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 DEFINE_SPINLOCK(offload_lock);
180 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
181 static struct list_head ptype_all __read_mostly; /* Taps */
182 static struct list_head offload_base __read_mostly;
185 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
188 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
190 * Writers must hold the rtnl semaphore while they loop through the
191 * dev_base_head list, and hold dev_base_lock for writing when they do the
192 * actual updates. This allows pure readers to access the list even
193 * while a writer is preparing to update it.
195 * To put it another way, dev_base_lock is held for writing only to
196 * protect against pure readers; the rtnl semaphore provides the
197 * protection against other writers.
199 * See, for example usages, register_netdevice() and
200 * unregister_netdevice(), which must be called with the rtnl
203 DEFINE_RWLOCK(dev_base_lock);
204 EXPORT_SYMBOL(dev_base_lock);
206 seqcount_t devnet_rename_seq;
208 static inline void dev_base_seq_inc(struct net *net)
210 while (++net->dev_base_seq == 0);
213 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
215 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
217 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
220 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
222 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
225 static inline void rps_lock(struct softnet_data *sd)
228 spin_lock(&sd->input_pkt_queue.lock);
232 static inline void rps_unlock(struct softnet_data *sd)
235 spin_unlock(&sd->input_pkt_queue.lock);
239 /* Device list insertion */
240 static int list_netdevice(struct net_device *dev)
242 struct net *net = dev_net(dev);
246 write_lock_bh(&dev_base_lock);
247 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
248 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
249 hlist_add_head_rcu(&dev->index_hlist,
250 dev_index_hash(net, dev->ifindex));
251 write_unlock_bh(&dev_base_lock);
253 dev_base_seq_inc(net);
258 /* Device list removal
259 * caller must respect a RCU grace period before freeing/reusing dev
261 static void unlist_netdevice(struct net_device *dev)
265 /* Unlink dev from the device chain */
266 write_lock_bh(&dev_base_lock);
267 list_del_rcu(&dev->dev_list);
268 hlist_del_rcu(&dev->name_hlist);
269 hlist_del_rcu(&dev->index_hlist);
270 write_unlock_bh(&dev_base_lock);
272 dev_base_seq_inc(dev_net(dev));
279 static RAW_NOTIFIER_HEAD(netdev_chain);
282 * Device drivers call our routines to queue packets here. We empty the
283 * queue in the local softnet handler.
286 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
287 EXPORT_PER_CPU_SYMBOL(softnet_data);
289 #ifdef CONFIG_LOCKDEP
291 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
292 * according to dev->type
294 static const unsigned short netdev_lock_type[] =
295 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
296 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
297 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
298 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
299 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
300 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
301 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
302 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
303 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
304 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
305 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
306 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
307 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
308 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
309 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
311 static const char *const netdev_lock_name[] =
312 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
313 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
314 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
315 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
316 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
317 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
318 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
319 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
320 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
321 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
322 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
323 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
324 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
325 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
326 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
328 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
329 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
331 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
335 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
336 if (netdev_lock_type[i] == dev_type)
338 /* the last key is used by default */
339 return ARRAY_SIZE(netdev_lock_type) - 1;
342 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
343 unsigned short dev_type)
347 i = netdev_lock_pos(dev_type);
348 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
349 netdev_lock_name[i]);
352 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
356 i = netdev_lock_pos(dev->type);
357 lockdep_set_class_and_name(&dev->addr_list_lock,
358 &netdev_addr_lock_key[i],
359 netdev_lock_name[i]);
362 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
363 unsigned short dev_type)
366 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
371 /*******************************************************************************
373 Protocol management and registration routines
375 *******************************************************************************/
378 * Add a protocol ID to the list. Now that the input handler is
379 * smarter we can dispense with all the messy stuff that used to be
382 * BEWARE!!! Protocol handlers, mangling input packets,
383 * MUST BE last in hash buckets and checking protocol handlers
384 * MUST start from promiscuous ptype_all chain in net_bh.
385 * It is true now, do not change it.
386 * Explanation follows: if protocol handler, mangling packet, will
387 * be the first on list, it is not able to sense, that packet
388 * is cloned and should be copied-on-write, so that it will
389 * change it and subsequent readers will get broken packet.
393 static inline struct list_head *ptype_head(const struct packet_type *pt)
395 if (pt->type == htons(ETH_P_ALL))
398 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
402 * dev_add_pack - add packet handler
403 * @pt: packet type declaration
405 * Add a protocol handler to the networking stack. The passed &packet_type
406 * is linked into kernel lists and may not be freed until it has been
407 * removed from the kernel lists.
409 * This call does not sleep therefore it can not
410 * guarantee all CPU's that are in middle of receiving packets
411 * will see the new packet type (until the next received packet).
414 void dev_add_pack(struct packet_type *pt)
416 struct list_head *head = ptype_head(pt);
418 spin_lock(&ptype_lock);
419 list_add_rcu(&pt->list, head);
420 spin_unlock(&ptype_lock);
422 EXPORT_SYMBOL(dev_add_pack);
425 * __dev_remove_pack - remove packet handler
426 * @pt: packet type declaration
428 * Remove a protocol handler that was previously added to the kernel
429 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
430 * from the kernel lists and can be freed or reused once this function
433 * The packet type might still be in use by receivers
434 * and must not be freed until after all the CPU's have gone
435 * through a quiescent state.
437 void __dev_remove_pack(struct packet_type *pt)
439 struct list_head *head = ptype_head(pt);
440 struct packet_type *pt1;
442 spin_lock(&ptype_lock);
444 list_for_each_entry(pt1, head, list) {
446 list_del_rcu(&pt->list);
451 pr_warn("dev_remove_pack: %p not found\n", pt);
453 spin_unlock(&ptype_lock);
455 EXPORT_SYMBOL(__dev_remove_pack);
458 * dev_remove_pack - remove packet handler
459 * @pt: packet type declaration
461 * Remove a protocol handler that was previously added to the kernel
462 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
463 * from the kernel lists and can be freed or reused once this function
466 * This call sleeps to guarantee that no CPU is looking at the packet
469 void dev_remove_pack(struct packet_type *pt)
471 __dev_remove_pack(pt);
475 EXPORT_SYMBOL(dev_remove_pack);
479 * dev_add_offload - register offload handlers
480 * @po: protocol offload declaration
482 * Add protocol offload handlers to the networking stack. The passed
483 * &proto_offload is linked into kernel lists and may not be freed until
484 * it has been removed from the kernel lists.
486 * This call does not sleep therefore it can not
487 * guarantee all CPU's that are in middle of receiving packets
488 * will see the new offload handlers (until the next received packet).
490 void dev_add_offload(struct packet_offload *po)
492 struct list_head *head = &offload_base;
494 spin_lock(&offload_lock);
495 list_add_rcu(&po->list, head);
496 spin_unlock(&offload_lock);
498 EXPORT_SYMBOL(dev_add_offload);
501 * __dev_remove_offload - remove offload handler
502 * @po: packet offload declaration
504 * Remove a protocol offload handler that was previously added to the
505 * kernel offload handlers by dev_add_offload(). The passed &offload_type
506 * is removed from the kernel lists and can be freed or reused once this
509 * The packet type might still be in use by receivers
510 * and must not be freed until after all the CPU's have gone
511 * through a quiescent state.
513 void __dev_remove_offload(struct packet_offload *po)
515 struct list_head *head = &offload_base;
516 struct packet_offload *po1;
518 spin_lock(&offload_lock);
520 list_for_each_entry(po1, head, list) {
522 list_del_rcu(&po->list);
527 pr_warn("dev_remove_offload: %p not found\n", po);
529 spin_unlock(&offload_lock);
531 EXPORT_SYMBOL(__dev_remove_offload);
534 * dev_remove_offload - remove packet offload handler
535 * @po: packet offload declaration
537 * Remove a packet offload handler that was previously added to the kernel
538 * offload handlers by dev_add_offload(). The passed &offload_type is
539 * removed from the kernel lists and can be freed or reused once this
542 * This call sleeps to guarantee that no CPU is looking at the packet
545 void dev_remove_offload(struct packet_offload *po)
547 __dev_remove_offload(po);
551 EXPORT_SYMBOL(dev_remove_offload);
553 /******************************************************************************
555 Device Boot-time Settings Routines
557 *******************************************************************************/
559 /* Boot time configuration table */
560 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
563 * netdev_boot_setup_add - add new setup entry
564 * @name: name of the device
565 * @map: configured settings for the device
567 * Adds new setup entry to the dev_boot_setup list. The function
568 * returns 0 on error and 1 on success. This is a generic routine to
571 static int netdev_boot_setup_add(char *name, struct ifmap *map)
573 struct netdev_boot_setup *s;
577 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
578 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
579 memset(s[i].name, 0, sizeof(s[i].name));
580 strlcpy(s[i].name, name, IFNAMSIZ);
581 memcpy(&s[i].map, map, sizeof(s[i].map));
586 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
590 * netdev_boot_setup_check - check boot time settings
591 * @dev: the netdevice
593 * Check boot time settings for the device.
594 * The found settings are set for the device to be used
595 * later in the device probing.
596 * Returns 0 if no settings found, 1 if they are.
598 int netdev_boot_setup_check(struct net_device *dev)
600 struct netdev_boot_setup *s = dev_boot_setup;
603 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
604 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
605 !strcmp(dev->name, s[i].name)) {
606 dev->irq = s[i].map.irq;
607 dev->base_addr = s[i].map.base_addr;
608 dev->mem_start = s[i].map.mem_start;
609 dev->mem_end = s[i].map.mem_end;
615 EXPORT_SYMBOL(netdev_boot_setup_check);
619 * netdev_boot_base - get address from boot time settings
620 * @prefix: prefix for network device
621 * @unit: id for network device
623 * Check boot time settings for the base address of device.
624 * The found settings are set for the device to be used
625 * later in the device probing.
626 * Returns 0 if no settings found.
628 unsigned long netdev_boot_base(const char *prefix, int unit)
630 const struct netdev_boot_setup *s = dev_boot_setup;
634 sprintf(name, "%s%d", prefix, unit);
637 * If device already registered then return base of 1
638 * to indicate not to probe for this interface
640 if (__dev_get_by_name(&init_net, name))
643 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
644 if (!strcmp(name, s[i].name))
645 return s[i].map.base_addr;
650 * Saves at boot time configured settings for any netdevice.
652 int __init netdev_boot_setup(char *str)
657 str = get_options(str, ARRAY_SIZE(ints), ints);
662 memset(&map, 0, sizeof(map));
666 map.base_addr = ints[2];
668 map.mem_start = ints[3];
670 map.mem_end = ints[4];
672 /* Add new entry to the list */
673 return netdev_boot_setup_add(str, &map);
676 __setup("netdev=", netdev_boot_setup);
678 /*******************************************************************************
680 Device Interface Subroutines
682 *******************************************************************************/
685 * __dev_get_by_name - find a device by its name
686 * @net: the applicable net namespace
687 * @name: name to find
689 * Find an interface by name. Must be called under RTNL semaphore
690 * or @dev_base_lock. If the name is found a pointer to the device
691 * is returned. If the name is not found then %NULL is returned. The
692 * reference counters are not incremented so the caller must be
693 * careful with locks.
696 struct net_device *__dev_get_by_name(struct net *net, const char *name)
698 struct hlist_node *p;
699 struct net_device *dev;
700 struct hlist_head *head = dev_name_hash(net, name);
702 hlist_for_each_entry(dev, p, head, name_hlist)
703 if (!strncmp(dev->name, name, IFNAMSIZ))
708 EXPORT_SYMBOL(__dev_get_by_name);
711 * dev_get_by_name_rcu - find a device by its name
712 * @net: the applicable net namespace
713 * @name: name to find
715 * Find an interface by name.
716 * If the name is found a pointer to the device is returned.
717 * If the name is not found then %NULL is returned.
718 * The reference counters are not incremented so the caller must be
719 * careful with locks. The caller must hold RCU lock.
722 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
724 struct hlist_node *p;
725 struct net_device *dev;
726 struct hlist_head *head = dev_name_hash(net, name);
728 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
729 if (!strncmp(dev->name, name, IFNAMSIZ))
734 EXPORT_SYMBOL(dev_get_by_name_rcu);
737 * dev_get_by_name - find a device by its name
738 * @net: the applicable net namespace
739 * @name: name to find
741 * Find an interface by name. This can be called from any
742 * context and does its own locking. The returned handle has
743 * the usage count incremented and the caller must use dev_put() to
744 * release it when it is no longer needed. %NULL is returned if no
745 * matching device is found.
748 struct net_device *dev_get_by_name(struct net *net, const char *name)
750 struct net_device *dev;
753 dev = dev_get_by_name_rcu(net, name);
759 EXPORT_SYMBOL(dev_get_by_name);
762 * __dev_get_by_index - find a device by its ifindex
763 * @net: the applicable net namespace
764 * @ifindex: index of device
766 * Search for an interface by index. Returns %NULL if the device
767 * is not found or a pointer to the device. The device has not
768 * had its reference counter increased so the caller must be careful
769 * about locking. The caller must hold either the RTNL semaphore
773 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
775 struct hlist_node *p;
776 struct net_device *dev;
777 struct hlist_head *head = dev_index_hash(net, ifindex);
779 hlist_for_each_entry(dev, p, head, index_hlist)
780 if (dev->ifindex == ifindex)
785 EXPORT_SYMBOL(__dev_get_by_index);
788 * dev_get_by_index_rcu - find a device by its ifindex
789 * @net: the applicable net namespace
790 * @ifindex: index of device
792 * Search for an interface by index. Returns %NULL if the device
793 * is not found or a pointer to the device. The device has not
794 * had its reference counter increased so the caller must be careful
795 * about locking. The caller must hold RCU lock.
798 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
800 struct hlist_node *p;
801 struct net_device *dev;
802 struct hlist_head *head = dev_index_hash(net, ifindex);
804 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
805 if (dev->ifindex == ifindex)
810 EXPORT_SYMBOL(dev_get_by_index_rcu);
814 * dev_get_by_index - find a device by its ifindex
815 * @net: the applicable net namespace
816 * @ifindex: index of device
818 * Search for an interface by index. Returns NULL if the device
819 * is not found or a pointer to the device. The device returned has
820 * had a reference added and the pointer is safe until the user calls
821 * dev_put to indicate they have finished with it.
824 struct net_device *dev_get_by_index(struct net *net, int ifindex)
826 struct net_device *dev;
829 dev = dev_get_by_index_rcu(net, ifindex);
835 EXPORT_SYMBOL(dev_get_by_index);
838 * dev_getbyhwaddr_rcu - find a device by its hardware address
839 * @net: the applicable net namespace
840 * @type: media type of device
841 * @ha: hardware address
843 * Search for an interface by MAC address. Returns NULL if the device
844 * is not found or a pointer to the device.
845 * The caller must hold RCU or RTNL.
