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 <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/net_tstamp.h>
136 #include <linux/static_key.h>
137 #include <net/flow_keys.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 * The list of packet types we will receive (as opposed to discard)
149 * and the routines to invoke.
151 * Why 16. Because with 16 the only overlap we get on a hash of the
152 * low nibble of the protocol value is RARP/SNAP/X.25.
154 * NOTE: That is no longer true with the addition of VLAN tags. Not
155 * sure which should go first, but I bet it won't make much
156 * difference if we are running VLANs. The good news is that
157 * this protocol won't be in the list unless compiled in, so
158 * the average user (w/out VLANs) will not be adversely affected.
175 #define PTYPE_HASH_SIZE (16)
176 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
178 static DEFINE_SPINLOCK(ptype_lock);
179 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
180 static struct list_head ptype_all __read_mostly; /* Taps */
183 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
186 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
188 * Writers must hold the rtnl semaphore while they loop through the
189 * dev_base_head list, and hold dev_base_lock for writing when they do the
190 * actual updates. This allows pure readers to access the list even
191 * while a writer is preparing to update it.
193 * To put it another way, dev_base_lock is held for writing only to
194 * protect against pure readers; the rtnl semaphore provides the
195 * protection against other writers.
197 * See, for example usages, register_netdevice() and
198 * unregister_netdevice(), which must be called with the rtnl
201 DEFINE_RWLOCK(dev_base_lock);
202 EXPORT_SYMBOL(dev_base_lock);
204 static inline void dev_base_seq_inc(struct net *net)
206 while (++net->dev_base_seq == 0);
209 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
211 unsigned int 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_IEEE80211, ARPHRD_IEEE80211_PRISM,
304 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
305 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
307 static const char *const netdev_lock_name[] =
308 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
309 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
310 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
311 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
312 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
313 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
314 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
315 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
316 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
317 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
318 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
319 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
320 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
321 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
322 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
324 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
325 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
327 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
331 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
332 if (netdev_lock_type[i] == dev_type)
334 /* the last key is used by default */
335 return ARRAY_SIZE(netdev_lock_type) - 1;
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
343 i = netdev_lock_pos(dev_type);
344 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
345 netdev_lock_name[i]);
348 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
352 i = netdev_lock_pos(dev->type);
353 lockdep_set_class_and_name(&dev->addr_list_lock,
354 &netdev_addr_lock_key[i],
355 netdev_lock_name[i]);
358 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
359 unsigned short dev_type)
362 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
367 /*******************************************************************************
369 Protocol management and registration routines
371 *******************************************************************************/
374 * Add a protocol ID to the list. Now that the input handler is
375 * smarter we can dispense with all the messy stuff that used to be
378 * BEWARE!!! Protocol handlers, mangling input packets,
379 * MUST BE last in hash buckets and checking protocol handlers
380 * MUST start from promiscuous ptype_all chain in net_bh.
381 * It is true now, do not change it.
382 * Explanation follows: if protocol handler, mangling packet, will
383 * be the first on list, it is not able to sense, that packet
384 * is cloned and should be copied-on-write, so that it will
385 * change it and subsequent readers will get broken packet.
389 static inline struct list_head *ptype_head(const struct packet_type *pt)
391 if (pt->type == htons(ETH_P_ALL))
394 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
398 * dev_add_pack - add packet handler
399 * @pt: packet type declaration
401 * Add a protocol handler to the networking stack. The passed &packet_type
402 * is linked into kernel lists and may not be freed until it has been
403 * removed from the kernel lists.
405 * This call does not sleep therefore it can not
406 * guarantee all CPU's that are in middle of receiving packets
407 * will see the new packet type (until the next received packet).
410 void dev_add_pack(struct packet_type *pt)
412 struct list_head *head = ptype_head(pt);
414 spin_lock(&ptype_lock);
415 list_add_rcu(&pt->list, head);
416 spin_unlock(&ptype_lock);
418 EXPORT_SYMBOL(dev_add_pack);
421 * __dev_remove_pack - remove packet handler
422 * @pt: packet type declaration
424 * Remove a protocol handler that was previously added to the kernel
425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
426 * from the kernel lists and can be freed or reused once this function
429 * The packet type might still be in use by receivers
430 * and must not be freed until after all the CPU's have gone
431 * through a quiescent state.
433 void __dev_remove_pack(struct packet_type *pt)
435 struct list_head *head = ptype_head(pt);
436 struct packet_type *pt1;
438 spin_lock(&ptype_lock);
440 list_for_each_entry(pt1, head, list) {
442 list_del_rcu(&pt->list);
447 pr_warn("dev_remove_pack: %p not found\n", pt);
449 spin_unlock(&ptype_lock);
451 EXPORT_SYMBOL(__dev_remove_pack);
454 * dev_remove_pack - remove packet handler
455 * @pt: packet type declaration
457 * Remove a protocol handler that was previously added to the kernel
458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
459 * from the kernel lists and can be freed or reused once this function
462 * This call sleeps to guarantee that no CPU is looking at the packet
465 void dev_remove_pack(struct packet_type *pt)
467 __dev_remove_pack(pt);
471 EXPORT_SYMBOL(dev_remove_pack);
473 /******************************************************************************
475 Device Boot-time Settings Routines
477 *******************************************************************************/
479 /* Boot time configuration table */
480 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
483 * netdev_boot_setup_add - add new setup entry
484 * @name: name of the device
485 * @map: configured settings for the device
487 * Adds new setup entry to the dev_boot_setup list. The function
488 * returns 0 on error and 1 on success. This is a generic routine to
491 static int netdev_boot_setup_add(char *name, struct ifmap *map)
493 struct netdev_boot_setup *s;
497 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
498 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
499 memset(s[i].name, 0, sizeof(s[i].name));
500 strlcpy(s[i].name, name, IFNAMSIZ);
501 memcpy(&s[i].map, map, sizeof(s[i].map));
506 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
510 * netdev_boot_setup_check - check boot time settings
511 * @dev: the netdevice
513 * Check boot time settings for the device.
514 * The found settings are set for the device to be used
515 * later in the device probing.
516 * Returns 0 if no settings found, 1 if they are.
518 int netdev_boot_setup_check(struct net_device *dev)
520 struct netdev_boot_setup *s = dev_boot_setup;
523 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
524 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
525 !strcmp(dev->name, s[i].name)) {
526 dev->irq = s[i].map.irq;
527 dev->base_addr = s[i].map.base_addr;
528 dev->mem_start = s[i].map.mem_start;
529 dev->mem_end = s[i].map.mem_end;
535 EXPORT_SYMBOL(netdev_boot_setup_check);
539 * netdev_boot_base - get address from boot time settings
540 * @prefix: prefix for network device
541 * @unit: id for network device
543 * Check boot time settings for the base address of device.
544 * The found settings are set for the device to be used
545 * later in the device probing.
546 * Returns 0 if no settings found.
548 unsigned long netdev_boot_base(const char *prefix, int unit)
550 const struct netdev_boot_setup *s = dev_boot_setup;
554 sprintf(name, "%s%d", prefix, unit);
557 * If device already registered then return base of 1
558 * to indicate not to probe for this interface
560 if (__dev_get_by_name(&init_net, name))
563 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
564 if (!strcmp(name, s[i].name))
565 return s[i].map.base_addr;
570 * Saves at boot time configured settings for any netdevice.
572 int __init netdev_boot_setup(char *str)
577 str = get_options(str, ARRAY_SIZE(ints), ints);
582 memset(&map, 0, sizeof(map));
586 map.base_addr = ints[2];
588 map.mem_start = ints[3];
590 map.mem_end = ints[4];
592 /* Add new entry to the list */
593 return netdev_boot_setup_add(str, &map);
596 __setup("netdev=", netdev_boot_setup);
598 /*******************************************************************************
600 Device Interface Subroutines
602 *******************************************************************************/
605 * __dev_get_by_name - find a device by its name
606 * @net: the applicable net namespace
607 * @name: name to find
609 * Find an interface by name. Must be called under RTNL semaphore
610 * or @dev_base_lock. If the name is found a pointer to the device
611 * is returned. If the name is not found then %NULL is returned. The
612 * reference counters are not incremented so the caller must be
613 * careful with locks.
616 struct net_device *__dev_get_by_name(struct net *net, const char *name)
618 struct hlist_node *p;
619 struct net_device *dev;
620 struct hlist_head *head = dev_name_hash(net, name);
622 hlist_for_each_entry(dev, p, head, name_hlist)
623 if (!strncmp(dev->name, name, IFNAMSIZ))
628 EXPORT_SYMBOL(__dev_get_by_name);
631 * dev_get_by_name_rcu - find a device by its name
632 * @net: the applicable net namespace
633 * @name: name to find
635 * Find an interface by name.
636 * If the name is found a pointer to the device is returned.
637 * If the name is not found then %NULL is returned.
638 * The reference counters are not incremented so the caller must be
639 * careful with locks. The caller must hold RCU lock.
642 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
644 struct hlist_node *p;
645 struct net_device *dev;
646 struct hlist_head *head = dev_name_hash(net, name);
648 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
649 if (!strncmp(dev->name, name, IFNAMSIZ))
654 EXPORT_SYMBOL(dev_get_by_name_rcu);
657 * dev_get_by_name - find a device by its name
658 * @net: the applicable net namespace
659 * @name: name to find
661 * Find an interface by name. This can be called from any
662 * context and does its own locking. The returned handle has
663 * the usage count incremented and the caller must use dev_put() to
664 * release it when it is no longer needed. %NULL is returned if no
665 * matching device is found.
668 struct net_device *dev_get_by_name(struct net *net, const char *name)
670 struct net_device *dev;
673 dev = dev_get_by_name_rcu(net, name);
679 EXPORT_SYMBOL(dev_get_by_name);
682 * __dev_get_by_index - find a device by its ifindex
683 * @net: the applicable net namespace
684 * @ifindex: index of device
686 * Search for an interface by index. Returns %NULL if the device
687 * is not found or a pointer to the device. The device has not
688 * had its reference counter increased so the caller must be careful
689 * about locking. The caller must hold either the RTNL semaphore
693 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
695 struct hlist_node *p;
696 struct net_device *dev;
697 struct hlist_head *head = dev_index_hash(net, ifindex);
699 hlist_for_each_entry(dev, p, head, index_hlist)
700 if (dev->ifindex == ifindex)
705 EXPORT_SYMBOL(__dev_get_by_index);
708 * dev_get_by_index_rcu - find a device by its ifindex
709 * @net: the applicable net namespace
710 * @ifindex: index of device
712 * Search for an interface by index. Returns %NULL if the device
713 * is not found or a pointer to the device. The device has not
714 * had its reference counter increased so the caller must be careful
715 * about locking. The caller must hold RCU lock.
718 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
720 struct hlist_node *p;
721 struct net_device *dev;
722 struct hlist_head *head = dev_index_hash(net, ifindex);
724 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
725 if (dev->ifindex == ifindex)
730 EXPORT_SYMBOL(dev_get_by_index_rcu);
734 * dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns NULL if the device
739 * is not found or a pointer to the device. The device returned has
740 * had a reference added and the pointer is safe until the user calls
741 * dev_put to indicate they have finished with it.
744 struct net_device *dev_get_by_index(struct net *net, int ifindex)
746 struct net_device *dev;
749 dev = dev_get_by_index_rcu(net, ifindex);
755 EXPORT_SYMBOL(dev_get_by_index);
758 * dev_getbyhwaddr_rcu - find a device by its hardware address
759 * @net: the applicable net namespace
760 * @type: media type of device
761 * @ha: hardware address
763 * Search for an interface by MAC address. Returns NULL if the device
764 * is not found or a pointer to the device.
765 * The caller must hold RCU or RTNL.
766 * The returned device has not had its ref count increased
767 * and the caller must therefore be careful about locking
771 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
774 struct net_device *dev;
776 for_each_netdev_rcu(net, dev)
777 if (dev->type == type &&
778 !memcmp(dev->dev_addr, ha, dev->addr_len))
783 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
785 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
787 struct net_device *dev;
790 for_each_netdev(net, dev)
791 if (dev->type == type)
796 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
798 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
800 struct net_device *dev, *ret = NULL;
803 for_each_netdev_rcu(net, dev)
804 if (dev->type == type) {
812 EXPORT_SYMBOL(dev_getfirstbyhwtype);
815 * dev_get_by_flags_rcu - find any device with given flags
816 * @net: the applicable net namespace
817 * @if_flags: IFF_* values
818 * @mask: bitmask of bits in if_flags to check
820 * Search for any interface with the given flags. Returns NULL if a device
821 * is not found or a pointer to the device. Must be called inside
822 * rcu_read_lock(), and result refcount is unchanged.
825 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
828 struct net_device *dev, *ret;
831 for_each_netdev_rcu(net, dev) {
832 if (((dev->flags ^ if_flags) & mask) == 0) {
839 EXPORT_SYMBOL(dev_get_by_flags_rcu);
842 * dev_valid_name - check if name is okay for network device
845 * Network device names need to be valid file names to
846 * to allow sysfs to work. We also disallow any kind of
849 bool dev_valid_name(const char *name)
853 if (strlen(name) >= IFNAMSIZ)
855 if (!strcmp(name, ".") || !strcmp(name, ".."))
859 if (*name == '/' || isspace(*name))
865 EXPORT_SYMBOL(dev_valid_name);
868 * __dev_alloc_name - allocate a name for a device
869 * @net: network namespace to allocate the device name in
870 * @name: name format string
871 * @buf: scratch buffer and result name string
873 * Passed a format string - eg "lt%d" it will try and find a suitable
874 * id. It scans list of devices to build up a free map, then chooses
875 * the first empty slot. The caller must hold the dev_base or rtnl lock
876 * while allocating the name and adding the device in order to avoid
878 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
879 * Returns the number of the unit assigned or a negative errno code.
882 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
886 const int max_netdevices = 8*PAGE_SIZE;
887 unsigned long *inuse;
888 struct net_device *d;
890 p = strnchr(name, IFNAMSIZ-1, '%');
893 * Verify the string as this thing may have come from
894 * the user. There must be either one "%d" and no other "%"
897 if (p[1] != 'd' || strchr(p + 2, '%'))
900 /* Use one page as a bit array of possible slots */
901 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
905 for_each_netdev(net, d) {
906 if (!sscanf(d->name, name, &i))
908 if (i < 0 || i >= max_netdevices)
911 /* avoid cases where sscanf is not exact inverse of printf */
912 snprintf(buf, IFNAMSIZ, name, i);
913 if (!strncmp(buf, d->name, IFNAMSIZ))
917 i = find_first_zero_bit(inuse, max_netdevices);
918 free_page((unsigned long) inuse);
922 snprintf(buf, IFNAMSIZ, name, i);
923 if (!__dev_get_by_name(net, buf))
926 /* It is possible to run out of possible slots
927 * when the name is long and there isn't enough space left
928 * for the digits, or if all bits are used.
934 * dev_alloc_name - allocate a name for a device
936 * @name: name format string
938 * Passed a format string - eg "lt%d" it will try and find a suitable
939 * id. It scans list of devices to build up a free map, then chooses
940 * the first empty slot. The caller must hold the dev_base or rtnl lock
941 * while allocating the name and adding the device in order to avoid
943 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
944 * Returns the number of the unit assigned or a negative errno code.
