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 hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
212 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
217 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
220 static inline void rps_lock(struct softnet_data *sd)
223 spin_lock(&sd->input_pkt_queue.lock);
227 static inline void rps_unlock(struct softnet_data *sd)
230 spin_unlock(&sd->input_pkt_queue.lock);
234 /* Device list insertion */
235 static int list_netdevice(struct net_device *dev)
237 struct net *net = dev_net(dev);
241 write_lock_bh(&dev_base_lock);
242 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
243 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
244 hlist_add_head_rcu(&dev->index_hlist,
245 dev_index_hash(net, dev->ifindex));
246 write_unlock_bh(&dev_base_lock);
248 dev_base_seq_inc(net);
253 /* Device list removal
254 * caller must respect a RCU grace period before freeing/reusing dev
256 static void unlist_netdevice(struct net_device *dev)
260 /* Unlink dev from the device chain */
261 write_lock_bh(&dev_base_lock);
262 list_del_rcu(&dev->dev_list);
263 hlist_del_rcu(&dev->name_hlist);
264 hlist_del_rcu(&dev->index_hlist);
265 write_unlock_bh(&dev_base_lock);
267 dev_base_seq_inc(dev_net(dev));
274 static RAW_NOTIFIER_HEAD(netdev_chain);
277 * Device drivers call our routines to queue packets here. We empty the
278 * queue in the local softnet handler.
281 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
282 EXPORT_PER_CPU_SYMBOL(softnet_data);
284 #ifdef CONFIG_LOCKDEP
286 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
287 * according to dev->type
289 static const unsigned short netdev_lock_type[] =
290 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
291 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
292 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
293 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
294 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
295 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
296 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
297 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
298 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
299 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
300 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
301 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
302 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
303 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
304 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
305 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_IEEE802_TR", "_xmit_IEEE80211",
321 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
322 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
323 "_xmit_VOID", "_xmit_NONE"};
325 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
326 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
328 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
332 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
333 if (netdev_lock_type[i] == dev_type)
335 /* the last key is used by default */
336 return ARRAY_SIZE(netdev_lock_type) - 1;
339 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
340 unsigned short dev_type)
344 i = netdev_lock_pos(dev_type);
345 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
346 netdev_lock_name[i]);
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
353 i = netdev_lock_pos(dev->type);
354 lockdep_set_class_and_name(&dev->addr_list_lock,
355 &netdev_addr_lock_key[i],
356 netdev_lock_name[i]);
359 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
360 unsigned short dev_type)
363 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
368 /*******************************************************************************
370 Protocol management and registration routines
372 *******************************************************************************/
375 * Add a protocol ID to the list. Now that the input handler is
376 * smarter we can dispense with all the messy stuff that used to be
379 * BEWARE!!! Protocol handlers, mangling input packets,
380 * MUST BE last in hash buckets and checking protocol handlers
381 * MUST start from promiscuous ptype_all chain in net_bh.
382 * It is true now, do not change it.
383 * Explanation follows: if protocol handler, mangling packet, will
384 * be the first on list, it is not able to sense, that packet
385 * is cloned and should be copied-on-write, so that it will
386 * change it and subsequent readers will get broken packet.
390 static inline struct list_head *ptype_head(const struct packet_type *pt)
392 if (pt->type == htons(ETH_P_ALL))
395 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
399 * dev_add_pack - add packet handler
400 * @pt: packet type declaration
402 * Add a protocol handler to the networking stack. The passed &packet_type
403 * is linked into kernel lists and may not be freed until it has been
404 * removed from the kernel lists.
406 * This call does not sleep therefore it can not
407 * guarantee all CPU's that are in middle of receiving packets
408 * will see the new packet type (until the next received packet).
411 void dev_add_pack(struct packet_type *pt)
413 struct list_head *head = ptype_head(pt);
415 spin_lock(&ptype_lock);
416 list_add_rcu(&pt->list, head);
417 spin_unlock(&ptype_lock);
419 EXPORT_SYMBOL(dev_add_pack);
422 * __dev_remove_pack - remove packet handler
423 * @pt: packet type declaration
425 * Remove a protocol handler that was previously added to the kernel
426 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
427 * from the kernel lists and can be freed or reused once this function
430 * The packet type might still be in use by receivers
431 * and must not be freed until after all the CPU's have gone
432 * through a quiescent state.
434 void __dev_remove_pack(struct packet_type *pt)
436 struct list_head *head = ptype_head(pt);
437 struct packet_type *pt1;
439 spin_lock(&ptype_lock);
441 list_for_each_entry(pt1, head, list) {
443 list_del_rcu(&pt->list);
448 pr_warn("dev_remove_pack: %p not found\n", pt);
450 spin_unlock(&ptype_lock);
452 EXPORT_SYMBOL(__dev_remove_pack);
455 * dev_remove_pack - remove packet handler
456 * @pt: packet type declaration
458 * Remove a protocol handler that was previously added to the kernel
459 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
460 * from the kernel lists and can be freed or reused once this function
463 * This call sleeps to guarantee that no CPU is looking at the packet
466 void dev_remove_pack(struct packet_type *pt)
468 __dev_remove_pack(pt);
472 EXPORT_SYMBOL(dev_remove_pack);
474 /******************************************************************************
476 Device Boot-time Settings Routines
478 *******************************************************************************/
480 /* Boot time configuration table */
481 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
484 * netdev_boot_setup_add - add new setup entry
485 * @name: name of the device
486 * @map: configured settings for the device
488 * Adds new setup entry to the dev_boot_setup list. The function
489 * returns 0 on error and 1 on success. This is a generic routine to
492 static int netdev_boot_setup_add(char *name, struct ifmap *map)
494 struct netdev_boot_setup *s;
498 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
499 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
500 memset(s[i].name, 0, sizeof(s[i].name));
501 strlcpy(s[i].name, name, IFNAMSIZ);
502 memcpy(&s[i].map, map, sizeof(s[i].map));
507 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
511 * netdev_boot_setup_check - check boot time settings
512 * @dev: the netdevice
514 * Check boot time settings for the device.
515 * The found settings are set for the device to be used
516 * later in the device probing.
517 * Returns 0 if no settings found, 1 if they are.
519 int netdev_boot_setup_check(struct net_device *dev)
521 struct netdev_boot_setup *s = dev_boot_setup;
524 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
525 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
526 !strcmp(dev->name, s[i].name)) {
527 dev->irq = s[i].map.irq;
528 dev->base_addr = s[i].map.base_addr;
529 dev->mem_start = s[i].map.mem_start;
530 dev->mem_end = s[i].map.mem_end;
536 EXPORT_SYMBOL(netdev_boot_setup_check);
540 * netdev_boot_base - get address from boot time settings
541 * @prefix: prefix for network device
542 * @unit: id for network device
544 * Check boot time settings for the base address of device.
545 * The found settings are set for the device to be used
546 * later in the device probing.
547 * Returns 0 if no settings found.
549 unsigned long netdev_boot_base(const char *prefix, int unit)
551 const struct netdev_boot_setup *s = dev_boot_setup;
555 sprintf(name, "%s%d", prefix, unit);
558 * If device already registered then return base of 1
559 * to indicate not to probe for this interface
561 if (__dev_get_by_name(&init_net, name))
564 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
565 if (!strcmp(name, s[i].name))
566 return s[i].map.base_addr;
571 * Saves at boot time configured settings for any netdevice.
573 int __init netdev_boot_setup(char *str)
578 str = get_options(str, ARRAY_SIZE(ints), ints);
583 memset(&map, 0, sizeof(map));
587 map.base_addr = ints[2];
589 map.mem_start = ints[3];
591 map.mem_end = ints[4];
593 /* Add new entry to the list */
594 return netdev_boot_setup_add(str, &map);
597 __setup("netdev=", netdev_boot_setup);
599 /*******************************************************************************
601 Device Interface Subroutines
603 *******************************************************************************/
606 * __dev_get_by_name - find a device by its name
607 * @net: the applicable net namespace
608 * @name: name to find
610 * Find an interface by name. Must be called under RTNL semaphore
611 * or @dev_base_lock. If the name is found a pointer to the device
612 * is returned. If the name is not found then %NULL is returned. The
613 * reference counters are not incremented so the caller must be
614 * careful with locks.
617 struct net_device *__dev_get_by_name(struct net *net, const char *name)
619 struct hlist_node *p;
620 struct net_device *dev;
621 struct hlist_head *head = dev_name_hash(net, name);
623 hlist_for_each_entry(dev, p, head, name_hlist)
624 if (!strncmp(dev->name, name, IFNAMSIZ))
629 EXPORT_SYMBOL(__dev_get_by_name);
632 * dev_get_by_name_rcu - find a device by its name
633 * @net: the applicable net namespace
634 * @name: name to find
636 * Find an interface by name.
637 * If the name is found a pointer to the device is returned.
638 * If the name is not found then %NULL is returned.
639 * The reference counters are not incremented so the caller must be
640 * careful with locks. The caller must hold RCU lock.
643 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
645 struct hlist_node *p;
646 struct net_device *dev;
647 struct hlist_head *head = dev_name_hash(net, name);
649 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
650 if (!strncmp(dev->name, name, IFNAMSIZ))
655 EXPORT_SYMBOL(dev_get_by_name_rcu);
658 * dev_get_by_name - find a device by its name
659 * @net: the applicable net namespace
660 * @name: name to find
662 * Find an interface by name. This can be called from any
663 * context and does its own locking. The returned handle has
664 * the usage count incremented and the caller must use dev_put() to
665 * release it when it is no longer needed. %NULL is returned if no
666 * matching device is found.
669 struct net_device *dev_get_by_name(struct net *net, const char *name)
671 struct net_device *dev;
674 dev = dev_get_by_name_rcu(net, name);
680 EXPORT_SYMBOL(dev_get_by_name);
683 * __dev_get_by_index - find a device by its ifindex
684 * @net: the applicable net namespace
685 * @ifindex: index of device
687 * Search for an interface by index. Returns %NULL if the device
688 * is not found or a pointer to the device. The device has not
689 * had its reference counter increased so the caller must be careful
690 * about locking. The caller must hold either the RTNL semaphore
694 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
696 struct hlist_node *p;
697 struct net_device *dev;
698 struct hlist_head *head = dev_index_hash(net, ifindex);
700 hlist_for_each_entry(dev, p, head, index_hlist)
701 if (dev->ifindex == ifindex)
706 EXPORT_SYMBOL(__dev_get_by_index);
709 * dev_get_by_index_rcu - find a device by its ifindex
710 * @net: the applicable net namespace
711 * @ifindex: index of device
713 * Search for an interface by index. Returns %NULL if the device
714 * is not found or a pointer to the device. The device has not
715 * had its reference counter increased so the caller must be careful
716 * about locking. The caller must hold RCU lock.
719 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
721 struct hlist_node *p;
722 struct net_device *dev;
723 struct hlist_head *head = dev_index_hash(net, ifindex);
725 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
726 if (dev->ifindex == ifindex)
731 EXPORT_SYMBOL(dev_get_by_index_rcu);
735 * dev_get_by_index - find a device by its ifindex
736 * @net: the applicable net namespace
737 * @ifindex: index of device
739 * Search for an interface by index. Returns NULL if the device
740 * is not found or a pointer to the device. The device returned has
741 * had a reference added and the pointer is safe until the user calls
742 * dev_put to indicate they have finished with it.
745 struct net_device *dev_get_by_index(struct net *net, int ifindex)
747 struct net_device *dev;
750 dev = dev_get_by_index_rcu(net, ifindex);
756 EXPORT_SYMBOL(dev_get_by_index);
759 * dev_getbyhwaddr_rcu - find a device by its hardware address
760 * @net: the applicable net namespace
761 * @type: media type of device
762 * @ha: hardware address
764 * Search for an interface by MAC address. Returns NULL if the device
765 * is not found or a pointer to the device.
766 * The caller must hold RCU or RTNL.
767 * The returned device has not had its ref count increased
768 * and the caller must therefore be careful about locking
772 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
775 struct net_device *dev;
777 for_each_netdev_rcu(net, dev)
778 if (dev->type == type &&
779 !memcmp(dev->dev_addr, ha, dev->addr_len))
784 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
786 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
788 struct net_device *dev;
791 for_each_netdev(net, dev)
792 if (dev->type == type)
797 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
799 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
801 struct net_device *dev, *ret = NULL;
804 for_each_netdev_rcu(net, dev)
805 if (dev->type == type) {
813 EXPORT_SYMBOL(dev_getfirstbyhwtype);
816 * dev_get_by_flags_rcu - find any device with given flags
817 * @net: the applicable net namespace
818 * @if_flags: IFF_* values
819 * @mask: bitmask of bits in if_flags to check
821 * Search for any interface with the given flags. Returns NULL if a device
822 * is not found or a pointer to the device. Must be called inside
823 * rcu_read_lock(), and result refcount is unchanged.
826 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
829 struct net_device *dev, *ret;
832 for_each_netdev_rcu(net, dev) {
833 if (((dev->flags ^ if_flags) & mask) == 0) {
840 EXPORT_SYMBOL(dev_get_by_flags_rcu);
843 * dev_valid_name - check if name is okay for network device
846 * Network device names need to be valid file names to
847 * to allow sysfs to work. We also disallow any kind of
850 bool dev_valid_name(const char *name)
854 if (strlen(name) >= IFNAMSIZ)
856 if (!strcmp(name, ".") || !strcmp(name, ".."))
860 if (*name == '/' || isspace(*name))
866 EXPORT_SYMBOL(dev_valid_name);
869 * __dev_alloc_name - allocate a name for a device
870 * @net: network namespace to allocate the device name in
871 * @name: name format string
872 * @buf: scratch buffer and result name string
874 * Passed a format string - eg "lt%d" it will try and find a suitable
875 * id. It scans list of devices to build up a free map, then chooses
876 * the first empty slot. The caller must hold the dev_base or rtnl lock
877 * while allocating the name and adding the device in order to avoid
879 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
880 * Returns the number of the unit assigned or a negative errno code.
883 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
887 const int max_netdevices = 8*PAGE_SIZE;
888 unsigned long *inuse;
889 struct net_device *d;
891 p = strnchr(name, IFNAMSIZ-1, '%');
894 * Verify the string as this thing may have come from
895 * the user. There must be either one "%d" and no other "%"
898 if (p[1] != 'd' || strchr(p + 2, '%'))
901 /* Use one page as a bit array of possible slots */
902 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
906 for_each_netdev(net, d) {
907 if (!sscanf(d->name, name, &i))
909 if (i < 0 || i >= max_netdevices)
912 /* avoid cases where sscanf is not exact inverse of printf */
913 snprintf(buf, IFNAMSIZ, name, i);
914 if (!strncmp(buf, d->name, IFNAMSIZ))
918 i = find_first_zero_bit(inuse, max_netdevices);
919 free_page((unsigned long) inuse);
923 snprintf(buf, IFNAMSIZ, name, i);
924 if (!__dev_get_by_name(net, buf))
927 /* It is possible to run out of possible slots
928 * when the name is long and there isn't enough space left
929 * for the digits, or if all bits are used.
935 * dev_alloc_name - allocate a name for a device
937 * @name: name format string
939 * Passed a format string - eg "lt%d" it will try and find a suitable
940 * id. It scans list of devices to build up a free map, then chooses
941 * the first empty slot. The caller must hold the dev_base or rtnl lock
942 * while allocating the name and adding the device in order to avoid
944 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
945 * Returns the number of the unit assigned or a negative errno code.
948 int dev_alloc_name(struct net_device *dev, const char *name)
954 BUG_ON(!dev_net(dev));
956 ret = __dev_alloc_name(net, name, buf);
958 strlcpy(dev->name, buf, IFNAMSIZ);
961 EXPORT_SYMBOL(dev_alloc_name);
963 static int dev_get_valid_name(struct net_device *dev, const char *name)
967 BUG_ON(!dev_net(dev));
970 if (!dev_valid_name(name))
973 if (strchr(name, '%'))
974 return dev_alloc_name(dev, name);
975 else if (__dev_get_by_name(net, name))
977 else if (dev->name != name)
978 strlcpy(dev->name, name, IFNAMSIZ);
984 * dev_change_name - change name of a device
986 * @newname: name (or format string) must be at least IFNAMSIZ
988 * Change name of a device, can pass format strings "eth%d".
