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_err("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);
1175 add_device_randomness(dev->dev_addr, dev->addr_len);
1182 * dev_open - prepare an interface for use.
1183 * @dev: device to open
1185 * Takes a device from down to up state. The device's private open
1186 * function is invoked and then the multicast lists are loaded. Finally
1187 * the device is moved into the up state and a %NETDEV_UP message is
1188 * sent to the netdev notifier chain.
1190 * Calling this function on an active interface is a nop. On a failure
1191 * a negative errno code is returned.
1193 int dev_open(struct net_device *dev)
1197 if (dev->flags & IFF_UP)
1200 ret = __dev_open(dev);
1204 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1205 call_netdevice_notifiers(NETDEV_UP, dev);
1209 EXPORT_SYMBOL(dev_open);
1211 static int __dev_close_many(struct list_head *head)
1213 struct net_device *dev;
1218 list_for_each_entry(dev, head, unreg_list) {
1219 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1221 clear_bit(__LINK_STATE_START, &dev->state);
1223 /* Synchronize to scheduled poll. We cannot touch poll list, it
1224 * can be even on different cpu. So just clear netif_running().
1226 * dev->stop() will invoke napi_disable() on all of it's
1227 * napi_struct instances on this device.
1229 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1232 dev_deactivate_many(head);
1234 list_for_each_entry(dev, head, unreg_list) {
1235 const struct net_device_ops *ops = dev->netdev_ops;
1238 * Call the device specific close. This cannot fail.
1239 * Only if device is UP
1241 * We allow it to be called even after a DETACH hot-plug
1247 dev->flags &= ~IFF_UP;
1248 net_dmaengine_put();
1254 static int __dev_close(struct net_device *dev)
1259 list_add(&dev->unreg_list, &single);
1260 retval = __dev_close_many(&single);
1265 static int dev_close_many(struct list_head *head)
1267 struct net_device *dev, *tmp;
1268 LIST_HEAD(tmp_list);
1270 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1271 if (!(dev->flags & IFF_UP))
1272 list_move(&dev->unreg_list, &tmp_list);
1274 __dev_close_many(head);
1276 list_for_each_entry(dev, head, unreg_list) {
1277 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1278 call_netdevice_notifiers(NETDEV_DOWN, dev);
1281 /* rollback_registered_many needs the complete original list */
1282 list_splice(&tmp_list, head);
1287 * dev_close - shutdown an interface.
1288 * @dev: device to shutdown
1290 * This function moves an active device into down state. A
1291 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1292 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1295 int dev_close(struct net_device *dev)
1297 if (dev->flags & IFF_UP) {
1300 list_add(&dev->unreg_list, &single);
1301 dev_close_many(&single);
1306 EXPORT_SYMBOL(dev_close);
1310 * dev_disable_lro - disable Large Receive Offload on a device
1313 * Disable Large Receive Offload (LRO) on a net device. Must be
1314 * called under RTNL. This is needed if received packets may be
1315 * forwarded to another interface.
1317 void dev_disable_lro(struct net_device *dev)
1320 * If we're trying to disable lro on a vlan device
1321 * use the underlying physical device instead
1323 if (is_vlan_dev(dev))
1324 dev = vlan_dev_real_dev(dev);
1326 dev->wanted_features &= ~NETIF_F_LRO;
1327 netdev_update_features(dev);
1329 if (unlikely(dev->features & NETIF_F_LRO))
1330 netdev_WARN(dev, "failed to disable LRO!\n");
1332 EXPORT_SYMBOL(dev_disable_lro);
1335 static int dev_boot_phase = 1;
1338 * register_netdevice_notifier - register a network notifier block
1341 * Register a notifier to be called when network device events occur.
1342 * The notifier passed is linked into the kernel structures and must
1343 * not be reused until it has been unregistered. A negative errno code
1344 * is returned on a failure.
1346 * When registered all registration and up events are replayed
1347 * to the new notifier to allow device to have a race free
1348 * view of the network device list.
1351 int register_netdevice_notifier(struct notifier_block *nb)
1353 struct net_device *dev;
1354 struct net_device *last;
1359 err = raw_notifier_chain_register(&netdev_chain, nb);
1365 for_each_netdev(net, dev) {
1366 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1367 err = notifier_to_errno(err);
1371 if (!(dev->flags & IFF_UP))
1374 nb->notifier_call(nb, NETDEV_UP, dev);
1385 for_each_netdev(net, dev) {
1389 if (dev->flags & IFF_UP) {
1390 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1391 nb->notifier_call(nb, NETDEV_DOWN, dev);
1393 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1394 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1399 raw_notifier_chain_unregister(&netdev_chain, nb);
1402 EXPORT_SYMBOL(register_netdevice_notifier);
1405 * unregister_netdevice_notifier - unregister a network notifier block
1408 * Unregister a notifier previously registered by
1409 * register_netdevice_notifier(). The notifier is unlinked into the
1410 * kernel structures and may then be reused. A negative errno code
1411 * is returned on a failure.
1413 * After unregistering unregister and down device events are synthesized
1414 * for all devices on the device list to the removed notifier to remove
1415 * the need for special case cleanup code.
1418 int unregister_netdevice_notifier(struct notifier_block *nb)
1420 struct net_device *dev;
1425 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1430 for_each_netdev(net, dev) {
1431 if (dev->flags & IFF_UP) {
1432 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1433 nb->notifier_call(nb, NETDEV_DOWN, dev);
1435 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1436 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1443 EXPORT_SYMBOL(unregister_netdevice_notifier);
1446 * call_netdevice_notifiers - call all network notifier blocks
1447 * @val: value passed unmodified to notifier function
1448 * @dev: net_device pointer passed unmodified to notifier function
1450 * Call all network notifier blocks. Parameters and return value
1451 * are as for raw_notifier_call_chain().
1454 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1457 return raw_notifier_call_chain(&netdev_chain, val, dev);
1459 EXPORT_SYMBOL(call_netdevice_notifiers);
1461 static struct static_key netstamp_needed __read_mostly;
1462 #ifdef HAVE_JUMP_LABEL
1463 /* We are not allowed to call static_key_slow_dec() from irq context
1464 * If net_disable_timestamp() is called from irq context, defer the
1465 * static_key_slow_dec() calls.
1467 static atomic_t netstamp_needed_deferred;
1470 void net_enable_timestamp(void)
1472 #ifdef HAVE_JUMP_LABEL
1473 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1477 static_key_slow_dec(&netstamp_needed);
1481 WARN_ON(in_interrupt());
1482 static_key_slow_inc(&netstamp_needed);
1484 EXPORT_SYMBOL(net_enable_timestamp);
1486 void net_disable_timestamp(void)
1488 #ifdef HAVE_JUMP_LABEL
1489 if (in_interrupt()) {
1490 atomic_inc(&netstamp_needed_deferred);
1494 static_key_slow_dec(&netstamp_needed);
1496 EXPORT_SYMBOL(net_disable_timestamp);
1498 static inline void net_timestamp_set(struct sk_buff *skb)
1500 skb->tstamp.tv64 = 0;
1501 if (static_key_false(&netstamp_needed))
1502 __net_timestamp(skb);
1505 #define net_timestamp_check(COND, SKB) \
1506 if (static_key_false(&netstamp_needed)) { \
1507 if ((COND) && !(SKB)->tstamp.tv64) \
1508 __net_timestamp(SKB); \
1511 static int net_hwtstamp_validate(struct ifreq *ifr)
1513 struct hwtstamp_config cfg;
1514 enum hwtstamp_tx_types tx_type;
1515 enum hwtstamp_rx_filters rx_filter;
1516 int tx_type_valid = 0;
1517 int rx_filter_valid = 0;
1519 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1522 if (cfg.flags) /* reserved for future extensions */
1525 tx_type = cfg.tx_type;
1526 rx_filter = cfg.rx_filter;
1529 case HWTSTAMP_TX_OFF:
1530 case HWTSTAMP_TX_ON:
1531 case HWTSTAMP_TX_ONESTEP_SYNC:
1536 switch (rx_filter) {
1537 case HWTSTAMP_FILTER_NONE:
1538 case HWTSTAMP_FILTER_ALL:
1539 case HWTSTAMP_FILTER_SOME:
1540 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1542 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1543 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1545 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1546 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1548 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1549 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1550 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1551 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1552 rx_filter_valid = 1;
1556 if (!tx_type_valid || !rx_filter_valid)
1562 static inline bool is_skb_forwardable(struct net_device *dev,
1563 struct sk_buff *skb)
1567 if (!(dev->flags & IFF_UP))
1570 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1571 if (skb->len <= len)
1574 /* if TSO is enabled, we don't care about the length as the packet
1575 * could be forwarded without being segmented before
1577 if (skb_is_gso(skb))
1584 * dev_forward_skb - loopback an skb to another netif
1586 * @dev: destination network device
1587 * @skb: buffer to forward
1590 * NET_RX_SUCCESS (no congestion)
1591 * NET_RX_DROP (packet was dropped, but freed)
1593 * dev_forward_skb can be used for injecting an skb from the
1594 * start_xmit function of one device into the receive queue
1595 * of another device.
1597 * The receiving device may be in another namespace, so
1598 * we have to clear all information in the skb that could
1599 * impact namespace isolation.
1601 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1603 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1604 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1605 atomic_long_inc(&dev->rx_dropped);
1614 if (unlikely(!is_skb_forwardable(dev, skb))) {
1615 atomic_long_inc(&dev->rx_dropped);
1622 skb->tstamp.tv64 = 0;
1623 skb->pkt_type = PACKET_HOST;
1624 skb->protocol = eth_type_trans(skb, dev);
1628 return netif_rx(skb);
1630 EXPORT_SYMBOL_GPL(dev_forward_skb);
1632 static inline int deliver_skb(struct sk_buff *skb,
1633 struct packet_type *pt_prev,
1634 struct net_device *orig_dev)
1636 atomic_inc(&skb->users);
1637 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1641 * Support routine. Sends outgoing frames to any network
1642 * taps currently in use.
1645 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1647 struct packet_type *ptype;
1648 struct sk_buff *skb2 = NULL;
1649 struct packet_type *pt_prev = NULL;
1652 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1653 /* Never send packets back to the socket
1654 * they originated from - MvS (miquels@drinkel.ow.org)
1656 if ((ptype->dev == dev || !ptype->dev) &&
1657 (ptype->af_packet_priv == NULL ||
1658 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1660 deliver_skb(skb2, pt_prev, skb->dev);
1665 skb2 = skb_clone(skb, GFP_ATOMIC);
1669 net_timestamp_set(skb2);
1671 /* skb->nh should be correctly
1672 set by sender, so that the second statement is
1673 just protection against buggy protocols.
1675 skb_reset_mac_header(skb2);
1677 if (skb_network_header(skb2) < skb2->data ||
1678 skb2->network_header > skb2->tail) {
1679 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1680 ntohs(skb2->protocol),
1682 skb_reset_network_header(skb2);
1685 skb2->transport_header = skb2->network_header;
1686 skb2->pkt_type = PACKET_OUTGOING;
1691 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);
1797 static inline void __netif_reschedule(struct Qdisc *q)
1799 struct softnet_data *sd;
1800 unsigned long flags;
1802 local_irq_save(flags);
1803 sd = &__get_cpu_var(softnet_data);
1804 q->next_sched = NULL;
1805 *sd->output_queue_tailp = q;
1806 sd->output_queue_tailp = &q->next_sched;
1807 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1808 local_irq_restore(flags);
1811 void __netif_schedule(struct Qdisc *q)
1813 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1814 __netif_reschedule(q);
1816 EXPORT_SYMBOL(__netif_schedule);
1818 void dev_kfree_skb_irq(struct sk_buff *skb)
1820 if (atomic_dec_and_test(&skb->users)) {
1821 struct softnet_data *sd;
1822 unsigned long flags;
1824 local_irq_save(flags);
1825 sd = &__get_cpu_var(softnet_data);
1826 skb->next = sd->completion_queue;
1827 sd->completion_queue = skb;
1828 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1829 local_irq_restore(flags);
1832 EXPORT_SYMBOL(dev_kfree_skb_irq);
1834 void dev_kfree_skb_any(struct sk_buff *skb)
1836 if (in_irq() || irqs_disabled())
1837 dev_kfree_skb_irq(skb);
1841 EXPORT_SYMBOL(dev_kfree_skb_any);
1845 * netif_device_detach - mark device as removed
1846 * @dev: network device
1848 * Mark device as removed from system and therefore no longer available.
1850 void netif_device_detach(struct net_device *dev)
1852 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1853 netif_running(dev)) {
1854 netif_tx_stop_all_queues(dev);
1857 EXPORT_SYMBOL(netif_device_detach);
1860 * netif_device_attach - mark device as attached
1861 * @dev: network device
1863 * Mark device as attached from system and restart if needed.
1865 void netif_device_attach(struct net_device *dev)
1867 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1868 netif_running(dev)) {
1869 netif_tx_wake_all_queues(dev);
1870 __netdev_watchdog_up(dev);
1873 EXPORT_SYMBOL(netif_device_attach);
1875 static void skb_warn_bad_offload(const struct sk_buff *skb)
1877 static const netdev_features_t null_features = 0;
1878 struct net_device *dev = skb->dev;
1879 const char *driver = "";
1881 if (dev && dev->dev.parent)
1882 driver = dev_driver_string(dev->dev.parent);
1884 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1885 "gso_type=%d ip_summed=%d\n",
1886 driver, dev ? &dev->features : &null_features,
1887 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1888 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1889 skb_shinfo(skb)->gso_type, skb->ip_summed);
1893 * Invalidate hardware checksum when packet is to be mangled, and
1894 * complete checksum manually on outgoing path.
