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.
1414 int unregister_netdevice_notifier(struct notifier_block *nb)
1419 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1423 EXPORT_SYMBOL(unregister_netdevice_notifier);
1426 * call_netdevice_notifiers - call all network notifier blocks
1427 * @val: value passed unmodified to notifier function
1428 * @dev: net_device pointer passed unmodified to notifier function
1430 * Call all network notifier blocks. Parameters and return value
1431 * are as for raw_notifier_call_chain().
1434 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1437 return raw_notifier_call_chain(&netdev_chain, val, dev);
1439 EXPORT_SYMBOL(call_netdevice_notifiers);
1441 static struct static_key netstamp_needed __read_mostly;
1442 #ifdef HAVE_JUMP_LABEL
1443 /* We are not allowed to call static_key_slow_dec() from irq context
1444 * If net_disable_timestamp() is called from irq context, defer the
1445 * static_key_slow_dec() calls.
1447 static atomic_t netstamp_needed_deferred;
1450 void net_enable_timestamp(void)
1452 #ifdef HAVE_JUMP_LABEL
1453 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1457 static_key_slow_dec(&netstamp_needed);
1461 WARN_ON(in_interrupt());
1462 static_key_slow_inc(&netstamp_needed);
1464 EXPORT_SYMBOL(net_enable_timestamp);
1466 void net_disable_timestamp(void)
1468 #ifdef HAVE_JUMP_LABEL
1469 if (in_interrupt()) {
1470 atomic_inc(&netstamp_needed_deferred);
1474 static_key_slow_dec(&netstamp_needed);
1476 EXPORT_SYMBOL(net_disable_timestamp);
1478 static inline void net_timestamp_set(struct sk_buff *skb)
1480 skb->tstamp.tv64 = 0;
1481 if (static_key_false(&netstamp_needed))
1482 __net_timestamp(skb);
1485 #define net_timestamp_check(COND, SKB) \
1486 if (static_key_false(&netstamp_needed)) { \
1487 if ((COND) && !(SKB)->tstamp.tv64) \
1488 __net_timestamp(SKB); \
1491 static int net_hwtstamp_validate(struct ifreq *ifr)
1493 struct hwtstamp_config cfg;
1494 enum hwtstamp_tx_types tx_type;
1495 enum hwtstamp_rx_filters rx_filter;
1496 int tx_type_valid = 0;
1497 int rx_filter_valid = 0;
1499 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1502 if (cfg.flags) /* reserved for future extensions */
1505 tx_type = cfg.tx_type;
1506 rx_filter = cfg.rx_filter;
1509 case HWTSTAMP_TX_OFF:
1510 case HWTSTAMP_TX_ON:
1511 case HWTSTAMP_TX_ONESTEP_SYNC:
1516 switch (rx_filter) {
1517 case HWTSTAMP_FILTER_NONE:
1518 case HWTSTAMP_FILTER_ALL:
1519 case HWTSTAMP_FILTER_SOME:
1520 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1521 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1522 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1523 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1524 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1525 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1526 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1527 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1528 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1529 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1530 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1531 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1532 rx_filter_valid = 1;
1536 if (!tx_type_valid || !rx_filter_valid)
1542 static inline bool is_skb_forwardable(struct net_device *dev,
1543 struct sk_buff *skb)
1547 if (!(dev->flags & IFF_UP))
1550 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1551 if (skb->len <= len)
1554 /* if TSO is enabled, we don't care about the length as the packet
1555 * could be forwarded without being segmented before
1557 if (skb_is_gso(skb))
1564 * dev_forward_skb - loopback an skb to another netif
1566 * @dev: destination network device
1567 * @skb: buffer to forward
1570 * NET_RX_SUCCESS (no congestion)
1571 * NET_RX_DROP (packet was dropped, but freed)
1573 * dev_forward_skb can be used for injecting an skb from the
1574 * start_xmit function of one device into the receive queue
1575 * of another device.
1577 * The receiving device may be in another namespace, so
1578 * we have to clear all information in the skb that could
1579 * impact namespace isolation.
1581 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1583 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1584 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1585 atomic_long_inc(&dev->rx_dropped);
1594 if (unlikely(!is_skb_forwardable(dev, skb))) {
1595 atomic_long_inc(&dev->rx_dropped);
1599 skb_set_dev(skb, dev);
1600 skb->tstamp.tv64 = 0;
1601 skb->pkt_type = PACKET_HOST;
1602 skb->protocol = eth_type_trans(skb, dev);
1603 return netif_rx(skb);
1605 EXPORT_SYMBOL_GPL(dev_forward_skb);
1607 static inline int deliver_skb(struct sk_buff *skb,
1608 struct packet_type *pt_prev,
1609 struct net_device *orig_dev)
1611 atomic_inc(&skb->users);
1612 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1616 * Support routine. Sends outgoing frames to any network
1617 * taps currently in use.
1620 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1622 struct packet_type *ptype;
1623 struct sk_buff *skb2 = NULL;
1624 struct packet_type *pt_prev = NULL;
1627 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1628 /* Never send packets back to the socket
1629 * they originated from - MvS (miquels@drinkel.ow.org)
1631 if ((ptype->dev == dev || !ptype->dev) &&
1632 (ptype->af_packet_priv == NULL ||
1633 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1635 deliver_skb(skb2, pt_prev, skb->dev);
1640 skb2 = skb_clone(skb, GFP_ATOMIC);
1644 net_timestamp_set(skb2);
1646 /* skb->nh should be correctly
1647 set by sender, so that the second statement is
1648 just protection against buggy protocols.
1650 skb_reset_mac_header(skb2);
1652 if (skb_network_header(skb2) < skb2->data ||
1653 skb2->network_header > skb2->tail) {
1654 if (net_ratelimit())
1655 pr_crit("protocol %04x is buggy, dev %s\n",
1656 ntohs(skb2->protocol),
1658 skb_reset_network_header(skb2);
1661 skb2->transport_header = skb2->network_header;
1662 skb2->pkt_type = PACKET_OUTGOING;
1667 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1671 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1672 * @dev: Network device
1673 * @txq: number of queues available
1675 * If real_num_tx_queues is changed the tc mappings may no longer be
1676 * valid. To resolve this verify the tc mapping remains valid and if
1677 * not NULL the mapping. With no priorities mapping to this
1678 * offset/count pair it will no longer be used. In the worst case TC0
1679 * is invalid nothing can be done so disable priority mappings. If is
1680 * expected that drivers will fix this mapping if they can before
1681 * calling netif_set_real_num_tx_queues.
1683 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1686 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1688 /* If TC0 is invalidated disable TC mapping */
1689 if (tc->offset + tc->count > txq) {
1690 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1695 /* Invalidated prio to tc mappings set to TC0 */
1696 for (i = 1; i < TC_BITMASK + 1; i++) {
1697 int q = netdev_get_prio_tc_map(dev, i);
1699 tc = &dev->tc_to_txq[q];
1700 if (tc->offset + tc->count > txq) {
1701 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1703 netdev_set_prio_tc_map(dev, i, 0);
1709 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1710 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1712 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1716 if (txq < 1 || txq > dev->num_tx_queues)
1719 if (dev->reg_state == NETREG_REGISTERED ||
1720 dev->reg_state == NETREG_UNREGISTERING) {
1723 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1729 netif_setup_tc(dev, txq);
1731 if (txq < dev->real_num_tx_queues)
1732 qdisc_reset_all_tx_gt(dev, txq);
1735 dev->real_num_tx_queues = txq;
1738 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1742 * netif_set_real_num_rx_queues - set actual number of RX queues used
1743 * @dev: Network device
1744 * @rxq: Actual number of RX queues
1746 * This must be called either with the rtnl_lock held or before
1747 * registration of the net device. Returns 0 on success, or a
1748 * negative error code. If called before registration, it always
1751 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1755 if (rxq < 1 || rxq > dev->num_rx_queues)
1758 if (dev->reg_state == NETREG_REGISTERED) {
1761 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1767 dev->real_num_rx_queues = rxq;
1770 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1773 static inline void __netif_reschedule(struct Qdisc *q)
1775 struct softnet_data *sd;
1776 unsigned long flags;
1778 local_irq_save(flags);
1779 sd = &__get_cpu_var(softnet_data);
1780 q->next_sched = NULL;
1781 *sd->output_queue_tailp = q;
1782 sd->output_queue_tailp = &q->next_sched;
1783 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1784 local_irq_restore(flags);
1787 void __netif_schedule(struct Qdisc *q)
1789 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1790 __netif_reschedule(q);
1792 EXPORT_SYMBOL(__netif_schedule);
1794 void dev_kfree_skb_irq(struct sk_buff *skb)
1796 if (atomic_dec_and_test(&skb->users)) {
1797 struct softnet_data *sd;
1798 unsigned long flags;
1800 local_irq_save(flags);
1801 sd = &__get_cpu_var(softnet_data);
1802 skb->next = sd->completion_queue;
1803 sd->completion_queue = skb;
1804 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1805 local_irq_restore(flags);
1808 EXPORT_SYMBOL(dev_kfree_skb_irq);
1810 void dev_kfree_skb_any(struct sk_buff *skb)
1812 if (in_irq() || irqs_disabled())
1813 dev_kfree_skb_irq(skb);
1817 EXPORT_SYMBOL(dev_kfree_skb_any);
1821 * netif_device_detach - mark device as removed
1822 * @dev: network device
1824 * Mark device as removed from system and therefore no longer available.
1826 void netif_device_detach(struct net_device *dev)
1828 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1829 netif_running(dev)) {
1830 netif_tx_stop_all_queues(dev);
1833 EXPORT_SYMBOL(netif_device_detach);
1836 * netif_device_attach - mark device as attached
1837 * @dev: network device
1839 * Mark device as attached from system and restart if needed.
1841 void netif_device_attach(struct net_device *dev)
1843 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1844 netif_running(dev)) {
1845 netif_tx_wake_all_queues(dev);
1846 __netdev_watchdog_up(dev);
1849 EXPORT_SYMBOL(netif_device_attach);
1852 * skb_dev_set -- assign a new device to a buffer
1853 * @skb: buffer for the new device
1854 * @dev: network device
1856 * If an skb is owned by a device already, we have to reset
1857 * all data private to the namespace a device belongs to
1858 * before assigning it a new device.
1860 #ifdef CONFIG_NET_NS
1861 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1864 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1867 skb_init_secmark(skb);
1871 skb->ipvs_property = 0;
1872 #ifdef CONFIG_NET_SCHED
1878 EXPORT_SYMBOL(skb_set_dev);
1879 #endif /* CONFIG_NET_NS */
1881 static void skb_warn_bad_offload(const struct sk_buff *skb)
1883 static const netdev_features_t null_features = 0;
1884 struct net_device *dev = skb->dev;
1885 const char *driver = "";
1887 if (dev && dev->dev.parent)
1888 driver = dev_driver_string(dev->dev.parent);
1890 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1891 "gso_type=%d ip_summed=%d\n",
1892 driver, dev ? &dev->features : &null_features,
1893 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1894 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1895 skb_shinfo(skb)->gso_type, skb->ip_summed);
1899 * Invalidate hardware checksum when packet is to be mangled, and
1900 * complete checksum manually on outgoing path.
1902 int skb_checksum_help(struct sk_buff *skb)
1905 int ret = 0, offset;
1907 if (skb->ip_summed == CHECKSUM_COMPLETE)
1908 goto out_set_summed;
1910 if (unlikely(skb_shinfo(skb)->gso_size)) {
1911 skb_warn_bad_offload(skb);
1915 offset = skb_checksum_start_offset(skb);
1916 BUG_ON(offset >= skb_headlen(skb));
1917 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1919 offset += skb->csum_offset;
1920 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1922 if (skb_cloned(skb) &&
1923 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1924 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1929 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1931 skb->ip_summed = CHECKSUM_NONE;
1935 EXPORT_SYMBOL(skb_checksum_help);
1938 * skb_gso_segment - Perform segmentation on skb.
1939 * @skb: buffer to segment
1940 * @features: features for the output path (see dev->features)
1942 * This function segments the given skb and returns a list of segments.
1944 * It may return NULL if the skb requires no segmentation. This is
1945 * only possible when GSO is used for verifying header integrity.
1947 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1948 netdev_features_t features)
1950 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1951 struct packet_type *ptype;
1952 __be16 type = skb->protocol;
1953 int vlan_depth = ETH_HLEN;
1956 while (type == htons(ETH_P_8021Q)) {
1957 struct vlan_hdr *vh;
1959 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1960 return ERR_PTR(-EINVAL);
1962 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1963 type = vh->h_vlan_encapsulated_proto;
1964 vlan_depth += VLAN_HLEN;
1967 skb_reset_mac_header(skb);
1968 skb->mac_len = skb->network_header - skb->mac_header;
1969 __skb_pull(skb, skb->mac_len);
1971 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1972 skb_warn_bad_offload(skb);
1974 if (skb_header_cloned(skb) &&
1975 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1976 return ERR_PTR(err);
1980 list_for_each_entry_rcu(ptype,
1981 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1982 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1983 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1984 err = ptype->gso_send_check(skb);
1985 segs = ERR_PTR(err);
1986 if (err || skb_gso_ok(skb, features))
1988 __skb_push(skb, (skb->data -
1989 skb_network_header(skb)));
1991 segs = ptype->gso_segment(skb, features);
1997 __skb_push(skb, skb->data - skb_mac_header(skb));
2001 EXPORT_SYMBOL(skb_gso_segment);
2003 /* Take action when hardware reception checksum errors are detected. */
2005 void netdev_rx_csum_fault(struct net_device *dev)
2007 if (net_ratelimit()) {
2008 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2012 EXPORT_SYMBOL(netdev_rx_csum_fault);
2015 /* Actually, we should eliminate this check as soon as we know, that:
2016 * 1. IOMMU is present and allows to map all the memory.