991 int dev_change_name(struct net_device *dev, const char *newname)
993 char oldname[IFNAMSIZ];
999 BUG_ON(!dev_net(dev));
1002 if (dev->flags & IFF_UP)
1005 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1008 memcpy(oldname, dev->name, IFNAMSIZ);
1010 err = dev_get_valid_name(dev, newname);
1015 ret = device_rename(&dev->dev, dev->name);
1017 memcpy(dev->name, oldname, IFNAMSIZ);
1021 write_lock_bh(&dev_base_lock);
1022 hlist_del_rcu(&dev->name_hlist);
1023 write_unlock_bh(&dev_base_lock);
1027 write_lock_bh(&dev_base_lock);
1028 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1029 write_unlock_bh(&dev_base_lock);
1031 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1032 ret = notifier_to_errno(ret);
1035 /* err >= 0 after dev_alloc_name() or stores the first errno */
1038 memcpy(dev->name, oldname, IFNAMSIZ);
1041 pr_err("%s: name change rollback failed: %d\n",
1050 * dev_set_alias - change ifalias of a device
1052 * @alias: name up to IFALIASZ
1053 * @len: limit of bytes to copy from info
1055 * Set ifalias for a device,
1057 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1061 if (len >= IFALIASZ)
1066 kfree(dev->ifalias);
1067 dev->ifalias = NULL;
1072 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1076 strlcpy(dev->ifalias, alias, len+1);
1082 * netdev_features_change - device changes features
1083 * @dev: device to cause notification
1085 * Called to indicate a device has changed features.
1087 void netdev_features_change(struct net_device *dev)
1089 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1091 EXPORT_SYMBOL(netdev_features_change);
1094 * netdev_state_change - device changes state
1095 * @dev: device to cause notification
1097 * Called to indicate a device has changed state. This function calls
1098 * the notifier chains for netdev_chain and sends a NEWLINK message
1099 * to the routing socket.
1101 void netdev_state_change(struct net_device *dev)
1103 if (dev->flags & IFF_UP) {
1104 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1105 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1108 EXPORT_SYMBOL(netdev_state_change);
1110 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1112 return call_netdevice_notifiers(event, dev);
1114 EXPORT_SYMBOL(netdev_bonding_change);
1117 * dev_load - load a network module
1118 * @net: the applicable net namespace
1119 * @name: name of interface
1121 * If a network interface is not present and the process has suitable
1122 * privileges this function loads the module. If module loading is not
1123 * available in this kernel then it becomes a nop.
1126 void dev_load(struct net *net, const char *name)
1128 struct net_device *dev;
1132 dev = dev_get_by_name_rcu(net, name);
1136 if (no_module && capable(CAP_NET_ADMIN))
1137 no_module = request_module("netdev-%s", name);
1138 if (no_module && capable(CAP_SYS_MODULE)) {
1139 if (!request_module("%s", name))
1140 pr_err("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1144 EXPORT_SYMBOL(dev_load);
1146 static int __dev_open(struct net_device *dev)
1148 const struct net_device_ops *ops = dev->netdev_ops;
1153 if (!netif_device_present(dev))
1156 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1157 ret = notifier_to_errno(ret);
1161 set_bit(__LINK_STATE_START, &dev->state);
1163 if (ops->ndo_validate_addr)
1164 ret = ops->ndo_validate_addr(dev);
1166 if (!ret && ops->ndo_open)
1167 ret = ops->ndo_open(dev);
1170 clear_bit(__LINK_STATE_START, &dev->state);
1172 dev->flags |= IFF_UP;
1173 net_dmaengine_get();
1174 dev_set_rx_mode(dev);
1182 * dev_open - prepare an interface for use.
1183 * @dev: device to open
1185 * Takes a device from down to up state. The device's private open
1186 * function is invoked and then the multicast lists are loaded. Finally
1187 * the device is moved into the up state and a %NETDEV_UP message is
1188 * sent to the netdev notifier chain.
1190 * Calling this function on an active interface is a nop. On a failure
1191 * a negative errno code is returned.
1193 int dev_open(struct net_device *dev)
1197 if (dev->flags & IFF_UP)
1200 ret = __dev_open(dev);
1204 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1205 call_netdevice_notifiers(NETDEV_UP, dev);
1209 EXPORT_SYMBOL(dev_open);
1211 static int __dev_close_many(struct list_head *head)
1213 struct net_device *dev;
1218 list_for_each_entry(dev, head, unreg_list) {
1219 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1221 clear_bit(__LINK_STATE_START, &dev->state);
1223 /* Synchronize to scheduled poll. We cannot touch poll list, it
1224 * can be even on different cpu. So just clear netif_running().
1226 * dev->stop() will invoke napi_disable() on all of it's
1227 * napi_struct instances on this device.
1229 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1232 dev_deactivate_many(head);
1234 list_for_each_entry(dev, head, unreg_list) {
1235 const struct net_device_ops *ops = dev->netdev_ops;
1238 * Call the device specific close. This cannot fail.
1239 * Only if device is UP
1241 * We allow it to be called even after a DETACH hot-plug
1247 dev->flags &= ~IFF_UP;
1248 net_dmaengine_put();
1254 static int __dev_close(struct net_device *dev)
1259 list_add(&dev->unreg_list, &single);
1260 retval = __dev_close_many(&single);
1265 static int dev_close_many(struct list_head *head)
1267 struct net_device *dev, *tmp;
1268 LIST_HEAD(tmp_list);
1270 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1271 if (!(dev->flags & IFF_UP))
1272 list_move(&dev->unreg_list, &tmp_list);
1274 __dev_close_many(head);
1276 list_for_each_entry(dev, head, unreg_list) {
1277 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1278 call_netdevice_notifiers(NETDEV_DOWN, dev);
1281 /* rollback_registered_many needs the complete original list */
1282 list_splice(&tmp_list, head);
1287 * dev_close - shutdown an interface.
1288 * @dev: device to shutdown
1290 * This function moves an active device into down state. A
1291 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1292 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1295 int dev_close(struct net_device *dev)
1297 if (dev->flags & IFF_UP) {
1300 list_add(&dev->unreg_list, &single);
1301 dev_close_many(&single);
1306 EXPORT_SYMBOL(dev_close);
1310 * dev_disable_lro - disable Large Receive Offload on a device
1313 * Disable Large Receive Offload (LRO) on a net device. Must be
1314 * called under RTNL. This is needed if received packets may be
1315 * forwarded to another interface.
1317 void dev_disable_lro(struct net_device *dev)
1320 * If we're trying to disable lro on a vlan device
1321 * use the underlying physical device instead
1323 if (is_vlan_dev(dev))
1324 dev = vlan_dev_real_dev(dev);
1326 dev->wanted_features &= ~NETIF_F_LRO;
1327 netdev_update_features(dev);
1329 if (unlikely(dev->features & NETIF_F_LRO))
1330 netdev_WARN(dev, "failed to disable LRO!\n");
1332 EXPORT_SYMBOL(dev_disable_lro);
1335 static int dev_boot_phase = 1;
1338 * register_netdevice_notifier - register a network notifier block
1341 * Register a notifier to be called when network device events occur.
1342 * The notifier passed is linked into the kernel structures and must
1343 * not be reused until it has been unregistered. A negative errno code
1344 * is returned on a failure.
1346 * When registered all registration and up events are replayed
1347 * to the new notifier to allow device to have a race free
1348 * view of the network device list.
1351 int register_netdevice_notifier(struct notifier_block *nb)
1353 struct net_device *dev;
1354 struct net_device *last;
1359 err = raw_notifier_chain_register(&netdev_chain, nb);
1365 for_each_netdev(net, dev) {
1366 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1367 err = notifier_to_errno(err);
1371 if (!(dev->flags & IFF_UP))
1374 nb->notifier_call(nb, NETDEV_UP, dev);
1385 for_each_netdev(net, dev) {
1389 if (dev->flags & IFF_UP) {
1390 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1391 nb->notifier_call(nb, NETDEV_DOWN, dev);
1393 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1394 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1399 raw_notifier_chain_unregister(&netdev_chain, nb);
1402 EXPORT_SYMBOL(register_netdevice_notifier);
1405 * unregister_netdevice_notifier - unregister a network notifier block
1408 * Unregister a notifier previously registered by
1409 * register_netdevice_notifier(). The notifier is unlinked into the
1410 * kernel structures and may then be reused. A negative errno code
1411 * is returned on a failure.
1413 * After unregistering unregister and down device events are synthesized
1414 * for all devices on the device list to the removed notifier to remove
1415 * the need for special case cleanup code.
1418 int unregister_netdevice_notifier(struct notifier_block *nb)
1420 struct net_device *dev;
1425 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1430 for_each_netdev(net, dev) {
1431 if (dev->flags & IFF_UP) {
1432 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1433 nb->notifier_call(nb, NETDEV_DOWN, dev);
1435 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1436 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1443 EXPORT_SYMBOL(unregister_netdevice_notifier);
1446 * call_netdevice_notifiers - call all network notifier blocks
1447 * @val: value passed unmodified to notifier function
1448 * @dev: net_device pointer passed unmodified to notifier function
1450 * Call all network notifier blocks. Parameters and return value
1451 * are as for raw_notifier_call_chain().
1454 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1457 return raw_notifier_call_chain(&netdev_chain, val, dev);
1459 EXPORT_SYMBOL(call_netdevice_notifiers);
1461 static struct static_key netstamp_needed __read_mostly;
1462 #ifdef HAVE_JUMP_LABEL
1463 /* We are not allowed to call static_key_slow_dec() from irq context
1464 * If net_disable_timestamp() is called from irq context, defer the
1465 * static_key_slow_dec() calls.
1467 static atomic_t netstamp_needed_deferred;
1470 void net_enable_timestamp(void)
1472 #ifdef HAVE_JUMP_LABEL
1473 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1477 static_key_slow_dec(&netstamp_needed);
1481 WARN_ON(in_interrupt());
1482 static_key_slow_inc(&netstamp_needed);
1484 EXPORT_SYMBOL(net_enable_timestamp);
1486 void net_disable_timestamp(void)
1488 #ifdef HAVE_JUMP_LABEL
1489 if (in_interrupt()) {
1490 atomic_inc(&netstamp_needed_deferred);
1494 static_key_slow_dec(&netstamp_needed);
1496 EXPORT_SYMBOL(net_disable_timestamp);
1498 static inline void net_timestamp_set(struct sk_buff *skb)
1500 skb->tstamp.tv64 = 0;
1501 if (static_key_false(&netstamp_needed))
1502 __net_timestamp(skb);
1505 #define net_timestamp_check(COND, SKB) \
1506 if (static_key_false(&netstamp_needed)) { \
1507 if ((COND) && !(SKB)->tstamp.tv64) \
1508 __net_timestamp(SKB); \
1511 static int net_hwtstamp_validate(struct ifreq *ifr)
1513 struct hwtstamp_config cfg;
1514 enum hwtstamp_tx_types tx_type;
1515 enum hwtstamp_rx_filters rx_filter;
1516 int tx_type_valid = 0;
1517 int rx_filter_valid = 0;
1519 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1522 if (cfg.flags) /* reserved for future extensions */
1525 tx_type = cfg.tx_type;
1526 rx_filter = cfg.rx_filter;
1529 case HWTSTAMP_TX_OFF:
1530 case HWTSTAMP_TX_ON:
1531 case HWTSTAMP_TX_ONESTEP_SYNC:
1536 switch (rx_filter) {
1537 case HWTSTAMP_FILTER_NONE:
1538 case HWTSTAMP_FILTER_ALL:
1539 case HWTSTAMP_FILTER_SOME:
1540 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1542 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1543 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1545 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1546 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1548 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1549 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1550 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1551 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1552 rx_filter_valid = 1;
1556 if (!tx_type_valid || !rx_filter_valid)
1562 static inline bool is_skb_forwardable(struct net_device *dev,
1563 struct sk_buff *skb)
1567 if (!(dev->flags & IFF_UP))
1570 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1571 if (skb->len <= len)
1574 /* if TSO is enabled, we don't care about the length as the packet
1575 * could be forwarded without being segmented before
1577 if (skb_is_gso(skb))
1584 * dev_forward_skb - loopback an skb to another netif
1586 * @dev: destination network device
1587 * @skb: buffer to forward
1590 * NET_RX_SUCCESS (no congestion)
1591 * NET_RX_DROP (packet was dropped, but freed)
1593 * dev_forward_skb can be used for injecting an skb from the
1594 * start_xmit function of one device into the receive queue
1595 * of another device.
1597 * The receiving device may be in another namespace, so
1598 * we have to clear all information in the skb that could
1599 * impact namespace isolation.
1601 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1603 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1604 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1605 atomic_long_inc(&dev->rx_dropped);
1614 if (unlikely(!is_skb_forwardable(dev, skb))) {
1615 atomic_long_inc(&dev->rx_dropped);
1622 skb->tstamp.tv64 = 0;
1623 skb->pkt_type = PACKET_HOST;
1624 skb->protocol = eth_type_trans(skb, dev);
1628 return netif_rx(skb);
1630 EXPORT_SYMBOL_GPL(dev_forward_skb);
1632 static inline int deliver_skb(struct sk_buff *skb,
1633 struct packet_type *pt_prev,
1634 struct net_device *orig_dev)
1636 atomic_inc(&skb->users);
1637 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1641 * Support routine. Sends outgoing frames to any network
1642 * taps currently in use.
1645 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1647 struct packet_type *ptype;
1648 struct sk_buff *skb2 = NULL;
1649 struct packet_type *pt_prev = NULL;
1652 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1653 /* Never send packets back to the socket
1654 * they originated from - MvS (miquels@drinkel.ow.org)
1656 if ((ptype->dev == dev || !ptype->dev) &&
1657 (ptype->af_packet_priv == NULL ||
1658 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1660 deliver_skb(skb2, pt_prev, skb->dev);
1665 skb2 = skb_clone(skb, GFP_ATOMIC);
1669 net_timestamp_set(skb2);
1671 /* skb->nh should be correctly
1672 set by sender, so that the second statement is
1673 just protection against buggy protocols.
1675 skb_reset_mac_header(skb2);
1677 if (skb_network_header(skb2) < skb2->data ||
1678 skb2->network_header > skb2->tail) {
1679 if (net_ratelimit())
1680 pr_crit("protocol %04x is buggy, dev %s\n",
1681 ntohs(skb2->protocol),
1683 skb_reset_network_header(skb2);
1686 skb2->transport_header = skb2->network_header;
1687 skb2->pkt_type = PACKET_OUTGOING;
1692 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1696 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1697 * @dev: Network device
1698 * @txq: number of queues available
1700 * If real_num_tx_queues is changed the tc mappings may no longer be
1701 * valid. To resolve this verify the tc mapping remains valid and if
1702 * not NULL the mapping. With no priorities mapping to this
1703 * offset/count pair it will no longer be used. In the worst case TC0
1704 * is invalid nothing can be done so disable priority mappings. If is
1705 * expected that drivers will fix this mapping if they can before
1706 * calling netif_set_real_num_tx_queues.
1708 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1711 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1713 /* If TC0 is invalidated disable TC mapping */
1714 if (tc->offset + tc->count > txq) {
1715 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1720 /* Invalidated prio to tc mappings set to TC0 */
1721 for (i = 1; i < TC_BITMASK + 1; i++) {
1722 int q = netdev_get_prio_tc_map(dev, i);
1724 tc = &dev->tc_to_txq[q];
1725 if (tc->offset + tc->count > txq) {
1726 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1728 netdev_set_prio_tc_map(dev, i, 0);
1734 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1735 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1737 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1741 if (txq < 1 || txq > dev->num_tx_queues)
1744 if (dev->reg_state == NETREG_REGISTERED ||
1745 dev->reg_state == NETREG_UNREGISTERING) {
1748 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1754 netif_setup_tc(dev, txq);
1756 if (txq < dev->real_num_tx_queues)
1757 qdisc_reset_all_tx_gt(dev, txq);
1760 dev->real_num_tx_queues = txq;
1763 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1767 * netif_set_real_num_rx_queues - set actual number of RX queues used
1768 * @dev: Network device
1769 * @rxq: Actual number of RX queues
1771 * This must be called either with the rtnl_lock held or before
1772 * registration of the net device. Returns 0 on success, or a
1773 * negative error code. If called before registration, it always
1776 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1780 if (rxq < 1 || rxq > dev->num_rx_queues)
1783 if (dev->reg_state == NETREG_REGISTERED) {
1786 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1792 dev->real_num_rx_queues = rxq;
1795 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1798 static inline void __netif_reschedule(struct Qdisc *q)
1800 struct softnet_data *sd;
1801 unsigned long flags;
1803 local_irq_save(flags);
1804 sd = &__get_cpu_var(softnet_data);
1805 q->next_sched = NULL;
1806 *sd->output_queue_tailp = q;
1807 sd->output_queue_tailp = &q->next_sched;
1808 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1809 local_irq_restore(flags);
1812 void __netif_schedule(struct Qdisc *q)
1814 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1815 __netif_reschedule(q);
1817 EXPORT_SYMBOL(__netif_schedule);
1819 void dev_kfree_skb_irq(struct sk_buff *skb)
1821 if (atomic_dec_and_test(&skb->users)) {
1822 struct softnet_data *sd;
1823 unsigned long flags;
1825 local_irq_save(flags);
1826 sd = &__get_cpu_var(softnet_data);
1827 skb->next = sd->completion_queue;
1828 sd->completion_queue = skb;
1829 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1830 local_irq_restore(flags);
1833 EXPORT_SYMBOL(dev_kfree_skb_irq);
1835 void dev_kfree_skb_any(struct sk_buff *skb)
1837 if (in_irq() || irqs_disabled())
1838 dev_kfree_skb_irq(skb);
1842 EXPORT_SYMBOL(dev_kfree_skb_any);
1846 * netif_device_detach - mark device as removed
1847 * @dev: network device
1849 * Mark device as removed from system and therefore no longer available.
