queues it for
2917 * the upper (protocol) levels to process. It always succeeds. The buffer
2918 * may be dropped during processing for congestion control or by the
2922 * NET_RX_SUCCESS (no congestion)
2923 * NET_RX_DROP (packet was dropped)
2927 int netif_rx(struct sk_buff *skb)
2931 /* if netpoll wants it, pretend we never saw it */
2932 if (netpoll_rx(skb))
2935 net_timestamp_check(netdev_tstamp_prequeue, skb);
2937 trace_netif_rx(skb);
2939 if (static_key_false(&rps_needed)) {
2940 struct rps_dev_flow voidflow, *rflow = &voidflow;
2946 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2948 cpu = smp_processor_id();
2950 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2958 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2963 EXPORT_SYMBOL(netif_rx);
2965 int netif_rx_ni(struct sk_buff *skb)
2970 err = netif_rx(skb);
2971 if (local_softirq_pending())
2977 EXPORT_SYMBOL(netif_rx_ni);
2979 static void net_tx_action(struct softirq_action *h)
2981 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2983 if (sd->completion_queue) {
2984 struct sk_buff *clist;
2986 local_irq_disable();
2987 clist = sd->completion_queue;
2988 sd->completion_queue = NULL;
2992 struct sk_buff *skb = clist;
2993 clist = clist->next;
2995 WARN_ON(atomic_read(&skb->users));
2996 trace_kfree_skb(skb, net_tx_action);
3001 if (sd->output_queue) {
3004 local_irq_disable();
3005 head = sd->output_queue;
3006 sd->output_queue = NULL;
3007 sd->output_queue_tailp = &sd->output_queue;
3011 struct Qdisc *q = head;
3012 spinlock_t *root_lock;
3014 head = head->next_sched;
3016 root_lock = qdisc_lock(q);
3017 if (spin_trylock(root_lock)) {
3018 smp_mb__before_clear_bit();
3019 clear_bit(__QDISC_STATE_SCHED,
3022 spin_unlock(root_lock);
3024 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3026 __netif_reschedule(q);
3028 smp_mb__before_clear_bit();
3029 clear_bit(__QDISC_STATE_SCHED,
3037 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3038 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3039 /* This hook is defined here for ATM LANE */
3040 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3041 unsigned char *addr) __read_mostly;
3042 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3045 #ifdef CONFIG_NET_CLS_ACT
3046 /* TODO: Maybe we should just force sch_ingress to be compiled in
3047 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3048 * a compare and 2 stores extra right now if we dont have it on
3049 * but have CONFIG_NET_CLS_ACT
3050 * NOTE: This doesn't stop any functionality; if you dont have
3051 * the ingress scheduler, you just can't add policies on ingress.
3054 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3056 struct net_device *dev = skb->dev;
3057 u32 ttl = G_TC_RTTL(skb->tc_verd);
3058 int result = TC_ACT_OK;
3061 if (unlikely(MAX_RED_LOOP < ttl++)) {
3062 if (net_ratelimit())
3063 pr_warn("Redir loop detected Dropping packet (%d->%d)\n",
3064 skb->skb_iif, dev->ifindex);
3068 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3069 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3072 if (q != &noop_qdisc) {
3073 spin_lock(qdisc_lock(q));
3074 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3075 result = qdisc_enqueue_root(skb, q);
3076 spin_unlock(qdisc_lock(q));
3082 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3083 struct packet_type **pt_prev,
3084 int *ret, struct net_device *orig_dev)
3086 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3088 if (!rxq || rxq->qdisc == &noop_qdisc)
3092 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3096 switch (ing_filter(skb, rxq)) {
3110 * netdev_rx_handler_register - register receive handler
3111 * @dev: device to register a handler for
3112 * @rx_handler: receive handler to register
3113 * @rx_handler_data: data pointer that is used by rx handler
3115 * Register a receive hander for a device. This handler will then be
3116 * called from __netif_receive_skb. A negative errno code is returned
3119 * The caller must hold the rtnl_mutex.
3121 * For a general description of rx_handler, see enum rx_handler_result.
3123 int netdev_rx_handler_register(struct net_device *dev,
3124 rx_handler_func_t *rx_handler,
3125 void *rx_handler_data)
3129 if (dev->rx_handler)
3132 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3133 rcu_assign_pointer(dev->rx_handler, rx_handler);
3137 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3140 * netdev_rx_handler_unregister - unregister receive handler
3141 * @dev: device to unregister a handler from
3143 * Unregister a receive hander from a device.
3145 * The caller must hold the rtnl_mutex.
3147 void netdev_rx_handler_unregister(struct net_device *dev)
3151 RCU_INIT_POINTER(dev->rx_handler, NULL);
3152 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3154 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3156 static int __netif_receive_skb(struct sk_buff *skb)
3158 struct packet_type *ptype, *pt_prev;
3159 rx_handler_func_t *rx_handler;
3160 struct net_device *orig_dev;
3161 struct net_device *null_or_dev;
3162 bool deliver_exact = false;
3163 int ret = NET_RX_DROP;
3166 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3168 trace_netif_receive_skb(skb);
3170 /* if we've gotten here through NAPI, check netpoll */
3171 if (netpoll_receive_skb(skb))
3175 skb->skb_iif = skb->dev->ifindex;
3176 orig_dev = skb->dev;
3178 skb_reset_network_header(skb);
3179 skb_reset_transport_header(skb);
3180 skb_reset_mac_len(skb);
3188 __this_cpu_inc(softnet_data.processed);
3190 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3191 skb = vlan_untag(skb);
3196 #ifdef CONFIG_NET_CLS_ACT
3197 if (skb->tc_verd & TC_NCLS) {
3198 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3203 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3204 if (!ptype->dev || ptype->dev == skb->dev) {
3206 ret = deliver_skb(skb, pt_prev, orig_dev);
3211 #ifdef CONFIG_NET_CLS_ACT
3212 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3218 rx_handler = rcu_dereference(skb->dev->rx_handler);
3219 if (vlan_tx_tag_present(skb)) {
3221 ret = deliver_skb(skb, pt_prev, orig_dev);
3224 if (vlan_do_receive(&skb, !rx_handler))
3226 else if (unlikely(!skb))
3232 ret = deliver_skb(skb, pt_prev, orig_dev);
3235 switch (rx_handler(&skb)) {
3236 case RX_HANDLER_CONSUMED:
3238 case RX_HANDLER_ANOTHER:
3240 case RX_HANDLER_EXACT:
3241 deliver_exact = true;
3242 case RX_HANDLER_PASS:
3249 /* deliver only exact match when indicated */
3250 null_or_dev = deliver_exact ? skb->dev : NULL;
3252 type = skb->protocol;
3253 list_for_each_entry_rcu(ptype,
3254 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3255 if (ptype->type == type &&
3256 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3257 ptype->dev == orig_dev)) {
3259 ret = deliver_skb(skb, pt_prev, orig_dev);
3265 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3267 atomic_long_inc(&skb->dev->rx_dropped);
3269 /* Jamal, now you will not able to escape explaining
3270 * me how you were going to use this. :-)
3281 * netif_receive_skb - process receive buffer from network
3282 * @skb: buffer to process
3284 * netif_receive_skb() is the main receive data processing function.
3285 * It always succeeds. The buffer may be dropped during processing
3286 * for congestion control or by the protocol layers.
3288 * This function may only be called from softirq context and interrupts
3289 * should be enabled.
3291 * Return values (usually ignored):
3292 * NET_RX_SUCCESS: no congestion
3293 * NET_RX_DROP: packet was dropped
3295 int netif_receive_skb(struct sk_buff *skb)
3297 net_timestamp_check(netdev_tstamp_prequeue, skb);
3299 if (skb_defer_rx_timestamp(skb))
3300 return NET_RX_SUCCESS;
3303 if (static_key_false(&rps_needed)) {
3304 struct rps_dev_flow voidflow, *rflow = &voidflow;
3309 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3312 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3319 return __netif_receive_skb(skb);
3321 EXPORT_SYMBOL(netif_receive_skb);
3323 /* Network device is going away, flush any packets still pending
3324 * Called with irqs disabled.