2017 * 2. No high memory really exists on this machine.
2020 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2022 #ifdef CONFIG_HIGHMEM
2024 if (!(dev->features & NETIF_F_HIGHDMA)) {
2025 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2026 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2027 if (PageHighMem(skb_frag_page(frag)))
2032 if (PCI_DMA_BUS_IS_PHYS) {
2033 struct device *pdev = dev->dev.parent;
2037 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2038 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2039 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2040 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2049 void (*destructor)(struct sk_buff *skb);
2052 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2054 static void dev_gso_skb_destructor(struct sk_buff *skb)
2056 struct dev_gso_cb *cb;
2059 struct sk_buff *nskb = skb->next;
2061 skb->next = nskb->next;
2064 } while (skb->next);
2066 cb = DEV_GSO_CB(skb);
2068 cb->destructor(skb);
2072 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2073 * @skb: buffer to segment
2074 * @features: device features as applicable to this skb
2076 * This function segments the given skb and stores the list of segments
2079 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2081 struct sk_buff *segs;
2083 segs = skb_gso_segment(skb, features);
2085 /* Verifying header integrity only. */
2090 return PTR_ERR(segs);
2093 DEV_GSO_CB(skb)->destructor = skb->destructor;
2094 skb->destructor = dev_gso_skb_destructor;
2100 * Try to orphan skb early, right before transmission by the device.
2101 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2102 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2104 static inline void skb_orphan_try(struct sk_buff *skb)
2106 struct sock *sk = skb->sk;
2108 if (sk && !skb_shinfo(skb)->tx_flags) {
2109 /* skb_tx_hash() wont be able to get sk.
2110 * We copy sk_hash into skb->rxhash
2113 skb->rxhash = sk->sk_hash;
2118 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2120 return ((features & NETIF_F_GEN_CSUM) ||
2121 ((features & NETIF_F_V4_CSUM) &&
2122 protocol == htons(ETH_P_IP)) ||
2123 ((features & NETIF_F_V6_CSUM) &&
2124 protocol == htons(ETH_P_IPV6)) ||
2125 ((features & NETIF_F_FCOE_CRC) &&
2126 protocol == htons(ETH_P_FCOE)));
2129 static netdev_features_t harmonize_features(struct sk_buff *skb,
2130 __be16 protocol, netdev_features_t features)
2132 if (!can_checksum_protocol(features, protocol)) {
2133 features &= ~NETIF_F_ALL_CSUM;
2134 features &= ~NETIF_F_SG;
2135 } else if (illegal_highdma(skb->dev, skb)) {
2136 features &= ~NETIF_F_SG;
2142 netdev_features_t netif_skb_features(struct sk_buff *skb)
2144 __be16 protocol = skb->protocol;
2145 netdev_features_t features = skb->dev->features;
2147 if (protocol == htons(ETH_P_8021Q)) {
2148 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2149 protocol = veh->h_vlan_encapsulated_proto;
2150 } else if (!vlan_tx_tag_present(skb)) {
2151 return harmonize_features(skb, protocol, features);
2154 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2156 if (protocol != htons(ETH_P_8021Q)) {
2157 return harmonize_features(skb, protocol, features);
2159 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2160 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2161 return harmonize_features(skb, protocol, features);
2164 EXPORT_SYMBOL(netif_skb_features);
2167 * Returns true if either:
2168 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2169 * 2. skb is fragmented and the device does not support SG, or if
2170 * at least one of fragments is in highmem and device does not
2171 * support DMA from it.
2173 static inline int skb_needs_linearize(struct sk_buff *skb,
2176 return skb_is_nonlinear(skb) &&
2177 ((skb_has_frag_list(skb) &&
2178 !(features & NETIF_F_FRAGLIST)) ||
2179 (skb_shinfo(skb)->nr_frags &&
2180 !(features & NETIF_F_SG)));
2183 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2184 struct netdev_queue *txq)
2186 const struct net_device_ops *ops = dev->netdev_ops;
2187 int rc = NETDEV_TX_OK;
2188 unsigned int skb_len;
2190 if (likely(!skb->next)) {
2191 netdev_features_t features;
2194 * If device doesn't need skb->dst, release it right now while
2195 * its hot in this cpu cache
2197 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2200 if (!list_empty(&ptype_all))
2201 dev_queue_xmit_nit(skb, dev);
2203 skb_orphan_try(skb);
2205 features = netif_skb_features(skb);
2207 if (vlan_tx_tag_present(skb) &&
2208 !(features & NETIF_F_HW_VLAN_TX)) {
2209 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2216 if (netif_needs_gso(skb, features)) {
2217 if (unlikely(dev_gso_segment(skb, features)))
2222 if (skb_needs_linearize(skb, features) &&
2223 __skb_linearize(skb))
2226 /* If packet is not checksummed and device does not
2227 * support checksumming for this protocol, complete
2228 * checksumming here.
2230 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2231 skb_set_transport_header(skb,
2232 skb_checksum_start_offset(skb));
2233 if (!(features & NETIF_F_ALL_CSUM) &&
2234 skb_checksum_help(skb))
2240 rc = ops->ndo_start_xmit(skb, dev);
2241 trace_net_dev_xmit(skb, rc, dev, skb_len);
2242 if (rc == NETDEV_TX_OK)
2243 txq_trans_update(txq);
2249 struct sk_buff *nskb = skb->next;
2251 skb->next = nskb->next;
2255 * If device doesn't need nskb->dst, release it right now while
2256 * its hot in this cpu cache
2258 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2261 skb_len = nskb->len;
2262 rc = ops->ndo_start_xmit(nskb, dev);
2263 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2264 if (unlikely(rc != NETDEV_TX_OK)) {
2265 if (rc & ~NETDEV_TX_MASK)
2266 goto out_kfree_gso_skb;
2267 nskb->next = skb->next;
2271 txq_trans_update(txq);
2272 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2273 return NETDEV_TX_BUSY;
2274 } while (skb->next);
2277 if (likely(skb->next == NULL))
2278 skb->destructor = DEV_GSO_CB(skb)->destructor;
2285 static u32 hashrnd __read_mostly;
2288 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2289 * to be used as a distribution range.
2291 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2292 unsigned int num_tx_queues)
2296 u16 qcount = num_tx_queues;
2298 if (skb_rx_queue_recorded(skb)) {
2299 hash = skb_get_rx_queue(skb);
2300 while (unlikely(hash >= num_tx_queues))
2301 hash -= num_tx_queues;
2306 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2307 qoffset = dev->tc_to_txq[tc].offset;
2308 qcount = dev->tc_to_txq[tc].count;
2311 if (skb->sk && skb->sk->sk_hash)
2312 hash = skb->sk->sk_hash;
2314 hash = (__force u16) skb->protocol ^ skb->rxhash;
2315 hash = jhash_1word(hash, hashrnd);
2317 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2319 EXPORT_SYMBOL(__skb_tx_hash);
2321 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2323 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2324 if (net_ratelimit()) {
2325 pr_warn("%s selects TX queue %d, but real number of TX queues is %d\n",
2326 dev->name, queue_index,
2327 dev->real_num_tx_queues);
2334 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2337 struct xps_dev_maps *dev_maps;
2338 struct xps_map *map;
2339 int queue_index = -1;
2342 dev_maps = rcu_dereference(dev->xps_maps);
2344 map = rcu_dereference(
2345 dev_maps->cpu_map[raw_smp_processor_id()]);
2348 queue_index = map->queues[0];
2351 if (skb->sk && skb->sk->sk_hash)
2352 hash = skb->sk->sk_hash;
2354 hash = (__force u16) skb->protocol ^
2356 hash = jhash_1word(hash, hashrnd);
2357 queue_index = map->queues[
2358 ((u64)hash * map->len) >> 32];
2360 if (unlikely(queue_index >= dev->real_num_tx_queues))
2372 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2373 struct sk_buff *skb)
2376 const struct net_device_ops *ops = dev->netdev_ops;
2378 if (dev->real_num_tx_queues == 1)
2380 else if (ops->ndo_select_queue) {
2381 queue_index = ops->ndo_select_queue(dev, skb);
2382 queue_index = dev_cap_txqueue(dev, queue_index);
2384 struct sock *sk = skb->sk;
2385 queue_index = sk_tx_queue_get(sk);
2387 if (queue_index < 0 || skb->ooo_okay ||
2388 queue_index >= dev->real_num_tx_queues) {
2389 int old_index = queue_index;
2391 queue_index = get_xps_queue(dev, skb);
2392 if (queue_index < 0)
2393 queue_index = skb_tx_hash(dev, skb);
2395 if (queue_index != old_index && sk) {
2396 struct dst_entry *dst =
2397 rcu_dereference_check(sk->sk_dst_cache, 1);
2399 if (dst && skb_dst(skb) == dst)
2400 sk_tx_queue_set(sk, queue_index);
2405 skb_set_queue_mapping(skb, queue_index);
2406 return netdev_get_tx_queue(dev, queue_index);
2409 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2410 struct net_device *dev,
2411 struct netdev_queue *txq)
2413 spinlock_t *root_lock = qdisc_lock(q);
2417 qdisc_skb_cb(skb)->pkt_len = skb->len;
2418 qdisc_calculate_pkt_len(skb, q);
2420 * Heuristic to force contended enqueues to serialize on a
2421 * separate lock before trying to get qdisc main lock.
2422 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2423 * and dequeue packets faster.
2425 contended = qdisc_is_running(q);
2426 if (unlikely(contended))
2427 spin_lock(&q->busylock);
2429 spin_lock(root_lock);
2430 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2433 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2434 qdisc_run_begin(q)) {
2436 * This is a work-conserving queue; there are no old skbs
2437 * waiting to be sent out; and the qdisc is not running -
2438 * xmit the skb directly.
2440 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2443 qdisc_bstats_update(q, skb);
2445 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2446 if (unlikely(contended)) {
2447 spin_unlock(&q->busylock);
2454 rc = NET_XMIT_SUCCESS;
2457 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2458 if (qdisc_run_begin(q)) {
2459 if (unlikely(contended)) {
2460 spin_unlock(&q->busylock);
2466 spin_unlock(root_lock);
2467 if (unlikely(contended))
2468 spin_unlock(&q->busylock);
2472 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2473 static void skb_update_prio(struct sk_buff *skb)
2475 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2477 if ((!skb->priority) && (skb->sk) && map)
2478 skb->priority = map->priomap[skb->sk->sk_cgrp_prioidx];
2481 #define skb_update_prio(skb)
2484 static DEFINE_PER_CPU(int, xmit_recursion);
2485 #define RECURSION_LIMIT 10
2488 * dev_queue_xmit - transmit a buffer
2489 * @skb: buffer to transmit
2491 * Queue a buffer for transmission to a network device. The caller must
2492 * have set the device and priority and built the buffer before calling
2493 * this function. The function can be called from an interrupt.
2495 * A negative errno code is returned on a failure. A success does not
2496 * guarantee the frame will be transmitted as it may be dropped due
2497 * to congestion or traffic shaping.
2499 * -----------------------------------------------------------------------------------
2500 * I notice this method can also return errors from the queue disciplines,
2501 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2504 * Regardless of the return value, the skb is consumed, so it is currently
2505 * difficult to retry a send to this method. (You can bump the ref count
2506 * before sending to hold a reference for retry if you are careful.)
2508 * When calling this method, interrupts MUST be enabled. This is because
2509 * the BH enable code must have IRQs enabled so that it will not deadlock.
2512 int dev_queue_xmit(struct sk_buff *skb)
2514 struct net_device *dev = skb->dev;
2515 struct netdev_queue *txq;
2519 /* Disable soft irqs for various locks below. Also
2520 * stops preemption for RCU.
2524 skb_update_prio(skb);
2526 txq = dev_pick_tx(dev, skb);
2527 q = rcu_dereference_bh(txq->qdisc);
2529 #ifdef CONFIG_NET_CLS_ACT
2530 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2532 trace_net_dev_queue(skb);
2534 rc = __dev_xmit_skb(skb, q, dev, txq);
2538 /* The device has no queue. Common case for software devices:
2539 loopback, all the sorts of tunnels...
2541 Really, it is unlikely that netif_tx_lock protection is necessary
2542 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2544 However, it is possible, that they rely on protection
2547 Check this and shot the lock. It is not prone from deadlocks.
2548 Either shot noqueue qdisc, it is even simpler 8)
2550 if (dev->flags & IFF_UP) {
2551 int cpu = smp_processor_id(); /* ok because BHs are off */
2553 if (txq->xmit_lock_owner != cpu) {
2555 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2556 goto recursion_alert;
2558 HARD_TX_LOCK(dev, txq, cpu);
2560 if (!netif_xmit_stopped(txq)) {
2561 __this_cpu_inc(xmit_recursion);
2562 rc = dev_hard_start_xmit(skb, dev, txq);
2563 __this_cpu_dec(xmit_recursion);
2564 if (dev_xmit_complete(rc)) {
2565 HARD_TX_UNLOCK(dev, txq);
2569 HARD_TX_UNLOCK(dev, txq);
2570 if (net_ratelimit())
2571 pr_crit("Virtual device %s asks to queue packet!\n",
2574 /* Recursion is detected! It is possible,
2578 if (net_ratelimit())
2579 pr_crit("Dead loop on virtual device %s, fix it urgently!\n",
2585 rcu_read_unlock_bh();
2590 rcu_read_unlock_bh();
2593 EXPORT_SYMBOL(dev_queue_xmit);
2596 /*=======================================================================
2598 =======================================================================*/
2600 int netdev_max_backlog __read_mostly = 1000;
2601 int netdev_tstamp_prequeue __read_mostly = 1;
2602 int netdev_budget __read_mostly = 300;
2603 int weight_p __read_mostly = 64; /* old backlog weight */
2605 /* Called with irq disabled */
2606 static inline void ____napi_schedule(struct softnet_data *sd,
2607 struct napi_struct *napi)
2609 list_add_tail(&napi->poll_list, &sd->poll_list);
2610 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2614 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2615 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2616 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2617 * if hash is a canonical 4-tuple hash over transport ports.