947 int dev_alloc_name(struct net_device *dev, const char *name)
953 BUG_ON(!dev_net(dev));
955 ret = __dev_alloc_name(net, name, buf);
957 strlcpy(dev->name, buf, IFNAMSIZ);
960 EXPORT_SYMBOL(dev_alloc_name);
962 static int dev_get_valid_name(struct net_device *dev, const char *name)
966 BUG_ON(!dev_net(dev));
969 if (!dev_valid_name(name))
972 if (strchr(name, '%'))
973 return dev_alloc_name(dev, name);
974 else if (__dev_get_by_name(net, name))
976 else if (dev->name != name)
977 strlcpy(dev->name, name, IFNAMSIZ);
983 * dev_change_name - change name of a device
985 * @newname: name (or format string) must be at least IFNAMSIZ
987 * Change name of a device, can pass format strings "eth%d".
990 int dev_change_name(struct net_device *dev, const char *newname)
992 char oldname[IFNAMSIZ];
998 BUG_ON(!dev_net(dev));
1001 if (dev->flags & IFF_UP)
1004 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1007 memcpy(oldname, dev->name, IFNAMSIZ);
1009 err = dev_get_valid_name(dev, newname);
1014 ret = device_rename(&dev->dev, dev->name);
1016 memcpy(dev->name, oldname, IFNAMSIZ);
1020 write_lock_bh(&dev_base_lock);
1021 hlist_del_rcu(&dev->name_hlist);
1022 write_unlock_bh(&dev_base_lock);
1026 write_lock_bh(&dev_base_lock);
1027 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1028 write_unlock_bh(&dev_base_lock);
1030 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1031 ret = notifier_to_errno(ret);
1034 /* err >= 0 after dev_alloc_name() or stores the first errno */
1037 memcpy(dev->name, oldname, IFNAMSIZ);
1040 pr_err("%s: name change rollback failed: %d\n",
1049 * dev_set_alias - change ifalias of a device
1051 * @alias: name up to IFALIASZ
1052 * @len: limit of bytes to copy from info
1054 * Set ifalias for a device,
1056 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1060 if (len >= IFALIASZ)
1065 kfree(dev->ifalias);
1066 dev->ifalias = NULL;
1071 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1075 strlcpy(dev->ifalias, alias, len+1);
1081 * netdev_features_change - device changes features
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed features.
1086 void netdev_features_change(struct net_device *dev)
1088 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1090 EXPORT_SYMBOL(netdev_features_change);
1093 * netdev_state_change - device changes state
1094 * @dev: device to cause notification
1096 * Called to indicate a device has changed state. This function calls
1097 * the notifier chains for netdev_chain and sends a NEWLINK message
1098 * to the routing socket.
1100 void netdev_state_change(struct net_device *dev)
1102 if (dev->flags & IFF_UP) {
1103 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1104 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1107 EXPORT_SYMBOL(netdev_state_change);
1109 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1111 return call_netdevice_notifiers(event, dev);
1113 EXPORT_SYMBOL(netdev_bonding_change);
1116 * dev_load - load a network module
1117 * @net: the applicable net namespace
1118 * @name: name of interface
1120 * If a network interface is not present and the process has suitable
1121 * privileges this function loads the module. If module loading is not
1122 * available in this kernel then it becomes a nop.
1125 void dev_load(struct net *net, const char *name)
1127 struct net_device *dev;
1131 dev = dev_get_by_name_rcu(net, name);
1135 if (no_module && capable(CAP_NET_ADMIN))
1136 no_module = request_module("netdev-%s", name);
1137 if (no_module && capable(CAP_SYS_MODULE)) {
1138 if (!request_module("%s", name))
1139 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1143 EXPORT_SYMBOL(dev_load);
1145 static int __dev_open(struct net_device *dev)
1147 const struct net_device_ops *ops = dev->netdev_ops;
1152 if (!netif_device_present(dev))
1155 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1156 ret = notifier_to_errno(ret);
1160 set_bit(__LINK_STATE_START, &dev->state);
1162 if (ops->ndo_validate_addr)
1163 ret = ops->ndo_validate_addr(dev);
1165 if (!ret && ops->ndo_open)
1166 ret = ops->ndo_open(dev);
1169 clear_bit(__LINK_STATE_START, &dev->state);
1171 dev->flags |= IFF_UP;
1172 net_dmaengine_get();
1173 dev_set_rx_mode(dev);
1181 * dev_open - prepare an interface for use.
1182 * @dev: device to open
1184 * Takes a device from down to up state. The device's private open
1185 * function is invoked and then the multicast lists are loaded. Finally
1186 * the device is moved into the up state and a %NETDEV_UP message is
1187 * sent to the netdev notifier chain.
1189 * Calling this function on an active interface is a nop. On a failure
1190 * a negative errno code is returned.
1192 int dev_open(struct net_device *dev)
1196 if (dev->flags & IFF_UP)
1199 ret = __dev_open(dev);
1203 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1204 call_netdevice_notifiers(NETDEV_UP, dev);
1208 EXPORT_SYMBOL(dev_open);
1210 static int __dev_close_many(struct list_head *head)
1212 struct net_device *dev;
1217 list_for_each_entry(dev, head, unreg_list) {
1218 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1220 clear_bit(__LINK_STATE_START, &dev->state);
1222 /* Synchronize to scheduled poll. We cannot touch poll list, it
1223 * can be even on different cpu. So just clear netif_running().
1225 * dev->stop() will invoke napi_disable() on all of it's
1226 * napi_struct instances on this device.
1228 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1231 dev_deactivate_many(head);
1233 list_for_each_entry(dev, head, unreg_list) {
1234 const struct net_device_ops *ops = dev->netdev_ops;
1237 * Call the device specific close. This cannot fail.
1238 * Only if device is UP
1240 * We allow it to be called even after a DETACH hot-plug
1246 dev->flags &= ~IFF_UP;
1247 net_dmaengine_put();
1253 static int __dev_close(struct net_device *dev)
1258 list_add(&dev->unreg_list, &single);
1259 retval = __dev_close_many(&single);
1264 static int dev_close_many(struct list_head *head)
1266 struct net_device *dev, *tmp;
1267 LIST_HEAD(tmp_list);
1269 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1270 if (!(dev->flags & IFF_UP))
1271 list_move(&dev->unreg_list, &tmp_list);
1273 __dev_close_many(head);
1275 list_for_each_entry(dev, head, unreg_list) {
1276 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1277 call_netdevice_notifiers(NETDEV_DOWN, dev);
1280 /* rollback_registered_many needs the complete original list */
1281 list_splice(&tmp_list, head);
1286 * dev_close - shutdown an interface.
1287 * @dev: device to shutdown
1289 * This function moves an active device into down state. A
1290 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1291 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1294 int dev_close(struct net_device *dev)
1296 if (dev->flags & IFF_UP) {
1299 list_add(&dev->unreg_list, &single);
1300 dev_close_many(&single);
1305 EXPORT_SYMBOL(dev_close);
1309 * dev_disable_lro - disable Large Receive Offload on a device
1312 * Disable Large Receive Offload (LRO) on a net device. Must be
1313 * called under RTNL. This is needed if received packets may be
1314 * forwarded to another interface.
1316 void dev_disable_lro(struct net_device *dev)
1319 * If we're trying to disable lro on a vlan device
1320 * use the underlying physical device instead
1322 if (is_vlan_dev(dev))
1323 dev = vlan_dev_real_dev(dev);
1325 dev->wanted_features &= ~NETIF_F_LRO;
1326 netdev_update_features(dev);
1328 if (unlikely(dev->features & NETIF_F_LRO))
1329 netdev_WARN(dev, "failed to disable LRO!\n");
1331 EXPORT_SYMBOL(dev_disable_lro);
1334 static int dev_boot_phase = 1;
1337 * register_netdevice_notifier - register a network notifier block
1340 * Register a notifier to be called when network device events occur.
1341 * The notifier passed is linked into the kernel structures and must
1342 * not be reused until it has been unregistered. A negative errno code
1343 * is returned on a failure.
1345 * When registered all registration and up events are replayed
1346 * to the new notifier to allow device to have a race free
1347 * view of the network device list.
1350 int register_netdevice_notifier(struct notifier_block *nb)
1352 struct net_device *dev;
1353 struct net_device *last;
1358 err = raw_notifier_chain_register(&netdev_chain, nb);
1364 for_each_netdev(net, dev) {
1365 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1366 err = notifier_to_errno(err);
1370 if (!(dev->flags & IFF_UP))
1373 nb->notifier_call(nb, NETDEV_UP, dev);
1384 for_each_netdev(net, dev) {
1388 if (dev->flags & IFF_UP) {
1389 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1390 nb->notifier_call(nb, NETDEV_DOWN, dev);
1392 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1393 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1398 raw_notifier_chain_unregister(&netdev_chain, nb);
1401 EXPORT_SYMBOL(register_netdevice_notifier);
1404 * unregister_netdevice_notifier - unregister a network notifier block
1407 * Unregister a notifier previously registered by
1408 * register_netdevice_notifier(). The notifier is unlinked into the
1409 * kernel structures and may then be reused. A negative errno code
1410 * is returned on a failure.
1412 * After unregistering unregister and down device events are synthesized
1413 * for all devices on the device list to the removed notifier to remove
1414 * the need for special case cleanup code.
1417 int unregister_netdevice_notifier(struct notifier_block *nb)
1419 struct net_device *dev;
1424 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1429 for_each_netdev(net, dev) {
1430 if (dev->flags & IFF_UP) {
1431 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1432 nb->notifier_call(nb, NETDEV_DOWN, dev);
1434 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1435 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1442 EXPORT_SYMBOL(unregister_netdevice_notifier);
1445 * call_netdevice_notifiers - call all network notifier blocks
1446 * @val: value passed unmodified to notifier function
1447 * @dev: net_device pointer passed unmodified to notifier function
1449 * Call all network notifier blocks. Parameters and return value
1450 * are as for raw_notifier_call_chain().
1453 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1456 return raw_notifier_call_chain(&netdev_chain, val, dev);
1458 EXPORT_SYMBOL(call_netdevice_notifiers);
1460 static struct static_key netstamp_needed __read_mostly;
1461 #ifdef HAVE_JUMP_LABEL
1462 /* We are not allowed to call static_key_slow_dec() from irq context
1463 * If net_disable_timestamp() is called from irq context, defer the
1464 * static_key_slow_dec() calls.
1466 static atomic_t netstamp_needed_deferred;
1469 void net_enable_timestamp(void)
1471 #ifdef HAVE_JUMP_LABEL
1472 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1476 static_key_slow_dec(&netstamp_needed);
1480 WARN_ON(in_interrupt());
1481 static_key_slow_inc(&netstamp_needed);
1483 EXPORT_SYMBOL(net_enable_timestamp);
1485 void net_disable_timestamp(void)
1487 #ifdef HAVE_JUMP_LABEL
1488 if (in_interrupt()) {
1489 atomic_inc(&netstamp_needed_deferred);
1493 static_key_slow_dec(&netstamp_needed);
1495 EXPORT_SYMBOL(net_disable_timestamp);
1497 static inline void net_timestamp_set(struct sk_buff *skb)
1499 skb->tstamp.tv64 = 0;
1500 if (static_key_false(&netstamp_needed))
1501 __net_timestamp(skb);
1504 #define net_timestamp_check(COND, SKB) \
1505 if (static_key_false(&netstamp_needed)) { \
1506 if ((COND) && !(SKB)->tstamp.tv64) \
1507 __net_timestamp(SKB); \
1510 static int net_hwtstamp_validate(struct ifreq *ifr)
1512 struct hwtstamp_config cfg;
1513 enum hwtstamp_tx_types tx_type;
1514 enum hwtstamp_rx_filters rx_filter;
1515 int tx_type_valid = 0;
1516 int rx_filter_valid = 0;
1518 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1521 if (cfg.flags) /* reserved for future extensions */
1524 tx_type = cfg.tx_type;
1525 rx_filter = cfg.rx_filter;
1528 case HWTSTAMP_TX_OFF:
1529 case HWTSTAMP_TX_ON:
1530 case HWTSTAMP_TX_ONESTEP_SYNC:
1535 switch (rx_filter) {
1536 case HWTSTAMP_FILTER_NONE:
1537 case HWTSTAMP_FILTER_ALL:
1538 case HWTSTAMP_FILTER_SOME:
1539 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1540 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1542 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1543 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1545 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1546 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1548 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1549 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1550 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1551 rx_filter_valid = 1;
1555 if (!tx_type_valid || !rx_filter_valid)
1561 static inline bool is_skb_forwardable(struct net_device *dev,
1562 struct sk_buff *skb)
1566 if (!(dev->flags & IFF_UP))
1569 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1570 if (skb->len <= len)
1573 /* if TSO is enabled, we don't care about the length as the packet
1574 * could be forwarded without being segmented before
1576 if (skb_is_gso(skb))
1583 * dev_forward_skb - loopback an skb to another netif
1585 * @dev: destination network device
1586 * @skb: buffer to forward
1589 * NET_RX_SUCCESS (no congestion)
1590 * NET_RX_DROP (packet was dropped, but freed)
1592 * dev_forward_skb can be used for injecting an skb from the
1593 * start_xmit function of one device into the receive queue
1594 * of another device.
1596 * The receiving device may be in another namespace, so
1597 * we have to clear all information in the skb that could
1598 * impact namespace isolation.
1600 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1602 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1603 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1604 atomic_long_inc(&dev->rx_dropped);
1613 if (unlikely(!is_skb_forwardable(dev, skb))) {
1614 atomic_long_inc(&dev->rx_dropped);
1621 skb->tstamp.tv64 = 0;
1622 skb->pkt_type = PACKET_HOST;
1623 skb->protocol = eth_type_trans(skb, dev);
1627 return netif_rx(skb);
1629 EXPORT_SYMBOL_GPL(dev_forward_skb);
1631 static inline int deliver_skb(struct sk_buff *skb,
1632 struct packet_type *pt_prev,
1633 struct net_device *orig_dev)
1635 atomic_inc(&skb->users);
1636 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1640 * Support routine. Sends outgoing frames to any network
1641 * taps currently in use.
1644 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1646 struct packet_type *ptype;
1647 struct sk_buff *skb2 = NULL;
1648 struct packet_type *pt_prev = NULL;
1651 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1652 /* Never send packets back to the socket
1653 * they originated from - MvS (miquels@drinkel.ow.org)
1655 if ((ptype->dev == dev || !ptype->dev) &&
1656 (ptype->af_packet_priv == NULL ||
1657 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1659 deliver_skb(skb2, pt_prev, skb->dev);
1664 skb2 = skb_clone(skb, GFP_ATOMIC);
1668 net_timestamp_set(skb2);
1670 /* skb->nh should be correctly
1671 set by sender, so that the second statement is
1672 just protection against buggy protocols.
1674 skb_reset_mac_header(skb2);
1676 if (skb_network_header(skb2) < skb2->data ||
1677 skb2->network_header > skb2->tail) {
1678 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1679 ntohs(skb2->protocol),
1681 skb_reset_network_header(skb2);
1684 skb2->transport_header = skb2->network_header;
1685 skb2->pkt_type = PACKET_OUTGOING;
1690 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1695 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1696 * @dev: Network device
1697 * @txq: number of queues available
1699 * If real_num_tx_queues is changed the tc mappings may no longer be
1700 * valid. To resolve this verify the tc mapping remains valid and if
1701 * not NULL the mapping. With no priorities mapping to this
1702 * offset/count pair it will no longer be used. In the worst case TC0
1703 * is invalid nothing can be done so disable priority mappings. If is
1704 * expected that drivers will fix this mapping if they can before
1705 * calling netif_set_real_num_tx_queues.
