2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/net_tstamp.h>
136 #include <linux/static_key.h>
137 #include <net/flow_keys.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 * The list of packet types we will receive (as opposed to discard)
149 * and the routines to invoke.
151 * Why 16. Because with 16 the only overlap we get on a hash of the
152 * low nibble of the protocol value is RARP/SNAP/X.25.
154 * NOTE: That is no longer true with the addition of VLAN tags. Not
155 * sure which should go first, but I bet it won't make much
156 * difference if we are running VLANs. The good news is that
157 * this protocol won't be in the list unless compiled in, so
158 * the average user (w/out VLANs) will not be adversely affected.
175 #define PTYPE_HASH_SIZE (16)
176 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
178 static DEFINE_SPINLOCK(ptype_lock);
179 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
180 static struct list_head ptype_all __read_mostly; /* Taps */
183 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
186 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
188 * Writers must hold the rtnl semaphore while they loop through the
189 * dev_base_head list, and hold dev_base_lock for writing when they do the
190 * actual updates. This allows pure readers to access the list even
191 * while a writer is preparing to update it.
193 * To put it another way, dev_base_lock is held for writing only to
194 * protect against pure readers; the rtnl semaphore provides the
195 * protection against other writers.
197 * See, for example usages, register_netdevice() and
198 * unregister_netdevice(), which must be called with the rtnl
201 DEFINE_RWLOCK(dev_base_lock);
202 EXPORT_SYMBOL(dev_base_lock);
204 static inline void dev_base_seq_inc(struct net *net)
206 while (++net->dev_base_seq == 0);
209 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
211 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
213 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
216 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
218 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
221 static inline void rps_lock(struct softnet_data *sd)
224 spin_lock(&sd->input_pkt_queue.lock);
228 static inline void rps_unlock(struct softnet_data *sd)
231 spin_unlock(&sd->input_pkt_queue.lock);
235 /* Device list insertion */
236 static int list_netdevice(struct net_device *dev)
238 struct net *net = dev_net(dev);
242 write_lock_bh(&dev_base_lock);
243 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
244 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
245 hlist_add_head_rcu(&dev->index_hlist,
246 dev_index_hash(net, dev->ifindex));
247 write_unlock_bh(&dev_base_lock);
249 dev_base_seq_inc(net);
254 /* Device list removal
255 * caller must respect a RCU grace period before freeing/reusing dev
257 static void unlist_netdevice(struct net_device *dev)
261 /* Unlink dev from the device chain */
262 write_lock_bh(&dev_base_lock);
263 list_del_rcu(&dev->dev_list);
264 hlist_del_rcu(&dev->name_hlist);
265 hlist_del_rcu(&dev->index_hlist);
266 write_unlock_bh(&dev_base_lock);
268 dev_base_seq_inc(dev_net(dev));
275 static RAW_NOTIFIER_HEAD(netdev_chain);
278 * Device drivers call our routines to queue packets here. We empty the
279 * queue in the local softnet handler.
282 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
283 EXPORT_PER_CPU_SYMBOL(softnet_data);
285 #ifdef CONFIG_LOCKDEP
287 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
288 * according to dev->type
290 static const unsigned short netdev_lock_type[] =
291 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
292 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
293 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
294 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
295 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
296 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
297 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
298 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
299 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
300 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
301 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
302 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
303 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
304 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
305 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
307 static const char *const netdev_lock_name[] =
308 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
309 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
310 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
311 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
312 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
313 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
314 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
315 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
316 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
317 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
318 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
319 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
320 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
321 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
322 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
324 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
325 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
327 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
331 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
332 if (netdev_lock_type[i] == dev_type)
334 /* the last key is used by default */
335 return ARRAY_SIZE(netdev_lock_type) - 1;
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
343 i = netdev_lock_pos(dev_type);
344 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
345 netdev_lock_name[i]);
348 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
352 i = netdev_lock_pos(dev->type);
353 lockdep_set_class_and_name(&dev->addr_list_lock,
354 &netdev_addr_lock_key[i],
355 netdev_lock_name[i]);
358 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
359 unsigned short dev_type)
362 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
367 /*******************************************************************************
369 Protocol management and registration routines
371 *******************************************************************************/
374 * Add a protocol ID to the list. Now that the input handler is
375 * smarter we can dispense with all the messy stuff that used to be
378 * BEWARE!!! Protocol handlers, mangling input packets,
379 * MUST BE last in hash buckets and checking protocol handlers
380 * MUST start from promiscuous ptype_all chain in net_bh.
381 * It is true now, do not change it.
382 * Explanation follows: if protocol handler, mangling packet, will
383 * be the first on list, it is not able to sense, that packet
384 * is cloned and should be copied-on-write, so that it will
385 * change it and subsequent readers will get broken packet.
389 static inline struct list_head *ptype_head(const struct packet_type *pt)
391 if (pt->type == htons(ETH_P_ALL))
394 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
398 * dev_add_pack - add packet handler
399 * @pt: packet type declaration
401 * Add a protocol handler to the networking stack. The passed &packet_type
402 * is linked into kernel lists and may not be freed until it has been
403 * removed from the kernel lists.
405 * This call does not sleep therefore it can not
406 * guarantee all CPU's that are in middle of receiving packets
407 * will see the new packet type (until the next received packet).
410 void dev_add_pack(struct packet_type *pt)
412 struct list_head *head = ptype_head(pt);
414 spin_lock(&ptype_lock);
415 list_add_rcu(&pt->list, head);
416 spin_unlock(&ptype_lock);
418 EXPORT_SYMBOL(dev_add_pack);
421 * __dev_remove_pack - remove packet handler
422 * @pt: packet type declaration
424 * Remove a protocol handler that was previously added to the kernel
425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
426 * from the kernel lists and can be freed or reused once this function
429 * The packet type might still be in use by receivers
430 * and must not be freed until after all the CPU's have gone
431 * through a quiescent state.
433 void __dev_remove_pack(struct packet_type *pt)
435 struct list_head *head = ptype_head(pt);
436 struct packet_type *pt1;
438 spin_lock(&ptype_lock);
440 list_for_each_entry(pt1, head, list) {
442 list_del_rcu(&pt->list);
447 pr_warn("dev_remove_pack: %p not found\n", pt);
449 spin_unlock(&ptype_lock);
451 EXPORT_SYMBOL(__dev_remove_pack);
454 * dev_remove_pack - remove packet handler
455 * @pt: packet type declaration
457 * Remove a protocol handler that was previously added to the kernel
458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
459 * from the kernel lists and can be freed or reused once this function
462 * This call sleeps to guarantee that no CPU is looking at the packet
465 void dev_remove_pack(struct packet_type *pt)
467 __dev_remove_pack(pt);
471 EXPORT_SYMBOL(dev_remove_pack);
473 /******************************************************************************
475 Device Boot-time Settings Routines
477 *******************************************************************************/
479 /* Boot time configuration table */
480 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
483 * netdev_boot_setup_add - add new setup entry
484 * @name: name of the device
485 * @map: configured settings for the device
487 * Adds new setup entry to the dev_boot_setup list. The function
488 * returns 0 on error and 1 on success. This is a generic routine to
491 static int netdev_boot_setup_add(char *name, struct ifmap *map)
493 struct netdev_boot_setup *s;
497 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
498 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
499 memset(s[i].name, 0, sizeof(s[i].name));
500 strlcpy(s[i].name, name, IFNAMSIZ);
501 memcpy(&s[i].map, map, sizeof(s[i].map));
506 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
510 * netdev_boot_setup_check - check boot time settings
511 * @dev: the netdevice
513 * Check boot time settings for the device.
514 * The found settings are set for the device to be used
515 * later in the device probing.
516 * Returns 0 if no settings found, 1 if they are.
518 int netdev_boot_setup_check(struct net_device *dev)
520 struct netdev_boot_setup *s = dev_boot_setup;
523 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
524 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
525 !strcmp(dev->name, s[i].name)) {
526 dev->irq = s[i].map.irq;
527 dev->base_addr = s[i].map.base_addr;
528 dev->mem_start = s[i].map.mem_start;
529 dev->mem_end = s[i].map.mem_end;
535 EXPORT_SYMBOL(netdev_boot_setup_check);
539 * netdev_boot_base - get address from boot time settings
540 * @prefix: prefix for network device
541 * @unit: id for network device
543 * Check boot time settings for the base address of device.
544 * The found settings are set for the device to be used
545 * later in the device probing.
546 * Returns 0 if no settings found.
548 unsigned long netdev_boot_base(const char *prefix, int unit)
550 const struct netdev_boot_setup *s = dev_boot_setup;
554 sprintf(name, "%s%d", prefix, unit);
557 * If device already registered then return base of 1
558 * to indicate not to probe for this interface
560 if (__dev_get_by_name(&init_net, name))
563 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
564 if (!strcmp(name, s[i].name))
565 return s[i].map.base_addr;
570 * Saves at boot time configured settings for any netdevice.
572 int __init netdev_boot_setup(char *str)
577 str = get_options(str, ARRAY_SIZE(ints), ints);
582 memset(&map, 0, sizeof(map));
586 map.base_addr = ints[2];
588 map.mem_start = ints[3];
590 map.mem_end = ints[4];
592 /* Add new entry to the list */
593 return netdev_boot_setup_add(str, &map);
596 __setup("netdev=", netdev_boot_setup);
598 /*******************************************************************************
600 Device Interface Subroutines
602 *******************************************************************************/
605 * __dev_get_by_name - find a device by its name
606 * @net: the applicable net namespace
607 * @name: name to find
609 * Find an interface by name. Must be called under RTNL semaphore
610 * or @dev_base_lock. If the name is found a pointer to the device
611 * is returned. If the name is not found then %NULL is returned. The
612 * reference counters are not incremented so the caller must be
613 * careful with locks.
616 struct net_device *__dev_get_by_name(struct net *net, const char *name)
618 struct hlist_node *p;
619 struct net_device *dev;
620 struct hlist_head *head = dev_name_hash(net, name);
622 hlist_for_each_entry(dev, p, head, name_hlist)
623 if (!strncmp(dev->name, name, IFNAMSIZ))
628 EXPORT_SYMBOL(__dev_get_by_name);
631 * dev_get_by_name_rcu - find a device by its name
632 * @net: the applicable net namespace
633 * @name: name to find
635 * Find an interface by name.
636 * If the name is found a pointer to the device is returned.
637 * If the name is not found then %NULL is returned.
638 * The reference counters are not incremented so the caller must be
639 * careful with locks. The caller must hold RCU lock.
642 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
644 struct hlist_node *p;
645 struct net_device *dev;
646 struct hlist_head *head = dev_name_hash(net, name);
648 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
649 if (!strncmp(dev->name, name, IFNAMSIZ))
654 EXPORT_SYMBOL(dev_get_by_name_rcu);
657 * dev_get_by_name - find a device by its name
658 * @net: the applicable net namespace
659 * @name: name to find
661 * Find an interface by name. This can be called from any
662 * context and does its own locking. The returned handle has
663 * the usage count incremented and the caller must use dev_put() to
664 * release it when it is no longer needed. %NULL is returned if no
665 * matching device is found.
668 struct net_device *dev_get_by_name(struct net *net, const char *name)
670 struct net_device *dev;
673 dev = dev_get_by_name_rcu(net, name);
679 EXPORT_SYMBOL(dev_get_by_name);
682 * __dev_get_by_index - find a device by its ifindex
683 * @net: the applicable net namespace
684 * @ifindex: index of device
686 * Search for an interface by index. Returns %NULL if the device
687 * is not found or a pointer to the device. The device has not
688 * had its reference counter increased so the caller must be careful
689 * about locking. The caller must hold either the RTNL semaphore
693 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
695 struct hlist_node *p;
696 struct net_device *dev;
697 struct hlist_head *head = dev_index_hash(net, ifindex);
699 hlist_for_each_entry(dev, p, head, index_hlist)
700 if (dev->ifindex == ifindex)
705 EXPORT_SYMBOL(__dev_get_by_index);
708 * dev_get_by_index_rcu - find a device by its ifindex
709 * @net: the applicable net namespace
710 * @ifindex: index of device
712 * Search for an interface by index. Returns %NULL if the device
713 * is not found or a pointer to the device. The device has not
714 * had its reference counter increased so the caller must be careful
715 * about locking. The caller must hold RCU lock.
718 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
720 struct hlist_node *p;
721 struct net_device *dev;
722 struct hlist_head *head = dev_index_hash(net, ifindex);
724 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
725 if (dev->ifindex == ifindex)
730 EXPORT_SYMBOL(dev_get_by_index_rcu);
734 * dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns NULL if the device
739 * is not found or a pointer to the device. The device returned has
740 * had a reference added and the pointer is safe until the user calls
741 * dev_put to indicate they have finished with it.
744 struct net_device *dev_get_by_index(struct net *net, int ifindex)
746 struct net_device *dev;
749 dev = dev_get_by_index_rcu(net, ifindex);
755 EXPORT_SYMBOL(dev_get_by_index);
758 * dev_getbyhwaddr_rcu - find a device by its hardware address
759 * @net: the applicable net namespace
760 * @type: media type of device
761 * @ha: hardware address
763 * Search for an interface by MAC address. Returns NULL if the device
764 * is not found or a pointer to the device.
765 * The caller must hold RCU or RTNL.
766 * The returned device has not had its ref count increased
767 * and the caller must therefore be careful about locking
771 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
774 struct net_device *dev;
776 for_each_netdev_rcu(net, dev)
777 if (dev->type == type &&
778 !memcmp(dev->dev_addr, ha, dev->addr_len))
783 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
785 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
787 struct net_device *dev;
790 for_each_netdev(net, dev)
791 if (dev->type == type)
796 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
798 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
800 struct net_device *dev, *ret = NULL;
803 for_each_netdev_rcu(net, dev)
804 if (dev->type == type) {
812 EXPORT_SYMBOL(dev_getfirstbyhwtype);
815 * dev_get_by_flags_rcu - find any device with given flags
816 * @net: the applicable net namespace
817 * @if_flags: IFF_* values
818 * @mask: bitmask of bits in if_flags to check
820 * Search for any interface with the given flags. Returns NULL if a device
821 * is not found or a pointer to the device. Must be called inside
822 * rcu_read_lock(), and result refcount is unchanged.
825 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
828 struct net_device *dev, *ret;
831 for_each_netdev_rcu(net, dev) {
832 if (((dev->flags ^ if_flags) & mask) == 0) {
839 EXPORT_SYMBOL(dev_get_by_flags_rcu);
842 * dev_valid_name - check if name is okay for network device
845 * Network device names need to be valid file names to
846 * to allow sysfs to work. We also disallow any kind of
849 bool dev_valid_name(const char *name)
853 if (strlen(name) >= IFNAMSIZ)
855 if (!strcmp(name, ".") || !strcmp(name, ".."))
859 if (*name == '/' || isspace(*name))
865 EXPORT_SYMBOL(dev_valid_name);
868 * __dev_alloc_name - allocate a name for a device
869 * @net: network namespace to allocate the device name in
870 * @name: name format string
871 * @buf: scratch buffer and result name string
873 * Passed a format string - eg "lt%d" it will try and find a suitable
874 * id. It scans list of devices to build up a free map, then chooses
875 * the first empty slot. The caller must hold the dev_base or rtnl lock
876 * while allocating the name and adding the device in order to avoid
878 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
879 * Returns the number of the unit assigned or a negative errno code.
882 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
886 const int max_netdevices = 8*PAGE_SIZE;
887 unsigned long *inuse;
888 struct net_device *d;
890 p = strnchr(name, IFNAMSIZ-1, '%');
893 * Verify the string as this thing may have come from
894 * the user. There must be either one "%d" and no other "%"
897 if (p[1] != 'd' || strchr(p + 2, '%'))
900 /* Use one page as a bit array of possible slots */
901 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
905 for_each_netdev(net, d) {
906 if (!sscanf(d->name, name, &i))
908 if (i < 0 || i >= max_netdevices)
911 /* avoid cases where sscanf is not exact inverse of printf */
912 snprintf(buf, IFNAMSIZ, name, i);
913 if (!strncmp(buf, d->name, IFNAMSIZ))
917 i = find_first_zero_bit(inuse, max_netdevices);
918 free_page((unsigned long) inuse);
922 snprintf(buf, IFNAMSIZ, name, i);
923 if (!__dev_get_by_name(net, buf))
926 /* It is possible to run out of possible slots
927 * when the name is long and there isn't enough space left
928 * for the digits, or if all bits are used.
934 * dev_alloc_name - allocate a name for a device
936 * @name: name format string
938 * Passed a format string - eg "lt%d" it will try and find a suitable
939 * id. It scans list of devices to build up a free map, then chooses
940 * the first empty slot. The caller must hold the dev_base or rtnl lock
941 * while allocating the name and adding the device in order to avoid
943 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
944 * Returns the number of the unit assigned or a negative errno code.