2619 void __skb_get_rxhash(struct sk_buff *skb)
2621 struct flow_keys keys;
2624 if (!skb_flow_dissect(skb, &keys))
2628 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2629 swap(keys.port16[0], keys.port16[1]);
2633 /* get a consistent hash (same value on both flow directions) */
2634 if ((__force u32)keys.dst < (__force u32)keys.src)
2635 swap(keys.dst, keys.src);
2637 hash = jhash_3words((__force u32)keys.dst,
2638 (__force u32)keys.src,
2639 (__force u32)keys.ports, hashrnd);
2645 EXPORT_SYMBOL(__skb_get_rxhash);
2649 /* One global table that all flow-based protocols share. */
2650 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2651 EXPORT_SYMBOL(rps_sock_flow_table);
2653 struct static_key rps_needed __read_mostly;
2655 static struct rps_dev_flow *
2656 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2657 struct rps_dev_flow *rflow, u16 next_cpu)
2659 if (next_cpu != RPS_NO_CPU) {
2660 #ifdef CONFIG_RFS_ACCEL
2661 struct netdev_rx_queue *rxqueue;
2662 struct rps_dev_flow_table *flow_table;
2663 struct rps_dev_flow *old_rflow;
2668 /* Should we steer this flow to a different hardware queue? */
2669 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2670 !(dev->features & NETIF_F_NTUPLE))
2672 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2673 if (rxq_index == skb_get_rx_queue(skb))
2676 rxqueue = dev->_rx + rxq_index;
2677 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2680 flow_id = skb->rxhash & flow_table->mask;
2681 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2682 rxq_index, flow_id);
2686 rflow = &flow_table->flows[flow_id];
2688 if (old_rflow->filter == rflow->filter)
2689 old_rflow->filter = RPS_NO_FILTER;
2693 per_cpu(softnet_data, next_cpu).input_queue_head;
2696 rflow->cpu = next_cpu;
2701 * get_rps_cpu is called from netif_receive_skb and returns the target
2702 * CPU from the RPS map of the receiving queue for a given skb.
2703 * rcu_read_lock must be held on entry.
2705 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2706 struct rps_dev_flow **rflowp)
2708 struct netdev_rx_queue *rxqueue;
2709 struct rps_map *map;
2710 struct rps_dev_flow_table *flow_table;
2711 struct rps_sock_flow_table *sock_flow_table;
2715 if (skb_rx_queue_recorded(skb)) {
2716 u16 index = skb_get_rx_queue(skb);
2717 if (unlikely(index >= dev->real_num_rx_queues)) {
2718 WARN_ONCE(dev->real_num_rx_queues > 1,
2719 "%s received packet on queue %u, but number "
2720 "of RX queues is %u\n",
2721 dev->name, index, dev->real_num_rx_queues);
2724 rxqueue = dev->_rx + index;
2728 map = rcu_dereference(rxqueue->rps_map);
2730 if (map->len == 1 &&
2731 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2732 tcpu = map->cpus[0];
2733 if (cpu_online(tcpu))
2737 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2741 skb_reset_network_header(skb);
2742 if (!skb_get_rxhash(skb))
2745 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2746 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2747 if (flow_table && sock_flow_table) {
2749 struct rps_dev_flow *rflow;
2751 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2754 next_cpu = sock_flow_table->ents[skb->rxhash &
2755 sock_flow_table->mask];
2758 * If the desired CPU (where last recvmsg was done) is
2759 * different from current CPU (one in the rx-queue flow
2760 * table entry), switch if one of the following holds:
2761 * - Current CPU is unset (equal to RPS_NO_CPU).
2762 * - Current CPU is offline.
2763 * - The current CPU's queue tail has advanced beyond the
2764 * last packet that was enqueued using this table entry.
2765 * This guarantees that all previous packets for the flow
2766 * have been dequeued, thus preserving in order delivery.
2768 if (unlikely(tcpu != next_cpu) &&
2769 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2770 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2771 rflow->last_qtail)) >= 0))
2772 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2774 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2782 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2784 if (cpu_online(tcpu)) {
2794 #ifdef CONFIG_RFS_ACCEL
2797 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2798 * @dev: Device on which the filter was set
2799 * @rxq_index: RX queue index
2800 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2801 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2803 * Drivers that implement ndo_rx_flow_steer() should periodically call
2804 * this function for each installed filter and remove the filters for
2805 * which it returns %true.
2807 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2808 u32 flow_id, u16 filter_id)
2810 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2811 struct rps_dev_flow_table *flow_table;
2812 struct rps_dev_flow *rflow;
2817 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2818 if (flow_table && flow_id <= flow_table->mask) {
2819 rflow = &flow_table->flows[flow_id];
2820 cpu = ACCESS_ONCE(rflow->cpu);
2821 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2822 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2823 rflow->last_qtail) <
2824 (int)(10 * flow_table->mask)))
2830 EXPORT_SYMBOL(rps_may_expire_flow);
2832 #endif /* CONFIG_RFS_ACCEL */
2834 /* Called from hardirq (IPI) context */
2835 static void rps_trigger_softirq(void *data)
2837 struct softnet_data *sd = data;
2839 ____napi_schedule(sd, &sd->backlog);
2843 #endif /* CONFIG_RPS */
2846 * Check if this softnet_data structure is another cpu one
2847 * If yes, queue it to our IPI list and return 1
2850 static int rps_ipi_queued(struct softnet_data *sd)
2853 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2856 sd->rps_ipi_next = mysd->rps_ipi_list;
2857 mysd->rps_ipi_list = sd;
2859 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2862 #endif /* CONFIG_RPS */
2867 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2868 * queue (may be a remote CPU queue).
2870 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2871 unsigned int *qtail)
2873 struct softnet_data *sd;
2874 unsigned long flags;
2876 sd = &per_cpu(softnet_data, cpu);
2878 local_irq_save(flags);
2881 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2882 if (skb_queue_len(&sd->input_pkt_queue)) {
2884 __skb_queue_tail(&sd->input_pkt_queue, skb);
2885 input_queue_tail_incr_save(sd, qtail);
2887 local_irq_restore(flags);
2888 return NET_RX_SUCCESS;
2891 /* Schedule NAPI for backlog device
2892 * We can use non atomic operation since we own the queue lock
2894 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2895 if (!rps_ipi_queued(sd))
2896 ____napi_schedule(sd, &sd->backlog);
2904 local_irq_restore(flags);
2906 atomic_long_inc(&skb->dev->rx_dropped);
2912 * netif_rx - post buffer to the network code
2913 * @skb: buffer to post
2915 * This function receives a packet from a device driver and queues it for
2916 * the upper (protocol) levels to process. It always succeeds. The buffer
2917 * may be dropped during processing for congestion control or by the
2921 * NET_RX_SUCCESS (no congestion)
2922 * NET_RX_DROP (packet was dropped)
2926 int netif_rx(struct sk_buff *skb)
2930 /* if netpoll wants it, pretend we never saw it */
2931 if (netpoll_rx(skb))
2934 net_timestamp_check(netdev_tstamp_prequeue, skb);
2936 trace_netif_rx(skb);
2938 if (static_key_false(&rps_needed)) {
2939 struct rps_dev_flow voidflow, *rflow = &voidflow;
2945 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2947 cpu = smp_processor_id();
2949 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2957 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2962 EXPORT_SYMBOL(netif_rx);
2964 int netif_rx_ni(struct sk_buff *skb)
2969 err = netif_rx(skb);
2970 if (local_softirq_pending())
2976 EXPORT_SYMBOL(netif_rx_ni);
2978 static void net_tx_action(struct softirq_action *h)
2980 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2982 if (sd->completion_queue) {
2983 struct sk_buff *clist;
2985 local_irq_disable();
2986 clist = sd->completion_queue;
2987 sd->completion_queue = NULL;
2991 struct sk_buff *skb = clist;
2992 clist = clist->next;
2994 WARN_ON(atomic_read(&skb->users));
2995 trace_kfree_skb(skb, net_tx_action);
3000 if (sd->output_queue) {
3003 local_irq_disable();
3004 head = sd->output_queue;
3005 sd->output_queue = NULL;
3006 sd->output_queue_tailp = &sd->output_queue;
3010 struct Qdisc *q = head;
3011 spinlock_t *root_lock;
3013 head = head->next_sched;
3015 root_lock = qdisc_lock(q);
3016 if (spin_trylock(root_lock)) {
3017 smp_mb__before_clear_bit();
3018 clear_bit(__QDISC_STATE_SCHED,
3021 spin_unlock(root_lock);
3023 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3025 __netif_reschedule(q);
3027 smp_mb__before_clear_bit();
3028 clear_bit(__QDISC_STATE_SCHED,
3036 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3037 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3038 /* This hook is defined here for ATM LANE */
3039 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3040 unsigned char *addr) __read_mostly;
3041 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3044 #ifdef CONFIG_NET_CLS_ACT
3045 /* TODO: Maybe we should just force sch_ingress to be compiled in
3046 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3047 * a compare and 2 stores extra right now if we dont have it on
3048 * but have CONFIG_NET_CLS_ACT
3049 * NOTE: This doesn't stop any functionality; if you dont have
3050 * the ingress scheduler, you just can't add policies on ingress.
3053 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3055 struct net_device *dev = skb->dev;
3056 u32 ttl = G_TC_RTTL(skb->tc_verd);
3057 int result = TC_ACT_OK;
3060 if (unlikely(MAX_RED_LOOP < ttl++)) {
3061 if (net_ratelimit())
3062 pr_warn("Redir loop detected Dropping packet (%d->%d)\n",
3063 skb->skb_iif, dev->ifindex);
3067 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3068 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3071 if (q != &noop_qdisc) {
3072 spin_lock(qdisc_lock(q));
3073 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3074 result = qdisc_enqueue_root(skb, q);
3075 spin_unlock(qdisc_lock(q));
3081 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3082 struct packet_type **pt_prev,
3083 int *ret, struct net_device *orig_dev)
3085 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3087 if (!rxq || rxq->qdisc == &noop_qdisc)
3091 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3095 switch (ing_filter(skb, rxq)) {
3109 * netdev_rx_handler_register - register receive handler
3110 * @dev: device to register a handler for
3111 * @rx_handler: receive handler to register
3112 * @rx_handler_data: data pointer that is used by rx handler
3114 * Register a receive hander for a device. This handler will then be
3115 * called from __netif_receive_skb. A negative errno code is returned
3118 * The caller must hold the rtnl_mutex.
3120 * For a general description of rx_handler, see enum rx_handler_result.
3122 int netdev_rx_handler_register(struct net_device *dev,
3123 rx_handler_func_t *rx_handler,
3124 void *rx_handler_data)
3128 if (dev->rx_handler)
3131 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3132 rcu_assign_pointer(dev->rx_handler, rx_handler);
3136 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3139 * netdev_rx_handler_unregister - unregister receive handler
3140 * @dev: device to unregister a handler from
3142 * Unregister a receive hander from a device.
3144 * The caller must hold the rtnl_mutex.
3146 void netdev_rx_handler_unregister(struct net_device *dev)
3150 RCU_INIT_POINTER(dev->rx_handler, NULL);
3151 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3153 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3155 static int __netif_receive_skb(struct sk_buff *skb)
3157 struct packet_type *ptype, *pt_prev;
3158 rx_handler_func_t *rx_handler;
3159 struct net_device *orig_dev;
3160 struct net_device *null_or_dev;
3161 bool deliver_exact = false;
3162 int ret = NET_RX_DROP;
3165 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3167 trace_netif_receive_skb(skb);
3169 /* if we've gotten here through NAPI, check netpoll */
3170 if (netpoll_receive_skb(skb))
3174 skb->skb_iif = skb->dev->ifindex;
3175 orig_dev = skb->dev;
3177 skb_reset_network_header(skb);
3178 skb_reset_transport_header(skb);
3179 skb_reset_mac_len(skb);
3187 __this_cpu_inc(softnet_data.processed);
3189 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3190 skb = vlan_untag(skb);
3195 #ifdef CONFIG_NET_CLS_ACT
3196 if (skb->tc_verd & TC_NCLS) {
3197 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3202 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3203 if (!ptype->dev || ptype->dev == skb->dev) {
3205 ret = deliver_skb(skb, pt_prev, orig_dev);
3210 #ifdef CONFIG_NET_CLS_ACT
3211 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3217 rx_handler = rcu_dereference(skb->dev->rx_handler);
3218 if (vlan_tx_tag_present(skb)) {
3220 ret = deliver_skb(skb, pt_prev, orig_dev);
3223 if (vlan_do_receive(&skb, !rx_handler))
3225 else if (unlikely(!skb))
3231 ret = deliver_skb(skb, pt_prev, orig_dev);
3234 switch (rx_handler(&skb)) {
3235 case RX_HANDLER_CONSUMED:
3237 case RX_HANDLER_ANOTHER:
3239 case RX_HANDLER_EXACT:
3240 deliver_exact = true;
3241 case RX_HANDLER_PASS:
3248 /* deliver only exact match when indicated */
3249 null_or_dev = deliver_exact ? skb->dev : NULL;
3251 type = skb->protocol;
3252 list_for_each_entry_rcu(ptype,
3253 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3254 if (ptype->type == type &&
3255 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3256 ptype->dev == orig_dev)) {
3258 ret = deliver_skb(skb, pt_prev, orig_dev);
3264 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3266 atomic_long_inc(&skb->dev->rx_dropped);
3268 /* Jamal, now you will not able to escape explaining
3269 * me how you were going to use this. :-)
3280 * netif_receive_skb - process receive buffer from network
3281 * @skb: buffer to process
3283 * netif_receive_skb() is the main receive data processing function.