1851 void netif_device_detach(struct net_device *dev)
1853 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1854 netif_running(dev)) {
1855 netif_tx_stop_all_queues(dev);
1858 EXPORT_SYMBOL(netif_device_detach);
1861 * netif_device_attach - mark device as attached
1862 * @dev: network device
1864 * Mark device as attached from system and restart if needed.
1866 void netif_device_attach(struct net_device *dev)
1868 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1869 netif_running(dev)) {
1870 netif_tx_wake_all_queues(dev);
1871 __netdev_watchdog_up(dev);
1874 EXPORT_SYMBOL(netif_device_attach);
1876 static void skb_warn_bad_offload(const struct sk_buff *skb)
1878 static const netdev_features_t null_features = 0;
1879 struct net_device *dev = skb->dev;
1880 const char *driver = "";
1882 if (dev && dev->dev.parent)
1883 driver = dev_driver_string(dev->dev.parent);
1885 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1886 "gso_type=%d ip_summed=%d\n",
1887 driver, dev ? &dev->features : &null_features,
1888 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1889 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1890 skb_shinfo(skb)->gso_type, skb->ip_summed);
1894 * Invalidate hardware checksum when packet is to be mangled, and
1895 * complete checksum manually on outgoing path.
1897 int skb_checksum_help(struct sk_buff *skb)
1900 int ret = 0, offset;
1902 if (skb->ip_summed == CHECKSUM_COMPLETE)
1903 goto out_set_summed;
1905 if (unlikely(skb_shinfo(skb)->gso_size)) {
1906 skb_warn_bad_offload(skb);
1910 offset = skb_checksum_start_offset(skb);
1911 BUG_ON(offset >= skb_headlen(skb));
1912 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1914 offset += skb->csum_offset;
1915 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1917 if (skb_cloned(skb) &&
1918 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1919 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1924 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1926 skb->ip_summed = CHECKSUM_NONE;
1930 EXPORT_SYMBOL(skb_checksum_help);
1933 * skb_gso_segment - Perform segmentation on skb.
1934 * @skb: buffer to segment
1935 * @features: features for the output path (see dev->features)
1937 * This function segments the given skb and returns a list of segments.
1939 * It may return NULL if the skb requires no segmentation. This is
1940 * only possible when GSO is used for verifying header integrity.
1942 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1943 netdev_features_t features)
1945 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1946 struct packet_type *ptype;
1947 __be16 type = skb->protocol;
1948 int vlan_depth = ETH_HLEN;
1951 while (type == htons(ETH_P_8021Q)) {
1952 struct vlan_hdr *vh;
1954 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1955 return ERR_PTR(-EINVAL);
1957 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1958 type = vh->h_vlan_encapsulated_proto;
1959 vlan_depth += VLAN_HLEN;
1962 skb_reset_mac_header(skb);
1963 skb->mac_len = skb->network_header - skb->mac_header;
1964 __skb_pull(skb, skb->mac_len);
1966 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1967 skb_warn_bad_offload(skb);
1969 if (skb_header_cloned(skb) &&
1970 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1971 return ERR_PTR(err);
1975 list_for_each_entry_rcu(ptype,
1976 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1977 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1978 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1979 err = ptype->gso_send_check(skb);
1980 segs = ERR_PTR(err);
1981 if (err || skb_gso_ok(skb, features))
1983 __skb_push(skb, (skb->data -
1984 skb_network_header(skb)));
1986 segs = ptype->gso_segment(skb, features);
1992 __skb_push(skb, skb->data - skb_mac_header(skb));
1996 EXPORT_SYMBOL(skb_gso_segment);
1998 /* Take action when hardware reception checksum errors are detected. */
2000 void netdev_rx_csum_fault(struct net_device *dev)
2002 if (net_ratelimit()) {
2003 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2007 EXPORT_SYMBOL(netdev_rx_csum_fault);
2010 /* Actually, we should eliminate this check as soon as we know, that:
2011 * 1. IOMMU is present and allows to map all the memory.
2012 * 2. No high memory really exists on this machine.
2015 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2017 #ifdef CONFIG_HIGHMEM
2019 if (!(dev->features & NETIF_F_HIGHDMA)) {
2020 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2021 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2022 if (PageHighMem(skb_frag_page(frag)))
2027 if (PCI_DMA_BUS_IS_PHYS) {
2028 struct device *pdev = dev->dev.parent;
2032 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2033 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2034 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2035 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2044 void (*destructor)(struct sk_buff *skb);
2047 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2049 static void dev_gso_skb_destructor(struct sk_buff *skb)
2051 struct dev_gso_cb *cb;
2054 struct sk_buff *nskb = skb->next;
2056 skb->next = nskb->next;
2059 } while (skb->next);
2061 cb = DEV_GSO_CB(skb);
2063 cb->destructor(skb);
2067 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2068 * @skb: buffer to segment
2069 * @features: device features as applicable to this skb
2071 * This function segments the given skb and stores the list of segments
2074 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2076 struct sk_buff *segs;
2078 segs = skb_gso_segment(skb, features);
2080 /* Verifying header integrity only. */
2085 return PTR_ERR(segs);
2088 DEV_GSO_CB(skb)->destructor = skb->destructor;
2089 skb->destructor = dev_gso_skb_destructor;
2094 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2096 return ((features & NETIF_F_GEN_CSUM) ||
2097 ((features & NETIF_F_V4_CSUM) &&
2098 protocol == htons(ETH_P_IP)) ||
2099 ((features & NETIF_F_V6_CSUM) &&
2100 protocol == htons(ETH_P_IPV6)) ||
2101 ((features & NETIF_F_FCOE_CRC) &&
2102 protocol == htons(ETH_P_FCOE)));
2105 static netdev_features_t harmonize_features(struct sk_buff *skb,
2106 __be16 protocol, netdev_features_t features)
2108 if (!can_checksum_protocol(features, protocol)) {
2109 features &= ~NETIF_F_ALL_CSUM;
2110 features &= ~NETIF_F_SG;
2111 } else if (illegal_highdma(skb->dev, skb)) {
2112 features &= ~NETIF_F_SG;
2118 netdev_features_t netif_skb_features(struct sk_buff *skb)
2120 __be16 protocol = skb->protocol;
2121 netdev_features_t features = skb->dev->features;
2123 if (protocol == htons(ETH_P_8021Q)) {
2124 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2125 protocol = veh->h_vlan_encapsulated_proto;
2126 } else if (!vlan_tx_tag_present(skb)) {
2127 return harmonize_features(skb, protocol, features);
2130 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2132 if (protocol != htons(ETH_P_8021Q)) {
2133 return harmonize_features(skb, protocol, features);
2135 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2136 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2137 return harmonize_features(skb, protocol, features);
2140 EXPORT_SYMBOL(netif_skb_features);
2143 * Returns true if either:
2144 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2145 * 2. skb is fragmented and the device does not support SG, or if
2146 * at least one of fragments is in highmem and device does not
2147 * support DMA from it.
2149 static inline int skb_needs_linearize(struct sk_buff *skb,
2152 return skb_is_nonlinear(skb) &&
2153 ((skb_has_frag_list(skb) &&
2154 !(features & NETIF_F_FRAGLIST)) ||
2155 (skb_shinfo(skb)->nr_frags &&
2156 !(features & NETIF_F_SG)));
2159 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2160 struct netdev_queue *txq)
2162 const struct net_device_ops *ops = dev->netdev_ops;
2163 int rc = NETDEV_TX_OK;
2164 unsigned int skb_len;
2166 if (likely(!skb->next)) {
2167 netdev_features_t features;
2170 * If device doesn't need skb->dst, release it right now while
2171 * its hot in this cpu cache
2173 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2176 if (!list_empty(&ptype_all))
2177 dev_queue_xmit_nit(skb, dev);
2179 features = netif_skb_features(skb);
2181 if (vlan_tx_tag_present(skb) &&
2182 !(features & NETIF_F_HW_VLAN_TX)) {
2183 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2190 if (netif_needs_gso(skb, features)) {
2191 if (unlikely(dev_gso_segment(skb, features)))
2196 if (skb_needs_linearize(skb, features) &&
2197 __skb_linearize(skb))
2200 /* If packet is not checksummed and device does not
2201 * support checksumming for this protocol, complete
2202 * checksumming here.
2204 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2205 skb_set_transport_header(skb,
2206 skb_checksum_start_offset(skb));
2207 if (!(features & NETIF_F_ALL_CSUM) &&
2208 skb_checksum_help(skb))
2214 rc = ops->ndo_start_xmit(skb, dev);
2215 trace_net_dev_xmit(skb, rc, dev, skb_len);
2216 if (rc == NETDEV_TX_OK)
2217 txq_trans_update(txq);
2223 struct sk_buff *nskb = skb->next;
2225 skb->next = nskb->next;
2229 * If device doesn't need nskb->dst, release it right now while
2230 * its hot in this cpu cache
2232 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2235 skb_len = nskb->len;
2236 rc = ops->ndo_start_xmit(nskb, dev);
2237 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2238 if (unlikely(rc != NETDEV_TX_OK)) {
2239 if (rc & ~NETDEV_TX_MASK)
2240 goto out_kfree_gso_skb;
2241 nskb->next = skb->next;
2245 txq_trans_update(txq);
2246 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2247 return NETDEV_TX_BUSY;
2248 } while (skb->next);
2251 if (likely(skb->next == NULL))
2252 skb->destructor = DEV_GSO_CB(skb)->destructor;
2259 static u32 hashrnd __read_mostly;
2262 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2263 * to be used as a distribution range.
2265 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2266 unsigned int num_tx_queues)
2270 u16 qcount = num_tx_queues;
2272 if (skb_rx_queue_recorded(skb)) {
2273 hash = skb_get_rx_queue(skb);
2274 while (unlikely(hash >= num_tx_queues))
2275 hash -= num_tx_queues;
2280 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2281 qoffset = dev->tc_to_txq[tc].offset;
2282 qcount = dev->tc_to_txq[tc].count;
2285 if (skb->sk && skb->sk->sk_hash)
2286 hash = skb->sk->sk_hash;
2288 hash = (__force u16) skb->protocol;
2289 hash = jhash_1word(hash, hashrnd);
2291 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2293 EXPORT_SYMBOL(__skb_tx_hash);
2295 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2297 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2298 if (net_ratelimit()) {
2299 pr_warn("%s selects TX queue %d, but real number of TX queues is %d\n",
2300 dev->name, queue_index,
2301 dev->real_num_tx_queues);
2308 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2311 struct xps_dev_maps *dev_maps;
2312 struct xps_map *map;
2313 int queue_index = -1;
2316 dev_maps = rcu_dereference(dev->xps_maps);
2318 map = rcu_dereference(
2319 dev_maps->cpu_map[raw_smp_processor_id()]);
2322 queue_index = map->queues[0];
2325 if (skb->sk && skb->sk->sk_hash)
2326 hash = skb->sk->sk_hash;
2328 hash = (__force u16) skb->protocol ^
2330 hash = jhash_1word(hash, hashrnd);
2331 queue_index = map->queues[
2332 ((u64)hash * map->len) >> 32];
2334 if (unlikely(queue_index >= dev->real_num_tx_queues))
2346 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2347 struct sk_buff *skb)
2350 const struct net_device_ops *ops = dev->netdev_ops;
2352 if (dev->real_num_tx_queues == 1)
2354 else if (ops->ndo_select_queue) {
2355 queue_index = ops->ndo_select_queue(dev, skb);
2356 queue_index = dev_cap_txqueue(dev, queue_index);
2358 struct sock *sk = skb->sk;
2359 queue_index = sk_tx_queue_get(sk);
2361 if (queue_index < 0 || skb->ooo_okay ||
2362 queue_index >= dev->real_num_tx_queues) {
2363 int old_index = queue_index;
2365 queue_index = get_xps_queue(dev, skb);
2366 if (queue_index < 0)
2367 queue_index = skb_tx_hash(dev, skb);
2369 if (queue_index != old_index && sk) {
2370 struct dst_entry *dst =
2371 rcu_dereference_check(sk->sk_dst_cache, 1);
2373 if (dst && skb_dst(skb) == dst)
2374 sk_tx_queue_set(sk, queue_index);
2379 skb_set_queue_mapping(skb, queue_index);
2380 return netdev_get_tx_queue(dev, queue_index);
2383 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2384 struct net_device *dev,
2385 struct netdev_queue *txq)
2387 spinlock_t *root_lock = qdisc_lock(q);
2391 qdisc_skb_cb(skb)->pkt_len = skb->len;
2392 qdisc_calculate_pkt_len(skb, q);
2394 * Heuristic to force contended enqueues to serialize on a
2395 * separate lock before trying to get qdisc main lock.
2396 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2397 * and dequeue packets faster.
2399 contended = qdisc_is_running(q);
2400 if (unlikely(contended))
2401 spin_lock(&q->busylock);
2403 spin_lock(root_lock);
2404 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2407 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2408 qdisc_run_begin(q)) {
2410 * This is a work-conserving queue; there are no old skbs
2411 * waiting to be sent out; and the qdisc is not running -
2412 * xmit the skb directly.
2414 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2417 qdisc_bstats_update(q, skb);
2419 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2420 if (unlikely(contended)) {
2421 spin_unlock(&q->busylock);
2428 rc = NET_XMIT_SUCCESS;
2431 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2432 if (qdisc_run_begin(q)) {
2433 if (unlikely(contended)) {
2434 spin_unlock(&q->busylock);
2440 spin_unlock(root_lock);
2441 if (unlikely(contended))
2442 spin_unlock(&q->busylock);
2446 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2447 static void skb_update_prio(struct sk_buff *skb)
2449 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2451 if ((!skb->priority) && (skb->sk) && map)
2452 skb->priority = map->priomap[skb->sk->sk_cgrp_prioidx];
2455 #define skb_update_prio(skb)
2458 static DEFINE_PER_CPU(int, xmit_recursion);
2459 #define RECURSION_LIMIT 10
2462 * dev_queue_xmit - transmit a buffer
2463 * @skb: buffer to transmit
2465 * Queue a buffer for transmission to a network device. The caller must
2466 * have set the device and priority and built the buffer before calling
2467 * this function. The function can be called from an interrupt.
2469 * A negative errno code is returned on a failure. A success does not
2470 * guarantee the frame will be transmitted as it may be dropped due
2471 * to congestion or traffic shaping.
2473 * -----------------------------------------------------------------------------------
2474 * I notice this method can also return errors from the queue disciplines,
2475 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2478 * Regardless of the return value, the skb is consumed, so it is currently
2479 * difficult to retry a send to this method. (You can bump the ref count
2480 * before sending to hold a reference for retry if you are careful.)
2482 * When calling this method, interrupts MUST be enabled. This is because
2483 * the BH enable code must have IRQs enabled so that it will not deadlock.
2486 int dev_queue_xmit(struct sk_buff *skb)
2488 struct net_device *dev = skb->dev;
2489 struct netdev_queue *txq;
2493 /* Disable soft irqs for various locks below. Also
2494 * stops preemption for RCU.
2498 skb_update_prio(skb);
2500 txq = dev_pick_tx(dev, skb);
2501 q = rcu_dereference_bh(txq->qdisc);
2503 #ifdef CONFIG_NET_CLS_ACT
2504 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2506 trace_net_dev_queue(skb);
2508 rc = __dev_xmit_skb(skb, q, dev, txq);
2512 /* The device has no queue. Common case for software devices:
2513 loopback, all the sorts of tunnels...