947 int dev_alloc_name(struct net_device *dev, const char *name)
953 BUG_ON(!dev_net(dev));
955 ret = __dev_alloc_name(net, name, buf);
957 strlcpy(dev->name, buf, IFNAMSIZ);
960 EXPORT_SYMBOL(dev_alloc_name);
962 static int dev_get_valid_name(struct net_device *dev, const char *name)
966 BUG_ON(!dev_net(dev));
969 if (!dev_valid_name(name))
972 if (strchr(name, '%'))
973 return dev_alloc_name(dev, name);
974 else if (__dev_get_by_name(net, name))
976 else if (dev->name != name)
977 strlcpy(dev->name, name, IFNAMSIZ);
983 * dev_change_name - change name of a device
985 * @newname: name (or format string) must be at least IFNAMSIZ
987 * Change name of a device, can pass format strings "eth%d".
990 int dev_change_name(struct net_device *dev, const char *newname)
992 char oldname[IFNAMSIZ];
998 BUG_ON(!dev_net(dev));
1001 if (dev->flags & IFF_UP)
1004 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1007 memcpy(oldname, dev->name, IFNAMSIZ);
1009 err = dev_get_valid_name(dev, newname);
1014 ret = device_rename(&dev->dev, dev->name);
1016 memcpy(dev->name, oldname, IFNAMSIZ);
1020 write_lock_bh(&dev_base_lock);
1021 hlist_del_rcu(&dev->name_hlist);
1022 write_unlock_bh(&dev_base_lock);
1026 write_lock_bh(&dev_base_lock);
1027 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1028 write_unlock_bh(&dev_base_lock);
1030 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1031 ret = notifier_to_errno(ret);
1034 /* err >= 0 after dev_alloc_name() or stores the first errno */
1037 memcpy(dev->name, oldname, IFNAMSIZ);
1040 pr_err("%s: name change rollback failed: %d\n",
1049 * dev_set_alias - change ifalias of a device
1051 * @alias: name up to IFALIASZ
1052 * @len: limit of bytes to copy from info
1054 * Set ifalias for a device,
1056 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1060 if (len >= IFALIASZ)
1065 kfree(dev->ifalias);
1066 dev->ifalias = NULL;
1071 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1075 strlcpy(dev->ifalias, alias, len+1);
1081 * netdev_features_change - device changes features
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed features.
1086 void netdev_features_change(struct net_device *dev)
1088 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1090 EXPORT_SYMBOL(netdev_features_change);
1093 * netdev_state_change - device changes state
1094 * @dev: device to cause notification
1096 * Called to indicate a device has changed state. This function calls
1097 * the notifier chains for netdev_chain and sends a NEWLINK message
1098 * to the routing socket.
1100 void netdev_state_change(struct net_device *dev)
1102 if (dev->flags & IFF_UP) {
1103 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1104 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1107 EXPORT_SYMBOL(netdev_state_change);
1109 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1111 return call_netdevice_notifiers(event, dev);
1113 EXPORT_SYMBOL(netdev_bonding_change);
1116 * dev_load - load a network module
1117 * @net: the applicable net namespace
1118 * @name: name of interface
1120 * If a network interface is not present and the process has suitable
1121 * privileges this function loads the module. If module loading is not
1122 * available in this kernel then it becomes a nop.
1125 void dev_load(struct net *net, const char *name)
1127 struct net_device *dev;
1131 dev = dev_get_by_name_rcu(net, name);
1135 if (no_module && capable(CAP_NET_ADMIN))
1136 no_module = request_module("netdev-%s", name);
1137 if (no_module && capable(CAP_SYS_MODULE)) {
1138 if (!request_module("%s", name))
1139 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1143 EXPORT_SYMBOL(dev_load);
1145 static int __dev_open(struct net_device *dev)
1147 const struct net_device_ops *ops = dev->netdev_ops;
1152 if (!netif_device_present(dev))
1155 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1156 ret = notifier_to_errno(ret);
1160 set_bit(__LINK_STATE_START, &dev->state);
1162 if (ops->ndo_validate_addr)
1163 ret = ops->ndo_validate_addr(dev);
1165 if (!ret && ops->ndo_open)
1166 ret = ops->ndo_open(dev);
1169 clear_bit(__LINK_STATE_START, &dev->state);
1171 dev->flags |= IFF_UP;
1172 net_dmaengine_get();
1173 dev_set_rx_mode(dev);
1181 * dev_open - prepare an interface for use.
1182 * @dev: device to open
1184 * Takes a device from down to up state. The device's private open
1185 * function is invoked and then the multicast lists are loaded. Finally
1186 * the device is moved into the up state and a %NETDEV_UP message is
1187 * sent to the netdev notifier chain.
1189 * Calling this function on an active interface is a nop. On a failure
1190 * a negative errno code is returned.
1192 int dev_open(struct net_device *dev)
1196 if (dev->flags & IFF_UP)
1199 ret = __dev_open(dev);
1203 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1204 call_netdevice_notifiers(NETDEV_UP, dev);
1208 EXPORT_SYMBOL(dev_open);
1210 static int __dev_close_many(struct list_head *head)
1212 struct net_device *dev;
1217 list_for_each_entry(dev, head, unreg_list) {
1218 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1220 clear_bit(__LINK_STATE_START, &dev->state);
1222 /* Synchronize to scheduled poll. We cannot touch poll list, it
1223 * can be even on different cpu. So just clear netif_running().
1225 * dev->stop() will invoke napi_disable() on all of it's
1226 * napi_struct instances on this device.
1228 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1231 dev_deactivate_many(head);
1233 list_for_each_entry(dev, head, unreg_list) {
1234 const struct net_device_ops *ops = dev->netdev_ops;
1237 * Call the device specific close. This cannot fail.
1238 * Only if device is UP
1240 * We allow it to be called even after a DETACH hot-plug
1246 dev->flags &= ~IFF_UP;
1247 net_dmaengine_put();
1253 static int __dev_close(struct net_device *dev)
1258 list_add(&dev->unreg_list, &single);
1259 retval = __dev_close_many(&single);
1264 static int dev_close_many(struct list_head *head)
1266 struct net_device *dev, *tmp;
1267 LIST_HEAD(tmp_list);
1269 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1270 if (!(dev->flags & IFF_UP))
1271 list_move(&dev->unreg_list, &tmp_list);
1273 __dev_close_many(head);
1275 list_for_each_entry(dev, head, unreg_list) {
1276 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1277 call_netdevice_notifiers(NETDEV_DOWN, dev);
1280 /* rollback_registered_many needs the complete original list */
1281 list_splice(&tmp_list, head);
1286 * dev_close - shutdown an interface.
1287 * @dev: device to shutdown
1289 * This function moves an active device into down state. A
1290 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1291 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1294 int dev_close(struct net_device *dev)
1296 if (dev->flags & IFF_UP) {
1299 list_add(&dev->unreg_list, &single);
1300 dev_close_many(&single);
1305 EXPORT_SYMBOL(dev_close);
1309 * dev_disable_lro - disable Large Receive Offload on a device
1312 * Disable Large Receive Offload (LRO) on a net device. Must be
1313 * called under RTNL. This is needed if received packets may be
1314 * forwarded to another interface.
1316 void dev_disable_lro(struct net_device *dev)
1319 * If we're trying to disable lro on a vlan device
1320 * use the underlying physical device instead
1322 if (is_vlan_dev(dev))
1323 dev = vlan_dev_real_dev(dev);
1325 dev->wanted_features &= ~NETIF_F_LRO;
1326 netdev_update_features(dev);
1328 if (unlikely(dev->features & NETIF_F_LRO))
1329 netdev_WARN(dev, "failed to disable LRO!\n");
1331 EXPORT_SYMBOL(dev_disable_lro);
1334 static int dev_boot_phase = 1;
1337 * register_netdevice_notifier - register a network notifier block
1340 * Register a notifier to be called when network device events occur.
1341 * The notifier passed is linked into the kernel structures and must
1342 * not be reused until it has been unregistered. A negative errno code
1343 * is returned on a failure.
1345 * When registered all registration and up events are replayed
1346 * to the new notifier to allow device to have a race free
1347 * view of the network device list.
1350 int register_netdevice_notifier(struct notifier_block *nb)
1352 struct net_device *dev;
1353 struct net_device *last;
1358 err = raw_notifier_chain_register(&netdev_chain, nb);
1364 for_each_netdev(net, dev) {
1365 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1366 err = notifier_to_errno(err);
1370 if (!(dev->flags & IFF_UP))
1373 nb->notifier_call(nb, NETDEV_UP, dev);
1384 for_each_netdev(net, dev) {
1388 if (dev->flags & IFF_UP) {
1389 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1390 nb->notifier_call(nb, NETDEV_DOWN, dev);
1392 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1393 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1398 raw_notifier_chain_unregister(&netdev_chain, nb);
1401 EXPORT_SYMBOL(register_netdevice_notifier);
1404 * unregister_netdevice_notifier - unregister a network notifier block
1407 * Unregister a notifier previously registered by
1408 * register_netdevice_notifier(). The notifier is unlinked into the
1409 * kernel structures and may then be reused. A negative errno code
1410 * is returned on a failure.
1412 * After unregistering unregister and down device events are synthesized
1413 * for all devices on the device list to the removed notifier to remove
1414 * the need for special case cleanup code.
1417 int unregister_netdevice_notifier(struct notifier_block *nb)
1419 struct net_device *dev;
1424 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1429 for_each_netdev(net, dev) {
1430 if (dev->flags & IFF_UP) {
1431 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1432 nb->notifier_call(nb, NETDEV_DOWN, dev);
1434 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1435 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1442 EXPORT_SYMBOL(unregister_netdevice_notifier);
1445 * call_netdevice_notifiers - call all network notifier blocks
1446 * @val: value passed unmodified to notifier function
1447 * @dev: net_device pointer passed unmodified to notifier function
1449 * Call all network notifier blocks. Parameters and return value
1450 * are as for raw_notifier_call_chain().
1453 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1456 return raw_notifier_call_chain(&netdev_chain, val, dev);
1458 EXPORT_SYMBOL(call_netdevice_notifiers);
1460 static struct static_key netstamp_needed __read_mostly;
1461 #ifdef HAVE_JUMP_LABEL
1462 /* We are not allowed to call static_key_slow_dec() from irq context
1463 * If net_disable_timestamp() is called from irq context, defer the
1464 * static_key_slow_dec() calls.
1466 static atomic_t netstamp_needed_deferred;
1469 void net_enable_timestamp(void)
1471 #ifdef HAVE_JUMP_LABEL
1472 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1476 static_key_slow_dec(&netstamp_needed);
1480 WARN_ON(in_interrupt());
1481 static_key_slow_inc(&netstamp_needed);
1483 EXPORT_SYMBOL(net_enable_timestamp);
1485 void net_disable_timestamp(void)
1487 #ifdef HAVE_JUMP_LABEL
1488 if (in_interrupt()) {
1489 atomic_inc(&netstamp_needed_deferred);
1493 static_key_slow_dec(&netstamp_needed);
1495 EXPORT_SYMBOL(net_disable_timestamp);
1497 static inline void net_timestamp_set(struct sk_buff *skb)
1499 skb->tstamp.tv64 = 0;
1500 if (static_key_false(&netstamp_needed))
1501 __net_timestamp(skb);
1504 #define net_timestamp_check(COND, SKB) \
1505 if (static_key_false(&netstamp_needed)) { \
1506 if ((COND) && !(SKB)->tstamp.tv64) \
1507 __net_timestamp(SKB); \
1510 static int net_hwtstamp_validate(struct ifreq *ifr)
1512 struct hwtstamp_config cfg;
1513 enum hwtstamp_tx_types tx_type;
1514 enum hwtstamp_rx_filters rx_filter;
1515 int tx_type_valid = 0;
1516 int rx_filter_valid = 0;
1518 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1521 if (cfg.flags) /* reserved for future extensions */
1524 tx_type = cfg.tx_type;
1525 rx_filter = cfg.rx_filter;
1528 case HWTSTAMP_TX_OFF:
1529 case HWTSTAMP_TX_ON:
1530 case HWTSTAMP_TX_ONESTEP_SYNC:
1535 switch (rx_filter) {
1536 case HWTSTAMP_FILTER_NONE:
1537 case HWTSTAMP_FILTER_ALL:
1538 case HWTSTAMP_FILTER_SOME:
1539 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1540 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1542 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1543 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1545 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1546 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1548 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1549 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1550 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1551 rx_filter_valid = 1;
1555 if (!tx_type_valid || !rx_filter_valid)
1561 static inline bool is_skb_forwardable(struct net_device *dev,
1562 struct sk_buff *skb)
1566 if (!(dev->flags & IFF_UP))
1569 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1570 if (skb->len <= len)
1573 /* if TSO is enabled, we don't care about the length as the packet
1574 * could be forwarded without being segmented before
1576 if (skb_is_gso(skb))
1583 * dev_forward_skb - loopback an skb to another netif
1585 * @dev: destination network device
1586 * @skb: buffer to forward
1589 * NET_RX_SUCCESS (no congestion)
1590 * NET_RX_DROP (packet was dropped, but freed)
1592 * dev_forward_skb can be used for injecting an skb from the
1593 * start_xmit function of one device into the receive queue
1594 * of another device.
1596 * The receiving device may be in another namespace, so
1597 * we have to clear all information in the skb that could
1598 * impact namespace isolation.
1600 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1602 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1603 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1604 atomic_long_inc(&dev->rx_dropped);
1613 if (unlikely(!is_skb_forwardable(dev, skb))) {
1614 atomic_long_inc(&dev->rx_dropped);
1621 skb->tstamp.tv64 = 0;
1622 skb->pkt_type = PACKET_HOST;
1623 skb->protocol = eth_type_trans(skb, dev);
1627 return netif_rx(skb);
1629 EXPORT_SYMBOL_GPL(dev_forward_skb);
1631 static inline int deliver_skb(struct sk_buff *skb,
1632 struct packet_type *pt_prev,
1633 struct net_device *orig_dev)
1635 atomic_inc(&skb->users);
1636 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1640 * Support routine. Sends outgoing frames to any network
1641 * taps currently in use.
1644 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1646 struct packet_type *ptype;
1647 struct sk_buff *skb2 = NULL;
1648 struct packet_type *pt_prev = NULL;
1651 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1652 /* Never send packets back to the socket
1653 * they originated from - MvS (miquels@drinkel.ow.org)
1655 if ((ptype->dev == dev || !ptype->dev) &&
1656 (ptype->af_packet_priv == NULL ||
1657 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1659 deliver_skb(skb2, pt_prev, skb->dev);
1664 skb2 = skb_clone(skb, GFP_ATOMIC);
1668 net_timestamp_set(skb2);
1670 /* skb->nh should be correctly
1671 set by sender, so that the second statement is
1672 just protection against buggy protocols.
1674 skb_reset_mac_header(skb2);
1676 if (skb_network_header(skb2) < skb2->data ||
1677 skb2->network_header > skb2->tail) {
1678 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1679 ntohs(skb2->protocol),
1681 skb_reset_network_header(skb2);
1684 skb2->transport_header = skb2->network_header;
1685 skb2->pkt_type = PACKET_OUTGOING;
1690 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1694 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1695 * @dev: Network device
1696 * @txq: number of queues available
1698 * If real_num_tx_queues is changed the tc mappings may no longer be
1699 * valid. To resolve this verify the tc mapping remains valid and if
1700 * not NULL the mapping. With no priorities mapping to this
1701 * offset/count pair it will no longer be used. In the worst case TC0
1702 * is invalid nothing can be done so disable priority mappings. If is
1703 * expected that drivers will fix this mapping if they can before
1704 * calling netif_set_real_num_tx_queues.