3326 static void flush_backlog(void *arg)
3328 struct net_device *dev = arg;
3329 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3330 struct sk_buff *skb, *tmp;
3333 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3334 if (skb->dev == dev) {
3335 __skb_unlink(skb, &sd->input_pkt_queue);
3337 input_queue_head_incr(sd);
3342 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3343 if (skb->dev == dev) {
3344 __skb_unlink(skb, &sd->process_queue);
3346 input_queue_head_incr(sd);
3351 static int napi_gro_complete(struct sk_buff *skb)
3353 struct packet_type *ptype;
3354 __be16 type = skb->protocol;
3355 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3358 if (NAPI_GRO_CB(skb)->count == 1) {
3359 skb_shinfo(skb)->gso_size = 0;
3364 list_for_each_entry_rcu(ptype, head, list) {
3365 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3368 err = ptype->gro_complete(skb);
3374 WARN_ON(&ptype->list == head);
3376 return NET_RX_SUCCESS;
3380 return netif_receive_skb(skb);
3383 inline void napi_gro_flush(struct napi_struct *napi)
3385 struct sk_buff *skb, *next;
3387 for (skb = napi->gro_list; skb; skb = next) {
3390 napi_gro_complete(skb);
3393 napi->gro_count = 0;
3394 napi->gro_list = NULL;
3396 EXPORT_SYMBOL(napi_gro_flush);
3398 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3400 struct sk_buff **pp = NULL;
3401 struct packet_type *ptype;
3402 __be16 type = skb->protocol;
3403 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3406 enum gro_result ret;
3408 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3411 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3415 list_for_each_entry_rcu(ptype, head, list) {
3416 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3419 skb_set_network_header(skb, skb_gro_offset(skb));
3420 mac_len = skb->network_header - skb->mac_header;
3421 skb->mac_len = mac_len;
3422 NAPI_GRO_CB(skb)->same_flow = 0;
3423 NAPI_GRO_CB(skb)->flush = 0;
3424 NAPI_GRO_CB(skb)->free = 0;
3426 pp = ptype->gro_receive(&napi->gro_list, skb);
3431 if (&ptype->list == head)
3434 same_flow = NAPI_GRO_CB(skb)->same_flow;
3435 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3438 struct sk_buff *nskb = *pp;
3442 napi_gro_complete(nskb);
3449 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3453 NAPI_GRO_CB(skb)->count = 1;
3454 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3455 skb->next = napi->gro_list;
3456 napi->gro_list = skb;
3460 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3461 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3463 BUG_ON(skb->end - skb->tail < grow);
3465 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3468 skb->data_len -= grow;
3470 skb_shinfo(skb)->frags[0].page_offset += grow;
3471 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3473 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3474 skb_frag_unref(skb, 0);
3475 memmove(skb_shinfo(skb)->frags,
3476 skb_shinfo(skb)->frags + 1,
3477 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3488 EXPORT_SYMBOL(dev_gro_receive);
3490 static inline gro_result_t
3491 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3494 unsigned int maclen = skb->dev->hard_header_len;
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 if (maclen == ETH_HLEN)
3502 diffs |= compare_ether_header(skb_mac_header(p),
3503 skb_gro_mac_header(skb));
3505 diffs = memcmp(skb_mac_header(p),
3506 skb_gro_mac_header(skb),
3508 NAPI_GRO_CB(p)->same_flow = !diffs;
3509 NAPI_GRO_CB(p)->flush = 0;
3512 return dev_gro_receive(napi, skb);
3515 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3519 if (netif_receive_skb(skb))
3524 case GRO_MERGED_FREE:
3535 EXPORT_SYMBOL(napi_skb_finish);
3537 void skb_gro_reset_offset(struct sk_buff *skb)
3539 NAPI_GRO_CB(skb)->data_offset = 0;
3540 NAPI_GRO_CB(skb)->frag0 = NULL;
3541 NAPI_GRO_CB(skb)->frag0_len = 0;
3543 if (skb->mac_header == skb->tail &&
3544 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3545 NAPI_GRO_CB(skb)->frag0 =
3546 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3547 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3550 EXPORT_SYMBOL(skb_gro_reset_offset);
3552 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3554 skb_gro_reset_offset(skb);
3556 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3558 EXPORT_SYMBOL(napi_gro_receive);
3560 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3562 __skb_pull(skb, skb_headlen(skb));
3563 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3565 skb->dev = napi->dev;
3571 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3573 struct sk_buff *skb = napi->skb;
3576 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3582 EXPORT_SYMBOL(napi_get_frags);
3584 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3590 skb->protocol = eth_type_trans(skb, skb->dev);
3592 if (ret == GRO_HELD)
3593 skb_gro_pull(skb, -ETH_HLEN);
3594 else if (netif_receive_skb(skb))
3599 case GRO_MERGED_FREE:
3600 napi_reuse_skb(napi, skb);
3609 EXPORT_SYMBOL(napi_frags_finish);
3611 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3613 struct sk_buff *skb = napi->skb;
3620 skb_reset_mac_header(skb);
3621 skb_gro_reset_offset(skb);
3623 off = skb_gro_offset(skb);
3624 hlen = off + sizeof(*eth);
3625 eth = skb_gro_header_fast(skb, off);
3626 if (skb_gro_header_hard(skb, hlen)) {
3627 eth = skb_gro_header_slow(skb, hlen, off);
3628 if (unlikely(!eth)) {
3629 napi_reuse_skb(napi, skb);
3635 skb_gro_pull(skb, sizeof(*eth));
3638 * This works because the only protocols we care about don't require
3639 * special handling. We'll fix it up properly at the end.
3641 skb->protocol = eth->h_proto;
3646 EXPORT_SYMBOL(napi_frags_skb);
3648 gro_result_t napi_gro_frags(struct napi_struct *napi)
3650 struct sk_buff *skb = napi_frags_skb(napi);
3655 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3657 EXPORT_SYMBOL(napi_gro_frags);
3660 * net_rps_action sends any pending IPI's for rps.
3661 * Note: called with local irq disabled, but exits with local irq enabled.
3663 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3666 struct softnet_data *remsd = sd->rps_ipi_list;
3669 sd->rps_ipi_list = NULL;
3673 /* Send pending IPI's to kick RPS processing on remote cpus. */
3675 struct softnet_data *next = remsd->rps_ipi_next;
3677 if (cpu_online(remsd->cpu))
3678 __smp_call_function_single(remsd->cpu,
3687 static int process_backlog(struct napi_struct *napi, int quota)
3690 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3693 /* Check if we have pending ipi, its better to send them now,
3694 * not waiting net_rx_action() end.
3696 if (sd->rps_ipi_list) {
3697 local_irq_disable();
3698 net_rps_action_and_irq_enable(sd);
3701 napi->weight = weight_p;
3702 local_irq_disable();
3703 while (work < quota) {
3704 struct sk_buff *skb;
3707 while ((skb = __skb_dequeue(&sd->process_queue))) {
3709 __netif_receive_skb(skb);
3710 local_irq_disable();
3711 input_queue_head_incr(sd);
3712 if (++work >= quota) {
3719 qlen = skb_queue_len(&sd->input_pkt_queue);
3721 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3722 &sd->process_queue);
3724 if (qlen < quota - work) {
3726 * Inline a custom version of __napi_complete().
3727 * only current cpu owns and manipulates this napi,
3728 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3729 * we can use a plain write instead of clear_bit(),
3730 * and we dont need an smp_mb() memory barrier.
3732 list_del(&napi->poll_list);
3735 quota = work + qlen;
3745 * __napi_schedule - schedule for receive
3746 * @n: entry to schedule
3748 * The entry's receive function will be scheduled to run
3750 void __napi_schedule(struct napi_struct *n)
3752 unsigned long flags;
3754 local_irq_save(flags);
3755 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3756 local_irq_restore(flags);
3758 EXPORT_SYMBOL(__napi_schedule);
3760 void __napi_complete(struct napi_struct *n)
3762 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3763 BUG_ON(n->gro_list);
3765 list_del(&n->poll_list);
3766 smp_mb__before_clear_bit();
3767 clear_bit(NAPI_STATE_SCHED, &n->state);
3769 EXPORT_SYMBOL(__napi_complete);
3771 void napi_complete(struct napi_struct *n)
3773 unsigned long flags;
3776 * don't let napi dequeue from the cpu poll list
3777 * just in case its running on a different cpu
3779 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3783 local_irq_save(flags);
3785 local_irq_restore(flags);
3787 EXPORT_SYMBOL(napi_complete);
3789 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3790 int (*poll)(struct napi_struct *, int), int weight)
3792 INIT_LIST_HEAD(&napi->poll_list);
3793 napi->gro_count = 0;
3794 napi->gro_list = NULL;
3797 napi->weight = weight;
3798 list_add(&napi->dev_list, &dev->napi_list);
3800 #ifdef CONFIG_NETPOLL
3801 spin_lock_init(&napi->poll_lock);
3802 napi->poll_owner = -1;
3804 set_bit(NAPI_STATE_SCHED, &napi->state);
3806 EXPORT_SYMBOL(netif_napi_add);
3808 void netif_napi_del(struct napi_struct *napi)
3810 struct sk_buff *skb, *next;
3812 list_del_init(&napi->dev_list);
3813 napi_free_frags(napi);
3815 for (skb = napi->gro_list; skb; skb = next) {
3821 napi->gro_list = NULL;
3822 napi->gro_count = 0;
3824 EXPORT_SYMBOL(netif_napi_del);
3826 static void net_rx_action(struct softirq_action *h)
3828 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3829 unsigned long time_limit = jiffies + 2;
3830 int budget = netdev_budget;
3833 local_irq_disable();
3835 while (!list_empty(&sd->poll_list)) {
3836 struct napi_struct *n;
3839 /* If softirq window is exhuasted then punt.
3840 * Allow this to run for 2 jiffies since which will allow
3841 * an average latency of 1.5/HZ.
3843 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3848 /* Even though interrupts have been re-enabled, this
3849 * access is safe because interrupts can only add new
3850 * entries to the tail of this list, and only ->poll()
3851 * calls can remove this head entry from the list.
3853 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3855 have = netpoll_poll_lock(n);
3859 /* This NAPI_STATE_SCHED test is for avoiding a race
3860 * with netpoll's poll_napi(). Only the entity which
3861 * obtains the lock and sees NAPI_STATE_SCHED set will
3862 * actually make the ->poll() call. Therefore we avoid
3863 * accidentally calling ->poll() when NAPI is not scheduled.