846 * The returned device has not had its ref count increased
847 * and the caller must therefore be careful about locking
851 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
854 struct net_device *dev;
856 for_each_netdev_rcu(net, dev)
857 if (dev->type == type &&
858 !memcmp(dev->dev_addr, ha, dev->addr_len))
863 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
865 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
867 struct net_device *dev;
870 for_each_netdev(net, dev)
871 if (dev->type == type)
876 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
878 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
880 struct net_device *dev, *ret = NULL;
883 for_each_netdev_rcu(net, dev)
884 if (dev->type == type) {
892 EXPORT_SYMBOL(dev_getfirstbyhwtype);
895 * dev_get_by_flags_rcu - find any device with given flags
896 * @net: the applicable net namespace
897 * @if_flags: IFF_* values
898 * @mask: bitmask of bits in if_flags to check
900 * Search for any interface with the given flags. Returns NULL if a device
901 * is not found or a pointer to the device. Must be called inside
902 * rcu_read_lock(), and result refcount is unchanged.
905 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
908 struct net_device *dev, *ret;
911 for_each_netdev_rcu(net, dev) {
912 if (((dev->flags ^ if_flags) & mask) == 0) {
919 EXPORT_SYMBOL(dev_get_by_flags_rcu);
922 * dev_valid_name - check if name is okay for network device
925 * Network device names need to be valid file names to
926 * to allow sysfs to work. We also disallow any kind of
929 bool dev_valid_name(const char *name)
933 if (strlen(name) >= IFNAMSIZ)
935 if (!strcmp(name, ".") || !strcmp(name, ".."))
939 if (*name == '/' || isspace(*name))
945 EXPORT_SYMBOL(dev_valid_name);
948 * __dev_alloc_name - allocate a name for a device
949 * @net: network namespace to allocate the device name in
950 * @name: name format string
951 * @buf: scratch buffer and result name string
953 * Passed a format string - eg "lt%d" it will try and find a suitable
954 * id. It scans list of devices to build up a free map, then chooses
955 * the first empty slot. The caller must hold the dev_base or rtnl lock
956 * while allocating the name and adding the device in order to avoid
958 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
959 * Returns the number of the unit assigned or a negative errno code.
962 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
966 const int max_netdevices = 8*PAGE_SIZE;
967 unsigned long *inuse;
968 struct net_device *d;
970 p = strnchr(name, IFNAMSIZ-1, '%');
973 * Verify the string as this thing may have come from
974 * the user. There must be either one "%d" and no other "%"
977 if (p[1] != 'd' || strchr(p + 2, '%'))
980 /* Use one page as a bit array of possible slots */
981 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
985 for_each_netdev(net, d) {
986 if (!sscanf(d->name, name, &i))
988 if (i < 0 || i >= max_netdevices)
991 /* avoid cases where sscanf is not exact inverse of printf */
992 snprintf(buf, IFNAMSIZ, name, i);
993 if (!strncmp(buf, d->name, IFNAMSIZ))
997 i = find_first_zero_bit(inuse, max_netdevices);
998 free_page((unsigned long) inuse);
1002 snprintf(buf, IFNAMSIZ, name, i);
1003 if (!__dev_get_by_name(net, buf))
1006 /* It is possible to run out of possible slots
1007 * when the name is long and there isn't enough space left
1008 * for the digits, or if all bits are used.
1014 * dev_alloc_name - allocate a name for a device
1016 * @name: name format string
1018 * Passed a format string - eg "lt%d" it will try and find a suitable
1019 * id. It scans list of devices to build up a free map, then chooses
1020 * the first empty slot. The caller must hold the dev_base or rtnl lock
1021 * while allocating the name and adding the device in order to avoid
1023 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1024 * Returns the number of the unit assigned or a negative errno code.
1027 int dev_alloc_name(struct net_device *dev, const char *name)
1033 BUG_ON(!dev_net(dev));
1035 ret = __dev_alloc_name(net, name, buf);
1037 strlcpy(dev->name, buf, IFNAMSIZ);
1040 EXPORT_SYMBOL(dev_alloc_name);
1042 static int dev_alloc_name_ns(struct net *net,
1043 struct net_device *dev,
1049 ret = __dev_alloc_name(net, name, buf);
1051 strlcpy(dev->name, buf, IFNAMSIZ);
1055 static int dev_get_valid_name(struct net *net,
1056 struct net_device *dev,
1061 if (!dev_valid_name(name))
1064 if (strchr(name, '%'))
1065 return dev_alloc_name_ns(net, dev, name);
1066 else if (__dev_get_by_name(net, name))
1068 else if (dev->name != name)
1069 strlcpy(dev->name, name, IFNAMSIZ);
1075 * dev_change_name - change name of a device
1077 * @newname: name (or format string) must be at least IFNAMSIZ
1079 * Change name of a device, can pass format strings "eth%d".
1082 int dev_change_name(struct net_device *dev, const char *newname)
1084 char oldname[IFNAMSIZ];
1090 BUG_ON(!dev_net(dev));
1093 if (dev->flags & IFF_UP)
1096 write_seqcount_begin(&devnet_rename_seq);
1098 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1099 write_seqcount_end(&devnet_rename_seq);
1103 memcpy(oldname, dev->name, IFNAMSIZ);
1105 err = dev_get_valid_name(net, dev, newname);
1107 write_seqcount_end(&devnet_rename_seq);
1112 ret = device_rename(&dev->dev, dev->name);
1114 memcpy(dev->name, oldname, IFNAMSIZ);
1115 write_seqcount_end(&devnet_rename_seq);
1119 write_seqcount_end(&devnet_rename_seq);
1121 write_lock_bh(&dev_base_lock);
1122 hlist_del_rcu(&dev->name_hlist);
1123 write_unlock_bh(&dev_base_lock);
1127 write_lock_bh(&dev_base_lock);
1128 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1129 write_unlock_bh(&dev_base_lock);
1131 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1132 ret = notifier_to_errno(ret);
1135 /* err >= 0 after dev_alloc_name() or stores the first errno */
1138 write_seqcount_begin(&devnet_rename_seq);
1139 memcpy(dev->name, oldname, IFNAMSIZ);
1142 pr_err("%s: name change rollback failed: %d\n",
1151 * dev_set_alias - change ifalias of a device
1153 * @alias: name up to IFALIASZ
1154 * @len: limit of bytes to copy from info
1156 * Set ifalias for a device,
1158 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1164 if (len >= IFALIASZ)
1168 kfree(dev->ifalias);
1169 dev->ifalias = NULL;
1173 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1176 dev->ifalias = new_ifalias;
1178 strlcpy(dev->ifalias, alias, len+1);
1184 * netdev_features_change - device changes features
1185 * @dev: device to cause notification
1187 * Called to indicate a device has changed features.
1189 void netdev_features_change(struct net_device *dev)
1191 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1193 EXPORT_SYMBOL(netdev_features_change);
1196 * netdev_state_change - device changes state
1197 * @dev: device to cause notification
1199 * Called to indicate a device has changed state. This function calls
1200 * the notifier chains for netdev_chain and sends a NEWLINK message
1201 * to the routing socket.
1203 void netdev_state_change(struct net_device *dev)
1205 if (dev->flags & IFF_UP) {
1206 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1207 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1210 EXPORT_SYMBOL(netdev_state_change);
1213 * netdev_notify_peers - notify network peers about existence of @dev
1214 * @dev: network device
1216 * Generate traffic such that interested network peers are aware of
1217 * @dev, such as by generating a gratuitous ARP. This may be used when
1218 * a device wants to inform the rest of the network about some sort of
1219 * reconfiguration such as a failover event or virtual machine
1222 void netdev_notify_peers(struct net_device *dev)
1225 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1228 EXPORT_SYMBOL(netdev_notify_peers);
1231 * dev_load - load a network module
1232 * @net: the applicable net namespace
1233 * @name: name of interface
1235 * If a network interface is not present and the process has suitable
1236 * privileges this function loads the module. If module loading is not
1237 * available in this kernel then it becomes a nop.
1240 void dev_load(struct net *net, const char *name)
1242 struct net_device *dev;
1246 dev = dev_get_by_name_rcu(net, name);
1250 if (no_module && capable(CAP_NET_ADMIN))
1251 no_module = request_module("netdev-%s", name);
1252 if (no_module && capable(CAP_SYS_MODULE)) {
1253 if (!request_module("%s", name))
1254 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1258 EXPORT_SYMBOL(dev_load);
1260 static int __dev_open(struct net_device *dev)
1262 const struct net_device_ops *ops = dev->netdev_ops;
1267 if (!netif_device_present(dev))
1270 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1271 ret = notifier_to_errno(ret);
1275 set_bit(__LINK_STATE_START, &dev->state);
1277 if (ops->ndo_validate_addr)
1278 ret = ops->ndo_validate_addr(dev);
1280 if (!ret && ops->ndo_open)
1281 ret = ops->ndo_open(dev);
1284 clear_bit(__LINK_STATE_START, &dev->state);
1286 dev->flags |= IFF_UP;
1287 net_dmaengine_get();
1288 dev_set_rx_mode(dev);
1290 add_device_randomness(dev->dev_addr, dev->addr_len);
1297 * dev_open - prepare an interface for use.
1298 * @dev: device to open
1300 * Takes a device from down to up state. The device's private open
1301 * function is invoked and then the multicast lists are loaded. Finally
1302 * the device is moved into the up state and a %NETDEV_UP message is
1303 * sent to the netdev notifier chain.
1305 * Calling this function on an active interface is a nop. On a failure
1306 * a negative errno code is returned.
1308 int dev_open(struct net_device *dev)
1312 if (dev->flags & IFF_UP)
1315 ret = __dev_open(dev);
1319 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1320 call_netdevice_notifiers(NETDEV_UP, dev);
1324 EXPORT_SYMBOL(dev_open);
1326 static int __dev_close_many(struct list_head *head)
1328 struct net_device *dev;
1333 list_for_each_entry(dev, head, unreg_list) {
1334 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1336 clear_bit(__LINK_STATE_START, &dev->state);
1338 /* Synchronize to scheduled poll. We cannot touch poll list, it
1339 * can be even on different cpu. So just clear netif_running().
1341 * dev->stop() will invoke napi_disable() on all of it's
1342 * napi_struct instances on this device.
1344 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1347 dev_deactivate_many(head);
1349 list_for_each_entry(dev, head, unreg_list) {
1350 const struct net_device_ops *ops = dev->netdev_ops;
1353 * Call the device specific close. This cannot fail.
1354 * Only if device is UP
1356 * We allow it to be called even after a DETACH hot-plug
1362 dev->flags &= ~IFF_UP;
1363 net_dmaengine_put();
1369 static int __dev_close(struct net_device *dev)
1374 list_add(&dev->unreg_list, &single);
1375 retval = __dev_close_many(&single);
1380 static int dev_close_many(struct list_head *head)
1382 struct net_device *dev, *tmp;
1383 LIST_HEAD(tmp_list);
1385 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1386 if (!(dev->flags & IFF_UP))
1387 list_move(&dev->unreg_list, &tmp_list);
1389 __dev_close_many(head);
1391 list_for_each_entry(dev, head, unreg_list) {
1392 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1393 call_netdevice_notifiers(NETDEV_DOWN, dev);
1396 /* rollback_registered_many needs the complete original list */
1397 list_splice(&tmp_list, head);
1402 * dev_close - shutdown an interface.
1403 * @dev: device to shutdown
1405 * This function moves an active device into down state. A
1406 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1407 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1410 int dev_close(struct net_device *dev)
1412 if (dev->flags & IFF_UP) {
1415 list_add(&dev->unreg_list, &single);
1416 dev_close_many(&single);
1421 EXPORT_SYMBOL(dev_close);
1425 * dev_disable_lro - disable Large Receive Offload on a device
1428 * Disable Large Receive Offload (LRO) on a net device. Must be
1429 * called under RTNL. This is needed if received packets may be
1430 * forwarded to another interface.
1432 void dev_disable_lro(struct net_device *dev)
1435 * If we're trying to disable lro on a vlan device
1436 * use the underlying physical device instead
1438 if (is_vlan_dev(dev))
1439 dev = vlan_dev_real_dev(dev);
1441 dev->wanted_features &= ~NETIF_F_LRO;
1442 netdev_update_features(dev);
1444 if (unlikely(dev->features & NETIF_F_LRO))
1445 netdev_WARN(dev, "failed to disable LRO!\n");
1447 EXPORT_SYMBOL(dev_disable_lro);
1450 static int dev_boot_phase = 1;
1453 * register_netdevice_notifier - register a network notifier block
1456 * Register a notifier to be called when network device events occur.
1457 * The notifier passed is linked into the kernel structures and must
1458 * not be reused until it has been unregistered. A negative errno code
1459 * is returned on a failure.
1461 * When registered all registration and up events are replayed
1462 * to the new notifier to allow device to have a race free
1463 * view of the network device list.
1466 int register_netdevice_notifier(struct notifier_block *nb)
1468 struct net_device *dev;
1469 struct net_device *last;
1474 err = raw_notifier_chain_register(&netdev_chain, nb);
1480 for_each_netdev(net, dev) {
1481 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1482 err = notifier_to_errno(err);
1486 if (!(dev->flags & IFF_UP))
1489 nb->notifier_call(nb, NETDEV_UP, dev);
1500 for_each_netdev(net, dev) {
1504 if (dev->flags & IFF_UP) {
1505 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1506 nb->notifier_call(nb, NETDEV_DOWN, dev);
1508 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1513 raw_notifier_chain_unregister(&netdev_chain, nb);
1516 EXPORT_SYMBOL(register_netdevice_notifier);
1519 * unregister_netdevice_notifier - unregister a network notifier block
1522 * Unregister a notifier previously registered by
1523 * register_netdevice_notifier(). The notifier is unlinked into the
1524 * kernel structures and may then be reused. A negative errno code
1525 * is returned on a failure.
1527 * After unregistering unregister and down device events are synthesized
1528 * for all devices on the device list to the removed notifier to remove
1529 * the need for special case cleanup code.
1532 int unregister_netdevice_notifier(struct notifier_block *nb)
1534 struct net_device *dev;
1539 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1544 for_each_netdev(net, dev) {
1545 if (dev->flags & IFF_UP) {
1546 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1547 nb->notifier_call(nb, NETDEV_DOWN, dev);
1549 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1556 EXPORT_SYMBOL(unregister_netdevice_notifier);
1559 * call_netdevice_notifiers - call all network notifier blocks
1560 * @val: value passed unmodified to notifier function
1561 * @dev: net_device pointer passed unmodified to notifier function
1563 * Call all network notifier blocks. Parameters and return value
1564 * are as for raw_notifier_call_chain().
1567 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1570 return raw_notifier_call_chain(&netdev_chain, val, dev);
1572 EXPORT_SYMBOL(call_netdevice_notifiers);
1574 static struct static_key netstamp_needed __read_mostly;
1575 #ifdef HAVE_JUMP_LABEL
1576 /* We are not allowed to call static_key_slow_dec() from irq context
1577 * If net_disable_timestamp() is called from irq context, defer the
1578 * static_key_slow_dec() calls.