1707 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1710 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1712 /* If TC0 is invalidated disable TC mapping */
1713 if (tc->offset + tc->count > txq) {
1714 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1719 /* Invalidated prio to tc mappings set to TC0 */
1720 for (i = 1; i < TC_BITMASK + 1; i++) {
1721 int q = netdev_get_prio_tc_map(dev, i);
1723 tc = &dev->tc_to_txq[q];
1724 if (tc->offset + tc->count > txq) {
1725 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1727 netdev_set_prio_tc_map(dev, i, 0);
1733 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1734 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1736 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1740 if (txq < 1 || txq > dev->num_tx_queues)
1743 if (dev->reg_state == NETREG_REGISTERED ||
1744 dev->reg_state == NETREG_UNREGISTERING) {
1747 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1753 netif_setup_tc(dev, txq);
1755 if (txq < dev->real_num_tx_queues)
1756 qdisc_reset_all_tx_gt(dev, txq);
1759 dev->real_num_tx_queues = txq;
1762 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1766 * netif_set_real_num_rx_queues - set actual number of RX queues used
1767 * @dev: Network device
1768 * @rxq: Actual number of RX queues
1770 * This must be called either with the rtnl_lock held or before
1771 * registration of the net device. Returns 0 on success, or a
1772 * negative error code. If called before registration, it always
1775 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1779 if (rxq < 1 || rxq > dev->num_rx_queues)
1782 if (dev->reg_state == NETREG_REGISTERED) {
1785 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1791 dev->real_num_rx_queues = rxq;
1794 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1798 * netif_get_num_default_rss_queues - default number of RSS queues
1800 * This routine should set an upper limit on the number of RSS queues
1801 * used by default by multiqueue devices.
1803 int netif_get_num_default_rss_queues(void)
1805 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
1807 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
1809 static inline void __netif_reschedule(struct Qdisc *q)
1811 struct softnet_data *sd;
1812 unsigned long flags;
1814 local_irq_save(flags);
1815 sd = &__get_cpu_var(softnet_data);
1816 q->next_sched = NULL;
1817 *sd->output_queue_tailp = q;
1818 sd->output_queue_tailp = &q->next_sched;
1819 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1820 local_irq_restore(flags);
1823 void __netif_schedule(struct Qdisc *q)
1825 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1826 __netif_reschedule(q);
1828 EXPORT_SYMBOL(__netif_schedule);
1830 void dev_kfree_skb_irq(struct sk_buff *skb)
1832 if (atomic_dec_and_test(&skb->users)) {
1833 struct softnet_data *sd;
1834 unsigned long flags;
1836 local_irq_save(flags);
1837 sd = &__get_cpu_var(softnet_data);
1838 skb->next = sd->completion_queue;
1839 sd->completion_queue = skb;
1840 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1841 local_irq_restore(flags);
1844 EXPORT_SYMBOL(dev_kfree_skb_irq);
1846 void dev_kfree_skb_any(struct sk_buff *skb)
1848 if (in_irq() || irqs_disabled())
1849 dev_kfree_skb_irq(skb);
1853 EXPORT_SYMBOL(dev_kfree_skb_any);
1857 * netif_device_detach - mark device as removed
1858 * @dev: network device
1860 * Mark device as removed from system and therefore no longer available.
1862 void netif_device_detach(struct net_device *dev)
1864 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1865 netif_running(dev)) {
1866 netif_tx_stop_all_queues(dev);
1869 EXPORT_SYMBOL(netif_device_detach);
1872 * netif_device_attach - mark device as attached
1873 * @dev: network device
1875 * Mark device as attached from system and restart if needed.
1877 void netif_device_attach(struct net_device *dev)
1879 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1880 netif_running(dev)) {
1881 netif_tx_wake_all_queues(dev);
1882 __netdev_watchdog_up(dev);
1885 EXPORT_SYMBOL(netif_device_attach);
1887 static void skb_warn_bad_offload(const struct sk_buff *skb)
1889 static const netdev_features_t null_features = 0;
1890 struct net_device *dev = skb->dev;
1891 const char *driver = "";
1893 if (dev && dev->dev.parent)
1894 driver = dev_driver_string(dev->dev.parent);
1896 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1897 "gso_type=%d ip_summed=%d\n",
1898 driver, dev ? &dev->features : &null_features,
1899 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1900 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1901 skb_shinfo(skb)->gso_type, skb->ip_summed);
1905 * Invalidate hardware checksum when packet is to be mangled, and
1906 * complete checksum manually on outgoing path.
1908 int skb_checksum_help(struct sk_buff *skb)
1911 int ret = 0, offset;
1913 if (skb->ip_summed == CHECKSUM_COMPLETE)
1914 goto out_set_summed;
1916 if (unlikely(skb_shinfo(skb)->gso_size)) {
1917 skb_warn_bad_offload(skb);
1921 offset = skb_checksum_start_offset(skb);
1922 BUG_ON(offset >= skb_headlen(skb));
1923 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1925 offset += skb->csum_offset;
1926 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1928 if (skb_cloned(skb) &&
1929 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1930 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1935 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1937 skb->ip_summed = CHECKSUM_NONE;
1941 EXPORT_SYMBOL(skb_checksum_help);
1944 * skb_gso_segment - Perform segmentation on skb.
1945 * @skb: buffer to segment
1946 * @features: features for the output path (see dev->features)
1948 * This function segments the given skb and returns a list of segments.
1950 * It may return NULL if the skb requires no segmentation. This is
1951 * only possible when GSO is used for verifying header integrity.
1953 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1954 netdev_features_t features)
1956 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1957 struct packet_type *ptype;
1958 __be16 type = skb->protocol;
1959 int vlan_depth = ETH_HLEN;
1962 while (type == htons(ETH_P_8021Q)) {
1963 struct vlan_hdr *vh;
1965 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1966 return ERR_PTR(-EINVAL);
1968 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1969 type = vh->h_vlan_encapsulated_proto;
1970 vlan_depth += VLAN_HLEN;
1973 skb_reset_mac_header(skb);
1974 skb->mac_len = skb->network_header - skb->mac_header;
1975 __skb_pull(skb, skb->mac_len);
1977 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1978 skb_warn_bad_offload(skb);
1980 if (skb_header_cloned(skb) &&
1981 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1982 return ERR_PTR(err);
1986 list_for_each_entry_rcu(ptype,
1987 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1988 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1989 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1990 err = ptype->gso_send_check(skb);
1991 segs = ERR_PTR(err);
1992 if (err || skb_gso_ok(skb, features))
1994 __skb_push(skb, (skb->data -
1995 skb_network_header(skb)));
1997 segs = ptype->gso_segment(skb, features);
2003 __skb_push(skb, skb->data - skb_mac_header(skb));
2007 EXPORT_SYMBOL(skb_gso_segment);
2009 /* Take action when hardware reception checksum errors are detected. */
2011 void netdev_rx_csum_fault(struct net_device *dev)
2013 if (net_ratelimit()) {
2014 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2018 EXPORT_SYMBOL(netdev_rx_csum_fault);
2021 /* Actually, we should eliminate this check as soon as we know, that:
2022 * 1. IOMMU is present and allows to map all the memory.
2023 * 2. No high memory really exists on this machine.
2026 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2028 #ifdef CONFIG_HIGHMEM
2030 if (!(dev->features & NETIF_F_HIGHDMA)) {
2031 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2032 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2033 if (PageHighMem(skb_frag_page(frag)))
2038 if (PCI_DMA_BUS_IS_PHYS) {
2039 struct device *pdev = dev->dev.parent;
2043 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2044 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2045 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2046 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2055 void (*destructor)(struct sk_buff *skb);
2058 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2060 static void dev_gso_skb_destructor(struct sk_buff *skb)
2062 struct dev_gso_cb *cb;
2065 struct sk_buff *nskb = skb->next;
2067 skb->next = nskb->next;
2070 } while (skb->next);
2072 cb = DEV_GSO_CB(skb);
2074 cb->destructor(skb);
2078 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2079 * @skb: buffer to segment
2080 * @features: device features as applicable to this skb
2082 * This function segments the given skb and stores the list of segments
2085 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2087 struct sk_buff *segs;
2089 segs = skb_gso_segment(skb, features);
2091 /* Verifying header integrity only. */
2096 return PTR_ERR(segs);
2099 DEV_GSO_CB(skb)->destructor = skb->destructor;
2100 skb->destructor = dev_gso_skb_destructor;
2105 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2107 return ((features & NETIF_F_GEN_CSUM) ||
2108 ((features & NETIF_F_V4_CSUM) &&
2109 protocol == htons(ETH_P_IP)) ||
2110 ((features & NETIF_F_V6_CSUM) &&
2111 protocol == htons(ETH_P_IPV6)) ||
2112 ((features & NETIF_F_FCOE_CRC) &&
2113 protocol == htons(ETH_P_FCOE)));
2116 static netdev_features_t harmonize_features(struct sk_buff *skb,
2117 __be16 protocol, netdev_features_t features)
2119 if (!can_checksum_protocol(features, protocol)) {
2120 features &= ~NETIF_F_ALL_CSUM;
2121 features &= ~NETIF_F_SG;
2122 } else if (illegal_highdma(skb->dev, skb)) {
2123 features &= ~NETIF_F_SG;
2129 netdev_features_t netif_skb_features(struct sk_buff *skb)
2131 __be16 protocol = skb->protocol;
2132 netdev_features_t features = skb->dev->features;
2134 if (protocol == htons(ETH_P_8021Q)) {
2135 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2136 protocol = veh->h_vlan_encapsulated_proto;
2137 } else if (!vlan_tx_tag_present(skb)) {
2138 return harmonize_features(skb, protocol, features);
2141 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2143 if (protocol != htons(ETH_P_8021Q)) {
2144 return harmonize_features(skb, protocol, features);
2146 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2147 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2148 return harmonize_features(skb, protocol, features);
2151 EXPORT_SYMBOL(netif_skb_features);
2154 * Returns true if either:
2155 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2156 * 2. skb is fragmented and the device does not support SG, or if
2157 * at least one of fragments is in highmem and device does not
2158 * support DMA from it.
2160 static inline int skb_needs_linearize(struct sk_buff *skb,
2163 return skb_is_nonlinear(skb) &&
2164 ((skb_has_frag_list(skb) &&
2165 !(features & NETIF_F_FRAGLIST)) ||
2166 (skb_shinfo(skb)->nr_frags &&
2167 !(features & NETIF_F_SG)));
2170 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2171 struct netdev_queue *txq)
2173 const struct net_device_ops *ops = dev->netdev_ops;
2174 int rc = NETDEV_TX_OK;
2175 unsigned int skb_len;
2177 if (likely(!skb->next)) {
2178 netdev_features_t features;
2181 * If device doesn't need skb->dst, release it right now while
2182 * its hot in this cpu cache
2184 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2187 if (!list_empty(&ptype_all))
2188 dev_queue_xmit_nit(skb, dev);
2190 features = netif_skb_features(skb);
2192 if (vlan_tx_tag_present(skb) &&
2193 !(features & NETIF_F_HW_VLAN_TX)) {
2194 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2201 if (netif_needs_gso(skb, features)) {
2202 if (unlikely(dev_gso_segment(skb, features)))
2207 if (skb_needs_linearize(skb, features) &&
2208 __skb_linearize(skb))
2211 /* If packet is not checksummed and device does not
2212 * support checksumming for this protocol, complete
2213 * checksumming here.
2215 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2216 skb_set_transport_header(skb,
2217 skb_checksum_start_offset(skb));
2218 if (!(features & NETIF_F_ALL_CSUM) &&
2219 skb_checksum_help(skb))
2225 rc = ops->ndo_start_xmit(skb, dev);
2226 trace_net_dev_xmit(skb, rc, dev, skb_len);
2227 if (rc == NETDEV_TX_OK)
2228 txq_trans_update(txq);
2234 struct sk_buff *nskb = skb->next;
2236 skb->next = nskb->next;
2240 * If device doesn't need nskb->dst, release it right now while
2241 * its hot in this cpu cache
2243 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2246 skb_len = nskb->len;
2247 rc = ops->ndo_start_xmit(nskb, dev);
2248 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2249 if (unlikely(rc != NETDEV_TX_OK)) {
2250 if (rc & ~NETDEV_TX_MASK)
2251 goto out_kfree_gso_skb;
2252 nskb->next = skb->next;
2256 txq_trans_update(txq);
2257 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2258 return NETDEV_TX_BUSY;
2259 } while (skb->next);
2262 if (likely(skb->next == NULL))
2263 skb->destructor = DEV_GSO_CB(skb)->destructor;
2270 static u32 hashrnd __read_mostly;
2273 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2274 * to be used as a distribution range.
2276 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2277 unsigned int num_tx_queues)
2281 u16 qcount = num_tx_queues;
2283 if (skb_rx_queue_recorded(skb)) {
2284 hash = skb_get_rx_queue(skb);
2285 while (unlikely(hash >= num_tx_queues))
2286 hash -= num_tx_queues;
2291 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2292 qoffset = dev->tc_to_txq[tc].offset;
2293 qcount = dev->tc_to_txq[tc].count;
2296 if (skb->sk && skb->sk->sk_hash)
2297 hash = skb->sk->sk_hash;
2299 hash = (__force u16) skb->protocol;
2300 hash = jhash_1word(hash, hashrnd);
2302 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2304 EXPORT_SYMBOL(__skb_tx_hash);
2306 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2308 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2309 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2310 dev->name, queue_index,
2311 dev->real_num_tx_queues);
2317 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2320 struct xps_dev_maps *dev_maps;
2321 struct xps_map *map;
2322 int queue_index = -1;
2325 dev_maps = rcu_dereference(dev->xps_maps);
2327 map = rcu_dereference(
2328 dev_maps->cpu_map[raw_smp_processor_id()]);
2331 queue_index = map->queues[0];
2334 if (skb->sk && skb->sk->sk_hash)
2335 hash = skb->sk->sk_hash;
2337 hash = (__force u16) skb->protocol ^
2339 hash = jhash_1word(hash, hashrnd);
2340 queue_index = map->queues[
2341 ((u64)hash * map->len) >> 32];
2343 if (unlikely(queue_index >= dev->real_num_tx_queues))
2355 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2356 struct sk_buff *skb)
2359 const struct net_device_ops *ops = dev->netdev_ops;
2361 if (dev->real_num_tx_queues == 1)
2363 else if (ops->ndo_select_queue) {
2364 queue_index = ops->ndo_select_queue(dev, skb);
2365 queue_index = dev_cap_txqueue(dev, queue_index);
2367 struct sock *sk = skb->sk;
2368 queue_index = sk_tx_queue_get(sk);
2370 if (queue_index < 0 || skb->ooo_okay ||
2371 queue_index >= dev->real_num_tx_queues) {
2372 int old_index = queue_index;
2374 queue_index = get_xps_queue(dev, skb);
2375 if (queue_index < 0)
2376 queue_index = skb_tx_hash(dev, skb);
2378 if (queue_index != old_index && sk) {
2379 struct dst_entry *dst =
2380 rcu_dereference_check(sk->sk_dst_cache, 1);
2382 if (dst && skb_dst(skb) == dst)
2383 sk_tx_queue_set(sk, queue_index);
2388 skb_set_queue_mapping(skb, queue_index);
2389 return netdev_get_tx_queue(dev, queue_index);
2392 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2393 struct net_device *dev,
2394 struct netdev_queue *txq)
2396 spinlock_t *root_lock = qdisc_lock(q);
2400 qdisc_skb_cb(skb)->pkt_len = skb->len;
2401 qdisc_calculate_pkt_len(skb, q);
2403 * Heuristic to force contended enqueues to serialize on a
2404 * separate lock before trying to get qdisc main lock.
2405 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2406 * and dequeue packets faster.