3284 * It always succeeds. The buffer may be dropped during processing
3285 * for congestion control or by the protocol layers.
3287 * This function may only be called from softirq context and interrupts
3288 * should be enabled.
3290 * Return values (usually ignored):
3291 * NET_RX_SUCCESS: no congestion
3292 * NET_RX_DROP: packet was dropped
3294 int netif_receive_skb(struct sk_buff *skb)
3296 net_timestamp_check(netdev_tstamp_prequeue, skb);
3298 if (skb_defer_rx_timestamp(skb))
3299 return NET_RX_SUCCESS;
3302 if (static_key_false(&rps_needed)) {
3303 struct rps_dev_flow voidflow, *rflow = &voidflow;
3308 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3311 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3318 return __netif_receive_skb(skb);
3320 EXPORT_SYMBOL(netif_receive_skb);
3322 /* Network device is going away, flush any packets still pending
3323 * Called with irqs disabled.
3325 static void flush_backlog(void *arg)
3327 struct net_device *dev = arg;
3328 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3329 struct sk_buff *skb, *tmp;
3332 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3333 if (skb->dev == dev) {
3334 __skb_unlink(skb, &sd->input_pkt_queue);
3336 input_queue_head_incr(sd);
3341 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3342 if (skb->dev == dev) {
3343 __skb_unlink(skb, &sd->process_queue);
3345 input_queue_head_incr(sd);
3350 static int napi_gro_complete(struct sk_buff *skb)
3352 struct packet_type *ptype;
3353 __be16 type = skb->protocol;
3354 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3357 if (NAPI_GRO_CB(skb)->count == 1) {
3358 skb_shinfo(skb)->gso_size = 0;
3363 list_for_each_entry_rcu(ptype, head, list) {
3364 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3367 err = ptype->gro_complete(skb);
3373 WARN_ON(&ptype->list == head);
3375 return NET_RX_SUCCESS;
3379 return netif_receive_skb(skb);
3382 inline void napi_gro_flush(struct napi_struct *napi)
3384 struct sk_buff *skb, *next;
3386 for (skb = napi->gro_list; skb; skb = next) {
3389 napi_gro_complete(skb);
3392 napi->gro_count = 0;
3393 napi->gro_list = NULL;
3395 EXPORT_SYMBOL(napi_gro_flush);
3397 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3399 struct sk_buff **pp = NULL;
3400 struct packet_type *ptype;
3401 __be16 type = skb->protocol;
3402 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3405 enum gro_result ret;
3407 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3410 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3414 list_for_each_entry_rcu(ptype, head, list) {
3415 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3418 skb_set_network_header(skb, skb_gro_offset(skb));
3419 mac_len = skb->network_header - skb->mac_header;
3420 skb->mac_len = mac_len;
3421 NAPI_GRO_CB(skb)->same_flow = 0;
3422 NAPI_GRO_CB(skb)->flush = 0;
3423 NAPI_GRO_CB(skb)->free = 0;
3425 pp = ptype->gro_receive(&napi->gro_list, skb);
3430 if (&ptype->list == head)
3433 same_flow = NAPI_GRO_CB(skb)->same_flow;
3434 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3437 struct sk_buff *nskb = *pp;
3441 napi_gro_complete(nskb);
3448 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3452 NAPI_GRO_CB(skb)->count = 1;
3453 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3454 skb->next = napi->gro_list;
3455 napi->gro_list = skb;
3459 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3460 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3462 BUG_ON(skb->end - skb->tail < grow);
3464 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3467 skb->data_len -= grow;
3469 skb_shinfo(skb)->frags[0].page_offset += grow;
3470 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3472 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3473 skb_frag_unref(skb, 0);
3474 memmove(skb_shinfo(skb)->frags,
3475 skb_shinfo(skb)->frags + 1,
3476 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3487 EXPORT_SYMBOL(dev_gro_receive);
3489 static inline gro_result_t
3490 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3493 unsigned int maclen = skb->dev->hard_header_len;
3495 for (p = napi->gro_list; p; p = p->next) {
3496 unsigned long diffs;
3498 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3499 diffs |= p->vlan_tci ^ skb->vlan_tci;
3500 if (maclen == ETH_HLEN)
3501 diffs |= compare_ether_header(skb_mac_header(p),
3502 skb_gro_mac_header(skb));
3504 diffs = memcmp(skb_mac_header(p),
3505 skb_gro_mac_header(skb),
3507 NAPI_GRO_CB(p)->same_flow = !diffs;
3508 NAPI_GRO_CB(p)->flush = 0;
3511 return dev_gro_receive(napi, skb);
3514 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3518 if (netif_receive_skb(skb))
3523 case GRO_MERGED_FREE:
3534 EXPORT_SYMBOL(napi_skb_finish);
3536 void skb_gro_reset_offset(struct sk_buff *skb)
3538 NAPI_GRO_CB(skb)->data_offset = 0;
3539 NAPI_GRO_CB(skb)->frag0 = NULL;
3540 NAPI_GRO_CB(skb)->frag0_len = 0;
3542 if (skb->mac_header == skb->tail &&
3543 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3544 NAPI_GRO_CB(skb)->frag0 =
3545 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3546 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3549 EXPORT_SYMBOL(skb_gro_reset_offset);
3551 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3553 skb_gro_reset_offset(skb);
3555 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3557 EXPORT_SYMBOL(napi_gro_receive);
3559 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3561 __skb_pull(skb, skb_headlen(skb));
3562 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3563 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3565 skb->dev = napi->dev;
3571 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3573 struct sk_buff *skb = napi->skb;
3576 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3582 EXPORT_SYMBOL(napi_get_frags);
3584 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3590 skb->protocol = eth_type_trans(skb, skb->dev);
3592 if (ret == GRO_HELD)
3593 skb_gro_pull(skb, -ETH_HLEN);
3594 else if (netif_receive_skb(skb))
3599 case GRO_MERGED_FREE:
3600 napi_reuse_skb(napi, skb);
3609 EXPORT_SYMBOL(napi_frags_finish);
3611 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3613 struct sk_buff *skb = napi->skb;
3620 skb_reset_mac_header(skb);
3621 skb_gro_reset_offset(skb);
3623 off = skb_gro_offset(skb);
3624 hlen = off + sizeof(*eth);
3625 eth = skb_gro_header_fast(skb, off);
3626 if (skb_gro_header_hard(skb, hlen)) {
3627 eth = skb_gro_header_slow(skb, hlen, off);
3628 if (unlikely(!eth)) {
3629 napi_reuse_skb(napi, skb);
3635 skb_gro_pull(skb, sizeof(*eth));
3638 * This works because the only protocols we care about don't require
3639 * special handling. We'll fix it up properly at the end.
3641 skb->protocol = eth->h_proto;
3646 EXPORT_SYMBOL(napi_frags_skb);
3648 gro_result_t napi_gro_frags(struct napi_struct *napi)
3650 struct sk_buff *skb = napi_frags_skb(napi);
3655 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3657 EXPORT_SYMBOL(napi_gro_frags);
3660 * net_rps_action sends any pending IPI's for rps.
3661 * Note: called with local irq disabled, but exits with local irq enabled.
3663 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3666 struct softnet_data *remsd = sd->rps_ipi_list;
3669 sd->rps_ipi_list = NULL;
3673 /* Send pending IPI's to kick RPS processing on remote cpus. */
3675 struct softnet_data *next = remsd->rps_ipi_next;
3677 if (cpu_online(remsd->cpu))
3678 __smp_call_function_single(remsd->cpu,
3687 static int process_backlog(struct napi_struct *napi, int quota)
3690 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3693 /* Check if we have pending ipi, its better to send them now,
3694 * not waiting net_rx_action() end.
3696 if (sd->rps_ipi_list) {
3697 local_irq_disable();
3698 net_rps_action_and_irq_enable(sd);
3701 napi->weight = weight_p;
3702 local_irq_disable();
3703 while (work < quota) {
3704 struct sk_buff *skb;
3707 while ((skb = __skb_dequeue(&sd->process_queue))) {
3709 __netif_receive_skb(skb);
3710 local_irq_disable();
3711 input_queue_head_incr(sd);
3712 if (++work >= quota) {
3719 qlen = skb_queue_len(&sd->input_pkt_queue);
3721 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3722 &sd->process_queue);
3724 if (qlen < quota - work) {
3726 * Inline a custom version of __napi_complete().
3727 * only current cpu owns and manipulates this napi,
3728 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3729 * we can use a plain write instead of clear_bit(),
3730 * and we dont need an smp_mb() memory barrier.
3732 list_del(&napi->poll_list);
3735 quota = work + qlen;
3745 * __napi_schedule - schedule for receive
3746 * @n: entry to schedule
3748 * The entry's receive function will be scheduled to run
3750 void __napi_schedule(struct napi_struct *n)
3752 unsigned long flags;
3754 local_irq_save(flags);
3755 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3756 local_irq_restore(flags);
3758 EXPORT_SYMBOL(__napi_schedule);
3760 void __napi_complete(struct napi_struct *n)
3762 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3763 BUG_ON(n->gro_list);
3765 list_del(&n->poll_list);
3766 smp_mb__before_clear_bit();
3767 clear_bit(NAPI_STATE_SCHED, &n->state);
3769 EXPORT_SYMBOL(__napi_complete);
3771 void napi_complete(struct napi_struct *n)
3773 unsigned long flags;
3776 * don't let napi dequeue from the cpu poll list
3777 * just in case its running on a different cpu
3779 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3783 local_irq_save(flags);
3785 local_irq_restore(flags);
3787 EXPORT_SYMBOL(napi_complete);
3789 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3790 int (*poll)(struct napi_struct *, int), int weight)
3792 INIT_LIST_HEAD(&napi->poll_list);
3793 napi->gro_count = 0;
3794 napi->gro_list = NULL;
3797 napi->weight = weight;
3798 list_add(&napi->dev_list, &dev->napi_list);
3800 #ifdef CONFIG_NETPOLL
3801 spin_lock_init(&napi->poll_lock);
3802 napi->poll_owner = -1;
3804 set_bit(NAPI_STATE_SCHED, &napi->state);
3806 EXPORT_SYMBOL(netif_napi_add);
3808 void netif_napi_del(struct napi_struct *napi)
3810 struct sk_buff *skb, *next;
3812 list_del_init(&napi->dev_list);
3813 napi_free_frags(napi);
3815 for (skb = napi->gro_list; skb; skb = next) {
3821 napi->gro_list = NULL;
3822 napi->gro_count = 0;
3824 EXPORT_SYMBOL(netif_napi_del);
3826 static void net_rx_action(struct softirq_action *h)
3828 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3829 unsigned long time_limit = jiffies + 2;
3830 int budget = netdev_budget;
3833 local_irq_disable();
3835 while (!list_empty(&sd->poll_list)) {
3836 struct napi_struct *n;
3839 /* If softirq window is exhuasted then punt.
3840 * Allow this to run for 2 jiffies since which will allow
3841 * an average latency of 1.5/HZ.
3843 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3848 /* Even though interrupts have been re-enabled, this
3849 * access is safe because interrupts can only add new
3850 * entries to the tail of this list, and only ->poll()
3851 * calls can remove this head entry from the list.
3853 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3855 have = netpoll_poll_lock(n);
3859 /* This NAPI_STATE_SCHED test is for avoiding a race
3860 * with netpoll's poll_napi(). Only the entity which
3861 * obtains the lock and sees NAPI_STATE_SCHED set will
3862 * actually make the ->poll() call. Therefore we avoid
3863 * accidentally calling ->poll() when NAPI is not scheduled.
3866 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3867 work = n->poll(n, weight);
3871 WARN_ON_ONCE(work > weight);
3875 local_irq_disable();
3877 /* Drivers must not modify the NAPI state if they
3878 * consume the entire weight. In such cases this code
3879 * still "owns" the NAPI instance and therefore can
3880 * move the instance around on the list at-will.
3882 if (unlikely(work == weight)) {
3883 if (unlikely(napi_disable_pending(n))) {
3886 local_irq_disable();
3888 list_move_tail(&n->poll_list, &sd->poll_list);
3891 netpoll_poll_unlock(have);
3894 net_rps_action_and_irq_enable(sd);
3896 #ifdef CONFIG_NET_DMA
3898 * There may not be any more sk_buffs coming right now, so push
3899 * any pending DMA copies to hardware
3901 dma_issue_pending_all();
3908 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3912 static gifconf_func_t *gifconf_list[NPROTO];
3915 * register_gifconf - register a SIOCGIF handler
3916 * @family: Address family
3917 * @gifconf: Function handler
3919 * Register protocol dependent address dumping routines. The handler
3920 * that is passed must not be freed or reused until it has been replaced
3921 * by another handler.
3923 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3925 if (family >= NPROTO)
3927 gifconf_list[family] = gifconf;
3930 EXPORT_SYMBOL(register_gifconf);
3934 * Map an interface index to its name (SIOCGIFNAME)
3938 * We need this ioctl for efficient implementation of the
3939 * if_indextoname() function required by the IPv6 API. Without
3940 * it, we would have to search all the interfaces to find a
3944 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3946 struct net_device *dev;
3950 * Fetch the caller's info block.
3953 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3957 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3963 strcpy(ifr.ifr_name, dev->name);
3966 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3972 * Perform a SIOCGIFCONF call. This structure will change
3973 * size eventually, and there is nothing I can do about it.
3974 * Thus we will need a 'compatibility mode'.
3977 static int dev_ifconf(struct net *net, char __user *arg)
3980 struct net_device *dev;
3987 * Fetch the caller's info block.
3990 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3997 * Loop over the interfaces, and write an info block for each.
4001 for_each_netdev(net, dev) {
4002 for (i = 0; i < NPROTO; i++) {
4003 if (gifconf_list[i]) {
4006 done = gifconf_list[i](dev, NULL, 0);
4008 done = gifconf_list[i](dev, pos + total,
4018 * All done. Write the updated control block back to the caller.