1580 static atomic_t netstamp_needed_deferred;
1583 void net_enable_timestamp(void)
1585 #ifdef HAVE_JUMP_LABEL
1586 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1590 static_key_slow_dec(&netstamp_needed);
1594 WARN_ON(in_interrupt());
1595 static_key_slow_inc(&netstamp_needed);
1597 EXPORT_SYMBOL(net_enable_timestamp);
1599 void net_disable_timestamp(void)
1601 #ifdef HAVE_JUMP_LABEL
1602 if (in_interrupt()) {
1603 atomic_inc(&netstamp_needed_deferred);
1607 static_key_slow_dec(&netstamp_needed);
1609 EXPORT_SYMBOL(net_disable_timestamp);
1611 static inline void net_timestamp_set(struct sk_buff *skb)
1613 skb->tstamp.tv64 = 0;
1614 if (static_key_false(&netstamp_needed))
1615 __net_timestamp(skb);
1618 #define net_timestamp_check(COND, SKB) \
1619 if (static_key_false(&netstamp_needed)) { \
1620 if ((COND) && !(SKB)->tstamp.tv64) \
1621 __net_timestamp(SKB); \
1624 static int net_hwtstamp_validate(struct ifreq *ifr)
1626 struct hwtstamp_config cfg;
1627 enum hwtstamp_tx_types tx_type;
1628 enum hwtstamp_rx_filters rx_filter;
1629 int tx_type_valid = 0;
1630 int rx_filter_valid = 0;
1632 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1635 if (cfg.flags) /* reserved for future extensions */
1638 tx_type = cfg.tx_type;
1639 rx_filter = cfg.rx_filter;
1642 case HWTSTAMP_TX_OFF:
1643 case HWTSTAMP_TX_ON:
1644 case HWTSTAMP_TX_ONESTEP_SYNC:
1649 switch (rx_filter) {
1650 case HWTSTAMP_FILTER_NONE:
1651 case HWTSTAMP_FILTER_ALL:
1652 case HWTSTAMP_FILTER_SOME:
1653 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1654 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1655 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1656 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1657 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1658 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1659 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1660 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1661 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1662 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1663 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1664 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1665 rx_filter_valid = 1;
1669 if (!tx_type_valid || !rx_filter_valid)
1675 static inline bool is_skb_forwardable(struct net_device *dev,
1676 struct sk_buff *skb)
1680 if (!(dev->flags & IFF_UP))
1683 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1684 if (skb->len <= len)
1687 /* if TSO is enabled, we don't care about the length as the packet
1688 * could be forwarded without being segmented before
1690 if (skb_is_gso(skb))
1697 * dev_forward_skb - loopback an skb to another netif
1699 * @dev: destination network device
1700 * @skb: buffer to forward
1703 * NET_RX_SUCCESS (no congestion)
1704 * NET_RX_DROP (packet was dropped, but freed)
1706 * dev_forward_skb can be used for injecting an skb from the
1707 * start_xmit function of one device into the receive queue
1708 * of another device.
1710 * The receiving device may be in another namespace, so
1711 * we have to clear all information in the skb that could
1712 * impact namespace isolation.
1714 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1716 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1717 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1718 atomic_long_inc(&dev->rx_dropped);
1727 if (unlikely(!is_skb_forwardable(dev, skb))) {
1728 atomic_long_inc(&dev->rx_dropped);
1735 skb->tstamp.tv64 = 0;
1736 skb->pkt_type = PACKET_HOST;
1737 skb->protocol = eth_type_trans(skb, dev);
1741 return netif_rx(skb);
1743 EXPORT_SYMBOL_GPL(dev_forward_skb);
1745 static inline int deliver_skb(struct sk_buff *skb,
1746 struct packet_type *pt_prev,
1747 struct net_device *orig_dev)
1749 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1751 atomic_inc(&skb->users);
1752 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1755 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1757 if (!ptype->af_packet_priv || !skb->sk)
1760 if (ptype->id_match)
1761 return ptype->id_match(ptype, skb->sk);
1762 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1769 * Support routine. Sends outgoing frames to any network
1770 * taps currently in use.
1773 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1775 struct packet_type *ptype;
1776 struct sk_buff *skb2 = NULL;
1777 struct packet_type *pt_prev = NULL;
1780 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1781 /* Never send packets back to the socket
1782 * they originated from - MvS (miquels@drinkel.ow.org)
1784 if ((ptype->dev == dev || !ptype->dev) &&
1785 (!skb_loop_sk(ptype, skb))) {
1787 deliver_skb(skb2, pt_prev, skb->dev);
1792 skb2 = skb_clone(skb, GFP_ATOMIC);
1796 net_timestamp_set(skb2);
1798 /* skb->nh should be correctly
1799 set by sender, so that the second statement is
1800 just protection against buggy protocols.
1802 skb_reset_mac_header(skb2);
1804 if (skb_network_header(skb2) < skb2->data ||
1805 skb2->network_header > skb2->tail) {
1806 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1807 ntohs(skb2->protocol),
1809 skb_reset_network_header(skb2);
1812 skb2->transport_header = skb2->network_header;
1813 skb2->pkt_type = PACKET_OUTGOING;
1818 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1823 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1824 * @dev: Network device
1825 * @txq: number of queues available
1827 * If real_num_tx_queues is changed the tc mappings may no longer be
1828 * valid. To resolve this verify the tc mapping remains valid and if
1829 * not NULL the mapping. With no priorities mapping to this
1830 * offset/count pair it will no longer be used. In the worst case TC0
1831 * is invalid nothing can be done so disable priority mappings. If is
1832 * expected that drivers will fix this mapping if they can before
1833 * calling netif_set_real_num_tx_queues.
1835 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1838 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1840 /* If TC0 is invalidated disable TC mapping */
1841 if (tc->offset + tc->count > txq) {
1842 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1847 /* Invalidated prio to tc mappings set to TC0 */
1848 for (i = 1; i < TC_BITMASK + 1; i++) {
1849 int q = netdev_get_prio_tc_map(dev, i);
1851 tc = &dev->tc_to_txq[q];
1852 if (tc->offset + tc->count > txq) {
1853 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1855 netdev_set_prio_tc_map(dev, i, 0);
1861 static DEFINE_MUTEX(xps_map_mutex);
1862 #define xmap_dereference(P) \
1863 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1865 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1868 struct xps_map *map = NULL;
1872 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1874 for (pos = 0; map && pos < map->len; pos++) {
1875 if (map->queues[pos] == index) {
1877 map->queues[pos] = map->queues[--map->len];
1879 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1880 kfree_rcu(map, rcu);
1890 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1892 struct xps_dev_maps *dev_maps;
1894 bool active = false;
1896 mutex_lock(&xps_map_mutex);
1897 dev_maps = xmap_dereference(dev->xps_maps);
1902 for_each_possible_cpu(cpu) {
1903 for (i = index; i < dev->num_tx_queues; i++) {
1904 if (!remove_xps_queue(dev_maps, cpu, i))
1907 if (i == dev->num_tx_queues)
1912 RCU_INIT_POINTER(dev->xps_maps, NULL);
1913 kfree_rcu(dev_maps, rcu);
1916 for (i = index; i < dev->num_tx_queues; i++)
1917 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1921 mutex_unlock(&xps_map_mutex);
1924 static struct xps_map *expand_xps_map(struct xps_map *map,
1927 struct xps_map *new_map;
1928 int alloc_len = XPS_MIN_MAP_ALLOC;
1931 for (pos = 0; map && pos < map->len; pos++) {
1932 if (map->queues[pos] != index)
1937 /* Need to add queue to this CPU's existing map */
1939 if (pos < map->alloc_len)
1942 alloc_len = map->alloc_len * 2;
1945 /* Need to allocate new map to store queue on this CPU's map */
1946 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1951 for (i = 0; i < pos; i++)
1952 new_map->queues[i] = map->queues[i];
1953 new_map->alloc_len = alloc_len;
1959 int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index)
1961 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1962 struct xps_map *map, *new_map;
1963 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1964 int cpu, numa_node_id = -2;
1965 bool active = false;
1967 mutex_lock(&xps_map_mutex);
1969 dev_maps = xmap_dereference(dev->xps_maps);
1971 /* allocate memory for queue storage */
1972 for_each_online_cpu(cpu) {
1973 if (!cpumask_test_cpu(cpu, mask))
1977 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1981 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1984 map = expand_xps_map(map, cpu, index);
1988 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1992 goto out_no_new_maps;
1994 for_each_possible_cpu(cpu) {
1995 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1996 /* add queue to CPU maps */
1999 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2000 while ((pos < map->len) && (map->queues[pos] != index))
2003 if (pos == map->len)
2004 map->queues[map->len++] = index;
2006 if (numa_node_id == -2)
2007 numa_node_id = cpu_to_node(cpu);
2008 else if (numa_node_id != cpu_to_node(cpu))
2011 } else if (dev_maps) {
2012 /* fill in the new device map from the old device map */
2013 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2014 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2019 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2021 /* Cleanup old maps */
2023 for_each_possible_cpu(cpu) {
2024 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2025 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2026 if (map && map != new_map)
2027 kfree_rcu(map, rcu);
2030 kfree_rcu(dev_maps, rcu);
2033 dev_maps = new_dev_maps;
2037 /* update Tx queue numa node */
2038 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2039 (numa_node_id >= 0) ? numa_node_id :
2045 /* removes queue from unused CPUs */
2046 for_each_possible_cpu(cpu) {
2047 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2050 if (remove_xps_queue(dev_maps, cpu, index))
2054 /* free map if not active */
2056 RCU_INIT_POINTER(dev->xps_maps, NULL);
2057 kfree_rcu(dev_maps, rcu);
2061 mutex_unlock(&xps_map_mutex);
2065 /* remove any maps that we added */
2066 for_each_possible_cpu(cpu) {
2067 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2068 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2070 if (new_map && new_map != map)
2074 mutex_unlock(&xps_map_mutex);
2076 kfree(new_dev_maps);
2079 EXPORT_SYMBOL(netif_set_xps_queue);
2083 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2084 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2086 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2090 if (txq < 1 || txq > dev->num_tx_queues)
2093 if (dev->reg_state == NETREG_REGISTERED ||
2094 dev->reg_state == NETREG_UNREGISTERING) {
2097 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2103 netif_setup_tc(dev, txq);
2105 if (txq < dev->real_num_tx_queues) {
2106 qdisc_reset_all_tx_gt(dev, txq);
2108 netif_reset_xps_queues_gt(dev, txq);
2113 dev->real_num_tx_queues = txq;
2116 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2120 * netif_set_real_num_rx_queues - set actual number of RX queues used
2121 * @dev: Network device
2122 * @rxq: Actual number of RX queues
2124 * This must be called either with the rtnl_lock held or before
2125 * registration of the net device. Returns 0 on success, or a
2126 * negative error code. If called before registration, it always
2129 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2133 if (rxq < 1 || rxq > dev->num_rx_queues)
2136 if (dev->reg_state == NETREG_REGISTERED) {
2139 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2145 dev->real_num_rx_queues = rxq;
2148 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2152 * netif_get_num_default_rss_queues - default number of RSS queues
2154 * This routine should set an upper limit on the number of RSS queues
2155 * used by default by multiqueue devices.
2157 int netif_get_num_default_rss_queues(void)
2159 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2161 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2163 static inline void __netif_reschedule(struct Qdisc *q)
2165 struct softnet_data *sd;
2166 unsigned long flags;
2168 local_irq_save(flags);
2169 sd = &__get_cpu_var(softnet_data);
2170 q->next_sched = NULL;
2171 *sd->output_queue_tailp = q;
2172 sd->output_queue_tailp = &q->next_sched;
2173 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2174 local_irq_restore(flags);
2177 void __netif_schedule(struct Qdisc *q)
2179 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2180 __netif_reschedule(q);
2182 EXPORT_SYMBOL(__netif_schedule);
2184 void dev_kfree_skb_irq(struct sk_buff *skb)
2186 if (atomic_dec_and_test(&skb->users)) {
2187 struct softnet_data *sd;
2188 unsigned long flags;
2190 local_irq_save(flags);
2191 sd = &__get_cpu_var(softnet_data);
2192 skb->next = sd->completion_queue;
2193 sd->completion_queue = skb;
2194 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2195 local_irq_restore(flags);
2198 EXPORT_SYMBOL(dev_kfree_skb_irq);
2200 void dev_kfree_skb_any(struct sk_buff *skb)
2202 if (in_irq() || irqs_disabled())
2203 dev_kfree_skb_irq(skb);
2207 EXPORT_SYMBOL(dev_kfree_skb_any);
2211 * netif_device_detach - mark device as removed
2212 * @dev: network device
2214 * Mark device as removed from system and therefore no longer available.
2216 void netif_device_detach(struct net_device *dev)
2218 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2219 netif_running(dev)) {
2220 netif_tx_stop_all_queues(dev);
2223 EXPORT_SYMBOL(netif_device_detach);
2226 * netif_device_attach - mark device as attached
2227 * @dev: network device
2229 * Mark device as attached from system and restart if needed.
2231 void netif_device_attach(struct net_device *dev)
2233 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2234 netif_running(dev)) {
2235 netif_tx_wake_all_queues(dev);
2236 __netdev_watchdog_up(dev);
2239 EXPORT_SYMBOL(netif_device_attach);
2241 static void skb_warn_bad_offload(const struct sk_buff *skb)
2243 static const netdev_features_t null_features = 0;
2244 struct net_device *dev = skb->dev;
2245 const char *driver = "";
2247 if (dev && dev->dev.parent)
2248 driver = dev_driver_string(dev->dev.parent);
2250 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2251 "gso_type=%d ip_summed=%d\n",
2252 driver, dev ? &dev->features : &null_features,
2253 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2254 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2255 skb_shinfo(skb)->gso_type, skb->ip_summed);
2259 * Invalidate hardware checksum when packet is to be mangled, and
2260 * complete checksum manually on outgoing path.
2262 int skb_checksum_help(struct sk_buff *skb)
2265 int ret = 0, offset;
2267 if (skb->ip_summed == CHECKSUM_COMPLETE)
2268 goto out_set_summed;
2270 if (unlikely(skb_shinfo(skb)->gso_size)) {
2271 skb_warn_bad_offload(skb);
2275 offset = skb_checksum_start_offset(skb);
2276 BUG_ON(offset >= skb_headlen(skb));
2277 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2279 offset += skb->csum_offset;
2280 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2282 if (skb_cloned(skb) &&
2283 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2284 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2289 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2291 skb->ip_summed = CHECKSUM_NONE;
2295 EXPORT_SYMBOL(skb_checksum_help);
2298 * skb_gso_segment - Perform segmentation on skb.
2299 * @skb: buffer to segment
2300 * @features: features for the output path (see dev->features)
2302 * This function segments the given skb and returns a list of segments.
2304 * It may return NULL if the skb requires no segmentation. This is
2305 * only possible when GSO is used for verifying header integrity.