2408 contended = qdisc_is_running(q);
2409 if (unlikely(contended))
2410 spin_lock(&q->busylock);
2412 spin_lock(root_lock);
2413 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2416 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2417 qdisc_run_begin(q)) {
2419 * This is a work-conserving queue; there are no old skbs
2420 * waiting to be sent out; and the qdisc is not running -
2421 * xmit the skb directly.
2423 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2426 qdisc_bstats_update(q, skb);
2428 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2429 if (unlikely(contended)) {
2430 spin_unlock(&q->busylock);
2437 rc = NET_XMIT_SUCCESS;
2440 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2441 if (qdisc_run_begin(q)) {
2442 if (unlikely(contended)) {
2443 spin_unlock(&q->busylock);
2449 spin_unlock(root_lock);
2450 if (unlikely(contended))
2451 spin_unlock(&q->busylock);
2455 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2456 static void skb_update_prio(struct sk_buff *skb)
2458 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2460 if (!skb->priority && skb->sk && map) {
2461 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2463 if (prioidx < map->priomap_len)
2464 skb->priority = map->priomap[prioidx];
2468 #define skb_update_prio(skb)
2471 static DEFINE_PER_CPU(int, xmit_recursion);
2472 #define RECURSION_LIMIT 10
2475 * dev_loopback_xmit - loop back @skb
2476 * @skb: buffer to transmit
2478 int dev_loopback_xmit(struct sk_buff *skb)
2480 skb_reset_mac_header(skb);
2481 __skb_pull(skb, skb_network_offset(skb));
2482 skb->pkt_type = PACKET_LOOPBACK;
2483 skb->ip_summed = CHECKSUM_UNNECESSARY;
2484 WARN_ON(!skb_dst(skb));
2489 EXPORT_SYMBOL(dev_loopback_xmit);
2492 * dev_queue_xmit - transmit a buffer
2493 * @skb: buffer to transmit
2495 * Queue a buffer for transmission to a network device. The caller must
2496 * have set the device and priority and built the buffer before calling
2497 * this function. The function can be called from an interrupt.
2499 * A negative errno code is returned on a failure. A success does not
2500 * guarantee the frame will be transmitted as it may be dropped due
2501 * to congestion or traffic shaping.
2503 * -----------------------------------------------------------------------------------
2504 * I notice this method can also return errors from the queue disciplines,
2505 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2508 * Regardless of the return value, the skb is consumed, so it is currently
2509 * difficult to retry a send to this method. (You can bump the ref count
2510 * before sending to hold a reference for retry if you are careful.)
2512 * When calling this method, interrupts MUST be enabled. This is because
2513 * the BH enable code must have IRQs enabled so that it will not deadlock.
2516 int dev_queue_xmit(struct sk_buff *skb)
2518 struct net_device *dev = skb->dev;
2519 struct netdev_queue *txq;
2523 /* Disable soft irqs for various locks below. Also
2524 * stops preemption for RCU.
2528 skb_update_prio(skb);
2530 txq = dev_pick_tx(dev, skb);
2531 q = rcu_dereference_bh(txq->qdisc);
2533 #ifdef CONFIG_NET_CLS_ACT
2534 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2536 trace_net_dev_queue(skb);
2538 rc = __dev_xmit_skb(skb, q, dev, txq);
2542 /* The device has no queue. Common case for software devices:
2543 loopback, all the sorts of tunnels...
2545 Really, it is unlikely that netif_tx_lock protection is necessary
2546 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2548 However, it is possible, that they rely on protection
2551 Check this and shot the lock. It is not prone from deadlocks.
2552 Either shot noqueue qdisc, it is even simpler 8)
2554 if (dev->flags & IFF_UP) {
2555 int cpu = smp_processor_id(); /* ok because BHs are off */
2557 if (txq->xmit_lock_owner != cpu) {
2559 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2560 goto recursion_alert;
2562 HARD_TX_LOCK(dev, txq, cpu);
2564 if (!netif_xmit_stopped(txq)) {
2565 __this_cpu_inc(xmit_recursion);
2566 rc = dev_hard_start_xmit(skb, dev, txq);
2567 __this_cpu_dec(xmit_recursion);
2568 if (dev_xmit_complete(rc)) {
2569 HARD_TX_UNLOCK(dev, txq);
2573 HARD_TX_UNLOCK(dev, txq);
2574 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2577 /* Recursion is detected! It is possible,
2581 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2587 rcu_read_unlock_bh();
2592 rcu_read_unlock_bh();
2595 EXPORT_SYMBOL(dev_queue_xmit);
2598 /*=======================================================================
2600 =======================================================================*/
2602 int netdev_max_backlog __read_mostly = 1000;
2603 int netdev_tstamp_prequeue __read_mostly = 1;
2604 int netdev_budget __read_mostly = 300;
2605 int weight_p __read_mostly = 64; /* old backlog weight */
2607 /* Called with irq disabled */
2608 static inline void ____napi_schedule(struct softnet_data *sd,
2609 struct napi_struct *napi)
2611 list_add_tail(&napi->poll_list, &sd->poll_list);
2612 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2616 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2617 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2618 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2619 * if hash is a canonical 4-tuple hash over transport ports.
2621 void __skb_get_rxhash(struct sk_buff *skb)
2623 struct flow_keys keys;
2626 if (!skb_flow_dissect(skb, &keys))
2630 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2631 swap(keys.port16[0], keys.port16[1]);
2635 /* get a consistent hash (same value on both flow directions) */
2636 if ((__force u32)keys.dst < (__force u32)keys.src)
2637 swap(keys.dst, keys.src);
2639 hash = jhash_3words((__force u32)keys.dst,
2640 (__force u32)keys.src,
2641 (__force u32)keys.ports, hashrnd);
2647 EXPORT_SYMBOL(__skb_get_rxhash);
2651 /* One global table that all flow-based protocols share. */
2652 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2653 EXPORT_SYMBOL(rps_sock_flow_table);
2655 struct static_key rps_needed __read_mostly;
2657 static struct rps_dev_flow *
2658 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2659 struct rps_dev_flow *rflow, u16 next_cpu)
2661 if (next_cpu != RPS_NO_CPU) {
2662 #ifdef CONFIG_RFS_ACCEL
2663 struct netdev_rx_queue *rxqueue;
2664 struct rps_dev_flow_table *flow_table;
2665 struct rps_dev_flow *old_rflow;
2670 /* Should we steer this flow to a different hardware queue? */
2671 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2672 !(dev->features & NETIF_F_NTUPLE))
2674 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2675 if (rxq_index == skb_get_rx_queue(skb))
2678 rxqueue = dev->_rx + rxq_index;
2679 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2682 flow_id = skb->rxhash & flow_table->mask;
2683 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2684 rxq_index, flow_id);
2688 rflow = &flow_table->flows[flow_id];
2690 if (old_rflow->filter == rflow->filter)
2691 old_rflow->filter = RPS_NO_FILTER;
2695 per_cpu(softnet_data, next_cpu).input_queue_head;
2698 rflow->cpu = next_cpu;
2703 * get_rps_cpu is called from netif_receive_skb and returns the target
2704 * CPU from the RPS map of the receiving queue for a given skb.
2705 * rcu_read_lock must be held on entry.
2707 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2708 struct rps_dev_flow **rflowp)
2710 struct netdev_rx_queue *rxqueue;
2711 struct rps_map *map;
2712 struct rps_dev_flow_table *flow_table;
2713 struct rps_sock_flow_table *sock_flow_table;
2717 if (skb_rx_queue_recorded(skb)) {
2718 u16 index = skb_get_rx_queue(skb);
2719 if (unlikely(index >= dev->real_num_rx_queues)) {
2720 WARN_ONCE(dev->real_num_rx_queues > 1,
2721 "%s received packet on queue %u, but number "
2722 "of RX queues is %u\n",
2723 dev->name, index, dev->real_num_rx_queues);
2726 rxqueue = dev->_rx + index;
2730 map = rcu_dereference(rxqueue->rps_map);
2732 if (map->len == 1 &&
2733 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2734 tcpu = map->cpus[0];
2735 if (cpu_online(tcpu))
2739 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2743 skb_reset_network_header(skb);
2744 if (!skb_get_rxhash(skb))
2747 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2748 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2749 if (flow_table && sock_flow_table) {
2751 struct rps_dev_flow *rflow;
2753 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2756 next_cpu = sock_flow_table->ents[skb->rxhash &
2757 sock_flow_table->mask];
2760 * If the desired CPU (where last recvmsg was done) is
2761 * different from current CPU (one in the rx-queue flow
2762 * table entry), switch if one of the following holds:
2763 * - Current CPU is unset (equal to RPS_NO_CPU).
2764 * - Current CPU is offline.
2765 * - The current CPU's queue tail has advanced beyond the
2766 * last packet that was enqueued using this table entry.
2767 * This guarantees that all previous packets for the flow
2768 * have been dequeued, thus preserving in order delivery.
2770 if (unlikely(tcpu != next_cpu) &&
2771 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2772 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2773 rflow->last_qtail)) >= 0))
2774 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2776 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2784 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2786 if (cpu_online(tcpu)) {
2796 #ifdef CONFIG_RFS_ACCEL
2799 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2800 * @dev: Device on which the filter was set
2801 * @rxq_index: RX queue index
2802 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2803 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2805 * Drivers that implement ndo_rx_flow_steer() should periodically call
2806 * this function for each installed filter and remove the filters for
2807 * which it returns %true.
2809 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2810 u32 flow_id, u16 filter_id)
2812 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2813 struct rps_dev_flow_table *flow_table;
2814 struct rps_dev_flow *rflow;
2819 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2820 if (flow_table && flow_id <= flow_table->mask) {
2821 rflow = &flow_table->flows[flow_id];
2822 cpu = ACCESS_ONCE(rflow->cpu);
2823 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2824 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2825 rflow->last_qtail) <
2826 (int)(10 * flow_table->mask)))
2832 EXPORT_SYMBOL(rps_may_expire_flow);
2834 #endif /* CONFIG_RFS_ACCEL */
2836 /* Called from hardirq (IPI) context */
2837 static void rps_trigger_softirq(void *data)
2839 struct softnet_data *sd = data;
2841 ____napi_schedule(sd, &sd->backlog);
2845 #endif /* CONFIG_RPS */
2848 * Check if this softnet_data structure is another cpu one
2849 * If yes, queue it to our IPI list and return 1
2852 static int rps_ipi_queued(struct softnet_data *sd)
2855 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2858 sd->rps_ipi_next = mysd->rps_ipi_list;
2859 mysd->rps_ipi_list = sd;
2861 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2864 #endif /* CONFIG_RPS */
2869 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2870 * queue (may be a remote CPU queue).
2872 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2873 unsigned int *qtail)
2875 struct softnet_data *sd;
2876 unsigned long flags;
2878 sd = &per_cpu(softnet_data, cpu);
2880 local_irq_save(flags);
2883 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2884 if (skb_queue_len(&sd->input_pkt_queue)) {
2886 __skb_queue_tail(&sd->input_pkt_queue, skb);
2887 input_queue_tail_incr_save(sd, qtail);
2889 local_irq_restore(flags);
2890 return NET_RX_SUCCESS;
2893 /* Schedule NAPI for backlog device
2894 * We can use non atomic operation since we own the queue lock
2896 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2897 if (!rps_ipi_queued(sd))
2898 ____napi_schedule(sd, &sd->backlog);
2906 local_irq_restore(flags);
2908 atomic_long_inc(&skb->dev->rx_dropped);
2914 * netif_rx - post buffer to the network code
2915 * @skb: buffer to post
2917 * This function receives a packet from a device driver and queues it for
2918 * the upper (protocol) levels to process. It always succeeds. The buffer
2919 * may be dropped during processing for congestion control or by the
2923 * NET_RX_SUCCESS (no congestion)
2924 * NET_RX_DROP (packet was dropped)
2928 int netif_rx(struct sk_buff *skb)
2932 /* if netpoll wants it, pretend we never saw it */
2933 if (netpoll_rx(skb))
2936 net_timestamp_check(netdev_tstamp_prequeue, skb);
2938 trace_netif_rx(skb);
2940 if (static_key_false(&rps_needed)) {
2941 struct rps_dev_flow voidflow, *rflow = &voidflow;
2947 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2949 cpu = smp_processor_id();
2951 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2959 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2964 EXPORT_SYMBOL(netif_rx);
2966 int netif_rx_ni(struct sk_buff *skb)
2971 err = netif_rx(skb);
2972 if (local_softirq_pending())
2978 EXPORT_SYMBOL(netif_rx_ni);
2980 static void net_tx_action(struct softirq_action *h)
2982 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2984 if (sd->completion_queue) {
2985 struct sk_buff *clist;
2987 local_irq_disable();
2988 clist = sd->completion_queue;
2989 sd->completion_queue = NULL;
2993 struct sk_buff *skb = clist;
2994 clist = clist->next;
2996 WARN_ON(atomic_read(&skb->users));
2997 trace_kfree_skb(skb, net_tx_action);
3002 if (sd->output_queue) {
3005 local_irq_disable();
3006 head = sd->output_queue;
3007 sd->output_queue = NULL;
3008 sd->output_queue_tailp = &sd->output_queue;
3012 struct Qdisc *q = head;
3013 spinlock_t *root_lock;
3015 head = head->next_sched;
3017 root_lock = qdisc_lock(q);
3018 if (spin_trylock(root_lock)) {
3019 smp_mb__before_clear_bit();
3020 clear_bit(__QDISC_STATE_SCHED,
3023 spin_unlock(root_lock);
3025 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3027 __netif_reschedule(q);
3029 smp_mb__before_clear_bit();
3030 clear_bit(__QDISC_STATE_SCHED,
3038 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3039 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3040 /* This hook is defined here for ATM LANE */
3041 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3042 unsigned char *addr) __read_mostly;
3043 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3046 #ifdef CONFIG_NET_CLS_ACT
3047 /* TODO: Maybe we should just force sch_ingress to be compiled in
3048 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3049 * a compare and 2 stores extra right now if we dont have it on
3050 * but have CONFIG_NET_CLS_ACT
3051 * NOTE: This doesn't stop any functionality; if you dont have
3052 * the ingress scheduler, you just can't add policies on ingress.
3055 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3057 struct net_device *dev = skb->dev;
3058 u32 ttl = G_TC_RTTL(skb->tc_verd);
3059 int result = TC_ACT_OK;
3062 if (unlikely(MAX_RED_LOOP < ttl++)) {
3063 net_warn_ratelimited("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))
3527 case GRO_MERGED_FREE:
3528 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3529 kmem_cache_free(skbuff_head_cache, skb);
3541 EXPORT_SYMBOL(napi_skb_finish);
3543 void skb_gro_reset_offset(struct sk_buff *skb)
3545 NAPI_GRO_CB(skb)->data_offset = 0;
3546 NAPI_GRO_CB(skb)->frag0 = NULL;
3547 NAPI_GRO_CB(skb)->frag0_len = 0;
3549 if (skb->mac_header == skb->tail &&
3550 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3551 NAPI_GRO_CB(skb)->frag0 =
3552 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3553 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3556 EXPORT_SYMBOL(skb_gro_reset_offset);
3558 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3560 skb_gro_reset_offset(skb);
3562 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3564 EXPORT_SYMBOL(napi_gro_receive);
3566 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3568 __skb_pull(skb, skb_headlen(skb));
3569 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3570 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3572 skb->dev = napi->dev;
3578 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3580 struct sk_buff *skb = napi->skb;
3583 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3589 EXPORT_SYMBOL(napi_get_frags);
3591 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3597 skb->protocol = eth_type_trans(skb, skb->dev);
3599 if (ret == GRO_HELD)
3600 skb_gro_pull(skb, -ETH_HLEN);
3601 else if (netif_receive_skb(skb))
3606 case GRO_MERGED_FREE:
3607 napi_reuse_skb(napi, skb);
3616 EXPORT_SYMBOL(napi_frags_finish);
3618 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3620 struct sk_buff *skb = napi->skb;
3627 skb_reset_mac_header(skb);
3628 skb_gro_reset_offset(skb);
3630 off = skb_gro_offset(skb);
3631 hlen = off + sizeof(*eth);
3632 eth = skb_gro_header_fast(skb, off);
3633 if (skb_gro_header_hard(skb, hlen)) {
3634 eth = skb_gro_header_slow(skb, hlen, off);
3635 if (unlikely(!eth)) {
3636 napi_reuse_skb(napi, skb);
3642 skb_gro_pull(skb, sizeof(*eth));
3645 * This works because the only protocols we care about don't require
3646 * special handling. We'll fix it up properly at the end.