3866 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3867 work = n->poll(n, weight);
3871 WARN_ON_ONCE(work > weight);
3875 local_irq_disable();
3877 /* Drivers must not modify the NAPI state if they
3878 * consume the entire weight. In such cases this code
3879 * still "owns" the NAPI instance and therefore can
3880 * move the instance around on the list at-will.
3882 if (unlikely(work == weight)) {
3883 if (unlikely(napi_disable_pending(n))) {
3886 local_irq_disable();
3888 list_move_tail(&n->poll_list, &sd->poll_list);
3891 netpoll_poll_unlock(have);
3894 net_rps_action_and_irq_enable(sd);
3896 #ifdef CONFIG_NET_DMA
3898 * There may not be any more sk_buffs coming right now, so push
3899 * any pending DMA copies to hardware
3901 dma_issue_pending_all();
3908 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3912 static gifconf_func_t *gifconf_list[NPROTO];
3915 * register_gifconf - register a SIOCGIF handler
3916 * @family: Address family
3917 * @gifconf: Function handler
3919 * Register protocol dependent address dumping routines. The handler
3920 * that is passed must not be freed or reused until it has been replaced
3921 * by another handler.
3923 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3925 if (family >= NPROTO)
3927 gifconf_list[family] = gifconf;
3930 EXPORT_SYMBOL(register_gifconf);
3934 * Map an interface index to its name (SIOCGIFNAME)
3938 * We need this ioctl for efficient implementation of the
3939 * if_indextoname() function required by the IPv6 API. Without
3940 * it, we would have to search all the interfaces to find a
3944 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3946 struct net_device *dev;
3950 * Fetch the caller's info block.
3953 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3957 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3963 strcpy(ifr.ifr_name, dev->name);
3966 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3972 * Perform a SIOCGIFCONF call. This structure will change
3973 * size eventually, and there is nothing I can do about it.
3974 * Thus we will need a 'compatibility mode'.
3977 static int dev_ifconf(struct net *net, char __user *arg)
3980 struct net_device *dev;
3987 * Fetch the caller's info block.
3990 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3997 * Loop over the interfaces, and write an info block for each.
4001 for_each_netdev(net, dev) {
4002 for (i = 0; i < NPROTO; i++) {
4003 if (gifconf_list[i]) {
4006 done = gifconf_list[i](dev, NULL, 0);
4008 done = gifconf_list[i](dev, pos + total,
4018 * All done. Write the updated control block back to the caller.
4020 ifc.ifc_len = total;
4023 * Both BSD and Solaris return 0 here, so we do too.
4025 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4028 #ifdef CONFIG_PROC_FS
4030 #define BUCKET_SPACE (32 - NETDEV_HASHBITS)
4032 struct dev_iter_state {
4033 struct seq_net_private p;
4034 unsigned int pos; /* bucket << BUCKET_SPACE + offset */
4037 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4038 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4039 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4041 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq)
4043 struct dev_iter_state *state = seq->private;
4044 struct net *net = seq_file_net(seq);
4045 struct net_device *dev;
4046 struct hlist_node *p;
4047 struct hlist_head *h;
4048 unsigned int count, bucket, offset;
4050 bucket = get_bucket(state->pos);
4051 offset = get_offset(state->pos);
4052 h = &net->dev_name_head[bucket];
4054 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4055 if (count++ == offset) {
4056 state->pos = set_bucket_offset(bucket, count);
4064 static inline struct net_device *dev_from_new_bucket(struct seq_file *seq)
4066 struct dev_iter_state *state = seq->private;
4067 struct net_device *dev;
4068 unsigned int bucket;
4070 bucket = get_bucket(state->pos);
4072 dev = dev_from_same_bucket(seq);
4077 state->pos = set_bucket_offset(bucket, 0);
4078 } while (bucket < NETDEV_HASHENTRIES);
4084 * This is invoked by the /proc filesystem handler to display a device
4087 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4090 struct dev_iter_state *state = seq->private;
4094 return SEQ_START_TOKEN;
4096 /* check for end of the hash */
4097 if (state->pos == 0 && *pos > 1)
4100 return dev_from_new_bucket(seq);
4103 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4105 struct net_device *dev;
4109 if (v == SEQ_START_TOKEN)
4110 return dev_from_new_bucket(seq);
4112 dev = dev_from_same_bucket(seq);
4116 return dev_from_new_bucket(seq);
4119 void dev_seq_stop(struct seq_file *seq, void *v)
4125 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4127 struct rtnl_link_stats64 temp;
4128 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4130 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4131 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4132 dev->name, stats->rx_bytes, stats->rx_packets,
4134 stats->rx_dropped + stats->rx_missed_errors,
4135 stats->rx_fifo_errors,
4136 stats->rx_length_errors + stats->rx_over_errors +
4137 stats->rx_crc_errors + stats->rx_frame_errors,
4138 stats->rx_compressed, stats->multicast,
4139 stats->tx_bytes, stats->tx_packets,
4140 stats->tx_errors, stats->tx_dropped,
4141 stats->tx_fifo_errors, stats->collisions,
4142 stats->tx_carrier_errors +
4143 stats->tx_aborted_errors +
4144 stats->tx_window_errors +
4145 stats->tx_heartbeat_errors,
4146 stats->tx_compressed);
4150 * Called from the PROCfs module. This now uses the new arbitrary sized
4151 * /proc/net interface to create /proc/net/dev
4153 static int dev_seq_show(struct seq_file *seq, void *v)
4155 if (v == SEQ_START_TOKEN)
4156 seq_puts(seq, "Inter-| Receive "
4158 " face |bytes packets errs drop fifo frame "
4159 "compressed multicast|bytes packets errs "
4160 "drop fifo colls carrier compressed\n");
4162 dev_seq_printf_stats(seq, v);
4166 static struct softnet_data *softnet_get_online(loff_t *pos)
4168 struct softnet_data *sd = NULL;
4170 while (*pos < nr_cpu_ids)
4171 if (cpu_online(*pos)) {
4172 sd = &per_cpu(softnet_data, *pos);
4179 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4181 return softnet_get_online(pos);
4184 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4187 return softnet_get_online(pos);
4190 static void softnet_seq_stop(struct seq_file *seq, void *v)
4194 static int softnet_seq_show(struct seq_file *seq, void *v)
4196 struct softnet_data *sd = v;
4198 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4199 sd->processed, sd->dropped, sd->time_squeeze, 0,
4200 0, 0, 0, 0, /* was fastroute */
4201 sd->cpu_collision, sd->received_rps);
4205 static const struct seq_operations dev_seq_ops = {
4206 .start = dev_seq_start,
4207 .next = dev_seq_next,
4208 .stop = dev_seq_stop,
4209 .show = dev_seq_show,
4212 static int dev_seq_open(struct inode *inode, struct file *file)
4214 return seq_open_net(inode, file, &dev_seq_ops,
4215 sizeof(struct dev_iter_state));
4218 int dev_seq_open_ops(struct inode *inode, struct file *file,
4219 const struct seq_operations *ops)
4221 return seq_open_net(inode, file, ops, sizeof(struct dev_iter_state));
4224 static const struct file_operations dev_seq_fops = {
4225 .owner = THIS_MODULE,
4226 .open = dev_seq_open,
4228 .llseek = seq_lseek,
4229 .release = seq_release_net,
4232 static const struct seq_operations softnet_seq_ops = {
4233 .start = softnet_seq_start,
4234 .next = softnet_seq_next,
4235 .stop = softnet_seq_stop,
4236 .show = softnet_seq_show,
4239 static int softnet_seq_open(struct inode *inode, struct file *file)
4241 return seq_open(file, &softnet_seq_ops);
4244 static const struct file_operations softnet_seq_fops = {
4245 .owner = THIS_MODULE,
4246 .open = softnet_seq_open,
4248 .llseek = seq_lseek,
4249 .release = seq_release,
4252 static void *ptype_get_idx(loff_t pos)
4254 struct packet_type *pt = NULL;
4258 list_for_each_entry_rcu(pt, &ptype_all, list) {
4264 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4265 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4274 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4278 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4281 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4283 struct packet_type *pt;
4284 struct list_head *nxt;
4288 if (v == SEQ_START_TOKEN)
4289 return ptype_get_idx(0);
4292 nxt = pt->list.next;
4293 if (pt->type == htons(ETH_P_ALL)) {
4294 if (nxt != &ptype_all)
4297 nxt = ptype_base[0].next;
4299 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4301 while (nxt == &ptype_base[hash]) {
4302 if (++hash >= PTYPE_HASH_SIZE)
4304 nxt = ptype_base[hash].next;
4307 return list_entry(nxt, struct packet_type, list);
4310 static void ptype_seq_stop(struct seq_file *seq, void *v)
4316 static int ptype_seq_show(struct seq_file *seq, void *v)
4318 struct packet_type *pt = v;
4320 if (v == SEQ_START_TOKEN)
4321 seq_puts(seq, "Type Device Function\n");
4322 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4323 if (pt->type == htons(ETH_P_ALL))
4324 seq_puts(seq, "ALL ");
4326 seq_printf(seq, "%04x", ntohs(pt->type));
4328 seq_printf(seq, " %-8s %pF\n",
4329 pt->dev ? pt->dev->name : "", pt->func);
4335 static const struct seq_operations ptype_seq_ops = {
4336 .start = ptype_seq_start,
4337 .next = ptype_seq_next,
4338 .stop = ptype_seq_stop,
4339 .show = ptype_seq_show,
4342 static int ptype_seq_open(struct inode *inode, struct file *file)
4344 return seq_open_net(inode, file, &ptype_seq_ops,
4345 sizeof(struct seq_net_private));
4348 static const struct file_operations ptype_seq_fops = {
4349 .owner = THIS_MODULE,
4350 .open = ptype_seq_open,
4352 .llseek = seq_lseek,
4353 .release = seq_release_net,
4357 static int __net_init dev_proc_net_init(struct net *net)
4361 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4363 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4365 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4368 if (wext_proc_init(net))
4374 proc_net_remove(net, "ptype");
4376 proc_net_remove(net, "softnet_stat");
4378 proc_net_remove(net, "dev");
4382 static void __net_exit dev_proc_net_exit(struct net *net)
4384 wext_proc_exit(net);
4386 proc_net_remove(net, "ptype");
4387 proc_net_remove(net, "softnet_stat");
4388 proc_net_remove(net, "dev");
4391 static struct pernet_operations __net_initdata dev_proc_ops = {
4392 .init = dev_proc_net_init,
4393 .exit = dev_proc_net_exit,
4396 static int __init dev_proc_init(void)
4398 return register_pernet_subsys(&dev_proc_ops);
4401 #define dev_proc_init() 0
4402 #endif /* CONFIG_PROC_FS */
4406 * netdev_set_master - set up master pointer
4407 * @slave: slave device
4408 * @master: new master device
4410 * Changes the master device of the slave. Pass %NULL to break the
4411 * bonding. The caller must hold the RTNL semaphore. On a failure
4412 * a negative errno code is returned. On success the reference counts
4413 * are adjusted and the function returns zero.