1706 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1709 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1711 /* If TC0 is invalidated disable TC mapping */
1712 if (tc->offset + tc->count > txq) {
1713 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1718 /* Invalidated prio to tc mappings set to TC0 */
1719 for (i = 1; i < TC_BITMASK + 1; i++) {
1720 int q = netdev_get_prio_tc_map(dev, i);
1722 tc = &dev->tc_to_txq[q];
1723 if (tc->offset + tc->count > txq) {
1724 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1726 netdev_set_prio_tc_map(dev, i, 0);
1732 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1733 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1735 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1739 if (txq < 1 || txq > dev->num_tx_queues)
1742 if (dev->reg_state == NETREG_REGISTERED ||
1743 dev->reg_state == NETREG_UNREGISTERING) {
1746 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1752 netif_setup_tc(dev, txq);
1754 if (txq < dev->real_num_tx_queues)
1755 qdisc_reset_all_tx_gt(dev, txq);
1758 dev->real_num_tx_queues = txq;
1761 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1765 * netif_set_real_num_rx_queues - set actual number of RX queues used
1766 * @dev: Network device
1767 * @rxq: Actual number of RX queues
1769 * This must be called either with the rtnl_lock held or before
1770 * registration of the net device. Returns 0 on success, or a
1771 * negative error code. If called before registration, it always
1774 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1778 if (rxq < 1 || rxq > dev->num_rx_queues)
1781 if (dev->reg_state == NETREG_REGISTERED) {
1784 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1790 dev->real_num_rx_queues = rxq;
1793 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1796 static inline void __netif_reschedule(struct Qdisc *q)
1798 struct softnet_data *sd;
1799 unsigned long flags;
1801 local_irq_save(flags);
1802 sd = &__get_cpu_var(softnet_data);
1803 q->next_sched = NULL;
1804 *sd->output_queue_tailp = q;
1805 sd->output_queue_tailp = &q->next_sched;
1806 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1807 local_irq_restore(flags);
1810 void __netif_schedule(struct Qdisc *q)
1812 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1813 __netif_reschedule(q);
1815 EXPORT_SYMBOL(__netif_schedule);
1817 void dev_kfree_skb_irq(struct sk_buff *skb)
1819 if (atomic_dec_and_test(&skb->users)) {
1820 struct softnet_data *sd;
1821 unsigned long flags;
1823 local_irq_save(flags);
1824 sd = &__get_cpu_var(softnet_data);
1825 skb->next = sd->completion_queue;
1826 sd->completion_queue = skb;
1827 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1828 local_irq_restore(flags);
1831 EXPORT_SYMBOL(dev_kfree_skb_irq);
1833 void dev_kfree_skb_any(struct sk_buff *skb)
1835 if (in_irq() || irqs_disabled())
1836 dev_kfree_skb_irq(skb);
1840 EXPORT_SYMBOL(dev_kfree_skb_any);
1844 * netif_device_detach - mark device as removed
1845 * @dev: network device
1847 * Mark device as removed from system and therefore no longer available.
1849 void netif_device_detach(struct net_device *dev)
1851 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1852 netif_running(dev)) {
1853 netif_tx_stop_all_queues(dev);
1856 EXPORT_SYMBOL(netif_device_detach);
1859 * netif_device_attach - mark device as attached
1860 * @dev: network device
1862 * Mark device as attached from system and restart if needed.
1864 void netif_device_attach(struct net_device *dev)
1866 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1867 netif_running(dev)) {
1868 netif_tx_wake_all_queues(dev);
1869 __netdev_watchdog_up(dev);
1872 EXPORT_SYMBOL(netif_device_attach);
1874 static void skb_warn_bad_offload(const struct sk_buff *skb)
1876 static const netdev_features_t null_features = 0;
1877 struct net_device *dev = skb->dev;
1878 const char *driver = "";
1880 if (dev && dev->dev.parent)
1881 driver = dev_driver_string(dev->dev.parent);
1883 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1884 "gso_type=%d ip_summed=%d\n",
1885 driver, dev ? &dev->features : &null_features,
1886 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1887 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1888 skb_shinfo(skb)->gso_type, skb->ip_summed);
1892 * Invalidate hardware checksum when packet is to be mangled, and
1893 * complete checksum manually on outgoing path.
1895 int skb_checksum_help(struct sk_buff *skb)
1898 int ret = 0, offset;
1900 if (skb->ip_summed == CHECKSUM_COMPLETE)
1901 goto out_set_summed;
1903 if (unlikely(skb_shinfo(skb)->gso_size)) {
1904 skb_warn_bad_offload(skb);
1908 offset = skb_checksum_start_offset(skb);
1909 BUG_ON(offset >= skb_headlen(skb));
1910 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1912 offset += skb->csum_offset;
1913 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1915 if (skb_cloned(skb) &&
1916 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1917 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1922 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1924 skb->ip_summed = CHECKSUM_NONE;
1928 EXPORT_SYMBOL(skb_checksum_help);
1931 * skb_gso_segment - Perform segmentation on skb.
1932 * @skb: buffer to segment
1933 * @features: features for the output path (see dev->features)
1935 * This function segments the given skb and returns a list of segments.
1937 * It may return NULL if the skb requires no segmentation. This is
1938 * only possible when GSO is used for verifying header integrity.
1940 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1941 netdev_features_t features)
1943 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1944 struct packet_type *ptype;
1945 __be16 type = skb->protocol;
1946 int vlan_depth = ETH_HLEN;
1949 while (type == htons(ETH_P_8021Q)) {
1950 struct vlan_hdr *vh;
1952 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1953 return ERR_PTR(-EINVAL);
1955 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1956 type = vh->h_vlan_encapsulated_proto;
1957 vlan_depth += VLAN_HLEN;
1960 skb_reset_mac_header(skb);
1961 skb->mac_len = skb->network_header - skb->mac_header;
1962 __skb_pull(skb, skb->mac_len);
1964 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1965 skb_warn_bad_offload(skb);
1967 if (skb_header_cloned(skb) &&
1968 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1969 return ERR_PTR(err);
1973 list_for_each_entry_rcu(ptype,
1974 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1975 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1976 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1977 err = ptype->gso_send_check(skb);
1978 segs = ERR_PTR(err);
1979 if (err || skb_gso_ok(skb, features))
1981 __skb_push(skb, (skb->data -
1982 skb_network_header(skb)));
1984 segs = ptype->gso_segment(skb, features);
1990 __skb_push(skb, skb->data - skb_mac_header(skb));
1994 EXPORT_SYMBOL(skb_gso_segment);
1996 /* Take action when hardware reception checksum errors are detected. */
1998 void netdev_rx_csum_fault(struct net_device *dev)
2000 if (net_ratelimit()) {
2001 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2005 EXPORT_SYMBOL(netdev_rx_csum_fault);
2008 /* Actually, we should eliminate this check as soon as we know, that:
2009 * 1. IOMMU is present and allows to map all the memory.
2010 * 2. No high memory really exists on this machine.
2013 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2015 #ifdef CONFIG_HIGHMEM
2017 if (!(dev->features & NETIF_F_HIGHDMA)) {
2018 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2019 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2020 if (PageHighMem(skb_frag_page(frag)))
2025 if (PCI_DMA_BUS_IS_PHYS) {
2026 struct device *pdev = dev->dev.parent;
2030 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2031 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2032 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2033 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2042 void (*destructor)(struct sk_buff *skb);
2045 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2047 static void dev_gso_skb_destructor(struct sk_buff *skb)
2049 struct dev_gso_cb *cb;
2052 struct sk_buff *nskb = skb->next;
2054 skb->next = nskb->next;
2057 } while (skb->next);
2059 cb = DEV_GSO_CB(skb);
2061 cb->destructor(skb);
2065 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2066 * @skb: buffer to segment
2067 * @features: device features as applicable to this skb
2069 * This function segments the given skb and stores the list of segments
2072 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2074 struct sk_buff *segs;
2076 segs = skb_gso_segment(skb, features);
2078 /* Verifying header integrity only. */
2083 return PTR_ERR(segs);
2086 DEV_GSO_CB(skb)->destructor = skb->destructor;
2087 skb->destructor = dev_gso_skb_destructor;
2092 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2094 return ((features & NETIF_F_GEN_CSUM) ||
2095 ((features & NETIF_F_V4_CSUM) &&
2096 protocol == htons(ETH_P_IP)) ||
2097 ((features & NETIF_F_V6_CSUM) &&
2098 protocol == htons(ETH_P_IPV6)) ||
2099 ((features & NETIF_F_FCOE_CRC) &&
2100 protocol == htons(ETH_P_FCOE)));
2103 static netdev_features_t harmonize_features(struct sk_buff *skb,
2104 __be16 protocol, netdev_features_t features)
2106 if (!can_checksum_protocol(features, protocol)) {
2107 features &= ~NETIF_F_ALL_CSUM;
2108 features &= ~NETIF_F_SG;
2109 } else if (illegal_highdma(skb->dev, skb)) {
2110 features &= ~NETIF_F_SG;
2116 netdev_features_t netif_skb_features(struct sk_buff *skb)
2118 __be16 protocol = skb->protocol;
2119 netdev_features_t features = skb->dev->features;
2121 if (protocol == htons(ETH_P_8021Q)) {
2122 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2123 protocol = veh->h_vlan_encapsulated_proto;
2124 } else if (!vlan_tx_tag_present(skb)) {
2125 return harmonize_features(skb, protocol, features);
2128 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2130 if (protocol != htons(ETH_P_8021Q)) {
2131 return harmonize_features(skb, protocol, features);
2133 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2134 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2135 return harmonize_features(skb, protocol, features);
2138 EXPORT_SYMBOL(netif_skb_features);
2141 * Returns true if either:
2142 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2143 * 2. skb is fragmented and the device does not support SG, or if
2144 * at least one of fragments is in highmem and device does not
2145 * support DMA from it.
2147 static inline int skb_needs_linearize(struct sk_buff *skb,
2150 return skb_is_nonlinear(skb) &&
2151 ((skb_has_frag_list(skb) &&
2152 !(features & NETIF_F_FRAGLIST)) ||
2153 (skb_shinfo(skb)->nr_frags &&
2154 !(features & NETIF_F_SG)));
2157 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2158 struct netdev_queue *txq)
2160 const struct net_device_ops *ops = dev->netdev_ops;
2161 int rc = NETDEV_TX_OK;
2162 unsigned int skb_len;
2164 if (likely(!skb->next)) {
2165 netdev_features_t features;
2168 * If device doesn't need skb->dst, release it right now while
2169 * its hot in this cpu cache
2171 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2174 if (!list_empty(&ptype_all))
2175 dev_queue_xmit_nit(skb, dev);
2177 features = netif_skb_features(skb);
2179 if (vlan_tx_tag_present(skb) &&
2180 !(features & NETIF_F_HW_VLAN_TX)) {
2181 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2188 if (netif_needs_gso(skb, features)) {
2189 if (unlikely(dev_gso_segment(skb, features)))
2194 if (skb_needs_linearize(skb, features) &&
2195 __skb_linearize(skb))
2198 /* If packet is not checksummed and device does not
2199 * support checksumming for this protocol, complete
2200 * checksumming here.
2202 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2203 skb_set_transport_header(skb,
2204 skb_checksum_start_offset(skb));
2205 if (!(features & NETIF_F_ALL_CSUM) &&
2206 skb_checksum_help(skb))
2212 rc = ops->ndo_start_xmit(skb, dev);
2213 trace_net_dev_xmit(skb, rc, dev, skb_len);
2214 if (rc == NETDEV_TX_OK)
2215 txq_trans_update(txq);
2221 struct sk_buff *nskb = skb->next;
2223 skb->next = nskb->next;
2227 * If device doesn't need nskb->dst, release it right now while
2228 * its hot in this cpu cache
2230 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2233 skb_len = nskb->len;
2234 rc = ops->ndo_start_xmit(nskb, dev);
2235 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2236 if (unlikely(rc != NETDEV_TX_OK)) {
2237 if (rc & ~NETDEV_TX_MASK)
2238 goto out_kfree_gso_skb;
2239 nskb->next = skb->next;
2243 txq_trans_update(txq);
2244 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2245 return NETDEV_TX_BUSY;
2246 } while (skb->next);
2249 if (likely(skb->next == NULL))
2250 skb->destructor = DEV_GSO_CB(skb)->destructor;
2257 static u32 hashrnd __read_mostly;
2260 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2261 * to be used as a distribution range.
2263 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2264 unsigned int num_tx_queues)
2268 u16 qcount = num_tx_queues;
2270 if (skb_rx_queue_recorded(skb)) {
2271 hash = skb_get_rx_queue(skb);
2272 while (unlikely(hash >= num_tx_queues))
2273 hash -= num_tx_queues;
2278 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2279 qoffset = dev->tc_to_txq[tc].offset;
2280 qcount = dev->tc_to_txq[tc].count;
2283 if (skb->sk && skb->sk->sk_hash)
2284 hash = skb->sk->sk_hash;
2286 hash = (__force u16) skb->protocol;
2287 hash = jhash_1word(hash, hashrnd);
2289 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2291 EXPORT_SYMBOL(__skb_tx_hash);
2293 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2295 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2296 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2297 dev->name, queue_index,
2298 dev->real_num_tx_queues);
2304 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2307 struct xps_dev_maps *dev_maps;
2308 struct xps_map *map;
2309 int queue_index = -1;
2312 dev_maps = rcu_dereference(dev->xps_maps);
2314 map = rcu_dereference(
2315 dev_maps->cpu_map[raw_smp_processor_id()]);
2318 queue_index = map->queues[0];
2321 if (skb->sk && skb->sk->sk_hash)
2322 hash = skb->sk->sk_hash;
2324 hash = (__force u16) skb->protocol ^
2326 hash = jhash_1word(hash, hashrnd);
2327 queue_index = map->queues[
2328 ((u64)hash * map->len) >> 32];
2330 if (unlikely(queue_index >= dev->real_num_tx_queues))
2342 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2343 struct sk_buff *skb)
2346 const struct net_device_ops *ops = dev->netdev_ops;
2348 if (dev->real_num_tx_queues == 1)
2350 else if (ops->ndo_select_queue) {
2351 queue_index = ops->ndo_select_queue(dev, skb);
2352 queue_index = dev_cap_txqueue(dev, queue_index);
2354 struct sock *sk = skb->sk;
2355 queue_index = sk_tx_queue_get(sk);
2357 if (queue_index < 0 || skb->ooo_okay ||
2358 queue_index >= dev->real_num_tx_queues) {
2359 int old_index = queue_index;
2361 queue_index = get_xps_queue(dev, skb);
2362 if (queue_index < 0)
2363 queue_index = skb_tx_hash(dev, skb);
2365 if (queue_index != old_index && sk) {
2366 struct dst_entry *dst =
2367 rcu_dereference_check(sk->sk_dst_cache, 1);
2369 if (dst && skb_dst(skb) == dst)
2370 sk_tx_queue_set(sk, queue_index);
2375 skb_set_queue_mapping(skb, queue_index);
2376 return netdev_get_tx_queue(dev, queue_index);
2379 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2380 struct net_device *dev,
2381 struct netdev_queue *txq)
2383 spinlock_t *root_lock = qdisc_lock(q);
2387 qdisc_skb_cb(skb)->pkt_len = skb->len;
2388 qdisc_calculate_pkt_len(skb, q);
2390 * Heuristic to force contended enqueues to serialize on a
2391 * separate lock before trying to get qdisc main lock.
2392 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2393 * and dequeue packets faster.
2395 contended = qdisc_is_running(q);
2396 if (unlikely(contended))
2397 spin_lock(&q->busylock);
2399 spin_lock(root_lock);
2400 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2403 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2404 qdisc_run_begin(q)) {
2406 * This is a work-conserving queue; there are no old skbs
2407 * waiting to be sent out; and the qdisc is not running -
2408 * xmit the skb directly.
2410 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2413 qdisc_bstats_update(q, skb);
2415 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2416 if (unlikely(contended)) {
2417 spin_unlock(&q->busylock);
2424 rc = NET_XMIT_SUCCESS;
2427 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2428 if (qdisc_run_begin(q)) {
2429 if (unlikely(contended)) {
2430 spin_unlock(&q->busylock);
2436 spin_unlock(root_lock);
2437 if (unlikely(contended))
2438 spin_unlock(&q->busylock);
2442 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2443 static void skb_update_prio(struct sk_buff *skb)
2445 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2447 if ((!skb->priority) && (skb->sk) && map)
2448 skb->priority = map->priomap[skb->sk->sk_cgrp_prioidx];
2451 #define skb_update_prio(skb)
2454 static DEFINE_PER_CPU(int, xmit_recursion);
2455 #define RECURSION_LIMIT 10
2458 * dev_queue_xmit - transmit a buffer
2459 * @skb: buffer to transmit
2461 * Queue a buffer for transmission to a network device. The caller must
2462 * have set the device and priority and built the buffer before calling
2463 * this function. The function can be called from an interrupt.
2465 * A negative errno code is returned on a failure. A success does not
2466 * guarantee the frame will be transmitted as it may be dropped due
2467 * to congestion or traffic shaping.
2469 * -----------------------------------------------------------------------------------
2470 * I notice this method can also return errors from the queue disciplines,
2471 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2474 * Regardless of the return value, the skb is consumed, so it is currently
2475 * difficult to retry a send to this method. (You can bump the ref count
2476 * before sending to hold a reference for retry if you are careful.)
2478 * When calling this method, interrupts MUST be enabled. This is because
2479 * the BH enable code must have IRQs enabled so that it will not deadlock.
2482 int dev_queue_xmit(struct sk_buff *skb)
2484 struct net_device *dev = skb->dev;
2485 struct netdev_queue *txq;
2489 /* Disable soft irqs for various locks below. Also
2490 * stops preemption for RCU.