4020 ifc.ifc_len = total;
4023 * Both BSD and Solaris return 0 here, so we do too.
4025 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4028 #ifdef CONFIG_PROC_FS
4030 #define BUCKET_SPACE (32 - NETDEV_HASHBITS)
4032 struct dev_iter_state {
4033 struct seq_net_private p;
4034 unsigned int pos; /* bucket << BUCKET_SPACE + offset */
4037 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4038 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4039 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4041 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq)
4043 struct dev_iter_state *state = seq->private;
4044 struct net *net = seq_file_net(seq);
4045 struct net_device *dev;
4046 struct hlist_node *p;
4047 struct hlist_head *h;
4048 unsigned int count, bucket, offset;
4050 bucket = get_bucket(state->pos);
4051 offset = get_offset(state->pos);
4052 h = &net->dev_name_head[bucket];
4054 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4055 if (count++ == offset) {
4056 state->pos = set_bucket_offset(bucket, count);
4064 static inline struct net_device *dev_from_new_bucket(struct seq_file *seq)
4066 struct dev_iter_state *state = seq->private;
4067 struct net_device *dev;
4068 unsigned int bucket;
4070 bucket = get_bucket(state->pos);
4072 dev = dev_from_same_bucket(seq);
4077 state->pos = set_bucket_offset(bucket, 0);
4078 } while (bucket < NETDEV_HASHENTRIES);
4084 * This is invoked by the /proc filesystem handler to display a device
4087 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4090 struct dev_iter_state *state = seq->private;
4094 return SEQ_START_TOKEN;
4096 /* check for end of the hash */
4097 if (state->pos == 0 && *pos > 1)
4100 return dev_from_new_bucket(seq);
4103 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4105 struct net_device *dev;
4109 if (v == SEQ_START_TOKEN)
4110 return dev_from_new_bucket(seq);
4112 dev = dev_from_same_bucket(seq);
4116 return dev_from_new_bucket(seq);
4119 void dev_seq_stop(struct seq_file *seq, void *v)
4125 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4127 struct rtnl_link_stats64 temp;
4128 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4130 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4131 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4132 dev->name, stats->rx_bytes, stats->rx_packets,
4134 stats->rx_dropped + stats->rx_missed_errors,
4135 stats->rx_fifo_errors,
4136 stats->rx_length_errors + stats->rx_over_errors +
4137 stats->rx_crc_errors + stats->rx_frame_errors,
4138 stats->rx_compressed, stats->multicast,
4139 stats->tx_bytes, stats->tx_packets,
4140 stats->tx_errors, stats->tx_dropped,
4141 stats->tx_fifo_errors, stats->collisions,
4142 stats->tx_carrier_errors +
4143 stats->tx_aborted_errors +
4144 stats->tx_window_errors +
4145 stats->tx_heartbeat_errors,
4146 stats->tx_compressed);
4150 * Called from the PROCfs module. This now uses the new arbitrary sized
4151 * /proc/net interface to create /proc/net/dev
4153 static int dev_seq_show(struct seq_file *seq, void *v)
4155 if (v == SEQ_START_TOKEN)
4156 seq_puts(seq, "Inter-| Receive "
4158 " face |bytes packets errs drop fifo frame "
4159 "compressed multicast|bytes packets errs "
4160 "drop fifo colls carrier compressed\n");
4162 dev_seq_printf_stats(seq, v);
4166 static struct softnet_data *softnet_get_online(loff_t *pos)
4168 struct softnet_data *sd = NULL;
4170 while (*pos < nr_cpu_ids)
4171 if (cpu_online(*pos)) {
4172 sd = &per_cpu(softnet_data, *pos);
4179 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4181 return softnet_get_online(pos);
4184 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4187 return softnet_get_online(pos);
4190 static void softnet_seq_stop(struct seq_file *seq, void *v)
4194 static int softnet_seq_show(struct seq_file *seq, void *v)
4196 struct softnet_data *sd = v;
4198 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4199 sd->processed, sd->dropped, sd->time_squeeze, 0,
4200 0, 0, 0, 0, /* was fastroute */
4201 sd->cpu_collision, sd->received_rps);
4205 static const struct seq_operations dev_seq_ops = {
4206 .start = dev_seq_start,
4207 .next = dev_seq_next,
4208 .stop = dev_seq_stop,
4209 .show = dev_seq_show,
4212 static int dev_seq_open(struct inode *inode, struct file *file)
4214 return seq_open_net(inode, file, &dev_seq_ops,
4215 sizeof(struct dev_iter_state));
4218 int dev_seq_open_ops(struct inode *inode, struct file *file,
4219 const struct seq_operations *ops)
4221 return seq_open_net(inode, file, ops, sizeof(struct dev_iter_state));
4224 static const struct file_operations dev_seq_fops = {
4225 .owner = THIS_MODULE,
4226 .open = dev_seq_open,
4228 .llseek = seq_lseek,
4229 .release = seq_release_net,
4232 static const struct seq_operations softnet_seq_ops = {
4233 .start = softnet_seq_start,
4234 .next = softnet_seq_next,
4235 .stop = softnet_seq_stop,
4236 .show = softnet_seq_show,
4239 static int softnet_seq_open(struct inode *inode, struct file *file)
4241 return seq_open(file, &softnet_seq_ops);
4244 static const struct file_operations softnet_seq_fops = {
4245 .owner = THIS_MODULE,
4246 .open = softnet_seq_open,
4248 .llseek = seq_lseek,
4249 .release = seq_release,
4252 static void *ptype_get_idx(loff_t pos)
4254 struct packet_type *pt = NULL;
4258 list_for_each_entry_rcu(pt, &ptype_all, list) {
4264 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4265 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4274 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4278 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4281 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4283 struct packet_type *pt;
4284 struct list_head *nxt;
4288 if (v == SEQ_START_TOKEN)
4289 return ptype_get_idx(0);
4292 nxt = pt->list.next;
4293 if (pt->type == htons(ETH_P_ALL)) {
4294 if (nxt != &ptype_all)
4297 nxt = ptype_base[0].next;
4299 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4301 while (nxt == &ptype_base[hash]) {
4302 if (++hash >= PTYPE_HASH_SIZE)
4304 nxt = ptype_base[hash].next;
4307 return list_entry(nxt, struct packet_type, list);
4310 static void ptype_seq_stop(struct seq_file *seq, void *v)
4316 static int ptype_seq_show(struct seq_file *seq, void *v)
4318 struct packet_type *pt = v;
4320 if (v == SEQ_START_TOKEN)
4321 seq_puts(seq, "Type Device Function\n");
4322 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4323 if (pt->type == htons(ETH_P_ALL))
4324 seq_puts(seq, "ALL ");
4326 seq_printf(seq, "%04x", ntohs(pt->type));
4328 seq_printf(seq, " %-8s %pF\n",
4329 pt->dev ? pt->dev->name : "", pt->func);
4335 static const struct seq_operations ptype_seq_ops = {
4336 .start = ptype_seq_start,
4337 .next = ptype_seq_next,
4338 .stop = ptype_seq_stop,
4339 .show = ptype_seq_show,
4342 static int ptype_seq_open(struct inode *inode, struct file *file)
4344 return seq_open_net(inode, file, &ptype_seq_ops,
4345 sizeof(struct seq_net_private));
4348 static const struct file_operations ptype_seq_fops = {
4349 .owner = THIS_MODULE,
4350 .open = ptype_seq_open,
4352 .llseek = seq_lseek,
4353 .release = seq_release_net,
4357 static int __net_init dev_proc_net_init(struct net *net)
4361 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4363 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4365 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4368 if (wext_proc_init(net))
4374 proc_net_remove(net, "ptype");
4376 proc_net_remove(net, "softnet_stat");
4378 proc_net_remove(net, "dev");
4382 static void __net_exit dev_proc_net_exit(struct net *net)
4384 wext_proc_exit(net);
4386 proc_net_remove(net, "ptype");
4387 proc_net_remove(net, "softnet_stat");
4388 proc_net_remove(net, "dev");
4391 static struct pernet_operations __net_initdata dev_proc_ops = {
4392 .init = dev_proc_net_init,
4393 .exit = dev_proc_net_exit,
4396 static int __init dev_proc_init(void)
4398 return register_pernet_subsys(&dev_proc_ops);
4401 #define dev_proc_init() 0
4402 #endif /* CONFIG_PROC_FS */
4406 * netdev_set_master - set up master pointer
4407 * @slave: slave device
4408 * @master: new master device
4410 * Changes the master device of the slave. Pass %NULL to break the
4411 * bonding. The caller must hold the RTNL semaphore. On a failure
4412 * a negative errno code is returned. On success the reference counts
4413 * are adjusted and the function returns zero.
4415 int netdev_set_master(struct net_device *slave, struct net_device *master)
4417 struct net_device *old = slave->master;
4427 slave->master = master;
4433 EXPORT_SYMBOL(netdev_set_master);
4436 * netdev_set_bond_master - set up bonding master/slave pair
4437 * @slave: slave device
4438 * @master: new master device
4440 * Changes the master device of the slave. Pass %NULL to break the
4441 * bonding. The caller must hold the RTNL semaphore. On a failure
4442 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4443 * to the routing socket and the function returns zero.
4445 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4451 err = netdev_set_master(slave, master);
4455 slave->flags |= IFF_SLAVE;
4457 slave->flags &= ~IFF_SLAVE;
4459 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4462 EXPORT_SYMBOL(netdev_set_bond_master);
4464 static void dev_change_rx_flags(struct net_device *dev, int flags)
4466 const struct net_device_ops *ops = dev->netdev_ops;
4468 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4469 ops->ndo_change_rx_flags(dev, flags);
4472 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4474 unsigned int old_flags = dev->flags;
4480 dev->flags |= IFF_PROMISC;
4481 dev->promiscuity += inc;
4482 if (dev->promiscuity == 0) {
4485 * If inc causes overflow, untouch promisc and return error.
4488 dev->flags &= ~IFF_PROMISC;
4490 dev->promiscuity -= inc;
4491 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4496 if (dev->flags != old_flags) {
4497 pr_info("device %s %s promiscuous mode\n",
4499 dev->flags & IFF_PROMISC ? "entered" : "left");
4500 if (audit_enabled) {
4501 current_uid_gid(&uid, &gid);
4502 audit_log(current->audit_context, GFP_ATOMIC,
4503 AUDIT_ANOM_PROMISCUOUS,
4504 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4505 dev->name, (dev->flags & IFF_PROMISC),
4506 (old_flags & IFF_PROMISC),
4507 audit_get_loginuid(current),
4509 audit_get_sessionid(current));
4512 dev_change_rx_flags(dev, IFF_PROMISC);
4518 * dev_set_promiscuity - update promiscuity count on a device
4522 * Add or remove promiscuity from a device. While the count in the device
4523 * remains above zero the interface remains promiscuous. Once it hits zero
4524 * the device reverts back to normal filtering operation. A negative inc
4525 * value is used to drop promiscuity on the device.
4526 * Return 0 if successful or a negative errno code on error.
4528 int dev_set_promiscuity(struct net_device *dev, int inc)
4530 unsigned int old_flags = dev->flags;
4533 err = __dev_set_promiscuity(dev, inc);
4536 if (dev->flags != old_flags)
4537 dev_set_rx_mode(dev);
4540 EXPORT_SYMBOL(dev_set_promiscuity);
4543 * dev_set_allmulti - update allmulti count on a device
4547 * Add or remove reception of all multicast frames to a device. While the
4548 * count in the device remains above zero the interface remains listening
4549 * to all interfaces. Once it hits zero the device reverts back to normal
4550 * filtering operation. A negative @inc value is used to drop the counter
4551 * when releasing a resource needing all multicasts.
4552 * Return 0 if successful or a negative errno code on error.
4555 int dev_set_allmulti(struct net_device *dev, int inc)
4557 unsigned int old_flags = dev->flags;
4561 dev->flags |= IFF_ALLMULTI;
4562 dev->allmulti += inc;
4563 if (dev->allmulti == 0) {
4566 * If inc causes overflow, untouch allmulti and return error.
4569 dev->flags &= ~IFF_ALLMULTI;
4571 dev->allmulti -= inc;
4572 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4577 if (dev->flags ^ old_flags) {
4578 dev_change_rx_flags(dev, IFF_ALLMULTI);
4579 dev_set_rx_mode(dev);
4583 EXPORT_SYMBOL(dev_set_allmulti);
4586 * Upload unicast and multicast address lists to device and
4587 * configure RX filtering. When the device doesn't support unicast
4588 * filtering it is put in promiscuous mode while unicast addresses
4591 void __dev_set_rx_mode(struct net_device *dev)
4593 const struct net_device_ops *ops = dev->netdev_ops;
4595 /* dev_open will call this function so the list will stay sane. */
4596 if (!(dev->flags&IFF_UP))
4599 if (!netif_device_present(dev))
4602 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4603 /* Unicast addresses changes may only happen under the rtnl,
4604 * therefore calling __dev_set_promiscuity here is safe.
4606 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4607 __dev_set_promiscuity(dev, 1);
4608 dev->uc_promisc = true;
4609 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4610 __dev_set_promiscuity(dev, -1);
4611 dev->uc_promisc = false;
4615 if (ops->ndo_set_rx_mode)
4616 ops->ndo_set_rx_mode(dev);
4619 void dev_set_rx_mode(struct net_device *dev)
4621 netif_addr_lock_bh(dev);
4622 __dev_set_rx_mode(dev);
4623 netif_addr_unlock_bh(dev);
4627 * dev_get_flags - get flags reported to userspace
4630 * Get the combination of flag bits exported through APIs to userspace.