2307 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
2308 netdev_features_t features)
2310 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2311 struct packet_offload *ptype;
2312 __be16 type = skb->protocol;
2313 int vlan_depth = ETH_HLEN;
2316 while (type == htons(ETH_P_8021Q)) {
2317 struct vlan_hdr *vh;
2319 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2320 return ERR_PTR(-EINVAL);
2322 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2323 type = vh->h_vlan_encapsulated_proto;
2324 vlan_depth += VLAN_HLEN;
2327 skb_reset_mac_header(skb);
2328 skb->mac_len = skb->network_header - skb->mac_header;
2329 __skb_pull(skb, skb->mac_len);
2331 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2332 skb_warn_bad_offload(skb);
2334 if (skb_header_cloned(skb) &&
2335 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2336 return ERR_PTR(err);
2340 list_for_each_entry_rcu(ptype, &offload_base, list) {
2341 if (ptype->type == type && ptype->callbacks.gso_segment) {
2342 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2343 err = ptype->callbacks.gso_send_check(skb);
2344 segs = ERR_PTR(err);
2345 if (err || skb_gso_ok(skb, features))
2347 __skb_push(skb, (skb->data -
2348 skb_network_header(skb)));
2350 segs = ptype->callbacks.gso_segment(skb, features);
2356 __skb_push(skb, skb->data - skb_mac_header(skb));
2360 EXPORT_SYMBOL(skb_gso_segment);
2362 /* Take action when hardware reception checksum errors are detected. */
2364 void netdev_rx_csum_fault(struct net_device *dev)
2366 if (net_ratelimit()) {
2367 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2371 EXPORT_SYMBOL(netdev_rx_csum_fault);
2374 /* Actually, we should eliminate this check as soon as we know, that:
2375 * 1. IOMMU is present and allows to map all the memory.
2376 * 2. No high memory really exists on this machine.
2379 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2381 #ifdef CONFIG_HIGHMEM
2383 if (!(dev->features & NETIF_F_HIGHDMA)) {
2384 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2385 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2386 if (PageHighMem(skb_frag_page(frag)))
2391 if (PCI_DMA_BUS_IS_PHYS) {
2392 struct device *pdev = dev->dev.parent;
2396 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2397 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2398 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2399 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2408 void (*destructor)(struct sk_buff *skb);
2411 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2413 static void dev_gso_skb_destructor(struct sk_buff *skb)
2415 struct dev_gso_cb *cb;
2418 struct sk_buff *nskb = skb->next;
2420 skb->next = nskb->next;
2423 } while (skb->next);
2425 cb = DEV_GSO_CB(skb);
2427 cb->destructor(skb);
2431 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2432 * @skb: buffer to segment
2433 * @features: device features as applicable to this skb
2435 * This function segments the given skb and stores the list of segments
2438 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2440 struct sk_buff *segs;
2442 segs = skb_gso_segment(skb, features);
2444 /* Verifying header integrity only. */
2449 return PTR_ERR(segs);
2452 DEV_GSO_CB(skb)->destructor = skb->destructor;
2453 skb->destructor = dev_gso_skb_destructor;
2458 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2460 return ((features & NETIF_F_GEN_CSUM) ||
2461 ((features & NETIF_F_V4_CSUM) &&
2462 protocol == htons(ETH_P_IP)) ||
2463 ((features & NETIF_F_V6_CSUM) &&
2464 protocol == htons(ETH_P_IPV6)) ||
2465 ((features & NETIF_F_FCOE_CRC) &&
2466 protocol == htons(ETH_P_FCOE)));
2469 static netdev_features_t harmonize_features(struct sk_buff *skb,
2470 __be16 protocol, netdev_features_t features)
2472 if (skb->ip_summed != CHECKSUM_NONE &&
2473 !can_checksum_protocol(features, protocol)) {
2474 features &= ~NETIF_F_ALL_CSUM;
2475 features &= ~NETIF_F_SG;
2476 } else if (illegal_highdma(skb->dev, skb)) {
2477 features &= ~NETIF_F_SG;
2483 netdev_features_t netif_skb_features(struct sk_buff *skb)
2485 __be16 protocol = skb->protocol;
2486 netdev_features_t features = skb->dev->features;
2488 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2489 features &= ~NETIF_F_GSO_MASK;
2491 if (protocol == htons(ETH_P_8021Q)) {
2492 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2493 protocol = veh->h_vlan_encapsulated_proto;
2494 } else if (!vlan_tx_tag_present(skb)) {
2495 return harmonize_features(skb, protocol, features);
2498 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2500 if (protocol != htons(ETH_P_8021Q)) {
2501 return harmonize_features(skb, protocol, features);
2503 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2504 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2505 return harmonize_features(skb, protocol, features);
2508 EXPORT_SYMBOL(netif_skb_features);
2511 * Returns true if either:
2512 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2513 * 2. skb is fragmented and the device does not support SG.
2515 static inline int skb_needs_linearize(struct sk_buff *skb,
2518 return skb_is_nonlinear(skb) &&
2519 ((skb_has_frag_list(skb) &&
2520 !(features & NETIF_F_FRAGLIST)) ||
2521 (skb_shinfo(skb)->nr_frags &&
2522 !(features & NETIF_F_SG)));
2525 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2526 struct netdev_queue *txq)
2528 const struct net_device_ops *ops = dev->netdev_ops;
2529 int rc = NETDEV_TX_OK;
2530 unsigned int skb_len;
2532 if (likely(!skb->next)) {
2533 netdev_features_t features;
2536 * If device doesn't need skb->dst, release it right now while
2537 * its hot in this cpu cache
2539 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2542 features = netif_skb_features(skb);
2544 if (vlan_tx_tag_present(skb) &&
2545 !(features & NETIF_F_HW_VLAN_TX)) {
2546 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2553 /* If encapsulation offload request, verify we are testing
2554 * hardware encapsulation features instead of standard
2555 * features for the netdev
2557 if (skb->encapsulation)
2558 features &= dev->hw_enc_features;
2560 if (netif_needs_gso(skb, features)) {
2561 if (unlikely(dev_gso_segment(skb, features)))
2566 if (skb_needs_linearize(skb, features) &&
2567 __skb_linearize(skb))
2570 /* If packet is not checksummed and device does not
2571 * support checksumming for this protocol, complete
2572 * checksumming here.
2574 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2575 if (skb->encapsulation)
2576 skb_set_inner_transport_header(skb,
2577 skb_checksum_start_offset(skb));
2579 skb_set_transport_header(skb,
2580 skb_checksum_start_offset(skb));
2581 if (!(features & NETIF_F_ALL_CSUM) &&
2582 skb_checksum_help(skb))
2587 if (!list_empty(&ptype_all))
2588 dev_queue_xmit_nit(skb, dev);
2591 rc = ops->ndo_start_xmit(skb, dev);
2592 trace_net_dev_xmit(skb, rc, dev, skb_len);
2593 if (rc == NETDEV_TX_OK)
2594 txq_trans_update(txq);
2600 struct sk_buff *nskb = skb->next;
2602 skb->next = nskb->next;
2606 * If device doesn't need nskb->dst, release it right now while
2607 * its hot in this cpu cache
2609 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2612 if (!list_empty(&ptype_all))
2613 dev_queue_xmit_nit(nskb, dev);
2615 skb_len = nskb->len;
2616 rc = ops->ndo_start_xmit(nskb, dev);
2617 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2618 if (unlikely(rc != NETDEV_TX_OK)) {
2619 if (rc & ~NETDEV_TX_MASK)
2620 goto out_kfree_gso_skb;
2621 nskb->next = skb->next;
2625 txq_trans_update(txq);
2626 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2627 return NETDEV_TX_BUSY;
2628 } while (skb->next);
2631 if (likely(skb->next == NULL))
2632 skb->destructor = DEV_GSO_CB(skb)->destructor;
2639 static u32 hashrnd __read_mostly;
2642 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2643 * to be used as a distribution range.
2645 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2646 unsigned int num_tx_queues)
2650 u16 qcount = num_tx_queues;
2652 if (skb_rx_queue_recorded(skb)) {
2653 hash = skb_get_rx_queue(skb);
2654 while (unlikely(hash >= num_tx_queues))
2655 hash -= num_tx_queues;
2660 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2661 qoffset = dev->tc_to_txq[tc].offset;
2662 qcount = dev->tc_to_txq[tc].count;
2665 if (skb->sk && skb->sk->sk_hash)
2666 hash = skb->sk->sk_hash;
2668 hash = (__force u16) skb->protocol;
2669 hash = jhash_1word(hash, hashrnd);
2671 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2673 EXPORT_SYMBOL(__skb_tx_hash);
2675 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2677 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2678 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2679 dev->name, queue_index,
2680 dev->real_num_tx_queues);
2686 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2689 struct xps_dev_maps *dev_maps;
2690 struct xps_map *map;
2691 int queue_index = -1;
2694 dev_maps = rcu_dereference(dev->xps_maps);
2696 map = rcu_dereference(
2697 dev_maps->cpu_map[raw_smp_processor_id()]);
2700 queue_index = map->queues[0];
2703 if (skb->sk && skb->sk->sk_hash)
2704 hash = skb->sk->sk_hash;
2706 hash = (__force u16) skb->protocol ^
2708 hash = jhash_1word(hash, hashrnd);
2709 queue_index = map->queues[
2710 ((u64)hash * map->len) >> 32];
2712 if (unlikely(queue_index >= dev->real_num_tx_queues))
2724 u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
2726 struct sock *sk = skb->sk;
2727 int queue_index = sk_tx_queue_get(sk);
2729 if (queue_index < 0 || skb->ooo_okay ||
2730 queue_index >= dev->real_num_tx_queues) {
2731 int new_index = get_xps_queue(dev, skb);
2733 new_index = skb_tx_hash(dev, skb);
2735 if (queue_index != new_index && sk) {
2736 struct dst_entry *dst =
2737 rcu_dereference_check(sk->sk_dst_cache, 1);
2739 if (dst && skb_dst(skb) == dst)
2740 sk_tx_queue_set(sk, queue_index);
2744 queue_index = new_index;
2750 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
2751 struct sk_buff *skb)
2753 int queue_index = 0;
2755 if (dev->real_num_tx_queues != 1) {
2756 const struct net_device_ops *ops = dev->netdev_ops;
2757 if (ops->ndo_select_queue)
2758 queue_index = ops->ndo_select_queue(dev, skb);
2760 queue_index = __netdev_pick_tx(dev, skb);
2761 queue_index = dev_cap_txqueue(dev, queue_index);
2764 skb_set_queue_mapping(skb, queue_index);
2765 return netdev_get_tx_queue(dev, queue_index);
2768 static void qdisc_pkt_len_init(struct sk_buff *skb)
2770 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2772 qdisc_skb_cb(skb)->pkt_len = skb->len;
2774 /* To get more precise estimation of bytes sent on wire,
2775 * we add to pkt_len the headers size of all segments
2777 if (shinfo->gso_size) {
2778 unsigned int hdr_len = skb_transport_offset(skb);
2780 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2781 hdr_len += tcp_hdrlen(skb);
2783 hdr_len += sizeof(struct udphdr);
2784 qdisc_skb_cb(skb)->pkt_len += (shinfo->gso_segs - 1) * hdr_len;
2788 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2789 struct net_device *dev,
2790 struct netdev_queue *txq)
2792 spinlock_t *root_lock = qdisc_lock(q);
2796 qdisc_pkt_len_init(skb);
2797 qdisc_calculate_pkt_len(skb, q);
2799 * Heuristic to force contended enqueues to serialize on a
2800 * separate lock before trying to get qdisc main lock.
2801 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2802 * and dequeue packets faster.
2804 contended = qdisc_is_running(q);
2805 if (unlikely(contended))
2806 spin_lock(&q->busylock);
2808 spin_lock(root_lock);
2809 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2812 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2813 qdisc_run_begin(q)) {
2815 * This is a work-conserving queue; there are no old skbs
2816 * waiting to be sent out; and the qdisc is not running -
2817 * xmit the skb directly.
2819 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2822 qdisc_bstats_update(q, skb);
2824 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2825 if (unlikely(contended)) {
2826 spin_unlock(&q->busylock);
2833 rc = NET_XMIT_SUCCESS;
2836 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2837 if (qdisc_run_begin(q)) {
2838 if (unlikely(contended)) {
2839 spin_unlock(&q->busylock);
2845 spin_unlock(root_lock);
2846 if (unlikely(contended))
2847 spin_unlock(&q->busylock);
2851 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2852 static void skb_update_prio(struct sk_buff *skb)
2854 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2856 if (!skb->priority && skb->sk && map) {
2857 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2859 if (prioidx < map->priomap_len)
2860 skb->priority = map->priomap[prioidx];
2864 #define skb_update_prio(skb)
2867 static DEFINE_PER_CPU(int, xmit_recursion);
2868 #define RECURSION_LIMIT 10
2871 * dev_loopback_xmit - loop back @skb
2872 * @skb: buffer to transmit
2874 int dev_loopback_xmit(struct sk_buff *skb)
2876 skb_reset_mac_header(skb);
2877 __skb_pull(skb, skb_network_offset(skb));
2878 skb->pkt_type = PACKET_LOOPBACK;
2879 skb->ip_summed = CHECKSUM_UNNECESSARY;
2880 WARN_ON(!skb_dst(skb));
2885 EXPORT_SYMBOL(dev_loopback_xmit);
2888 * dev_queue_xmit - transmit a buffer
2889 * @skb: buffer to transmit
2891 * Queue a buffer for transmission to a network device. The caller must
2892 * have set the device and priority and built the buffer before calling
2893 * this function. The function can be called from an interrupt.
2895 * A negative errno code is returned on a failure. A success does not
2896 * guarantee the frame will be transmitted as it may be dropped due
2897 * to congestion or traffic shaping.
2899 * -----------------------------------------------------------------------------------
2900 * I notice this method can also return errors from the queue disciplines,
2901 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2904 * Regardless of the return value, the skb is consumed, so it is currently
2905 * difficult to retry a send to this method. (You can bump the ref count
2906 * before sending to hold a reference for retry if you are careful.)
2908 * When calling this method, interrupts MUST be enabled. This is because
2909 * the BH enable code must have IRQs enabled so that it will not deadlock.
2912 int dev_queue_xmit(struct sk_buff *skb)
2914 struct net_device *dev = skb->dev;
2915 struct netdev_queue *txq;
2919 /* Disable soft irqs for various locks below. Also
2920 * stops preemption for RCU.
2924 skb_update_prio(skb);
2926 txq = netdev_pick_tx(dev, skb);
2927 q = rcu_dereference_bh(txq->qdisc);
2929 #ifdef CONFIG_NET_CLS_ACT
2930 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2932 trace_net_dev_queue(skb);
2934 rc = __dev_xmit_skb(skb, q, dev, txq);
2938 /* The device has no queue. Common case for software devices:
2939 loopback, all the sorts of tunnels...