2494 skb_update_prio(skb);
2496 txq = dev_pick_tx(dev, skb);
2497 q = rcu_dereference_bh(txq->qdisc);
2499 #ifdef CONFIG_NET_CLS_ACT
2500 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2502 trace_net_dev_queue(skb);
2504 rc = __dev_xmit_skb(skb, q, dev, txq);
2508 /* The device has no queue. Common case for software devices:
2509 loopback, all the sorts of tunnels...
2511 Really, it is unlikely that netif_tx_lock protection is necessary
2512 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2514 However, it is possible, that they rely on protection
2517 Check this and shot the lock. It is not prone from deadlocks.
2518 Either shot noqueue qdisc, it is even simpler 8)
2520 if (dev->flags & IFF_UP) {
2521 int cpu = smp_processor_id(); /* ok because BHs are off */
2523 if (txq->xmit_lock_owner != cpu) {
2525 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2526 goto recursion_alert;
2528 HARD_TX_LOCK(dev, txq, cpu);
2530 if (!netif_xmit_stopped(txq)) {
2531 __this_cpu_inc(xmit_recursion);
2532 rc = dev_hard_start_xmit(skb, dev, txq);
2533 __this_cpu_dec(xmit_recursion);
2534 if (dev_xmit_complete(rc)) {
2535 HARD_TX_UNLOCK(dev, txq);
2539 HARD_TX_UNLOCK(dev, txq);
2540 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2543 /* Recursion is detected! It is possible,
2547 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2553 rcu_read_unlock_bh();
2558 rcu_read_unlock_bh();
2561 EXPORT_SYMBOL(dev_queue_xmit);
2564 /*=======================================================================
2566 =======================================================================*/
2568 int netdev_max_backlog __read_mostly = 1000;
2569 int netdev_tstamp_prequeue __read_mostly = 1;
2570 int netdev_budget __read_mostly = 300;
2571 int weight_p __read_mostly = 64; /* old backlog weight */
2573 /* Called with irq disabled */
2574 static inline void ____napi_schedule(struct softnet_data *sd,
2575 struct napi_struct *napi)
2577 list_add_tail(&napi->poll_list, &sd->poll_list);
2578 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2582 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2583 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2584 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2585 * if hash is a canonical 4-tuple hash over transport ports.
2587 void __skb_get_rxhash(struct sk_buff *skb)
2589 struct flow_keys keys;
2592 if (!skb_flow_dissect(skb, &keys))
2596 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2597 swap(keys.port16[0], keys.port16[1]);
2601 /* get a consistent hash (same value on both flow directions) */
2602 if ((__force u32)keys.dst < (__force u32)keys.src)
2603 swap(keys.dst, keys.src);
2605 hash = jhash_3words((__force u32)keys.dst,
2606 (__force u32)keys.src,
2607 (__force u32)keys.ports, hashrnd);
2613 EXPORT_SYMBOL(__skb_get_rxhash);
2617 /* One global table that all flow-based protocols share. */
2618 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2619 EXPORT_SYMBOL(rps_sock_flow_table);
2621 struct static_key rps_needed __read_mostly;
2623 static struct rps_dev_flow *
2624 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2625 struct rps_dev_flow *rflow, u16 next_cpu)
2627 if (next_cpu != RPS_NO_CPU) {
2628 #ifdef CONFIG_RFS_ACCEL
2629 struct netdev_rx_queue *rxqueue;
2630 struct rps_dev_flow_table *flow_table;
2631 struct rps_dev_flow *old_rflow;
2636 /* Should we steer this flow to a different hardware queue? */
2637 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2638 !(dev->features & NETIF_F_NTUPLE))
2640 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2641 if (rxq_index == skb_get_rx_queue(skb))
2644 rxqueue = dev->_rx + rxq_index;
2645 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2648 flow_id = skb->rxhash & flow_table->mask;
2649 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2650 rxq_index, flow_id);
2654 rflow = &flow_table->flows[flow_id];
2656 if (old_rflow->filter == rflow->filter)
2657 old_rflow->filter = RPS_NO_FILTER;
2661 per_cpu(softnet_data, next_cpu).input_queue_head;
2664 rflow->cpu = next_cpu;
2669 * get_rps_cpu is called from netif_receive_skb and returns the target
2670 * CPU from the RPS map of the receiving queue for a given skb.
2671 * rcu_read_lock must be held on entry.
2673 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2674 struct rps_dev_flow **rflowp)
2676 struct netdev_rx_queue *rxqueue;
2677 struct rps_map *map;
2678 struct rps_dev_flow_table *flow_table;
2679 struct rps_sock_flow_table *sock_flow_table;
2683 if (skb_rx_queue_recorded(skb)) {
2684 u16 index = skb_get_rx_queue(skb);
2685 if (unlikely(index >= dev->real_num_rx_queues)) {
2686 WARN_ONCE(dev->real_num_rx_queues > 1,
2687 "%s received packet on queue %u, but number "
2688 "of RX queues is %u\n",
2689 dev->name, index, dev->real_num_rx_queues);
2692 rxqueue = dev->_rx + index;
2696 map = rcu_dereference(rxqueue->rps_map);
2698 if (map->len == 1 &&
2699 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2700 tcpu = map->cpus[0];
2701 if (cpu_online(tcpu))
2705 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2709 skb_reset_network_header(skb);
2710 if (!skb_get_rxhash(skb))
2713 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2714 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2715 if (flow_table && sock_flow_table) {
2717 struct rps_dev_flow *rflow;
2719 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2722 next_cpu = sock_flow_table->ents[skb->rxhash &
2723 sock_flow_table->mask];
2726 * If the desired CPU (where last recvmsg was done) is
2727 * different from current CPU (one in the rx-queue flow
2728 * table entry), switch if one of the following holds:
2729 * - Current CPU is unset (equal to RPS_NO_CPU).
2730 * - Current CPU is offline.
2731 * - The current CPU's queue tail has advanced beyond the
2732 * last packet that was enqueued using this table entry.
2733 * This guarantees that all previous packets for the flow
2734 * have been dequeued, thus preserving in order delivery.
2736 if (unlikely(tcpu != next_cpu) &&
2737 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2738 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2739 rflow->last_qtail)) >= 0))
2740 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2742 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2750 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2752 if (cpu_online(tcpu)) {
2762 #ifdef CONFIG_RFS_ACCEL
2765 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2766 * @dev: Device on which the filter was set
2767 * @rxq_index: RX queue index
2768 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2769 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2771 * Drivers that implement ndo_rx_flow_steer() should periodically call
2772 * this function for each installed filter and remove the filters for
2773 * which it returns %true.
2775 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2776 u32 flow_id, u16 filter_id)
2778 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2779 struct rps_dev_flow_table *flow_table;
2780 struct rps_dev_flow *rflow;
2785 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2786 if (flow_table && flow_id <= flow_table->mask) {
2787 rflow = &flow_table->flows[flow_id];
2788 cpu = ACCESS_ONCE(rflow->cpu);
2789 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2790 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2791 rflow->last_qtail) <
2792 (int)(10 * flow_table->mask)))
2798 EXPORT_SYMBOL(rps_may_expire_flow);
2800 #endif /* CONFIG_RFS_ACCEL */
2802 /* Called from hardirq (IPI) context */
2803 static void rps_trigger_softirq(void *data)
2805 struct softnet_data *sd = data;
2807 ____napi_schedule(sd, &sd->backlog);
2811 #endif /* CONFIG_RPS */
2814 * Check if this softnet_data structure is another cpu one
2815 * If yes, queue it to our IPI list and return 1
2818 static int rps_ipi_queued(struct softnet_data *sd)
2821 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2824 sd->rps_ipi_next = mysd->rps_ipi_list;
2825 mysd->rps_ipi_list = sd;
2827 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2830 #endif /* CONFIG_RPS */
2835 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2836 * queue (may be a remote CPU queue).
2838 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2839 unsigned int *qtail)
2841 struct softnet_data *sd;
2842 unsigned long flags;
2844 sd = &per_cpu(softnet_data, cpu);
2846 local_irq_save(flags);
2849 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2850 if (skb_queue_len(&sd->input_pkt_queue)) {
2852 __skb_queue_tail(&sd->input_pkt_queue, skb);
2853 input_queue_tail_incr_save(sd, qtail);
2855 local_irq_restore(flags);
2856 return NET_RX_SUCCESS;
2859 /* Schedule NAPI for backlog device
2860 * We can use non atomic operation since we own the queue lock
2862 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2863 if (!rps_ipi_queued(sd))
2864 ____napi_schedule(sd, &sd->backlog);
2872 local_irq_restore(flags);
2874 atomic_long_inc(&skb->dev->rx_dropped);
2880 * netif_rx - post buffer to the network code
2881 * @skb: buffer to post
2883 * This function receives a packet from a device driver and queues it for
2884 * the upper (protocol) levels to process. It always succeeds. The buffer
2885 * may be dropped during processing for congestion control or by the
2889 * NET_RX_SUCCESS (no congestion)
2890 * NET_RX_DROP (packet was dropped)
2894 int netif_rx(struct sk_buff *skb)
2898 /* if netpoll wants it, pretend we never saw it */
2899 if (netpoll_rx(skb))
2902 net_timestamp_check(netdev_tstamp_prequeue, skb);
2904 trace_netif_rx(skb);
2906 if (static_key_false(&rps_needed)) {
2907 struct rps_dev_flow voidflow, *rflow = &voidflow;
2913 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2915 cpu = smp_processor_id();
2917 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2925 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2930 EXPORT_SYMBOL(netif_rx);
2932 int netif_rx_ni(struct sk_buff *skb)
2937 err = netif_rx(skb);
2938 if (local_softirq_pending())
2944 EXPORT_SYMBOL(netif_rx_ni);
2946 static void net_tx_action(struct softirq_action *h)
2948 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2950 if (sd->completion_queue) {
2951 struct sk_buff *clist;
2953 local_irq_disable();
2954 clist = sd->completion_queue;
2955 sd->completion_queue = NULL;
2959 struct sk_buff *skb = clist;
2960 clist = clist->next;
2962 WARN_ON(atomic_read(&skb->users));
2963 trace_kfree_skb(skb, net_tx_action);
2968 if (sd->output_queue) {
2971 local_irq_disable();
2972 head = sd->output_queue;
2973 sd->output_queue = NULL;
2974 sd->output_queue_tailp = &sd->output_queue;
2978 struct Qdisc *q = head;
2979 spinlock_t *root_lock;
2981 head = head->next_sched;
2983 root_lock = qdisc_lock(q);
2984 if (spin_trylock(root_lock)) {
2985 smp_mb__before_clear_bit();
2986 clear_bit(__QDISC_STATE_SCHED,
2989 spin_unlock(root_lock);
2991 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2993 __netif_reschedule(q);
2995 smp_mb__before_clear_bit();
2996 clear_bit(__QDISC_STATE_SCHED,
3004 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3005 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3006 /* This hook is defined here for ATM LANE */
3007 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3008 unsigned char *addr) __read_mostly;
3009 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3012 #ifdef CONFIG_NET_CLS_ACT
3013 /* TODO: Maybe we should just force sch_ingress to be compiled in
3014 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3015 * a compare and 2 stores extra right now if we dont have it on
3016 * but have CONFIG_NET_CLS_ACT
3017 * NOTE: This doesn't stop any functionality; if you dont have
3018 * the ingress scheduler, you just can't add policies on ingress.
3021 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3023 struct net_device *dev = skb->dev;
3024 u32 ttl = G_TC_RTTL(skb->tc_verd);
3025 int result = TC_ACT_OK;
3028 if (unlikely(MAX_RED_LOOP < ttl++)) {
3029 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3030 skb->skb_iif, dev->ifindex);
3034 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3035 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3038 if (q != &noop_qdisc) {
3039 spin_lock(qdisc_lock(q));
3040 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3041 result = qdisc_enqueue_root(skb, q);
3042 spin_unlock(qdisc_lock(q));
3048 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3049 struct packet_type **pt_prev,
3050 int *ret, struct net_device *orig_dev)
3052 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3054 if (!rxq || rxq->qdisc == &noop_qdisc)
3058 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3062 switch (ing_filter(skb, rxq)) {
3076 * netdev_rx_handler_register - register receive handler
3077 * @dev: device to register a handler for
3078 * @rx_handler: receive handler to register
3079 * @rx_handler_data: data pointer that is used by rx handler
3081 * Register a receive hander for a device. This handler will then be
3082 * called from __netif_receive_skb. A negative errno code is returned
3085 * The caller must hold the rtnl_mutex.
3087 * For a general description of rx_handler, see enum rx_handler_result.
3089 int netdev_rx_handler_register(struct net_device *dev,
3090 rx_handler_func_t *rx_handler,
3091 void *rx_handler_data)
3095 if (dev->rx_handler)
3098 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3099 rcu_assign_pointer(dev->rx_handler, rx_handler);
3103 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3106 * netdev_rx_handler_unregister - unregister receive handler
3107 * @dev: device to unregister a handler from
3109 * Unregister a receive hander from a device.
3111 * The caller must hold the rtnl_mutex.
3113 void netdev_rx_handler_unregister(struct net_device *dev)
3117 RCU_INIT_POINTER(dev->rx_handler, NULL);
3118 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3120 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3122 static int __netif_receive_skb(struct sk_buff *skb)
3124 struct packet_type *ptype, *pt_prev;
3125 rx_handler_func_t *rx_handler;
3126 struct net_device *orig_dev;
3127 struct net_device *null_or_dev;
3128 bool deliver_exact = false;
3129 int ret = NET_RX_DROP;
3132 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3134 trace_netif_receive_skb(skb);
3136 /* if we've gotten here through NAPI, check netpoll */
3137 if (netpoll_receive_skb(skb))
3141 skb->skb_iif = skb->dev->ifindex;
3142 orig_dev = skb->dev;
3144 skb_reset_network_header(skb);
3145 skb_reset_transport_header(skb);
3146 skb_reset_mac_len(skb);
3154 __this_cpu_inc(softnet_data.processed);
3156 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3157 skb = vlan_untag(skb);
3162 #ifdef CONFIG_NET_CLS_ACT
3163 if (skb->tc_verd & TC_NCLS) {
3164 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3169 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3170 if (!ptype->dev || ptype->dev == skb->dev) {
3172 ret = deliver_skb(skb, pt_prev, orig_dev);
3177 #ifdef CONFIG_NET_CLS_ACT
3178 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3184 rx_handler = rcu_dereference(skb->dev->rx_handler);
3185 if (vlan_tx_tag_present(skb)) {
3187 ret = deliver_skb(skb, pt_prev, orig_dev);
3190 if (vlan_do_receive(&skb, !rx_handler))
3192 else if (unlikely(!skb))
3198 ret = deliver_skb(skb, pt_prev, orig_dev);
3201 switch (rx_handler(&skb)) {
3202 case RX_HANDLER_CONSUMED:
3204 case RX_HANDLER_ANOTHER:
3206 case RX_HANDLER_EXACT:
3207 deliver_exact = true;
3208 case RX_HANDLER_PASS:
3215 /* deliver only exact match when indicated */
3216 null_or_dev = deliver_exact ? skb->dev : NULL;
3218 type = skb->protocol;
3219 list_for_each_entry_rcu(ptype,
3220 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3221 if (ptype->type == type &&
3222 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3223 ptype->dev == orig_dev)) {
3225 ret = deliver_skb(skb, pt_prev, orig_dev);
3231 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3233 atomic_long_inc(&skb->dev->rx_dropped);
3235 /* Jamal, now you will not able to escape explaining
3236 * me how you were going to use this. :-)
3247 * netif_receive_skb - process receive buffer from network
3248 * @skb: buffer to process
3250 * netif_receive_skb() is the main receive data processing function.
3251 * It always succeeds. The buffer may be dropped during processing
3252 * for congestion control or by the protocol layers.
3254 * This function may only be called from softirq context and interrupts
3255 * should be enabled.
3257 * Return values (usually ignored):
3258 * NET_RX_SUCCESS: no congestion
3259 * NET_RX_DROP: packet was dropped
3261 int netif_receive_skb(struct sk_buff *skb)
3263 net_timestamp_check(netdev_tstamp_prequeue, skb);
3265 if (skb_defer_rx_timestamp(skb))
3266 return NET_RX_SUCCESS;
3269 if (static_key_false(&rps_needed)) {
3270 struct rps_dev_flow voidflow, *rflow = &voidflow;
3275 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3278 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3285 return __netif_receive_skb(skb);
3287 EXPORT_SYMBOL(netif_receive_skb);
3289 /* Network device is going away, flush any packets still pending
3290 * Called with irqs disabled.