4632 unsigned dev_get_flags(const struct net_device *dev)
4636 flags = (dev->flags & ~(IFF_PROMISC |
4641 (dev->gflags & (IFF_PROMISC |
4644 if (netif_running(dev)) {
4645 if (netif_oper_up(dev))
4646 flags |= IFF_RUNNING;
4647 if (netif_carrier_ok(dev))
4648 flags |= IFF_LOWER_UP;
4649 if (netif_dormant(dev))
4650 flags |= IFF_DORMANT;
4655 EXPORT_SYMBOL(dev_get_flags);
4657 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4659 unsigned int old_flags = dev->flags;
4665 * Set the flags on our device.
4668 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4669 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4671 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4675 * Load in the correct multicast list now the flags have changed.
4678 if ((old_flags ^ flags) & IFF_MULTICAST)
4679 dev_change_rx_flags(dev, IFF_MULTICAST);
4681 dev_set_rx_mode(dev);
4684 * Have we downed the interface. We handle IFF_UP ourselves
4685 * according to user attempts to set it, rather than blindly
4690 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4691 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4694 dev_set_rx_mode(dev);
4697 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4698 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4700 dev->gflags ^= IFF_PROMISC;
4701 dev_set_promiscuity(dev, inc);
4704 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4705 is important. Some (broken) drivers set IFF_PROMISC, when
4706 IFF_ALLMULTI is requested not asking us and not reporting.
4708 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4709 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4711 dev->gflags ^= IFF_ALLMULTI;
4712 dev_set_allmulti(dev, inc);
4718 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4720 unsigned int changes = dev->flags ^ old_flags;
4722 if (changes & IFF_UP) {
4723 if (dev->flags & IFF_UP)
4724 call_netdevice_notifiers(NETDEV_UP, dev);
4726 call_netdevice_notifiers(NETDEV_DOWN, dev);
4729 if (dev->flags & IFF_UP &&
4730 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4731 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4735 * dev_change_flags - change device settings
4737 * @flags: device state flags
4739 * Change settings on device based state flags. The flags are
4740 * in the userspace exported format.
4742 int dev_change_flags(struct net_device *dev, unsigned int flags)
4745 unsigned int changes, old_flags = dev->flags;
4747 ret = __dev_change_flags(dev, flags);
4751 changes = old_flags ^ dev->flags;
4753 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4755 __dev_notify_flags(dev, old_flags);
4758 EXPORT_SYMBOL(dev_change_flags);
4761 * dev_set_mtu - Change maximum transfer unit
4763 * @new_mtu: new transfer unit
4765 * Change the maximum transfer size of the network device.
4767 int dev_set_mtu(struct net_device *dev, int new_mtu)
4769 const struct net_device_ops *ops = dev->netdev_ops;
4772 if (new_mtu == dev->mtu)
4775 /* MTU must be positive. */
4779 if (!netif_device_present(dev))
4783 if (ops->ndo_change_mtu)
4784 err = ops->ndo_change_mtu(dev, new_mtu);
4788 if (!err && dev->flags & IFF_UP)
4789 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4792 EXPORT_SYMBOL(dev_set_mtu);
4795 * dev_set_group - Change group this device belongs to
4797 * @new_group: group this device should belong to
4799 void dev_set_group(struct net_device *dev, int new_group)
4801 dev->group = new_group;
4803 EXPORT_SYMBOL(dev_set_group);
4806 * dev_set_mac_address - Change Media Access Control Address
4810 * Change the hardware (MAC) address of the device
4812 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4814 const struct net_device_ops *ops = dev->netdev_ops;
4817 if (!ops->ndo_set_mac_address)
4819 if (sa->sa_family != dev->type)
4821 if (!netif_device_present(dev))
4823 err = ops->ndo_set_mac_address(dev, sa);
4825 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4828 EXPORT_SYMBOL(dev_set_mac_address);
4831 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4833 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4836 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4842 case SIOCGIFFLAGS: /* Get interface flags */
4843 ifr->ifr_flags = (short) dev_get_flags(dev);
4846 case SIOCGIFMETRIC: /* Get the metric on the interface
4847 (currently unused) */
4848 ifr->ifr_metric = 0;
4851 case SIOCGIFMTU: /* Get the MTU of a device */
4852 ifr->ifr_mtu = dev->mtu;
4857 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4859 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4860 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4861 ifr->ifr_hwaddr.sa_family = dev->type;
4869 ifr->ifr_map.mem_start = dev->mem_start;
4870 ifr->ifr_map.mem_end = dev->mem_end;
4871 ifr->ifr_map.base_addr = dev->base_addr;
4872 ifr->ifr_map.irq = dev->irq;
4873 ifr->ifr_map.dma = dev->dma;
4874 ifr->ifr_map.port = dev->if_port;
4878 ifr->ifr_ifindex = dev->ifindex;
4882 ifr->ifr_qlen = dev->tx_queue_len;
4886 /* dev_ioctl() should ensure this case
4898 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4900 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4903 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4904 const struct net_device_ops *ops;
4909 ops = dev->netdev_ops;
4912 case SIOCSIFFLAGS: /* Set interface flags */
4913 return dev_change_flags(dev, ifr->ifr_flags);
4915 case SIOCSIFMETRIC: /* Set the metric on the interface
4916 (currently unused) */
4919 case SIOCSIFMTU: /* Set the MTU of a device */
4920 return dev_set_mtu(dev, ifr->ifr_mtu);
4923 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4925 case SIOCSIFHWBROADCAST:
4926 if (ifr->ifr_hwaddr.sa_family != dev->type)
4928 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4929 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4930 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4934 if (ops->ndo_set_config) {
4935 if (!netif_device_present(dev))
4937 return ops->ndo_set_config(dev, &ifr->ifr_map);
4942 if (!ops->ndo_set_rx_mode ||
4943 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4945 if (!netif_device_present(dev))
4947 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4950 if (!ops->ndo_set_rx_mode ||
4951 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4953 if (!netif_device_present(dev))
4955 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4958 if (ifr->ifr_qlen < 0)
4960 dev->tx_queue_len = ifr->ifr_qlen;
4964 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4965 return dev_change_name(dev, ifr->ifr_newname);
4968 err = net_hwtstamp_validate(ifr);
4974 * Unknown or private ioctl
4977 if ((cmd >= SIOCDEVPRIVATE &&
4978 cmd <= SIOCDEVPRIVATE + 15) ||
4979 cmd == SIOCBONDENSLAVE ||
4980 cmd == SIOCBONDRELEASE ||
4981 cmd == SIOCBONDSETHWADDR ||
4982 cmd == SIOCBONDSLAVEINFOQUERY ||
4983 cmd == SIOCBONDINFOQUERY ||
4984 cmd == SIOCBONDCHANGEACTIVE ||
4985 cmd == SIOCGMIIPHY ||
4986 cmd == SIOCGMIIREG ||
4987 cmd == SIOCSMIIREG ||
4988 cmd == SIOCBRADDIF ||
4989 cmd == SIOCBRDELIF ||
4990 cmd == SIOCSHWTSTAMP ||
4991 cmd == SIOCWANDEV) {
4993 if (ops->ndo_do_ioctl) {
4994 if (netif_device_present(dev))
4995 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5007 * This function handles all "interface"-type I/O control requests. The actual
5008 * 'doing' part of this is dev_ifsioc above.
5012 * dev_ioctl - network device ioctl
5013 * @net: the applicable net namespace
5014 * @cmd: command to issue
5015 * @arg: pointer to a struct ifreq in user space
5017 * Issue ioctl functions to devices. This is normally called by the
5018 * user space syscall interfaces but can sometimes be useful for
5019 * other purposes. The return value is the return from the syscall if
5020 * positive or a negative errno code on error.
5023 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5029 /* One special case: SIOCGIFCONF takes ifconf argument
5030 and requires shared lock, because it sleeps writing
5034 if (cmd == SIOCGIFCONF) {
5036 ret = dev_ifconf(net, (char __user *) arg);
5040 if (cmd == SIOCGIFNAME)
5041 return dev_ifname(net, (struct ifreq __user *)arg);
5043 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5046 ifr.ifr_name[IFNAMSIZ-1] = 0;
5048 colon = strchr(ifr.ifr_name, ':');
5053 * See which interface the caller is talking about.
5058 * These ioctl calls:
5059 * - can be done by all.
5060 * - atomic and do not require locking.
5071 dev_load(net, ifr.ifr_name);
5073 ret = dev_ifsioc_locked(net, &ifr, cmd);
5078 if (copy_to_user(arg, &ifr,
5079 sizeof(struct ifreq)))
5085 dev_load(net, ifr.ifr_name);
5087 ret = dev_ethtool(net, &ifr);
5092 if (copy_to_user(arg, &ifr,
5093 sizeof(struct ifreq)))
5099 * These ioctl calls:
5100 * - require superuser power.
5101 * - require strict serialization.
5107 if (!capable(CAP_NET_ADMIN))
5109 dev_load(net, ifr.ifr_name);
5111 ret = dev_ifsioc(net, &ifr, cmd);
5116 if (copy_to_user(arg, &ifr,
5117 sizeof(struct ifreq)))
5123 * These ioctl calls:
5124 * - require superuser power.
5125 * - require strict serialization.
5126 * - do not return a value
5136 case SIOCSIFHWBROADCAST:
5139 case SIOCBONDENSLAVE:
5140 case SIOCBONDRELEASE:
5141 case SIOCBONDSETHWADDR:
5142 case SIOCBONDCHANGEACTIVE:
5146 if (!capable(CAP_NET_ADMIN))
5149 case SIOCBONDSLAVEINFOQUERY:
5150 case SIOCBONDINFOQUERY:
5151 dev_load(net, ifr.ifr_name);
5153 ret = dev_ifsioc(net, &ifr, cmd);
5158 /* Get the per device memory space. We can add this but
5159 * currently do not support it */
5161 /* Set the per device memory buffer space.
5162 * Not applicable in our case */
5167 * Unknown or private ioctl.
5170 if (cmd == SIOCWANDEV ||
5171 (cmd >= SIOCDEVPRIVATE &&
5172 cmd <= SIOCDEVPRIVATE + 15)) {
5173 dev_load(net, ifr.ifr_name);
5175 ret = dev_ifsioc(net, &ifr, cmd);
5177 if (!ret && copy_to_user(arg, &ifr,
5178 sizeof(struct ifreq)))
5182 /* Take care of Wireless Extensions */
5183 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5184 return wext_handle_ioctl(net, &ifr, cmd, arg);
5191 * dev_new_index - allocate an ifindex
5192 * @net: the applicable net namespace
5194 * Returns a suitable unique value for a new device interface
5195 * number. The caller must hold the rtnl semaphore or the
5196 * dev_base_lock to be sure it remains unique.
5198 static int dev_new_index(struct net *net)
5204 if (!__dev_get_by_index(net, ifindex))
5209 /* Delayed registration/unregisteration */
5210 static LIST_HEAD(net_todo_list);
5212 static void net_set_todo(struct net_device *dev)
5214 list_add_tail(&dev->todo_list, &net_todo_list);
5217 static void rollback_registered_many(struct list_head *head)
5219 struct net_device *dev, *tmp;
5221 BUG_ON(dev_boot_phase);
5224 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5225 /* Some devices call without registering
5226 * for initialization unwind. Remove those
5227 * devices and proceed with the remaining.
5229 if (dev->reg_state == NETREG_UNINITIALIZED) {
5230 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5234 list_del(&dev->unreg_list);
5237 dev->dismantle = true;
5238 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5241 /* If device is running, close it first. */
5242 dev_close_many(head);
5244 list_for_each_entry(dev, head, unreg_list) {
5245 /* And unlink it from device chain. */
5246 unlist_netdevice(dev);
5248 dev->reg_state = NETREG_UNREGISTERING;
5253 list_for_each_entry(dev, head, unreg_list) {
5254 /* Shutdown queueing discipline. */
5258 /* Notify protocols, that we are about to destroy
5259 this device. They should clean all the things.
5261 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5263 if (!dev->rtnl_link_ops ||
5264 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5265 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5268 * Flush the unicast and multicast chains
5273 if (dev->netdev_ops->ndo_uninit)
5274 dev->netdev_ops->ndo_uninit(dev);
5276 /* Notifier chain MUST detach us from master device. */
5277 WARN_ON(dev->master);
5279 /* Remove entries from kobject tree */
5280 netdev_unregister_kobject(dev);
5283 /* Process any work delayed until the end of the batch */
5284 dev = list_first_entry(head, struct net_device, unreg_list);
5285 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5289 list_for_each_entry(dev, head, unreg_list)
5293 static void rollback_registered(struct net_device *dev)
5297 list_add(&dev->unreg_list, &single);
5298 rollback_registered_many(&single);
5302 static netdev_features_t netdev_fix_features(struct net_device *dev,
5303 netdev_features_t features)
5305 /* Fix illegal checksum combinations */
5306 if ((features & NETIF_F_HW_CSUM) &&
5307 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5308 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5309 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5312 /* Fix illegal SG+CSUM combinations. */
5313 if ((features & NETIF_F_SG) &&
5314 !(features & NETIF_F_ALL_CSUM)) {
5316 "Dropping NETIF_F_SG since no checksum feature.\n");
5317 features &= ~NETIF_F_SG;
5320 /* TSO requires that SG is present as well. */
5321 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5322 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5323 features &= ~NETIF_F_ALL_TSO;
5326 /* TSO ECN requires that TSO is present as well. */
5327 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5328 features &= ~NETIF_F_TSO_ECN;
5330 /* Software GSO depends on SG. */
5331 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5332 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5333 features &= ~NETIF_F_GSO;
5336 /* UFO needs SG and checksumming */
5337 if (features & NETIF_F_UFO) {
5338 /* maybe split UFO into V4 and V6? */
5339 if (!((features & NETIF_F_GEN_CSUM) ||
5340 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5341 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5343 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5344 features &= ~NETIF_F_UFO;
5347 if (!(features & NETIF_F_SG)) {
5349 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5350 features &= ~NETIF_F_UFO;
5357 int __netdev_update_features(struct net_device *dev)
5359 netdev_features_t features;
5364 features = netdev_get_wanted_features(dev);
5366 if (dev->netdev_ops->ndo_fix_features)
5367 features = dev->netdev_ops->ndo_fix_features(dev, features);
5369 /* driver might be less strict about feature dependencies */
5370 features = netdev_fix_features(dev, features);
5372 if (dev->features == features)
5375 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5376 &dev->features, &features);
5378 if (dev->netdev_ops->ndo_set_features)
5379 err = dev->netdev_ops->ndo_set_features(dev, features);
5381 if (unlikely(err < 0)) {
5383 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5384 err, &features, &dev->features);
5389 dev->features = features;
5395 * netdev_update_features - recalculate device features
5396 * @dev: the device to check
5398 * Recalculate dev->features set and send notifications if it
5399 * has changed. Should be called after driver or hardware dependent
5400 * conditions might have changed that influence the features.