3648 skb->protocol = eth->h_proto;
3654 gro_result_t napi_gro_frags(struct napi_struct *napi)
3656 struct sk_buff *skb = napi_frags_skb(napi);
3661 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3663 EXPORT_SYMBOL(napi_gro_frags);
3666 * net_rps_action sends any pending IPI's for rps.
3667 * Note: called with local irq disabled, but exits with local irq enabled.
3669 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3672 struct softnet_data *remsd = sd->rps_ipi_list;
3675 sd->rps_ipi_list = NULL;
3679 /* Send pending IPI's to kick RPS processing on remote cpus. */
3681 struct softnet_data *next = remsd->rps_ipi_next;
3683 if (cpu_online(remsd->cpu))
3684 __smp_call_function_single(remsd->cpu,
3693 static int process_backlog(struct napi_struct *napi, int quota)
3696 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3699 /* Check if we have pending ipi, its better to send them now,
3700 * not waiting net_rx_action() end.
3702 if (sd->rps_ipi_list) {
3703 local_irq_disable();
3704 net_rps_action_and_irq_enable(sd);
3707 napi->weight = weight_p;
3708 local_irq_disable();
3709 while (work < quota) {
3710 struct sk_buff *skb;
3713 while ((skb = __skb_dequeue(&sd->process_queue))) {
3715 __netif_receive_skb(skb);
3716 local_irq_disable();
3717 input_queue_head_incr(sd);
3718 if (++work >= quota) {
3725 qlen = skb_queue_len(&sd->input_pkt_queue);
3727 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3728 &sd->process_queue);
3730 if (qlen < quota - work) {
3732 * Inline a custom version of __napi_complete().
3733 * only current cpu owns and manipulates this napi,
3734 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3735 * we can use a plain write instead of clear_bit(),
3736 * and we dont need an smp_mb() memory barrier.
3738 list_del(&napi->poll_list);
3741 quota = work + qlen;
3751 * __napi_schedule - schedule for receive
3752 * @n: entry to schedule
3754 * The entry's receive function will be scheduled to run
3756 void __napi_schedule(struct napi_struct *n)
3758 unsigned long flags;
3760 local_irq_save(flags);
3761 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3762 local_irq_restore(flags);
3764 EXPORT_SYMBOL(__napi_schedule);
3766 void __napi_complete(struct napi_struct *n)
3768 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3769 BUG_ON(n->gro_list);
3771 list_del(&n->poll_list);
3772 smp_mb__before_clear_bit();
3773 clear_bit(NAPI_STATE_SCHED, &n->state);
3775 EXPORT_SYMBOL(__napi_complete);
3777 void napi_complete(struct napi_struct *n)
3779 unsigned long flags;
3782 * don't let napi dequeue from the cpu poll list
3783 * just in case its running on a different cpu
3785 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3789 local_irq_save(flags);
3791 local_irq_restore(flags);
3793 EXPORT_SYMBOL(napi_complete);
3795 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3796 int (*poll)(struct napi_struct *, int), int weight)
3798 INIT_LIST_HEAD(&napi->poll_list);
3799 napi->gro_count = 0;
3800 napi->gro_list = NULL;
3803 napi->weight = weight;
3804 list_add(&napi->dev_list, &dev->napi_list);
3806 #ifdef CONFIG_NETPOLL
3807 spin_lock_init(&napi->poll_lock);
3808 napi->poll_owner = -1;
3810 set_bit(NAPI_STATE_SCHED, &napi->state);
3812 EXPORT_SYMBOL(netif_napi_add);
3814 void netif_napi_del(struct napi_struct *napi)
3816 struct sk_buff *skb, *next;
3818 list_del_init(&napi->dev_list);
3819 napi_free_frags(napi);
3821 for (skb = napi->gro_list; skb; skb = next) {
3827 napi->gro_list = NULL;
3828 napi->gro_count = 0;
3830 EXPORT_SYMBOL(netif_napi_del);
3832 static void net_rx_action(struct softirq_action *h)
3834 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3835 unsigned long time_limit = jiffies + 2;
3836 int budget = netdev_budget;
3839 local_irq_disable();
3841 while (!list_empty(&sd->poll_list)) {
3842 struct napi_struct *n;
3845 /* If softirq window is exhuasted then punt.
3846 * Allow this to run for 2 jiffies since which will allow
3847 * an average latency of 1.5/HZ.
3849 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3854 /* Even though interrupts have been re-enabled, this
3855 * access is safe because interrupts can only add new
3856 * entries to the tail of this list, and only ->poll()
3857 * calls can remove this head entry from the list.
3859 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3861 have = netpoll_poll_lock(n);
3865 /* This NAPI_STATE_SCHED test is for avoiding a race
3866 * with netpoll's poll_napi(). Only the entity which
3867 * obtains the lock and sees NAPI_STATE_SCHED set will
3868 * actually make the ->poll() call. Therefore we avoid
3869 * accidentally calling ->poll() when NAPI is not scheduled.
3872 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3873 work = n->poll(n, weight);
3877 WARN_ON_ONCE(work > weight);
3881 local_irq_disable();
3883 /* Drivers must not modify the NAPI state if they
3884 * consume the entire weight. In such cases this code
3885 * still "owns" the NAPI instance and therefore can
3886 * move the instance around on the list at-will.
3888 if (unlikely(work == weight)) {
3889 if (unlikely(napi_disable_pending(n))) {
3892 local_irq_disable();
3894 list_move_tail(&n->poll_list, &sd->poll_list);
3897 netpoll_poll_unlock(have);
3900 net_rps_action_and_irq_enable(sd);
3902 #ifdef CONFIG_NET_DMA
3904 * There may not be any more sk_buffs coming right now, so push
3905 * any pending DMA copies to hardware
3907 dma_issue_pending_all();
3914 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3918 static gifconf_func_t *gifconf_list[NPROTO];
3921 * register_gifconf - register a SIOCGIF handler
3922 * @family: Address family
3923 * @gifconf: Function handler
3925 * Register protocol dependent address dumping routines. The handler
3926 * that is passed must not be freed or reused until it has been replaced
3927 * by another handler.
3929 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3931 if (family >= NPROTO)
3933 gifconf_list[family] = gifconf;
3936 EXPORT_SYMBOL(register_gifconf);
3940 * Map an interface index to its name (SIOCGIFNAME)
3944 * We need this ioctl for efficient implementation of the
3945 * if_indextoname() function required by the IPv6 API. Without
3946 * it, we would have to search all the interfaces to find a
3950 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3952 struct net_device *dev;
3956 * Fetch the caller's info block.
3959 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3963 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3969 strcpy(ifr.ifr_name, dev->name);
3972 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3978 * Perform a SIOCGIFCONF call. This structure will change
3979 * size eventually, and there is nothing I can do about it.
3980 * Thus we will need a 'compatibility mode'.
3983 static int dev_ifconf(struct net *net, char __user *arg)
3986 struct net_device *dev;
3993 * Fetch the caller's info block.
3996 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4003 * Loop over the interfaces, and write an info block for each.
4007 for_each_netdev(net, dev) {
4008 for (i = 0; i < NPROTO; i++) {
4009 if (gifconf_list[i]) {
4012 done = gifconf_list[i](dev, NULL, 0);
4014 done = gifconf_list[i](dev, pos + total,
4024 * All done. Write the updated control block back to the caller.
4026 ifc.ifc_len = total;
4029 * Both BSD and Solaris return 0 here, so we do too.
4031 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4034 #ifdef CONFIG_PROC_FS
4036 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4038 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4039 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4040 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4042 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
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 = 0, offset = get_offset(*pos);
4050 h = &net->dev_name_head[get_bucket(*pos)];
4051 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4052 if (++count == offset)
4059 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4061 struct net_device *dev;
4062 unsigned int bucket;
4065 dev = dev_from_same_bucket(seq, pos);
4069 bucket = get_bucket(*pos) + 1;
4070 *pos = set_bucket_offset(bucket, 1);
4071 } while (bucket < NETDEV_HASHENTRIES);
4077 * This is invoked by the /proc filesystem handler to display a device
4080 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4085 return SEQ_START_TOKEN;
4087 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4090 return dev_from_bucket(seq, pos);
4093 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4096 return dev_from_bucket(seq, pos);
4099 void dev_seq_stop(struct seq_file *seq, void *v)
4105 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4107 struct rtnl_link_stats64 temp;
4108 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4110 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4111 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4112 dev->name, stats->rx_bytes, stats->rx_packets,
4114 stats->rx_dropped + stats->rx_missed_errors,
4115 stats->rx_fifo_errors,
4116 stats->rx_length_errors + stats->rx_over_errors +
4117 stats->rx_crc_errors + stats->rx_frame_errors,
4118 stats->rx_compressed, stats->multicast,
4119 stats->tx_bytes, stats->tx_packets,
4120 stats->tx_errors, stats->tx_dropped,
4121 stats->tx_fifo_errors, stats->collisions,
4122 stats->tx_carrier_errors +
4123 stats->tx_aborted_errors +
4124 stats->tx_window_errors +
4125 stats->tx_heartbeat_errors,
4126 stats->tx_compressed);
4130 * Called from the PROCfs module. This now uses the new arbitrary sized
4131 * /proc/net interface to create /proc/net/dev
4133 static int dev_seq_show(struct seq_file *seq, void *v)
4135 if (v == SEQ_START_TOKEN)
4136 seq_puts(seq, "Inter-| Receive "
4138 " face |bytes packets errs drop fifo frame "
4139 "compressed multicast|bytes packets errs "
4140 "drop fifo colls carrier compressed\n");
4142 dev_seq_printf_stats(seq, v);
4146 static struct softnet_data *softnet_get_online(loff_t *pos)
4148 struct softnet_data *sd = NULL;
4150 while (*pos < nr_cpu_ids)
4151 if (cpu_online(*pos)) {
4152 sd = &per_cpu(softnet_data, *pos);
4159 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4161 return softnet_get_online(pos);
4164 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4167 return softnet_get_online(pos);
4170 static void softnet_seq_stop(struct seq_file *seq, void *v)
4174 static int softnet_seq_show(struct seq_file *seq, void *v)
4176 struct softnet_data *sd = v;
4178 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4179 sd->processed, sd->dropped, sd->time_squeeze, 0,
4180 0, 0, 0, 0, /* was fastroute */
4181 sd->cpu_collision, sd->received_rps);
4185 static const struct seq_operations dev_seq_ops = {
4186 .start = dev_seq_start,
4187 .next = dev_seq_next,
4188 .stop = dev_seq_stop,
4189 .show = dev_seq_show,
4192 static int dev_seq_open(struct inode *inode, struct file *file)
4194 return seq_open_net(inode, file, &dev_seq_ops,
4195 sizeof(struct seq_net_private));
4198 static const struct file_operations dev_seq_fops = {
4199 .owner = THIS_MODULE,
4200 .open = dev_seq_open,
4202 .llseek = seq_lseek,
4203 .release = seq_release_net,
4206 static const struct seq_operations softnet_seq_ops = {
4207 .start = softnet_seq_start,
4208 .next = softnet_seq_next,
4209 .stop = softnet_seq_stop,
4210 .show = softnet_seq_show,
4213 static int softnet_seq_open(struct inode *inode, struct file *file)
4215 return seq_open(file, &softnet_seq_ops);
4218 static const struct file_operations softnet_seq_fops = {
4219 .owner = THIS_MODULE,
4220 .open = softnet_seq_open,
4222 .llseek = seq_lseek,
4223 .release = seq_release,
4226 static void *ptype_get_idx(loff_t pos)
4228 struct packet_type *pt = NULL;
4232 list_for_each_entry_rcu(pt, &ptype_all, list) {
4238 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4239 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4248 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4252 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4255 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4257 struct packet_type *pt;
4258 struct list_head *nxt;
4262 if (v == SEQ_START_TOKEN)
4263 return ptype_get_idx(0);
4266 nxt = pt->list.next;
4267 if (pt->type == htons(ETH_P_ALL)) {
4268 if (nxt != &ptype_all)
4271 nxt = ptype_base[0].next;
4273 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4275 while (nxt == &ptype_base[hash]) {
4276 if (++hash >= PTYPE_HASH_SIZE)
4278 nxt = ptype_base[hash].next;
4281 return list_entry(nxt, struct packet_type, list);
4284 static void ptype_seq_stop(struct seq_file *seq, void *v)
4290 static int ptype_seq_show(struct seq_file *seq, void *v)
4292 struct packet_type *pt = v;
4294 if (v == SEQ_START_TOKEN)
4295 seq_puts(seq, "Type Device Function\n");
4296 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4297 if (pt->type == htons(ETH_P_ALL))
4298 seq_puts(seq, "ALL ");
4300 seq_printf(seq, "%04x", ntohs(pt->type));
4302 seq_printf(seq, " %-8s %pF\n",
4303 pt->dev ? pt->dev->name : "", pt->func);
4309 static const struct seq_operations ptype_seq_ops = {
4310 .start = ptype_seq_start,
4311 .next = ptype_seq_next,
4312 .stop = ptype_seq_stop,
4313 .show = ptype_seq_show,
4316 static int ptype_seq_open(struct inode *inode, struct file *file)
4318 return seq_open_net(inode, file, &ptype_seq_ops,
4319 sizeof(struct seq_net_private));
4322 static const struct file_operations ptype_seq_fops = {
4323 .owner = THIS_MODULE,
4324 .open = ptype_seq_open,
4326 .llseek = seq_lseek,
4327 .release = seq_release_net,
4331 static int __net_init dev_proc_net_init(struct net *net)
4335 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4337 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4339 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4342 if (wext_proc_init(net))
4348 proc_net_remove(net, "ptype");
4350 proc_net_remove(net, "softnet_stat");
4352 proc_net_remove(net, "dev");
4356 static void __net_exit dev_proc_net_exit(struct net *net)
4358 wext_proc_exit(net);
4360 proc_net_remove(net, "ptype");
4361 proc_net_remove(net, "softnet_stat");
4362 proc_net_remove(net, "dev");
4365 static struct pernet_operations __net_initdata dev_proc_ops = {
4366 .init = dev_proc_net_init,
4367 .exit = dev_proc_net_exit,
4370 static int __init dev_proc_init(void)
4372 return register_pernet_subsys(&dev_proc_ops);
4375 #define dev_proc_init() 0
4376 #endif /* CONFIG_PROC_FS */
4380 * netdev_set_master - set up master pointer
4381 * @slave: slave device
4382 * @master: new master device
4384 * Changes the master device of the slave. Pass %NULL to break the
4385 * bonding. The caller must hold the RTNL semaphore. On a failure
4386 * a negative errno code is returned. On success the reference counts
4387 * are adjusted and the function returns zero.