4415 int netdev_set_master(struct net_device *slave, struct net_device *master)
4417 struct net_device *old = slave->master;
4427 slave->master = master;
4433 EXPORT_SYMBOL(netdev_set_master);
4436 * netdev_set_bond_master - set up bonding master/slave pair
4437 * @slave: slave device
4438 * @master: new master device
4440 * Changes the master device of the slave. Pass %NULL to break the
4441 * bonding. The caller must hold the RTNL semaphore. On a failure
4442 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4443 * to the routing socket and the function returns zero.
4445 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4451 err = netdev_set_master(slave, master);
4455 slave->flags |= IFF_SLAVE;
4457 slave->flags &= ~IFF_SLAVE;
4459 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4462 EXPORT_SYMBOL(netdev_set_bond_master);
4464 static void dev_change_rx_flags(struct net_device *dev, int flags)
4466 const struct net_device_ops *ops = dev->netdev_ops;
4468 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4469 ops->ndo_change_rx_flags(dev, flags);
4472 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4474 unsigned int old_flags = dev->flags;
4480 dev->flags |= IFF_PROMISC;
4481 dev->promiscuity += inc;
4482 if (dev->promiscuity == 0) {
4485 * If inc causes overflow, untouch promisc and return error.
4488 dev->flags &= ~IFF_PROMISC;
4490 dev->promiscuity -= inc;
4491 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4496 if (dev->flags != old_flags) {
4497 pr_info("device %s %s promiscuous mode\n",
4499 dev->flags & IFF_PROMISC ? "entered" : "left");
4500 if (audit_enabled) {
4501 current_uid_gid(&uid, &gid);
4502 audit_log(current->audit_context, GFP_ATOMIC,
4503 AUDIT_ANOM_PROMISCUOUS,
4504 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4505 dev->name, (dev->flags & IFF_PROMISC),
4506 (old_flags & IFF_PROMISC),
4507 audit_get_loginuid(current),
4509 audit_get_sessionid(current));
4512 dev_change_rx_flags(dev, IFF_PROMISC);
4518 * dev_set_promiscuity - update promiscuity count on a device
4522 * Add or remove promiscuity from a device. While the count in the device
4523 * remains above zero the interface remains promiscuous. Once it hits zero
4524 * the device reverts back to normal filtering operation. A negative inc
4525 * value is used to drop promiscuity on the device.
4526 * Return 0 if successful or a negative errno code on error.
4528 int dev_set_promiscuity(struct net_device *dev, int inc)
4530 unsigned int old_flags = dev->flags;
4533 err = __dev_set_promiscuity(dev, inc);
4536 if (dev->flags != old_flags)
4537 dev_set_rx_mode(dev);
4540 EXPORT_SYMBOL(dev_set_promiscuity);
4543 * dev_set_allmulti - update allmulti count on a device
4547 * Add or remove reception of all multicast frames to a device. While the
4548 * count in the device remains above zero the interface remains listening
4549 * to all interfaces. Once it hits zero the device reverts back to normal
4550 * filtering operation. A negative @inc value is used to drop the counter
4551 * when releasing a resource needing all multicasts.
4552 * Return 0 if successful or a negative errno code on error.
4555 int dev_set_allmulti(struct net_device *dev, int inc)
4557 unsigned int old_flags = dev->flags;
4561 dev->flags |= IFF_ALLMULTI;
4562 dev->allmulti += inc;
4563 if (dev->allmulti == 0) {
4566 * If inc causes overflow, untouch allmulti and return error.
4569 dev->flags &= ~IFF_ALLMULTI;
4571 dev->allmulti -= inc;
4572 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4577 if (dev->flags ^ old_flags) {
4578 dev_change_rx_flags(dev, IFF_ALLMULTI);
4579 dev_set_rx_mode(dev);
4583 EXPORT_SYMBOL(dev_set_allmulti);
4586 * Upload unicast and multicast address lists to device and
4587 * configure RX filtering. When the device doesn't support unicast
4588 * filtering it is put in promiscuous mode while unicast addresses
4591 void __dev_set_rx_mode(struct net_device *dev)
4593 const struct net_device_ops *ops = dev->netdev_ops;
4595 /* dev_open will call this function so the list will stay sane. */
4596 if (!(dev->flags&IFF_UP))
4599 if (!netif_device_present(dev))
4602 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4603 /* Unicast addresses changes may only happen under the rtnl,
4604 * therefore calling __dev_set_promiscuity here is safe.
4606 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4607 __dev_set_promiscuity(dev, 1);
4608 dev->uc_promisc = true;
4609 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4610 __dev_set_promiscuity(dev, -1);
4611 dev->uc_promisc = false;
4615 if (ops->ndo_set_rx_mode)
4616 ops->ndo_set_rx_mode(dev);
4619 void dev_set_rx_mode(struct net_device *dev)
4621 netif_addr_lock_bh(dev);
4622 __dev_set_rx_mode(dev);
4623 netif_addr_unlock_bh(dev);
4627 * dev_get_flags - get flags reported to userspace
4630 * Get the combination of flag bits exported through APIs to userspace.
4632 unsigned dev_get_flags(const struct net_device *dev)
4636 flags = (dev->flags & ~(IFF_PROMISC |
4641 (dev->gflags & (IFF_PROMISC |
4644 if (netif_running(dev)) {
4645 if (netif_oper_up(dev))
4646 flags |= IFF_RUNNING;
4647 if (netif_carrier_ok(dev))
4648 flags |= IFF_LOWER_UP;
4649 if (netif_dormant(dev))
4650 flags |= IFF_DORMANT;
4655 EXPORT_SYMBOL(dev_get_flags);
4657 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4659 unsigned int old_flags = dev->flags;
4665 * Set the flags on our device.
4668 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4669 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4671 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4675 * Load in the correct multicast list now the flags have changed.
4678 if ((old_flags ^ flags) & IFF_MULTICAST)
4679 dev_change_rx_flags(dev, IFF_MULTICAST);
4681 dev_set_rx_mode(dev);
4684 * Have we downed the interface. We handle IFF_UP ourselves
4685 * according to user attempts to set it, rather than blindly
4690 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4691 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4694 dev_set_rx_mode(dev);
4697 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4698 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4700 dev->gflags ^= IFF_PROMISC;
4701 dev_set_promiscuity(dev, inc);
4704 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4705 is important. Some (broken) drivers set IFF_PROMISC, when
4706 IFF_ALLMULTI is requested not asking us and not reporting.
4708 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4709 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4711 dev->gflags ^= IFF_ALLMULTI;
4712 dev_set_allmulti(dev, inc);
4718 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4720 unsigned int changes = dev->flags ^ old_flags;
4722 if (changes & IFF_UP) {
4723 if (dev->flags & IFF_UP)
4724 call_netdevice_notifiers(NETDEV_UP, dev);
4726 call_netdevice_notifiers(NETDEV_DOWN, dev);
4729 if (dev->flags & IFF_UP &&
4730 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4731 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4735 * dev_change_flags - change device settings
4737 * @flags: device state flags
4739 * Change settings on device based state flags. The flags are
4740 * in the userspace exported format.
4742 int dev_change_flags(struct net_device *dev, unsigned int flags)
4745 unsigned int changes, old_flags = dev->flags;
4747 ret = __dev_change_flags(dev, flags);
4751 changes = old_flags ^ dev->flags;
4753 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4755 __dev_notify_flags(dev, old_flags);
4758 EXPORT_SYMBOL(dev_change_flags);
4761 * dev_set_mtu - Change maximum transfer unit
4763 * @new_mtu: new transfer unit
4765 * Change the maximum transfer size of the network device.