2941 Really, it is unlikely that netif_tx_lock protection is necessary
2942 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2944 However, it is possible, that they rely on protection
2947 Check this and shot the lock. It is not prone from deadlocks.
2948 Either shot noqueue qdisc, it is even simpler 8)
2950 if (dev->flags & IFF_UP) {
2951 int cpu = smp_processor_id(); /* ok because BHs are off */
2953 if (txq->xmit_lock_owner != cpu) {
2955 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2956 goto recursion_alert;
2958 HARD_TX_LOCK(dev, txq, cpu);
2960 if (!netif_xmit_stopped(txq)) {
2961 __this_cpu_inc(xmit_recursion);
2962 rc = dev_hard_start_xmit(skb, dev, txq);
2963 __this_cpu_dec(xmit_recursion);
2964 if (dev_xmit_complete(rc)) {
2965 HARD_TX_UNLOCK(dev, txq);
2969 HARD_TX_UNLOCK(dev, txq);
2970 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2973 /* Recursion is detected! It is possible,
2977 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2983 rcu_read_unlock_bh();
2988 rcu_read_unlock_bh();
2991 EXPORT_SYMBOL(dev_queue_xmit);
2994 /*=======================================================================
2996 =======================================================================*/
2998 int netdev_max_backlog __read_mostly = 1000;
2999 EXPORT_SYMBOL(netdev_max_backlog);
3001 int netdev_tstamp_prequeue __read_mostly = 1;
3002 int netdev_budget __read_mostly = 300;
3003 int weight_p __read_mostly = 64; /* old backlog weight */
3005 /* Called with irq disabled */
3006 static inline void ____napi_schedule(struct softnet_data *sd,
3007 struct napi_struct *napi)
3009 list_add_tail(&napi->poll_list, &sd->poll_list);
3010 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3014 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
3015 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
3016 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
3017 * if hash is a canonical 4-tuple hash over transport ports.
3019 void __skb_get_rxhash(struct sk_buff *skb)
3021 struct flow_keys keys;
3024 if (!skb_flow_dissect(skb, &keys))
3030 /* get a consistent hash (same value on both flow directions) */
3031 if (((__force u32)keys.dst < (__force u32)keys.src) ||
3032 (((__force u32)keys.dst == (__force u32)keys.src) &&
3033 ((__force u16)keys.port16[1] < (__force u16)keys.port16[0]))) {
3034 swap(keys.dst, keys.src);
3035 swap(keys.port16[0], keys.port16[1]);
3038 hash = jhash_3words((__force u32)keys.dst,
3039 (__force u32)keys.src,
3040 (__force u32)keys.ports, hashrnd);
3046 EXPORT_SYMBOL(__skb_get_rxhash);
3050 /* One global table that all flow-based protocols share. */
3051 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3052 EXPORT_SYMBOL(rps_sock_flow_table);
3054 struct static_key rps_needed __read_mostly;
3056 static struct rps_dev_flow *
3057 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3058 struct rps_dev_flow *rflow, u16 next_cpu)
3060 if (next_cpu != RPS_NO_CPU) {
3061 #ifdef CONFIG_RFS_ACCEL
3062 struct netdev_rx_queue *rxqueue;
3063 struct rps_dev_flow_table *flow_table;
3064 struct rps_dev_flow *old_rflow;
3069 /* Should we steer this flow to a different hardware queue? */
3070 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3071 !(dev->features & NETIF_F_NTUPLE))
3073 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3074 if (rxq_index == skb_get_rx_queue(skb))
3077 rxqueue = dev->_rx + rxq_index;
3078 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3081 flow_id = skb->rxhash & flow_table->mask;
3082 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3083 rxq_index, flow_id);
3087 rflow = &flow_table->flows[flow_id];
3089 if (old_rflow->filter == rflow->filter)
3090 old_rflow->filter = RPS_NO_FILTER;
3094 per_cpu(softnet_data, next_cpu).input_queue_head;
3097 rflow->cpu = next_cpu;
3102 * get_rps_cpu is called from netif_receive_skb and returns the target
3103 * CPU from the RPS map of the receiving queue for a given skb.
3104 * rcu_read_lock must be held on entry.
3106 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3107 struct rps_dev_flow **rflowp)
3109 struct netdev_rx_queue *rxqueue;
3110 struct rps_map *map;
3111 struct rps_dev_flow_table *flow_table;
3112 struct rps_sock_flow_table *sock_flow_table;
3116 if (skb_rx_queue_recorded(skb)) {
3117 u16 index = skb_get_rx_queue(skb);
3118 if (unlikely(index >= dev->real_num_rx_queues)) {
3119 WARN_ONCE(dev->real_num_rx_queues > 1,
3120 "%s received packet on queue %u, but number "
3121 "of RX queues is %u\n",
3122 dev->name, index, dev->real_num_rx_queues);
3125 rxqueue = dev->_rx + index;
3129 map = rcu_dereference(rxqueue->rps_map);
3131 if (map->len == 1 &&
3132 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3133 tcpu = map->cpus[0];
3134 if (cpu_online(tcpu))
3138 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3142 skb_reset_network_header(skb);
3143 if (!skb_get_rxhash(skb))
3146 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3147 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3148 if (flow_table && sock_flow_table) {
3150 struct rps_dev_flow *rflow;
3152 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3155 next_cpu = sock_flow_table->ents[skb->rxhash &
3156 sock_flow_table->mask];
3159 * If the desired CPU (where last recvmsg was done) is
3160 * different from current CPU (one in the rx-queue flow
3161 * table entry), switch if one of the following holds:
3162 * - Current CPU is unset (equal to RPS_NO_CPU).
3163 * - Current CPU is offline.
3164 * - The current CPU's queue tail has advanced beyond the
3165 * last packet that was enqueued using this table entry.
3166 * This guarantees that all previous packets for the flow
3167 * have been dequeued, thus preserving in order delivery.
3169 if (unlikely(tcpu != next_cpu) &&
3170 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3171 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3172 rflow->last_qtail)) >= 0)) {
3174 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3177 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3185 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3187 if (cpu_online(tcpu)) {
3197 #ifdef CONFIG_RFS_ACCEL
3200 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3201 * @dev: Device on which the filter was set
3202 * @rxq_index: RX queue index
3203 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3204 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3206 * Drivers that implement ndo_rx_flow_steer() should periodically call
3207 * this function for each installed filter and remove the filters for
3208 * which it returns %true.
3210 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3211 u32 flow_id, u16 filter_id)
3213 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3214 struct rps_dev_flow_table *flow_table;
3215 struct rps_dev_flow *rflow;
3220 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3221 if (flow_table && flow_id <= flow_table->mask) {
3222 rflow = &flow_table->flows[flow_id];
3223 cpu = ACCESS_ONCE(rflow->cpu);
3224 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3225 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3226 rflow->last_qtail) <
3227 (int)(10 * flow_table->mask)))
3233 EXPORT_SYMBOL(rps_may_expire_flow);
3235 #endif /* CONFIG_RFS_ACCEL */
3237 /* Called from hardirq (IPI) context */
3238 static void rps_trigger_softirq(void *data)
3240 struct softnet_data *sd = data;
3242 ____napi_schedule(sd, &sd->backlog);
3246 #endif /* CONFIG_RPS */
3249 * Check if this softnet_data structure is another cpu one
3250 * If yes, queue it to our IPI list and return 1
3253 static int rps_ipi_queued(struct softnet_data *sd)
3256 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3259 sd->rps_ipi_next = mysd->rps_ipi_list;
3260 mysd->rps_ipi_list = sd;
3262 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3265 #endif /* CONFIG_RPS */
3270 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3271 * queue (may be a remote CPU queue).
3273 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3274 unsigned int *qtail)
3276 struct softnet_data *sd;
3277 unsigned long flags;
3279 sd = &per_cpu(softnet_data, cpu);
3281 local_irq_save(flags);
3284 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
3285 if (skb_queue_len(&sd->input_pkt_queue)) {
3287 __skb_queue_tail(&sd->input_pkt_queue, skb);
3288 input_queue_tail_incr_save(sd, qtail);
3290 local_irq_restore(flags);
3291 return NET_RX_SUCCESS;
3294 /* Schedule NAPI for backlog device
3295 * We can use non atomic operation since we own the queue lock
3297 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3298 if (!rps_ipi_queued(sd))
3299 ____napi_schedule(sd, &sd->backlog);
3307 local_irq_restore(flags);
3309 atomic_long_inc(&skb->dev->rx_dropped);
3315 * netif_rx - post buffer to the network code
3316 * @skb: buffer to post
3318 * This function receives a packet from a device driver and queues it for
3319 * the upper (protocol) levels to process. It always succeeds. The buffer
3320 * may be dropped during processing for congestion control or by the
3324 * NET_RX_SUCCESS (no congestion)
3325 * NET_RX_DROP (packet was dropped)
3329 int netif_rx(struct sk_buff *skb)
3333 /* if netpoll wants it, pretend we never saw it */
3334 if (netpoll_rx(skb))
3337 net_timestamp_check(netdev_tstamp_prequeue, skb);
3339 trace_netif_rx(skb);
3341 if (static_key_false(&rps_needed)) {
3342 struct rps_dev_flow voidflow, *rflow = &voidflow;
3348 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3350 cpu = smp_processor_id();
3352 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3360 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3365 EXPORT_SYMBOL(netif_rx);
3367 int netif_rx_ni(struct sk_buff *skb)
3372 err = netif_rx(skb);
3373 if (local_softirq_pending())
3379 EXPORT_SYMBOL(netif_rx_ni);
3381 static void net_tx_action(struct softirq_action *h)
3383 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3385 if (sd->completion_queue) {
3386 struct sk_buff *clist;
3388 local_irq_disable();
3389 clist = sd->completion_queue;
3390 sd->completion_queue = NULL;
3394 struct sk_buff *skb = clist;
3395 clist = clist->next;
3397 WARN_ON(atomic_read(&skb->users));
3398 trace_kfree_skb(skb, net_tx_action);
3403 if (sd->output_queue) {
3406 local_irq_disable();
3407 head = sd->output_queue;
3408 sd->output_queue = NULL;
3409 sd->output_queue_tailp = &sd->output_queue;
3413 struct Qdisc *q = head;
3414 spinlock_t *root_lock;
3416 head = head->next_sched;
3418 root_lock = qdisc_lock(q);
3419 if (spin_trylock(root_lock)) {
3420 smp_mb__before_clear_bit();
3421 clear_bit(__QDISC_STATE_SCHED,
3424 spin_unlock(root_lock);
3426 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3428 __netif_reschedule(q);
3430 smp_mb__before_clear_bit();
3431 clear_bit(__QDISC_STATE_SCHED,
3439 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3440 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3441 /* This hook is defined here for ATM LANE */
3442 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3443 unsigned char *addr) __read_mostly;
3444 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3447 #ifdef CONFIG_NET_CLS_ACT
3448 /* TODO: Maybe we should just force sch_ingress to be compiled in
3449 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3450 * a compare and 2 stores extra right now if we dont have it on
3451 * but have CONFIG_NET_CLS_ACT
3452 * NOTE: This doesn't stop any functionality; if you dont have
3453 * the ingress scheduler, you just can't add policies on ingress.
3456 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3458 struct net_device *dev = skb->dev;
3459 u32 ttl = G_TC_RTTL(skb->tc_verd);
3460 int result = TC_ACT_OK;
3463 if (unlikely(MAX_RED_LOOP < ttl++)) {
3464 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3465 skb->skb_iif, dev->ifindex);
3469 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3470 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3473 if (q != &noop_qdisc) {
3474 spin_lock(qdisc_lock(q));
3475 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3476 result = qdisc_enqueue_root(skb, q);
3477 spin_unlock(qdisc_lock(q));
3483 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3484 struct packet_type **pt_prev,
3485 int *ret, struct net_device *orig_dev)
3487 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3489 if (!rxq || rxq->qdisc == &noop_qdisc)
3493 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3497 switch (ing_filter(skb, rxq)) {
3511 * netdev_rx_handler_register - register receive handler
3512 * @dev: device to register a handler for
3513 * @rx_handler: receive handler to register
3514 * @rx_handler_data: data pointer that is used by rx handler
3516 * Register a receive hander for a device. This handler will then be
3517 * called from __netif_receive_skb. A negative errno code is returned
3520 * The caller must hold the rtnl_mutex.
3522 * For a general description of rx_handler, see enum rx_handler_result.
3524 int netdev_rx_handler_register(struct net_device *dev,
3525 rx_handler_func_t *rx_handler,
3526 void *rx_handler_data)
3530 if (dev->rx_handler)
3533 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3534 rcu_assign_pointer(dev->rx_handler, rx_handler);
3538 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3541 * netdev_rx_handler_unregister - unregister receive handler
3542 * @dev: device to unregister a handler from
3544 * Unregister a receive hander from a device.
3546 * The caller must hold the rtnl_mutex.
3548 void netdev_rx_handler_unregister(struct net_device *dev)
3552 RCU_INIT_POINTER(dev->rx_handler, NULL);
3553 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3555 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3558 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3559 * the special handling of PFMEMALLOC skbs.
3561 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3563 switch (skb->protocol) {
3564 case __constant_htons(ETH_P_ARP):
3565 case __constant_htons(ETH_P_IP):
3566 case __constant_htons(ETH_P_IPV6):
3567 case __constant_htons(ETH_P_8021Q):
3574 static int __netif_receive_skb(struct sk_buff *skb)
3576 struct packet_type *ptype, *pt_prev;
3577 rx_handler_func_t *rx_handler;
3578 struct net_device *orig_dev;
3579 struct net_device *null_or_dev;
3580 bool deliver_exact = false;
3581 int ret = NET_RX_DROP;
3583 unsigned long pflags = current->flags;
3585 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3587 trace_netif_receive_skb(skb);
3590 * PFMEMALLOC skbs are special, they should
3591 * - be delivered to SOCK_MEMALLOC sockets only
3592 * - stay away from userspace
3593 * - have bounded memory usage
3595 * Use PF_MEMALLOC as this saves us from propagating the allocation
3596 * context down to all allocation sites.
3598 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3599 current->flags |= PF_MEMALLOC;
3601 /* if we've gotten here through NAPI, check netpoll */
3602 if (netpoll_receive_skb(skb))
3605 orig_dev = skb->dev;
3607 skb_reset_network_header(skb);
3608 if (!skb_transport_header_was_set(skb))
3609 skb_reset_transport_header(skb);
3610 skb_reset_mac_len(skb);
3617 skb->skb_iif = skb->dev->ifindex;
3619 __this_cpu_inc(softnet_data.processed);
3621 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3622 skb = vlan_untag(skb);
3627 #ifdef CONFIG_NET_CLS_ACT
3628 if (skb->tc_verd & TC_NCLS) {
3629 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3634 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3637 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3638 if (!ptype->dev || ptype->dev == skb->dev) {
3640 ret = deliver_skb(skb, pt_prev, orig_dev);
3646 #ifdef CONFIG_NET_CLS_ACT
3647 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3653 if (sk_memalloc_socks() && skb_pfmemalloc(skb)
3654 && !skb_pfmemalloc_protocol(skb))
3657 if (vlan_tx_tag_present(skb)) {
3659 ret = deliver_skb(skb, pt_prev, orig_dev);
3662 if (vlan_do_receive(&skb))
3664 else if (unlikely(!skb))
3668 rx_handler = rcu_dereference(skb->dev->rx_handler);
3671 ret = deliver_skb(skb, pt_prev, orig_dev);
3674 switch (rx_handler(&skb)) {
3675 case RX_HANDLER_CONSUMED:
3677 case RX_HANDLER_ANOTHER:
3679 case RX_HANDLER_EXACT:
3680 deliver_exact = true;
3681 case RX_HANDLER_PASS:
3688 if (vlan_tx_nonzero_tag_present(skb))
3689 skb->pkt_type = PACKET_OTHERHOST;
3691 /* deliver only exact match when indicated */
3692 null_or_dev = deliver_exact ? skb->dev : NULL;
3694 type = skb->protocol;
3695 list_for_each_entry_rcu(ptype,
3696 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3697 if (ptype->type == type &&
3698 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3699 ptype->dev == orig_dev)) {
3701 ret = deliver_skb(skb, pt_prev, orig_dev);
3707 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3710 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3713 atomic_long_inc(&skb->dev->rx_dropped);
3715 /* Jamal, now you will not able to escape explaining
3716 * me how you were going to use this. :-)
3724 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3729 * netif_receive_skb - process receive buffer from network
3730 * @skb: buffer to process
3732 * netif_receive_skb() is the main receive data processing function.