3292 static void flush_backlog(void *arg)
3294 struct net_device *dev = arg;
3295 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3296 struct sk_buff *skb, *tmp;
3299 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3300 if (skb->dev == dev) {
3301 __skb_unlink(skb, &sd->input_pkt_queue);
3303 input_queue_head_incr(sd);
3308 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3309 if (skb->dev == dev) {
3310 __skb_unlink(skb, &sd->process_queue);
3312 input_queue_head_incr(sd);
3317 static int napi_gro_complete(struct sk_buff *skb)
3319 struct packet_type *ptype;
3320 __be16 type = skb->protocol;
3321 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3324 if (NAPI_GRO_CB(skb)->count == 1) {
3325 skb_shinfo(skb)->gso_size = 0;
3330 list_for_each_entry_rcu(ptype, head, list) {
3331 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3334 err = ptype->gro_complete(skb);
3340 WARN_ON(&ptype->list == head);
3342 return NET_RX_SUCCESS;
3346 return netif_receive_skb(skb);
3349 inline void napi_gro_flush(struct napi_struct *napi)
3351 struct sk_buff *skb, *next;
3353 for (skb = napi->gro_list; skb; skb = next) {
3356 napi_gro_complete(skb);
3359 napi->gro_count = 0;
3360 napi->gro_list = NULL;
3362 EXPORT_SYMBOL(napi_gro_flush);
3364 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3366 struct sk_buff **pp = NULL;
3367 struct packet_type *ptype;
3368 __be16 type = skb->protocol;
3369 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3372 enum gro_result ret;
3374 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3377 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3381 list_for_each_entry_rcu(ptype, head, list) {
3382 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3385 skb_set_network_header(skb, skb_gro_offset(skb));
3386 mac_len = skb->network_header - skb->mac_header;
3387 skb->mac_len = mac_len;
3388 NAPI_GRO_CB(skb)->same_flow = 0;
3389 NAPI_GRO_CB(skb)->flush = 0;
3390 NAPI_GRO_CB(skb)->free = 0;
3392 pp = ptype->gro_receive(&napi->gro_list, skb);
3397 if (&ptype->list == head)
3400 same_flow = NAPI_GRO_CB(skb)->same_flow;
3401 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3404 struct sk_buff *nskb = *pp;
3408 napi_gro_complete(nskb);
3415 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3419 NAPI_GRO_CB(skb)->count = 1;
3420 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3421 skb->next = napi->gro_list;
3422 napi->gro_list = skb;
3426 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3427 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3429 BUG_ON(skb->end - skb->tail < grow);
3431 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3434 skb->data_len -= grow;
3436 skb_shinfo(skb)->frags[0].page_offset += grow;
3437 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3439 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3440 skb_frag_unref(skb, 0);
3441 memmove(skb_shinfo(skb)->frags,
3442 skb_shinfo(skb)->frags + 1,
3443 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3454 EXPORT_SYMBOL(dev_gro_receive);
3456 static inline gro_result_t
3457 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3460 unsigned int maclen = skb->dev->hard_header_len;
3462 for (p = napi->gro_list; p; p = p->next) {
3463 unsigned long diffs;
3465 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3466 diffs |= p->vlan_tci ^ skb->vlan_tci;
3467 if (maclen == ETH_HLEN)
3468 diffs |= compare_ether_header(skb_mac_header(p),
3469 skb_gro_mac_header(skb));
3471 diffs = memcmp(skb_mac_header(p),
3472 skb_gro_mac_header(skb),
3474 NAPI_GRO_CB(p)->same_flow = !diffs;
3475 NAPI_GRO_CB(p)->flush = 0;
3478 return dev_gro_receive(napi, skb);
3481 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3485 if (netif_receive_skb(skb))
3493 case GRO_MERGED_FREE:
3494 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3495 kmem_cache_free(skbuff_head_cache, skb);
3507 EXPORT_SYMBOL(napi_skb_finish);
3509 void skb_gro_reset_offset(struct sk_buff *skb)
3511 NAPI_GRO_CB(skb)->data_offset = 0;
3512 NAPI_GRO_CB(skb)->frag0 = NULL;
3513 NAPI_GRO_CB(skb)->frag0_len = 0;
3515 if (skb->mac_header == skb->tail &&
3516 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3517 NAPI_GRO_CB(skb)->frag0 =
3518 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3519 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3522 EXPORT_SYMBOL(skb_gro_reset_offset);
3524 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3526 skb_gro_reset_offset(skb);
3528 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3530 EXPORT_SYMBOL(napi_gro_receive);
3532 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3534 __skb_pull(skb, skb_headlen(skb));
3535 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3536 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3538 skb->dev = napi->dev;
3544 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3546 struct sk_buff *skb = napi->skb;
3549 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3555 EXPORT_SYMBOL(napi_get_frags);
3557 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3563 skb->protocol = eth_type_trans(skb, skb->dev);
3565 if (ret == GRO_HELD)
3566 skb_gro_pull(skb, -ETH_HLEN);
3567 else if (netif_receive_skb(skb))
3572 case GRO_MERGED_FREE:
3573 napi_reuse_skb(napi, skb);
3582 EXPORT_SYMBOL(napi_frags_finish);
3584 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3586 struct sk_buff *skb = napi->skb;
3593 skb_reset_mac_header(skb);
3594 skb_gro_reset_offset(skb);
3596 off = skb_gro_offset(skb);
3597 hlen = off + sizeof(*eth);
3598 eth = skb_gro_header_fast(skb, off);
3599 if (skb_gro_header_hard(skb, hlen)) {
3600 eth = skb_gro_header_slow(skb, hlen, off);
3601 if (unlikely(!eth)) {
3602 napi_reuse_skb(napi, skb);
3608 skb_gro_pull(skb, sizeof(*eth));
3611 * This works because the only protocols we care about don't require
3612 * special handling. We'll fix it up properly at the end.
3614 skb->protocol = eth->h_proto;
3620 gro_result_t napi_gro_frags(struct napi_struct *napi)
3622 struct sk_buff *skb = napi_frags_skb(napi);
3627 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3629 EXPORT_SYMBOL(napi_gro_frags);
3632 * net_rps_action sends any pending IPI's for rps.
3633 * Note: called with local irq disabled, but exits with local irq enabled.
3635 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3638 struct softnet_data *remsd = sd->rps_ipi_list;
3641 sd->rps_ipi_list = NULL;
3645 /* Send pending IPI's to kick RPS processing on remote cpus. */
3647 struct softnet_data *next = remsd->rps_ipi_next;
3649 if (cpu_online(remsd->cpu))
3650 __smp_call_function_single(remsd->cpu,
3659 static int process_backlog(struct napi_struct *napi, int quota)
3662 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3665 /* Check if we have pending ipi, its better to send them now,
3666 * not waiting net_rx_action() end.
3668 if (sd->rps_ipi_list) {
3669 local_irq_disable();
3670 net_rps_action_and_irq_enable(sd);
3673 napi->weight = weight_p;
3674 local_irq_disable();
3675 while (work < quota) {
3676 struct sk_buff *skb;
3679 while ((skb = __skb_dequeue(&sd->process_queue))) {
3681 __netif_receive_skb(skb);
3682 local_irq_disable();
3683 input_queue_head_incr(sd);
3684 if (++work >= quota) {
3691 qlen = skb_queue_len(&sd->input_pkt_queue);
3693 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3694 &sd->process_queue);
3696 if (qlen < quota - work) {
3698 * Inline a custom version of __napi_complete().
3699 * only current cpu owns and manipulates this napi,
3700 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3701 * we can use a plain write instead of clear_bit(),
3702 * and we dont need an smp_mb() memory barrier.
3704 list_del(&napi->poll_list);
3707 quota = work + qlen;
3717 * __napi_schedule - schedule for receive
3718 * @n: entry to schedule
3720 * The entry's receive function will be scheduled to run
3722 void __napi_schedule(struct napi_struct *n)
3724 unsigned long flags;
3726 local_irq_save(flags);
3727 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3728 local_irq_restore(flags);
3730 EXPORT_SYMBOL(__napi_schedule);
3732 void __napi_complete(struct napi_struct *n)
3734 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3735 BUG_ON(n->gro_list);
3737 list_del(&n->poll_list);
3738 smp_mb__before_clear_bit();
3739 clear_bit(NAPI_STATE_SCHED, &n->state);
3741 EXPORT_SYMBOL(__napi_complete);
3743 void napi_complete(struct napi_struct *n)
3745 unsigned long flags;
3748 * don't let napi dequeue from the cpu poll list
3749 * just in case its running on a different cpu
3751 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3755 local_irq_save(flags);
3757 local_irq_restore(flags);
3759 EXPORT_SYMBOL(napi_complete);
3761 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3762 int (*poll)(struct napi_struct *, int), int weight)
3764 INIT_LIST_HEAD(&napi->poll_list);
3765 napi->gro_count = 0;
3766 napi->gro_list = NULL;
3769 napi->weight = weight;
3770 list_add(&napi->dev_list, &dev->napi_list);
3772 #ifdef CONFIG_NETPOLL
3773 spin_lock_init(&napi->poll_lock);
3774 napi->poll_owner = -1;
3776 set_bit(NAPI_STATE_SCHED, &napi->state);
3778 EXPORT_SYMBOL(netif_napi_add);
3780 void netif_napi_del(struct napi_struct *napi)
3782 struct sk_buff *skb, *next;
3784 list_del_init(&napi->dev_list);
3785 napi_free_frags(napi);
3787 for (skb = napi->gro_list; skb; skb = next) {
3793 napi->gro_list = NULL;
3794 napi->gro_count = 0;
3796 EXPORT_SYMBOL(netif_napi_del);
3798 static void net_rx_action(struct softirq_action *h)
3800 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3801 unsigned long time_limit = jiffies + 2;
3802 int budget = netdev_budget;
3805 local_irq_disable();
3807 while (!list_empty(&sd->poll_list)) {
3808 struct napi_struct *n;
3811 /* If softirq window is exhuasted then punt.
3812 * Allow this to run for 2 jiffies since which will allow
3813 * an average latency of 1.5/HZ.
3815 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3820 /* Even though interrupts have been re-enabled, this
3821 * access is safe because interrupts can only add new
3822 * entries to the tail of this list, and only ->poll()
3823 * calls can remove this head entry from the list.
3825 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3827 have = netpoll_poll_lock(n);
3831 /* This NAPI_STATE_SCHED test is for avoiding a race
3832 * with netpoll's poll_napi(). Only the entity which
3833 * obtains the lock and sees NAPI_STATE_SCHED set will
3834 * actually make the ->poll() call. Therefore we avoid
3835 * accidentally calling ->poll() when NAPI is not scheduled.
3838 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3839 work = n->poll(n, weight);
3843 WARN_ON_ONCE(work > weight);
3847 local_irq_disable();
3849 /* Drivers must not modify the NAPI state if they
3850 * consume the entire weight. In such cases this code
3851 * still "owns" the NAPI instance and therefore can
3852 * move the instance around on the list at-will.
3854 if (unlikely(work == weight)) {
3855 if (unlikely(napi_disable_pending(n))) {
3858 local_irq_disable();
3860 list_move_tail(&n->poll_list, &sd->poll_list);
3863 netpoll_poll_unlock(have);
3866 net_rps_action_and_irq_enable(sd);
3868 #ifdef CONFIG_NET_DMA
3870 * There may not be any more sk_buffs coming right now, so push
3871 * any pending DMA copies to hardware
3873 dma_issue_pending_all();
3880 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3884 static gifconf_func_t *gifconf_list[NPROTO];
3887 * register_gifconf - register a SIOCGIF handler
3888 * @family: Address family
3889 * @gifconf: Function handler
3891 * Register protocol dependent address dumping routines. The handler
3892 * that is passed must not be freed or reused until it has been replaced
3893 * by another handler.
3895 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3897 if (family >= NPROTO)
3899 gifconf_list[family] = gifconf;
3902 EXPORT_SYMBOL(register_gifconf);
3906 * Map an interface index to its name (SIOCGIFNAME)
3910 * We need this ioctl for efficient implementation of the
3911 * if_indextoname() function required by the IPv6 API. Without
3912 * it, we would have to search all the interfaces to find a
3916 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3918 struct net_device *dev;
3922 * Fetch the caller's info block.
3925 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3929 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3935 strcpy(ifr.ifr_name, dev->name);
3938 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3944 * Perform a SIOCGIFCONF call. This structure will change
3945 * size eventually, and there is nothing I can do about it.
3946 * Thus we will need a 'compatibility mode'.
3949 static int dev_ifconf(struct net *net, char __user *arg)
3952 struct net_device *dev;
3959 * Fetch the caller's info block.
3962 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3969 * Loop over the interfaces, and write an info block for each.
3973 for_each_netdev(net, dev) {
3974 for (i = 0; i < NPROTO; i++) {
3975 if (gifconf_list[i]) {
3978 done = gifconf_list[i](dev, NULL, 0);
3980 done = gifconf_list[i](dev, pos + total,
3990 * All done. Write the updated control block back to the caller.
3992 ifc.ifc_len = total;
3995 * Both BSD and Solaris return 0 here, so we do too.
3997 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4000 #ifdef CONFIG_PROC_FS
4002 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4004 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4005 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4006 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4008 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4010 struct net *net = seq_file_net(seq);
4011 struct net_device *dev;
4012 struct hlist_node *p;
4013 struct hlist_head *h;
4014 unsigned int count = 0, offset = get_offset(*pos);
4016 h = &net->dev_name_head[get_bucket(*pos)];
4017 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4018 if (++count == offset)
4025 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4027 struct net_device *dev;
4028 unsigned int bucket;
4031 dev = dev_from_same_bucket(seq, pos);
4035 bucket = get_bucket(*pos) + 1;
4036 *pos = set_bucket_offset(bucket, 1);
4037 } while (bucket < NETDEV_HASHENTRIES);
4043 * This is invoked by the /proc filesystem handler to display a device
4046 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4051 return SEQ_START_TOKEN;
4053 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4056 return dev_from_bucket(seq, pos);
4059 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4062 return dev_from_bucket(seq, pos);
4065 void dev_seq_stop(struct seq_file *seq, void *v)
4071 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4073 struct rtnl_link_stats64 temp;
4074 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4076 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4077 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4078 dev->name, stats->rx_bytes, stats->rx_packets,
4080 stats->rx_dropped + stats->rx_missed_errors,
4081 stats->rx_fifo_errors,
4082 stats->rx_length_errors + stats->rx_over_errors +
4083 stats->rx_crc_errors + stats->rx_frame_errors,
4084 stats->rx_compressed, stats->multicast,
4085 stats->tx_bytes, stats->tx_packets,
4086 stats->tx_errors, stats->tx_dropped,
4087 stats->tx_fifo_errors, stats->collisions,
4088 stats->tx_carrier_errors +
4089 stats->tx_aborted_errors +
4090 stats->tx_window_errors +
4091 stats->tx_heartbeat_errors,
4092 stats->tx_compressed);
4096 * Called from the PROCfs module. This now uses the new arbitrary sized
4097 * /proc/net interface to create /proc/net/dev
4099 static int dev_seq_show(struct seq_file *seq, void *v)
4101 if (v == SEQ_START_TOKEN)
4102 seq_puts(seq, "Inter-| Receive "
4104 " face |bytes packets errs drop fifo frame "
4105 "compressed multicast|bytes packets errs "
4106 "drop fifo colls carrier compressed\n");
4108 dev_seq_printf_stats(seq, v);
4112 static struct softnet_data *softnet_get_online(loff_t *pos)
4114 struct softnet_data *sd = NULL;
4116 while (*pos < nr_cpu_ids)
4117 if (cpu_online(*pos)) {
4118 sd = &per_cpu(softnet_data, *pos);
4125 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4127 return softnet_get_online(pos);
4130 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4133 return softnet_get_online(pos);
4136 static void softnet_seq_stop(struct seq_file *seq, void *v)
4140 static int softnet_seq_show(struct seq_file *seq, void *v)
4142 struct softnet_data *sd = v;
4144 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4145 sd->processed, sd->dropped, sd->time_squeeze, 0,
4146 0, 0, 0, 0, /* was fastroute */
4147 sd->cpu_collision, sd->received_rps);
4151 static const struct seq_operations dev_seq_ops = {
4152 .start = dev_seq_start,
4153 .next = dev_seq_next,
4154 .stop = dev_seq_stop,
4155 .show = dev_seq_show,
4158 static int dev_seq_open(struct inode *inode, struct file *file)
4160 return seq_open_net(inode, file, &dev_seq_ops,
4161 sizeof(struct seq_net_private));
4164 static const struct file_operations dev_seq_fops = {
4165 .owner = THIS_MODULE,
4166 .open = dev_seq_open,
4168 .llseek = seq_lseek,
4169 .release = seq_release_net,
4172 static const struct seq_operations softnet_seq_ops = {
4173 .start = softnet_seq_start,
4174 .next = softnet_seq_next,
4175 .stop = softnet_seq_stop,
4176 .show = softnet_seq_show,
4179 static int softnet_seq_open(struct inode *inode, struct file *file)
4181 return seq_open(file, &softnet_seq_ops);
4184 static const struct file_operations softnet_seq_fops = {
4185 .owner = THIS_MODULE,
4186 .open = softnet_seq_open,
4188 .llseek = seq_lseek,
4189 .release = seq_release,
4192 static void *ptype_get_idx(loff_t pos)
4194 struct packet_type *pt = NULL;
4198 list_for_each_entry_rcu(pt, &ptype_all, list) {
4204 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4205 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4214 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4218 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4221 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4223 struct packet_type *pt;
4224 struct list_head *nxt;
4228 if (v == SEQ_START_TOKEN)
4229 return ptype_get_idx(0);
4232 nxt = pt->list.next;
4233 if (pt->type == htons(ETH_P_ALL)) {
4234 if (nxt != &ptype_all)
4237 nxt = ptype_base[0].next;
4239 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4241 while (nxt == &ptype_base[hash]) {
4242 if (++hash >= PTYPE_HASH_SIZE)
4244 nxt = ptype_base[hash].next;
4247 return list_entry(nxt, struct packet_type, list);
4250 static void ptype_seq_stop(struct seq_file *seq, void *v)
4256 static int ptype_seq_show(struct seq_file *seq, void *v)
4258 struct packet_type *pt = v;
4260 if (v == SEQ_START_TOKEN)
4261 seq_puts(seq, "Type Device Function\n");
4262 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4263 if (pt->type == htons(ETH_P_ALL))
4264 seq_puts(seq, "ALL ");
4266 seq_printf(seq, "%04x", ntohs(pt->type));
4268 seq_printf(seq, " %-8s %pF\n",
4269 pt->dev ? pt->dev->name : "", pt->func);
4275 static const struct seq_operations ptype_seq_ops = {
4276 .start = ptype_seq_start,
4277 .next = ptype_seq_next,
4278 .stop = ptype_seq_stop,
4279 .show = ptype_seq_show,
4282 static int ptype_seq_open(struct inode *inode, struct file *file)
4284 return seq_open_net(inode, file, &ptype_seq_ops,
4285 sizeof(struct seq_net_private));
4288 static const struct file_operations ptype_seq_fops = {
4289 .owner = THIS_MODULE,
4290 .open = ptype_seq_open,
4292 .llseek = seq_lseek,
4293 .release = seq_release_net,
4297 static int __net_init dev_proc_net_init(struct net *net)
4301 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4303 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4305 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4308 if (wext_proc_init(net))
4314 proc_net_remove(net, "ptype");
4316 proc_net_remove(net, "softnet_stat");
4318 proc_net_remove(net, "dev");
4322 static void __net_exit dev_proc_net_exit(struct net *net)
4324 wext_proc_exit(net);
4326 proc_net_remove(net, "ptype");
4327 proc_net_remove(net, "softnet_stat");
4328 proc_net_remove(net, "dev");
4331 static struct pernet_operations __net_initdata dev_proc_ops = {
4332 .init = dev_proc_net_init,
4333 .exit = dev_proc_net_exit,
4336 static int __init dev_proc_init(void)
4338 return register_pernet_subsys(&dev_proc_ops);
4341 #define dev_proc_init() 0
4342 #endif /* CONFIG_PROC_FS */
4346 * netdev_set_master - set up master pointer
4347 * @slave: slave device
4348 * @master: new master device
4350 * Changes the master device of the slave. Pass %NULL to break the
4351 * bonding. The caller must hold the RTNL semaphore. On a failure
4352 * a negative errno code is returned. On success the reference counts
4353 * are adjusted and the function returns zero.