1896 int skb_checksum_help(struct sk_buff *skb)
1899 int ret = 0, offset;
1901 if (skb->ip_summed == CHECKSUM_COMPLETE)
1902 goto out_set_summed;
1904 if (unlikely(skb_shinfo(skb)->gso_size)) {
1905 skb_warn_bad_offload(skb);
1909 offset = skb_checksum_start_offset(skb);
1910 BUG_ON(offset >= skb_headlen(skb));
1911 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1913 offset += skb->csum_offset;
1914 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1916 if (skb_cloned(skb) &&
1917 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1918 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1923 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1925 skb->ip_summed = CHECKSUM_NONE;
1929 EXPORT_SYMBOL(skb_checksum_help);
1932 * skb_gso_segment - Perform segmentation on skb.
1933 * @skb: buffer to segment
1934 * @features: features for the output path (see dev->features)
1936 * This function segments the given skb and returns a list of segments.
1938 * It may return NULL if the skb requires no segmentation. This is
1939 * only possible when GSO is used for verifying header integrity.
1941 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1942 netdev_features_t features)
1944 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1945 struct packet_type *ptype;
1946 __be16 type = skb->protocol;
1947 int vlan_depth = ETH_HLEN;
1950 while (type == htons(ETH_P_8021Q)) {
1951 struct vlan_hdr *vh;
1953 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1954 return ERR_PTR(-EINVAL);
1956 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1957 type = vh->h_vlan_encapsulated_proto;
1958 vlan_depth += VLAN_HLEN;
1961 skb_reset_mac_header(skb);
1962 skb->mac_len = skb->network_header - skb->mac_header;
1963 __skb_pull(skb, skb->mac_len);
1965 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1966 skb_warn_bad_offload(skb);
1968 if (skb_header_cloned(skb) &&
1969 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1970 return ERR_PTR(err);
1974 list_for_each_entry_rcu(ptype,
1975 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1976 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1977 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1978 err = ptype->gso_send_check(skb);
1979 segs = ERR_PTR(err);
1980 if (err || skb_gso_ok(skb, features))
1982 __skb_push(skb, (skb->data -
1983 skb_network_header(skb)));
1985 segs = ptype->gso_segment(skb, features);
1991 __skb_push(skb, skb->data - skb_mac_header(skb));
1995 EXPORT_SYMBOL(skb_gso_segment);
1997 /* Take action when hardware reception checksum errors are detected. */
1999 void netdev_rx_csum_fault(struct net_device *dev)
2001 if (net_ratelimit()) {
2002 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2006 EXPORT_SYMBOL(netdev_rx_csum_fault);
2009 /* Actually, we should eliminate this check as soon as we know, that:
2010 * 1. IOMMU is present and allows to map all the memory.
2011 * 2. No high memory really exists on this machine.
2014 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2016 #ifdef CONFIG_HIGHMEM
2018 if (!(dev->features & NETIF_F_HIGHDMA)) {
2019 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2020 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2021 if (PageHighMem(skb_frag_page(frag)))
2026 if (PCI_DMA_BUS_IS_PHYS) {
2027 struct device *pdev = dev->dev.parent;
2031 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2032 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2033 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2034 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2043 void (*destructor)(struct sk_buff *skb);
2046 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2048 static void dev_gso_skb_destructor(struct sk_buff *skb)
2050 struct dev_gso_cb *cb;
2053 struct sk_buff *nskb = skb->next;
2055 skb->next = nskb->next;
2058 } while (skb->next);
2060 cb = DEV_GSO_CB(skb);
2062 cb->destructor(skb);
2066 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2067 * @skb: buffer to segment
2068 * @features: device features as applicable to this skb
2070 * This function segments the given skb and stores the list of segments
2073 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2075 struct sk_buff *segs;
2077 segs = skb_gso_segment(skb, features);
2079 /* Verifying header integrity only. */
2084 return PTR_ERR(segs);
2087 DEV_GSO_CB(skb)->destructor = skb->destructor;
2088 skb->destructor = dev_gso_skb_destructor;
2093 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2095 return ((features & NETIF_F_GEN_CSUM) ||
2096 ((features & NETIF_F_V4_CSUM) &&
2097 protocol == htons(ETH_P_IP)) ||
2098 ((features & NETIF_F_V6_CSUM) &&
2099 protocol == htons(ETH_P_IPV6)) ||
2100 ((features & NETIF_F_FCOE_CRC) &&
2101 protocol == htons(ETH_P_FCOE)));
2104 static netdev_features_t harmonize_features(struct sk_buff *skb,
2105 __be16 protocol, netdev_features_t features)
2107 if (!can_checksum_protocol(features, protocol)) {
2108 features &= ~NETIF_F_ALL_CSUM;
2109 features &= ~NETIF_F_SG;
2110 } else if (illegal_highdma(skb->dev, skb)) {
2111 features &= ~NETIF_F_SG;
2117 netdev_features_t netif_skb_features(struct sk_buff *skb)
2119 __be16 protocol = skb->protocol;
2120 netdev_features_t features = skb->dev->features;
2122 if (protocol == htons(ETH_P_8021Q)) {
2123 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2124 protocol = veh->h_vlan_encapsulated_proto;
2125 } else if (!vlan_tx_tag_present(skb)) {
2126 return harmonize_features(skb, protocol, features);
2129 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2131 if (protocol != htons(ETH_P_8021Q)) {
2132 return harmonize_features(skb, protocol, features);
2134 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2135 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2136 return harmonize_features(skb, protocol, features);
2139 EXPORT_SYMBOL(netif_skb_features);
2142 * Returns true if either:
2143 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2144 * 2. skb is fragmented and the device does not support SG, or if
2145 * at least one of fragments is in highmem and device does not
2146 * support DMA from it.
2148 static inline int skb_needs_linearize(struct sk_buff *skb,
2151 return skb_is_nonlinear(skb) &&
2152 ((skb_has_frag_list(skb) &&
2153 !(features & NETIF_F_FRAGLIST)) ||
2154 (skb_shinfo(skb)->nr_frags &&
2155 !(features & NETIF_F_SG)));
2158 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2159 struct netdev_queue *txq)
2161 const struct net_device_ops *ops = dev->netdev_ops;
2162 int rc = NETDEV_TX_OK;
2163 unsigned int skb_len;
2165 if (likely(!skb->next)) {
2166 netdev_features_t features;
2169 * If device doesn't need skb->dst, release it right now while
2170 * its hot in this cpu cache
2172 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2175 if (!list_empty(&ptype_all))
2176 dev_queue_xmit_nit(skb, dev);
2178 features = netif_skb_features(skb);
2180 if (vlan_tx_tag_present(skb) &&
2181 !(features & NETIF_F_HW_VLAN_TX)) {
2182 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2189 if (netif_needs_gso(skb, features)) {
2190 if (unlikely(dev_gso_segment(skb, features)))
2195 if (skb_needs_linearize(skb, features) &&
2196 __skb_linearize(skb))
2199 /* If packet is not checksummed and device does not
2200 * support checksumming for this protocol, complete
2201 * checksumming here.
2203 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2204 skb_set_transport_header(skb,
2205 skb_checksum_start_offset(skb));
2206 if (!(features & NETIF_F_ALL_CSUM) &&
2207 skb_checksum_help(skb))
2213 rc = ops->ndo_start_xmit(skb, dev);
2214 trace_net_dev_xmit(skb, rc, dev, skb_len);
2215 if (rc == NETDEV_TX_OK)
2216 txq_trans_update(txq);
2222 struct sk_buff *nskb = skb->next;
2224 skb->next = nskb->next;
2228 * If device doesn't need nskb->dst, release it right now while
2229 * its hot in this cpu cache
2231 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2234 skb_len = nskb->len;
2235 rc = ops->ndo_start_xmit(nskb, dev);
2236 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2237 if (unlikely(rc != NETDEV_TX_OK)) {
2238 if (rc & ~NETDEV_TX_MASK)
2239 goto out_kfree_gso_skb;
2240 nskb->next = skb->next;
2244 txq_trans_update(txq);
2245 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2246 return NETDEV_TX_BUSY;
2247 } while (skb->next);
2250 if (likely(skb->next == NULL))
2251 skb->destructor = DEV_GSO_CB(skb)->destructor;
2258 static u32 hashrnd __read_mostly;
2261 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2262 * to be used as a distribution range.
2264 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2265 unsigned int num_tx_queues)
2269 u16 qcount = num_tx_queues;
2271 if (skb_rx_queue_recorded(skb)) {
2272 hash = skb_get_rx_queue(skb);
2273 while (unlikely(hash >= num_tx_queues))
2274 hash -= num_tx_queues;
2279 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2280 qoffset = dev->tc_to_txq[tc].offset;
2281 qcount = dev->tc_to_txq[tc].count;
2284 if (skb->sk && skb->sk->sk_hash)
2285 hash = skb->sk->sk_hash;
2287 hash = (__force u16) skb->protocol;
2288 hash = jhash_1word(hash, hashrnd);
2290 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2292 EXPORT_SYMBOL(__skb_tx_hash);
2294 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2296 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2297 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2298 dev->name, queue_index,
2299 dev->real_num_tx_queues);
2305 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2308 struct xps_dev_maps *dev_maps;
2309 struct xps_map *map;
2310 int queue_index = -1;
2313 dev_maps = rcu_dereference(dev->xps_maps);
2315 map = rcu_dereference(
2316 dev_maps->cpu_map[raw_smp_processor_id()]);
2319 queue_index = map->queues[0];
2322 if (skb->sk && skb->sk->sk_hash)
2323 hash = skb->sk->sk_hash;
2325 hash = (__force u16) skb->protocol ^
2327 hash = jhash_1word(hash, hashrnd);
2328 queue_index = map->queues[
2329 ((u64)hash * map->len) >> 32];
2331 if (unlikely(queue_index >= dev->real_num_tx_queues))
2343 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2344 struct sk_buff *skb)
2347 const struct net_device_ops *ops = dev->netdev_ops;
2349 if (dev->real_num_tx_queues == 1)
2351 else if (ops->ndo_select_queue) {
2352 queue_index = ops->ndo_select_queue(dev, skb);
2353 queue_index = dev_cap_txqueue(dev, queue_index);
2355 struct sock *sk = skb->sk;
2356 queue_index = sk_tx_queue_get(sk);
2358 if (queue_index < 0 || skb->ooo_okay ||
2359 queue_index >= dev->real_num_tx_queues) {
2360 int old_index = queue_index;
2362 queue_index = get_xps_queue(dev, skb);
2363 if (queue_index < 0)
2364 queue_index = skb_tx_hash(dev, skb);
2366 if (queue_index != old_index && sk) {
2367 struct dst_entry *dst =
2368 rcu_dereference_check(sk->sk_dst_cache, 1);
2370 if (dst && skb_dst(skb) == dst)
2371 sk_tx_queue_set(sk, queue_index);
2376 skb_set_queue_mapping(skb, queue_index);
2377 return netdev_get_tx_queue(dev, queue_index);
2380 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2381 struct net_device *dev,
2382 struct netdev_queue *txq)
2384 spinlock_t *root_lock = qdisc_lock(q);
2388 qdisc_skb_cb(skb)->pkt_len = skb->len;
2389 qdisc_calculate_pkt_len(skb, q);
2391 * Heuristic to force contended enqueues to serialize on a
2392 * separate lock before trying to get qdisc main lock.
2393 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2394 * and dequeue packets faster.
2396 contended = qdisc_is_running(q);
2397 if (unlikely(contended))
2398 spin_lock(&q->busylock);
2400 spin_lock(root_lock);
2401 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2404 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2405 qdisc_run_begin(q)) {
2407 * This is a work-conserving queue; there are no old skbs
2408 * waiting to be sent out; and the qdisc is not running -
2409 * xmit the skb directly.
2411 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2414 qdisc_bstats_update(q, skb);
2416 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2417 if (unlikely(contended)) {
2418 spin_unlock(&q->busylock);
2425 rc = NET_XMIT_SUCCESS;
2428 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2429 if (qdisc_run_begin(q)) {
2430 if (unlikely(contended)) {
2431 spin_unlock(&q->busylock);
2437 spin_unlock(root_lock);
2438 if (unlikely(contended))
2439 spin_unlock(&q->busylock);
2443 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2444 static void skb_update_prio(struct sk_buff *skb)
2446 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2448 if ((!skb->priority) && (skb->sk) && map)
2449 skb->priority = map->priomap[skb->sk->sk_cgrp_prioidx];
2452 #define skb_update_prio(skb)
2455 static DEFINE_PER_CPU(int, xmit_recursion);
2456 #define RECURSION_LIMIT 10
2459 * dev_queue_xmit - transmit a buffer
2460 * @skb: buffer to transmit
2462 * Queue a buffer for transmission to a network device. The caller must
2463 * have set the device and priority and built the buffer before calling
2464 * this function. The function can be called from an interrupt.
2466 * A negative errno code is returned on a failure. A success does not
2467 * guarantee the frame will be transmitted as it may be dropped due
2468 * to congestion or traffic shaping.
2470 * -----------------------------------------------------------------------------------
2471 * I notice this method can also return errors from the queue disciplines,
2472 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2475 * Regardless of the return value, the skb is consumed, so it is currently
2476 * difficult to retry a send to this method. (You can bump the ref count
2477 * before sending to hold a reference for retry if you are careful.)
2479 * When calling this method, interrupts MUST be enabled. This is because
2480 * the BH enable code must have IRQs enabled so that it will not deadlock.
2483 int dev_queue_xmit(struct sk_buff *skb)
2485 struct net_device *dev = skb->dev;
2486 struct netdev_queue *txq;
2490 /* Disable soft irqs for various locks below. Also
2491 * stops preemption for RCU.
2495 skb_update_prio(skb);
2497 txq = dev_pick_tx(dev, skb);
2498 q = rcu_dereference_bh(txq->qdisc);
2500 #ifdef CONFIG_NET_CLS_ACT
2501 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2503 trace_net_dev_queue(skb);
2505 rc = __dev_xmit_skb(skb, q, dev, txq);
2509 /* The device has no queue. Common case for software devices:
2510 loopback, all the sorts of tunnels...