3733 * It always succeeds. The buffer may be dropped during processing
3734 * for congestion control or by the protocol layers.
3736 * This function may only be called from softirq context and interrupts
3737 * should be enabled.
3739 * Return values (usually ignored):
3740 * NET_RX_SUCCESS: no congestion
3741 * NET_RX_DROP: packet was dropped
3743 int netif_receive_skb(struct sk_buff *skb)
3745 net_timestamp_check(netdev_tstamp_prequeue, skb);
3747 if (skb_defer_rx_timestamp(skb))
3748 return NET_RX_SUCCESS;
3751 if (static_key_false(&rps_needed)) {
3752 struct rps_dev_flow voidflow, *rflow = &voidflow;
3757 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3760 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3767 return __netif_receive_skb(skb);
3769 EXPORT_SYMBOL(netif_receive_skb);
3771 /* Network device is going away, flush any packets still pending
3772 * Called with irqs disabled.
3774 static void flush_backlog(void *arg)
3776 struct net_device *dev = arg;
3777 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3778 struct sk_buff *skb, *tmp;
3781 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3782 if (skb->dev == dev) {
3783 __skb_unlink(skb, &sd->input_pkt_queue);
3785 input_queue_head_incr(sd);
3790 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3791 if (skb->dev == dev) {
3792 __skb_unlink(skb, &sd->process_queue);
3794 input_queue_head_incr(sd);
3799 static int napi_gro_complete(struct sk_buff *skb)
3801 struct packet_offload *ptype;
3802 __be16 type = skb->protocol;
3803 struct list_head *head = &offload_base;
3806 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3808 if (NAPI_GRO_CB(skb)->count == 1) {
3809 skb_shinfo(skb)->gso_size = 0;
3814 list_for_each_entry_rcu(ptype, head, list) {
3815 if (ptype->type != type || !ptype->callbacks.gro_complete)
3818 err = ptype->callbacks.gro_complete(skb);
3824 WARN_ON(&ptype->list == head);
3826 return NET_RX_SUCCESS;
3830 return netif_receive_skb(skb);
3833 /* napi->gro_list contains packets ordered by age.
3834 * youngest packets at the head of it.
3835 * Complete skbs in reverse order to reduce latencies.
3837 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3839 struct sk_buff *skb, *prev = NULL;
3841 /* scan list and build reverse chain */
3842 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3847 for (skb = prev; skb; skb = prev) {
3850 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3854 napi_gro_complete(skb);
3858 napi->gro_list = NULL;
3860 EXPORT_SYMBOL(napi_gro_flush);
3862 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3865 unsigned int maclen = skb->dev->hard_header_len;
3867 for (p = napi->gro_list; p; p = p->next) {
3868 unsigned long diffs;
3870 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3871 diffs |= p->vlan_tci ^ skb->vlan_tci;
3872 if (maclen == ETH_HLEN)
3873 diffs |= compare_ether_header(skb_mac_header(p),
3874 skb_gro_mac_header(skb));
3876 diffs = memcmp(skb_mac_header(p),
3877 skb_gro_mac_header(skb),
3879 NAPI_GRO_CB(p)->same_flow = !diffs;
3880 NAPI_GRO_CB(p)->flush = 0;
3884 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3886 struct sk_buff **pp = NULL;
3887 struct packet_offload *ptype;
3888 __be16 type = skb->protocol;
3889 struct list_head *head = &offload_base;
3892 enum gro_result ret;
3894 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3897 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3900 gro_list_prepare(napi, skb);
3903 list_for_each_entry_rcu(ptype, head, list) {
3904 if (ptype->type != type || !ptype->callbacks.gro_receive)
3907 skb_set_network_header(skb, skb_gro_offset(skb));
3908 mac_len = skb->network_header - skb->mac_header;
3909 skb->mac_len = mac_len;
3910 NAPI_GRO_CB(skb)->same_flow = 0;
3911 NAPI_GRO_CB(skb)->flush = 0;
3912 NAPI_GRO_CB(skb)->free = 0;
3914 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3919 if (&ptype->list == head)
3922 same_flow = NAPI_GRO_CB(skb)->same_flow;
3923 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3926 struct sk_buff *nskb = *pp;
3930 napi_gro_complete(nskb);
3937 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3941 NAPI_GRO_CB(skb)->count = 1;
3942 NAPI_GRO_CB(skb)->age = jiffies;
3943 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3944 skb->next = napi->gro_list;
3945 napi->gro_list = skb;
3949 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3950 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3952 BUG_ON(skb->end - skb->tail < grow);
3954 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3957 skb->data_len -= grow;
3959 skb_shinfo(skb)->frags[0].page_offset += grow;
3960 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3962 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3963 skb_frag_unref(skb, 0);
3964 memmove(skb_shinfo(skb)->frags,
3965 skb_shinfo(skb)->frags + 1,
3966 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3979 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3983 if (netif_receive_skb(skb))
3991 case GRO_MERGED_FREE:
3992 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3993 kmem_cache_free(skbuff_head_cache, skb);
4006 static void skb_gro_reset_offset(struct sk_buff *skb)
4008 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4009 const skb_frag_t *frag0 = &pinfo->frags[0];
4011 NAPI_GRO_CB(skb)->data_offset = 0;
4012 NAPI_GRO_CB(skb)->frag0 = NULL;
4013 NAPI_GRO_CB(skb)->frag0_len = 0;
4015 if (skb->mac_header == skb->tail &&
4017 !PageHighMem(skb_frag_page(frag0))) {
4018 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4019 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4023 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4025 skb_gro_reset_offset(skb);
4027 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4029 EXPORT_SYMBOL(napi_gro_receive);
4031 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4033 __skb_pull(skb, skb_headlen(skb));
4034 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4035 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4037 skb->dev = napi->dev;
4043 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4045 struct sk_buff *skb = napi->skb;
4048 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4054 EXPORT_SYMBOL(napi_get_frags);
4056 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
4062 skb->protocol = eth_type_trans(skb, skb->dev);
4064 if (ret == GRO_HELD)
4065 skb_gro_pull(skb, -ETH_HLEN);
4066 else if (netif_receive_skb(skb))
4071 case GRO_MERGED_FREE:
4072 napi_reuse_skb(napi, skb);
4082 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4084 struct sk_buff *skb = napi->skb;
4091 skb_reset_mac_header(skb);
4092 skb_gro_reset_offset(skb);
4094 off = skb_gro_offset(skb);
4095 hlen = off + sizeof(*eth);
4096 eth = skb_gro_header_fast(skb, off);
4097 if (skb_gro_header_hard(skb, hlen)) {
4098 eth = skb_gro_header_slow(skb, hlen, off);
4099 if (unlikely(!eth)) {
4100 napi_reuse_skb(napi, skb);
4106 skb_gro_pull(skb, sizeof(*eth));
4109 * This works because the only protocols we care about don't require
4110 * special handling. We'll fix it up properly at the end.
4112 skb->protocol = eth->h_proto;
4118 gro_result_t napi_gro_frags(struct napi_struct *napi)
4120 struct sk_buff *skb = napi_frags_skb(napi);
4125 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4127 EXPORT_SYMBOL(napi_gro_frags);
4130 * net_rps_action sends any pending IPI's for rps.
4131 * Note: called with local irq disabled, but exits with local irq enabled.
4133 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4136 struct softnet_data *remsd = sd->rps_ipi_list;
4139 sd->rps_ipi_list = NULL;
4143 /* Send pending IPI's to kick RPS processing on remote cpus. */
4145 struct softnet_data *next = remsd->rps_ipi_next;
4147 if (cpu_online(remsd->cpu))
4148 __smp_call_function_single(remsd->cpu,
4157 static int process_backlog(struct napi_struct *napi, int quota)
4160 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4163 /* Check if we have pending ipi, its better to send them now,
4164 * not waiting net_rx_action() end.
4166 if (sd->rps_ipi_list) {
4167 local_irq_disable();
4168 net_rps_action_and_irq_enable(sd);
4171 napi->weight = weight_p;
4172 local_irq_disable();
4173 while (work < quota) {
4174 struct sk_buff *skb;
4177 while ((skb = __skb_dequeue(&sd->process_queue))) {
4179 __netif_receive_skb(skb);
4180 local_irq_disable();
4181 input_queue_head_incr(sd);
4182 if (++work >= quota) {
4189 qlen = skb_queue_len(&sd->input_pkt_queue);
4191 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4192 &sd->process_queue);
4194 if (qlen < quota - work) {
4196 * Inline a custom version of __napi_complete().
4197 * only current cpu owns and manipulates this napi,
4198 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4199 * we can use a plain write instead of clear_bit(),
4200 * and we dont need an smp_mb() memory barrier.
4202 list_del(&napi->poll_list);
4205 quota = work + qlen;
4215 * __napi_schedule - schedule for receive
4216 * @n: entry to schedule
4218 * The entry's receive function will be scheduled to run
4220 void __napi_schedule(struct napi_struct *n)
4222 unsigned long flags;
4224 local_irq_save(flags);
4225 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4226 local_irq_restore(flags);
4228 EXPORT_SYMBOL(__napi_schedule);
4230 void __napi_complete(struct napi_struct *n)
4232 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4233 BUG_ON(n->gro_list);
4235 list_del(&n->poll_list);
4236 smp_mb__before_clear_bit();
4237 clear_bit(NAPI_STATE_SCHED, &n->state);
4239 EXPORT_SYMBOL(__napi_complete);
4241 void napi_complete(struct napi_struct *n)
4243 unsigned long flags;
4246 * don't let napi dequeue from the cpu poll list
4247 * just in case its running on a different cpu
4249 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4252 napi_gro_flush(n, false);
4253 local_irq_save(flags);
4255 local_irq_restore(flags);
4257 EXPORT_SYMBOL(napi_complete);
4259 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4260 int (*poll)(struct napi_struct *, int), int weight)
4262 INIT_LIST_HEAD(&napi->poll_list);
4263 napi->gro_count = 0;
4264 napi->gro_list = NULL;
4267 napi->weight = weight;
4268 list_add(&napi->dev_list, &dev->napi_list);
4270 #ifdef CONFIG_NETPOLL
4271 spin_lock_init(&napi->poll_lock);
4272 napi->poll_owner = -1;
4274 set_bit(NAPI_STATE_SCHED, &napi->state);
4276 EXPORT_SYMBOL(netif_napi_add);
4278 void netif_napi_del(struct napi_struct *napi)
4280 struct sk_buff *skb, *next;
4282 list_del_init(&napi->dev_list);
4283 napi_free_frags(napi);
4285 for (skb = napi->gro_list; skb; skb = next) {
4291 napi->gro_list = NULL;
4292 napi->gro_count = 0;
4294 EXPORT_SYMBOL(netif_napi_del);
4296 static void net_rx_action(struct softirq_action *h)
4298 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4299 unsigned long time_limit = jiffies + 2;
4300 int budget = netdev_budget;
4303 local_irq_disable();
4305 while (!list_empty(&sd->poll_list)) {
4306 struct napi_struct *n;
4309 /* If softirq window is exhuasted then punt.
4310 * Allow this to run for 2 jiffies since which will allow
4311 * an average latency of 1.5/HZ.
4313 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
4318 /* Even though interrupts have been re-enabled, this
4319 * access is safe because interrupts can only add new
4320 * entries to the tail of this list, and only ->poll()
4321 * calls can remove this head entry from the list.
4323 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4325 have = netpoll_poll_lock(n);
4329 /* This NAPI_STATE_SCHED test is for avoiding a race
4330 * with netpoll's poll_napi(). Only the entity which
4331 * obtains the lock and sees NAPI_STATE_SCHED set will
4332 * actually make the ->poll() call. Therefore we avoid
4333 * accidentally calling ->poll() when NAPI is not scheduled.
4336 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4337 work = n->poll(n, weight);
4341 WARN_ON_ONCE(work > weight);
4345 local_irq_disable();
4347 /* Drivers must not modify the NAPI state if they
4348 * consume the entire weight. In such cases this code
4349 * still "owns" the NAPI instance and therefore can
4350 * move the instance around on the list at-will.
4352 if (unlikely(work == weight)) {
4353 if (unlikely(napi_disable_pending(n))) {
4356 local_irq_disable();
4359 /* flush too old packets
4360 * If HZ < 1000, flush all packets.
4363 napi_gro_flush(n, HZ >= 1000);
4364 local_irq_disable();
4366 list_move_tail(&n->poll_list, &sd->poll_list);
4370 netpoll_poll_unlock(have);
4373 net_rps_action_and_irq_enable(sd);
4375 #ifdef CONFIG_NET_DMA
4377 * There may not be any more sk_buffs coming right now, so push
4378 * any pending DMA copies to hardware
4380 dma_issue_pending_all();
4387 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4391 static gifconf_func_t *gifconf_list[NPROTO];
4394 * register_gifconf - register a SIOCGIF handler
4395 * @family: Address family
4396 * @gifconf: Function handler
4398 * Register protocol dependent address dumping routines. The handler
4399 * that is passed must not be freed or reused until it has been replaced
4400 * by another handler.
4402 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4404 if (family >= NPROTO)
4406 gifconf_list[family] = gifconf;
4409 EXPORT_SYMBOL(register_gifconf);
4413 * Map an interface index to its name (SIOCGIFNAME)
4417 * We need this ioctl for efficient implementation of the
4418 * if_indextoname() function required by the IPv6 API. Without
4419 * it, we would have to search all the interfaces to find a
4423 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4425 struct net_device *dev;
4430 * Fetch the caller's info block.
4433 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4437 seq = read_seqcount_begin(&devnet_rename_seq);
4439 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4445 strcpy(ifr.ifr_name, dev->name);
4447 if (read_seqcount_retry(&devnet_rename_seq, seq))
4450 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4456 * Perform a SIOCGIFCONF call. This structure will change
4457 * size eventually, and there is nothing I can do about it.
4458 * Thus we will need a 'compatibility mode'.
4461 static int dev_ifconf(struct net *net, char __user *arg)
4464 struct net_device *dev;
4471 * Fetch the caller's info block.
4474 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4481 * Loop over the interfaces, and write an info block for each.