4389 int netdev_set_master(struct net_device *slave, struct net_device *master)
4391 struct net_device *old = slave->master;
4401 slave->master = master;
4407 EXPORT_SYMBOL(netdev_set_master);
4410 * netdev_set_bond_master - set up bonding master/slave pair
4411 * @slave: slave device
4412 * @master: new master device
4414 * Changes the master device of the slave. Pass %NULL to break the
4415 * bonding. The caller must hold the RTNL semaphore. On a failure
4416 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4417 * to the routing socket and the function returns zero.
4419 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4425 err = netdev_set_master(slave, master);
4429 slave->flags |= IFF_SLAVE;
4431 slave->flags &= ~IFF_SLAVE;
4433 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4436 EXPORT_SYMBOL(netdev_set_bond_master);
4438 static void dev_change_rx_flags(struct net_device *dev, int flags)
4440 const struct net_device_ops *ops = dev->netdev_ops;
4442 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4443 ops->ndo_change_rx_flags(dev, flags);
4446 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4448 unsigned int old_flags = dev->flags;
4454 dev->flags |= IFF_PROMISC;
4455 dev->promiscuity += inc;
4456 if (dev->promiscuity == 0) {
4459 * If inc causes overflow, untouch promisc and return error.
4462 dev->flags &= ~IFF_PROMISC;
4464 dev->promiscuity -= inc;
4465 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4470 if (dev->flags != old_flags) {
4471 pr_info("device %s %s promiscuous mode\n",
4473 dev->flags & IFF_PROMISC ? "entered" : "left");
4474 if (audit_enabled) {
4475 current_uid_gid(&uid, &gid);
4476 audit_log(current->audit_context, GFP_ATOMIC,
4477 AUDIT_ANOM_PROMISCUOUS,
4478 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4479 dev->name, (dev->flags & IFF_PROMISC),
4480 (old_flags & IFF_PROMISC),
4481 audit_get_loginuid(current),
4483 audit_get_sessionid(current));
4486 dev_change_rx_flags(dev, IFF_PROMISC);
4492 * dev_set_promiscuity - update promiscuity count on a device
4496 * Add or remove promiscuity from a device. While the count in the device
4497 * remains above zero the interface remains promiscuous. Once it hits zero
4498 * the device reverts back to normal filtering operation. A negative inc
4499 * value is used to drop promiscuity on the device.
4500 * Return 0 if successful or a negative errno code on error.
4502 int dev_set_promiscuity(struct net_device *dev, int inc)
4504 unsigned int old_flags = dev->flags;
4507 err = __dev_set_promiscuity(dev, inc);
4510 if (dev->flags != old_flags)
4511 dev_set_rx_mode(dev);
4514 EXPORT_SYMBOL(dev_set_promiscuity);
4517 * dev_set_allmulti - update allmulti count on a device
4521 * Add or remove reception of all multicast frames to a device. While the
4522 * count in the device remains above zero the interface remains listening
4523 * to all interfaces. Once it hits zero the device reverts back to normal
4524 * filtering operation. A negative @inc value is used to drop the counter
4525 * when releasing a resource needing all multicasts.
4526 * Return 0 if successful or a negative errno code on error.
4529 int dev_set_allmulti(struct net_device *dev, int inc)
4531 unsigned int old_flags = dev->flags;
4535 dev->flags |= IFF_ALLMULTI;
4536 dev->allmulti += inc;
4537 if (dev->allmulti == 0) {
4540 * If inc causes overflow, untouch allmulti and return error.
4543 dev->flags &= ~IFF_ALLMULTI;
4545 dev->allmulti -= inc;
4546 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4551 if (dev->flags ^ old_flags) {
4552 dev_change_rx_flags(dev, IFF_ALLMULTI);
4553 dev_set_rx_mode(dev);
4557 EXPORT_SYMBOL(dev_set_allmulti);
4560 * Upload unicast and multicast address lists to device and
4561 * configure RX filtering. When the device doesn't support unicast
4562 * filtering it is put in promiscuous mode while unicast addresses
4565 void __dev_set_rx_mode(struct net_device *dev)
4567 const struct net_device_ops *ops = dev->netdev_ops;
4569 /* dev_open will call this function so the list will stay sane. */
4570 if (!(dev->flags&IFF_UP))
4573 if (!netif_device_present(dev))
4576 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4577 /* Unicast addresses changes may only happen under the rtnl,
4578 * therefore calling __dev_set_promiscuity here is safe.
4580 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4581 __dev_set_promiscuity(dev, 1);
4582 dev->uc_promisc = true;
4583 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4584 __dev_set_promiscuity(dev, -1);
4585 dev->uc_promisc = false;
4589 if (ops->ndo_set_rx_mode)
4590 ops->ndo_set_rx_mode(dev);
4593 void dev_set_rx_mode(struct net_device *dev)
4595 netif_addr_lock_bh(dev);
4596 __dev_set_rx_mode(dev);
4597 netif_addr_unlock_bh(dev);
4601 * dev_get_flags - get flags reported to userspace
4604 * Get the combination of flag bits exported through APIs to userspace.
4606 unsigned int dev_get_flags(const struct net_device *dev)
4610 flags = (dev->flags & ~(IFF_PROMISC |
4615 (dev->gflags & (IFF_PROMISC |
4618 if (netif_running(dev)) {
4619 if (netif_oper_up(dev))
4620 flags |= IFF_RUNNING;
4621 if (netif_carrier_ok(dev))
4622 flags |= IFF_LOWER_UP;
4623 if (netif_dormant(dev))
4624 flags |= IFF_DORMANT;
4629 EXPORT_SYMBOL(dev_get_flags);
4631 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4633 unsigned int old_flags = dev->flags;
4639 * Set the flags on our device.
4642 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4643 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4645 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4649 * Load in the correct multicast list now the flags have changed.
4652 if ((old_flags ^ flags) & IFF_MULTICAST)
4653 dev_change_rx_flags(dev, IFF_MULTICAST);
4655 dev_set_rx_mode(dev);
4658 * Have we downed the interface. We handle IFF_UP ourselves
4659 * according to user attempts to set it, rather than blindly
4664 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4665 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4668 dev_set_rx_mode(dev);
4671 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4672 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4674 dev->gflags ^= IFF_PROMISC;
4675 dev_set_promiscuity(dev, inc);
4678 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4679 is important. Some (broken) drivers set IFF_PROMISC, when
4680 IFF_ALLMULTI is requested not asking us and not reporting.
4682 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4683 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4685 dev->gflags ^= IFF_ALLMULTI;
4686 dev_set_allmulti(dev, inc);
4692 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4694 unsigned int changes = dev->flags ^ old_flags;
4696 if (changes & IFF_UP) {
4697 if (dev->flags & IFF_UP)
4698 call_netdevice_notifiers(NETDEV_UP, dev);
4700 call_netdevice_notifiers(NETDEV_DOWN, dev);
4703 if (dev->flags & IFF_UP &&
4704 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4705 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4709 * dev_change_flags - change device settings
4711 * @flags: device state flags
4713 * Change settings on device based state flags. The flags are
4714 * in the userspace exported format.
4716 int dev_change_flags(struct net_device *dev, unsigned int flags)
4719 unsigned int changes, old_flags = dev->flags;
4721 ret = __dev_change_flags(dev, flags);
4725 changes = old_flags ^ dev->flags;
4727 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4729 __dev_notify_flags(dev, old_flags);
4732 EXPORT_SYMBOL(dev_change_flags);
4735 * dev_set_mtu - Change maximum transfer unit
4737 * @new_mtu: new transfer unit
4739 * Change the maximum transfer size of the network device.
4741 int dev_set_mtu(struct net_device *dev, int new_mtu)
4743 const struct net_device_ops *ops = dev->netdev_ops;
4746 if (new_mtu == dev->mtu)
4749 /* MTU must be positive. */
4753 if (!netif_device_present(dev))
4757 if (ops->ndo_change_mtu)
4758 err = ops->ndo_change_mtu(dev, new_mtu);
4762 if (!err && dev->flags & IFF_UP)
4763 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4766 EXPORT_SYMBOL(dev_set_mtu);
4769 * dev_set_group - Change group this device belongs to
4771 * @new_group: group this device should belong to
4773 void dev_set_group(struct net_device *dev, int new_group)
4775 dev->group = new_group;
4777 EXPORT_SYMBOL(dev_set_group);
4780 * dev_set_mac_address - Change Media Access Control Address
4784 * Change the hardware (MAC) address of the device
4786 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4788 const struct net_device_ops *ops = dev->netdev_ops;
4791 if (!ops->ndo_set_mac_address)
4793 if (sa->sa_family != dev->type)
4795 if (!netif_device_present(dev))
4797 err = ops->ndo_set_mac_address(dev, sa);
4799 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4802 EXPORT_SYMBOL(dev_set_mac_address);
4805 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4807 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4810 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4816 case SIOCGIFFLAGS: /* Get interface flags */
4817 ifr->ifr_flags = (short) dev_get_flags(dev);
4820 case SIOCGIFMETRIC: /* Get the metric on the interface
4821 (currently unused) */
4822 ifr->ifr_metric = 0;
4825 case SIOCGIFMTU: /* Get the MTU of a device */
4826 ifr->ifr_mtu = dev->mtu;
4831 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4833 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4834 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4835 ifr->ifr_hwaddr.sa_family = dev->type;
4843 ifr->ifr_map.mem_start = dev->mem_start;
4844 ifr->ifr_map.mem_end = dev->mem_end;
4845 ifr->ifr_map.base_addr = dev->base_addr;
4846 ifr->ifr_map.irq = dev->irq;
4847 ifr->ifr_map.dma = dev->dma;
4848 ifr->ifr_map.port = dev->if_port;
4852 ifr->ifr_ifindex = dev->ifindex;
4856 ifr->ifr_qlen = dev->tx_queue_len;
4860 /* dev_ioctl() should ensure this case
4872 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4874 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4877 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4878 const struct net_device_ops *ops;
4883 ops = dev->netdev_ops;
4886 case SIOCSIFFLAGS: /* Set interface flags */
4887 return dev_change_flags(dev, ifr->ifr_flags);
4889 case SIOCSIFMETRIC: /* Set the metric on the interface
4890 (currently unused) */
4893 case SIOCSIFMTU: /* Set the MTU of a device */
4894 return dev_set_mtu(dev, ifr->ifr_mtu);
4897 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4899 case SIOCSIFHWBROADCAST:
4900 if (ifr->ifr_hwaddr.sa_family != dev->type)
4902 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4903 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4904 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4908 if (ops->ndo_set_config) {
4909 if (!netif_device_present(dev))
4911 return ops->ndo_set_config(dev, &ifr->ifr_map);
4916 if (!ops->ndo_set_rx_mode ||
4917 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4919 if (!netif_device_present(dev))
4921 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4924 if (!ops->ndo_set_rx_mode ||
4925 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4927 if (!netif_device_present(dev))
4929 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4932 if (ifr->ifr_qlen < 0)
4934 dev->tx_queue_len = ifr->ifr_qlen;
4938 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4939 return dev_change_name(dev, ifr->ifr_newname);
4942 err = net_hwtstamp_validate(ifr);
4948 * Unknown or private ioctl
4951 if ((cmd >= SIOCDEVPRIVATE &&
4952 cmd <= SIOCDEVPRIVATE + 15) ||
4953 cmd == SIOCBONDENSLAVE ||
4954 cmd == SIOCBONDRELEASE ||
4955 cmd == SIOCBONDSETHWADDR ||
4956 cmd == SIOCBONDSLAVEINFOQUERY ||
4957 cmd == SIOCBONDINFOQUERY ||
4958 cmd == SIOCBONDCHANGEACTIVE ||
4959 cmd == SIOCGMIIPHY ||
4960 cmd == SIOCGMIIREG ||
4961 cmd == SIOCSMIIREG ||
4962 cmd == SIOCBRADDIF ||
4963 cmd == SIOCBRDELIF ||
4964 cmd == SIOCSHWTSTAMP ||
4965 cmd == SIOCWANDEV) {
4967 if (ops->ndo_do_ioctl) {
4968 if (netif_device_present(dev))
4969 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4981 * This function handles all "interface"-type I/O control requests. The actual
4982 * 'doing' part of this is dev_ifsioc above.
4986 * dev_ioctl - network device ioctl
4987 * @net: the applicable net namespace
4988 * @cmd: command to issue
4989 * @arg: pointer to a struct ifreq in user space
4991 * Issue ioctl functions to devices. This is normally called by the
4992 * user space syscall interfaces but can sometimes be useful for
4993 * other purposes. The return value is the return from the syscall if
4994 * positive or a negative errno code on error.
4997 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5003 /* One special case: SIOCGIFCONF takes ifconf argument
5004 and requires shared lock, because it sleeps writing
5008 if (cmd == SIOCGIFCONF) {
5010 ret = dev_ifconf(net, (char __user *) arg);
5014 if (cmd == SIOCGIFNAME)
5015 return dev_ifname(net, (struct ifreq __user *)arg);
5017 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5020 ifr.ifr_name[IFNAMSIZ-1] = 0;
5022 colon = strchr(ifr.ifr_name, ':');
5027 * See which interface the caller is talking about.
5032 * These ioctl calls:
5033 * - can be done by all.
5034 * - atomic and do not require locking.
5045 dev_load(net, ifr.ifr_name);
5047 ret = dev_ifsioc_locked(net, &ifr, cmd);
5052 if (copy_to_user(arg, &ifr,
5053 sizeof(struct ifreq)))
5059 dev_load(net, ifr.ifr_name);
5061 ret = dev_ethtool(net, &ifr);
5066 if (copy_to_user(arg, &ifr,
5067 sizeof(struct ifreq)))
5073 * These ioctl calls:
5074 * - require superuser power.
5075 * - require strict serialization.
5081 if (!capable(CAP_NET_ADMIN))
5083 dev_load(net, ifr.ifr_name);
5085 ret = dev_ifsioc(net, &ifr, cmd);
5090 if (copy_to_user(arg, &ifr,
5091 sizeof(struct ifreq)))
5097 * These ioctl calls:
5098 * - require superuser power.
5099 * - require strict serialization.
5100 * - do not return a value
5110 case SIOCSIFHWBROADCAST:
5113 case SIOCBONDENSLAVE:
5114 case SIOCBONDRELEASE:
5115 case SIOCBONDSETHWADDR:
5116 case SIOCBONDCHANGEACTIVE:
5120 if (!capable(CAP_NET_ADMIN))
5123 case SIOCBONDSLAVEINFOQUERY:
5124 case SIOCBONDINFOQUERY:
5125 dev_load(net, ifr.ifr_name);
5127 ret = dev_ifsioc(net, &ifr, cmd);
5132 /* Get the per device memory space. We can add this but
5133 * currently do not support it */
5135 /* Set the per device memory buffer space.
5136 * Not applicable in our case */
5141 * Unknown or private ioctl.
5144 if (cmd == SIOCWANDEV ||
5145 (cmd >= SIOCDEVPRIVATE &&
5146 cmd <= SIOCDEVPRIVATE + 15)) {
5147 dev_load(net, ifr.ifr_name);
5149 ret = dev_ifsioc(net, &ifr, cmd);
5151 if (!ret && copy_to_user(arg, &ifr,
5152 sizeof(struct ifreq)))
5156 /* Take care of Wireless Extensions */
5157 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5158 return wext_handle_ioctl(net, &ifr, cmd, arg);
5165 * dev_new_index - allocate an ifindex
5166 * @net: the applicable net namespace
5168 * Returns a suitable unique value for a new device interface
5169 * number. The caller must hold the rtnl semaphore or the
5170 * dev_base_lock to be sure it remains unique.