4767 int dev_set_mtu(struct net_device *dev, int new_mtu)
4769 const struct net_device_ops *ops = dev->netdev_ops;
4772 if (new_mtu == dev->mtu)
4775 /* MTU must be positive. */
4779 if (!netif_device_present(dev))
4783 if (ops->ndo_change_mtu)
4784 err = ops->ndo_change_mtu(dev, new_mtu);
4788 if (!err && dev->flags & IFF_UP)
4789 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4792 EXPORT_SYMBOL(dev_set_mtu);
4795 * dev_set_group - Change group this device belongs to
4797 * @new_group: group this device should belong to
4799 void dev_set_group(struct net_device *dev, int new_group)
4801 dev->group = new_group;
4803 EXPORT_SYMBOL(dev_set_group);
4806 * dev_set_mac_address - Change Media Access Control Address
4810 * Change the hardware (MAC) address of the device
4812 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4814 const struct net_device_ops *ops = dev->netdev_ops;
4817 if (!ops->ndo_set_mac_address)
4819 if (sa->sa_family != dev->type)
4821 if (!netif_device_present(dev))
4823 err = ops->ndo_set_mac_address(dev, sa);
4825 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4828 EXPORT_SYMBOL(dev_set_mac_address);
4831 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4833 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4836 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4842 case SIOCGIFFLAGS: /* Get interface flags */
4843 ifr->ifr_flags = (short) dev_get_flags(dev);
4846 case SIOCGIFMETRIC: /* Get the metric on the interface
4847 (currently unused) */
4848 ifr->ifr_metric = 0;
4851 case SIOCGIFMTU: /* Get the MTU of a device */
4852 ifr->ifr_mtu = dev->mtu;
4857 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4859 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4860 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4861 ifr->ifr_hwaddr.sa_family = dev->type;
4869 ifr->ifr_map.mem_start = dev->mem_start;
4870 ifr->ifr_map.mem_end = dev->mem_end;
4871 ifr->ifr_map.base_addr = dev->base_addr;
4872 ifr->ifr_map.irq = dev->irq;
4873 ifr->ifr_map.dma = dev->dma;
4874 ifr->ifr_map.port = dev->if_port;
4878 ifr->ifr_ifindex = dev->ifindex;
4882 ifr->ifr_qlen = dev->tx_queue_len;
4886 /* dev_ioctl() should ensure this case
4898 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4900 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4903 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4904 const struct net_device_ops *ops;
4909 ops = dev->netdev_ops;
4912 case SIOCSIFFLAGS: /* Set interface flags */
4913 return dev_change_flags(dev, ifr->ifr_flags);
4915 case SIOCSIFMETRIC: /* Set the metric on the interface
4916 (currently unused) */
4919 case SIOCSIFMTU: /* Set the MTU of a device */
4920 return dev_set_mtu(dev, ifr->ifr_mtu);
4923 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4925 case SIOCSIFHWBROADCAST:
4926 if (ifr->ifr_hwaddr.sa_family != dev->type)
4928 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4929 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4930 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4934 if (ops->ndo_set_config) {
4935 if (!netif_device_present(dev))
4937 return ops->ndo_set_config(dev, &ifr->ifr_map);
4942 if (!ops->ndo_set_rx_mode ||
4943 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4945 if (!netif_device_present(dev))
4947 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4950 if (!ops->ndo_set_rx_mode ||
4951 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4953 if (!netif_device_present(dev))
4955 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4958 if (ifr->ifr_qlen < 0)
4960 dev->tx_queue_len = ifr->ifr_qlen;
4964 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4965 return dev_change_name(dev, ifr->ifr_newname);
4968 err = net_hwtstamp_validate(ifr);
4974 * Unknown or private ioctl
4977 if ((cmd >= SIOCDEVPRIVATE &&
4978 cmd <= SIOCDEVPRIVATE + 15) ||
4979 cmd == SIOCBONDENSLAVE ||
4980 cmd == SIOCBONDRELEASE ||
4981 cmd == SIOCBONDSETHWADDR ||
4982 cmd == SIOCBONDSLAVEINFOQUERY ||
4983 cmd == SIOCBONDINFOQUERY ||
4984 cmd == SIOCBONDCHANGEACTIVE ||
4985 cmd == SIOCGMIIPHY ||
4986 cmd == SIOCGMIIREG ||
4987 cmd == SIOCSMIIREG ||
4988 cmd == SIOCBRADDIF ||
4989 cmd == SIOCBRDELIF ||
4990 cmd == SIOCSHWTSTAMP ||
4991 cmd == SIOCWANDEV) {
4993 if (ops->ndo_do_ioctl) {
4994 if (netif_device_present(dev))
4995 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5007 * This function handles all "interface"-type I/O control requests. The actual
5008 * 'doing' part of this is dev_ifsioc above.
5012 * dev_ioctl - network device ioctl
5013 * @net: the applicable net namespace
5014 * @cmd: command to issue
5015 * @arg: pointer to a struct ifreq in user space
5017 * Issue ioctl functions to devices. This is normally called by the
5018 * user space syscall interfaces but can sometimes be useful for
5019 * other purposes. The return value is the return from the syscall if
5020 * positive or a negative errno code on error.
5023 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5029 /* One special case: SIOCGIFCONF takes ifconf argument
5030 and requires shared lock, because it sleeps writing
5034 if (cmd == SIOCGIFCONF) {
5036 ret = dev_ifconf(net, (char __user *) arg);
5040 if (cmd == SIOCGIFNAME)
5041 return dev_ifname(net, (struct ifreq __user *)arg);
5043 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5046 ifr.ifr_name[IFNAMSIZ-1] = 0;
5048 colon = strchr(ifr.ifr_name, ':');
5053 * See which interface the caller is talking about.
5058 * These ioctl calls:
5059 * - can be done by all.
5060 * - atomic and do not require locking.
5071 dev_load(net, ifr.ifr_name);
5073 ret = dev_ifsioc_locked(net, &ifr, cmd);
5078 if (copy_to_user(arg, &ifr,
5079 sizeof(struct ifreq)))
5085 dev_load(net, ifr.ifr_name);
5087 ret = dev_ethtool(net, &ifr);
5092 if (copy_to_user(arg, &ifr,
5093 sizeof(struct ifreq)))
5099 * These ioctl calls:
5100 * - require superuser power.
5101 * - require strict serialization.
5107 if (!capable(CAP_NET_ADMIN))
5109 dev_load(net, ifr.ifr_name);
5111 ret = dev_ifsioc(net, &ifr, cmd);
5116 if (copy_to_user(arg, &ifr,
5117 sizeof(struct ifreq)))
5123 * These ioctl calls:
5124 * - require superuser power.
5125 * - require strict serialization.
5126 * - do not return a value
5136 case SIOCSIFHWBROADCAST:
5139 case SIOCBONDENSLAVE:
5140 case SIOCBONDRELEASE:
5141 case SIOCBONDSETHWADDR:
5142 case SIOCBONDCHANGEACTIVE:
5146 if (!capable(CAP_NET_ADMIN))
5149 case SIOCBONDSLAVEINFOQUERY:
5150 case SIOCBONDINFOQUERY:
5151 dev_load(net, ifr.ifr_name);
5153 ret = dev_ifsioc(net, &ifr, cmd);
5158 /* Get the per device memory space. We can add this but
5159 * currently do not support it */
5161 /* Set the per device memory buffer space.
5162 * Not applicable in our case */
5167 * Unknown or private ioctl.
5170 if (cmd == SIOCWANDEV ||
5171 (cmd >= SIOCDEVPRIVATE &&
5172 cmd <= SIOCDEVPRIVATE + 15)) {
5173 dev_load(net, ifr.ifr_name);
5175 ret = dev_ifsioc(net, &ifr, cmd);
5177 if (!ret && copy_to_user(arg, &ifr,
5178 sizeof(struct ifreq)))
5182 /* Take care of Wireless Extensions */
5183 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5184 return wext_handle_ioctl(net, &ifr, cmd, arg);
5191 * dev_new_index - allocate an ifindex
5192 * @net: the applicable net namespace
5194 * Returns a suitable unique value for a new device interface
5195 * number. The caller must hold the rtnl semaphore or the
5196 * dev_base_lock to be sure it remains unique.
5198 static int dev_new_index(struct net *net)
5204 if (!__dev_get_by_index(net, ifindex))
5209 /* Delayed registration/unregisteration */
5210 static LIST_HEAD(net_todo_list);
5212 static void net_set_todo(struct net_device *dev)
5214 list_add_tail(&dev->todo_list, &net_todo_list);
5217 static void rollback_registered_many(struct list_head *head)
5219 struct net_device *dev, *tmp;
5221 BUG_ON(dev_boot_phase);
5224 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5225 /* Some devices call without registering
5226 * for initialization unwind. Remove those
5227 * devices and proceed with the remaining.
5229 if (dev->reg_state == NETREG_UNINITIALIZED) {
5230 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5234 list_del(&dev->unreg_list);
5237 dev->dismantle = true;
5238 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5241 /* If device is running, close it first. */
5242 dev_close_many(head);
5244 list_for_each_entry(dev, head, unreg_list) {
5245 /* And unlink it from device chain. */
5246 unlist_netdevice(dev);
5248 dev->reg_state = NETREG_UNREGISTERING;
5253 list_for_each_entry(dev, head, unreg_list) {
5254 /* Shutdown queueing discipline. */
5258 /* Notify protocols, that we are about to destroy
5259 this device. They should clean all the things.