4485 for_each_netdev(net, dev) {
4486 for (i = 0; i < NPROTO; i++) {
4487 if (gifconf_list[i]) {
4490 done = gifconf_list[i](dev, NULL, 0);
4492 done = gifconf_list[i](dev, pos + total,
4502 * All done. Write the updated control block back to the caller.
4504 ifc.ifc_len = total;
4507 * Both BSD and Solaris return 0 here, so we do too.
4509 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4512 #ifdef CONFIG_PROC_FS
4514 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4516 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4517 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4518 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4520 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4522 struct net *net = seq_file_net(seq);
4523 struct net_device *dev;
4524 struct hlist_node *p;
4525 struct hlist_head *h;
4526 unsigned int count = 0, offset = get_offset(*pos);
4528 h = &net->dev_name_head[get_bucket(*pos)];
4529 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4530 if (++count == offset)
4537 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4539 struct net_device *dev;
4540 unsigned int bucket;
4543 dev = dev_from_same_bucket(seq, pos);
4547 bucket = get_bucket(*pos) + 1;
4548 *pos = set_bucket_offset(bucket, 1);
4549 } while (bucket < NETDEV_HASHENTRIES);
4555 * This is invoked by the /proc filesystem handler to display a device
4558 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4563 return SEQ_START_TOKEN;
4565 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4568 return dev_from_bucket(seq, pos);
4571 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4574 return dev_from_bucket(seq, pos);
4577 void dev_seq_stop(struct seq_file *seq, void *v)
4583 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4585 struct rtnl_link_stats64 temp;
4586 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4588 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4589 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4590 dev->name, stats->rx_bytes, stats->rx_packets,
4592 stats->rx_dropped + stats->rx_missed_errors,
4593 stats->rx_fifo_errors,
4594 stats->rx_length_errors + stats->rx_over_errors +
4595 stats->rx_crc_errors + stats->rx_frame_errors,
4596 stats->rx_compressed, stats->multicast,
4597 stats->tx_bytes, stats->tx_packets,
4598 stats->tx_errors, stats->tx_dropped,
4599 stats->tx_fifo_errors, stats->collisions,
4600 stats->tx_carrier_errors +
4601 stats->tx_aborted_errors +
4602 stats->tx_window_errors +
4603 stats->tx_heartbeat_errors,
4604 stats->tx_compressed);
4608 * Called from the PROCfs module. This now uses the new arbitrary sized
4609 * /proc/net interface to create /proc/net/dev
4611 static int dev_seq_show(struct seq_file *seq, void *v)
4613 if (v == SEQ_START_TOKEN)
4614 seq_puts(seq, "Inter-| Receive "
4616 " face |bytes packets errs drop fifo frame "
4617 "compressed multicast|bytes packets errs "
4618 "drop fifo colls carrier compressed\n");
4620 dev_seq_printf_stats(seq, v);
4624 static struct softnet_data *softnet_get_online(loff_t *pos)
4626 struct softnet_data *sd = NULL;
4628 while (*pos < nr_cpu_ids)
4629 if (cpu_online(*pos)) {
4630 sd = &per_cpu(softnet_data, *pos);
4637 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4639 return softnet_get_online(pos);
4642 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4645 return softnet_get_online(pos);
4648 static void softnet_seq_stop(struct seq_file *seq, void *v)
4652 static int softnet_seq_show(struct seq_file *seq, void *v)
4654 struct softnet_data *sd = v;
4656 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4657 sd->processed, sd->dropped, sd->time_squeeze, 0,
4658 0, 0, 0, 0, /* was fastroute */
4659 sd->cpu_collision, sd->received_rps);
4663 static const struct seq_operations dev_seq_ops = {
4664 .start = dev_seq_start,
4665 .next = dev_seq_next,
4666 .stop = dev_seq_stop,
4667 .show = dev_seq_show,
4670 static int dev_seq_open(struct inode *inode, struct file *file)
4672 return seq_open_net(inode, file, &dev_seq_ops,
4673 sizeof(struct seq_net_private));
4676 static const struct file_operations dev_seq_fops = {
4677 .owner = THIS_MODULE,
4678 .open = dev_seq_open,
4680 .llseek = seq_lseek,
4681 .release = seq_release_net,
4684 static const struct seq_operations softnet_seq_ops = {
4685 .start = softnet_seq_start,
4686 .next = softnet_seq_next,
4687 .stop = softnet_seq_stop,
4688 .show = softnet_seq_show,
4691 static int softnet_seq_open(struct inode *inode, struct file *file)
4693 return seq_open(file, &softnet_seq_ops);
4696 static const struct file_operations softnet_seq_fops = {
4697 .owner = THIS_MODULE,
4698 .open = softnet_seq_open,
4700 .llseek = seq_lseek,
4701 .release = seq_release,
4704 static void *ptype_get_idx(loff_t pos)
4706 struct packet_type *pt = NULL;
4710 list_for_each_entry_rcu(pt, &ptype_all, list) {
4716 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4717 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4726 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4730 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4733 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4735 struct packet_type *pt;
4736 struct list_head *nxt;
4740 if (v == SEQ_START_TOKEN)
4741 return ptype_get_idx(0);
4744 nxt = pt->list.next;
4745 if (pt->type == htons(ETH_P_ALL)) {
4746 if (nxt != &ptype_all)
4749 nxt = ptype_base[0].next;
4751 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4753 while (nxt == &ptype_base[hash]) {
4754 if (++hash >= PTYPE_HASH_SIZE)
4756 nxt = ptype_base[hash].next;
4759 return list_entry(nxt, struct packet_type, list);
4762 static void ptype_seq_stop(struct seq_file *seq, void *v)
4768 static int ptype_seq_show(struct seq_file *seq, void *v)
4770 struct packet_type *pt = v;
4772 if (v == SEQ_START_TOKEN)
4773 seq_puts(seq, "Type Device Function\n");
4774 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4775 if (pt->type == htons(ETH_P_ALL))
4776 seq_puts(seq, "ALL ");
4778 seq_printf(seq, "%04x", ntohs(pt->type));
4780 seq_printf(seq, " %-8s %pF\n",
4781 pt->dev ? pt->dev->name : "", pt->func);
4787 static const struct seq_operations ptype_seq_ops = {
4788 .start = ptype_seq_start,
4789 .next = ptype_seq_next,
4790 .stop = ptype_seq_stop,
4791 .show = ptype_seq_show,
4794 static int ptype_seq_open(struct inode *inode, struct file *file)
4796 return seq_open_net(inode, file, &ptype_seq_ops,
4797 sizeof(struct seq_net_private));
4800 static const struct file_operations ptype_seq_fops = {
4801 .owner = THIS_MODULE,
4802 .open = ptype_seq_open,
4804 .llseek = seq_lseek,
4805 .release = seq_release_net,
4809 static int __net_init dev_proc_net_init(struct net *net)
4813 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4815 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4817 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4820 if (wext_proc_init(net))
4826 proc_net_remove(net, "ptype");
4828 proc_net_remove(net, "softnet_stat");
4830 proc_net_remove(net, "dev");
4834 static void __net_exit dev_proc_net_exit(struct net *net)
4836 wext_proc_exit(net);
4838 proc_net_remove(net, "ptype");
4839 proc_net_remove(net, "softnet_stat");
4840 proc_net_remove(net, "dev");
4843 static struct pernet_operations __net_initdata dev_proc_ops = {
4844 .init = dev_proc_net_init,
4845 .exit = dev_proc_net_exit,
4848 static int __init dev_proc_init(void)
4850 return register_pernet_subsys(&dev_proc_ops);
4853 #define dev_proc_init() 0
4854 #endif /* CONFIG_PROC_FS */
4857 struct netdev_upper {
4858 struct net_device *dev;
4860 struct list_head list;
4861 struct rcu_head rcu;
4862 struct list_head search_list;
4865 static void __append_search_uppers(struct list_head *search_list,
4866 struct net_device *dev)
4868 struct netdev_upper *upper;
4870 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4871 /* check if this upper is not already in search list */
4872 if (list_empty(&upper->search_list))
4873 list_add_tail(&upper->search_list, search_list);
4877 static bool __netdev_search_upper_dev(struct net_device *dev,
4878 struct net_device *upper_dev)
4880 LIST_HEAD(search_list);
4881 struct netdev_upper *upper;
4882 struct netdev_upper *tmp;
4885 __append_search_uppers(&search_list, dev);
4886 list_for_each_entry(upper, &search_list, search_list) {
4887 if (upper->dev == upper_dev) {
4891 __append_search_uppers(&search_list, upper->dev);
4893 list_for_each_entry_safe(upper, tmp, &search_list, search_list)
4894 INIT_LIST_HEAD(&upper->search_list);
4898 static struct netdev_upper *__netdev_find_upper(struct net_device *dev,
4899 struct net_device *upper_dev)
4901 struct netdev_upper *upper;
4903 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4904 if (upper->dev == upper_dev)
4911 * netdev_has_upper_dev - Check if device is linked to an upper device
4913 * @upper_dev: upper device to check
4915 * Find out if a device is linked to specified upper device and return true
4916 * in case it is. Note that this checks only immediate upper device,
4917 * not through a complete stack of devices. The caller must hold the RTNL lock.
4919 bool netdev_has_upper_dev(struct net_device *dev,
4920 struct net_device *upper_dev)
4924 return __netdev_find_upper(dev, upper_dev);
4926 EXPORT_SYMBOL(netdev_has_upper_dev);
4929 * netdev_has_any_upper_dev - Check if device is linked to some device
4932 * Find out if a device is linked to an upper device and return true in case
4933 * it is. The caller must hold the RTNL lock.
4935 bool netdev_has_any_upper_dev(struct net_device *dev)
4939 return !list_empty(&dev->upper_dev_list);
4941 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4944 * netdev_master_upper_dev_get - Get master upper device
4947 * Find a master upper device and return pointer to it or NULL in case
4948 * it's not there. The caller must hold the RTNL lock.
4950 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4952 struct netdev_upper *upper;
4956 if (list_empty(&dev->upper_dev_list))
4959 upper = list_first_entry(&dev->upper_dev_list,
4960 struct netdev_upper, list);
4961 if (likely(upper->master))
4965 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4968 * netdev_master_upper_dev_get_rcu - Get master upper device
4971 * Find a master upper device and return pointer to it or NULL in case
4972 * it's not there. The caller must hold the RCU read lock.
4974 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4976 struct netdev_upper *upper;
4978 upper = list_first_or_null_rcu(&dev->upper_dev_list,
4979 struct netdev_upper, list);
4980 if (upper && likely(upper->master))
4984 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4986 static int __netdev_upper_dev_link(struct net_device *dev,
4987 struct net_device *upper_dev, bool master)
4989 struct netdev_upper *upper;
4993 if (dev == upper_dev)
4996 /* To prevent loops, check if dev is not upper device to upper_dev. */
4997 if (__netdev_search_upper_dev(upper_dev, dev))
5000 if (__netdev_find_upper(dev, upper_dev))
5003 if (master && netdev_master_upper_dev_get(dev))
5006 upper = kmalloc(sizeof(*upper), GFP_KERNEL);
5010 upper->dev = upper_dev;
5011 upper->master = master;
5012 INIT_LIST_HEAD(&upper->search_list);
5014 /* Ensure that master upper link is always the first item in list. */
5016 list_add_rcu(&upper->list, &dev->upper_dev_list);
5018 list_add_tail_rcu(&upper->list, &dev->upper_dev_list);
5019 dev_hold(upper_dev);
5025 * netdev_upper_dev_link - Add a link to the upper device
5027 * @upper_dev: new upper device
5029 * Adds a link to device which is upper to this one. The caller must hold
5030 * the RTNL lock. On a failure a negative errno code is returned.
5031 * On success the reference counts are adjusted and the function
5034 int netdev_upper_dev_link(struct net_device *dev,
5035 struct net_device *upper_dev)
5037 return __netdev_upper_dev_link(dev, upper_dev, false);
5039 EXPORT_SYMBOL(netdev_upper_dev_link);
5042 * netdev_master_upper_dev_link - Add a master link to the upper device
5044 * @upper_dev: new upper device
5046 * Adds a link to device which is upper to this one. In this case, only
5047 * one master upper device can be linked, although other non-master devices
5048 * might be linked as well. The caller must hold the RTNL lock.
5049 * On a failure a negative errno code is returned. On success the reference
5050 * counts are adjusted and the function returns zero.
5052 int netdev_master_upper_dev_link(struct net_device *dev,
5053 struct net_device *upper_dev)
5055 return __netdev_upper_dev_link(dev, upper_dev, true);
5057 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5060 * netdev_upper_dev_unlink - Removes a link to upper device
5062 * @upper_dev: new upper device
5064 * Removes a link to device which is upper to this one. The caller must hold
5067 void netdev_upper_dev_unlink(struct net_device *dev,
5068 struct net_device *upper_dev)
5070 struct netdev_upper *upper;
5074 upper = __netdev_find_upper(dev, upper_dev);
5077 list_del_rcu(&upper->list);
5079 kfree_rcu(upper, rcu);
5081 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5083 static void dev_change_rx_flags(struct net_device *dev, int flags)
5085 const struct net_device_ops *ops = dev->netdev_ops;
5087 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
5088 ops->ndo_change_rx_flags(dev, flags);
5091 static int __dev_set_promiscuity(struct net_device *dev, int inc)
5093 unsigned int old_flags = dev->flags;
5099 dev->flags |= IFF_PROMISC;
5100 dev->promiscuity += inc;
5101 if (dev->promiscuity == 0) {
5104 * If inc causes overflow, untouch promisc and return error.
5107 dev->flags &= ~IFF_PROMISC;
5109 dev->promiscuity -= inc;
5110 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5115 if (dev->flags != old_flags) {
5116 pr_info("device %s %s promiscuous mode\n",
5118 dev->flags & IFF_PROMISC ? "entered" : "left");
5119 if (audit_enabled) {
5120 current_uid_gid(&uid, &gid);
5121 audit_log(current->audit_context, GFP_ATOMIC,
5122 AUDIT_ANOM_PROMISCUOUS,
5123 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5124 dev->name, (dev->flags & IFF_PROMISC),
5125 (old_flags & IFF_PROMISC),
5126 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5127 from_kuid(&init_user_ns, uid),
5128 from_kgid(&init_user_ns, gid),
5129 audit_get_sessionid(current));
5132 dev_change_rx_flags(dev, IFF_PROMISC);
5138 * dev_set_promiscuity - update promiscuity count on a device
5142 * Add or remove promiscuity from a device. While the count in the device
5143 * remains above zero the interface remains promiscuous. Once it hits zero
5144 * the device reverts back to normal filtering operation. A negative inc
5145 * value is used to drop promiscuity on the device.
5146 * Return 0 if successful or a negative errno code on error.
5148 int dev_set_promiscuity(struct net_device *dev, int inc)
5150 unsigned int old_flags = dev->flags;
5153 err = __dev_set_promiscuity(dev, inc);
5156 if (dev->flags != old_flags)
5157 dev_set_rx_mode(dev);
5160 EXPORT_SYMBOL(dev_set_promiscuity);
5163 * dev_set_allmulti - update allmulti count on a device
5167 * Add or remove reception of all multicast frames to a device. While the
5168 * count in the device remains above zero the interface remains listening
5169 * to all interfaces. Once it hits zero the device reverts back to normal
5170 * filtering operation. A negative @inc value is used to drop the counter
5171 * when releasing a resource needing all multicasts.