2515 Really, it is unlikely that netif_tx_lock protection is necessary
2516 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2518 However, it is possible, that they rely on protection
2521 Check this and shot the lock. It is not prone from deadlocks.
2522 Either shot noqueue qdisc, it is even simpler 8)
2524 if (dev->flags & IFF_UP) {
2525 int cpu = smp_processor_id(); /* ok because BHs are off */
2527 if (txq->xmit_lock_owner != cpu) {
2529 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2530 goto recursion_alert;
2532 HARD_TX_LOCK(dev, txq, cpu);
2534 if (!netif_xmit_stopped(txq)) {
2535 __this_cpu_inc(xmit_recursion);
2536 rc = dev_hard_start_xmit(skb, dev, txq);
2537 __this_cpu_dec(xmit_recursion);
2538 if (dev_xmit_complete(rc)) {
2539 HARD_TX_UNLOCK(dev, txq);
2543 HARD_TX_UNLOCK(dev, txq);
2544 if (net_ratelimit())
2545 pr_crit("Virtual device %s asks to queue packet!\n",
2548 /* Recursion is detected! It is possible,
2552 if (net_ratelimit())
2553 pr_crit("Dead loop on virtual device %s, fix it urgently!\n",
2559 rcu_read_unlock_bh();
2564 rcu_read_unlock_bh();
2567 EXPORT_SYMBOL(dev_queue_xmit);
2570 /*=======================================================================
2572 =======================================================================*/
2574 int netdev_max_backlog __read_mostly = 1000;
2575 int netdev_tstamp_prequeue __read_mostly = 1;
2576 int netdev_budget __read_mostly = 300;
2577 int weight_p __read_mostly = 64; /* old backlog weight */
2579 /* Called with irq disabled */
2580 static inline void ____napi_schedule(struct softnet_data *sd,
2581 struct napi_struct *napi)
2583 list_add_tail(&napi->poll_list, &sd->poll_list);
2584 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2588 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2589 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2590 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2591 * if hash is a canonical 4-tuple hash over transport ports.
2593 void __skb_get_rxhash(struct sk_buff *skb)
2595 struct flow_keys keys;
2598 if (!skb_flow_dissect(skb, &keys))
2602 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2603 swap(keys.port16[0], keys.port16[1]);
2607 /* get a consistent hash (same value on both flow directions) */
2608 if ((__force u32)keys.dst < (__force u32)keys.src)
2609 swap(keys.dst, keys.src);
2611 hash = jhash_3words((__force u32)keys.dst,
2612 (__force u32)keys.src,
2613 (__force u32)keys.ports, hashrnd);
2619 EXPORT_SYMBOL(__skb_get_rxhash);
2623 /* One global table that all flow-based protocols share. */
2624 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2625 EXPORT_SYMBOL(rps_sock_flow_table);
2627 struct static_key rps_needed __read_mostly;
2629 static struct rps_dev_flow *
2630 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2631 struct rps_dev_flow *rflow, u16 next_cpu)
2633 if (next_cpu != RPS_NO_CPU) {
2634 #ifdef CONFIG_RFS_ACCEL
2635 struct netdev_rx_queue *rxqueue;
2636 struct rps_dev_flow_table *flow_table;
2637 struct rps_dev_flow *old_rflow;
2642 /* Should we steer this flow to a different hardware queue? */
2643 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2644 !(dev->features & NETIF_F_NTUPLE))
2646 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2647 if (rxq_index == skb_get_rx_queue(skb))
2650 rxqueue = dev->_rx + rxq_index;
2651 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2654 flow_id = skb->rxhash & flow_table->mask;
2655 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2656 rxq_index, flow_id);
2660 rflow = &flow_table->flows[flow_id];
2662 if (old_rflow->filter == rflow->filter)
2663 old_rflow->filter = RPS_NO_FILTER;
2667 per_cpu(softnet_data, next_cpu).input_queue_head;
2670 rflow->cpu = next_cpu;
2675 * get_rps_cpu is called from netif_receive_skb and returns the target
2676 * CPU from the RPS map of the receiving queue for a given skb.
2677 * rcu_read_lock must be held on entry.
2679 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2680 struct rps_dev_flow **rflowp)
2682 struct netdev_rx_queue *rxqueue;
2683 struct rps_map *map;
2684 struct rps_dev_flow_table *flow_table;
2685 struct rps_sock_flow_table *sock_flow_table;
2689 if (skb_rx_queue_recorded(skb)) {
2690 u16 index = skb_get_rx_queue(skb);
2691 if (unlikely(index >= dev->real_num_rx_queues)) {
2692 WARN_ONCE(dev->real_num_rx_queues > 1,
2693 "%s received packet on queue %u, but number "
2694 "of RX queues is %u\n",
2695 dev->name, index, dev->real_num_rx_queues);
2698 rxqueue = dev->_rx + index;
2702 map = rcu_dereference(rxqueue->rps_map);
2704 if (map->len == 1 &&
2705 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2706 tcpu = map->cpus[0];
2707 if (cpu_online(tcpu))
2711 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2715 skb_reset_network_header(skb);
2716 if (!skb_get_rxhash(skb))
2719 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2720 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2721 if (flow_table && sock_flow_table) {
2723 struct rps_dev_flow *rflow;
2725 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2728 next_cpu = sock_flow_table->ents[skb->rxhash &
2729 sock_flow_table->mask];
2732 * If the desired CPU (where last recvmsg was done) is
2733 * different from current CPU (one in the rx-queue flow
2734 * table entry), switch if one of the following holds:
2735 * - Current CPU is unset (equal to RPS_NO_CPU).
2736 * - Current CPU is offline.
2737 * - The current CPU's queue tail has advanced beyond the
2738 * last packet that was enqueued using this table entry.
2739 * This guarantees that all previous packets for the flow
2740 * have been dequeued, thus preserving in order delivery.
2742 if (unlikely(tcpu != next_cpu) &&
2743 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2744 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2745 rflow->last_qtail)) >= 0))
2746 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2748 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2756 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2758 if (cpu_online(tcpu)) {
2768 #ifdef CONFIG_RFS_ACCEL
2771 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2772 * @dev: Device on which the filter was set
2773 * @rxq_index: RX queue index
2774 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2775 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2777 * Drivers that implement ndo_rx_flow_steer() should periodically call
2778 * this function for each installed filter and remove the filters for
2779 * which it returns %true.
2781 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2782 u32 flow_id, u16 filter_id)
2784 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2785 struct rps_dev_flow_table *flow_table;
2786 struct rps_dev_flow *rflow;
2791 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2792 if (flow_table && flow_id <= flow_table->mask) {
2793 rflow = &flow_table->flows[flow_id];
2794 cpu = ACCESS_ONCE(rflow->cpu);
2795 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2796 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2797 rflow->last_qtail) <
2798 (int)(10 * flow_table->mask)))
2804 EXPORT_SYMBOL(rps_may_expire_flow);
2806 #endif /* CONFIG_RFS_ACCEL */
2808 /* Called from hardirq (IPI) context */
2809 static void rps_trigger_softirq(void *data)
2811 struct softnet_data *sd = data;
2813 ____napi_schedule(sd, &sd->backlog);
2817 #endif /* CONFIG_RPS */
2820 * Check if this softnet_data structure is another cpu one
2821 * If yes, queue it to our IPI list and return 1
2824 static int rps_ipi_queued(struct softnet_data *sd)
2827 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2830 sd->rps_ipi_next = mysd->rps_ipi_list;
2831 mysd->rps_ipi_list = sd;
2833 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2836 #endif /* CONFIG_RPS */
2841 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2842 * queue (may be a remote CPU queue).
2844 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2845 unsigned int *qtail)
2847 struct softnet_data *sd;
2848 unsigned long flags;
2850 sd = &per_cpu(softnet_data, cpu);
2852 local_irq_save(flags);
2855 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2856 if (skb_queue_len(&sd->input_pkt_queue)) {
2858 __skb_queue_tail(&sd->input_pkt_queue, skb);
2859 input_queue_tail_incr_save(sd, qtail);
2861 local_irq_restore(flags);
2862 return NET_RX_SUCCESS;
2865 /* Schedule NAPI for backlog device
2866 * We can use non atomic operation since we own the queue lock
2868 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2869 if (!rps_ipi_queued(sd))
2870 ____napi_schedule(sd, &sd->backlog);
2878 local_irq_restore(flags);
2880 atomic_long_inc(&skb->dev->rx_dropped);
2886 * netif_rx - post buffer to the network code
2887 * @skb: buffer to post
2889 * This function receives a packet from a device driver and queues it for
2890 * the upper (protocol) levels to process. It always succeeds. The buffer
2891 * may be dropped during processing for congestion control or by the
2895 * NET_RX_SUCCESS (no congestion)
2896 * NET_RX_DROP (packet was dropped)
2900 int netif_rx(struct sk_buff *skb)
2904 /* if netpoll wants it, pretend we never saw it */
2905 if (netpoll_rx(skb))
2908 net_timestamp_check(netdev_tstamp_prequeue, skb);
2910 trace_netif_rx(skb);
2912 if (static_key_false(&rps_needed)) {
2913 struct rps_dev_flow voidflow, *rflow = &voidflow;
2919 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2921 cpu = smp_processor_id();
2923 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2931 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2936 EXPORT_SYMBOL(netif_rx);
2938 int netif_rx_ni(struct sk_buff *skb)
2943 err = netif_rx(skb);
2944 if (local_softirq_pending())
2950 EXPORT_SYMBOL(netif_rx_ni);
2952 static void net_tx_action(struct softirq_action *h)
2954 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2956 if (sd->completion_queue) {
2957 struct sk_buff *clist;
2959 local_irq_disable();
2960 clist = sd->completion_queue;
2961 sd->completion_queue = NULL;
2965 struct sk_buff *skb = clist;
2966 clist = clist->next;
2968 WARN_ON(atomic_read(&skb->users));
2969 trace_kfree_skb(skb, net_tx_action);
2974 if (sd->output_queue) {
2977 local_irq_disable();
2978 head = sd->output_queue;
2979 sd->output_queue = NULL;
2980 sd->output_queue_tailp = &sd->output_queue;
2984 struct Qdisc *q = head;
2985 spinlock_t *root_lock;
2987 head = head->next_sched;
2989 root_lock = qdisc_lock(q);
2990 if (spin_trylock(root_lock)) {
2991 smp_mb__before_clear_bit();
2992 clear_bit(__QDISC_STATE_SCHED,
2995 spin_unlock(root_lock);
2997 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2999 __netif_reschedule(q);
3001 smp_mb__before_clear_bit();
3002 clear_bit(__QDISC_STATE_SCHED,
3010 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3011 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3012 /* This hook is defined here for ATM LANE */
3013 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3014 unsigned char *addr) __read_mostly;
3015 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3018 #ifdef CONFIG_NET_CLS_ACT
3019 /* TODO: Maybe we should just force sch_ingress to be compiled in
3020 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3021 * a compare and 2 stores extra right now if we dont have it on
3022 * but have CONFIG_NET_CLS_ACT
3023 * NOTE: This doesn't stop any functionality; if you dont have
3024 * the ingress scheduler, you just can't add policies on ingress.
3027 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3029 struct net_device *dev = skb->dev;
3030 u32 ttl = G_TC_RTTL(skb->tc_verd);
3031 int result = TC_ACT_OK;
3034 if (unlikely(MAX_RED_LOOP < ttl++)) {
3035 if (net_ratelimit())
3036 pr_warn("Redir loop detected Dropping packet (%d->%d)\n",
3037 skb->skb_iif, dev->ifindex);
3041 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3042 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3045 if (q != &noop_qdisc) {
3046 spin_lock(qdisc_lock(q));
3047 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3048 result = qdisc_enqueue_root(skb, q);
3049 spin_unlock(qdisc_lock(q));
3055 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3056 struct packet_type **pt_prev,
3057 int *ret, struct net_device *orig_dev)
3059 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3061 if (!rxq || rxq->qdisc == &noop_qdisc)
3065 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3069 switch (ing_filter(skb, rxq)) {
3083 * netdev_rx_handler_register - register receive handler
3084 * @dev: device to register a handler for
3085 * @rx_handler: receive handler to register
3086 * @rx_handler_data: data pointer that is used by rx handler
3088 * Register a receive hander for a device. This handler will then be
3089 * called from __netif_receive_skb. A negative errno code is returned
3092 * The caller must hold the rtnl_mutex.
3094 * For a general description of rx_handler, see enum rx_handler_result.
3096 int netdev_rx_handler_register(struct net_device *dev,
3097 rx_handler_func_t *rx_handler,
3098 void *rx_handler_data)
3102 if (dev->rx_handler)
3105 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3106 rcu_assign_pointer(dev->rx_handler, rx_handler);
3110 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3113 * netdev_rx_handler_unregister - unregister receive handler
3114 * @dev: device to unregister a handler from
3116 * Unregister a receive hander from a device.
3118 * The caller must hold the rtnl_mutex.
3120 void netdev_rx_handler_unregister(struct net_device *dev)
3124 RCU_INIT_POINTER(dev->rx_handler, NULL);
3125 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3127 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3129 static int __netif_receive_skb(struct sk_buff *skb)
3131 struct packet_type *ptype, *pt_prev;
3132 rx_handler_func_t *rx_handler;
3133 struct net_device *orig_dev;
3134 struct net_device *null_or_dev;
3135 bool deliver_exact = false;
3136 int ret = NET_RX_DROP;
3139 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3141 trace_netif_receive_skb(skb);
3143 /* if we've gotten here through NAPI, check netpoll */
3144 if (netpoll_receive_skb(skb))
3148 skb->skb_iif = skb->dev->ifindex;
3149 orig_dev = skb->dev;
3151 skb_reset_network_header(skb);
3152 skb_reset_transport_header(skb);
3153 skb_reset_mac_len(skb);
3161 __this_cpu_inc(softnet_data.processed);
3163 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3164 skb = vlan_untag(skb);
3169 #ifdef CONFIG_NET_CLS_ACT
3170 if (skb->tc_verd & TC_NCLS) {
3171 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3176 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3177 if (!ptype->dev || ptype->dev == skb->dev) {
3179 ret = deliver_skb(skb, pt_prev, orig_dev);
3184 #ifdef CONFIG_NET_CLS_ACT
3185 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3191 rx_handler = rcu_dereference(skb->dev->rx_handler);
3192 if (vlan_tx_tag_present(skb)) {
3194 ret = deliver_skb(skb, pt_prev, orig_dev);
3197 if (vlan_do_receive(&skb, !rx_handler))
3199 else if (unlikely(!skb))
3205 ret = deliver_skb(skb, pt_prev, orig_dev);
3208 switch (rx_handler(&skb)) {
3209 case RX_HANDLER_CONSUMED:
3211 case RX_HANDLER_ANOTHER:
3213 case RX_HANDLER_EXACT:
3214 deliver_exact = true;
3215 case RX_HANDLER_PASS:
3222 /* deliver only exact match when indicated */
3223 null_or_dev = deliver_exact ? skb->dev : NULL;
3225 type = skb->protocol;
3226 list_for_each_entry_rcu(ptype,
3227 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3228 if (ptype->type == type &&
3229 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3230 ptype->dev == orig_dev)) {
3232 ret = deliver_skb(skb, pt_prev, orig_dev);
3238 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3240 atomic_long_inc(&skb->dev->rx_dropped);
3242 /* Jamal, now you will not able to escape explaining
3243 * me how you were going to use this. :-)
3254 * netif_receive_skb - process receive buffer from network
3255 * @skb: buffer to process
3257 * netif_receive_skb() is the main receive data processing function.
3258 * It always succeeds. The buffer may be dropped during processing
3259 * for congestion control or by the protocol layers.
3261 * This function may only be called from softirq context and interrupts
3262 * should be enabled.
3264 * Return values (usually ignored):
3265 * NET_RX_SUCCESS: no congestion
3266 * NET_RX_DROP: packet was dropped
3268 int netif_receive_skb(struct sk_buff *skb)
3270 net_timestamp_check(netdev_tstamp_prequeue, skb);
3272 if (skb_defer_rx_timestamp(skb))
3273 return NET_RX_SUCCESS;
3276 if (static_key_false(&rps_needed)) {
3277 struct rps_dev_flow voidflow, *rflow = &voidflow;
3282 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3285 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3292 return __netif_receive_skb(skb);
3294 EXPORT_SYMBOL(netif_receive_skb);
3296 /* Network device is going away, flush any packets still pending
3297 * Called with irqs disabled.