4355 int netdev_set_master(struct net_device *slave, struct net_device *master)
4357 struct net_device *old = slave->master;
4367 slave->master = master;
4373 EXPORT_SYMBOL(netdev_set_master);
4376 * netdev_set_bond_master - set up bonding master/slave pair
4377 * @slave: slave device
4378 * @master: new master device
4380 * Changes the master device of the slave. Pass %NULL to break the
4381 * bonding. The caller must hold the RTNL semaphore. On a failure
4382 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4383 * to the routing socket and the function returns zero.
4385 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4391 err = netdev_set_master(slave, master);
4395 slave->flags |= IFF_SLAVE;
4397 slave->flags &= ~IFF_SLAVE;
4399 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4402 EXPORT_SYMBOL(netdev_set_bond_master);
4404 static void dev_change_rx_flags(struct net_device *dev, int flags)
4406 const struct net_device_ops *ops = dev->netdev_ops;
4408 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4409 ops->ndo_change_rx_flags(dev, flags);
4412 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4414 unsigned int old_flags = dev->flags;
4420 dev->flags |= IFF_PROMISC;
4421 dev->promiscuity += inc;
4422 if (dev->promiscuity == 0) {
4425 * If inc causes overflow, untouch promisc and return error.
4428 dev->flags &= ~IFF_PROMISC;
4430 dev->promiscuity -= inc;
4431 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4436 if (dev->flags != old_flags) {
4437 pr_info("device %s %s promiscuous mode\n",
4439 dev->flags & IFF_PROMISC ? "entered" : "left");
4440 if (audit_enabled) {
4441 current_uid_gid(&uid, &gid);
4442 audit_log(current->audit_context, GFP_ATOMIC,
4443 AUDIT_ANOM_PROMISCUOUS,
4444 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4445 dev->name, (dev->flags & IFF_PROMISC),
4446 (old_flags & IFF_PROMISC),
4447 audit_get_loginuid(current),
4449 audit_get_sessionid(current));
4452 dev_change_rx_flags(dev, IFF_PROMISC);
4458 * dev_set_promiscuity - update promiscuity count on a device
4462 * Add or remove promiscuity from a device. While the count in the device
4463 * remains above zero the interface remains promiscuous. Once it hits zero
4464 * the device reverts back to normal filtering operation. A negative inc
4465 * value is used to drop promiscuity on the device.
4466 * Return 0 if successful or a negative errno code on error.
4468 int dev_set_promiscuity(struct net_device *dev, int inc)
4470 unsigned int old_flags = dev->flags;
4473 err = __dev_set_promiscuity(dev, inc);
4476 if (dev->flags != old_flags)
4477 dev_set_rx_mode(dev);
4480 EXPORT_SYMBOL(dev_set_promiscuity);
4483 * dev_set_allmulti - update allmulti count on a device
4487 * Add or remove reception of all multicast frames to a device. While the
4488 * count in the device remains above zero the interface remains listening
4489 * to all interfaces. Once it hits zero the device reverts back to normal
4490 * filtering operation. A negative @inc value is used to drop the counter
4491 * when releasing a resource needing all multicasts.
4492 * Return 0 if successful or a negative errno code on error.
4495 int dev_set_allmulti(struct net_device *dev, int inc)
4497 unsigned int old_flags = dev->flags;
4501 dev->flags |= IFF_ALLMULTI;
4502 dev->allmulti += inc;
4503 if (dev->allmulti == 0) {
4506 * If inc causes overflow, untouch allmulti and return error.
4509 dev->flags &= ~IFF_ALLMULTI;
4511 dev->allmulti -= inc;
4512 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4517 if (dev->flags ^ old_flags) {
4518 dev_change_rx_flags(dev, IFF_ALLMULTI);
4519 dev_set_rx_mode(dev);
4523 EXPORT_SYMBOL(dev_set_allmulti);
4526 * Upload unicast and multicast address lists to device and
4527 * configure RX filtering. When the device doesn't support unicast
4528 * filtering it is put in promiscuous mode while unicast addresses
4531 void __dev_set_rx_mode(struct net_device *dev)
4533 const struct net_device_ops *ops = dev->netdev_ops;
4535 /* dev_open will call this function so the list will stay sane. */
4536 if (!(dev->flags&IFF_UP))
4539 if (!netif_device_present(dev))
4542 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4543 /* Unicast addresses changes may only happen under the rtnl,
4544 * therefore calling __dev_set_promiscuity here is safe.
4546 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4547 __dev_set_promiscuity(dev, 1);
4548 dev->uc_promisc = true;
4549 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4550 __dev_set_promiscuity(dev, -1);
4551 dev->uc_promisc = false;
4555 if (ops->ndo_set_rx_mode)
4556 ops->ndo_set_rx_mode(dev);
4559 void dev_set_rx_mode(struct net_device *dev)
4561 netif_addr_lock_bh(dev);
4562 __dev_set_rx_mode(dev);
4563 netif_addr_unlock_bh(dev);
4567 * dev_get_flags - get flags reported to userspace
4570 * Get the combination of flag bits exported through APIs to userspace.
4572 unsigned int dev_get_flags(const struct net_device *dev)
4576 flags = (dev->flags & ~(IFF_PROMISC |
4581 (dev->gflags & (IFF_PROMISC |
4584 if (netif_running(dev)) {
4585 if (netif_oper_up(dev))
4586 flags |= IFF_RUNNING;
4587 if (netif_carrier_ok(dev))
4588 flags |= IFF_LOWER_UP;
4589 if (netif_dormant(dev))
4590 flags |= IFF_DORMANT;
4595 EXPORT_SYMBOL(dev_get_flags);
4597 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4599 unsigned int old_flags = dev->flags;
4605 * Set the flags on our device.
4608 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4609 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4611 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4615 * Load in the correct multicast list now the flags have changed.
4618 if ((old_flags ^ flags) & IFF_MULTICAST)
4619 dev_change_rx_flags(dev, IFF_MULTICAST);
4621 dev_set_rx_mode(dev);
4624 * Have we downed the interface. We handle IFF_UP ourselves
4625 * according to user attempts to set it, rather than blindly
4630 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4631 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4634 dev_set_rx_mode(dev);
4637 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4638 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4640 dev->gflags ^= IFF_PROMISC;
4641 dev_set_promiscuity(dev, inc);
4644 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4645 is important. Some (broken) drivers set IFF_PROMISC, when
4646 IFF_ALLMULTI is requested not asking us and not reporting.
4648 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4649 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4651 dev->gflags ^= IFF_ALLMULTI;
4652 dev_set_allmulti(dev, inc);
4658 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4660 unsigned int changes = dev->flags ^ old_flags;
4662 if (changes & IFF_UP) {
4663 if (dev->flags & IFF_UP)
4664 call_netdevice_notifiers(NETDEV_UP, dev);
4666 call_netdevice_notifiers(NETDEV_DOWN, dev);
4669 if (dev->flags & IFF_UP &&
4670 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4671 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4675 * dev_change_flags - change device settings
4677 * @flags: device state flags
4679 * Change settings on device based state flags. The flags are
4680 * in the userspace exported format.
4682 int dev_change_flags(struct net_device *dev, unsigned int flags)
4685 unsigned int changes, old_flags = dev->flags;
4687 ret = __dev_change_flags(dev, flags);
4691 changes = old_flags ^ dev->flags;
4693 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4695 __dev_notify_flags(dev, old_flags);
4698 EXPORT_SYMBOL(dev_change_flags);
4701 * dev_set_mtu - Change maximum transfer unit
4703 * @new_mtu: new transfer unit
4705 * Change the maximum transfer size of the network device.
4707 int dev_set_mtu(struct net_device *dev, int new_mtu)
4709 const struct net_device_ops *ops = dev->netdev_ops;
4712 if (new_mtu == dev->mtu)
4715 /* MTU must be positive. */
4719 if (!netif_device_present(dev))
4723 if (ops->ndo_change_mtu)
4724 err = ops->ndo_change_mtu(dev, new_mtu);
4728 if (!err && dev->flags & IFF_UP)
4729 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4732 EXPORT_SYMBOL(dev_set_mtu);
4735 * dev_set_group - Change group this device belongs to
4737 * @new_group: group this device should belong to
4739 void dev_set_group(struct net_device *dev, int new_group)
4741 dev->group = new_group;
4743 EXPORT_SYMBOL(dev_set_group);
4746 * dev_set_mac_address - Change Media Access Control Address
4750 * Change the hardware (MAC) address of the device
4752 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4754 const struct net_device_ops *ops = dev->netdev_ops;
4757 if (!ops->ndo_set_mac_address)
4759 if (sa->sa_family != dev->type)
4761 if (!netif_device_present(dev))
4763 err = ops->ndo_set_mac_address(dev, sa);
4765 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4768 EXPORT_SYMBOL(dev_set_mac_address);
4771 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4773 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4776 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4782 case SIOCGIFFLAGS: /* Get interface flags */
4783 ifr->ifr_flags = (short) dev_get_flags(dev);
4786 case SIOCGIFMETRIC: /* Get the metric on the interface
4787 (currently unused) */
4788 ifr->ifr_metric = 0;
4791 case SIOCGIFMTU: /* Get the MTU of a device */
4792 ifr->ifr_mtu = dev->mtu;
4797 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4799 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4800 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4801 ifr->ifr_hwaddr.sa_family = dev->type;
4809 ifr->ifr_map.mem_start = dev->mem_start;
4810 ifr->ifr_map.mem_end = dev->mem_end;
4811 ifr->ifr_map.base_addr = dev->base_addr;
4812 ifr->ifr_map.irq = dev->irq;
4813 ifr->ifr_map.dma = dev->dma;
4814 ifr->ifr_map.port = dev->if_port;
4818 ifr->ifr_ifindex = dev->ifindex;
4822 ifr->ifr_qlen = dev->tx_queue_len;
4826 /* dev_ioctl() should ensure this case
4838 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4840 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4843 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4844 const struct net_device_ops *ops;
4849 ops = dev->netdev_ops;
4852 case SIOCSIFFLAGS: /* Set interface flags */
4853 return dev_change_flags(dev, ifr->ifr_flags);
4855 case SIOCSIFMETRIC: /* Set the metric on the interface
4856 (currently unused) */
4859 case SIOCSIFMTU: /* Set the MTU of a device */
4860 return dev_set_mtu(dev, ifr->ifr_mtu);
4863 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4865 case SIOCSIFHWBROADCAST:
4866 if (ifr->ifr_hwaddr.sa_family != dev->type)
4868 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4869 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4870 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4874 if (ops->ndo_set_config) {
4875 if (!netif_device_present(dev))
4877 return ops->ndo_set_config(dev, &ifr->ifr_map);
4882 if (!ops->ndo_set_rx_mode ||
4883 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4885 if (!netif_device_present(dev))
4887 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4890 if (!ops->ndo_set_rx_mode ||
4891 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4893 if (!netif_device_present(dev))
4895 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4898 if (ifr->ifr_qlen < 0)
4900 dev->tx_queue_len = ifr->ifr_qlen;
4904 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4905 return dev_change_name(dev, ifr->ifr_newname);
4908 err = net_hwtstamp_validate(ifr);
4914 * Unknown or private ioctl
4917 if ((cmd >= SIOCDEVPRIVATE &&
4918 cmd <= SIOCDEVPRIVATE + 15) ||
4919 cmd == SIOCBONDENSLAVE ||
4920 cmd == SIOCBONDRELEASE ||
4921 cmd == SIOCBONDSETHWADDR ||
4922 cmd == SIOCBONDSLAVEINFOQUERY ||
4923 cmd == SIOCBONDINFOQUERY ||
4924 cmd == SIOCBONDCHANGEACTIVE ||
4925 cmd == SIOCGMIIPHY ||
4926 cmd == SIOCGMIIREG ||
4927 cmd == SIOCSMIIREG ||
4928 cmd == SIOCBRADDIF ||
4929 cmd == SIOCBRDELIF ||
4930 cmd == SIOCSHWTSTAMP ||
4931 cmd == SIOCWANDEV) {
4933 if (ops->ndo_do_ioctl) {
4934 if (netif_device_present(dev))
4935 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4947 * This function handles all "interface"-type I/O control requests. The actual
4948 * 'doing' part of this is dev_ifsioc above.
4952 * dev_ioctl - network device ioctl
4953 * @net: the applicable net namespace
4954 * @cmd: command to issue
4955 * @arg: pointer to a struct ifreq in user space
4957 * Issue ioctl functions to devices. This is normally called by the
4958 * user space syscall interfaces but can sometimes be useful for
4959 * other purposes. The return value is the return from the syscall if
4960 * positive or a negative errno code on error.
4963 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4969 /* One special case: SIOCGIFCONF takes ifconf argument
4970 and requires shared lock, because it sleeps writing
4974 if (cmd == SIOCGIFCONF) {
4976 ret = dev_ifconf(net, (char __user *) arg);
4980 if (cmd == SIOCGIFNAME)
4981 return dev_ifname(net, (struct ifreq __user *)arg);
4983 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4986 ifr.ifr_name[IFNAMSIZ-1] = 0;
4988 colon = strchr(ifr.ifr_name, ':');
4993 * See which interface the caller is talking about.