5172 static int dev_new_index(struct net *net)
5178 if (!__dev_get_by_index(net, ifindex))
5183 /* Delayed registration/unregisteration */
5184 static LIST_HEAD(net_todo_list);
5186 static void net_set_todo(struct net_device *dev)
5188 list_add_tail(&dev->todo_list, &net_todo_list);
5191 static void rollback_registered_many(struct list_head *head)
5193 struct net_device *dev, *tmp;
5195 BUG_ON(dev_boot_phase);
5198 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5199 /* Some devices call without registering
5200 * for initialization unwind. Remove those
5201 * devices and proceed with the remaining.
5203 if (dev->reg_state == NETREG_UNINITIALIZED) {
5204 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5208 list_del(&dev->unreg_list);
5211 dev->dismantle = true;
5212 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5215 /* If device is running, close it first. */
5216 dev_close_many(head);
5218 list_for_each_entry(dev, head, unreg_list) {
5219 /* And unlink it from device chain. */
5220 unlist_netdevice(dev);
5222 dev->reg_state = NETREG_UNREGISTERING;
5227 list_for_each_entry(dev, head, unreg_list) {
5228 /* Shutdown queueing discipline. */
5232 /* Notify protocols, that we are about to destroy
5233 this device. They should clean all the things.
5235 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5237 if (!dev->rtnl_link_ops ||
5238 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5239 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5242 * Flush the unicast and multicast chains
5247 if (dev->netdev_ops->ndo_uninit)
5248 dev->netdev_ops->ndo_uninit(dev);
5250 /* Notifier chain MUST detach us from master device. */
5251 WARN_ON(dev->master);
5253 /* Remove entries from kobject tree */
5254 netdev_unregister_kobject(dev);
5257 /* Process any work delayed until the end of the batch */
5258 dev = list_first_entry(head, struct net_device, unreg_list);
5259 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5263 list_for_each_entry(dev, head, unreg_list)
5267 static void rollback_registered(struct net_device *dev)
5271 list_add(&dev->unreg_list, &single);
5272 rollback_registered_many(&single);
5276 static netdev_features_t netdev_fix_features(struct net_device *dev,
5277 netdev_features_t features)
5279 /* Fix illegal checksum combinations */
5280 if ((features & NETIF_F_HW_CSUM) &&
5281 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5282 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5283 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5286 /* Fix illegal SG+CSUM combinations. */
5287 if ((features & NETIF_F_SG) &&
5288 !(features & NETIF_F_ALL_CSUM)) {
5290 "Dropping NETIF_F_SG since no checksum feature.\n");
5291 features &= ~NETIF_F_SG;
5294 /* TSO requires that SG is present as well. */
5295 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5296 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5297 features &= ~NETIF_F_ALL_TSO;
5300 /* TSO ECN requires that TSO is present as well. */
5301 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5302 features &= ~NETIF_F_TSO_ECN;
5304 /* Software GSO depends on SG. */
5305 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5306 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5307 features &= ~NETIF_F_GSO;
5310 /* UFO needs SG and checksumming */
5311 if (features & NETIF_F_UFO) {
5312 /* maybe split UFO into V4 and V6? */
5313 if (!((features & NETIF_F_GEN_CSUM) ||
5314 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5315 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5317 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5318 features &= ~NETIF_F_UFO;
5321 if (!(features & NETIF_F_SG)) {
5323 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5324 features &= ~NETIF_F_UFO;
5331 int __netdev_update_features(struct net_device *dev)
5333 netdev_features_t features;
5338 features = netdev_get_wanted_features(dev);
5340 if (dev->netdev_ops->ndo_fix_features)
5341 features = dev->netdev_ops->ndo_fix_features(dev, features);
5343 /* driver might be less strict about feature dependencies */
5344 features = netdev_fix_features(dev, features);
5346 if (dev->features == features)
5349 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5350 &dev->features, &features);
5352 if (dev->netdev_ops->ndo_set_features)
5353 err = dev->netdev_ops->ndo_set_features(dev, features);
5355 if (unlikely(err < 0)) {
5357 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5358 err, &features, &dev->features);
5363 dev->features = features;
5369 * netdev_update_features - recalculate device features
5370 * @dev: the device to check
5372 * Recalculate dev->features set and send notifications if it
5373 * has changed. Should be called after driver or hardware dependent
5374 * conditions might have changed that influence the features.
5376 void netdev_update_features(struct net_device *dev)
5378 if (__netdev_update_features(dev))
5379 netdev_features_change(dev);
5381 EXPORT_SYMBOL(netdev_update_features);
5384 * netdev_change_features - recalculate device features
5385 * @dev: the device to check
5387 * Recalculate dev->features set and send notifications even
5388 * if they have not changed. Should be called instead of
5389 * netdev_update_features() if also dev->vlan_features might
5390 * have changed to allow the changes to be propagated to stacked
5393 void netdev_change_features(struct net_device *dev)
5395 __netdev_update_features(dev);
5396 netdev_features_change(dev);
5398 EXPORT_SYMBOL(netdev_change_features);
5401 * netif_stacked_transfer_operstate - transfer operstate
5402 * @rootdev: the root or lower level device to transfer state from
5403 * @dev: the device to transfer operstate to
5405 * Transfer operational state from root to device. This is normally
5406 * called when a stacking relationship exists between the root
5407 * device and the device(a leaf device).
5409 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5410 struct net_device *dev)
5412 if (rootdev->operstate == IF_OPER_DORMANT)
5413 netif_dormant_on(dev);
5415 netif_dormant_off(dev);
5417 if (netif_carrier_ok(rootdev)) {
5418 if (!netif_carrier_ok(dev))
5419 netif_carrier_on(dev);
5421 if (netif_carrier_ok(dev))
5422 netif_carrier_off(dev);
5425 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5428 static int netif_alloc_rx_queues(struct net_device *dev)
5430 unsigned int i, count = dev->num_rx_queues;
5431 struct netdev_rx_queue *rx;
5435 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5437 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5442 for (i = 0; i < count; i++)
5448 static void netdev_init_one_queue(struct net_device *dev,
5449 struct netdev_queue *queue, void *_unused)
5451 /* Initialize queue lock */
5452 spin_lock_init(&queue->_xmit_lock);
5453 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5454 queue->xmit_lock_owner = -1;
5455 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5458 dql_init(&queue->dql, HZ);
5462 static int netif_alloc_netdev_queues(struct net_device *dev)
5464 unsigned int count = dev->num_tx_queues;
5465 struct netdev_queue *tx;
5469 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5471 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5476 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5477 spin_lock_init(&dev->tx_global_lock);
5483 * register_netdevice - register a network device
5484 * @dev: device to register
5486 * Take a completed network device structure and add it to the kernel
5487 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5488 * chain. 0 is returned on success. A negative errno code is returned
5489 * on a failure to set up the device, or if the name is a duplicate.
5491 * Callers must hold the rtnl semaphore. You may want
5492 * register_netdev() instead of this.
5495 * The locking appears insufficient to guarantee two parallel registers
5496 * will not get the same name.
5499 int register_netdevice(struct net_device *dev)
5502 struct net *net = dev_net(dev);
5504 BUG_ON(dev_boot_phase);
5509 /* When net_device's are persistent, this will be fatal. */
5510 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5513 spin_lock_init(&dev->addr_list_lock);
5514 netdev_set_addr_lockdep_class(dev);
5518 ret = dev_get_valid_name(dev, dev->name);
5522 /* Init, if this function is available */
5523 if (dev->netdev_ops->ndo_init) {
5524 ret = dev->netdev_ops->ndo_init(dev);
5532 dev->ifindex = dev_new_index(net);
5533 if (dev->iflink == -1)
5534 dev->iflink = dev->ifindex;
5536 /* Transfer changeable features to wanted_features and enable
5537 * software offloads (GSO and GRO).
5539 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5540 dev->features |= NETIF_F_SOFT_FEATURES;
5541 dev->wanted_features = dev->features & dev->hw_features;
5543 /* Turn on no cache copy if HW is doing checksum */
5544 if (!(dev->flags & IFF_LOOPBACK)) {
5545 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5546 if (dev->features & NETIF_F_ALL_CSUM) {
5547 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5548 dev->features |= NETIF_F_NOCACHE_COPY;
5552 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5554 dev->vlan_features |= NETIF_F_HIGHDMA;
5556 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5557 ret = notifier_to_errno(ret);
5561 ret = netdev_register_kobject(dev);
5564 dev->reg_state = NETREG_REGISTERED;
5566 __netdev_update_features(dev);
5569 * Default initial state at registry is that the
5570 * device is present.
5573 set_bit(__LINK_STATE_PRESENT, &dev->state);
5575 dev_init_scheduler(dev);
5577 list_netdevice(dev);
5579 /* Notify protocols, that a new device appeared. */
5580 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5581 ret = notifier_to_errno(ret);
5583 rollback_registered(dev);
5584 dev->reg_state = NETREG_UNREGISTERED;
5587 * Prevent userspace races by waiting until the network
5588 * device is fully setup before sending notifications.
5590 if (!dev->rtnl_link_ops ||
5591 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5592 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5598 if (dev->netdev_ops->ndo_uninit)
5599 dev->netdev_ops->ndo_uninit(dev);
5602 EXPORT_SYMBOL(register_netdevice);
5605 * init_dummy_netdev - init a dummy network device for NAPI
5606 * @dev: device to init
5608 * This takes a network device structure and initialize the minimum
5609 * amount of fields so it can be used to schedule NAPI polls without
5610 * registering a full blown interface. This is to be used by drivers
5611 * that need to tie several hardware interfaces to a single NAPI
5612 * poll scheduler due to HW limitations.
5614 int init_dummy_netdev(struct net_device *dev)
5616 /* Clear everything. Note we don't initialize spinlocks
5617 * are they aren't supposed to be taken by any of the
5618 * NAPI code and this dummy netdev is supposed to be
5619 * only ever used for NAPI polls
5621 memset(dev, 0, sizeof(struct net_device));
5623 /* make sure we BUG if trying to hit standard
5624 * register/unregister code path
5626 dev->reg_state = NETREG_DUMMY;
5628 /* NAPI wants this */
5629 INIT_LIST_HEAD(&dev->napi_list);
5631 /* a dummy interface is started by default */
5632 set_bit(__LINK_STATE_PRESENT, &dev->state);
5633 set_bit(__LINK_STATE_START, &dev->state);
5635 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5636 * because users of this 'device' dont need to change
5642 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5646 * register_netdev - register a network device
5647 * @dev: device to register
5649 * Take a completed network device structure and add it to the kernel
5650 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5651 * chain. 0 is returned on success. A negative errno code is returned
5652 * on a failure to set up the device, or if the name is a duplicate.
5654 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5655 * and expands the device name if you passed a format string to
5658 int register_netdev(struct net_device *dev)
5663 err = register_netdevice(dev);
5667 EXPORT_SYMBOL(register_netdev);
5669 int netdev_refcnt_read(const struct net_device *dev)
5673 for_each_possible_cpu(i)
5674 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5677 EXPORT_SYMBOL(netdev_refcnt_read);
5680 * netdev_wait_allrefs - wait until all references are gone.
5682 * This is called when unregistering network devices.
5684 * Any protocol or device that holds a reference should register
5685 * for netdevice notification, and cleanup and put back the
5686 * reference if they receive an UNREGISTER event.
5687 * We can get stuck here if buggy protocols don't correctly
5690 static void netdev_wait_allrefs(struct net_device *dev)
5692 unsigned long rebroadcast_time, warning_time;
5695 linkwatch_forget_dev(dev);
5697 rebroadcast_time = warning_time = jiffies;
5698 refcnt = netdev_refcnt_read(dev);
5700 while (refcnt != 0) {
5701 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5704 /* Rebroadcast unregister notification */
5705 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5706 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5707 * should have already handle it the first time */
5709 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5711 /* We must not have linkwatch events
5712 * pending on unregister. If this
5713 * happens, we simply run the queue
5714 * unscheduled, resulting in a noop
5717 linkwatch_run_queue();
5722 rebroadcast_time = jiffies;
5727 refcnt = netdev_refcnt_read(dev);
5729 if (time_after(jiffies, warning_time + 10 * HZ)) {
5730 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5732 warning_time = jiffies;
5741 * register_netdevice(x1);
5742 * register_netdevice(x2);
5744 * unregister_netdevice(y1);
5745 * unregister_netdevice(y2);
5751 * We are invoked by rtnl_unlock().
5752 * This allows us to deal with problems:
5753 * 1) We can delete sysfs objects which invoke hotplug
5754 * without deadlocking with linkwatch via keventd.
5755 * 2) Since we run with the RTNL semaphore not held, we can sleep
5756 * safely in order to wait for the netdev refcnt to drop to zero.
5758 * We must not return until all unregister events added during
5759 * the interval the lock was held have been completed.
5761 void netdev_run_todo(void)
5763 struct list_head list;
5765 /* Snapshot list, allow later requests */
5766 list_replace_init(&net_todo_list, &list);
5770 /* Wait for rcu callbacks to finish before attempting to drain
5771 * the device list. This usually avoids a 250ms wait.
5773 if (!list_empty(&list))
5776 while (!list_empty(&list)) {
5777 struct net_device *dev
5778 = list_first_entry(&list, struct net_device, todo_list);
5779 list_del(&dev->todo_list);
5781 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5782 pr_err("network todo '%s' but state %d\n",
5783 dev->name, dev->reg_state);
5788 dev->reg_state = NETREG_UNREGISTERED;
5790 on_each_cpu(flush_backlog, dev, 1);
5792 netdev_wait_allrefs(dev);
5795 BUG_ON(netdev_refcnt_read(dev));
5796 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5797 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5798 WARN_ON(dev->dn_ptr);
5800 if (dev->destructor)
5801 dev->destructor(dev);
5803 /* Free network device */
5804 kobject_put(&dev->dev.kobj);
5808 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5809 * fields in the same order, with only the type differing.
5811 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5812 const struct net_device_stats *netdev_stats)
5814 #if BITS_PER_LONG == 64
5815 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5816 memcpy(stats64, netdev_stats, sizeof(*stats64));
5818 size_t i, n = sizeof(*stats64) / sizeof(u64);
5819 const unsigned long *src = (const unsigned long *)netdev_stats;
5820 u64 *dst = (u64 *)stats64;
5822 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5823 sizeof(*stats64) / sizeof(u64));
5824 for (i = 0; i < n; i++)
5828 EXPORT_SYMBOL(netdev_stats_to_stats64);
5831 * dev_get_stats - get network device statistics
5832 * @dev: device to get statistics from
5833 * @storage: place to store stats
5835 * Get network statistics from device. Return @storage.
5836 * The device driver may provide its own method by setting
5837 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5838 * otherwise the internal statistics structure is used.
5840 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5841 struct rtnl_link_stats64 *storage)
5843 const struct net_device_ops *ops = dev->netdev_ops;
5845 if (ops->ndo_get_stats64) {
5846 memset(storage, 0, sizeof(*storage));
5847 ops->ndo_get_stats64(dev, storage);
5848 } else if (ops->ndo_get_stats) {
5849 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5851 netdev_stats_to_stats64(storage, &dev->stats);
5853 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5856 EXPORT_SYMBOL(dev_get_stats);
5858 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5860 struct netdev_queue *queue = dev_ingress_queue(dev);
5862 #ifdef CONFIG_NET_CLS_ACT
5865 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5868 netdev_init_one_queue(dev, queue, NULL);
5869 queue->qdisc = &noop_qdisc;
5870 queue->qdisc_sleeping = &noop_qdisc;
5871 rcu_assign_pointer(dev->ingress_queue, queue);
5877 * alloc_netdev_mqs - allocate network device
5878 * @sizeof_priv: size of private data to allocate space for
5879 * @name: device name format string
5880 * @setup: callback to initialize device
5881 * @txqs: the number of TX subqueues to allocate
5882 * @rxqs: the number of RX subqueues to allocate
5884 * Allocates a struct net_device with private data area for driver use
5885 * and performs basic initialization. Also allocates subquue structs
5886 * for each queue on the device.