3299 static void flush_backlog(void *arg)
3301 struct net_device *dev = arg;
3302 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3303 struct sk_buff *skb, *tmp;
3306 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3307 if (skb->dev == dev) {
3308 __skb_unlink(skb, &sd->input_pkt_queue);
3310 input_queue_head_incr(sd);
3315 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3316 if (skb->dev == dev) {
3317 __skb_unlink(skb, &sd->process_queue);
3319 input_queue_head_incr(sd);
3324 static int napi_gro_complete(struct sk_buff *skb)
3326 struct packet_type *ptype;
3327 __be16 type = skb->protocol;
3328 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3331 if (NAPI_GRO_CB(skb)->count == 1) {
3332 skb_shinfo(skb)->gso_size = 0;
3337 list_for_each_entry_rcu(ptype, head, list) {
3338 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3341 err = ptype->gro_complete(skb);
3347 WARN_ON(&ptype->list == head);
3349 return NET_RX_SUCCESS;
3353 return netif_receive_skb(skb);
3356 inline void napi_gro_flush(struct napi_struct *napi)
3358 struct sk_buff *skb, *next;
3360 for (skb = napi->gro_list; skb; skb = next) {
3363 napi_gro_complete(skb);
3366 napi->gro_count = 0;
3367 napi->gro_list = NULL;
3369 EXPORT_SYMBOL(napi_gro_flush);
3371 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3373 struct sk_buff **pp = NULL;
3374 struct packet_type *ptype;
3375 __be16 type = skb->protocol;
3376 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3379 enum gro_result ret;
3381 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3384 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3388 list_for_each_entry_rcu(ptype, head, list) {
3389 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3392 skb_set_network_header(skb, skb_gro_offset(skb));
3393 mac_len = skb->network_header - skb->mac_header;
3394 skb->mac_len = mac_len;
3395 NAPI_GRO_CB(skb)->same_flow = 0;
3396 NAPI_GRO_CB(skb)->flush = 0;
3397 NAPI_GRO_CB(skb)->free = 0;
3399 pp = ptype->gro_receive(&napi->gro_list, skb);
3404 if (&ptype->list == head)
3407 same_flow = NAPI_GRO_CB(skb)->same_flow;
3408 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3411 struct sk_buff *nskb = *pp;
3415 napi_gro_complete(nskb);
3422 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3426 NAPI_GRO_CB(skb)->count = 1;
3427 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3428 skb->next = napi->gro_list;
3429 napi->gro_list = skb;
3433 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3434 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3436 BUG_ON(skb->end - skb->tail < grow);
3438 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3441 skb->data_len -= grow;
3443 skb_shinfo(skb)->frags[0].page_offset += grow;
3444 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3446 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3447 skb_frag_unref(skb, 0);
3448 memmove(skb_shinfo(skb)->frags,
3449 skb_shinfo(skb)->frags + 1,
3450 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3461 EXPORT_SYMBOL(dev_gro_receive);
3463 static inline gro_result_t
3464 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3467 unsigned int maclen = skb->dev->hard_header_len;
3469 for (p = napi->gro_list; p; p = p->next) {
3470 unsigned long diffs;
3472 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3473 diffs |= p->vlan_tci ^ skb->vlan_tci;
3474 if (maclen == ETH_HLEN)
3475 diffs |= compare_ether_header(skb_mac_header(p),
3476 skb_gro_mac_header(skb));
3478 diffs = memcmp(skb_mac_header(p),
3479 skb_gro_mac_header(skb),
3481 NAPI_GRO_CB(p)->same_flow = !diffs;
3482 NAPI_GRO_CB(p)->flush = 0;
3485 return dev_gro_receive(napi, skb);
3488 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3492 if (netif_receive_skb(skb))
3497 case GRO_MERGED_FREE:
3508 EXPORT_SYMBOL(napi_skb_finish);
3510 void skb_gro_reset_offset(struct sk_buff *skb)
3512 NAPI_GRO_CB(skb)->data_offset = 0;
3513 NAPI_GRO_CB(skb)->frag0 = NULL;
3514 NAPI_GRO_CB(skb)->frag0_len = 0;
3516 if (skb->mac_header == skb->tail &&
3517 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3518 NAPI_GRO_CB(skb)->frag0 =
3519 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3520 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3523 EXPORT_SYMBOL(skb_gro_reset_offset);
3525 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3527 skb_gro_reset_offset(skb);
3529 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3531 EXPORT_SYMBOL(napi_gro_receive);
3533 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3535 __skb_pull(skb, skb_headlen(skb));
3536 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3537 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3539 skb->dev = napi->dev;
3545 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3547 struct sk_buff *skb = napi->skb;
3550 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3556 EXPORT_SYMBOL(napi_get_frags);
3558 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3564 skb->protocol = eth_type_trans(skb, skb->dev);
3566 if (ret == GRO_HELD)
3567 skb_gro_pull(skb, -ETH_HLEN);
3568 else if (netif_receive_skb(skb))
3573 case GRO_MERGED_FREE:
3574 napi_reuse_skb(napi, skb);
3583 EXPORT_SYMBOL(napi_frags_finish);
3585 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3587 struct sk_buff *skb = napi->skb;
3594 skb_reset_mac_header(skb);
3595 skb_gro_reset_offset(skb);
3597 off = skb_gro_offset(skb);
3598 hlen = off + sizeof(*eth);
3599 eth = skb_gro_header_fast(skb, off);
3600 if (skb_gro_header_hard(skb, hlen)) {
3601 eth = skb_gro_header_slow(skb, hlen, off);
3602 if (unlikely(!eth)) {
3603 napi_reuse_skb(napi, skb);
3609 skb_gro_pull(skb, sizeof(*eth));
3612 * This works because the only protocols we care about don't require
3613 * special handling. We'll fix it up properly at the end.
3615 skb->protocol = eth->h_proto;
3620 EXPORT_SYMBOL(napi_frags_skb);
3622 gro_result_t napi_gro_frags(struct napi_struct *napi)
3624 struct sk_buff *skb = napi_frags_skb(napi);
3629 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3631 EXPORT_SYMBOL(napi_gro_frags);
3634 * net_rps_action sends any pending IPI's for rps.
3635 * Note: called with local irq disabled, but exits with local irq enabled.
3637 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3640 struct softnet_data *remsd = sd->rps_ipi_list;
3643 sd->rps_ipi_list = NULL;
3647 /* Send pending IPI's to kick RPS processing on remote cpus. */
3649 struct softnet_data *next = remsd->rps_ipi_next;
3651 if (cpu_online(remsd->cpu))
3652 __smp_call_function_single(remsd->cpu,
3661 static int process_backlog(struct napi_struct *napi, int quota)
3664 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3667 /* Check if we have pending ipi, its better to send them now,
3668 * not waiting net_rx_action() end.
3670 if (sd->rps_ipi_list) {
3671 local_irq_disable();
3672 net_rps_action_and_irq_enable(sd);
3675 napi->weight = weight_p;
3676 local_irq_disable();
3677 while (work < quota) {
3678 struct sk_buff *skb;
3681 while ((skb = __skb_dequeue(&sd->process_queue))) {
3683 __netif_receive_skb(skb);
3684 local_irq_disable();
3685 input_queue_head_incr(sd);
3686 if (++work >= quota) {
3693 qlen = skb_queue_len(&sd->input_pkt_queue);
3695 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3696 &sd->process_queue);
3698 if (qlen < quota - work) {
3700 * Inline a custom version of __napi_complete().
3701 * only current cpu owns and manipulates this napi,
3702 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3703 * we can use a plain write instead of clear_bit(),
3704 * and we dont need an smp_mb() memory barrier.
3706 list_del(&napi->poll_list);
3709 quota = work + qlen;
3719 * __napi_schedule - schedule for receive
3720 * @n: entry to schedule
3722 * The entry's receive function will be scheduled to run
3724 void __napi_schedule(struct napi_struct *n)
3726 unsigned long flags;
3728 local_irq_save(flags);
3729 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3730 local_irq_restore(flags);
3732 EXPORT_SYMBOL(__napi_schedule);
3734 void __napi_complete(struct napi_struct *n)
3736 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3737 BUG_ON(n->gro_list);
3739 list_del(&n->poll_list);
3740 smp_mb__before_clear_bit();
3741 clear_bit(NAPI_STATE_SCHED, &n->state);
3743 EXPORT_SYMBOL(__napi_complete);
3745 void napi_complete(struct napi_struct *n)
3747 unsigned long flags;
3750 * don't let napi dequeue from the cpu poll list
3751 * just in case its running on a different cpu
3753 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3757 local_irq_save(flags);
3759 local_irq_restore(flags);
3761 EXPORT_SYMBOL(napi_complete);
3763 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3764 int (*poll)(struct napi_struct *, int), int weight)
3766 INIT_LIST_HEAD(&napi->poll_list);
3767 napi->gro_count = 0;
3768 napi->gro_list = NULL;
3771 napi->weight = weight;
3772 list_add(&napi->dev_list, &dev->napi_list);
3774 #ifdef CONFIG_NETPOLL
3775 spin_lock_init(&napi->poll_lock);
3776 napi->poll_owner = -1;
3778 set_bit(NAPI_STATE_SCHED, &napi->state);
3780 EXPORT_SYMBOL(netif_napi_add);
3782 void netif_napi_del(struct napi_struct *napi)
3784 struct sk_buff *skb, *next;
3786 list_del_init(&napi->dev_list);
3787 napi_free_frags(napi);
3789 for (skb = napi->gro_list; skb; skb = next) {
3795 napi->gro_list = NULL;
3796 napi->gro_count = 0;
3798 EXPORT_SYMBOL(netif_napi_del);
3800 static void net_rx_action(struct softirq_action *h)
3802 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3803 unsigned long time_limit = jiffies + 2;
3804 int budget = netdev_budget;
3807 local_irq_disable();
3809 while (!list_empty(&sd->poll_list)) {
3810 struct napi_struct *n;
3813 /* If softirq window is exhuasted then punt.
3814 * Allow this to run for 2 jiffies since which will allow
3815 * an average latency of 1.5/HZ.
3817 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3822 /* Even though interrupts have been re-enabled, this
3823 * access is safe because interrupts can only add new
3824 * entries to the tail of this list, and only ->poll()
3825 * calls can remove this head entry from the list.
3827 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3829 have = netpoll_poll_lock(n);
3833 /* This NAPI_STATE_SCHED test is for avoiding a race
3834 * with netpoll's poll_napi(). Only the entity which
3835 * obtains the lock and sees NAPI_STATE_SCHED set will
3836 * actually make the ->poll() call. Therefore we avoid
3837 * accidentally calling ->poll() when NAPI is not scheduled.
3840 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3841 work = n->poll(n, weight);
3845 WARN_ON_ONCE(work > weight);
3849 local_irq_disable();
3851 /* Drivers must not modify the NAPI state if they
3852 * consume the entire weight. In such cases this code
3853 * still "owns" the NAPI instance and therefore can
3854 * move the instance around on the list at-will.
3856 if (unlikely(work == weight)) {
3857 if (unlikely(napi_disable_pending(n))) {
3860 local_irq_disable();
3862 list_move_tail(&n->poll_list, &sd->poll_list);
3865 netpoll_poll_unlock(have);
3868 net_rps_action_and_irq_enable(sd);
3870 #ifdef CONFIG_NET_DMA
3872 * There may not be any more sk_buffs coming right now, so push
3873 * any pending DMA copies to hardware
3875 dma_issue_pending_all();
3882 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3886 static gifconf_func_t *gifconf_list[NPROTO];
3889 * register_gifconf - register a SIOCGIF handler
3890 * @family: Address family
3891 * @gifconf: Function handler
3893 * Register protocol dependent address dumping routines. The handler
3894 * that is passed must not be freed or reused until it has been replaced
3895 * by another handler.
3897 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3899 if (family >= NPROTO)
3901 gifconf_list[family] = gifconf;
3904 EXPORT_SYMBOL(register_gifconf);
3908 * Map an interface index to its name (SIOCGIFNAME)
3912 * We need this ioctl for efficient implementation of the
3913 * if_indextoname() function required by the IPv6 API. Without
3914 * it, we would have to search all the interfaces to find a
3918 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3920 struct net_device *dev;
3924 * Fetch the caller's info block.
3927 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3931 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3937 strcpy(ifr.ifr_name, dev->name);
3940 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3946 * Perform a SIOCGIFCONF call. This structure will change
3947 * size eventually, and there is nothing I can do about it.
3948 * Thus we will need a 'compatibility mode'.
3951 static int dev_ifconf(struct net *net, char __user *arg)
3954 struct net_device *dev;
3961 * Fetch the caller's info block.
3964 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3971 * Loop over the interfaces, and write an info block for each.
3975 for_each_netdev(net, dev) {
3976 for (i = 0; i < NPROTO; i++) {
3977 if (gifconf_list[i]) {
3980 done = gifconf_list[i](dev, NULL, 0);
3982 done = gifconf_list[i](dev, pos + total,
3992 * All done. Write the updated control block back to the caller.
3994 ifc.ifc_len = total;
3997 * Both BSD and Solaris return 0 here, so we do too.
3999 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4002 #ifdef CONFIG_PROC_FS
4004 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4006 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4007 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4008 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4010 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4012 struct net *net = seq_file_net(seq);
4013 struct net_device *dev;
4014 struct hlist_node *p;
4015 struct hlist_head *h;
4016 unsigned int count = 0, offset = get_offset(*pos);
4018 h = &net->dev_name_head[get_bucket(*pos)];
4019 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4020 if (++count == offset)
4027 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4029 struct net_device *dev;
4030 unsigned int bucket;
4033 dev = dev_from_same_bucket(seq, pos);
4037 bucket = get_bucket(*pos) + 1;
4038 *pos = set_bucket_offset(bucket, 1);
4039 } while (bucket < NETDEV_HASHENTRIES);
4045 * This is invoked by the /proc filesystem handler to display a device
4048 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4053 return SEQ_START_TOKEN;
4055 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4058 return dev_from_bucket(seq, pos);
4061 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4064 return dev_from_bucket(seq, pos);
4067 void dev_seq_stop(struct seq_file *seq, void *v)
4073 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4075 struct rtnl_link_stats64 temp;
4076 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4078 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4079 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4080 dev->name, stats->rx_bytes, stats->rx_packets,
4082 stats->rx_dropped + stats->rx_missed_errors,
4083 stats->rx_fifo_errors,
4084 stats->rx_length_errors + stats->rx_over_errors +
4085 stats->rx_crc_errors + stats->rx_frame_errors,
4086 stats->rx_compressed, stats->multicast,
4087 stats->tx_bytes, stats->tx_packets,
4088 stats->tx_errors, stats->tx_dropped,
4089 stats->tx_fifo_errors, stats->collisions,
4090 stats->tx_carrier_errors +
4091 stats->tx_aborted_errors +
4092 stats->tx_window_errors +
4093 stats->tx_heartbeat_errors,
4094 stats->tx_compressed);
4098 * Called from the PROCfs module. This now uses the new arbitrary sized
4099 * /proc/net interface to create /proc/net/dev
4101 static int dev_seq_show(struct seq_file *seq, void *v)
4103 if (v == SEQ_START_TOKEN)
4104 seq_puts(seq, "Inter-| Receive "
4106 " face |bytes packets errs drop fifo frame "
4107 "compressed multicast|bytes packets errs "
4108 "drop fifo colls carrier compressed\n");
4110 dev_seq_printf_stats(seq, v);
4114 static struct softnet_data *softnet_get_online(loff_t *pos)
4116 struct softnet_data *sd = NULL;
4118 while (*pos < nr_cpu_ids)
4119 if (cpu_online(*pos)) {
4120 sd = &per_cpu(softnet_data, *pos);
4127 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4129 return softnet_get_online(pos);
4132 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4135 return softnet_get_online(pos);
4138 static void softnet_seq_stop(struct seq_file *seq, void *v)
4142 static int softnet_seq_show(struct seq_file *seq, void *v)
4144 struct softnet_data *sd = v;
4146 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4147 sd->processed, sd->dropped, sd->time_squeeze, 0,
4148 0, 0, 0, 0, /* was fastroute */
4149 sd->cpu_collision, sd->received_rps);
4153 static const struct seq_operations dev_seq_ops = {
4154 .start = dev_seq_start,
4155 .next = dev_seq_next,
4156 .stop = dev_seq_stop,
4157 .show = dev_seq_show,
4160 static int dev_seq_open(struct inode *inode, struct file *file)
4162 return seq_open_net(inode, file, &dev_seq_ops,
4163 sizeof(struct seq_net_private));
4166 static const struct file_operations dev_seq_fops = {
4167 .owner = THIS_MODULE,
4168 .open = dev_seq_open,
4170 .llseek = seq_lseek,
4171 .release = seq_release_net,
4174 static const struct seq_operations softnet_seq_ops = {
4175 .start = softnet_seq_start,
4176 .next = softnet_seq_next,
4177 .stop = softnet_seq_stop,
4178 .show = softnet_seq_show,
4181 static int softnet_seq_open(struct inode *inode, struct file *file)
4183 return seq_open(file, &softnet_seq_ops);
4186 static const struct file_operations softnet_seq_fops = {
4187 .owner = THIS_MODULE,
4188 .open = softnet_seq_open,
4190 .llseek = seq_lseek,
4191 .release = seq_release,
4194 static void *ptype_get_idx(loff_t pos)
4196 struct packet_type *pt = NULL;
4200 list_for_each_entry_rcu(pt, &ptype_all, list) {
4206 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4207 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4216 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4220 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4223 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4225 struct packet_type *pt;
4226 struct list_head *nxt;
4230 if (v == SEQ_START_TOKEN)
4231 return ptype_get_idx(0);
4234 nxt = pt->list.next;
4235 if (pt->type == htons(ETH_P_ALL)) {
4236 if (nxt != &ptype_all)
4239 nxt = ptype_base[0].next;
4241 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4243 while (nxt == &ptype_base[hash]) {
4244 if (++hash >= PTYPE_HASH_SIZE)
4246 nxt = ptype_base[hash].next;
4249 return list_entry(nxt, struct packet_type, list);
4252 static void ptype_seq_stop(struct seq_file *seq, void *v)
4258 static int ptype_seq_show(struct seq_file *seq, void *v)
4260 struct packet_type *pt = v;
4262 if (v == SEQ_START_TOKEN)
4263 seq_puts(seq, "Type Device Function\n");
4264 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4265 if (pt->type == htons(ETH_P_ALL))
4266 seq_puts(seq, "ALL ");
4268 seq_printf(seq, "%04x", ntohs(pt->type));
4270 seq_printf(seq, " %-8s %pF\n",
4271 pt->dev ? pt->dev->name : "", pt->func);
4277 static const struct seq_operations ptype_seq_ops = {
4278 .start = ptype_seq_start,
4279 .next = ptype_seq_next,
4280 .stop = ptype_seq_stop,
4281 .show = ptype_seq_show,
4284 static int ptype_seq_open(struct inode *inode, struct file *file)
4286 return seq_open_net(inode, file, &ptype_seq_ops,
4287 sizeof(struct seq_net_private));
4290 static const struct file_operations ptype_seq_fops = {
4291 .owner = THIS_MODULE,
4292 .open = ptype_seq_open,
4294 .llseek = seq_lseek,
4295 .release = seq_release_net,
4299 static int __net_init dev_proc_net_init(struct net *net)
4303 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4305 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4307 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4310 if (wext_proc_init(net))
4316 proc_net_remove(net, "ptype");
4318 proc_net_remove(net, "softnet_stat");
4320 proc_net_remove(net, "dev");
4324 static void __net_exit dev_proc_net_exit(struct net *net)
4326 wext_proc_exit(net);
4328 proc_net_remove(net, "ptype");
4329 proc_net_remove(net, "softnet_stat");
4330 proc_net_remove(net, "dev");
4333 static struct pernet_operations __net_initdata dev_proc_ops = {
4334 .init = dev_proc_net_init,
4335 .exit = dev_proc_net_exit,
4338 static int __init dev_proc_init(void)
4340 return register_pernet_subsys(&dev_proc_ops);
4343 #define dev_proc_init() 0
4344 #endif /* CONFIG_PROC_FS */
4348 * netdev_set_master - set up master pointer
4349 * @slave: slave device
4350 * @master: new master device
4352 * Changes the master device of the slave. Pass %NULL to break the
4353 * bonding. The caller must hold the RTNL semaphore. On a failure
4354 * a negative errno code is returned. On success the reference counts
4355 * are adjusted and the function returns zero.