5402 void netdev_update_features(struct net_device *dev)
5404 if (__netdev_update_features(dev))
5405 netdev_features_change(dev);
5407 EXPORT_SYMBOL(netdev_update_features);
5410 * netdev_change_features - recalculate device features
5411 * @dev: the device to check
5413 * Recalculate dev->features set and send notifications even
5414 * if they have not changed. Should be called instead of
5415 * netdev_update_features() if also dev->vlan_features might
5416 * have changed to allow the changes to be propagated to stacked
5419 void netdev_change_features(struct net_device *dev)
5421 __netdev_update_features(dev);
5422 netdev_features_change(dev);
5424 EXPORT_SYMBOL(netdev_change_features);
5427 * netif_stacked_transfer_operstate - transfer operstate
5428 * @rootdev: the root or lower level device to transfer state from
5429 * @dev: the device to transfer operstate to
5431 * Transfer operational state from root to device. This is normally
5432 * called when a stacking relationship exists between the root
5433 * device and the device(a leaf device).
5435 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5436 struct net_device *dev)
5438 if (rootdev->operstate == IF_OPER_DORMANT)
5439 netif_dormant_on(dev);
5441 netif_dormant_off(dev);
5443 if (netif_carrier_ok(rootdev)) {
5444 if (!netif_carrier_ok(dev))
5445 netif_carrier_on(dev);
5447 if (netif_carrier_ok(dev))
5448 netif_carrier_off(dev);
5451 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5454 static int netif_alloc_rx_queues(struct net_device *dev)
5456 unsigned int i, count = dev->num_rx_queues;
5457 struct netdev_rx_queue *rx;
5461 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5463 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5468 for (i = 0; i < count; i++)
5474 static void netdev_init_one_queue(struct net_device *dev,
5475 struct netdev_queue *queue, void *_unused)
5477 /* Initialize queue lock */
5478 spin_lock_init(&queue->_xmit_lock);
5479 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5480 queue->xmit_lock_owner = -1;
5481 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5484 dql_init(&queue->dql, HZ);
5488 static int netif_alloc_netdev_queues(struct net_device *dev)
5490 unsigned int count = dev->num_tx_queues;
5491 struct netdev_queue *tx;
5495 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5497 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5502 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5503 spin_lock_init(&dev->tx_global_lock);
5509 * register_netdevice - register a network device
5510 * @dev: device to register
5512 * Take a completed network device structure and add it to the kernel
5513 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5514 * chain. 0 is returned on success. A negative errno code is returned
5515 * on a failure to set up the device, or if the name is a duplicate.
5517 * Callers must hold the rtnl semaphore. You may want
5518 * register_netdev() instead of this.
5521 * The locking appears insufficient to guarantee two parallel registers
5522 * will not get the same name.
5525 int register_netdevice(struct net_device *dev)
5528 struct net *net = dev_net(dev);
5530 BUG_ON(dev_boot_phase);
5535 /* When net_device's are persistent, this will be fatal. */
5536 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5539 spin_lock_init(&dev->addr_list_lock);
5540 netdev_set_addr_lockdep_class(dev);
5544 ret = dev_get_valid_name(dev, dev->name);
5548 /* Init, if this function is available */
5549 if (dev->netdev_ops->ndo_init) {
5550 ret = dev->netdev_ops->ndo_init(dev);
5558 dev->ifindex = dev_new_index(net);
5559 if (dev->iflink == -1)
5560 dev->iflink = dev->ifindex;
5562 /* Transfer changeable features to wanted_features and enable
5563 * software offloads (GSO and GRO).
5565 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5566 dev->features |= NETIF_F_SOFT_FEATURES;
5567 dev->wanted_features = dev->features & dev->hw_features;
5569 /* Turn on no cache copy if HW is doing checksum */
5570 if (!(dev->flags & IFF_LOOPBACK)) {
5571 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5572 if (dev->features & NETIF_F_ALL_CSUM) {
5573 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5574 dev->features |= NETIF_F_NOCACHE_COPY;
5578 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5580 dev->vlan_features |= NETIF_F_HIGHDMA;
5582 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5583 ret = notifier_to_errno(ret);
5587 ret = netdev_register_kobject(dev);
5590 dev->reg_state = NETREG_REGISTERED;
5592 __netdev_update_features(dev);
5595 * Default initial state at registry is that the
5596 * device is present.
5599 set_bit(__LINK_STATE_PRESENT, &dev->state);
5601 dev_init_scheduler(dev);
5603 list_netdevice(dev);
5605 /* Notify protocols, that a new device appeared. */
5606 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5607 ret = notifier_to_errno(ret);
5609 rollback_registered(dev);
5610 dev->reg_state = NETREG_UNREGISTERED;
5613 * Prevent userspace races by waiting until the network
5614 * device is fully setup before sending notifications.
5616 if (!dev->rtnl_link_ops ||
5617 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5618 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5624 if (dev->netdev_ops->ndo_uninit)
5625 dev->netdev_ops->ndo_uninit(dev);
5628 EXPORT_SYMBOL(register_netdevice);
5631 * init_dummy_netdev - init a dummy network device for NAPI
5632 * @dev: device to init
5634 * This takes a network device structure and initialize the minimum
5635 * amount of fields so it can be used to schedule NAPI polls without
5636 * registering a full blown interface. This is to be used by drivers
5637 * that need to tie several hardware interfaces to a single NAPI
5638 * poll scheduler due to HW limitations.
5640 int init_dummy_netdev(struct net_device *dev)
5642 /* Clear everything. Note we don't initialize spinlocks
5643 * are they aren't supposed to be taken by any of the
5644 * NAPI code and this dummy netdev is supposed to be
5645 * only ever used for NAPI polls
5647 memset(dev, 0, sizeof(struct net_device));
5649 /* make sure we BUG if trying to hit standard
5650 * register/unregister code path
5652 dev->reg_state = NETREG_DUMMY;
5654 /* NAPI wants this */
5655 INIT_LIST_HEAD(&dev->napi_list);
5657 /* a dummy interface is started by default */
5658 set_bit(__LINK_STATE_PRESENT, &dev->state);
5659 set_bit(__LINK_STATE_START, &dev->state);
5661 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5662 * because users of this 'device' dont need to change
5668 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5672 * register_netdev - register a network device
5673 * @dev: device to register
5675 * Take a completed network device structure and add it to the kernel
5676 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5677 * chain. 0 is returned on success. A negative errno code is returned
5678 * on a failure to set up the device, or if the name is a duplicate.
5680 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5681 * and expands the device name if you passed a format string to
5684 int register_netdev(struct net_device *dev)
5689 err = register_netdevice(dev);
5693 EXPORT_SYMBOL(register_netdev);
5695 int netdev_refcnt_read(const struct net_device *dev)
5699 for_each_possible_cpu(i)
5700 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5703 EXPORT_SYMBOL(netdev_refcnt_read);
5706 * netdev_wait_allrefs - wait until all references are gone.
5708 * This is called when unregistering network devices.
5710 * Any protocol or device that holds a reference should register
5711 * for netdevice notification, and cleanup and put back the
5712 * reference if they receive an UNREGISTER event.
5713 * We can get stuck here if buggy protocols don't correctly
5716 static void netdev_wait_allrefs(struct net_device *dev)
5718 unsigned long rebroadcast_time, warning_time;
5721 linkwatch_forget_dev(dev);
5723 rebroadcast_time = warning_time = jiffies;
5724 refcnt = netdev_refcnt_read(dev);
5726 while (refcnt != 0) {
5727 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5730 /* Rebroadcast unregister notification */
5731 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5732 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5733 * should have already handle it the first time */
5735 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5737 /* We must not have linkwatch events
5738 * pending on unregister. If this
5739 * happens, we simply run the queue
5740 * unscheduled, resulting in a noop
5743 linkwatch_run_queue();
5748 rebroadcast_time = jiffies;
5753 refcnt = netdev_refcnt_read(dev);
5755 if (time_after(jiffies, warning_time + 10 * HZ)) {
5756 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5758 warning_time = jiffies;
5767 * register_netdevice(x1);
5768 * register_netdevice(x2);
5770 * unregister_netdevice(y1);
5771 * unregister_netdevice(y2);
5777 * We are invoked by rtnl_unlock().
5778 * This allows us to deal with problems:
5779 * 1) We can delete sysfs objects which invoke hotplug
5780 * without deadlocking with linkwatch via keventd.
5781 * 2) Since we run with the RTNL semaphore not held, we can sleep
5782 * safely in order to wait for the netdev refcnt to drop to zero.
5784 * We must not return until all unregister events added during
5785 * the interval the lock was held have been completed.
5787 void netdev_run_todo(void)
5789 struct list_head list;
5791 /* Snapshot list, allow later requests */
5792 list_replace_init(&net_todo_list, &list);
5796 /* Wait for rcu callbacks to finish before attempting to drain
5797 * the device list. This usually avoids a 250ms wait.
5799 if (!list_empty(&list))
5802 while (!list_empty(&list)) {
5803 struct net_device *dev
5804 = list_first_entry(&list, struct net_device, todo_list);
5805 list_del(&dev->todo_list);
5807 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5808 pr_err("network todo '%s' but state %d\n",
5809 dev->name, dev->reg_state);
5814 dev->reg_state = NETREG_UNREGISTERED;
5816 on_each_cpu(flush_backlog, dev, 1);
5818 netdev_wait_allrefs(dev);
5821 BUG_ON(netdev_refcnt_read(dev));
5822 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5823 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5824 WARN_ON(dev->dn_ptr);
5826 if (dev->destructor)
5827 dev->destructor(dev);
5829 /* Free network device */
5830 kobject_put(&dev->dev.kobj);
5834 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5835 * fields in the same order, with only the type differing.
5837 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5838 const struct net_device_stats *netdev_stats)
5840 #if BITS_PER_LONG == 64
5841 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5842 memcpy(stats64, netdev_stats, sizeof(*stats64));
5844 size_t i, n = sizeof(*stats64) / sizeof(u64);
5845 const unsigned long *src = (const unsigned long *)netdev_stats;
5846 u64 *dst = (u64 *)stats64;
5848 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5849 sizeof(*stats64) / sizeof(u64));
5850 for (i = 0; i < n; i++)
5854 EXPORT_SYMBOL(netdev_stats_to_stats64);
5857 * dev_get_stats - get network device statistics
5858 * @dev: device to get statistics from
5859 * @storage: place to store stats
5861 * Get network statistics from device. Return @storage.
5862 * The device driver may provide its own method by setting
5863 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5864 * otherwise the internal statistics structure is used.
5866 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5867 struct rtnl_link_stats64 *storage)
5869 const struct net_device_ops *ops = dev->netdev_ops;
5871 if (ops->ndo_get_stats64) {
5872 memset(storage, 0, sizeof(*storage));
5873 ops->ndo_get_stats64(dev, storage);
5874 } else if (ops->ndo_get_stats) {
5875 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5877 netdev_stats_to_stats64(storage, &dev->stats);
5879 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5882 EXPORT_SYMBOL(dev_get_stats);
5884 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5886 struct netdev_queue *queue = dev_ingress_queue(dev);
5888 #ifdef CONFIG_NET_CLS_ACT
5891 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5894 netdev_init_one_queue(dev, queue, NULL);
5895 queue->qdisc = &noop_qdisc;
5896 queue->qdisc_sleeping = &noop_qdisc;
5897 rcu_assign_pointer(dev->ingress_queue, queue);
5903 * alloc_netdev_mqs - allocate network device
5904 * @sizeof_priv: size of private data to allocate space for
5905 * @name: device name format string
5906 * @setup: callback to initialize device
5907 * @txqs: the number of TX subqueues to allocate
5908 * @rxqs: the number of RX subqueues to allocate
5910 * Allocates a struct net_device with private data area for driver use
5911 * and performs basic initialization. Also allocates subquue structs
5912 * for each queue on the device.
5914 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5915 void (*setup)(struct net_device *),
5916 unsigned int txqs, unsigned int rxqs)
5918 struct net_device *dev;
5920 struct net_device *p;
5922 BUG_ON(strlen(name) >= sizeof(dev->name));
5925 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5931 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5936 alloc_size = sizeof(struct net_device);
5938 /* ensure 32-byte alignment of private area */
5939 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5940 alloc_size += sizeof_priv;
5942 /* ensure 32-byte alignment of whole construct */
5943 alloc_size += NETDEV_ALIGN - 1;
5945 p = kzalloc(alloc_size, GFP_KERNEL);
5947 pr_err("alloc_netdev: Unable to allocate device\n");
5951 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5952 dev->padded = (char *)dev - (char *)p;
5954 dev->pcpu_refcnt = alloc_percpu(int);
5955 if (!dev->pcpu_refcnt)
5958 if (dev_addr_init(dev))
5964 dev_net_set(dev, &init_net);
5966 dev->gso_max_size = GSO_MAX_SIZE;
5968 INIT_LIST_HEAD(&dev->napi_list);
5969 INIT_LIST_HEAD(&dev->unreg_list);
5970 INIT_LIST_HEAD(&dev->link_watch_list);
5971 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5974 dev->num_tx_queues = txqs;
5975 dev->real_num_tx_queues = txqs;
5976 if (netif_alloc_netdev_queues(dev))
5980 dev->num_rx_queues = rxqs;
5981 dev->real_num_rx_queues = rxqs;
5982 if (netif_alloc_rx_queues(dev))
5986 strcpy(dev->name, name);
5987 dev->group = INIT_NETDEV_GROUP;
5995 free_percpu(dev->pcpu_refcnt);
6005 EXPORT_SYMBOL(alloc_netdev_mqs);
6008 * free_netdev - free network device
6011 * This function does the last stage of destroying an allocated device
6012 * interface. The reference to the device object is released.