5888 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5889 void (*setup)(struct net_device *),
5890 unsigned int txqs, unsigned int rxqs)
5892 struct net_device *dev;
5894 struct net_device *p;
5896 BUG_ON(strlen(name) >= sizeof(dev->name));
5899 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5905 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5910 alloc_size = sizeof(struct net_device);
5912 /* ensure 32-byte alignment of private area */
5913 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5914 alloc_size += sizeof_priv;
5916 /* ensure 32-byte alignment of whole construct */
5917 alloc_size += NETDEV_ALIGN - 1;
5919 p = kzalloc(alloc_size, GFP_KERNEL);
5921 pr_err("alloc_netdev: Unable to allocate device\n");
5925 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5926 dev->padded = (char *)dev - (char *)p;
5928 dev->pcpu_refcnt = alloc_percpu(int);
5929 if (!dev->pcpu_refcnt)
5932 if (dev_addr_init(dev))
5938 dev_net_set(dev, &init_net);
5940 dev->gso_max_size = GSO_MAX_SIZE;
5942 INIT_LIST_HEAD(&dev->napi_list);
5943 INIT_LIST_HEAD(&dev->unreg_list);
5944 INIT_LIST_HEAD(&dev->link_watch_list);
5945 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5948 dev->num_tx_queues = txqs;
5949 dev->real_num_tx_queues = txqs;
5950 if (netif_alloc_netdev_queues(dev))
5954 dev->num_rx_queues = rxqs;
5955 dev->real_num_rx_queues = rxqs;
5956 if (netif_alloc_rx_queues(dev))
5960 strcpy(dev->name, name);
5961 dev->group = INIT_NETDEV_GROUP;
5969 free_percpu(dev->pcpu_refcnt);
5979 EXPORT_SYMBOL(alloc_netdev_mqs);
5982 * free_netdev - free network device
5985 * This function does the last stage of destroying an allocated device
5986 * interface. The reference to the device object is released.
5987 * If this is the last reference then it will be freed.
5989 void free_netdev(struct net_device *dev)
5991 struct napi_struct *p, *n;
5993 release_net(dev_net(dev));
6000 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6002 /* Flush device addresses */
6003 dev_addr_flush(dev);
6005 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6008 free_percpu(dev->pcpu_refcnt);
6009 dev->pcpu_refcnt = NULL;
6011 /* Compatibility with error handling in drivers */
6012 if (dev->reg_state == NETREG_UNINITIALIZED) {
6013 kfree((char *)dev - dev->padded);
6017 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6018 dev->reg_state = NETREG_RELEASED;
6020 /* will free via device release */
6021 put_device(&dev->dev);
6023 EXPORT_SYMBOL(free_netdev);
6026 * synchronize_net - Synchronize with packet receive processing
6028 * Wait for packets currently being received to be done.
6029 * Does not block later packets from starting.
6031 void synchronize_net(void)
6034 if (rtnl_is_locked())
6035 synchronize_rcu_expedited();
6039 EXPORT_SYMBOL(synchronize_net);
6042 * unregister_netdevice_queue - remove device from the kernel
6046 * This function shuts down a device interface and removes it
6047 * from the kernel tables.
6048 * If head not NULL, device is queued to be unregistered later.
6050 * Callers must hold the rtnl semaphore. You may want
6051 * unregister_netdev() instead of this.
6054 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6059 list_move_tail(&dev->unreg_list, head);
6061 rollback_registered(dev);
6062 /* Finish processing unregister after unlock */
6066 EXPORT_SYMBOL(unregister_netdevice_queue);
6069 * unregister_netdevice_many - unregister many devices
6070 * @head: list of devices
6072 void unregister_netdevice_many(struct list_head *head)
6074 struct net_device *dev;
6076 if (!list_empty(head)) {
6077 rollback_registered_many(head);
6078 list_for_each_entry(dev, head, unreg_list)
6082 EXPORT_SYMBOL(unregister_netdevice_many);
6085 * unregister_netdev - remove device from the kernel
6088 * This function shuts down a device interface and removes it
6089 * from the kernel tables.
6091 * This is just a wrapper for unregister_netdevice that takes
6092 * the rtnl semaphore. In general you want to use this and not
6093 * unregister_netdevice.
6095 void unregister_netdev(struct net_device *dev)
6098 unregister_netdevice(dev);
6101 EXPORT_SYMBOL(unregister_netdev);
6104 * dev_change_net_namespace - move device to different nethost namespace
6106 * @net: network namespace
6107 * @pat: If not NULL name pattern to try if the current device name
6108 * is already taken in the destination network namespace.
6110 * This function shuts down a device interface and moves it
6111 * to a new network namespace. On success 0 is returned, on
6112 * a failure a netagive errno code is returned.
6114 * Callers must hold the rtnl semaphore.
6117 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6123 /* Don't allow namespace local devices to be moved. */
6125 if (dev->features & NETIF_F_NETNS_LOCAL)
6128 /* Ensure the device has been registrered */
6130 if (dev->reg_state != NETREG_REGISTERED)
6133 /* Get out if there is nothing todo */
6135 if (net_eq(dev_net(dev), net))
6138 /* Pick the destination device name, and ensure
6139 * we can use it in the destination network namespace.
6142 if (__dev_get_by_name(net, dev->name)) {
6143 /* We get here if we can't use the current device name */
6146 if (dev_get_valid_name(dev, pat) < 0)
6151 * And now a mini version of register_netdevice unregister_netdevice.
6154 /* If device is running close it first. */
6157 /* And unlink it from device chain */
6159 unlist_netdevice(dev);
6163 /* Shutdown queueing discipline. */
6166 /* Notify protocols, that we are about to destroy
6167 this device. They should clean all the things.
6169 Note that dev->reg_state stays at NETREG_REGISTERED.
6170 This is wanted because this way 8021q and macvlan know
6171 the device is just moving and can keep their slaves up.
6173 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6174 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6175 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6178 * Flush the unicast and multicast chains
6183 /* Actually switch the network namespace */
6184 dev_net_set(dev, net);
6186 /* If there is an ifindex conflict assign a new one */
6187 if (__dev_get_by_index(net, dev->ifindex)) {
6188 int iflink = (dev->iflink == dev->ifindex);
6189 dev->ifindex = dev_new_index(net);
6191 dev->iflink = dev->ifindex;
6194 /* Fixup kobjects */
6195 err = device_rename(&dev->dev, dev->name);
6198 /* Add the device back in the hashes */
6199 list_netdevice(dev);
6201 /* Notify protocols, that a new device appeared. */
6202 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6205 * Prevent userspace races by waiting until the network
6206 * device is fully setup before sending notifications.
6208 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6215 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6217 static int dev_cpu_callback(struct notifier_block *nfb,
6218 unsigned long action,
6221 struct sk_buff **list_skb;
6222 struct sk_buff *skb;
6223 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6224 struct softnet_data *sd, *oldsd;
6226 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6229 local_irq_disable();
6230 cpu = smp_processor_id();
6231 sd = &per_cpu(softnet_data, cpu);
6232 oldsd = &per_cpu(softnet_data, oldcpu);
6234 /* Find end of our completion_queue. */
6235 list_skb = &sd->completion_queue;
6237 list_skb = &(*list_skb)->next;
6238 /* Append completion queue from offline CPU. */
6239 *list_skb = oldsd->completion_queue;
6240 oldsd->completion_queue = NULL;
6242 /* Append output queue from offline CPU. */
6243 if (oldsd->output_queue) {
6244 *sd->output_queue_tailp = oldsd->output_queue;
6245 sd->output_queue_tailp = oldsd->output_queue_tailp;
6246 oldsd->output_queue = NULL;
6247 oldsd->output_queue_tailp = &oldsd->output_queue;
6249 /* Append NAPI poll list from offline CPU. */
6250 if (!list_empty(&oldsd->poll_list)) {
6251 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6252 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6255 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6258 /* Process offline CPU's input_pkt_queue */
6259 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6261 input_queue_head_incr(oldsd);
6263 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6265 input_queue_head_incr(oldsd);
6273 * netdev_increment_features - increment feature set by one
6274 * @all: current feature set
6275 * @one: new feature set
6276 * @mask: mask feature set
6278 * Computes a new feature set after adding a device with feature set
6279 * @one to the master device with current feature set @all. Will not
6280 * enable anything that is off in @mask. Returns the new feature set.
6282 netdev_features_t netdev_increment_features(netdev_features_t all,
6283 netdev_features_t one, netdev_features_t mask)
6285 if (mask & NETIF_F_GEN_CSUM)
6286 mask |= NETIF_F_ALL_CSUM;
6287 mask |= NETIF_F_VLAN_CHALLENGED;
6289 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6290 all &= one | ~NETIF_F_ALL_FOR_ALL;
6292 /* If one device supports hw checksumming, set for all. */
6293 if (all & NETIF_F_GEN_CSUM)
6294 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6298 EXPORT_SYMBOL(netdev_increment_features);
6300 static struct hlist_head *netdev_create_hash(void)
6303 struct hlist_head *hash;
6305 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6307 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6308 INIT_HLIST_HEAD(&hash[i]);
6313 /* Initialize per network namespace state */
6314 static int __net_init netdev_init(struct net *net)
6316 if (net != &init_net)
6317 INIT_LIST_HEAD(&net->dev_base_head);
6319 net->dev_name_head = netdev_create_hash();
6320 if (net->dev_name_head == NULL)
6323 net->dev_index_head = netdev_create_hash();
6324 if (net->dev_index_head == NULL)
6330 kfree(net->dev_name_head);
6336 * netdev_drivername - network driver for the device
6337 * @dev: network device
6339 * Determine network driver for device.
6341 const char *netdev_drivername(const struct net_device *dev)
6343 const struct device_driver *driver;
6344 const struct device *parent;
6345 const char *empty = "";
6347 parent = dev->dev.parent;
6351 driver = parent->driver;
6352 if (driver && driver->name)
6353 return driver->name;
6357 int __netdev_printk(const char *level, const struct net_device *dev,
6358 struct va_format *vaf)
6362 if (dev && dev->dev.parent)
6363 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6364 netdev_name(dev), vaf);
6366 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6368 r = printk("%s(NULL net_device): %pV", level, vaf);
6372 EXPORT_SYMBOL(__netdev_printk);
6374 int netdev_printk(const char *level, const struct net_device *dev,
6375 const char *format, ...)
6377 struct va_format vaf;
6381 va_start(args, format);
6386 r = __netdev_printk(level, dev, &vaf);
6391 EXPORT_SYMBOL(netdev_printk);
6393 #define define_netdev_printk_level(func, level) \
6394 int func(const struct net_device *dev, const char *fmt, ...) \
6397 struct va_format vaf; \
6400 va_start(args, fmt); \
6405 r = __netdev_printk(level, dev, &vaf); \
6410 EXPORT_SYMBOL(func);
6412 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6413 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6414 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6415 define_netdev_printk_level(netdev_err, KERN_ERR);
6416 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6417 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6418 define_netdev_printk_level(netdev_info, KERN_INFO);
6420 static void __net_exit netdev_exit(struct net *net)
6422 kfree(net->dev_name_head);
6423 kfree(net->dev_index_head);
6426 static struct pernet_operations __net_initdata netdev_net_ops = {
6427 .init = netdev_init,
6428 .exit = netdev_exit,
6431 static void __net_exit default_device_exit(struct net *net)
6433 struct net_device *dev, *aux;
6435 * Push all migratable network devices back to the
6436 * initial network namespace
6439 for_each_netdev_safe(net, dev, aux) {
6441 char fb_name[IFNAMSIZ];
6443 /* Ignore unmoveable devices (i.e. loopback) */
6444 if (dev->features & NETIF_F_NETNS_LOCAL)
6447 /* Leave virtual devices for the generic cleanup */
6448 if (dev->rtnl_link_ops)
6451 /* Push remaining network devices to init_net */
6452 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6453 err = dev_change_net_namespace(dev, &init_net, fb_name);
6455 pr_emerg("%s: failed to move %s to init_net: %d\n",
6456 __func__, dev->name, err);
6463 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6465 /* At exit all network devices most be removed from a network
6466 * namespace. Do this in the reverse order of registration.
6467 * Do this across as many network namespaces as possible to
6468 * improve batching efficiency.
6470 struct net_device *dev;
6472 LIST_HEAD(dev_kill_list);
6475 list_for_each_entry(net, net_list, exit_list) {
6476 for_each_netdev_reverse(net, dev) {
6477 if (dev->rtnl_link_ops)
6478 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6480 unregister_netdevice_queue(dev, &dev_kill_list);
6483 unregister_netdevice_many(&dev_kill_list);
6484 list_del(&dev_kill_list);
6488 static struct pernet_operations __net_initdata default_device_ops = {
6489 .exit = default_device_exit,
6490 .exit_batch = default_device_exit_batch,
6494 * Initialize the DEV module. At boot time this walks the device list and
6495 * unhooks any devices that fail to initialise (normally hardware not
6496 * present) and leaves us with a valid list of present and active devices.
6501 * This is called single threaded during boot, so no need
6502 * to take the rtnl semaphore.
6504 static int __init net_dev_init(void)
6506 int i, rc = -ENOMEM;
6508 BUG_ON(!dev_boot_phase);
6510 if (dev_proc_init())
6513 if (netdev_kobject_init())
6516 INIT_LIST_HEAD(&ptype_all);
6517 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6518 INIT_LIST_HEAD(&ptype_base[i]);
6520 if (register_pernet_subsys(&netdev_net_ops))
6524 * Initialise the packet receive queues.
6527 for_each_possible_cpu(i) {
6528 struct softnet_data *sd = &per_cpu(softnet_data, i);
6530 memset(sd, 0, sizeof(*sd));
6531 skb_queue_head_init(&sd->input_pkt_queue);
6532 skb_queue_head_init(&sd->process_queue);
6533 sd->completion_queue = NULL;
6534 INIT_LIST_HEAD(&sd->poll_list);
6535 sd->output_queue = NULL;
6536 sd->output_queue_tailp = &sd->output_queue;
6538 sd->csd.func = rps_trigger_softirq;
6544 sd->backlog.poll = process_backlog;
6545 sd->backlog.weight = weight_p;
6546 sd->backlog.gro_list = NULL;
6547 sd->backlog.gro_count = 0;
6552 /* The loopback device is special if any other network devices
6553 * is present in a network namespace the loopback device must
6554 * be present. Since we now dynamically allocate and free the
6555 * loopback device ensure this invariant is maintained by
6556 * keeping the loopback device as the first device on the
6557 * list of network devices. Ensuring the loopback devices
6558 * is the first device that appears and the last network device
6561 if (register_pernet_device(&loopback_net_ops))
6564 if (register_pernet_device(&default_device_ops))
6567 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6568 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6570 hotcpu_notifier(dev_cpu_callback, 0);
6578 subsys_initcall(net_dev_init);
6580 static int __init initialize_hashrnd(void)
6582 get_random_bytes(&hashrnd, sizeof(hashrnd));
6586 late_initcall_sync(initialize_hashrnd);