4357 int netdev_set_master(struct net_device *slave, struct net_device *master)
4359 struct net_device *old = slave->master;
4369 slave->master = master;
4375 EXPORT_SYMBOL(netdev_set_master);
4378 * netdev_set_bond_master - set up bonding master/slave pair
4379 * @slave: slave device
4380 * @master: new master device
4382 * Changes the master device of the slave. Pass %NULL to break the
4383 * bonding. The caller must hold the RTNL semaphore. On a failure
4384 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4385 * to the routing socket and the function returns zero.
4387 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4393 err = netdev_set_master(slave, master);
4397 slave->flags |= IFF_SLAVE;
4399 slave->flags &= ~IFF_SLAVE;
4401 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4404 EXPORT_SYMBOL(netdev_set_bond_master);
4406 static void dev_change_rx_flags(struct net_device *dev, int flags)
4408 const struct net_device_ops *ops = dev->netdev_ops;
4410 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4411 ops->ndo_change_rx_flags(dev, flags);
4414 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4416 unsigned int old_flags = dev->flags;
4422 dev->flags |= IFF_PROMISC;
4423 dev->promiscuity += inc;
4424 if (dev->promiscuity == 0) {
4427 * If inc causes overflow, untouch promisc and return error.
4430 dev->flags &= ~IFF_PROMISC;
4432 dev->promiscuity -= inc;
4433 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4438 if (dev->flags != old_flags) {
4439 pr_info("device %s %s promiscuous mode\n",
4441 dev->flags & IFF_PROMISC ? "entered" : "left");
4442 if (audit_enabled) {
4443 current_uid_gid(&uid, &gid);
4444 audit_log(current->audit_context, GFP_ATOMIC,
4445 AUDIT_ANOM_PROMISCUOUS,
4446 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4447 dev->name, (dev->flags & IFF_PROMISC),
4448 (old_flags & IFF_PROMISC),
4449 audit_get_loginuid(current),
4451 audit_get_sessionid(current));
4454 dev_change_rx_flags(dev, IFF_PROMISC);
4460 * dev_set_promiscuity - update promiscuity count on a device
4464 * Add or remove promiscuity from a device. While the count in the device
4465 * remains above zero the interface remains promiscuous. Once it hits zero
4466 * the device reverts back to normal filtering operation. A negative inc
4467 * value is used to drop promiscuity on the device.
4468 * Return 0 if successful or a negative errno code on error.
4470 int dev_set_promiscuity(struct net_device *dev, int inc)
4472 unsigned int old_flags = dev->flags;
4475 err = __dev_set_promiscuity(dev, inc);
4478 if (dev->flags != old_flags)
4479 dev_set_rx_mode(dev);
4482 EXPORT_SYMBOL(dev_set_promiscuity);
4485 * dev_set_allmulti - update allmulti count on a device
4489 * Add or remove reception of all multicast frames to a device. While the
4490 * count in the device remains above zero the interface remains listening
4491 * to all interfaces. Once it hits zero the device reverts back to normal
4492 * filtering operation. A negative @inc value is used to drop the counter
4493 * when releasing a resource needing all multicasts.
4494 * Return 0 if successful or a negative errno code on error.
4497 int dev_set_allmulti(struct net_device *dev, int inc)
4499 unsigned int old_flags = dev->flags;
4503 dev->flags |= IFF_ALLMULTI;
4504 dev->allmulti += inc;
4505 if (dev->allmulti == 0) {
4508 * If inc causes overflow, untouch allmulti and return error.
4511 dev->flags &= ~IFF_ALLMULTI;
4513 dev->allmulti -= inc;
4514 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4519 if (dev->flags ^ old_flags) {
4520 dev_change_rx_flags(dev, IFF_ALLMULTI);
4521 dev_set_rx_mode(dev);
4525 EXPORT_SYMBOL(dev_set_allmulti);
4528 * Upload unicast and multicast address lists to device and
4529 * configure RX filtering. When the device doesn't support unicast
4530 * filtering it is put in promiscuous mode while unicast addresses
4533 void __dev_set_rx_mode(struct net_device *dev)
4535 const struct net_device_ops *ops = dev->netdev_ops;
4537 /* dev_open will call this function so the list will stay sane. */
4538 if (!(dev->flags&IFF_UP))
4541 if (!netif_device_present(dev))
4544 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4545 /* Unicast addresses changes may only happen under the rtnl,
4546 * therefore calling __dev_set_promiscuity here is safe.
4548 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4549 __dev_set_promiscuity(dev, 1);
4550 dev->uc_promisc = true;
4551 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4552 __dev_set_promiscuity(dev, -1);
4553 dev->uc_promisc = false;
4557 if (ops->ndo_set_rx_mode)
4558 ops->ndo_set_rx_mode(dev);
4561 void dev_set_rx_mode(struct net_device *dev)
4563 netif_addr_lock_bh(dev);
4564 __dev_set_rx_mode(dev);
4565 netif_addr_unlock_bh(dev);
4569 * dev_get_flags - get flags reported to userspace
4572 * Get the combination of flag bits exported through APIs to userspace.
4574 unsigned dev_get_flags(const struct net_device *dev)
4578 flags = (dev->flags & ~(IFF_PROMISC |
4583 (dev->gflags & (IFF_PROMISC |
4586 if (netif_running(dev)) {
4587 if (netif_oper_up(dev))
4588 flags |= IFF_RUNNING;
4589 if (netif_carrier_ok(dev))
4590 flags |= IFF_LOWER_UP;
4591 if (netif_dormant(dev))
4592 flags |= IFF_DORMANT;
4597 EXPORT_SYMBOL(dev_get_flags);
4599 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4601 unsigned int old_flags = dev->flags;
4607 * Set the flags on our device.
4610 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4611 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4613 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4617 * Load in the correct multicast list now the flags have changed.
4620 if ((old_flags ^ flags) & IFF_MULTICAST)
4621 dev_change_rx_flags(dev, IFF_MULTICAST);
4623 dev_set_rx_mode(dev);
4626 * Have we downed the interface. We handle IFF_UP ourselves
4627 * according to user attempts to set it, rather than blindly
4632 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4633 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4636 dev_set_rx_mode(dev);
4639 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4640 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4642 dev->gflags ^= IFF_PROMISC;
4643 dev_set_promiscuity(dev, inc);
4646 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4647 is important. Some (broken) drivers set IFF_PROMISC, when
4648 IFF_ALLMULTI is requested not asking us and not reporting.
4650 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4651 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4653 dev->gflags ^= IFF_ALLMULTI;
4654 dev_set_allmulti(dev, inc);
4660 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4662 unsigned int changes = dev->flags ^ old_flags;
4664 if (changes & IFF_UP) {
4665 if (dev->flags & IFF_UP)
4666 call_netdevice_notifiers(NETDEV_UP, dev);
4668 call_netdevice_notifiers(NETDEV_DOWN, dev);
4671 if (dev->flags & IFF_UP &&
4672 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4673 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4677 * dev_change_flags - change device settings
4679 * @flags: device state flags
4681 * Change settings on device based state flags. The flags are
4682 * in the userspace exported format.
4684 int dev_change_flags(struct net_device *dev, unsigned int flags)
4687 unsigned int changes, old_flags = dev->flags;
4689 ret = __dev_change_flags(dev, flags);
4693 changes = old_flags ^ dev->flags;
4695 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4697 __dev_notify_flags(dev, old_flags);
4700 EXPORT_SYMBOL(dev_change_flags);
4703 * dev_set_mtu - Change maximum transfer unit
4705 * @new_mtu: new transfer unit
4707 * Change the maximum transfer size of the network device.
4709 int dev_set_mtu(struct net_device *dev, int new_mtu)
4711 const struct net_device_ops *ops = dev->netdev_ops;
4714 if (new_mtu == dev->mtu)
4717 /* MTU must be positive. */
4721 if (!netif_device_present(dev))
4725 if (ops->ndo_change_mtu)
4726 err = ops->ndo_change_mtu(dev, new_mtu);
4730 if (!err && dev->flags & IFF_UP)
4731 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4734 EXPORT_SYMBOL(dev_set_mtu);
4737 * dev_set_group - Change group this device belongs to
4739 * @new_group: group this device should belong to
4741 void dev_set_group(struct net_device *dev, int new_group)
4743 dev->group = new_group;
4745 EXPORT_SYMBOL(dev_set_group);
4748 * dev_set_mac_address - Change Media Access Control Address
4752 * Change the hardware (MAC) address of the device
4754 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4756 const struct net_device_ops *ops = dev->netdev_ops;
4759 if (!ops->ndo_set_mac_address)
4761 if (sa->sa_family != dev->type)
4763 if (!netif_device_present(dev))
4765 err = ops->ndo_set_mac_address(dev, sa);
4767 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4770 EXPORT_SYMBOL(dev_set_mac_address);
4773 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4775 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4778 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4784 case SIOCGIFFLAGS: /* Get interface flags */
4785 ifr->ifr_flags = (short) dev_get_flags(dev);
4788 case SIOCGIFMETRIC: /* Get the metric on the interface
4789 (currently unused) */
4790 ifr->ifr_metric = 0;
4793 case SIOCGIFMTU: /* Get the MTU of a device */
4794 ifr->ifr_mtu = dev->mtu;
4799 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4801 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4802 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4803 ifr->ifr_hwaddr.sa_family = dev->type;
4811 ifr->ifr_map.mem_start = dev->mem_start;
4812 ifr->ifr_map.mem_end = dev->mem_end;
4813 ifr->ifr_map.base_addr = dev->base_addr;
4814 ifr->ifr_map.irq = dev->irq;
4815 ifr->ifr_map.dma = dev->dma;
4816 ifr->ifr_map.port = dev->if_port;
4820 ifr->ifr_ifindex = dev->ifindex;
4824 ifr->ifr_qlen = dev->tx_queue_len;
4828 /* dev_ioctl() should ensure this case
4840 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4842 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4845 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4846 const struct net_device_ops *ops;
4851 ops = dev->netdev_ops;
4854 case SIOCSIFFLAGS: /* Set interface flags */
4855 return dev_change_flags(dev, ifr->ifr_flags);
4857 case SIOCSIFMETRIC: /* Set the metric on the interface
4858 (currently unused) */
4861 case SIOCSIFMTU: /* Set the MTU of a device */
4862 return dev_set_mtu(dev, ifr->ifr_mtu);
4865 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4867 case SIOCSIFHWBROADCAST:
4868 if (ifr->ifr_hwaddr.sa_family != dev->type)
4870 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4871 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4872 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4876 if (ops->ndo_set_config) {
4877 if (!netif_device_present(dev))
4879 return ops->ndo_set_config(dev, &ifr->ifr_map);
4884 if (!ops->ndo_set_rx_mode ||
4885 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4887 if (!netif_device_present(dev))
4889 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4892 if (!ops->ndo_set_rx_mode ||
4893 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4895 if (!netif_device_present(dev))
4897 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4900 if (ifr->ifr_qlen < 0)
4902 dev->tx_queue_len = ifr->ifr_qlen;
4906 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4907 return dev_change_name(dev, ifr->ifr_newname);
4910 err = net_hwtstamp_validate(ifr);
4916 * Unknown or private ioctl
4919 if ((cmd >= SIOCDEVPRIVATE &&
4920 cmd <= SIOCDEVPRIVATE + 15) ||
4921 cmd == SIOCBONDENSLAVE ||
4922 cmd == SIOCBONDRELEASE ||
4923 cmd == SIOCBONDSETHWADDR ||
4924 cmd == SIOCBONDSLAVEINFOQUERY ||
4925 cmd == SIOCBONDINFOQUERY ||
4926 cmd == SIOCBONDCHANGEACTIVE ||
4927 cmd == SIOCGMIIPHY ||
4928 cmd == SIOCGMIIREG ||
4929 cmd == SIOCSMIIREG ||
4930 cmd == SIOCBRADDIF ||
4931 cmd == SIOCBRDELIF ||
4932 cmd == SIOCSHWTSTAMP ||
4933 cmd == SIOCWANDEV) {
4935 if (ops->ndo_do_ioctl) {
4936 if (netif_device_present(dev))
4937 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4949 * This function handles all "interface"-type I/O control requests. The actual
4950 * 'doing' part of this is dev_ifsioc above.
4954 * dev_ioctl - network device ioctl
4955 * @net: the applicable net namespace
4956 * @cmd: command to issue
4957 * @arg: pointer to a struct ifreq in user space
4959 * Issue ioctl functions to devices. This is normally called by the
4960 * user space syscall interfaces but can sometimes be useful for
4961 * other purposes. The return value is the return from the syscall if
4962 * positive or a negative errno code on error.
4965 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4971 /* One special case: SIOCGIFCONF takes ifconf argument
4972 and requires shared lock, because it sleeps writing
4976 if (cmd == SIOCGIFCONF) {
4978 ret = dev_ifconf(net, (char __user *) arg);
4982 if (cmd == SIOCGIFNAME)
4983 return dev_ifname(net, (struct ifreq __user *)arg);
4985 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4988 ifr.ifr_name[IFNAMSIZ-1] = 0;
4990 colon = strchr(ifr.ifr_name, ':');
4995 * See which interface the caller is talking about.
5000 * These ioctl calls:
5001 * - can be done by all.
5002 * - atomic and do not require locking.