5261 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5263 if (!dev->rtnl_link_ops ||
5264 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5265 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5268 * Flush the unicast and multicast chains
5273 if (dev->netdev_ops->ndo_uninit)
5274 dev->netdev_ops->ndo_uninit(dev);
5276 /* Notifier chain MUST detach us from master device. */
5277 WARN_ON(dev->master);
5279 /* Remove entries from kobject tree */
5280 netdev_unregister_kobject(dev);
5283 /* Process any work delayed until the end of the batch */
5284 dev = list_first_entry(head, struct net_device, unreg_list);
5285 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5289 list_for_each_entry(dev, head, unreg_list)
5293 static void rollback_registered(struct net_device *dev)
5297 list_add(&dev->unreg_list, &single);
5298 rollback_registered_many(&single);
5302 static netdev_features_t netdev_fix_features(struct net_device *dev,
5303 netdev_features_t features)
5305 /* Fix illegal checksum combinations */
5306 if ((features & NETIF_F_HW_CSUM) &&
5307 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5308 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5309 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5312 /* Fix illegal SG+CSUM combinations. */
5313 if ((features & NETIF_F_SG) &&
5314 !(features & NETIF_F_ALL_CSUM)) {
5316 "Dropping NETIF_F_SG since no checksum feature.\n");
5317 features &= ~NETIF_F_SG;
5320 /* TSO requires that SG is present as well. */
5321 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5322 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5323 features &= ~NETIF_F_ALL_TSO;
5326 /* TSO ECN requires that TSO is present as well. */
5327 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5328 features &= ~NETIF_F_TSO_ECN;
5330 /* Software GSO depends on SG. */
5331 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5332 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5333 features &= ~NETIF_F_GSO;
5336 /* UFO needs SG and checksumming */
5337 if (features & NETIF_F_UFO) {
5338 /* maybe split UFO into V4 and V6? */
5339 if (!((features & NETIF_F_GEN_CSUM) ||
5340 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5341 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5343 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5344 features &= ~NETIF_F_UFO;
5347 if (!(features & NETIF_F_SG)) {
5349 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5350 features &= ~NETIF_F_UFO;
5357 int __netdev_update_features(struct net_device *dev)
5359 netdev_features_t features;
5364 features = netdev_get_wanted_features(dev);
5366 if (dev->netdev_ops->ndo_fix_features)
5367 features = dev->netdev_ops->ndo_fix_features(dev, features);
5369 /* driver might be less strict about feature dependencies */
5370 features = netdev_fix_features(dev, features);
5372 if (dev->features == features)
5375 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5376 &dev->features, &features);
5378 if (dev->netdev_ops->ndo_set_features)
5379 err = dev->netdev_ops->ndo_set_features(dev, features);
5381 if (unlikely(err < 0)) {
5383 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5384 err, &features, &dev->features);
5389 dev->features = features;
5395 * netdev_update_features - recalculate device features
5396 * @dev: the device to check
5398 * Recalculate dev->features set and send notifications if it
5399 * has changed. Should be called after driver or hardware dependent
5400 * conditions might have changed that influence the features.
5402 void netdev_update_features(struct net_device *dev)
5404 if (__netdev_update_features(dev))
5405 netdev_features_change(dev);
5407 EXPORT_SYMBOL(netdev_update_features);
5410 * netdev_change_features - recalculate device features
5411 * @dev: the device to check
5413 * Recalculate dev->features set and send notifications even
5414 * if they have not changed. Should be called instead of
5415 * netdev_update_features() if also dev->vlan_features might
5416 * have changed to allow the changes to be propagated to stacked
5419 void netdev_change_features(struct net_device *dev)
5421 __netdev_update_features(dev);
5422 netdev_features_change(dev);
5424 EXPORT_SYMBOL(netdev_change_features);
5427 * netif_stacked_transfer_operstate - transfer operstate
5428 * @rootdev: the root or lower level device to transfer state from
5429 * @dev: the device to transfer operstate to
5431 * Transfer operational state from root to device. This is normally
5432 * called when a stacking relationship exists between the root
5433 * device and the device(a leaf device).
5435 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5436 struct net_device *dev)
5438 if (rootdev->operstate == IF_OPER_DORMANT)
5439 netif_dormant_on(dev);
5441 netif_dormant_off(dev);
5443 if (netif_carrier_ok(rootdev)) {
5444 if (!netif_carrier_ok(dev))
5445 netif_carrier_on(dev);
5447 if (netif_carrier_ok(dev))
5448 netif_carrier_off(dev);
5451 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5454 static int netif_alloc_rx_queues(struct net_device *dev)
5456 unsigned int i, count = dev->num_rx_queues;
5457 struct netdev_rx_queue *rx;
5461 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5463 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5468 for (i = 0; i < count; i++)
5474 static void netdev_init_one_queue(struct net_device *dev,
5475 struct netdev_queue *queue, void *_unused)
5477 /* Initialize queue lock */
5478 spin_lock_init(&queue->_xmit_lock);
5479 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5480 queue->xmit_lock_owner = -1;
5481 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5484 dql_init(&queue->dql, HZ);
5488 static int netif_alloc_netdev_queues(struct net_device *dev)
5490 unsigned int count = dev->num_tx_queues;
5491 struct netdev_queue *tx;
5495 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5497 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5502 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5503 spin_lock_init(&dev->tx_global_lock);
5509 * register_netdevice - register a network device
5510 * @dev: device to register
5512 * Take a completed network device structure and add it to the kernel
5513 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5514 * chain. 0 is returned on success. A negative errno code is returned
5515 * on a failure to set up the device, or if the name is a duplicate.
5517 * Callers must hold the rtnl semaphore. You may want
5518 * register_netdev() instead of this.
5521 * The locking appears insufficient to guarantee two parallel registers
5522 * will not get the same name.
5525 int register_netdevice(struct net_device *dev)
5528 struct net *net = dev_net(dev);
5530 BUG_ON(dev_boot_phase);
5535 /* When net_device's are persistent, this will be fatal. */
5536 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5539 spin_lock_init(&dev->addr_list_lock);
5540 netdev_set_addr_lockdep_class(dev);
5544 ret = dev_get_valid_name(dev, dev->name);
5548 /* Init, if this function is available */
5549 if (dev->netdev_ops->ndo_init) {
5550 ret = dev->netdev_ops->ndo_init(dev);
5558 dev->ifindex = dev_new_index(net);
5559 if (dev->iflink == -1)
5560 dev->iflink = dev->ifindex;
5562 /* Transfer changeable features to wanted_features and enable
5563 * software offloads (GSO and GRO).
5565 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5566 dev->features |= NETIF_F_SOFT_FEATURES;
5567 dev->wanted_features = dev->features & dev->hw_features;
5569 /* Turn on no cache copy if HW is doing checksum */
5570 if (!(dev->flags & IFF_LOOPBACK)) {
5571 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5572 if (dev->features & NETIF_F_ALL_CSUM) {
5573 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5574 dev->features |= NETIF_F_NOCACHE_COPY;
5578 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5580 dev->vlan_features |= NETIF_F_HIGHDMA;
5582 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5583 ret = notifier_to_errno(ret);
5587 ret = netdev_register_kobject(dev);
5590 dev->reg_state = NETREG_REGISTERED;
5592 __netdev_update_features(dev);
5595 * Default initial state at registry is that the
5596 * device is present.
5599 set_bit(__LINK_STATE_PRESENT, &dev->state);
5601 dev_init_scheduler(dev);
5603 list_netdevice(dev);
5605 /* Notify protocols, that a new device appeared. */
5606 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5607 ret = notifier_to_errno(ret);
5609 rollback_registered(dev);
5610 dev->reg_state = NETREG_UNREGISTERED;
5613 * Prevent userspace races by waiting until the network
5614 * device is fully setup before sending notifications.
5616 if (!dev->rtnl_link_ops ||
5617 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5618 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5624 if (dev->netdev_ops->ndo_uninit)
5625 dev->netdev_ops->ndo_uninit(dev);
5628 EXPORT_SYMBOL(register_netdevice);
5631 * init_dummy_netdev - init a dummy network device for NAPI
5632 * @dev: device to init
5634 * This takes a network device structure and initialize the minimum
5635 * amount of fields so it can be used to schedule NAPI polls without
5636 * registering a full blown interface. This is to be used by drivers
5637 * that need to tie several hardware interfaces to a single NAPI
5638 * poll scheduler due to HW limitations.
5640 int init_dummy_netdev(struct net_device *dev)
5642 /* Clear everything. Note we don't initialize spinlocks
5643 * are they aren't supposed to be taken by any of the
5644 * NAPI code and this dummy netdev is supposed to be
5645 * only ever used for NAPI polls
5647 memset(dev, 0, sizeof(struct net_device));
5649 /* make sure we BUG if trying to hit standard
5650 * register/unregister code path
5652 dev->reg_state = NETREG_DUMMY;
5654 /* NAPI wants this */
5655 INIT_LIST_HEAD(&dev->napi_list);
5657 /* a dummy interface is started by default */
5658 set_bit(__LINK_STATE_PRESENT, &dev->state);
5659 set_bit(__LINK_STATE_START, &dev->state);
5661 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5662 * because users of this 'device' dont need to change
5668 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5672 * register_netdev - register a network device
5673 * @dev: device to register
5675 * Take a completed network device structure and add it to the kernel
5676 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5677 * chain. 0 is returned on success. A negative errno code is returned
5678 * on a failure to set up the device, or if the name is a duplicate.