4998 * These ioctl calls:
4999 * - can be done by all.
5000 * - atomic and do not require locking.
5011 dev_load(net, ifr.ifr_name);
5013 ret = dev_ifsioc_locked(net, &ifr, cmd);
5018 if (copy_to_user(arg, &ifr,
5019 sizeof(struct ifreq)))
5025 dev_load(net, ifr.ifr_name);
5027 ret = dev_ethtool(net, &ifr);
5032 if (copy_to_user(arg, &ifr,
5033 sizeof(struct ifreq)))
5039 * These ioctl calls:
5040 * - require superuser power.
5041 * - require strict serialization.
5047 if (!capable(CAP_NET_ADMIN))
5049 dev_load(net, ifr.ifr_name);
5051 ret = dev_ifsioc(net, &ifr, cmd);
5056 if (copy_to_user(arg, &ifr,
5057 sizeof(struct ifreq)))
5063 * These ioctl calls:
5064 * - require superuser power.
5065 * - require strict serialization.
5066 * - do not return a value
5076 case SIOCSIFHWBROADCAST:
5079 case SIOCBONDENSLAVE:
5080 case SIOCBONDRELEASE:
5081 case SIOCBONDSETHWADDR:
5082 case SIOCBONDCHANGEACTIVE:
5086 if (!capable(CAP_NET_ADMIN))
5089 case SIOCBONDSLAVEINFOQUERY:
5090 case SIOCBONDINFOQUERY:
5091 dev_load(net, ifr.ifr_name);
5093 ret = dev_ifsioc(net, &ifr, cmd);
5098 /* Get the per device memory space. We can add this but
5099 * currently do not support it */
5101 /* Set the per device memory buffer space.
5102 * Not applicable in our case */
5107 * Unknown or private ioctl.
5110 if (cmd == SIOCWANDEV ||
5111 (cmd >= SIOCDEVPRIVATE &&
5112 cmd <= SIOCDEVPRIVATE + 15)) {
5113 dev_load(net, ifr.ifr_name);
5115 ret = dev_ifsioc(net, &ifr, cmd);
5117 if (!ret && copy_to_user(arg, &ifr,
5118 sizeof(struct ifreq)))
5122 /* Take care of Wireless Extensions */
5123 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5124 return wext_handle_ioctl(net, &ifr, cmd, arg);
5131 * dev_new_index - allocate an ifindex
5132 * @net: the applicable net namespace
5134 * Returns a suitable unique value for a new device interface
5135 * number. The caller must hold the rtnl semaphore or the
5136 * dev_base_lock to be sure it remains unique.
5138 static int dev_new_index(struct net *net)
5144 if (!__dev_get_by_index(net, ifindex))
5149 /* Delayed registration/unregisteration */
5150 static LIST_HEAD(net_todo_list);
5152 static void net_set_todo(struct net_device *dev)
5154 list_add_tail(&dev->todo_list, &net_todo_list);
5157 static void rollback_registered_many(struct list_head *head)
5159 struct net_device *dev, *tmp;
5161 BUG_ON(dev_boot_phase);
5164 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5165 /* Some devices call without registering
5166 * for initialization unwind. Remove those
5167 * devices and proceed with the remaining.
5169 if (dev->reg_state == NETREG_UNINITIALIZED) {
5170 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5174 list_del(&dev->unreg_list);
5177 dev->dismantle = true;
5178 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5181 /* If device is running, close it first. */
5182 dev_close_many(head);
5184 list_for_each_entry(dev, head, unreg_list) {
5185 /* And unlink it from device chain. */
5186 unlist_netdevice(dev);
5188 dev->reg_state = NETREG_UNREGISTERING;
5193 list_for_each_entry(dev, head, unreg_list) {
5194 /* Shutdown queueing discipline. */
5198 /* Notify protocols, that we are about to destroy
5199 this device. They should clean all the things.
5201 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5203 if (!dev->rtnl_link_ops ||
5204 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5205 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5208 * Flush the unicast and multicast chains
5213 if (dev->netdev_ops->ndo_uninit)
5214 dev->netdev_ops->ndo_uninit(dev);
5216 /* Notifier chain MUST detach us from master device. */
5217 WARN_ON(dev->master);
5219 /* Remove entries from kobject tree */
5220 netdev_unregister_kobject(dev);
5223 /* Process any work delayed until the end of the batch */
5224 dev = list_first_entry(head, struct net_device, unreg_list);
5225 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5229 list_for_each_entry(dev, head, unreg_list)
5233 static void rollback_registered(struct net_device *dev)
5237 list_add(&dev->unreg_list, &single);
5238 rollback_registered_many(&single);
5242 static netdev_features_t netdev_fix_features(struct net_device *dev,
5243 netdev_features_t features)
5245 /* Fix illegal checksum combinations */
5246 if ((features & NETIF_F_HW_CSUM) &&
5247 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5248 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5249 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5252 /* Fix illegal SG+CSUM combinations. */
5253 if ((features & NETIF_F_SG) &&
5254 !(features & NETIF_F_ALL_CSUM)) {
5256 "Dropping NETIF_F_SG since no checksum feature.\n");
5257 features &= ~NETIF_F_SG;
5260 /* TSO requires that SG is present as well. */
5261 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5262 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5263 features &= ~NETIF_F_ALL_TSO;
5266 /* TSO ECN requires that TSO is present as well. */
5267 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5268 features &= ~NETIF_F_TSO_ECN;
5270 /* Software GSO depends on SG. */
5271 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5272 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5273 features &= ~NETIF_F_GSO;
5276 /* UFO needs SG and checksumming */
5277 if (features & NETIF_F_UFO) {
5278 /* maybe split UFO into V4 and V6? */
5279 if (!((features & NETIF_F_GEN_CSUM) ||
5280 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5281 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5283 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5284 features &= ~NETIF_F_UFO;
5287 if (!(features & NETIF_F_SG)) {
5289 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5290 features &= ~NETIF_F_UFO;
5297 int __netdev_update_features(struct net_device *dev)
5299 netdev_features_t features;
5304 features = netdev_get_wanted_features(dev);
5306 if (dev->netdev_ops->ndo_fix_features)
5307 features = dev->netdev_ops->ndo_fix_features(dev, features);
5309 /* driver might be less strict about feature dependencies */
5310 features = netdev_fix_features(dev, features);
5312 if (dev->features == features)
5315 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5316 &dev->features, &features);
5318 if (dev->netdev_ops->ndo_set_features)
5319 err = dev->netdev_ops->ndo_set_features(dev, features);
5321 if (unlikely(err < 0)) {
5323 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5324 err, &features, &dev->features);
5329 dev->features = features;
5335 * netdev_update_features - recalculate device features
5336 * @dev: the device to check
5338 * Recalculate dev->features set and send notifications if it
5339 * has changed. Should be called after driver or hardware dependent
5340 * conditions might have changed that influence the features.
5342 void netdev_update_features(struct net_device *dev)
5344 if (__netdev_update_features(dev))
5345 netdev_features_change(dev);
5347 EXPORT_SYMBOL(netdev_update_features);
5350 * netdev_change_features - recalculate device features
5351 * @dev: the device to check
5353 * Recalculate dev->features set and send notifications even
5354 * if they have not changed. Should be called instead of
5355 * netdev_update_features() if also dev->vlan_features might
5356 * have changed to allow the changes to be propagated to stacked
5359 void netdev_change_features(struct net_device *dev)
5361 __netdev_update_features(dev);
5362 netdev_features_change(dev);
5364 EXPORT_SYMBOL(netdev_change_features);
5367 * netif_stacked_transfer_operstate - transfer operstate
5368 * @rootdev: the root or lower level device to transfer state from
5369 * @dev: the device to transfer operstate to
5371 * Transfer operational state from root to device. This is normally
5372 * called when a stacking relationship exists between the root
5373 * device and the device(a leaf device).
5375 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5376 struct net_device *dev)
5378 if (rootdev->operstate == IF_OPER_DORMANT)
5379 netif_dormant_on(dev);
5381 netif_dormant_off(dev);
5383 if (netif_carrier_ok(rootdev)) {
5384 if (!netif_carrier_ok(dev))
5385 netif_carrier_on(dev);
5387 if (netif_carrier_ok(dev))
5388 netif_carrier_off(dev);
5391 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5394 static int netif_alloc_rx_queues(struct net_device *dev)
5396 unsigned int i, count = dev->num_rx_queues;
5397 struct netdev_rx_queue *rx;
5401 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5403 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5408 for (i = 0; i < count; i++)
5414 static void netdev_init_one_queue(struct net_device *dev,
5415 struct netdev_queue *queue, void *_unused)
5417 /* Initialize queue lock */
5418 spin_lock_init(&queue->_xmit_lock);
5419 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5420 queue->xmit_lock_owner = -1;
5421 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5424 dql_init(&queue->dql, HZ);
5428 static int netif_alloc_netdev_queues(struct net_device *dev)
5430 unsigned int count = dev->num_tx_queues;
5431 struct netdev_queue *tx;
5435 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5437 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5442 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5443 spin_lock_init(&dev->tx_global_lock);
5449 * register_netdevice - register a network device
5450 * @dev: device to register
5452 * Take a completed network device structure and add it to the kernel
5453 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5454 * chain. 0 is returned on success. A negative errno code is returned
5455 * on a failure to set up the device, or if the name is a duplicate.
5457 * Callers must hold the rtnl semaphore. You may want
5458 * register_netdev() instead of this.
5461 * The locking appears insufficient to guarantee two parallel registers
5462 * will not get the same name.
5465 int register_netdevice(struct net_device *dev)
5468 struct net *net = dev_net(dev);
5470 BUG_ON(dev_boot_phase);
5475 /* When net_device's are persistent, this will be fatal. */
5476 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5479 spin_lock_init(&dev->addr_list_lock);
5480 netdev_set_addr_lockdep_class(dev);
5484 ret = dev_get_valid_name(dev, dev->name);
5488 /* Init, if this function is available */
5489 if (dev->netdev_ops->ndo_init) {
5490 ret = dev->netdev_ops->ndo_init(dev);
5498 dev->ifindex = dev_new_index(net);
5499 if (dev->iflink == -1)
5500 dev->iflink = dev->ifindex;
5502 /* Transfer changeable features to wanted_features and enable
5503 * software offloads (GSO and GRO).
5505 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5506 dev->features |= NETIF_F_SOFT_FEATURES;
5507 dev->wanted_features = dev->features & dev->hw_features;
5509 /* Turn on no cache copy if HW is doing checksum */
5510 if (!(dev->flags & IFF_LOOPBACK)) {
5511 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5512 if (dev->features & NETIF_F_ALL_CSUM) {
5513 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5514 dev->features |= NETIF_F_NOCACHE_COPY;
5518 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5520 dev->vlan_features |= NETIF_F_HIGHDMA;
5522 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5523 ret = notifier_to_errno(ret);
5527 ret = netdev_register_kobject(dev);
5530 dev->reg_state = NETREG_REGISTERED;
5532 __netdev_update_features(dev);
5535 * Default initial state at registry is that the
5536 * device is present.
5539 set_bit(__LINK_STATE_PRESENT, &dev->state);
5541 dev_init_scheduler(dev);
5543 list_netdevice(dev);
5545 /* Notify protocols, that a new device appeared. */
5546 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5547 ret = notifier_to_errno(ret);
5549 rollback_registered(dev);
5550 dev->reg_state = NETREG_UNREGISTERED;
5553 * Prevent userspace races by waiting until the network
5554 * device is fully setup before sending notifications.
5556 if (!dev->rtnl_link_ops ||
5557 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5558 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5564 if (dev->netdev_ops->ndo_uninit)
5565 dev->netdev_ops->ndo_uninit(dev);
5568 EXPORT_SYMBOL(register_netdevice);
5571 * init_dummy_netdev - init a dummy network device for NAPI
5572 * @dev: device to init
5574 * This takes a network device structure and initialize the minimum
5575 * amount of fields so it can be used to schedule NAPI polls without
5576 * registering a full blown interface. This is to be used by drivers
5577 * that need to tie several hardware interfaces to a single NAPI
5578 * poll scheduler due to HW limitations.
5580 int init_dummy_netdev(struct net_device *dev)
5582 /* Clear everything. Note we don't initialize spinlocks
5583 * are they aren't supposed to be taken by any of the
5584 * NAPI code and this dummy netdev is supposed to be
5585 * only ever used for NAPI polls
5587 memset(dev, 0, sizeof(struct net_device));
5589 /* make sure we BUG if trying to hit standard
5590 * register/unregister code path
5592 dev->reg_state = NETREG_DUMMY;
5594 /* NAPI wants this */
5595 INIT_LIST_HEAD(&dev->napi_list);
5597 /* a dummy interface is started by default */
5598 set_bit(__LINK_STATE_PRESENT, &dev->state);
5599 set_bit(__LINK_STATE_START, &dev->state);
5601 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5602 * because users of this 'device' dont need to change
5608 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5612 * register_netdev - register a network device
5613 * @dev: device to register
5615 * Take a completed network device structure and add it to the kernel
5616 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5617 * chain. 0 is returned on success. A negative errno code is returned
5618 * on a failure to set up the device, or if the name is a duplicate.
5620 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5621 * and expands the device name if you passed a format string to
5624 int register_netdev(struct net_device *dev)
5629 err = register_netdevice(dev);
5633 EXPORT_SYMBOL(register_netdev);
5635 int netdev_refcnt_read(const struct net_device *dev)
5639 for_each_possible_cpu(i)
5640 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5643 EXPORT_SYMBOL(netdev_refcnt_read);
5646 * netdev_wait_allrefs - wait until all references are gone.
5648 * This is called when unregistering network devices.
5650 * Any protocol or device that holds a reference should register
5651 * for netdevice notification, and cleanup and put back the
5652 * reference if they receive an UNREGISTER event.
5653 * We can get stuck here if buggy protocols don't correctly
5656 static void netdev_wait_allrefs(struct net_device *dev)
5658 unsigned long rebroadcast_time, warning_time;
5661 linkwatch_forget_dev(dev);
5663 rebroadcast_time = warning_time = jiffies;
5664 refcnt = netdev_refcnt_read(dev);
5666 while (refcnt != 0) {
5667 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5670 /* Rebroadcast unregister notification */
5671 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5672 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5673 * should have already handle it the first time */
5675 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5677 /* We must not have linkwatch events
5678 * pending on unregister. If this
5679 * happens, we simply run the queue
5680 * unscheduled, resulting in a noop
5683 linkwatch_run_queue();
5688 rebroadcast_time = jiffies;
5693 refcnt = netdev_refcnt_read(dev);
5695 if (time_after(jiffies, warning_time + 10 * HZ)) {
5696 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5698 warning_time = jiffies;
5707 * register_netdevice(x1);
5708 * register_netdevice(x2);
5710 * unregister_netdevice(y1);
5711 * unregister_netdevice(y2);
5717 * We are invoked by rtnl_unlock().
5718 * This allows us to deal with problems:
5719 * 1) We can delete sysfs objects which invoke hotplug
5720 * without deadlocking with linkwatch via keventd.
5721 * 2) Since we run with the RTNL semaphore not held, we can sleep
5722 * safely in order to wait for the netdev refcnt to drop to zero.
5724 * We must not return until all unregister events added during
5725 * the interval the lock was held have been completed.
5727 void netdev_run_todo(void)
5729 struct list_head list;
5731 /* Snapshot list, allow later requests */
5732 list_replace_init(&net_todo_list, &list);
5736 /* Wait for rcu callbacks to finish before attempting to drain
5737 * the device list. This usually avoids a 250ms wait.
5739 if (!list_empty(&list))
5742 while (!list_empty(&list)) {
5743 struct net_device *dev
5744 = list_first_entry(&list, struct net_device, todo_list);
5745 list_del(&dev->todo_list);
5747 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5748 pr_err("network todo '%s' but state %d\n",
5749 dev->name, dev->reg_state);
5754 dev->reg_state = NETREG_UNREGISTERED;
5756 on_each_cpu(flush_backlog, dev, 1);
5758 netdev_wait_allrefs(dev);
5761 BUG_ON(netdev_refcnt_read(dev));
5762 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5763 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5764 WARN_ON(dev->dn_ptr);
5766 if (dev->destructor)
5767 dev->destructor(dev);
5769 /* Free network device */
5770 kobject_put(&dev->dev.kobj);
5774 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5775 * fields in the same order, with only the type differing.