6013 * If this is the last reference then it will be freed.
6015 void free_netdev(struct net_device *dev)
6017 struct napi_struct *p, *n;
6019 release_net(dev_net(dev));
6026 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6028 /* Flush device addresses */
6029 dev_addr_flush(dev);
6031 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6034 free_percpu(dev->pcpu_refcnt);
6035 dev->pcpu_refcnt = NULL;
6037 /* Compatibility with error handling in drivers */
6038 if (dev->reg_state == NETREG_UNINITIALIZED) {
6039 kfree((char *)dev - dev->padded);
6043 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6044 dev->reg_state = NETREG_RELEASED;
6046 /* will free via device release */
6047 put_device(&dev->dev);
6049 EXPORT_SYMBOL(free_netdev);
6052 * synchronize_net - Synchronize with packet receive processing
6054 * Wait for packets currently being received to be done.
6055 * Does not block later packets from starting.
6057 void synchronize_net(void)
6060 if (rtnl_is_locked())
6061 synchronize_rcu_expedited();
6065 EXPORT_SYMBOL(synchronize_net);
6068 * unregister_netdevice_queue - remove device from the kernel
6072 * This function shuts down a device interface and removes it
6073 * from the kernel tables.
6074 * If head not NULL, device is queued to be unregistered later.
6076 * Callers must hold the rtnl semaphore. You may want
6077 * unregister_netdev() instead of this.
6080 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6085 list_move_tail(&dev->unreg_list, head);
6087 rollback_registered(dev);
6088 /* Finish processing unregister after unlock */
6092 EXPORT_SYMBOL(unregister_netdevice_queue);
6095 * unregister_netdevice_many - unregister many devices
6096 * @head: list of devices
6098 void unregister_netdevice_many(struct list_head *head)
6100 struct net_device *dev;
6102 if (!list_empty(head)) {
6103 rollback_registered_many(head);
6104 list_for_each_entry(dev, head, unreg_list)
6108 EXPORT_SYMBOL(unregister_netdevice_many);
6111 * unregister_netdev - remove device from the kernel
6114 * This function shuts down a device interface and removes it
6115 * from the kernel tables.
6117 * This is just a wrapper for unregister_netdevice that takes
6118 * the rtnl semaphore. In general you want to use this and not
6119 * unregister_netdevice.
6121 void unregister_netdev(struct net_device *dev)
6124 unregister_netdevice(dev);
6127 EXPORT_SYMBOL(unregister_netdev);
6130 * dev_change_net_namespace - move device to different nethost namespace
6132 * @net: network namespace
6133 * @pat: If not NULL name pattern to try if the current device name
6134 * is already taken in the destination network namespace.
6136 * This function shuts down a device interface and moves it
6137 * to a new network namespace. On success 0 is returned, on
6138 * a failure a netagive errno code is returned.
6140 * Callers must hold the rtnl semaphore.
6143 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6149 /* Don't allow namespace local devices to be moved. */
6151 if (dev->features & NETIF_F_NETNS_LOCAL)
6154 /* Ensure the device has been registrered */
6156 if (dev->reg_state != NETREG_REGISTERED)
6159 /* Get out if there is nothing todo */
6161 if (net_eq(dev_net(dev), net))
6164 /* Pick the destination device name, and ensure
6165 * we can use it in the destination network namespace.
6168 if (__dev_get_by_name(net, dev->name)) {
6169 /* We get here if we can't use the current device name */
6172 if (dev_get_valid_name(dev, pat) < 0)
6177 * And now a mini version of register_netdevice unregister_netdevice.
6180 /* If device is running close it first. */
6183 /* And unlink it from device chain */
6185 unlist_netdevice(dev);
6189 /* Shutdown queueing discipline. */
6192 /* Notify protocols, that we are about to destroy
6193 this device. They should clean all the things.
6195 Note that dev->reg_state stays at NETREG_REGISTERED.
6196 This is wanted because this way 8021q and macvlan know
6197 the device is just moving and can keep their slaves up.
6199 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6200 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6201 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6204 * Flush the unicast and multicast chains
6209 /* Actually switch the network namespace */
6210 dev_net_set(dev, net);
6212 /* If there is an ifindex conflict assign a new one */
6213 if (__dev_get_by_index(net, dev->ifindex)) {
6214 int iflink = (dev->iflink == dev->ifindex);
6215 dev->ifindex = dev_new_index(net);
6217 dev->iflink = dev->ifindex;
6220 /* Fixup kobjects */
6221 err = device_rename(&dev->dev, dev->name);
6224 /* Add the device back in the hashes */
6225 list_netdevice(dev);
6227 /* Notify protocols, that a new device appeared. */
6228 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6231 * Prevent userspace races by waiting until the network
6232 * device is fully setup before sending notifications.
6234 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6241 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6243 static int dev_cpu_callback(struct notifier_block *nfb,
6244 unsigned long action,
6247 struct sk_buff **list_skb;
6248 struct sk_buff *skb;
6249 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6250 struct softnet_data *sd, *oldsd;
6252 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6255 local_irq_disable();
6256 cpu = smp_processor_id();
6257 sd = &per_cpu(softnet_data, cpu);
6258 oldsd = &per_cpu(softnet_data, oldcpu);
6260 /* Find end of our completion_queue. */
6261 list_skb = &sd->completion_queue;
6263 list_skb = &(*list_skb)->next;
6264 /* Append completion queue from offline CPU. */
6265 *list_skb = oldsd->completion_queue;
6266 oldsd->completion_queue = NULL;
6268 /* Append output queue from offline CPU. */
6269 if (oldsd->output_queue) {
6270 *sd->output_queue_tailp = oldsd->output_queue;
6271 sd->output_queue_tailp = oldsd->output_queue_tailp;
6272 oldsd->output_queue = NULL;
6273 oldsd->output_queue_tailp = &oldsd->output_queue;
6275 /* Append NAPI poll list from offline CPU. */
6276 if (!list_empty(&oldsd->poll_list)) {
6277 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6278 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6281 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6284 /* Process offline CPU's input_pkt_queue */
6285 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6287 input_queue_head_incr(oldsd);
6289 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6291 input_queue_head_incr(oldsd);
6299 * netdev_increment_features - increment feature set by one
6300 * @all: current feature set
6301 * @one: new feature set
6302 * @mask: mask feature set
6304 * Computes a new feature set after adding a device with feature set
6305 * @one to the master device with current feature set @all. Will not
6306 * enable anything that is off in @mask. Returns the new feature set.
6308 netdev_features_t netdev_increment_features(netdev_features_t all,
6309 netdev_features_t one, netdev_features_t mask)
6311 if (mask & NETIF_F_GEN_CSUM)
6312 mask |= NETIF_F_ALL_CSUM;
6313 mask |= NETIF_F_VLAN_CHALLENGED;
6315 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6316 all &= one | ~NETIF_F_ALL_FOR_ALL;
6318 /* If one device supports hw checksumming, set for all. */
6319 if (all & NETIF_F_GEN_CSUM)
6320 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6324 EXPORT_SYMBOL(netdev_increment_features);
6326 static struct hlist_head *netdev_create_hash(void)
6329 struct hlist_head *hash;
6331 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6333 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6334 INIT_HLIST_HEAD(&hash[i]);
6339 /* Initialize per network namespace state */
6340 static int __net_init netdev_init(struct net *net)
6342 INIT_LIST_HEAD(&net->dev_base_head);
6344 net->dev_name_head = netdev_create_hash();
6345 if (net->dev_name_head == NULL)
6348 net->dev_index_head = netdev_create_hash();
6349 if (net->dev_index_head == NULL)
6355 kfree(net->dev_name_head);
6361 * netdev_drivername - network driver for the device
6362 * @dev: network device
6364 * Determine network driver for device.
6366 const char *netdev_drivername(const struct net_device *dev)
6368 const struct device_driver *driver;
6369 const struct device *parent;
6370 const char *empty = "";
6372 parent = dev->dev.parent;
6376 driver = parent->driver;
6377 if (driver && driver->name)
6378 return driver->name;
6382 int __netdev_printk(const char *level, const struct net_device *dev,
6383 struct va_format *vaf)
6387 if (dev && dev->dev.parent)
6388 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6389 netdev_name(dev), vaf);
6391 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6393 r = printk("%s(NULL net_device): %pV", level, vaf);
6397 EXPORT_SYMBOL(__netdev_printk);
6399 int netdev_printk(const char *level, const struct net_device *dev,
6400 const char *format, ...)
6402 struct va_format vaf;
6406 va_start(args, format);
6411 r = __netdev_printk(level, dev, &vaf);
6416 EXPORT_SYMBOL(netdev_printk);
6418 #define define_netdev_printk_level(func, level) \
6419 int func(const struct net_device *dev, const char *fmt, ...) \
6422 struct va_format vaf; \
6425 va_start(args, fmt); \
6430 r = __netdev_printk(level, dev, &vaf); \
6435 EXPORT_SYMBOL(func);
6437 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6438 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6439 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6440 define_netdev_printk_level(netdev_err, KERN_ERR);
6441 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6442 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6443 define_netdev_printk_level(netdev_info, KERN_INFO);
6445 static void __net_exit netdev_exit(struct net *net)
6447 kfree(net->dev_name_head);
6448 kfree(net->dev_index_head);
6451 static struct pernet_operations __net_initdata netdev_net_ops = {
6452 .init = netdev_init,
6453 .exit = netdev_exit,
6456 static void __net_exit default_device_exit(struct net *net)
6458 struct net_device *dev, *aux;
6460 * Push all migratable network devices back to the
6461 * initial network namespace
6464 for_each_netdev_safe(net, dev, aux) {
6466 char fb_name[IFNAMSIZ];
6468 /* Ignore unmoveable devices (i.e. loopback) */
6469 if (dev->features & NETIF_F_NETNS_LOCAL)
6472 /* Leave virtual devices for the generic cleanup */
6473 if (dev->rtnl_link_ops)
6476 /* Push remaining network devices to init_net */
6477 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6478 err = dev_change_net_namespace(dev, &init_net, fb_name);
6480 pr_emerg("%s: failed to move %s to init_net: %d\n",
6481 __func__, dev->name, err);
6488 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6490 /* At exit all network devices most be removed from a network
6491 * namespace. Do this in the reverse order of registration.
6492 * Do this across as many network namespaces as possible to
6493 * improve batching efficiency.
6495 struct net_device *dev;
6497 LIST_HEAD(dev_kill_list);
6500 list_for_each_entry(net, net_list, exit_list) {
6501 for_each_netdev_reverse(net, dev) {
6502 if (dev->rtnl_link_ops)
6503 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6505 unregister_netdevice_queue(dev, &dev_kill_list);
6508 unregister_netdevice_many(&dev_kill_list);
6509 list_del(&dev_kill_list);
6513 static struct pernet_operations __net_initdata default_device_ops = {
6514 .exit = default_device_exit,
6515 .exit_batch = default_device_exit_batch,
6519 * Initialize the DEV module. At boot time this walks the device list and
6520 * unhooks any devices that fail to initialise (normally hardware not
6521 * present) and leaves us with a valid list of present and active devices.
6526 * This is called single threaded during boot, so no need
6527 * to take the rtnl semaphore.
6529 static int __init net_dev_init(void)
6531 int i, rc = -ENOMEM;
6533 BUG_ON(!dev_boot_phase);
6535 if (dev_proc_init())
6538 if (netdev_kobject_init())
6541 INIT_LIST_HEAD(&ptype_all);
6542 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6543 INIT_LIST_HEAD(&ptype_base[i]);
6545 if (register_pernet_subsys(&netdev_net_ops))
6549 * Initialise the packet receive queues.
6552 for_each_possible_cpu(i) {
6553 struct softnet_data *sd = &per_cpu(softnet_data, i);
6555 memset(sd, 0, sizeof(*sd));
6556 skb_queue_head_init(&sd->input_pkt_queue);
6557 skb_queue_head_init(&sd->process_queue);
6558 sd->completion_queue = NULL;
6559 INIT_LIST_HEAD(&sd->poll_list);
6560 sd->output_queue = NULL;
6561 sd->output_queue_tailp = &sd->output_queue;
6563 sd->csd.func = rps_trigger_softirq;
6569 sd->backlog.poll = process_backlog;
6570 sd->backlog.weight = weight_p;
6571 sd->backlog.gro_list = NULL;
6572 sd->backlog.gro_count = 0;
6577 /* The loopback device is special if any other network devices
6578 * is present in a network namespace the loopback device must
6579 * be present. Since we now dynamically allocate and free the
6580 * loopback device ensure this invariant is maintained by
6581 * keeping the loopback device as the first device on the
6582 * list of network devices. Ensuring the loopback devices
6583 * is the first device that appears and the last network device
6586 if (register_pernet_device(&loopback_net_ops))
6589 if (register_pernet_device(&default_device_ops))
6592 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6593 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6595 hotcpu_notifier(dev_cpu_callback, 0);
6603 subsys_initcall(net_dev_init);
6605 static int __init initialize_hashrnd(void)
6607 get_random_bytes(&hashrnd, sizeof(hashrnd));
6611 late_initcall_sync(initialize_hashrnd);