2512 Really, it is unlikely that netif_tx_lock protection is necessary
2513 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2515 However, it is possible, that they rely on protection
2518 Check this and shot the lock. It is not prone from deadlocks.
2519 Either shot noqueue qdisc, it is even simpler 8)
2521 if (dev->flags & IFF_UP) {
2522 int cpu = smp_processor_id(); /* ok because BHs are off */
2524 if (txq->xmit_lock_owner != cpu) {
2526 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2527 goto recursion_alert;
2529 HARD_TX_LOCK(dev, txq, cpu);
2531 if (!netif_xmit_stopped(txq)) {
2532 __this_cpu_inc(xmit_recursion);
2533 rc = dev_hard_start_xmit(skb, dev, txq);
2534 __this_cpu_dec(xmit_recursion);
2535 if (dev_xmit_complete(rc)) {
2536 HARD_TX_UNLOCK(dev, txq);
2540 HARD_TX_UNLOCK(dev, txq);
2541 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2544 /* Recursion is detected! It is possible,
2548 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2554 rcu_read_unlock_bh();
2559 rcu_read_unlock_bh();
2562 EXPORT_SYMBOL(dev_queue_xmit);
2565 /*=======================================================================
2567 =======================================================================*/
2569 int netdev_max_backlog __read_mostly = 1000;
2570 int netdev_tstamp_prequeue __read_mostly = 1;
2571 int netdev_budget __read_mostly = 300;
2572 int weight_p __read_mostly = 64; /* old backlog weight */
2574 /* Called with irq disabled */
2575 static inline void ____napi_schedule(struct softnet_data *sd,
2576 struct napi_struct *napi)
2578 list_add_tail(&napi->poll_list, &sd->poll_list);
2579 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2583 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2584 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2585 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2586 * if hash is a canonical 4-tuple hash over transport ports.
2588 void __skb_get_rxhash(struct sk_buff *skb)
2590 struct flow_keys keys;
2593 if (!skb_flow_dissect(skb, &keys))
2597 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2598 swap(keys.port16[0], keys.port16[1]);
2602 /* get a consistent hash (same value on both flow directions) */
2603 if ((__force u32)keys.dst < (__force u32)keys.src)
2604 swap(keys.dst, keys.src);
2606 hash = jhash_3words((__force u32)keys.dst,
2607 (__force u32)keys.src,
2608 (__force u32)keys.ports, hashrnd);
2614 EXPORT_SYMBOL(__skb_get_rxhash);
2618 /* One global table that all flow-based protocols share. */
2619 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2620 EXPORT_SYMBOL(rps_sock_flow_table);
2622 struct static_key rps_needed __read_mostly;
2624 static struct rps_dev_flow *
2625 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2626 struct rps_dev_flow *rflow, u16 next_cpu)
2628 if (next_cpu != RPS_NO_CPU) {
2629 #ifdef CONFIG_RFS_ACCEL
2630 struct netdev_rx_queue *rxqueue;
2631 struct rps_dev_flow_table *flow_table;
2632 struct rps_dev_flow *old_rflow;
2637 /* Should we steer this flow to a different hardware queue? */
2638 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2639 !(dev->features & NETIF_F_NTUPLE))
2641 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2642 if (rxq_index == skb_get_rx_queue(skb))
2645 rxqueue = dev->_rx + rxq_index;
2646 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2649 flow_id = skb->rxhash & flow_table->mask;
2650 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2651 rxq_index, flow_id);
2655 rflow = &flow_table->flows[flow_id];
2657 if (old_rflow->filter == rflow->filter)
2658 old_rflow->filter = RPS_NO_FILTER;
2662 per_cpu(softnet_data, next_cpu).input_queue_head;
2665 rflow->cpu = next_cpu;
2670 * get_rps_cpu is called from netif_receive_skb and returns the target
2671 * CPU from the RPS map of the receiving queue for a given skb.
2672 * rcu_read_lock must be held on entry.
2674 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2675 struct rps_dev_flow **rflowp)
2677 struct netdev_rx_queue *rxqueue;
2678 struct rps_map *map;
2679 struct rps_dev_flow_table *flow_table;
2680 struct rps_sock_flow_table *sock_flow_table;
2684 if (skb_rx_queue_recorded(skb)) {
2685 u16 index = skb_get_rx_queue(skb);
2686 if (unlikely(index >= dev->real_num_rx_queues)) {
2687 WARN_ONCE(dev->real_num_rx_queues > 1,
2688 "%s received packet on queue %u, but number "
2689 "of RX queues is %u\n",
2690 dev->name, index, dev->real_num_rx_queues);
2693 rxqueue = dev->_rx + index;
2697 map = rcu_dereference(rxqueue->rps_map);
2699 if (map->len == 1 &&
2700 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2701 tcpu = map->cpus[0];
2702 if (cpu_online(tcpu))
2706 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2710 skb_reset_network_header(skb);
2711 if (!skb_get_rxhash(skb))
2714 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2715 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2716 if (flow_table && sock_flow_table) {
2718 struct rps_dev_flow *rflow;
2720 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2723 next_cpu = sock_flow_table->ents[skb->rxhash &
2724 sock_flow_table->mask];
2727 * If the desired CPU (where last recvmsg was done) is
2728 * different from current CPU (one in the rx-queue flow
2729 * table entry), switch if one of the following holds:
2730 * - Current CPU is unset (equal to RPS_NO_CPU).
2731 * - Current CPU is offline.
2732 * - The current CPU's queue tail has advanced beyond the
2733 * last packet that was enqueued using this table entry.
2734 * This guarantees that all previous packets for the flow
2735 * have been dequeued, thus preserving in order delivery.
2737 if (unlikely(tcpu != next_cpu) &&
2738 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2739 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2740 rflow->last_qtail)) >= 0))
2741 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2743 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2751 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2753 if (cpu_online(tcpu)) {
2763 #ifdef CONFIG_RFS_ACCEL
2766 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2767 * @dev: Device on which the filter was set
2768 * @rxq_index: RX queue index
2769 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2770 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2772 * Drivers that implement ndo_rx_flow_steer() should periodically call
2773 * this function for each installed filter and remove the filters for
2774 * which it returns %true.
2776 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2777 u32 flow_id, u16 filter_id)
2779 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2780 struct rps_dev_flow_table *flow_table;
2781 struct rps_dev_flow *rflow;
2786 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2787 if (flow_table && flow_id <= flow_table->mask) {
2788 rflow = &flow_table->flows[flow_id];
2789 cpu = ACCESS_ONCE(rflow->cpu);
2790 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2791 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2792 rflow->last_qtail) <
2793 (int)(10 * flow_table->mask)))
2799 EXPORT_SYMBOL(rps_may_expire_flow);
2801 #endif /* CONFIG_RFS_ACCEL */
2803 /* Called from hardirq (IPI) context */
2804 static void rps_trigger_softirq(void *data)
2806 struct softnet_data *sd = data;
2808 ____napi_schedule(sd, &sd->backlog);
2812 #endif /* CONFIG_RPS */
2815 * Check if this softnet_data structure is another cpu one
2816 * If yes, queue it to our IPI list and return 1
2819 static int rps_ipi_queued(struct softnet_data *sd)
2822 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2825 sd->rps_ipi_next = mysd->rps_ipi_list;
2826 mysd->rps_ipi_list = sd;
2828 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2831 #endif /* CONFIG_RPS */
2836 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2837 * queue (may be a remote CPU queue).
2839 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2840 unsigned int *qtail)
2842 struct softnet_data *sd;
2843 unsigned long flags;
2845 sd = &per_cpu(softnet_data, cpu);
2847 local_irq_save(flags);
2850 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2851 if (skb_queue_len(&sd->input_pkt_queue)) {
2853 __skb_queue_tail(&sd->input_pkt_queue, skb);
2854 input_queue_tail_incr_save(sd, qtail);
2856 local_irq_restore(flags);
2857 return NET_RX_SUCCESS;
2860 /* Schedule NAPI for backlog device
2861 * We can use non atomic operation since we own the queue lock
2863 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2864 if (!rps_ipi_queued(sd))
2865 ____napi_schedule(sd, &sd->backlog);
2873 local_irq_restore(flags);
2875 atomic_long_inc(&skb->dev->rx_dropped);
2881 * netif_rx - post buffer to the network code
2882 * @skb: buffer to post
2884 * This function receives a packet from a device driver and queues it for
2885 * the upper (protocol) levels to process. It always succeeds. The buffer
2886 * may be dropped during processing for congestion control or by the
2890 * NET_RX_SUCCESS (no congestion)
2891 * NET_RX_DROP (packet was dropped)
2895 int netif_rx(struct sk_buff *skb)
2899 /* if netpoll wants it, pretend we never saw it */
2900 if (netpoll_rx(skb))
2903 net_timestamp_check(netdev_tstamp_prequeue, skb);
2905 trace_netif_rx(skb);
2907 if (static_key_false(&rps_needed)) {
2908 struct rps_dev_flow voidflow, *rflow = &voidflow;
2914 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2916 cpu = smp_processor_id();
2918 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2926 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2931 EXPORT_SYMBOL(netif_rx);
2933 int netif_rx_ni(struct sk_buff *skb)
2938 err = netif_rx(skb);
2939 if (local_softirq_pending())
2945 EXPORT_SYMBOL(netif_rx_ni);
2947 static void net_tx_action(struct softirq_action *h)
2949 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2951 if (sd->completion_queue) {
2952 struct sk_buff *clist;
2954 local_irq_disable();
2955 clist = sd->completion_queue;
2956 sd->completion_queue = NULL;
2960 struct sk_buff *skb = clist;
2961 clist = clist->next;
2963 WARN_ON(atomic_read(&skb->users));
2964 trace_kfree_skb(skb, net_tx_action);
2969 if (sd->output_queue) {
2972 local_irq_disable();
2973 head = sd->output_queue;
2974 sd->output_queue = NULL;
2975 sd->output_queue_tailp = &sd->output_queue;
2979 struct Qdisc *q = head;
2980 spinlock_t *root_lock;
2982 head = head->next_sched;
2984 root_lock = qdisc_lock(q);
2985 if (spin_trylock(root_lock)) {
2986 smp_mb__before_clear_bit();
2987 clear_bit(__QDISC_STATE_SCHED,
2990 spin_unlock(root_lock);
2992 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2994 __netif_reschedule(q);
2996 smp_mb__before_clear_bit();
2997 clear_bit(__QDISC_STATE_SCHED,
3005 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3006 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3007 /* This hook is defined here for ATM LANE */
3008 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3009 unsigned char *addr) __read_mostly;
3010 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3013 #ifdef CONFIG_NET_CLS_ACT
3014 /* TODO: Maybe we should just force sch_ingress to be compiled in
3015 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3016 * a compare and 2 stores extra right now if we dont have it on
3017 * but have CONFIG_NET_CLS_ACT
3018 * NOTE: This doesn't stop any functionality; if you dont have
3019 * the ingress scheduler, you just can't add policies on ingress.
3022 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3024 struct net_device *dev = skb->dev;
3025 u32 ttl = G_TC_RTTL(skb->tc_verd);
3026 int result = TC_ACT_OK;
3029 if (unlikely(MAX_RED_LOOP < ttl++)) {
3030 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3031 skb->skb_iif, dev->ifindex);
3035 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3036 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3039 if (q != &noop_qdisc) {
3040 spin_lock(qdisc_lock(q));
3041 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3042 result = qdisc_enqueue_root(skb, q);
3043 spin_unlock(qdisc_lock(q));
3049 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3050 struct packet_type **pt_prev,
3051 int *ret, struct net_device *orig_dev)
3053 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3055 if (!rxq || rxq->qdisc == &noop_qdisc)
3059 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3063 switch (ing_filter(skb, rxq)) {
3077 * netdev_rx_handler_register - register receive handler
3078 * @dev: device to register a handler for
3079 * @rx_handler: receive handler to register
3080 * @rx_handler_data: data pointer that is used by rx handler
3082 * Register a receive hander for a device. This handler will then be
3083 * called from __netif_receive_skb. A negative errno code is returned
3086 * The caller must hold the rtnl_mutex.
3088 * For a general description of rx_handler, see enum rx_handler_result.
3090 int netdev_rx_handler_register(struct net_device *dev,
3091 rx_handler_func_t *rx_handler,
3092 void *rx_handler_data)
3096 if (dev->rx_handler)
3099 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3100 rcu_assign_pointer(dev->rx_handler, rx_handler);
3104 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3107 * netdev_rx_handler_unregister - unregister receive handler
3108 * @dev: device to unregister a handler from
3110 * Unregister a receive hander from a device.
3112 * The caller must hold the rtnl_mutex.
3114 void netdev_rx_handler_unregister(struct net_device *dev)
3118 RCU_INIT_POINTER(dev->rx_handler, NULL);
3119 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3121 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3123 static int __netif_receive_skb(struct sk_buff *skb)
3125 struct packet_type *ptype, *pt_prev;
3126 rx_handler_func_t *rx_handler;
3127 struct net_device *orig_dev;
3128 struct net_device *null_or_dev;
3129 bool deliver_exact = false;
3130 int ret = NET_RX_DROP;
3133 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3135 trace_netif_receive_skb(skb);
3137 /* if we've gotten here through NAPI, check netpoll */
3138 if (netpoll_receive_skb(skb))
3142 skb->skb_iif = skb->dev->ifindex;
3143 orig_dev = skb->dev;
3145 skb_reset_network_header(skb);
3146 skb_reset_transport_header(skb);
3147 skb_reset_mac_len(skb);
3155 __this_cpu_inc(softnet_data.processed);
3157 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3158 skb = vlan_untag(skb);
3163 #ifdef CONFIG_NET_CLS_ACT
3164 if (skb->tc_verd & TC_NCLS) {
3165 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3170 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3171 if (!ptype->dev || ptype->dev == skb->dev) {
3173 ret = deliver_skb(skb, pt_prev, orig_dev);
3178 #ifdef CONFIG_NET_CLS_ACT
3179 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3185 rx_handler = rcu_dereference(skb->dev->rx_handler);
3186 if (vlan_tx_tag_present(skb)) {
3188 ret = deliver_skb(skb, pt_prev, orig_dev);
3191 if (vlan_do_receive(&skb, !rx_handler))
3193 else if (unlikely(!skb))
3199 ret = deliver_skb(skb, pt_prev, orig_dev);
3202 switch (rx_handler(&skb)) {
3203 case RX_HANDLER_CONSUMED:
3205 case RX_HANDLER_ANOTHER:
3207 case RX_HANDLER_EXACT:
3208 deliver_exact = true;
3209 case RX_HANDLER_PASS:
3216 /* deliver only exact match when indicated */
3217 null_or_dev = deliver_exact ? skb->dev : NULL;
3219 type = skb->protocol;
3220 list_for_each_entry_rcu(ptype,
3221 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3222 if (ptype->type == type &&
3223 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3224 ptype->dev == orig_dev)) {
3226 ret = deliver_skb(skb, pt_prev, orig_dev);
3232 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3234 atomic_long_inc(&skb->dev->rx_dropped);
3236 /* Jamal, now you will not able to escape explaining
3237 * me how you were going to use this. :-)
3248 * netif_receive_skb - process receive buffer from network
3249 * @skb: buffer to process
3251 * netif_receive_skb() is the main receive data processing function.