5680 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5681 * and expands the device name if you passed a format string to
5684 int register_netdev(struct net_device *dev)
5689 err = register_netdevice(dev);
5693 EXPORT_SYMBOL(register_netdev);
5695 int netdev_refcnt_read(const struct net_device *dev)
5699 for_each_possible_cpu(i)
5700 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5703 EXPORT_SYMBOL(netdev_refcnt_read);
5706 * netdev_wait_allrefs - wait until all references are gone.
5708 * This is called when unregistering network devices.
5710 * Any protocol or device that holds a reference should register
5711 * for netdevice notification, and cleanup and put back the
5712 * reference if they receive an UNREGISTER event.
5713 * We can get stuck here if buggy protocols don't correctly
5716 static void netdev_wait_allrefs(struct net_device *dev)
5718 unsigned long rebroadcast_time, warning_time;
5721 linkwatch_forget_dev(dev);
5723 rebroadcast_time = warning_time = jiffies;
5724 refcnt = netdev_refcnt_read(dev);
5726 while (refcnt != 0) {
5727 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5730 /* Rebroadcast unregister notification */
5731 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5732 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5733 * should have already handle it the first time */
5735 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5737 /* We must not have linkwatch events
5738 * pending on unregister. If this
5739 * happens, we simply run the queue
5740 * unscheduled, resulting in a noop
5743 linkwatch_run_queue();
5748 rebroadcast_time = jiffies;
5753 refcnt = netdev_refcnt_read(dev);
5755 if (time_after(jiffies, warning_time + 10 * HZ)) {
5756 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
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 pr_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 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++)
5854 EXPORT_SYMBOL(netdev_stats_to_stats64);
5857 * dev_get_stats - get network device statistics
5858 * @dev: device to get statistics from
5859 * @storage: place to store stats
5861 * Get network statistics from device. Return @storage.
5862 * The device driver may provide its own method by setting
5863 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5864 * otherwise the internal statistics structure is used.
5866 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5867 struct rtnl_link_stats64 *storage)
5869 const struct net_device_ops *ops = dev->netdev_ops;
5871 if (ops->ndo_get_stats64) {
5872 memset(storage, 0, sizeof(*storage));
5873 ops->ndo_get_stats64(dev, storage);
5874 } else if (ops->ndo_get_stats) {
5875 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5877 netdev_stats_to_stats64(storage, &dev->stats);
5879 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5882 EXPORT_SYMBOL(dev_get_stats);
5884 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5886 struct netdev_queue *queue = dev_ingress_queue(dev);
5888 #ifdef CONFIG_NET_CLS_ACT
5891 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5894 netdev_init_one_queue(dev, queue, NULL);
5895 queue->qdisc = &noop_qdisc;
5896 queue->qdisc_sleeping = &noop_qdisc;
5897 rcu_assign_pointer(dev->ingress_queue, queue);
5903 * alloc_netdev_mqs - allocate network device
5904 * @sizeof_priv: size of private data to allocate space for
5905 * @name: device name format string
5906 * @setup: callback to initialize device
5907 * @txqs: the number of TX subqueues to allocate
5908 * @rxqs: the number of RX subqueues to allocate
5910 * Allocates a struct net_device with private data area for driver use
5911 * and performs basic initialization. Also allocates subquue structs
5912 * for each queue on the device.
5914 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5915 void (*setup)(struct net_device *),
5916 unsigned int txqs, unsigned int rxqs)
5918 struct net_device *dev;
5920 struct net_device *p;
5922 BUG_ON(strlen(name) >= sizeof(dev->name));
5925 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5931 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5936 alloc_size = sizeof(struct net_device);
5938 /* ensure 32-byte alignment of private area */
5939 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5940 alloc_size += sizeof_priv;
5942 /* ensure 32-byte alignment of whole construct */
5943 alloc_size += NETDEV_ALIGN - 1;
5945 p = kzalloc(alloc_size, GFP_KERNEL);
5947 pr_err("alloc_netdev: Unable to allocate device\n");
5951 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5952 dev->padded = (char *)dev - (char *)p;
5954 dev->pcpu_refcnt = alloc_percpu(int);
5955 if (!dev->pcpu_refcnt)
5958 if (dev_addr_init(dev))
5964 dev_net_set(dev, &init_net);
5966 dev->gso_max_size = GSO_MAX_SIZE;
5968 INIT_LIST_HEAD(&dev->napi_list);
5969 INIT_LIST_HEAD(&dev->unreg_list);
5970 INIT_LIST_HEAD(&dev->link_watch_list);
5971 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5974 dev->num_tx_queues = txqs;
5975 dev->real_num_tx_queues = txqs;
5976 if (netif_alloc_netdev_queues(dev))
5980 dev->num_rx_queues = rxqs;
5981 dev->real_num_rx_queues = rxqs;
5982 if (netif_alloc_rx_queues(dev))
5986 strcpy(dev->name, name);
5987 dev->group = INIT_NETDEV_GROUP;
5995 free_percpu(dev->pcpu_refcnt);
6005 EXPORT_SYMBOL(alloc_netdev_mqs);
6008 * free_netdev - free network device
6011 * This function does the last stage of destroying an allocated device
6012 * interface. The reference to the device object is released.
6013 * If this is the last reference then it will be freed.
6015 void free_netdev(struct net_device *dev)
6017 struct napi_struct *p, *n;
6019 release_net(dev_net(dev));
6026 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6028 /* Flush device addresses */
6029 dev_addr_flush(dev);
6031 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6034 free_percpu(dev->pcpu_refcnt);
6035 dev->pcpu_refcnt = NULL;
6037 /* Compatibility with error handling in drivers */
6038 if (dev->reg_state == NETREG_UNINITIALIZED) {
6039 kfree((char *)dev - dev->padded);
6043 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6044 dev->reg_state = NETREG_RELEASED;
6046 /* will free via device release */
6047 put_device(&dev->dev);
6049 EXPORT_SYMBOL(free_netdev);
6052 * synchronize_net - Synchronize with packet receive processing
6054 * Wait for packets currently being received to be done.
6055 * Does not block later packets from starting.
6057 void synchronize_net(void)
6060 if (rtnl_is_locked())
6061 synchronize_rcu_expedited();
6065 EXPORT_SYMBOL(synchronize_net);
6068 * unregister_netdevice_queue - remove device from the kernel
6072 * This function shuts down a device interface and removes it
6073 * from the kernel tables.
6074 * If head not NULL, device is queued to be unregistered later.
6076 * Callers must hold the rtnl semaphore. You may want
6077 * unregister_netdev() instead of this.
6080 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6085 list_move_tail(&dev->unreg_list, head);
6087 rollback_registered(dev);
6088 /* Finish processing unregister after unlock */
6092 EXPORT_SYMBOL(unregister_netdevice_queue);
6095 * unregister_netdevice_many - unregister many devices
6096 * @head: list of devices
6098 void unregister_netdevice_many(struct list_head *head)
6100 struct net_device *dev;
6102 if (!list_empty(head)) {
6103 rollback_registered_many(head);
6104 list_for_each_entry(dev, head, unreg_list)
6108 EXPORT_SYMBOL(unregister_netdevice_many);
6111 * unregister_netdev - remove device from the kernel
6114 * This function shuts down a device interface and removes it
6115 * from the kernel tables.
6117 * This is just a wrapper for unregister_netdevice that takes
6118 * the rtnl semaphore. In general you want to use this and not
6119 * unregister_netdevice.
6121 void unregister_netdev(struct net_device *dev)
6124 unregister_netdevice(dev);
6127 EXPORT_SYMBOL(unregister_netdev);
6130 * dev_change_net_namespace - move device to different nethost namespace
6132 * @net: network namespace
6133 * @pat: If not NULL name pattern to try if the current device name
6134 * is already taken in the destination network namespace.
6136 * This function shuts down a device interface and moves it
6137 * to a new network namespace. On success 0 is returned, on
6138 * a failure a netagive errno code is returned.
6140 * Callers must hold the rtnl semaphore.
6143 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6149 /* Don't allow namespace local devices to be moved. */
6151 if (dev->features & NETIF_F_NETNS_LOCAL)
6154 /* Ensure the device has been registrered */
6156 if (dev->reg_state != NETREG_REGISTERED)
6159 /* Get out if there is nothing todo */
6161 if (net_eq(dev_net(dev), net))
6164 /* Pick the destination device name, and ensure
6165 * we can use it in the destination network namespace.
6168 if (__dev_get_by_name(net, dev->name)) {
6169 /* We get here if we can't use the current device name */
6172 if (dev_get_valid_name(dev, pat) < 0)
6177 * And now a mini version of register_netdevice unregister_netdevice.
6180 /* If device is running close it first. */
6183 /* And unlink it from device chain */
6185 unlist_netdevice(dev);
6189 /* Shutdown queueing discipline. */
6192 /* Notify protocols, that we are about to destroy
6193 this device. They should clean all the things.
6195 Note that dev->reg_state stays at NETREG_REGISTERED.
6196 This is wanted because this way 8021q and macvlan know
6197 the device is just moving and can keep their slaves up.