5013 dev_load(net, ifr.ifr_name);
5015 ret = dev_ifsioc_locked(net, &ifr, cmd);
5020 if (copy_to_user(arg, &ifr,
5021 sizeof(struct ifreq)))
5027 dev_load(net, ifr.ifr_name);
5029 ret = dev_ethtool(net, &ifr);
5034 if (copy_to_user(arg, &ifr,
5035 sizeof(struct ifreq)))
5041 * These ioctl calls:
5042 * - require superuser power.
5043 * - require strict serialization.
5049 if (!capable(CAP_NET_ADMIN))
5051 dev_load(net, ifr.ifr_name);
5053 ret = dev_ifsioc(net, &ifr, cmd);
5058 if (copy_to_user(arg, &ifr,
5059 sizeof(struct ifreq)))
5065 * These ioctl calls:
5066 * - require superuser power.
5067 * - require strict serialization.
5068 * - do not return a value
5078 case SIOCSIFHWBROADCAST:
5081 case SIOCBONDENSLAVE:
5082 case SIOCBONDRELEASE:
5083 case SIOCBONDSETHWADDR:
5084 case SIOCBONDCHANGEACTIVE:
5088 if (!capable(CAP_NET_ADMIN))
5091 case SIOCBONDSLAVEINFOQUERY:
5092 case SIOCBONDINFOQUERY:
5093 dev_load(net, ifr.ifr_name);
5095 ret = dev_ifsioc(net, &ifr, cmd);
5100 /* Get the per device memory space. We can add this but
5101 * currently do not support it */
5103 /* Set the per device memory buffer space.
5104 * Not applicable in our case */
5109 * Unknown or private ioctl.
5112 if (cmd == SIOCWANDEV ||
5113 (cmd >= SIOCDEVPRIVATE &&
5114 cmd <= SIOCDEVPRIVATE + 15)) {
5115 dev_load(net, ifr.ifr_name);
5117 ret = dev_ifsioc(net, &ifr, cmd);
5119 if (!ret && copy_to_user(arg, &ifr,
5120 sizeof(struct ifreq)))
5124 /* Take care of Wireless Extensions */
5125 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5126 return wext_handle_ioctl(net, &ifr, cmd, arg);
5133 * dev_new_index - allocate an ifindex
5134 * @net: the applicable net namespace
5136 * Returns a suitable unique value for a new device interface
5137 * number. The caller must hold the rtnl semaphore or the
5138 * dev_base_lock to be sure it remains unique.
5140 static int dev_new_index(struct net *net)
5146 if (!__dev_get_by_index(net, ifindex))
5151 /* Delayed registration/unregisteration */
5152 static LIST_HEAD(net_todo_list);
5154 static void net_set_todo(struct net_device *dev)
5156 list_add_tail(&dev->todo_list, &net_todo_list);
5159 static void rollback_registered_many(struct list_head *head)
5161 struct net_device *dev, *tmp;
5163 BUG_ON(dev_boot_phase);
5166 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5167 /* Some devices call without registering
5168 * for initialization unwind. Remove those
5169 * devices and proceed with the remaining.
5171 if (dev->reg_state == NETREG_UNINITIALIZED) {
5172 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5176 list_del(&dev->unreg_list);
5179 dev->dismantle = true;
5180 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5183 /* If device is running, close it first. */
5184 dev_close_many(head);
5186 list_for_each_entry(dev, head, unreg_list) {
5187 /* And unlink it from device chain. */
5188 unlist_netdevice(dev);
5190 dev->reg_state = NETREG_UNREGISTERING;
5195 list_for_each_entry(dev, head, unreg_list) {
5196 /* Shutdown queueing discipline. */
5200 /* Notify protocols, that we are about to destroy
5201 this device. They should clean all the things.
5203 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5205 if (!dev->rtnl_link_ops ||
5206 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5207 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5210 * Flush the unicast and multicast chains
5215 if (dev->netdev_ops->ndo_uninit)
5216 dev->netdev_ops->ndo_uninit(dev);
5218 /* Notifier chain MUST detach us from master device. */
5219 WARN_ON(dev->master);
5221 /* Remove entries from kobject tree */
5222 netdev_unregister_kobject(dev);
5225 /* Process any work delayed until the end of the batch */
5226 dev = list_first_entry(head, struct net_device, unreg_list);
5227 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5231 list_for_each_entry(dev, head, unreg_list)
5235 static void rollback_registered(struct net_device *dev)
5239 list_add(&dev->unreg_list, &single);
5240 rollback_registered_many(&single);
5244 static netdev_features_t netdev_fix_features(struct net_device *dev,
5245 netdev_features_t features)
5247 /* Fix illegal checksum combinations */
5248 if ((features & NETIF_F_HW_CSUM) &&
5249 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5250 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5251 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5254 /* Fix illegal SG+CSUM combinations. */
5255 if ((features & NETIF_F_SG) &&
5256 !(features & NETIF_F_ALL_CSUM)) {
5258 "Dropping NETIF_F_SG since no checksum feature.\n");
5259 features &= ~NETIF_F_SG;
5262 /* TSO requires that SG is present as well. */
5263 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5264 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5265 features &= ~NETIF_F_ALL_TSO;
5268 /* TSO ECN requires that TSO is present as well. */
5269 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5270 features &= ~NETIF_F_TSO_ECN;
5272 /* Software GSO depends on SG. */
5273 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5274 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5275 features &= ~NETIF_F_GSO;
5278 /* UFO needs SG and checksumming */
5279 if (features & NETIF_F_UFO) {
5280 /* maybe split UFO into V4 and V6? */
5281 if (!((features & NETIF_F_GEN_CSUM) ||
5282 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5283 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5285 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5286 features &= ~NETIF_F_UFO;
5289 if (!(features & NETIF_F_SG)) {
5291 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5292 features &= ~NETIF_F_UFO;
5299 int __netdev_update_features(struct net_device *dev)
5301 netdev_features_t features;
5306 features = netdev_get_wanted_features(dev);
5308 if (dev->netdev_ops->ndo_fix_features)
5309 features = dev->netdev_ops->ndo_fix_features(dev, features);
5311 /* driver might be less strict about feature dependencies */
5312 features = netdev_fix_features(dev, features);
5314 if (dev->features == features)
5317 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5318 &dev->features, &features);
5320 if (dev->netdev_ops->ndo_set_features)
5321 err = dev->netdev_ops->ndo_set_features(dev, features);
5323 if (unlikely(err < 0)) {
5325 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5326 err, &features, &dev->features);
5331 dev->features = features;
5337 * netdev_update_features - recalculate device features
5338 * @dev: the device to check
5340 * Recalculate dev->features set and send notifications if it
5341 * has changed. Should be called after driver or hardware dependent
5342 * conditions might have changed that influence the features.
5344 void netdev_update_features(struct net_device *dev)
5346 if (__netdev_update_features(dev))
5347 netdev_features_change(dev);
5349 EXPORT_SYMBOL(netdev_update_features);
5352 * netdev_change_features - recalculate device features
5353 * @dev: the device to check
5355 * Recalculate dev->features set and send notifications even
5356 * if they have not changed. Should be called instead of
5357 * netdev_update_features() if also dev->vlan_features might
5358 * have changed to allow the changes to be propagated to stacked
5361 void netdev_change_features(struct net_device *dev)
5363 __netdev_update_features(dev);
5364 netdev_features_change(dev);
5366 EXPORT_SYMBOL(netdev_change_features);
5369 * netif_stacked_transfer_operstate - transfer operstate
5370 * @rootdev: the root or lower level device to transfer state from
5371 * @dev: the device to transfer operstate to
5373 * Transfer operational state from root to device. This is normally
5374 * called when a stacking relationship exists between the root
5375 * device and the device(a leaf device).
5377 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5378 struct net_device *dev)
5380 if (rootdev->operstate == IF_OPER_DORMANT)
5381 netif_dormant_on(dev);
5383 netif_dormant_off(dev);
5385 if (netif_carrier_ok(rootdev)) {
5386 if (!netif_carrier_ok(dev))
5387 netif_carrier_on(dev);
5389 if (netif_carrier_ok(dev))
5390 netif_carrier_off(dev);
5393 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5396 static int netif_alloc_rx_queues(struct net_device *dev)
5398 unsigned int i, count = dev->num_rx_queues;
5399 struct netdev_rx_queue *rx;
5403 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5405 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5410 for (i = 0; i < count; i++)
5416 static void netdev_init_one_queue(struct net_device *dev,
5417 struct netdev_queue *queue, void *_unused)
5419 /* Initialize queue lock */
5420 spin_lock_init(&queue->_xmit_lock);
5421 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5422 queue->xmit_lock_owner = -1;
5423 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5426 dql_init(&queue->dql, HZ);
5430 static int netif_alloc_netdev_queues(struct net_device *dev)
5432 unsigned int count = dev->num_tx_queues;
5433 struct netdev_queue *tx;
5437 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5439 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5444 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5445 spin_lock_init(&dev->tx_global_lock);
5451 * register_netdevice - register a network device
5452 * @dev: device to register
5454 * Take a completed network device structure and add it to the kernel
5455 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5456 * chain. 0 is returned on success. A negative errno code is returned
5457 * on a failure to set up the device, or if the name is a duplicate.
5459 * Callers must hold the rtnl semaphore. You may want
5460 * register_netdev() instead of this.
5463 * The locking appears insufficient to guarantee two parallel registers
5464 * will not get the same name.
5467 int register_netdevice(struct net_device *dev)
5470 struct net *net = dev_net(dev);
5472 BUG_ON(dev_boot_phase);
5477 /* When net_device's are persistent, this will be fatal. */
5478 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5481 spin_lock_init(&dev->addr_list_lock);
5482 netdev_set_addr_lockdep_class(dev);
5486 ret = dev_get_valid_name(dev, dev->name);
5490 /* Init, if this function is available */
5491 if (dev->netdev_ops->ndo_init) {
5492 ret = dev->netdev_ops->ndo_init(dev);
5500 dev->ifindex = dev_new_index(net);
5501 if (dev->iflink == -1)
5502 dev->iflink = dev->ifindex;
5504 /* Transfer changeable features to wanted_features and enable
5505 * software offloads (GSO and GRO).
5507 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5508 dev->features |= NETIF_F_SOFT_FEATURES;
5509 dev->wanted_features = dev->features & dev->hw_features;
5511 /* Turn on no cache copy if HW is doing checksum */
5512 if (!(dev->flags & IFF_LOOPBACK)) {
5513 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5514 if (dev->features & NETIF_F_ALL_CSUM) {
5515 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5516 dev->features |= NETIF_F_NOCACHE_COPY;
5520 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5522 dev->vlan_features |= NETIF_F_HIGHDMA;
5524 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5525 ret = notifier_to_errno(ret);
5529 ret = netdev_register_kobject(dev);
5532 dev->reg_state = NETREG_REGISTERED;
5534 __netdev_update_features(dev);
5537 * Default initial state at registry is that the
5538 * device is present.
5541 set_bit(__LINK_STATE_PRESENT, &dev->state);
5543 dev_init_scheduler(dev);
5545 list_netdevice(dev);
5547 /* Notify protocols, that a new device appeared. */
5548 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5549 ret = notifier_to_errno(ret);
5551 rollback_registered(dev);
5552 dev->reg_state = NETREG_UNREGISTERED;
5555 * Prevent userspace races by waiting until the network
5556 * device is fully setup before sending notifications.
5558 if (!dev->rtnl_link_ops ||
5559 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5560 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5566 if (dev->netdev_ops->ndo_uninit)
5567 dev->netdev_ops->ndo_uninit(dev);
5570 EXPORT_SYMBOL(register_netdevice);
5573 * init_dummy_netdev - init a dummy network device for NAPI
5574 * @dev: device to init
5576 * This takes a network device structure and initialize the minimum
5577 * amount of fields so it can be used to schedule NAPI polls without
5578 * registering a full blown interface. This is to be used by drivers
5579 * that need to tie several hardware interfaces to a single NAPI
5580 * poll scheduler due to HW limitations.
5582 int init_dummy_netdev(struct net_device *dev)
5584 /* Clear everything. Note we don't initialize spinlocks
5585 * are they aren't supposed to be taken by any of the
5586 * NAPI code and this dummy netdev is supposed to be
5587 * only ever used for NAPI polls
5589 memset(dev, 0, sizeof(struct net_device));
5591 /* make sure we BUG if trying to hit standard
5592 * register/unregister code path
5594 dev->reg_state = NETREG_DUMMY;
5596 /* NAPI wants this */
5597 INIT_LIST_HEAD(&dev->napi_list);
5599 /* a dummy interface is started by default */
5600 set_bit(__LINK_STATE_PRESENT, &dev->state);
5601 set_bit(__LINK_STATE_START, &dev->state);
5603 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5604 * because users of this 'device' dont need to change
5610 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5614 * register_netdev - register a network device
5615 * @dev: device to register
5617 * Take a completed network device structure and add it to the kernel
5618 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5619 * chain. 0 is returned on success. A negative errno code is returned
5620 * on a failure to set up the device, or if the name is a duplicate.
5622 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5623 * and expands the device name if you passed a format string to
5626 int register_netdev(struct net_device *dev)
5631 err = register_netdevice(dev);
5635 EXPORT_SYMBOL(register_netdev);
5637 int netdev_refcnt_read(const struct net_device *dev)
5641 for_each_possible_cpu(i)
5642 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5645 EXPORT_SYMBOL(netdev_refcnt_read);
5648 * netdev_wait_allrefs - wait until all references are gone.
5650 * This is called when unregistering network devices.
5652 * Any protocol or device that holds a reference should register
5653 * for netdevice notification, and cleanup and put back the
5654 * reference if they receive an UNREGISTER event.
5655 * We can get stuck here if buggy protocols don't correctly
5658 static void netdev_wait_allrefs(struct net_device *dev)
5660 unsigned long rebroadcast_time, warning_time;
5663 linkwatch_forget_dev(dev);
5665 rebroadcast_time = warning_time = jiffies;
5666 refcnt = netdev_refcnt_read(dev);
5668 while (refcnt != 0) {
5669 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5672 /* Rebroadcast unregister notification */
5673 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5674 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5675 * should have already handle it the first time */
5677 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5679 /* We must not have linkwatch events
5680 * pending on unregister. If this
5681 * happens, we simply run the queue
5682 * unscheduled, resulting in a noop
5685 linkwatch_run_queue();
5690 rebroadcast_time = jiffies;
5695 refcnt = netdev_refcnt_read(dev);
5697 if (time_after(jiffies, warning_time + 10 * HZ)) {
5698 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5700 warning_time = jiffies;
5709 * register_netdevice(x1);
5710 * register_netdevice(x2);
5712 * unregister_netdevice(y1);
5713 * unregister_netdevice(y2);
5719 * We are invoked by rtnl_unlock().
5720 * This allows us to deal with problems:
5721 * 1) We can delete sysfs objects which invoke hotplug
5722 * without deadlocking with linkwatch via keventd.
5723 * 2) Since we run with the RTNL semaphore not held, we can sleep
5724 * safely in order to wait for the netdev refcnt to drop to zero.
5726 * We must not return until all unregister events added during
5727 * the interval the lock was held have been completed.
5729 void netdev_run_todo(void)
5731 struct list_head list;
5733 /* Snapshot list, allow later requests */
5734 list_replace_init(&net_todo_list, &list);
5738 /* Wait for rcu callbacks to finish before attempting to drain
5739 * the device list. This usually avoids a 250ms wait.
5741 if (!list_empty(&list))
5744 while (!list_empty(&list)) {
5745 struct net_device *dev
5746 = list_first_entry(&list, struct net_device, todo_list);
5747 list_del(&dev->todo_list);
5749 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5750 pr_err("network todo '%s' but state %d\n",
5751 dev->name, dev->reg_state);
5756 dev->reg_state = NETREG_UNREGISTERED;
5758 on_each_cpu(flush_backlog, dev, 1);
5760 netdev_wait_allrefs(dev);
5763 BUG_ON(netdev_refcnt_read(dev));
5764 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5765 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5766 WARN_ON(dev->dn_ptr);
5768 if (dev->destructor)
5769 dev->destructor(dev);
5771 /* Free network device */
5772 kobject_put(&dev->dev.kobj);
5776 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5777 * fields in the same order, with only the type differing.
5779 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5780 const struct net_device_stats *netdev_stats)
5782 #if BITS_PER_LONG == 64
5783 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5784 memcpy(stats64, netdev_stats, sizeof(*stats64));
5786 size_t i, n = sizeof(*stats64) / sizeof(u64);
5787 const unsigned long *src = (const unsigned long *)netdev_stats;
5788 u64 *dst = (u64 *)stats64;
5790 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5791 sizeof(*stats64) / sizeof(u64));
5792 for (i = 0; i < n; i++)
5796 EXPORT_SYMBOL(netdev_stats_to_stats64);
5799 * dev_get_stats - get network device statistics
5800 * @dev: device to get statistics from
5801 * @storage: place to store stats
5803 * Get network statistics from device. Return @storage.