3252 * It always succeeds. The buffer may be dropped during processing
3253 * for congestion control or by the protocol layers.
3255 * This function may only be called from softirq context and interrupts
3256 * should be enabled.
3258 * Return values (usually ignored):
3259 * NET_RX_SUCCESS: no congestion
3260 * NET_RX_DROP: packet was dropped
3262 int netif_receive_skb(struct sk_buff *skb)
3264 net_timestamp_check(netdev_tstamp_prequeue, skb);
3266 if (skb_defer_rx_timestamp(skb))
3267 return NET_RX_SUCCESS;
3270 if (static_key_false(&rps_needed)) {
3271 struct rps_dev_flow voidflow, *rflow = &voidflow;
3276 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3279 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3286 return __netif_receive_skb(skb);
3288 EXPORT_SYMBOL(netif_receive_skb);
3290 /* Network device is going away, flush any packets still pending
3291 * Called with irqs disabled.
3293 static void flush_backlog(void *arg)
3295 struct net_device *dev = arg;
3296 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3297 struct sk_buff *skb, *tmp;
3300 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3301 if (skb->dev == dev) {
3302 __skb_unlink(skb, &sd->input_pkt_queue);
3304 input_queue_head_incr(sd);
3309 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3310 if (skb->dev == dev) {
3311 __skb_unlink(skb, &sd->process_queue);
3313 input_queue_head_incr(sd);
3318 static int napi_gro_complete(struct sk_buff *skb)
3320 struct packet_type *ptype;
3321 __be16 type = skb->protocol;
3322 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3325 if (NAPI_GRO_CB(skb)->count == 1) {
3326 skb_shinfo(skb)->gso_size = 0;
3331 list_for_each_entry_rcu(ptype, head, list) {
3332 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3335 err = ptype->gro_complete(skb);
3341 WARN_ON(&ptype->list == head);
3343 return NET_RX_SUCCESS;
3347 return netif_receive_skb(skb);
3350 inline void napi_gro_flush(struct napi_struct *napi)
3352 struct sk_buff *skb, *next;
3354 for (skb = napi->gro_list; skb; skb = next) {
3357 napi_gro_complete(skb);
3360 napi->gro_count = 0;
3361 napi->gro_list = NULL;
3363 EXPORT_SYMBOL(napi_gro_flush);
3365 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3367 struct sk_buff **pp = NULL;
3368 struct packet_type *ptype;
3369 __be16 type = skb->protocol;
3370 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3373 enum gro_result ret;
3375 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3378 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3382 list_for_each_entry_rcu(ptype, head, list) {
3383 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3386 skb_set_network_header(skb, skb_gro_offset(skb));
3387 mac_len = skb->network_header - skb->mac_header;
3388 skb->mac_len = mac_len;
3389 NAPI_GRO_CB(skb)->same_flow = 0;
3390 NAPI_GRO_CB(skb)->flush = 0;
3391 NAPI_GRO_CB(skb)->free = 0;
3393 pp = ptype->gro_receive(&napi->gro_list, skb);
3398 if (&ptype->list == head)
3401 same_flow = NAPI_GRO_CB(skb)->same_flow;
3402 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3405 struct sk_buff *nskb = *pp;
3409 napi_gro_complete(nskb);
3416 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3420 NAPI_GRO_CB(skb)->count = 1;
3421 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3422 skb->next = napi->gro_list;
3423 napi->gro_list = skb;
3427 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3428 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3430 BUG_ON(skb->end - skb->tail < grow);
3432 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3435 skb->data_len -= grow;
3437 skb_shinfo(skb)->frags[0].page_offset += grow;
3438 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3440 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3441 skb_frag_unref(skb, 0);
3442 memmove(skb_shinfo(skb)->frags,
3443 skb_shinfo(skb)->frags + 1,
3444 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3455 EXPORT_SYMBOL(dev_gro_receive);
3457 static inline gro_result_t
3458 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3461 unsigned int maclen = skb->dev->hard_header_len;
3463 for (p = napi->gro_list; p; p = p->next) {
3464 unsigned long diffs;
3466 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3467 diffs |= p->vlan_tci ^ skb->vlan_tci;
3468 if (maclen == ETH_HLEN)
3469 diffs |= compare_ether_header(skb_mac_header(p),
3470 skb_gro_mac_header(skb));
3472 diffs = memcmp(skb_mac_header(p),
3473 skb_gro_mac_header(skb),
3475 NAPI_GRO_CB(p)->same_flow = !diffs;
3476 NAPI_GRO_CB(p)->flush = 0;
3479 return dev_gro_receive(napi, skb);
3482 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3486 if (netif_receive_skb(skb))
3494 case GRO_MERGED_FREE:
3495 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3496 kmem_cache_free(skbuff_head_cache, skb);
3508 EXPORT_SYMBOL(napi_skb_finish);
3510 void skb_gro_reset_offset(struct sk_buff *skb)
3512 NAPI_GRO_CB(skb)->data_offset = 0;
3513 NAPI_GRO_CB(skb)->frag0 = NULL;
3514 NAPI_GRO_CB(skb)->frag0_len = 0;
3516 if (skb->mac_header == skb->tail &&
3517 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3518 NAPI_GRO_CB(skb)->frag0 =
3519 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3520 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3523 EXPORT_SYMBOL(skb_gro_reset_offset);
3525 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3527 skb_gro_reset_offset(skb);
3529 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3531 EXPORT_SYMBOL(napi_gro_receive);
3533 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3535 __skb_pull(skb, skb_headlen(skb));
3536 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3537 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3539 skb->dev = napi->dev;
3545 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3547 struct sk_buff *skb = napi->skb;
3550 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3556 EXPORT_SYMBOL(napi_get_frags);
3558 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3564 skb->protocol = eth_type_trans(skb, skb->dev);
3566 if (ret == GRO_HELD)
3567 skb_gro_pull(skb, -ETH_HLEN);
3568 else if (netif_receive_skb(skb))
3573 case GRO_MERGED_FREE:
3574 napi_reuse_skb(napi, skb);
3583 EXPORT_SYMBOL(napi_frags_finish);
3585 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3587 struct sk_buff *skb = napi->skb;
3594 skb_reset_mac_header(skb);
3595 skb_gro_reset_offset(skb);
3597 off = skb_gro_offset(skb);
3598 hlen = off + sizeof(*eth);
3599 eth = skb_gro_header_fast(skb, off);
3600 if (skb_gro_header_hard(skb, hlen)) {
3601 eth = skb_gro_header_slow(skb, hlen, off);
3602 if (unlikely(!eth)) {
3603 napi_reuse_skb(napi, skb);
3609 skb_gro_pull(skb, sizeof(*eth));
3612 * This works because the only protocols we care about don't require
3613 * special handling. We'll fix it up properly at the end.
3615 skb->protocol = eth->h_proto;
3621 gro_result_t napi_gro_frags(struct napi_struct *napi)
3623 struct sk_buff *skb = napi_frags_skb(napi);
3628 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3630 EXPORT_SYMBOL(napi_gro_frags);
3633 * net_rps_action sends any pending IPI's for rps.
3634 * Note: called with local irq disabled, but exits with local irq enabled.
3636 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3639 struct softnet_data *remsd = sd->rps_ipi_list;
3642 sd->rps_ipi_list = NULL;
3646 /* Send pending IPI's to kick RPS processing on remote cpus. */
3648 struct softnet_data *next = remsd->rps_ipi_next;
3650 if (cpu_online(remsd->cpu))
3651 __smp_call_function_single(remsd->cpu,
3660 static int process_backlog(struct napi_struct *napi, int quota)
3663 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3666 /* Check if we have pending ipi, its better to send them now,
3667 * not waiting net_rx_action() end.
3669 if (sd->rps_ipi_list) {
3670 local_irq_disable();
3671 net_rps_action_and_irq_enable(sd);
3674 napi->weight = weight_p;
3675 local_irq_disable();
3676 while (work < quota) {
3677 struct sk_buff *skb;
3680 while ((skb = __skb_dequeue(&sd->process_queue))) {
3682 __netif_receive_skb(skb);
3683 local_irq_disable();
3684 input_queue_head_incr(sd);
3685 if (++work >= quota) {
3692 qlen = skb_queue_len(&sd->input_pkt_queue);
3694 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3695 &sd->process_queue);
3697 if (qlen < quota - work) {
3699 * Inline a custom version of __napi_complete().
3700 * only current cpu owns and manipulates this napi,
3701 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3702 * we can use a plain write instead of clear_bit(),
3703 * and we dont need an smp_mb() memory barrier.
3705 list_del(&napi->poll_list);
3708 quota = work + qlen;
3718 * __napi_schedule - schedule for receive
3719 * @n: entry to schedule
3721 * The entry's receive function will be scheduled to run
3723 void __napi_schedule(struct napi_struct *n)
3725 unsigned long flags;
3727 local_irq_save(flags);
3728 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3729 local_irq_restore(flags);
3731 EXPORT_SYMBOL(__napi_schedule);
3733 void __napi_complete(struct napi_struct *n)
3735 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3736 BUG_ON(n->gro_list);
3738 list_del(&n->poll_list);
3739 smp_mb__before_clear_bit();
3740 clear_bit(NAPI_STATE_SCHED, &n->state);
3742 EXPORT_SYMBOL(__napi_complete);
3744 void napi_complete(struct napi_struct *n)
3746 unsigned long flags;
3749 * don't let napi dequeue from the cpu poll list
3750 * just in case its running on a different cpu
3752 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3756 local_irq_save(flags);
3758 local_irq_restore(flags);
3760 EXPORT_SYMBOL(napi_complete);
3762 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3763 int (*poll)(struct napi_struct *, int), int weight)
3765 INIT_LIST_HEAD(&napi->poll_list);
3766 napi->gro_count = 0;
3767 napi->gro_list = NULL;
3770 napi->weight = weight;
3771 list_add(&napi->dev_list, &dev->napi_list);
3773 #ifdef CONFIG_NETPOLL
3774 spin_lock_init(&napi->poll_lock);
3775 napi->poll_owner = -1;
3777 set_bit(NAPI_STATE_SCHED, &napi->state);
3779 EXPORT_SYMBOL(netif_napi_add);
3781 void netif_napi_del(struct napi_struct *napi)
3783 struct sk_buff *skb, *next;
3785 list_del_init(&napi->dev_list);
3786 napi_free_frags(napi);
3788 for (skb = napi->gro_list; skb; skb = next) {
3794 napi->gro_list = NULL;
3795 napi->gro_count = 0;
3797 EXPORT_SYMBOL(netif_napi_del);
3799 static void net_rx_action(struct softirq_action *h)
3801 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3802 unsigned long time_limit = jiffies + 2;
3803 int budget = netdev_budget;
3806 local_irq_disable();
3808 while (!list_empty(&sd->poll_list)) {
3809 struct napi_struct *n;
3812 /* If softirq window is exhuasted then punt.
3813 * Allow this to run for 2 jiffies since which will allow
3814 * an average latency of 1.5/HZ.
3816 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3821 /* Even though interrupts have been re-enabled, this
3822 * access is safe because interrupts can only add new
3823 * entries to the tail of this list, and only ->poll()
3824 * calls can remove this head entry from the list.
3826 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3828 have = netpoll_poll_lock(n);
3832 /* This NAPI_STATE_SCHED test is for avoiding a race
3833 * with netpoll's poll_napi(). Only the entity which
3834 * obtains the lock and sees NAPI_STATE_SCHED set will
3835 * actually make the ->poll() call. Therefore we avoid
3836 * accidentally calling ->poll() when NAPI is not scheduled.
3839 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3840 work = n->poll(n, weight);
3844 WARN_ON_ONCE(work > weight);
3848 local_irq_disable();
3850 /* Drivers must not modify the NAPI state if they
3851 * consume the entire weight. In such cases this code
3852 * still "owns" the NAPI instance and therefore can
3853 * move the instance around on the list at-will.
3855 if (unlikely(work == weight)) {
3856 if (unlikely(napi_disable_pending(n))) {
3859 local_irq_disable();
3861 list_move_tail(&n->poll_list, &sd->poll_list);
3864 netpoll_poll_unlock(have);
3867 net_rps_action_and_irq_enable(sd);
3869 #ifdef CONFIG_NET_DMA
3871 * There may not be any more sk_buffs coming right now, so push
3872 * any pending DMA copies to hardware
3874 dma_issue_pending_all();
3881 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3885 static gifconf_func_t *gifconf_list[NPROTO];
3888 * register_gifconf - register a SIOCGIF handler
3889 * @family: Address family
3890 * @gifconf: Function handler
3892 * Register protocol dependent address dumping routines. The handler
3893 * that is passed must not be freed or reused until it has been replaced
3894 * by another handler.