6199 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6200 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6201 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6204 * Flush the unicast and multicast chains
6209 /* Actually switch the network namespace */
6210 dev_net_set(dev, net);
6212 /* If there is an ifindex conflict assign a new one */
6213 if (__dev_get_by_index(net, dev->ifindex)) {
6214 int iflink = (dev->iflink == dev->ifindex);
6215 dev->ifindex = dev_new_index(net);
6217 dev->iflink = dev->ifindex;
6220 /* Fixup kobjects */
6221 err = device_rename(&dev->dev, dev->name);
6224 /* Add the device back in the hashes */
6225 list_netdevice(dev);
6227 /* Notify protocols, that a new device appeared. */
6228 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6231 * Prevent userspace races by waiting until the network
6232 * device is fully setup before sending notifications.
6234 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6241 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6243 static int dev_cpu_callback(struct notifier_block *nfb,
6244 unsigned long action,
6247 struct sk_buff **list_skb;
6248 struct sk_buff *skb;
6249 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6250 struct softnet_data *sd, *oldsd;
6252 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6255 local_irq_disable();
6256 cpu = smp_processor_id();
6257 sd = &per_cpu(softnet_data, cpu);
6258 oldsd = &per_cpu(softnet_data, oldcpu);
6260 /* Find end of our completion_queue. */
6261 list_skb = &sd->completion_queue;
6263 list_skb = &(*list_skb)->next;
6264 /* Append completion queue from offline CPU. */
6265 *list_skb = oldsd->completion_queue;
6266 oldsd->completion_queue = NULL;
6268 /* Append output queue from offline CPU. */
6269 if (oldsd->output_queue) {
6270 *sd->output_queue_tailp = oldsd->output_queue;
6271 sd->output_queue_tailp = oldsd->output_queue_tailp;
6272 oldsd->output_queue = NULL;
6273 oldsd->output_queue_tailp = &oldsd->output_queue;
6275 /* Append NAPI poll list from offline CPU. */
6276 if (!list_empty(&oldsd->poll_list)) {
6277 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6278 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6281 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6284 /* Process offline CPU's input_pkt_queue */
6285 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6287 input_queue_head_incr(oldsd);
6289 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6291 input_queue_head_incr(oldsd);
6299 * netdev_increment_features - increment feature set by one
6300 * @all: current feature set
6301 * @one: new feature set
6302 * @mask: mask feature set
6304 * Computes a new feature set after adding a device with feature set
6305 * @one to the master device with current feature set @all. Will not
6306 * enable anything that is off in @mask. Returns the new feature set.
6308 netdev_features_t netdev_increment_features(netdev_features_t all,
6309 netdev_features_t one, netdev_features_t mask)
6311 if (mask & NETIF_F_GEN_CSUM)
6312 mask |= NETIF_F_ALL_CSUM;
6313 mask |= NETIF_F_VLAN_CHALLENGED;
6315 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6316 all &= one | ~NETIF_F_ALL_FOR_ALL;
6318 /* If one device supports hw checksumming, set for all. */
6319 if (all & NETIF_F_GEN_CSUM)
6320 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6324 EXPORT_SYMBOL(netdev_increment_features);
6326 static struct hlist_head *netdev_create_hash(void)
6329 struct hlist_head *hash;
6331 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6333 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6334 INIT_HLIST_HEAD(&hash[i]);
6339 /* Initialize per network namespace state */
6340 static int __net_init netdev_init(struct net *net)
6342 INIT_LIST_HEAD(&net->dev_base_head);
6344 net->dev_name_head = netdev_create_hash();
6345 if (net->dev_name_head == NULL)
6348 net->dev_index_head = netdev_create_hash();
6349 if (net->dev_index_head == NULL)
6355 kfree(net->dev_name_head);
6361 * netdev_drivername - network driver for the device
6362 * @dev: network device
6364 * Determine network driver for device.
6366 const char *netdev_drivername(const struct net_device *dev)
6368 const struct device_driver *driver;
6369 const struct device *parent;
6370 const char *empty = "";
6372 parent = dev->dev.parent;
6376 driver = parent->driver;
6377 if (driver && driver->name)
6378 return driver->name;
6382 int __netdev_printk(const char *level, const struct net_device *dev,
6383 struct va_format *vaf)
6387 if (dev && dev->dev.parent)
6388 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6389 netdev_name(dev), vaf);
6391 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6393 r = printk("%s(NULL net_device): %pV", level, vaf);
6397 EXPORT_SYMBOL(__netdev_printk);
6399 int netdev_printk(const char *level, const struct net_device *dev,
6400 const char *format, ...)
6402 struct va_format vaf;
6406 va_start(args, format);
6411 r = __netdev_printk(level, dev, &vaf);
6416 EXPORT_SYMBOL(netdev_printk);
6418 #define define_netdev_printk_level(func, level) \
6419 int func(const struct net_device *dev, const char *fmt, ...) \
6422 struct va_format vaf; \
6425 va_start(args, fmt); \
6430 r = __netdev_printk(level, dev, &vaf); \
6435 EXPORT_SYMBOL(func);
6437 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6438 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6439 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6440 define_netdev_printk_level(netdev_err, KERN_ERR);
6441 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6442 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6443 define_netdev_printk_level(netdev_info, KERN_INFO);
6445 static void __net_exit netdev_exit(struct net *net)
6447 kfree(net->dev_name_head);
6448 kfree(net->dev_index_head);
6451 static struct pernet_operations __net_initdata netdev_net_ops = {
6452 .init = netdev_init,
6453 .exit = netdev_exit,
6456 static void __net_exit default_device_exit(struct net *net)
6458 struct net_device *dev, *aux;
6460 * Push all migratable network devices back to the
6461 * initial network namespace
6464 for_each_netdev_safe(net, dev, aux) {
6466 char fb_name[IFNAMSIZ];
6468 /* Ignore unmoveable devices (i.e. loopback) */
6469 if (dev->features & NETIF_F_NETNS_LOCAL)
6472 /* Leave virtual devices for the generic cleanup */
6473 if (dev->rtnl_link_ops)
6476 /* Push remaining network devices to init_net */
6477 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6478 err = dev_change_net_namespace(dev, &init_net, fb_name);
6480 pr_emerg("%s: failed to move %s to init_net: %d\n",
6481 __func__, dev->name, err);
6488 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6490 /* At exit all network devices most be removed from a network
6491 * namespace. Do this in the reverse order of registration.
6492 * Do this across as many network namespaces as possible to
6493 * improve batching efficiency.
6495 struct net_device *dev;
6497 LIST_HEAD(dev_kill_list);
6500 list_for_each_entry(net, net_list, exit_list) {
6501 for_each_netdev_reverse(net, dev) {
6502 if (dev->rtnl_link_ops)
6503 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6505 unregister_netdevice_queue(dev, &dev_kill_list);
6508 unregister_netdevice_many(&dev_kill_list);
6509 list_del(&dev_kill_list);
6513 static struct pernet_operations __net_initdata default_device_ops = {
6514 .exit = default_device_exit,
6515 .exit_batch = default_device_exit_batch,
6519 * Initialize the DEV module. At boot time this walks the device list and
6520 * unhooks any devices that fail to initialise (normally hardware not
6521 * present) and leaves us with a valid list of present and active devices.
6526 * This is called single threaded during boot, so no need
6527 * to take the rtnl semaphore.
6529 static int __init net_dev_init(void)
6531 int i, rc = -ENOMEM;
6533 BUG_ON(!dev_boot_phase);
6535 if (dev_proc_init())
6538 if (netdev_kobject_init())
6541 INIT_LIST_HEAD(&ptype_all);
6542 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6543 INIT_LIST_HEAD(&ptype_base[i]);
6545 if (register_pernet_subsys(&netdev_net_ops))
6549 * Initialise the packet receive queues.
6552 for_each_possible_cpu(i) {
6553 struct softnet_data *sd = &per_cpu(softnet_data, i);
6555 memset(sd, 0, sizeof(*sd));
6556 skb_queue_head_init(&sd->input_pkt_queue);
6557 skb_queue_head_init(&sd->process_queue);
6558 sd->completion_queue = NULL;
6559 INIT_LIST_HEAD(&sd->poll_list);
6560 sd->output_queue = NULL;
6561 sd->output_queue_tailp = &sd->output_queue;
6563 sd->csd.func = rps_trigger_softirq;
6569 sd->backlog.poll = process_backlog;
6570 sd->backlog.weight = weight_p;
6571 sd->backlog.gro_list = NULL;
6572 sd->backlog.gro_count = 0;
6577 /* The loopback device is special if any other network devices
6578 * is present in a network namespace the loopback device must
6579 * be present. Since we now dynamically allocate and free the
6580 * loopback device ensure this invariant is maintained by
6581 * keeping the loopback device as the first device on the
6582 * list of network devices. Ensuring the loopback devices
6583 * is the first device that appears and the last network device
6586 if (register_pernet_device(&loopback_net_ops))
6589 if (register_pernet_device(&default_device_ops))
6592 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6593 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6595 hotcpu_notifier(dev_cpu_callback, 0);
6603 subsys_initcall(net_dev_init);
6605 static int __init initialize_hashrnd(void)
6607 get_random_bytes(&hashrnd, sizeof(hashrnd));
6611 late_initcall_sync(initialize_hashrnd);