5804 * The device driver may provide its own method by setting
5805 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5806 * otherwise the internal statistics structure is used.
5808 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5809 struct rtnl_link_stats64 *storage)
5811 const struct net_device_ops *ops = dev->netdev_ops;
5813 if (ops->ndo_get_stats64) {
5814 memset(storage, 0, sizeof(*storage));
5815 ops->ndo_get_stats64(dev, storage);
5816 } else if (ops->ndo_get_stats) {
5817 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5819 netdev_stats_to_stats64(storage, &dev->stats);
5821 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5824 EXPORT_SYMBOL(dev_get_stats);
5826 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5828 struct netdev_queue *queue = dev_ingress_queue(dev);
5830 #ifdef CONFIG_NET_CLS_ACT
5833 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5836 netdev_init_one_queue(dev, queue, NULL);
5837 queue->qdisc = &noop_qdisc;
5838 queue->qdisc_sleeping = &noop_qdisc;
5839 rcu_assign_pointer(dev->ingress_queue, queue);
5845 * alloc_netdev_mqs - allocate network device
5846 * @sizeof_priv: size of private data to allocate space for
5847 * @name: device name format string
5848 * @setup: callback to initialize device
5849 * @txqs: the number of TX subqueues to allocate
5850 * @rxqs: the number of RX subqueues to allocate
5852 * Allocates a struct net_device with private data area for driver use
5853 * and performs basic initialization. Also allocates subquue structs
5854 * for each queue on the device.
5856 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5857 void (*setup)(struct net_device *),
5858 unsigned int txqs, unsigned int rxqs)
5860 struct net_device *dev;
5862 struct net_device *p;
5864 BUG_ON(strlen(name) >= sizeof(dev->name));
5867 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5873 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5878 alloc_size = sizeof(struct net_device);
5880 /* ensure 32-byte alignment of private area */
5881 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5882 alloc_size += sizeof_priv;
5884 /* ensure 32-byte alignment of whole construct */
5885 alloc_size += NETDEV_ALIGN - 1;
5887 p = kzalloc(alloc_size, GFP_KERNEL);
5889 pr_err("alloc_netdev: Unable to allocate device\n");
5893 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5894 dev->padded = (char *)dev - (char *)p;
5896 dev->pcpu_refcnt = alloc_percpu(int);
5897 if (!dev->pcpu_refcnt)
5900 if (dev_addr_init(dev))
5906 dev_net_set(dev, &init_net);
5908 dev->gso_max_size = GSO_MAX_SIZE;
5910 INIT_LIST_HEAD(&dev->napi_list);
5911 INIT_LIST_HEAD(&dev->unreg_list);
5912 INIT_LIST_HEAD(&dev->link_watch_list);
5913 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5916 dev->num_tx_queues = txqs;
5917 dev->real_num_tx_queues = txqs;
5918 if (netif_alloc_netdev_queues(dev))
5922 dev->num_rx_queues = rxqs;
5923 dev->real_num_rx_queues = rxqs;
5924 if (netif_alloc_rx_queues(dev))
5928 strcpy(dev->name, name);
5929 dev->group = INIT_NETDEV_GROUP;
5937 free_percpu(dev->pcpu_refcnt);
5947 EXPORT_SYMBOL(alloc_netdev_mqs);
5950 * free_netdev - free network device
5953 * This function does the last stage of destroying an allocated device
5954 * interface. The reference to the device object is released.
5955 * If this is the last reference then it will be freed.
5957 void free_netdev(struct net_device *dev)
5959 struct napi_struct *p, *n;
5961 release_net(dev_net(dev));
5968 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5970 /* Flush device addresses */
5971 dev_addr_flush(dev);
5973 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5976 free_percpu(dev->pcpu_refcnt);
5977 dev->pcpu_refcnt = NULL;
5979 /* Compatibility with error handling in drivers */
5980 if (dev->reg_state == NETREG_UNINITIALIZED) {
5981 kfree((char *)dev - dev->padded);
5985 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5986 dev->reg_state = NETREG_RELEASED;
5988 /* will free via device release */
5989 put_device(&dev->dev);
5991 EXPORT_SYMBOL(free_netdev);
5994 * synchronize_net - Synchronize with packet receive processing
5996 * Wait for packets currently being received to be done.
5997 * Does not block later packets from starting.
5999 void synchronize_net(void)
6002 if (rtnl_is_locked())
6003 synchronize_rcu_expedited();
6007 EXPORT_SYMBOL(synchronize_net);
6010 * unregister_netdevice_queue - remove device from the kernel
6014 * This function shuts down a device interface and removes it
6015 * from the kernel tables.
6016 * If head not NULL, device is queued to be unregistered later.
6018 * Callers must hold the rtnl semaphore. You may want
6019 * unregister_netdev() instead of this.
6022 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6027 list_move_tail(&dev->unreg_list, head);
6029 rollback_registered(dev);
6030 /* Finish processing unregister after unlock */
6034 EXPORT_SYMBOL(unregister_netdevice_queue);
6037 * unregister_netdevice_many - unregister many devices
6038 * @head: list of devices
6040 void unregister_netdevice_many(struct list_head *head)
6042 struct net_device *dev;
6044 if (!list_empty(head)) {
6045 rollback_registered_many(head);
6046 list_for_each_entry(dev, head, unreg_list)
6050 EXPORT_SYMBOL(unregister_netdevice_many);
6053 * unregister_netdev - remove device from the kernel
6056 * This function shuts down a device interface and removes it
6057 * from the kernel tables.
6059 * This is just a wrapper for unregister_netdevice that takes
6060 * the rtnl semaphore. In general you want to use this and not
6061 * unregister_netdevice.
6063 void unregister_netdev(struct net_device *dev)
6066 unregister_netdevice(dev);
6069 EXPORT_SYMBOL(unregister_netdev);
6072 * dev_change_net_namespace - move device to different nethost namespace
6074 * @net: network namespace
6075 * @pat: If not NULL name pattern to try if the current device name
6076 * is already taken in the destination network namespace.
6078 * This function shuts down a device interface and moves it
6079 * to a new network namespace. On success 0 is returned, on
6080 * a failure a netagive errno code is returned.
6082 * Callers must hold the rtnl semaphore.
6085 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6091 /* Don't allow namespace local devices to be moved. */
6093 if (dev->features & NETIF_F_NETNS_LOCAL)
6096 /* Ensure the device has been registrered */
6098 if (dev->reg_state != NETREG_REGISTERED)
6101 /* Get out if there is nothing todo */
6103 if (net_eq(dev_net(dev), net))
6106 /* Pick the destination device name, and ensure
6107 * we can use it in the destination network namespace.
6110 if (__dev_get_by_name(net, dev->name)) {
6111 /* We get here if we can't use the current device name */
6114 if (dev_get_valid_name(dev, pat) < 0)
6119 * And now a mini version of register_netdevice unregister_netdevice.
6122 /* If device is running close it first. */
6125 /* And unlink it from device chain */
6127 unlist_netdevice(dev);
6131 /* Shutdown queueing discipline. */
6134 /* Notify protocols, that we are about to destroy
6135 this device. They should clean all the things.
6137 Note that dev->reg_state stays at NETREG_REGISTERED.
6138 This is wanted because this way 8021q and macvlan know
6139 the device is just moving and can keep their slaves up.
6141 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6142 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6143 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6146 * Flush the unicast and multicast chains
6151 /* Actually switch the network namespace */
6152 dev_net_set(dev, net);
6154 /* If there is an ifindex conflict assign a new one */
6155 if (__dev_get_by_index(net, dev->ifindex)) {
6156 int iflink = (dev->iflink == dev->ifindex);
6157 dev->ifindex = dev_new_index(net);
6159 dev->iflink = dev->ifindex;
6162 /* Fixup kobjects */
6163 err = device_rename(&dev->dev, dev->name);
6166 /* Add the device back in the hashes */
6167 list_netdevice(dev);
6169 /* Notify protocols, that a new device appeared. */
6170 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6173 * Prevent userspace races by waiting until the network
6174 * device is fully setup before sending notifications.
6176 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6183 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6185 static int dev_cpu_callback(struct notifier_block *nfb,
6186 unsigned long action,
6189 struct sk_buff **list_skb;
6190 struct sk_buff *skb;
6191 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6192 struct softnet_data *sd, *oldsd;
6194 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6197 local_irq_disable();
6198 cpu = smp_processor_id();
6199 sd = &per_cpu(softnet_data, cpu);
6200 oldsd = &per_cpu(softnet_data, oldcpu);
6202 /* Find end of our completion_queue. */
6203 list_skb = &sd->completion_queue;
6205 list_skb = &(*list_skb)->next;
6206 /* Append completion queue from offline CPU. */
6207 *list_skb = oldsd->completion_queue;
6208 oldsd->completion_queue = NULL;
6210 /* Append output queue from offline CPU. */
6211 if (oldsd->output_queue) {
6212 *sd->output_queue_tailp = oldsd->output_queue;
6213 sd->output_queue_tailp = oldsd->output_queue_tailp;
6214 oldsd->output_queue = NULL;
6215 oldsd->output_queue_tailp = &oldsd->output_queue;
6217 /* Append NAPI poll list from offline CPU. */
6218 if (!list_empty(&oldsd->poll_list)) {
6219 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6220 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6223 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6226 /* Process offline CPU's input_pkt_queue */
6227 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6229 input_queue_head_incr(oldsd);
6231 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6233 input_queue_head_incr(oldsd);
6241 * netdev_increment_features - increment feature set by one
6242 * @all: current feature set
6243 * @one: new feature set
6244 * @mask: mask feature set
6246 * Computes a new feature set after adding a device with feature set
6247 * @one to the master device with current feature set @all. Will not
6248 * enable anything that is off in @mask. Returns the new feature set.
6250 netdev_features_t netdev_increment_features(netdev_features_t all,
6251 netdev_features_t one, netdev_features_t mask)
6253 if (mask & NETIF_F_GEN_CSUM)
6254 mask |= NETIF_F_ALL_CSUM;
6255 mask |= NETIF_F_VLAN_CHALLENGED;
6257 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6258 all &= one | ~NETIF_F_ALL_FOR_ALL;
6260 /* If one device supports hw checksumming, set for all. */
6261 if (all & NETIF_F_GEN_CSUM)
6262 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6266 EXPORT_SYMBOL(netdev_increment_features);
6268 static struct hlist_head *netdev_create_hash(void)
6271 struct hlist_head *hash;
6273 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6275 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6276 INIT_HLIST_HEAD(&hash[i]);
6281 /* Initialize per network namespace state */
6282 static int __net_init netdev_init(struct net *net)
6284 INIT_LIST_HEAD(&net->dev_base_head);
6286 net->dev_name_head = netdev_create_hash();
6287 if (net->dev_name_head == NULL)
6290 net->dev_index_head = netdev_create_hash();
6291 if (net->dev_index_head == NULL)
6297 kfree(net->dev_name_head);
6303 * netdev_drivername - network driver for the device
6304 * @dev: network device
6306 * Determine network driver for device.
6308 const char *netdev_drivername(const struct net_device *dev)
6310 const struct device_driver *driver;
6311 const struct device *parent;
6312 const char *empty = "";
6314 parent = dev->dev.parent;
6318 driver = parent->driver;
6319 if (driver && driver->name)
6320 return driver->name;
6324 int __netdev_printk(const char *level, const struct net_device *dev,
6325 struct va_format *vaf)
6329 if (dev && dev->dev.parent)
6330 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6331 netdev_name(dev), vaf);
6333 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6335 r = printk("%s(NULL net_device): %pV", level, vaf);
6339 EXPORT_SYMBOL(__netdev_printk);
6341 int netdev_printk(const char *level, const struct net_device *dev,
6342 const char *format, ...)
6344 struct va_format vaf;
6348 va_start(args, format);
6353 r = __netdev_printk(level, dev, &vaf);
6358 EXPORT_SYMBOL(netdev_printk);
6360 #define define_netdev_printk_level(func, level) \
6361 int func(const struct net_device *dev, const char *fmt, ...) \
6364 struct va_format vaf; \
6367 va_start(args, fmt); \
6372 r = __netdev_printk(level, dev, &vaf); \
6377 EXPORT_SYMBOL(func);
6379 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6380 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6381 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6382 define_netdev_printk_level(netdev_err, KERN_ERR);
6383 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6384 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6385 define_netdev_printk_level(netdev_info, KERN_INFO);
6387 static void __net_exit netdev_exit(struct net *net)
6389 kfree(net->dev_name_head);
6390 kfree(net->dev_index_head);
6393 static struct pernet_operations __net_initdata netdev_net_ops = {
6394 .init = netdev_init,
6395 .exit = netdev_exit,
6398 static void __net_exit default_device_exit(struct net *net)
6400 struct net_device *dev, *aux;
6402 * Push all migratable network devices back to the
6403 * initial network namespace
6406 for_each_netdev_safe(net, dev, aux) {
6408 char fb_name[IFNAMSIZ];
6410 /* Ignore unmoveable devices (i.e. loopback) */
6411 if (dev->features & NETIF_F_NETNS_LOCAL)
6414 /* Leave virtual devices for the generic cleanup */
6415 if (dev->rtnl_link_ops)
6418 /* Push remaining network devices to init_net */
6419 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6420 err = dev_change_net_namespace(dev, &init_net, fb_name);
6422 pr_emerg("%s: failed to move %s to init_net: %d\n",
6423 __func__, dev->name, err);
6430 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6432 /* At exit all network devices most be removed from a network
6433 * namespace. Do this in the reverse order of registration.
6434 * Do this across as many network namespaces as possible to
6435 * improve batching efficiency.
6437 struct net_device *dev;
6439 LIST_HEAD(dev_kill_list);
6442 list_for_each_entry(net, net_list, exit_list) {
6443 for_each_netdev_reverse(net, dev) {
6444 if (dev->rtnl_link_ops)
6445 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6447 unregister_netdevice_queue(dev, &dev_kill_list);
6450 unregister_netdevice_many(&dev_kill_list);
6451 list_del(&dev_kill_list);
6455 static struct pernet_operations __net_initdata default_device_ops = {
6456 .exit = default_device_exit,
6457 .exit_batch = default_device_exit_batch,
6461 * Initialize the DEV module. At boot time this walks the device list and
6462 * unhooks any devices that fail to initialise (normally hardware not
6463 * present) and leaves us with a valid list of present and active devices.
6468 * This is called single threaded during boot, so no need
6469 * to take the rtnl semaphore.
6471 static int __init net_dev_init(void)
6473 int i, rc = -ENOMEM;
6475 BUG_ON(!dev_boot_phase);
6477 if (dev_proc_init())
6480 if (netdev_kobject_init())
6483 INIT_LIST_HEAD(&ptype_all);
6484 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6485 INIT_LIST_HEAD(&ptype_base[i]);
6487 if (register_pernet_subsys(&netdev_net_ops))
6491 * Initialise the packet receive queues.
6494 for_each_possible_cpu(i) {
6495 struct softnet_data *sd = &per_cpu(softnet_data, i);
6497 memset(sd, 0, sizeof(*sd));
6498 skb_queue_head_init(&sd->input_pkt_queue);
6499 skb_queue_head_init(&sd->process_queue);
6500 sd->completion_queue = NULL;
6501 INIT_LIST_HEAD(&sd->poll_list);
6502 sd->output_queue = NULL;
6503 sd->output_queue_tailp = &sd->output_queue;
6505 sd->csd.func = rps_trigger_softirq;
6511 sd->backlog.poll = process_backlog;
6512 sd->backlog.weight = weight_p;
6513 sd->backlog.gro_list = NULL;
6514 sd->backlog.gro_count = 0;
6519 /* The loopback device is special if any other network devices
6520 * is present in a network namespace the loopback device must
6521 * be present. Since we now dynamically allocate and free the
6522 * loopback device ensure this invariant is maintained by
6523 * keeping the loopback device as the first device on the
6524 * list of network devices. Ensuring the loopback devices
6525 * is the first device that appears and the last network device
6528 if (register_pernet_device(&loopback_net_ops))
6531 if (register_pernet_device(&default_device_ops))
6534 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6535 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6537 hotcpu_notifier(dev_cpu_callback, 0);
6545 subsys_initcall(net_dev_init);
6547 static int __init initialize_hashrnd(void)
6549 get_random_bytes(&hashrnd, sizeof(hashrnd));
6553 late_initcall_sync(initialize_hashrnd);