3896 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3898 if (family >= NPROTO)
3900 gifconf_list[family] = gifconf;
3903 EXPORT_SYMBOL(register_gifconf);
3907 * Map an interface index to its name (SIOCGIFNAME)
3911 * We need this ioctl for efficient implementation of the
3912 * if_indextoname() function required by the IPv6 API. Without
3913 * it, we would have to search all the interfaces to find a
3917 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3919 struct net_device *dev;
3923 * Fetch the caller's info block.
3926 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3930 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3936 strcpy(ifr.ifr_name, dev->name);
3939 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3945 * Perform a SIOCGIFCONF call. This structure will change
3946 * size eventually, and there is nothing I can do about it.
3947 * Thus we will need a 'compatibility mode'.
3950 static int dev_ifconf(struct net *net, char __user *arg)
3953 struct net_device *dev;
3960 * Fetch the caller's info block.
3963 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3970 * Loop over the interfaces, and write an info block for each.
3974 for_each_netdev(net, dev) {
3975 for (i = 0; i < NPROTO; i++) {
3976 if (gifconf_list[i]) {
3979 done = gifconf_list[i](dev, NULL, 0);
3981 done = gifconf_list[i](dev, pos + total,
3991 * All done. Write the updated control block back to the caller.
3993 ifc.ifc_len = total;
3996 * Both BSD and Solaris return 0 here, so we do too.
3998 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4001 #ifdef CONFIG_PROC_FS
4003 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4005 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4006 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4007 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4009 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4011 struct net *net = seq_file_net(seq);
4012 struct net_device *dev;
4013 struct hlist_node *p;
4014 struct hlist_head *h;
4015 unsigned int count = 0, offset = get_offset(*pos);
4017 h = &net->dev_name_head[get_bucket(*pos)];
4018 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4019 if (++count == offset)
4026 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4028 struct net_device *dev;
4029 unsigned int bucket;
4032 dev = dev_from_same_bucket(seq, pos);
4036 bucket = get_bucket(*pos) + 1;
4037 *pos = set_bucket_offset(bucket, 1);
4038 } while (bucket < NETDEV_HASHENTRIES);
4044 * This is invoked by the /proc filesystem handler to display a device
4047 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4052 return SEQ_START_TOKEN;
4054 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4057 return dev_from_bucket(seq, pos);
4060 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4063 return dev_from_bucket(seq, pos);
4066 void dev_seq_stop(struct seq_file *seq, void *v)
4072 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4074 struct rtnl_link_stats64 temp;
4075 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4077 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4078 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4079 dev->name, stats->rx_bytes, stats->rx_packets,
4081 stats->rx_dropped + stats->rx_missed_errors,
4082 stats->rx_fifo_errors,
4083 stats->rx_length_errors + stats->rx_over_errors +
4084 stats->rx_crc_errors + stats->rx_frame_errors,
4085 stats->rx_compressed, stats->multicast,
4086 stats->tx_bytes, stats->tx_packets,
4087 stats->tx_errors, stats->tx_dropped,
4088 stats->tx_fifo_errors, stats->collisions,
4089 stats->tx_carrier_errors +
4090 stats->tx_aborted_errors +
4091 stats->tx_window_errors +
4092 stats->tx_heartbeat_errors,
4093 stats->tx_compressed);
4097 * Called from the PROCfs module. This now uses the new arbitrary sized
4098 * /proc/net interface to create /proc/net/dev
4100 static int dev_seq_show(struct seq_file *seq, void *v)
4102 if (v == SEQ_START_TOKEN)
4103 seq_puts(seq, "Inter-| Receive "
4105 " face |bytes packets errs drop fifo frame "
4106 "compressed multicast|bytes packets errs "
4107 "drop fifo colls carrier compressed\n");
4109 dev_seq_printf_stats(seq, v);
4113 static struct softnet_data *softnet_get_online(loff_t *pos)
4115 struct softnet_data *sd = NULL;
4117 while (*pos < nr_cpu_ids)
4118 if (cpu_online(*pos)) {
4119 sd = &per_cpu(softnet_data, *pos);
4126 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4128 return softnet_get_online(pos);
4131 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4134 return softnet_get_online(pos);
4137 static void softnet_seq_stop(struct seq_file *seq, void *v)
4141 static int softnet_seq_show(struct seq_file *seq, void *v)
4143 struct softnet_data *sd = v;
4145 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4146 sd->processed, sd->dropped, sd->time_squeeze, 0,
4147 0, 0, 0, 0, /* was fastroute */
4148 sd->cpu_collision, sd->received_rps);
4152 static const struct seq_operations dev_seq_ops = {
4153 .start = dev_seq_start,
4154 .next = dev_seq_next,
4155 .stop = dev_seq_stop,
4156 .show = dev_seq_show,
4159 static int dev_seq_open(struct inode *inode, struct file *file)
4161 return seq_open_net(inode, file, &dev_seq_ops,
4162 sizeof(struct seq_net_private));
4165 static const struct file_operations dev_seq_fops = {
4166 .owner = THIS_MODULE,
4167 .open = dev_seq_open,
4169 .llseek = seq_lseek,
4170 .release = seq_release_net,
4173 static const struct seq_operations softnet_seq_ops = {
4174 .start = softnet_seq_start,
4175 .next = softnet_seq_next,
4176 .stop = softnet_seq_stop,
4177 .show = softnet_seq_show,
4180 static int softnet_seq_open(struct inode *inode, struct file *file)
4182 return seq_open(file, &softnet_seq_ops);
4185 static const struct file_operations softnet_seq_fops = {
4186 .owner = THIS_MODULE,
4187 .open = softnet_seq_open,
4189 .llseek = seq_lseek,
4190 .release = seq_release,
4193 static void *ptype_get_idx(loff_t pos)
4195 struct packet_type *pt = NULL;
4199 list_for_each_entry_rcu(pt, &ptype_all, list) {
4205 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4206 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4215 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4219 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4222 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4224 struct packet_type *pt;
4225 struct list_head *nxt;
4229 if (v == SEQ_START_TOKEN)
4230 return ptype_get_idx(0);
4233 nxt = pt->list.next;
4234 if (pt->type == htons(ETH_P_ALL)) {
4235 if (nxt != &ptype_all)
4238 nxt = ptype_base[0].next;
4240 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4242 while (nxt == &ptype_base[hash]) {
4243 if (++hash >= PTYPE_HASH_SIZE)
4245 nxt = ptype_base[hash].next;
4248 return list_entry(nxt, struct packet_type, list);
4251 static void ptype_seq_stop(struct seq_file *seq, void *v)
4257 static int ptype_seq_show(struct seq_file *seq, void *v)
4259 struct packet_type *pt = v;
4261 if (v == SEQ_START_TOKEN)
4262 seq_puts(seq, "Type Device Function\n");
4263 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4264 if (pt->type == htons(ETH_P_ALL))
4265 seq_puts(seq, "ALL ");
4267 seq_printf(seq, "%04x", ntohs(pt->type));
4269 seq_printf(seq, " %-8s %pF\n",
4270 pt->dev ? pt->dev->name : "", pt->func);
4276 static const struct seq_operations ptype_seq_ops = {
4277 .start = ptype_seq_start,
4278 .next = ptype_seq_next,
4279 .stop = ptype_seq_stop,
4280 .show = ptype_seq_show,
4283 static int ptype_seq_open(struct inode *inode, struct file *file)
4285 return seq_open_net(inode, file, &ptype_seq_ops,
4286 sizeof(struct seq_net_private));
4289 static const struct file_operations ptype_seq_fops = {
4290 .owner = THIS_MODULE,
4291 .open = ptype_seq_open,
4293 .llseek = seq_lseek,
4294 .release = seq_release_net,
4298 static int __net_init dev_proc_net_init(struct net *net)
4302 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4304 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4306 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4309 if (wext_proc_init(net))
4315 proc_net_remove(net, "ptype");
4317 proc_net_remove(net, "softnet_stat");
4319 proc_net_remove(net, "dev");
4323 static void __net_exit dev_proc_net_exit(struct net *net)
4325 wext_proc_exit(net);
4327 proc_net_remove(net, "ptype");
4328 proc_net_remove(net, "softnet_stat");
4329 proc_net_remove(net, "dev");
4332 static struct pernet_operations __net_initdata dev_proc_ops = {
4333 .init = dev_proc_net_init,
4334 .exit = dev_proc_net_exit,
4337 static int __init dev_proc_init(void)
4339 return register_pernet_subsys(&dev_proc_ops);
4342 #define dev_proc_init() 0
4343 #endif /* CONFIG_PROC_FS */
4347 * netdev_set_master - set up master pointer
4348 * @slave: slave device
4349 * @master: new master device
4351 * Changes the master device of the slave. Pass %NULL to break the
4352 * bonding. The caller must hold the RTNL semaphore. On a failure
4353 * a negative errno code is returned. On success the reference counts
4354 * are adjusted and the function returns zero.
4356 int netdev_set_master(struct net_device *slave, struct net_device *master)
4358 struct net_device *old = slave->master;
4368 slave->master = master;
4374 EXPORT_SYMBOL(netdev_set_master);
4377 * netdev_set_bond_master - set up bonding master/slave pair
4378 * @slave: slave device
4379 * @master: new master device
4381 * Changes the master device of the slave. Pass %NULL to break the
4382 * bonding. The caller must hold the RTNL semaphore. On a failure
4383 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4384 * to the routing socket and the function returns zero.
4386 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4392 err = netdev_set_master(slave, master);
4396 slave->flags |= IFF_SLAVE;
4398 slave->flags &= ~IFF_SLAVE;
4400 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4403 EXPORT_SYMBOL(netdev_set_bond_master);
4405 static void dev_change_rx_flags(struct net_device *dev, int flags)
4407 const struct net_device_ops *ops = dev->netdev_ops;
4409 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4410 ops->ndo_change_rx_flags(dev, flags);
4413 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4415 unsigned int old_flags = dev->flags;
4421 dev->flags |= IFF_PROMISC;
4422 dev->promiscuity += inc;
4423 if (dev->promiscuity == 0) {
4426 * If inc causes overflow, untouch promisc and return error.
4429 dev->flags &= ~IFF_PROMISC;
4431 dev->promiscuity -= inc;
4432 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4437 if (dev->flags != old_flags) {
4438 pr_info("device %s %s promiscuous mode\n",
4440 dev->flags & IFF_PROMISC ? "entered" : "left");
4441 if (audit_enabled) {
4442 current_uid_gid(&uid, &gid);
4443 audit_log(current->audit_context, GFP_ATOMIC,
4444 AUDIT_ANOM_PROMISCUOUS,
4445 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4446 dev->name, (dev->flags & IFF_PROMISC),
4447 (old_flags & IFF_PROMISC),
4448 audit_get_loginuid(current),
4450 audit_get_sessionid(current));
4453 dev_change_rx_flags(dev, IFF_PROMISC);
4459 * dev_set_promiscuity - update promiscuity count on a device
4463 * Add or remove promiscuity from a device. While the count in the device
4464 * remains above zero the interface remains promiscuous. Once it hits zero
4465 * the device reverts back to normal filtering operation. A negative inc
4466 * value is used to drop promiscuity on the device.
4467 * Return 0 if successful or a negative errno code on error.
4469 int dev_set_promiscuity(struct net_device *dev, int inc)
4471 unsigned int old_flags = dev->flags;
4474 err = __dev_set_promiscuity(dev, inc);
4477 if (dev->flags != old_flags)
4478 dev_set_rx_mode(dev);
4481 EXPORT_SYMBOL(dev_set_promiscuity);
4484 * dev_set_allmulti - update allmulti count on a device
4488 * Add or remove reception of all multicast frames to a device. While the
4489 * count in the device remains above zero the interface remains listening
4490 * to all interfaces. Once it hits zero the device reverts back to normal
4491 * filtering operation. A negative @inc value is used to drop the counter
4492 * when releasing a resource needing all multicasts.
4493 * Return 0 if successful or a negative errno code on error.
4496 int dev_set_allmulti(struct net_device *dev, int inc)
4498 unsigned int old_flags = dev->flags;
4502 dev->flags |= IFF_ALLMULTI;
4503 dev->allmulti += inc;
4504 if (dev->allmulti == 0) {
4507 * If inc causes overflow, untouch allmulti and return error.
4510 dev->flags &= ~IFF_ALLMULTI;
4512 dev->allmulti -= inc;
4513 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4518 if (dev->flags ^ old_flags) {
4519 dev_change_rx_flags(dev, IFF_ALLMULTI);
4520 dev_set_rx_mode(dev);
4524 EXPORT_SYMBOL(dev_set_allmulti);
4527 * Upload unicast and multicast address lists to device and
4528 * configure RX filtering. When the device doesn't support unicast
4529 * filtering it is put in promiscuous mode while unicast addresses
4532 void __dev_set_rx_mode(struct net_device *dev)
4534 const struct net_device_ops *ops = dev->netdev_ops;
4536 /* dev_open will call this function so the list will stay sane. */
4537 if (!(dev->flags&IFF_UP))
4540 if (!netif_device_present(dev))
4543 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4544 /* Unicast addresses changes may only happen under the rtnl,
4545 * therefore calling __dev_set_promiscuity here is safe.
4547 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4548 __dev_set_promiscuity(dev, 1);
4549 dev->uc_promisc = true;
4550 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4551 __dev_set_promiscuity(dev, -1);
4552 dev->uc_promisc = false;
4556 if (ops->ndo_set_rx_mode)
4557 ops->ndo_set_rx_mode(dev);
4560 void dev_set_rx_mode(struct net_device *dev)
4562 netif_addr_lock_bh(dev);
4563 __dev_set_rx_mode(dev);
4564 netif_addr_unlock_bh(dev);
4568 * dev_get_flags - get flags reported to userspace
4571 * Get the combination of flag bits exported through APIs to userspace.