5777 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5778 const struct net_device_stats *netdev_stats)
5780 #if BITS_PER_LONG == 64
5781 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5782 memcpy(stats64, netdev_stats, sizeof(*stats64));
5784 size_t i, n = sizeof(*stats64) / sizeof(u64);
5785 const unsigned long *src = (const unsigned long *)netdev_stats;
5786 u64 *dst = (u64 *)stats64;
5788 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5789 sizeof(*stats64) / sizeof(u64));
5790 for (i = 0; i < n; i++)
5794 EXPORT_SYMBOL(netdev_stats_to_stats64);
5797 * dev_get_stats - get network device statistics
5798 * @dev: device to get statistics from
5799 * @storage: place to store stats
5801 * Get network statistics from device. Return @storage.
5802 * The device driver may provide its own method by setting
5803 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5804 * otherwise the internal statistics structure is used.
5806 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5807 struct rtnl_link_stats64 *storage)
5809 const struct net_device_ops *ops = dev->netdev_ops;
5811 if (ops->ndo_get_stats64) {
5812 memset(storage, 0, sizeof(*storage));
5813 ops->ndo_get_stats64(dev, storage);
5814 } else if (ops->ndo_get_stats) {
5815 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5817 netdev_stats_to_stats64(storage, &dev->stats);
5819 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5822 EXPORT_SYMBOL(dev_get_stats);
5824 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5826 struct netdev_queue *queue = dev_ingress_queue(dev);
5828 #ifdef CONFIG_NET_CLS_ACT
5831 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5834 netdev_init_one_queue(dev, queue, NULL);
5835 queue->qdisc = &noop_qdisc;
5836 queue->qdisc_sleeping = &noop_qdisc;
5837 rcu_assign_pointer(dev->ingress_queue, queue);
5843 * alloc_netdev_mqs - allocate network device
5844 * @sizeof_priv: size of private data to allocate space for
5845 * @name: device name format string
5846 * @setup: callback to initialize device
5847 * @txqs: the number of TX subqueues to allocate
5848 * @rxqs: the number of RX subqueues to allocate
5850 * Allocates a struct net_device with private data area for driver use
5851 * and performs basic initialization. Also allocates subquue structs
5852 * for each queue on the device.
5854 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5855 void (*setup)(struct net_device *),
5856 unsigned int txqs, unsigned int rxqs)
5858 struct net_device *dev;
5860 struct net_device *p;
5862 BUG_ON(strlen(name) >= sizeof(dev->name));
5865 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5871 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5876 alloc_size = sizeof(struct net_device);
5878 /* ensure 32-byte alignment of private area */
5879 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5880 alloc_size += sizeof_priv;
5882 /* ensure 32-byte alignment of whole construct */
5883 alloc_size += NETDEV_ALIGN - 1;
5885 p = kzalloc(alloc_size, GFP_KERNEL);
5887 pr_err("alloc_netdev: Unable to allocate device\n");
5891 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5892 dev->padded = (char *)dev - (char *)p;
5894 dev->pcpu_refcnt = alloc_percpu(int);
5895 if (!dev->pcpu_refcnt)
5898 if (dev_addr_init(dev))
5904 dev_net_set(dev, &init_net);
5906 dev->gso_max_size = GSO_MAX_SIZE;
5908 INIT_LIST_HEAD(&dev->napi_list);
5909 INIT_LIST_HEAD(&dev->unreg_list);
5910 INIT_LIST_HEAD(&dev->link_watch_list);
5911 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5914 dev->num_tx_queues = txqs;
5915 dev->real_num_tx_queues = txqs;
5916 if (netif_alloc_netdev_queues(dev))
5920 dev->num_rx_queues = rxqs;
5921 dev->real_num_rx_queues = rxqs;
5922 if (netif_alloc_rx_queues(dev))
5926 strcpy(dev->name, name);
5927 dev->group = INIT_NETDEV_GROUP;
5935 free_percpu(dev->pcpu_refcnt);
5945 EXPORT_SYMBOL(alloc_netdev_mqs);
5948 * free_netdev - free network device
5951 * This function does the last stage of destroying an allocated device
5952 * interface. The reference to the device object is released.
5953 * If this is the last reference then it will be freed.
5955 void free_netdev(struct net_device *dev)
5957 struct napi_struct *p, *n;
5959 release_net(dev_net(dev));
5966 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5968 /* Flush device addresses */
5969 dev_addr_flush(dev);
5971 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5974 free_percpu(dev->pcpu_refcnt);
5975 dev->pcpu_refcnt = NULL;
5977 /* Compatibility with error handling in drivers */
5978 if (dev->reg_state == NETREG_UNINITIALIZED) {
5979 kfree((char *)dev - dev->padded);
5983 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5984 dev->reg_state = NETREG_RELEASED;
5986 /* will free via device release */
5987 put_device(&dev->dev);
5989 EXPORT_SYMBOL(free_netdev);
5992 * synchronize_net - Synchronize with packet receive processing
5994 * Wait for packets currently being received to be done.
5995 * Does not block later packets from starting.
5997 void synchronize_net(void)
6000 if (rtnl_is_locked())
6001 synchronize_rcu_expedited();
6005 EXPORT_SYMBOL(synchronize_net);
6008 * unregister_netdevice_queue - remove device from the kernel
6012 * This function shuts down a device interface and removes it
6013 * from the kernel tables.
6014 * If head not NULL, device is queued to be unregistered later.
6016 * Callers must hold the rtnl semaphore. You may want
6017 * unregister_netdev() instead of this.
6020 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6025 list_move_tail(&dev->unreg_list, head);
6027 rollback_registered(dev);
6028 /* Finish processing unregister after unlock */
6032 EXPORT_SYMBOL(unregister_netdevice_queue);
6035 * unregister_netdevice_many - unregister many devices
6036 * @head: list of devices
6038 void unregister_netdevice_many(struct list_head *head)
6040 struct net_device *dev;
6042 if (!list_empty(head)) {
6043 rollback_registered_many(head);
6044 list_for_each_entry(dev, head, unreg_list)
6048 EXPORT_SYMBOL(unregister_netdevice_many);
6051 * unregister_netdev - remove device from the kernel
6054 * This function shuts down a device interface and removes it
6055 * from the kernel tables.
6057 * This is just a wrapper for unregister_netdevice that takes
6058 * the rtnl semaphore. In general you want to use this and not
6059 * unregister_netdevice.
6061 void unregister_netdev(struct net_device *dev)
6064 unregister_netdevice(dev);
6067 EXPORT_SYMBOL(unregister_netdev);
6070 * dev_change_net_namespace - move device to different nethost namespace
6072 * @net: network namespace
6073 * @pat: If not NULL name pattern to try if the current device name
6074 * is already taken in the destination network namespace.
6076 * This function shuts down a device interface and moves it
6077 * to a new network namespace. On success 0 is returned, on
6078 * a failure a netagive errno code is returned.
6080 * Callers must hold the rtnl semaphore.
6083 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6089 /* Don't allow namespace local devices to be moved. */
6091 if (dev->features & NETIF_F_NETNS_LOCAL)
6094 /* Ensure the device has been registrered */
6096 if (dev->reg_state != NETREG_REGISTERED)
6099 /* Get out if there is nothing todo */
6101 if (net_eq(dev_net(dev), net))
6104 /* Pick the destination device name, and ensure
6105 * we can use it in the destination network namespace.
6108 if (__dev_get_by_name(net, dev->name)) {
6109 /* We get here if we can't use the current device name */
6112 if (dev_get_valid_name(dev, pat) < 0)
6117 * And now a mini version of register_netdevice unregister_netdevice.
6120 /* If device is running close it first. */
6123 /* And unlink it from device chain */
6125 unlist_netdevice(dev);
6129 /* Shutdown queueing discipline. */
6132 /* Notify protocols, that we are about to destroy
6133 this device. They should clean all the things.
6135 Note that dev->reg_state stays at NETREG_REGISTERED.
6136 This is wanted because this way 8021q and macvlan know
6137 the device is just moving and can keep their slaves up.
6139 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6140 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6141 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6144 * Flush the unicast and multicast chains
6149 /* Actually switch the network namespace */
6150 dev_net_set(dev, net);
6152 /* If there is an ifindex conflict assign a new one */
6153 if (__dev_get_by_index(net, dev->ifindex)) {
6154 int iflink = (dev->iflink == dev->ifindex);
6155 dev->ifindex = dev_new_index(net);
6157 dev->iflink = dev->ifindex;
6160 /* Fixup kobjects */
6161 err = device_rename(&dev->dev, dev->name);
6164 /* Add the device back in the hashes */
6165 list_netdevice(dev);
6167 /* Notify protocols, that a new device appeared. */
6168 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6171 * Prevent userspace races by waiting until the network
6172 * device is fully setup before sending notifications.
6174 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6181 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6183 static int dev_cpu_callback(struct notifier_block *nfb,
6184 unsigned long action,
6187 struct sk_buff **list_skb;
6188 struct sk_buff *skb;
6189 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6190 struct softnet_data *sd, *oldsd;
6192 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6195 local_irq_disable();
6196 cpu = smp_processor_id();
6197 sd = &per_cpu(softnet_data, cpu);
6198 oldsd = &per_cpu(softnet_data, oldcpu);
6200 /* Find end of our completion_queue. */
6201 list_skb = &sd->completion_queue;
6203 list_skb = &(*list_skb)->next;
6204 /* Append completion queue from offline CPU. */
6205 *list_skb = oldsd->completion_queue;
6206 oldsd->completion_queue = NULL;
6208 /* Append output queue from offline CPU. */
6209 if (oldsd->output_queue) {
6210 *sd->output_queue_tailp = oldsd->output_queue;
6211 sd->output_queue_tailp = oldsd->output_queue_tailp;
6212 oldsd->output_queue = NULL;
6213 oldsd->output_queue_tailp = &oldsd->output_queue;
6215 /* Append NAPI poll list from offline CPU. */
6216 if (!list_empty(&oldsd->poll_list)) {
6217 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6218 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6221 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6224 /* Process offline CPU's input_pkt_queue */
6225 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6227 input_queue_head_incr(oldsd);
6229 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6231 input_queue_head_incr(oldsd);
6239 * netdev_increment_features - increment feature set by one
6240 * @all: current feature set
6241 * @one: new feature set
6242 * @mask: mask feature set
6244 * Computes a new feature set after adding a device with feature set
6245 * @one to the master device with current feature set @all. Will not
6246 * enable anything that is off in @mask. Returns the new feature set.
6248 netdev_features_t netdev_increment_features(netdev_features_t all,
6249 netdev_features_t one, netdev_features_t mask)
6251 if (mask & NETIF_F_GEN_CSUM)
6252 mask |= NETIF_F_ALL_CSUM;
6253 mask |= NETIF_F_VLAN_CHALLENGED;
6255 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6256 all &= one | ~NETIF_F_ALL_FOR_ALL;
6258 /* If one device supports hw checksumming, set for all. */
6259 if (all & NETIF_F_GEN_CSUM)
6260 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6264 EXPORT_SYMBOL(netdev_increment_features);
6266 static struct hlist_head *netdev_create_hash(void)
6269 struct hlist_head *hash;
6271 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6273 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6274 INIT_HLIST_HEAD(&hash[i]);
6279 /* Initialize per network namespace state */
6280 static int __net_init netdev_init(struct net *net)
6282 INIT_LIST_HEAD(&net->dev_base_head);
6284 net->dev_name_head = netdev_create_hash();
6285 if (net->dev_name_head == NULL)
6288 net->dev_index_head = netdev_create_hash();
6289 if (net->dev_index_head == NULL)
6295 kfree(net->dev_name_head);
6301 * netdev_drivername - network driver for the device
6302 * @dev: network device
6304 * Determine network driver for device.
6306 const char *netdev_drivername(const struct net_device *dev)
6308 const struct device_driver *driver;
6309 const struct device *parent;
6310 const char *empty = "";
6312 parent = dev->dev.parent;
6316 driver = parent->driver;
6317 if (driver && driver->name)
6318 return driver->name;
6322 int __netdev_printk(const char *level, const struct net_device *dev,
6323 struct va_format *vaf)
6327 if (dev && dev->dev.parent)
6328 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6329 netdev_name(dev), vaf);
6331 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6333 r = printk("%s(NULL net_device): %pV", level, vaf);
6337 EXPORT_SYMBOL(__netdev_printk);
6339 int netdev_printk(const char *level, const struct net_device *dev,
6340 const char *format, ...)
6342 struct va_format vaf;
6346 va_start(args, format);
6351 r = __netdev_printk(level, dev, &vaf);
6356 EXPORT_SYMBOL(netdev_printk);
6358 #define define_netdev_printk_level(func, level) \
6359 int func(const struct net_device *dev, const char *fmt, ...) \
6362 struct va_format vaf; \
6365 va_start(args, fmt); \
6370 r = __netdev_printk(level, dev, &vaf); \
6375 EXPORT_SYMBOL(func);
6377 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6378 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6379 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6380 define_netdev_printk_level(netdev_err, KERN_ERR);
6381 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6382 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6383 define_netdev_printk_level(netdev_info, KERN_INFO);
6385 static void __net_exit netdev_exit(struct net *net)
6387 kfree(net->dev_name_head);
6388 kfree(net->dev_index_head);
6391 static struct pernet_operations __net_initdata netdev_net_ops = {
6392 .init = netdev_init,
6393 .exit = netdev_exit,
6396 static void __net_exit default_device_exit(struct net *net)
6398 struct net_device *dev, *aux;
6400 * Push all migratable network devices back to the
6401 * initial network namespace
6404 for_each_netdev_safe(net, dev, aux) {
6406 char fb_name[IFNAMSIZ];
6408 /* Ignore unmoveable devices (i.e. loopback) */
6409 if (dev->features & NETIF_F_NETNS_LOCAL)
6412 /* Leave virtual devices for the generic cleanup */
6413 if (dev->rtnl_link_ops)
6416 /* Push remaining network devices to init_net */
6417 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6418 err = dev_change_net_namespace(dev, &init_net, fb_name);
6420 pr_emerg("%s: failed to move %s to init_net: %d\n",
6421 __func__, dev->name, err);
6428 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6430 /* At exit all network devices most be removed from a network
6431 * namespace. Do this in the reverse order of registration.
6432 * Do this across as many network namespaces as possible to
6433 * improve batching efficiency.
6435 struct net_device *dev;
6437 LIST_HEAD(dev_kill_list);
6440 list_for_each_entry(net, net_list, exit_list) {
6441 for_each_netdev_reverse(net, dev) {
6442 if (dev->rtnl_link_ops)
6443 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6445 unregister_netdevice_queue(dev, &dev_kill_list);
6448 unregister_netdevice_many(&dev_kill_list);
6449 list_del(&dev_kill_list);
6453 static struct pernet_operations __net_initdata default_device_ops = {
6454 .exit = default_device_exit,
6455 .exit_batch = default_device_exit_batch,
6459 * Initialize the DEV module. At boot time this walks the device list and
6460 * unhooks any devices that fail to initialise (normally hardware not
6461 * present) and leaves us with a valid list of present and active devices.
6466 * This is called single threaded during boot, so no need
6467 * to take the rtnl semaphore.
6469 static int __init net_dev_init(void)
6471 int i, rc = -ENOMEM;
6473 BUG_ON(!dev_boot_phase);
6475 if (dev_proc_init())
6478 if (netdev_kobject_init())
6481 INIT_LIST_HEAD(&ptype_all);
6482 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6483 INIT_LIST_HEAD(&ptype_base[i]);
6485 if (register_pernet_subsys(&netdev_net_ops))
6489 * Initialise the packet receive queues.
6492 for_each_possible_cpu(i) {
6493 struct softnet_data *sd = &per_cpu(softnet_data, i);
6495 memset(sd, 0, sizeof(*sd));
6496 skb_queue_head_init(&sd->input_pkt_queue);
6497 skb_queue_head_init(&sd->process_queue);
6498 sd->completion_queue = NULL;
6499 INIT_LIST_HEAD(&sd->poll_list);
6500 sd->output_queue = NULL;
6501 sd->output_queue_tailp = &sd->output_queue;
6503 sd->csd.func = rps_trigger_softirq;
6509 sd->backlog.poll = process_backlog;
6510 sd->backlog.weight = weight_p;
6511 sd->backlog.gro_list = NULL;
6512 sd->backlog.gro_count = 0;
6517 /* The loopback device is special if any other network devices
6518 * is present in a network namespace the loopback device must
6519 * be present. Since we now dynamically allocate and free the
6520 * loopback device ensure this invariant is maintained by
6521 * keeping the loopback device as the first device on the
6522 * list of network devices. Ensuring the loopback devices
6523 * is the first device that appears and the last network device
6526 if (register_pernet_device(&loopback_net_ops))
6529 if (register_pernet_device(&default_device_ops))
6532 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6533 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6535 hotcpu_notifier(dev_cpu_callback, 0);
6543 subsys_initcall(net_dev_init);
6545 static int __init initialize_hashrnd(void)
6547 get_random_bytes(&hashrnd, sizeof(hashrnd));
6551 late_initcall_sync(initialize_hashrnd);