4573 unsigned int dev_get_flags(const struct net_device *dev)
4577 flags = (dev->flags & ~(IFF_PROMISC |
4582 (dev->gflags & (IFF_PROMISC |
4585 if (netif_running(dev)) {
4586 if (netif_oper_up(dev))
4587 flags |= IFF_RUNNING;
4588 if (netif_carrier_ok(dev))
4589 flags |= IFF_LOWER_UP;
4590 if (netif_dormant(dev))
4591 flags |= IFF_DORMANT;
4596 EXPORT_SYMBOL(dev_get_flags);
4598 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4600 unsigned int old_flags = dev->flags;
4606 * Set the flags on our device.
4609 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4610 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4612 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4616 * Load in the correct multicast list now the flags have changed.
4619 if ((old_flags ^ flags) & IFF_MULTICAST)
4620 dev_change_rx_flags(dev, IFF_MULTICAST);
4622 dev_set_rx_mode(dev);
4625 * Have we downed the interface. We handle IFF_UP ourselves
4626 * according to user attempts to set it, rather than blindly
4631 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4632 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4635 dev_set_rx_mode(dev);
4638 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4639 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4641 dev->gflags ^= IFF_PROMISC;
4642 dev_set_promiscuity(dev, inc);
4645 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4646 is important. Some (broken) drivers set IFF_PROMISC, when
4647 IFF_ALLMULTI is requested not asking us and not reporting.
4649 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4650 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4652 dev->gflags ^= IFF_ALLMULTI;
4653 dev_set_allmulti(dev, inc);
4659 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4661 unsigned int changes = dev->flags ^ old_flags;
4663 if (changes & IFF_UP) {
4664 if (dev->flags & IFF_UP)
4665 call_netdevice_notifiers(NETDEV_UP, dev);
4667 call_netdevice_notifiers(NETDEV_DOWN, dev);
4670 if (dev->flags & IFF_UP &&
4671 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4672 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4676 * dev_change_flags - change device settings
4678 * @flags: device state flags
4680 * Change settings on device based state flags. The flags are
4681 * in the userspace exported format.
4683 int dev_change_flags(struct net_device *dev, unsigned int flags)
4686 unsigned int changes, old_flags = dev->flags;
4688 ret = __dev_change_flags(dev, flags);
4692 changes = old_flags ^ dev->flags;
4694 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4696 __dev_notify_flags(dev, old_flags);
4699 EXPORT_SYMBOL(dev_change_flags);
4702 * dev_set_mtu - Change maximum transfer unit
4704 * @new_mtu: new transfer unit
4706 * Change the maximum transfer size of the network device.
4708 int dev_set_mtu(struct net_device *dev, int new_mtu)
4710 const struct net_device_ops *ops = dev->netdev_ops;
4713 if (new_mtu == dev->mtu)
4716 /* MTU must be positive. */
4720 if (!netif_device_present(dev))
4724 if (ops->ndo_change_mtu)
4725 err = ops->ndo_change_mtu(dev, new_mtu);
4729 if (!err && dev->flags & IFF_UP)
4730 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4733 EXPORT_SYMBOL(dev_set_mtu);
4736 * dev_set_group - Change group this device belongs to
4738 * @new_group: group this device should belong to
4740 void dev_set_group(struct net_device *dev, int new_group)
4742 dev->group = new_group;
4744 EXPORT_SYMBOL(dev_set_group);
4747 * dev_set_mac_address - Change Media Access Control Address
4751 * Change the hardware (MAC) address of the device
4753 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4755 const struct net_device_ops *ops = dev->netdev_ops;
4758 if (!ops->ndo_set_mac_address)
4760 if (sa->sa_family != dev->type)
4762 if (!netif_device_present(dev))
4764 err = ops->ndo_set_mac_address(dev, sa);
4766 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4767 add_device_randomness(dev->dev_addr, dev->addr_len);
4770 EXPORT_SYMBOL(dev_set_mac_address);
4773 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4775 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4778 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4784 case SIOCGIFFLAGS: /* Get interface flags */
4785 ifr->ifr_flags = (short) dev_get_flags(dev);
4788 case SIOCGIFMETRIC: /* Get the metric on the interface
4789 (currently unused) */
4790 ifr->ifr_metric = 0;
4793 case SIOCGIFMTU: /* Get the MTU of a device */
4794 ifr->ifr_mtu = dev->mtu;
4799 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4801 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4802 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4803 ifr->ifr_hwaddr.sa_family = dev->type;
4811 ifr->ifr_map.mem_start = dev->mem_start;
4812 ifr->ifr_map.mem_end = dev->mem_end;
4813 ifr->ifr_map.base_addr = dev->base_addr;
4814 ifr->ifr_map.irq = dev->irq;
4815 ifr->ifr_map.dma = dev->dma;
4816 ifr->ifr_map.port = dev->if_port;
4820 ifr->ifr_ifindex = dev->ifindex;
4824 ifr->ifr_qlen = dev->tx_queue_len;
4828 /* dev_ioctl() should ensure this case
4840 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4842 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4845 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4846 const struct net_device_ops *ops;
4851 ops = dev->netdev_ops;
4854 case SIOCSIFFLAGS: /* Set interface flags */
4855 return dev_change_flags(dev, ifr->ifr_flags);
4857 case SIOCSIFMETRIC: /* Set the metric on the interface
4858 (currently unused) */
4861 case SIOCSIFMTU: /* Set the MTU of a device */
4862 return dev_set_mtu(dev, ifr->ifr_mtu);
4865 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4867 case SIOCSIFHWBROADCAST:
4868 if (ifr->ifr_hwaddr.sa_family != dev->type)
4870 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4871 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4872 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4876 if (ops->ndo_set_config) {
4877 if (!netif_device_present(dev))
4879 return ops->ndo_set_config(dev, &ifr->ifr_map);
4884 if (!ops->ndo_set_rx_mode ||
4885 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4887 if (!netif_device_present(dev))
4889 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4892 if (!ops->ndo_set_rx_mode ||
4893 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4895 if (!netif_device_present(dev))
4897 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4900 if (ifr->ifr_qlen < 0)
4902 dev->tx_queue_len = ifr->ifr_qlen;
4906 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4907 return dev_change_name(dev, ifr->ifr_newname);
4910 err = net_hwtstamp_validate(ifr);
4916 * Unknown or private ioctl
4919 if ((cmd >= SIOCDEVPRIVATE &&
4920 cmd <= SIOCDEVPRIVATE + 15) ||
4921 cmd == SIOCBONDENSLAVE ||
4922 cmd == SIOCBONDRELEASE ||
4923 cmd == SIOCBONDSETHWADDR ||
4924 cmd == SIOCBONDSLAVEINFOQUERY ||
4925 cmd == SIOCBONDINFOQUERY ||
4926 cmd == SIOCBONDCHANGEACTIVE ||
4927 cmd == SIOCGMIIPHY ||
4928 cmd == SIOCGMIIREG ||
4929 cmd == SIOCSMIIREG ||
4930 cmd == SIOCBRADDIF ||
4931 cmd == SIOCBRDELIF ||
4932 cmd == SIOCSHWTSTAMP ||
4933 cmd == SIOCWANDEV) {
4935 if (ops->ndo_do_ioctl) {
4936 if (netif_device_present(dev))
4937 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4949 * This function handles all "interface"-type I/O control requests. The actual
4950 * 'doing' part of this is dev_ifsioc above.
4954 * dev_ioctl - network device ioctl
4955 * @net: the applicable net namespace
4956 * @cmd: command to issue
4957 * @arg: pointer to a struct ifreq in user space
4959 * Issue ioctl functions to devices. This is normally called by the
4960 * user space syscall interfaces but can sometimes be useful for
4961 * other purposes. The return value is the return from the syscall if
4962 * positive or a negative errno code on error.
4965 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4971 /* One special case: SIOCGIFCONF takes ifconf argument
4972 and requires shared lock, because it sleeps writing
4976 if (cmd == SIOCGIFCONF) {
4978 ret = dev_ifconf(net, (char __user *) arg);
4982 if (cmd == SIOCGIFNAME)
4983 return dev_ifname(net, (struct ifreq __user *)arg);
4985 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4988 ifr.ifr_name[IFNAMSIZ-1] = 0;
4990 colon = strchr(ifr.ifr_name, ':');
4995 * See which interface the caller is talking about.
5000 * These ioctl calls:
5001 * - can be done by all.
5002 * - atomic and do not require locking.
5013 dev_load(net, ifr.ifr_name);
5015 ret = dev_ifsioc_locked(net, &ifr, cmd);
5020 if (copy_to_user(arg, &ifr,
5021 sizeof(struct ifreq)))
5027 dev_load(net, ifr.ifr_name);
5029 ret = dev_ethtool(net, &ifr);
5034 if (copy_to_user(arg, &ifr,
5035 sizeof(struct ifreq)))
5041 * These ioctl calls:
5042 * - require superuser power.
5043 * - require strict serialization.
5049 if (!capable(CAP_NET_ADMIN))
5051 dev_load(net, ifr.ifr_name);
5053 ret = dev_ifsioc(net, &ifr, cmd);
5058 if (copy_to_user(arg, &ifr,
5059 sizeof(struct ifreq)))
5065 * These ioctl calls:
5066 * - require superuser power.
5067 * - require strict serialization.
5068 * - do not return a value
5078 case SIOCSIFHWBROADCAST:
5081 case SIOCBONDENSLAVE:
5082 case SIOCBONDRELEASE:
5083 case SIOCBONDSETHWADDR:
5084 case SIOCBONDCHANGEACTIVE:
5088 if (!capable(CAP_NET_ADMIN))
5091 case SIOCBONDSLAVEINFOQUERY:
5092 case SIOCBONDINFOQUERY:
5093 dev_load(net, ifr.ifr_name);
5095 ret = dev_ifsioc(net, &ifr, cmd);
5100 /* Get the per device memory space. We can add this but
5101 * currently do not support it */
5103 /* Set the per device memory buffer space.
5104 * Not applicable in our case */
5109 * Unknown or private ioctl.
5112 if (cmd == SIOCWANDEV ||
5113 (cmd >= SIOCDEVPRIVATE &&
5114 cmd <= SIOCDEVPRIVATE + 15)) {
5115 dev_load(net, ifr.ifr_name);
5117 ret = dev_ifsioc(net, &ifr, cmd);
5119 if (!ret && copy_to_user(arg, &ifr,
5120 sizeof(struct ifreq)))
5124 /* Take care of Wireless Extensions */
5125 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5126 return wext_handle_ioctl(net, &ifr, cmd, arg);
5133 * dev_new_index - allocate an ifindex
5134 * @net: the applicable net namespace
5136 * Returns a suitable unique value for a new device interface
5137 * number. The caller must hold the rtnl semaphore or the
5138 * dev_base_lock to be sure it remains unique.
5140 static int dev_new_index(struct net *net)
5146 if (!__dev_get_by_index(net, ifindex))
5151 /* Delayed registration/unregisteration */
5152 static LIST_HEAD(net_todo_list);
5154 static void net_set_todo(struct net_device *dev)
5156 list_add_tail(&dev->todo_list, &net_todo_list);
5159 static void rollback_registered_many(struct list_head *head)
5161 struct net_device *dev, *tmp;
5163 BUG_ON(dev_boot_phase);
5166 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5167 /* Some devices call without registering
5168 * for initialization unwind. Remove those
5169 * devices and proceed with the remaining.
5171 if (dev->reg_state == NETREG_UNINITIALIZED) {
5172 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5176 list_del(&dev->unreg_list);
5179 dev->dismantle = true;
5180 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5183 /* If device is running, close it first. */
5184 dev_close_many(head);
5186 list_for_each_entry(dev, head, unreg_list) {
5187 /* And unlink it from device chain. */
5188 unlist_netdevice(dev);
5190 dev->reg_state = NETREG_UNREGISTERING;
5195 list_for_each_entry(dev, head, unreg_list) {
5196 /* Shutdown queueing discipline. */
5200 /* Notify protocols, that we are about to destroy
5201 this device. They should clean all the things.
5203 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5205 if (!dev->rtnl_link_ops ||
5206 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5207 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5210 * Flush the unicast and multicast chains
5215 if (dev->netdev_ops->ndo_uninit)
5216 dev->netdev_ops->ndo_uninit(dev);
5218 /* Notifier chain MUST detach us from master device. */
5219 WARN_ON(dev->master);
5221 /* Remove entries from kobject tree */
5222 netdev_unregister_kobject(dev);
5225 /* Process any work delayed until the end of the batch */
5226 dev = list_first_entry(head, struct net_device, unreg_list);
5227 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5231 list_for_each_entry(dev, head, unreg_list)
5235 static void rollback_registered(struct net_device *dev)
5239 list_add(&dev->unreg_list, &single);
5240 rollback_registered_many(&single);
5244 static netdev_features_t netdev_fix_features(struct net_device *dev,
5245 netdev_features_t features)
5247 /* Fix illegal checksum combinations */
5248 if ((features & NETIF_F_HW_CSUM) &&
5249 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5250 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5251 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5254 /* Fix illegal SG+CSUM combinations. */
5255 if ((features & NETIF_F_SG) &&
5256 !(features & NETIF_F_ALL_CSUM)) {
5258 "Dropping NETIF_F_SG since no checksum feature.\n");
5259 features &= ~NETIF_F_SG;
5262 /* TSO requires that SG is present as well. */
5263 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5264 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5265 features &= ~NETIF_F_ALL_TSO;
5268 /* TSO ECN requires that TSO is present as well. */
5269 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5270 features &= ~NETIF_F_TSO_ECN;
5272 /* Software GSO depends on SG. */
5273 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5274 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5275 features &= ~NETIF_F_GSO;
5278 /* UFO needs SG and checksumming */
5279 if (features & NETIF_F_UFO) {
5280 /* maybe split UFO into V4 and V6? */
5281 if (!((features & NETIF_F_GEN_CSUM) ||
5282 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5283 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5285 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5286 features &= ~NETIF_F_UFO;
5289 if (!(features & NETIF_F_SG)) {
5291 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5292 features &= ~NETIF_F_UFO;
5299 int __netdev_update_features(struct net_device *dev)
5301 netdev_features_t features;
5306 features = netdev_get_wanted_features(dev);
5308 if (dev->netdev_ops->ndo_fix_features)
5309 features = dev->netdev_ops->ndo_fix_features(dev, features);
5311 /* driver might be less strict about feature dependencies */
5312 features = netdev_fix_features(dev, features);
5314 if (dev->features == features)
5317 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5318 &dev->features, &features);
5320 if (dev->netdev_ops->ndo_set_features)
5321 err = dev->netdev_ops->ndo_set_features(dev, features);
5323 if (unlikely(err < 0)) {
5325 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5326 err, &features, &dev->features);
5331 dev->features = features;
5337 * netdev_update_features - recalculate device features
5338 * @dev: the device to check
5340 * Recalculate dev->features set and send notifications if it
5341 * has changed. Should be called after driver or hardware dependent
5342 * conditions might have changed that influence the features.
5344 void netdev_update_features(struct net_device *dev)
5346 if (__netdev_update_features(dev))
5347 netdev_features_change(dev);
5349 EXPORT_SYMBOL(netdev_update_features);
5352 * netdev_change_features - recalculate device features
5353 * @dev: the device to check
5355 * Recalculate dev->features set and send notifications even
5356 * if they have not changed. Should be called instead of
5357 * netdev_update_features() if also dev->vlan_features might
5358 * have changed to allow the changes to be propagated to stacked
5361 void netdev_change_features(struct net_device *dev)
5363 __netdev_update_features(dev);
5364 netdev_features_change(dev);
5366 EXPORT_SYMBOL(netdev_change_features);
5369 * netif_stacked_transfer_operstate - transfer operstate
5370 * @rootdev: the root or lower level device to transfer state from
5371 * @dev: the device to transfer operstate to
5373 * Transfer operational state from root to device. This is normally
5374 * called when a stacking relationship exists between the root
5375 * device and the device(a leaf device).
5377 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5378 struct net_device *dev)
5380 if (rootdev->operstate == IF_OPER_DORMANT)
5381 netif_dormant_on(dev);
5383 netif_dormant_off(dev);
5385 if (netif_carrier_ok(rootdev)) {
5386 if (!netif_carrier_ok(dev))
5387 netif_carrier_on(dev);
5389 if (netif_carrier_ok(dev))
5390 netif_carrier_off(dev);
5393 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5396 static int netif_alloc_rx_queues(struct net_device *dev)
5398 unsigned int i, count = dev->num_rx_queues;
5399 struct netdev_rx_queue *rx;
5403 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5405 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5410 for (i = 0; i < count; i++)
5416 static void netdev_init_one_queue(struct net_device *dev,
5417 struct netdev_queue *queue, void *_unused)
5419 /* Initialize queue lock */
5420 spin_lock_init(&queue->_xmit_lock);
5421 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5422 queue->xmit_lock_owner = -1;
5423 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5426 dql_init(&queue->dql, HZ);
5430 static int netif_alloc_netdev_queues(struct net_device *dev)
5432 unsigned int count = dev->num_tx_queues;
5433 struct netdev_queue *tx;
5437 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5439 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5444 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5445 spin_lock_init(&dev->tx_global_lock);
5451 * register_netdevice - register a network device
5452 * @dev: device to register
5454 * Take a completed network device structure and add it to the kernel
5455 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5456 * chain. 0 is returned on success. A negative errno code is returned
5457 * on a failure to set up the device, or if the name is a duplicate.
5459 * Callers must hold the rtnl semaphore. You may want
5460 * register_netdev() instead of this.
5463 * The locking appears insufficient to guarantee two parallel registers
5464 * will not get the same name.
5467 int register_netdevice(struct net_device *dev)
5470 struct net *net = dev_net(dev);
5472 BUG_ON(dev_boot_phase);
5477 /* When net_device's are persistent, this will be fatal. */
5478 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5481 spin_lock_init(&dev->addr_list_lock);
5482 netdev_set_addr_lockdep_class(dev);
5486 ret = dev_get_valid_name(dev, dev->name);
5490 /* Init, if this function is available */
5491 if (dev->netdev_ops->ndo_init) {
5492 ret = dev->netdev_ops->ndo_init(dev);
5500 dev->ifindex = dev_new_index(net);
5501 if (dev->iflink == -1)
5502 dev->iflink = dev->ifindex;
5504 /* Transfer changeable features to wanted_features and enable
5505 * software offloads (GSO and GRO).
5507 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5508 dev->features |= NETIF_F_SOFT_FEATURES;
5509 dev->wanted_features = dev->features & dev->hw_features;
5511 /* Turn on no cache copy if HW is doing checksum */
5512 if (!(dev->flags & IFF_LOOPBACK)) {
5513 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5514 if (dev->features & NETIF_F_ALL_CSUM) {
5515 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5516 dev->features |= NETIF_F_NOCACHE_COPY;
5520 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5522 dev->vlan_features |= NETIF_F_HIGHDMA;
5524 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5525 ret = notifier_to_errno(ret);
5529 ret = netdev_register_kobject(dev);
5532 dev->reg_state = NETREG_REGISTERED;
5534 __netdev_update_features(dev);
5537 * Default initial state at registry is that the
5538 * device is present.
5541 set_bit(__LINK_STATE_PRESENT, &dev->state);
5543 dev_init_scheduler(dev);
5545 list_netdevice(dev);
5546 add_device_randomness(dev->dev_addr, dev->addr_len);
5548 /* Notify protocols, that a new device appeared. */
5549 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5550 ret = notifier_to_errno(ret);
5552 rollback_registered(dev);
5553 dev->reg_state = NETREG_UNREGISTERED;
5556 * Prevent userspace races by waiting until the network
5557 * device is fully setup before sending notifications.
5559 if (!dev->rtnl_link_ops ||
5560 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5561 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5567 if (dev->netdev_ops->ndo_uninit)
5568 dev->netdev_ops->ndo_uninit(dev);
5571 EXPORT_SYMBOL(register_netdevice);
5574 * init_dummy_netdev - init a dummy network device for NAPI
5575 * @dev: device to init
5577 * This takes a network device structure and initialize the minimum
5578 * amount of fields so it can be used to schedule NAPI polls without
5579 * registering a full blown interface. This is to be used by drivers
5580 * that need to tie several hardware interfaces to a single NAPI
5581 * poll scheduler due to HW limitations.
5583 int init_dummy_netdev(struct net_device *dev)
5585 /* Clear everything. Note we don't initialize spinlocks
5586 * are they aren't supposed to be taken by any of the
5587 * NAPI code and this dummy netdev is supposed to be
5588 * only ever used for NAPI polls
5590 memset(dev, 0, sizeof(struct net_device));
5592 /* make sure we BUG if trying to hit standard
5593 * register/unregister code path
5595 dev->reg_state = NETREG_DUMMY;
5597 /* NAPI wants this */
5598 INIT_LIST_HEAD(&dev->napi_list);
5600 /* a dummy interface is started by default */
5601 set_bit(__LINK_STATE_PRESENT, &dev->state);
5602 set_bit(__LINK_STATE_START, &dev->state);
5604 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5605 * because users of this 'device' dont need to change
5611 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5615 * register_netdev - register a network device
5616 * @dev: device to register
5618 * Take a completed network device structure and add it to the kernel
5619 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5620 * chain. 0 is returned on success. A negative errno code is returned
5621 * on a failure to set up the device, or if the name is a duplicate.
5623 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5624 * and expands the device name if you passed a format string to
5627 int register_netdev(struct net_device *dev)
5632 err = register_netdevice(dev);
5636 EXPORT_SYMBOL(register_netdev);
5638 int netdev_refcnt_read(const struct net_device *dev)
5642 for_each_possible_cpu(i)
5643 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5646 EXPORT_SYMBOL(netdev_refcnt_read);
5649 * netdev_wait_allrefs - wait until all references are gone.
5651 * This is called when unregistering network devices.
5653 * Any protocol or device that holds a reference should register
5654 * for netdevice notification, and cleanup and put back the
5655 * reference if they receive an UNREGISTER event.
5656 * We can get stuck here if buggy protocols don't correctly
5659 static void netdev_wait_allrefs(struct net_device *dev)
5661 unsigned long rebroadcast_time, warning_time;
5664 linkwatch_forget_dev(dev);
5666 rebroadcast_time = warning_time = jiffies;
5667 refcnt = netdev_refcnt_read(dev);
5669 while (refcnt != 0) {
5670 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5673 /* Rebroadcast unregister notification */
5674 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5675 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5676 * should have already handle it the first time */
5678 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5680 /* We must not have linkwatch events
5681 * pending on unregister. If this
5682 * happens, we simply run the queue
5683 * unscheduled, resulting in a noop
5686 linkwatch_run_queue();
5691 rebroadcast_time = jiffies;
5696 refcnt = netdev_refcnt_read(dev);
5698 if (time_after(jiffies, warning_time + 10 * HZ)) {
5699 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5701 warning_time = jiffies;
5710 * register_netdevice(x1);
5711 * register_netdevice(x2);
5713 * unregister_netdevice(y1);
5714 * unregister_netdevice(y2);
5720 * We are invoked by rtnl_unlock().
5721 * This allows us to deal with problems:
5722 * 1) We can delete sysfs objects which invoke hotplug
5723 * without deadlocking with linkwatch via keventd.
5724 * 2) Since we run with the RTNL semaphore not held, we can sleep
5725 * safely in order to wait for the netdev refcnt to drop to zero.
5727 * We must not return until all unregister events added during
5728 * the interval the lock was held have been completed.
5730 void netdev_run_todo(void)
5732 struct list_head list;
5734 /* Snapshot list, allow later requests */
5735 list_replace_init(&net_todo_list, &list);
5739 /* Wait for rcu callbacks to finish before attempting to drain
5740 * the device list. This usually avoids a 250ms wait.
5742 if (!list_empty(&list))
5745 while (!list_empty(&list)) {
5746 struct net_device *dev
5747 = list_first_entry(&list, struct net_device, todo_list);
5748 list_del(&dev->todo_list);
5750 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5751 pr_err("network todo '%s' but state %d\n",
5752 dev->name, dev->reg_state);
5757 dev->reg_state = NETREG_UNREGISTERED;
5759 on_each_cpu(flush_backlog, dev, 1);
5761 netdev_wait_allrefs(dev);
5764 BUG_ON(netdev_refcnt_read(dev));
5765 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5766 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5767 WARN_ON(dev->dn_ptr);
5769 if (dev->destructor)
5770 dev->destructor(dev);
5772 /* Free network device */
5773 kobject_put(&dev->dev.kobj);
5777 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5778 * fields in the same order, with only the type differing.
5780 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5781 const struct net_device_stats *netdev_stats)
5783 #if BITS_PER_LONG == 64
5784 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5785 memcpy(stats64, netdev_stats, sizeof(*stats64));
5787 size_t i, n = sizeof(*stats64) / sizeof(u64);
5788 const unsigned long *src = (const unsigned long *)netdev_stats;
5789 u64 *dst = (u64 *)stats64;
5791 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5792 sizeof(*stats64) / sizeof(u64));
5793 for (i = 0; i < n; i++)
5797 EXPORT_SYMBOL(netdev_stats_to_stats64);
5800 * dev_get_stats - get network device statistics
5801 * @dev: device to get statistics from
5802 * @storage: place to store stats
5804 * Get network statistics from device. Return @storage.
5805 * The device driver may provide its own method by setting
5806 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5807 * otherwise the internal statistics structure is used.
5809 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5810 struct rtnl_link_stats64 *storage)
5812 const struct net_device_ops *ops = dev->netdev_ops;
5814 if (ops->ndo_get_stats64) {
5815 memset(storage, 0, sizeof(*storage));
5816 ops->ndo_get_stats64(dev, storage);
5817 } else if (ops->ndo_get_stats) {
5818 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5820 netdev_stats_to_stats64(storage, &dev->stats);
5822 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5825 EXPORT_SYMBOL(dev_get_stats);
5827 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5829 struct netdev_queue *queue = dev_ingress_queue(dev);
5831 #ifdef CONFIG_NET_CLS_ACT
5834 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5837 netdev_init_one_queue(dev, queue, NULL);
5838 queue->qdisc = &noop_qdisc;
5839 queue->qdisc_sleeping = &noop_qdisc;
5840 rcu_assign_pointer(dev->ingress_queue, queue);
5846 * alloc_netdev_mqs - allocate network device
5847 * @sizeof_priv: size of private data to allocate space for
5848 * @name: device name format string
5849 * @setup: callback to initialize device
5850 * @txqs: the number of TX subqueues to allocate
5851 * @rxqs: the number of RX subqueues to allocate
5853 * Allocates a struct net_device with private data area for driver use
5854 * and performs basic initialization. Also allocates subquue structs
5855 * for each queue on the device.
5857 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5858 void (*setup)(struct net_device *),
5859 unsigned int txqs, unsigned int rxqs)
5861 struct net_device *dev;
5863 struct net_device *p;
5865 BUG_ON(strlen(name) >= sizeof(dev->name));
5868 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5874 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5879 alloc_size = sizeof(struct net_device);
5881 /* ensure 32-byte alignment of private area */
5882 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5883 alloc_size += sizeof_priv;
5885 /* ensure 32-byte alignment of whole construct */
5886 alloc_size += NETDEV_ALIGN - 1;
5888 p = kzalloc(alloc_size, GFP_KERNEL);
5890 pr_err("alloc_netdev: Unable to allocate device\n");
5894 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5895 dev->padded = (char *)dev - (char *)p;
5897 dev->pcpu_refcnt = alloc_percpu(int);
5898 if (!dev->pcpu_refcnt)
5901 if (dev_addr_init(dev))
5907 dev_net_set(dev, &init_net);
5909 dev->gso_max_size = GSO_MAX_SIZE;
5911 INIT_LIST_HEAD(&dev->napi_list);
5912 INIT_LIST_HEAD(&dev->unreg_list);
5913 INIT_LIST_HEAD(&dev->link_watch_list);
5914 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5917 dev->num_tx_queues = txqs;
5918 dev->real_num_tx_queues = txqs;
5919 if (netif_alloc_netdev_queues(dev))
5923 dev->num_rx_queues = rxqs;
5924 dev->real_num_rx_queues = rxqs;
5925 if (netif_alloc_rx_queues(dev))
5929 strcpy(dev->name, name);
5930 dev->group = INIT_NETDEV_GROUP;
5938 free_percpu(dev->pcpu_refcnt);
5948 EXPORT_SYMBOL(alloc_netdev_mqs);
5951 * free_netdev - free network device
5954 * This function does the last stage of destroying an allocated device
5955 * interface. The reference to the device object is released.
5956 * If this is the last reference then it will be freed.
5958 void free_netdev(struct net_device *dev)
5960 struct napi_struct *p, *n;
5962 release_net(dev_net(dev));
5969 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5971 /* Flush device addresses */
5972 dev_addr_flush(dev);
5974 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5977 free_percpu(dev->pcpu_refcnt);
5978 dev->pcpu_refcnt = NULL;
5980 /* Compatibility with error handling in drivers */
5981 if (dev->reg_state == NETREG_UNINITIALIZED) {
5982 kfree((char *)dev - dev->padded);
5986 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5987 dev->reg_state = NETREG_RELEASED;
5989 /* will free via device release */
5990 put_device(&dev->dev);
5992 EXPORT_SYMBOL(free_netdev);
5995 * synchronize_net - Synchronize with packet receive processing
5997 * Wait for packets currently being received to be done.
5998 * Does not block later packets from starting.
6000 void synchronize_net(void)
6003 if (rtnl_is_locked())
6004 synchronize_rcu_expedited();
6008 EXPORT_SYMBOL(synchronize_net);
6011 * unregister_netdevice_queue - remove device from the kernel
6015 * This function shuts down a device interface and removes it
6016 * from the kernel tables.
6017 * If head not NULL, device is queued to be unregistered later.
6019 * Callers must hold the rtnl semaphore. You may want
6020 * unregister_netdev() instead of this.
6023 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6028 list_move_tail(&dev->unreg_list, head);
6030 rollback_registered(dev);
6031 /* Finish processing unregister after unlock */
6035 EXPORT_SYMBOL(unregister_netdevice_queue);
6038 * unregister_netdevice_many - unregister many devices
6039 * @head: list of devices
6041 void unregister_netdevice_many(struct list_head *head)
6043 struct net_device *dev;
6045 if (!list_empty(head)) {
6046 rollback_registered_many(head);
6047 list_for_each_entry(dev, head, unreg_list)
6051 EXPORT_SYMBOL(unregister_netdevice_many);
6054 * unregister_netdev - remove device from the kernel
6057 * This function shuts down a device interface and removes it
6058 * from the kernel tables.
6060 * This is just a wrapper for unregister_netdevice that takes
6061 * the rtnl semaphore. In general you want to use this and not
6062 * unregister_netdevice.
6064 void unregister_netdev(struct net_device *dev)
6067 unregister_netdevice(dev);
6070 EXPORT_SYMBOL(unregister_netdev);
6073 * dev_change_net_namespace - move device to different nethost namespace
6075 * @net: network namespace
6076 * @pat: If not NULL name pattern to try if the current device name
6077 * is already taken in the destination network namespace.
6079 * This function shuts down a device interface and moves it
6080 * to a new network namespace. On success 0 is returned, on
6081 * a failure a netagive errno code is returned.
6083 * Callers must hold the rtnl semaphore.
6086 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6092 /* Don't allow namespace local devices to be moved. */
6094 if (dev->features & NETIF_F_NETNS_LOCAL)
6097 /* Ensure the device has been registrered */
6099 if (dev->reg_state != NETREG_REGISTERED)
6102 /* Get out if there is nothing todo */
6104 if (net_eq(dev_net(dev), net))
6107 /* Pick the destination device name, and ensure
6108 * we can use it in the destination network namespace.
6111 if (__dev_get_by_name(net, dev->name)) {
6112 /* We get here if we can't use the current device name */
6115 if (dev_get_valid_name(dev, pat) < 0)
6120 * And now a mini version of register_netdevice unregister_netdevice.
6123 /* If device is running close it first. */
6126 /* And unlink it from device chain */
6128 unlist_netdevice(dev);
6132 /* Shutdown queueing discipline. */
6135 /* Notify protocols, that we are about to destroy
6136 this device. They should clean all the things.
6138 Note that dev->reg_state stays at NETREG_REGISTERED.
6139 This is wanted because this way 8021q and macvlan know
6140 the device is just moving and can keep their slaves up.
6142 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6143 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6144 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6147 * Flush the unicast and multicast chains
6152 /* Actually switch the network namespace */
6153 dev_net_set(dev, net);
6155 /* If there is an ifindex conflict assign a new one */
6156 if (__dev_get_by_index(net, dev->ifindex)) {
6157 int iflink = (dev->iflink == dev->ifindex);
6158 dev->ifindex = dev_new_index(net);
6160 dev->iflink = dev->ifindex;
6163 /* Fixup kobjects */
6164 err = device_rename(&dev->dev, dev->name);
6167 /* Add the device back in the hashes */
6168 list_netdevice(dev);
6170 /* Notify protocols, that a new device appeared. */
6171 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6174 * Prevent userspace races by waiting until the network
6175 * device is fully setup before sending notifications.
6177 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6184 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6186 static int dev_cpu_callback(struct notifier_block *nfb,
6187 unsigned long action,
6190 struct sk_buff **list_skb;
6191 struct sk_buff *skb;
6192 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6193 struct softnet_data *sd, *oldsd;
6195 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6198 local_irq_disable();
6199 cpu = smp_processor_id();
6200 sd = &per_cpu(softnet_data, cpu);
6201 oldsd = &per_cpu(softnet_data, oldcpu);
6203 /* Find end of our completion_queue. */
6204 list_skb = &sd->completion_queue;
6206 list_skb = &(*list_skb)->next;
6207 /* Append completion queue from offline CPU. */
6208 *list_skb = oldsd->completion_queue;
6209 oldsd->completion_queue = NULL;
6211 /* Append output queue from offline CPU. */
6212 if (oldsd->output_queue) {
6213 *sd->output_queue_tailp = oldsd->output_queue;
6214 sd->output_queue_tailp = oldsd->output_queue_tailp;
6215 oldsd->output_queue = NULL;
6216 oldsd->output_queue_tailp = &oldsd->output_queue;
6218 /* Append NAPI poll list from offline CPU. */
6219 if (!list_empty(&oldsd->poll_list)) {
6220 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6221 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6224 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6227 /* Process offline CPU's input_pkt_queue */
6228 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6230 input_queue_head_incr(oldsd);
6232 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6234 input_queue_head_incr(oldsd);
6242 * netdev_increment_features - increment feature set by one
6243 * @all: current feature set
6244 * @one: new feature set
6245 * @mask: mask feature set
6247 * Computes a new feature set after adding a device with feature set
6248 * @one to the master device with current feature set @all. Will not
6249 * enable anything that is off in @mask. Returns the new feature set.
6251 netdev_features_t netdev_increment_features(netdev_features_t all,
6252 netdev_features_t one, netdev_features_t mask)
6254 if (mask & NETIF_F_GEN_CSUM)
6255 mask |= NETIF_F_ALL_CSUM;
6256 mask |= NETIF_F_VLAN_CHALLENGED;
6258 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6259 all &= one | ~NETIF_F_ALL_FOR_ALL;
6261 /* If one device supports hw checksumming, set for all. */
6262 if (all & NETIF_F_GEN_CSUM)
6263 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6267 EXPORT_SYMBOL(netdev_increment_features);
6269 static struct hlist_head *netdev_create_hash(void)
6272 struct hlist_head *hash;
6274 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6276 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6277 INIT_HLIST_HEAD(&hash[i]);
6282 /* Initialize per network namespace state */
6283 static int __net_init netdev_init(struct net *net)
6285 INIT_LIST_HEAD(&net->dev_base_head);
6287 net->dev_name_head = netdev_create_hash();
6288 if (net->dev_name_head == NULL)
6291 net->dev_index_head = netdev_create_hash();
6292 if (net->dev_index_head == NULL)
6298 kfree(net->dev_name_head);
6304 * netdev_drivername - network driver for the device
6305 * @dev: network device
6307 * Determine network driver for device.
6309 const char *netdev_drivername(const struct net_device *dev)
6311 const struct device_driver *driver;
6312 const struct device *parent;
6313 const char *empty = "";
6315 parent = dev->dev.parent;
6319 driver = parent->driver;
6320 if (driver && driver->name)
6321 return driver->name;
6325 int __netdev_printk(const char *level, const struct net_device *dev,
6326 struct va_format *vaf)
6330 if (dev && dev->dev.parent)
6331 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6332 netdev_name(dev), vaf);
6334 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6336 r = printk("%s(NULL net_device): %pV", level, vaf);
6340 EXPORT_SYMBOL(__netdev_printk);
6342 int netdev_printk(const char *level, const struct net_device *dev,
6343 const char *format, ...)
6345 struct va_format vaf;
6349 va_start(args, format);
6354 r = __netdev_printk(level, dev, &vaf);
6359 EXPORT_SYMBOL(netdev_printk);
6361 #define define_netdev_printk_level(func, level) \
6362 int func(const struct net_device *dev, const char *fmt, ...) \
6365 struct va_format vaf; \
6368 va_start(args, fmt); \
6373 r = __netdev_printk(level, dev, &vaf); \
6378 EXPORT_SYMBOL(func);
6380 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6381 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6382 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6383 define_netdev_printk_level(netdev_err, KERN_ERR);
6384 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6385 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6386 define_netdev_printk_level(netdev_info, KERN_INFO);
6388 static void __net_exit netdev_exit(struct net *net)
6390 kfree(net->dev_name_head);
6391 kfree(net->dev_index_head);
6394 static struct pernet_operations __net_initdata netdev_net_ops = {
6395 .init = netdev_init,
6396 .exit = netdev_exit,
6399 static void __net_exit default_device_exit(struct net *net)
6401 struct net_device *dev, *aux;
6403 * Push all migratable network devices back to the
6404 * initial network namespace
6407 for_each_netdev_safe(net, dev, aux) {
6409 char fb_name[IFNAMSIZ];
6411 /* Ignore unmoveable devices (i.e. loopback) */
6412 if (dev->features & NETIF_F_NETNS_LOCAL)
6415 /* Leave virtual devices for the generic cleanup */
6416 if (dev->rtnl_link_ops)
6419 /* Push remaining network devices to init_net */
6420 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6421 err = dev_change_net_namespace(dev, &init_net, fb_name);
6423 pr_emerg("%s: failed to move %s to init_net: %d\n",
6424 __func__, dev->name, err);
6431 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6433 /* At exit all network devices most be removed from a network
6434 * namespace. Do this in the reverse order of registration.
6435 * Do this across as many network namespaces as possible to
6436 * improve batching efficiency.
6438 struct net_device *dev;
6440 LIST_HEAD(dev_kill_list);
6443 list_for_each_entry(net, net_list, exit_list) {
6444 for_each_netdev_reverse(net, dev) {
6445 if (dev->rtnl_link_ops)
6446 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6448 unregister_netdevice_queue(dev, &dev_kill_list);
6451 unregister_netdevice_many(&dev_kill_list);
6452 list_del(&dev_kill_list);
6456 static struct pernet_operations __net_initdata default_device_ops = {
6457 .exit = default_device_exit,
6458 .exit_batch = default_device_exit_batch,
6462 * Initialize the DEV module. At boot time this walks the device list and
6463 * unhooks any devices that fail to initialise (normally hardware not
6464 * present) and leaves us with a valid list of present and active devices.
6469 * This is called single threaded during boot, so no need
6470 * to take the rtnl semaphore.
6472 static int __init net_dev_init(void)
6474 int i, rc = -ENOMEM;
6476 BUG_ON(!dev_boot_phase);
6478 if (dev_proc_init())
6481 if (netdev_kobject_init())
6484 INIT_LIST_HEAD(&ptype_all);
6485 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6486 INIT_LIST_HEAD(&ptype_base[i]);
6488 if (register_pernet_subsys(&netdev_net_ops))
6492 * Initialise the packet receive queues.
6495 for_each_possible_cpu(i) {
6496 struct softnet_data *sd = &per_cpu(softnet_data, i);
6498 memset(sd, 0, sizeof(*sd));
6499 skb_queue_head_init(&sd->input_pkt_queue);
6500 skb_queue_head_init(&sd->process_queue);
6501 sd->completion_queue = NULL;
6502 INIT_LIST_HEAD(&sd->poll_list);
6503 sd->output_queue = NULL;
6504 sd->output_queue_tailp = &sd->output_queue;
6506 sd->csd.func = rps_trigger_softirq;
6512 sd->backlog.poll = process_backlog;
6513 sd->backlog.weight = weight_p;
6514 sd->backlog.gro_list = NULL;
6515 sd->backlog.gro_count = 0;
6520 /* The loopback device is special if any other network devices
6521 * is present in a network namespace the loopback device must
6522 * be present. Since we now dynamically allocate and free the
6523 * loopback device ensure this invariant is maintained by
6524 * keeping the loopback device as the first device on the
6525 * list of network devices. Ensuring the loopback devices
6526 * is the first device that appears and the last network device
6529 if (register_pernet_device(&loopback_net_ops))
6532 if (register_pernet_device(&default_device_ops))
6535 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6536 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6538 hotcpu_notifier(dev_cpu_callback, 0);
6546 subsys_initcall(net_dev_init);
6548 static int __init initialize_hashrnd(void)
6550 get_random_bytes(&hashrnd, sizeof(hashrnd));
6554 late_initcall_sync(initialize_hashrnd);