5172 * Return 0 if successful or a negative errno code on error.
5175 int dev_set_allmulti(struct net_device *dev, int inc)
5177 unsigned int old_flags = dev->flags;
5181 dev->flags |= IFF_ALLMULTI;
5182 dev->allmulti += inc;
5183 if (dev->allmulti == 0) {
5186 * If inc causes overflow, untouch allmulti and return error.
5189 dev->flags &= ~IFF_ALLMULTI;
5191 dev->allmulti -= inc;
5192 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5197 if (dev->flags ^ old_flags) {
5198 dev_change_rx_flags(dev, IFF_ALLMULTI);
5199 dev_set_rx_mode(dev);
5203 EXPORT_SYMBOL(dev_set_allmulti);
5206 * Upload unicast and multicast address lists to device and
5207 * configure RX filtering. When the device doesn't support unicast
5208 * filtering it is put in promiscuous mode while unicast addresses
5211 void __dev_set_rx_mode(struct net_device *dev)
5213 const struct net_device_ops *ops = dev->netdev_ops;
5215 /* dev_open will call this function so the list will stay sane. */
5216 if (!(dev->flags&IFF_UP))
5219 if (!netif_device_present(dev))
5222 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5223 /* Unicast addresses changes may only happen under the rtnl,
5224 * therefore calling __dev_set_promiscuity here is safe.
5226 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5227 __dev_set_promiscuity(dev, 1);
5228 dev->uc_promisc = true;
5229 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5230 __dev_set_promiscuity(dev, -1);
5231 dev->uc_promisc = false;
5235 if (ops->ndo_set_rx_mode)
5236 ops->ndo_set_rx_mode(dev);
5239 void dev_set_rx_mode(struct net_device *dev)
5241 netif_addr_lock_bh(dev);
5242 __dev_set_rx_mode(dev);
5243 netif_addr_unlock_bh(dev);
5247 * dev_get_flags - get flags reported to userspace
5250 * Get the combination of flag bits exported through APIs to userspace.
5252 unsigned int dev_get_flags(const struct net_device *dev)
5256 flags = (dev->flags & ~(IFF_PROMISC |
5261 (dev->gflags & (IFF_PROMISC |
5264 if (netif_running(dev)) {
5265 if (netif_oper_up(dev))
5266 flags |= IFF_RUNNING;
5267 if (netif_carrier_ok(dev))
5268 flags |= IFF_LOWER_UP;
5269 if (netif_dormant(dev))
5270 flags |= IFF_DORMANT;
5275 EXPORT_SYMBOL(dev_get_flags);
5277 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5279 unsigned int old_flags = dev->flags;
5285 * Set the flags on our device.
5288 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5289 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5291 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5295 * Load in the correct multicast list now the flags have changed.
5298 if ((old_flags ^ flags) & IFF_MULTICAST)
5299 dev_change_rx_flags(dev, IFF_MULTICAST);
5301 dev_set_rx_mode(dev);
5304 * Have we downed the interface. We handle IFF_UP ourselves
5305 * according to user attempts to set it, rather than blindly
5310 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5311 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5314 dev_set_rx_mode(dev);
5317 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5318 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5320 dev->gflags ^= IFF_PROMISC;
5321 dev_set_promiscuity(dev, inc);
5324 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5325 is important. Some (broken) drivers set IFF_PROMISC, when
5326 IFF_ALLMULTI is requested not asking us and not reporting.
5328 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5329 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5331 dev->gflags ^= IFF_ALLMULTI;
5332 dev_set_allmulti(dev, inc);
5338 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
5340 unsigned int changes = dev->flags ^ old_flags;
5342 if (changes & IFF_UP) {
5343 if (dev->flags & IFF_UP)
5344 call_netdevice_notifiers(NETDEV_UP, dev);
5346 call_netdevice_notifiers(NETDEV_DOWN, dev);
5349 if (dev->flags & IFF_UP &&
5350 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
5351 call_netdevice_notifiers(NETDEV_CHANGE, dev);
5355 * dev_change_flags - change device settings
5357 * @flags: device state flags
5359 * Change settings on device based state flags. The flags are
5360 * in the userspace exported format.
5362 int dev_change_flags(struct net_device *dev, unsigned int flags)
5365 unsigned int changes, old_flags = dev->flags;
5367 ret = __dev_change_flags(dev, flags);
5371 changes = old_flags ^ dev->flags;
5373 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
5375 __dev_notify_flags(dev, old_flags);
5378 EXPORT_SYMBOL(dev_change_flags);
5381 * dev_set_mtu - Change maximum transfer unit
5383 * @new_mtu: new transfer unit
5385 * Change the maximum transfer size of the network device.
5387 int dev_set_mtu(struct net_device *dev, int new_mtu)
5389 const struct net_device_ops *ops = dev->netdev_ops;
5392 if (new_mtu == dev->mtu)
5395 /* MTU must be positive. */
5399 if (!netif_device_present(dev))
5403 if (ops->ndo_change_mtu)
5404 err = ops->ndo_change_mtu(dev, new_mtu);
5409 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5412 EXPORT_SYMBOL(dev_set_mtu);
5415 * dev_set_group - Change group this device belongs to
5417 * @new_group: group this device should belong to
5419 void dev_set_group(struct net_device *dev, int new_group)
5421 dev->group = new_group;
5423 EXPORT_SYMBOL(dev_set_group);
5426 * dev_set_mac_address - Change Media Access Control Address
5430 * Change the hardware (MAC) address of the device
5432 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5434 const struct net_device_ops *ops = dev->netdev_ops;
5437 if (!ops->ndo_set_mac_address)
5439 if (sa->sa_family != dev->type)
5441 if (!netif_device_present(dev))
5443 err = ops->ndo_set_mac_address(dev, sa);
5446 dev->addr_assign_type = NET_ADDR_SET;
5447 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5448 add_device_randomness(dev->dev_addr, dev->addr_len);
5451 EXPORT_SYMBOL(dev_set_mac_address);
5454 * dev_change_carrier - Change device carrier
5456 * @new_carries: new value
5458 * Change device carrier
5460 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5462 const struct net_device_ops *ops = dev->netdev_ops;
5464 if (!ops->ndo_change_carrier)
5466 if (!netif_device_present(dev))
5468 return ops->ndo_change_carrier(dev, new_carrier);
5470 EXPORT_SYMBOL(dev_change_carrier);
5473 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
5475 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
5478 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
5484 case SIOCGIFFLAGS: /* Get interface flags */
5485 ifr->ifr_flags = (short) dev_get_flags(dev);
5488 case SIOCGIFMETRIC: /* Get the metric on the interface
5489 (currently unused) */
5490 ifr->ifr_metric = 0;
5493 case SIOCGIFMTU: /* Get the MTU of a device */
5494 ifr->ifr_mtu = dev->mtu;
5499 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
5501 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
5502 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
5503 ifr->ifr_hwaddr.sa_family = dev->type;
5511 ifr->ifr_map.mem_start = dev->mem_start;
5512 ifr->ifr_map.mem_end = dev->mem_end;
5513 ifr->ifr_map.base_addr = dev->base_addr;
5514 ifr->ifr_map.irq = dev->irq;
5515 ifr->ifr_map.dma = dev->dma;
5516 ifr->ifr_map.port = dev->if_port;
5520 ifr->ifr_ifindex = dev->ifindex;
5524 ifr->ifr_qlen = dev->tx_queue_len;
5528 /* dev_ioctl() should ensure this case
5540 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
5542 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
5545 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
5546 const struct net_device_ops *ops;
5551 ops = dev->netdev_ops;
5554 case SIOCSIFFLAGS: /* Set interface flags */
5555 return dev_change_flags(dev, ifr->ifr_flags);
5557 case SIOCSIFMETRIC: /* Set the metric on the interface
5558 (currently unused) */
5561 case SIOCSIFMTU: /* Set the MTU of a device */
5562 return dev_set_mtu(dev, ifr->ifr_mtu);
5565 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
5567 case SIOCSIFHWBROADCAST:
5568 if (ifr->ifr_hwaddr.sa_family != dev->type)
5570 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
5571 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
5572 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5576 if (ops->ndo_set_config) {
5577 if (!netif_device_present(dev))
5579 return ops->ndo_set_config(dev, &ifr->ifr_map);
5584 if (!ops->ndo_set_rx_mode ||
5585 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5587 if (!netif_device_present(dev))
5589 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5592 if (!ops->ndo_set_rx_mode ||
5593 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5595 if (!netif_device_present(dev))
5597 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5600 if (ifr->ifr_qlen < 0)
5602 dev->tx_queue_len = ifr->ifr_qlen;
5606 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5607 return dev_change_name(dev, ifr->ifr_newname);
5610 err = net_hwtstamp_validate(ifr);
5616 * Unknown or private ioctl
5619 if ((cmd >= SIOCDEVPRIVATE &&
5620 cmd <= SIOCDEVPRIVATE + 15) ||
5621 cmd == SIOCBONDENSLAVE ||
5622 cmd == SIOCBONDRELEASE ||
5623 cmd == SIOCBONDSETHWADDR ||
5624 cmd == SIOCBONDSLAVEINFOQUERY ||
5625 cmd == SIOCBONDINFOQUERY ||
5626 cmd == SIOCBONDCHANGEACTIVE ||
5627 cmd == SIOCGMIIPHY ||
5628 cmd == SIOCGMIIREG ||
5629 cmd == SIOCSMIIREG ||
5630 cmd == SIOCBRADDIF ||
5631 cmd == SIOCBRDELIF ||
5632 cmd == SIOCSHWTSTAMP ||
5633 cmd == SIOCWANDEV) {
5635 if (ops->ndo_do_ioctl) {
5636 if (netif_device_present(dev))
5637 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5649 * This function handles all "interface"-type I/O control requests. The actual
5650 * 'doing' part of this is dev_ifsioc above.
5654 * dev_ioctl - network device ioctl
5655 * @net: the applicable net namespace
5656 * @cmd: command to issue
5657 * @arg: pointer to a struct ifreq in user space
5659 * Issue ioctl functions to devices. This is normally called by the
5660 * user space syscall interfaces but can sometimes be useful for
5661 * other purposes. The return value is the return from the syscall if
5662 * positive or a negative errno code on error.
5665 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5671 /* One special case: SIOCGIFCONF takes ifconf argument
5672 and requires shared lock, because it sleeps writing
5676 if (cmd == SIOCGIFCONF) {
5678 ret = dev_ifconf(net, (char __user *) arg);
5682 if (cmd == SIOCGIFNAME)
5683 return dev_ifname(net, (struct ifreq __user *)arg);
5685 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5688 ifr.ifr_name[IFNAMSIZ-1] = 0;
5690 colon = strchr(ifr.ifr_name, ':');
5695 * See which interface the caller is talking about.
5700 * These ioctl calls:
5701 * - can be done by all.
5702 * - atomic and do not require locking.
5713 dev_load(net, ifr.ifr_name);
5715 ret = dev_ifsioc_locked(net, &ifr, cmd);
5720 if (copy_to_user(arg, &ifr,
5721 sizeof(struct ifreq)))
5727 dev_load(net, ifr.ifr_name);
5729 ret = dev_ethtool(net, &ifr);
5734 if (copy_to_user(arg, &ifr,
5735 sizeof(struct ifreq)))
5741 * These ioctl calls:
5742 * - require superuser power.
5743 * - require strict serialization.
5749 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
5751 dev_load(net, ifr.ifr_name);
5753 ret = dev_ifsioc(net, &ifr, cmd);
5758 if (copy_to_user(arg, &ifr,
5759 sizeof(struct ifreq)))
5765 * These ioctl calls:
5766 * - require superuser power.
5767 * - require strict serialization.
5768 * - do not return a value
5772 if (!capable(CAP_NET_ADMIN))
5776 * These ioctl calls:
5777 * - require local superuser power.
5778 * - require strict serialization.
5779 * - do not return a value
5788 case SIOCSIFHWBROADCAST:
5790 case SIOCBONDENSLAVE:
5791 case SIOCBONDRELEASE:
5792 case SIOCBONDSETHWADDR:
5793 case SIOCBONDCHANGEACTIVE:
5797 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
5800 case SIOCBONDSLAVEINFOQUERY:
5801 case SIOCBONDINFOQUERY:
5802 dev_load(net, ifr.ifr_name);
5804 ret = dev_ifsioc(net, &ifr, cmd);
5809 /* Get the per device memory space. We can add this but
5810 * currently do not support it */
5812 /* Set the per device memory buffer space.
5813 * Not applicable in our case */
5818 * Unknown or private ioctl.
5821 if (cmd == SIOCWANDEV ||
5822 (cmd >= SIOCDEVPRIVATE &&
5823 cmd <= SIOCDEVPRIVATE + 15)) {
5824 dev_load(net, ifr.ifr_name);
5826 ret = dev_ifsioc(net, &ifr, cmd);
5828 if (!ret && copy_to_user(arg, &ifr,
5829 sizeof(struct ifreq)))
5833 /* Take care of Wireless Extensions */
5834 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5835 return wext_handle_ioctl(net, &ifr, cmd, arg);
5842 * dev_new_index - allocate an ifindex
5843 * @net: the applicable net namespace
5845 * Returns a suitable unique value for a new device interface
5846 * number. The caller must hold the rtnl semaphore or the
5847 * dev_base_lock to be sure it remains unique.
5849 static int dev_new_index(struct net *net)
5851 int ifindex = net->ifindex;
5855 if (!__dev_get_by_index(net, ifindex))
5856 return net->ifindex = ifindex;
5860 /* Delayed registration/unregisteration */
5861 static LIST_HEAD(net_todo_list);
5863 static void net_set_todo(struct net_device *dev)
5865 list_add_tail(&dev->todo_list, &net_todo_list);
5868 static void rollback_registered_many(struct list_head *head)
5870 struct net_device *dev, *tmp;
5872 BUG_ON(dev_boot_phase);
5875 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5876 /* Some devices call without registering
5877 * for initialization unwind. Remove those
5878 * devices and proceed with the remaining.
5880 if (dev->reg_state == NETREG_UNINITIALIZED) {
5881 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5885 list_del(&dev->unreg_list);
5888 dev->dismantle = true;
5889 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5892 /* If device is running, close it first. */
5893 dev_close_many(head);
5895 list_for_each_entry(dev, head, unreg_list) {
5896 /* And unlink it from device chain. */
5897 unlist_netdevice(dev);
5899 dev->reg_state = NETREG_UNREGISTERING;
5904 list_for_each_entry(dev, head, unreg_list) {
5905 /* Shutdown queueing discipline. */
5909 /* Notify protocols, that we are about to destroy
5910 this device. They should clean all the things.
5912 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5914 if (!dev->rtnl_link_ops ||
5915 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5916 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5919 * Flush the unicast and multicast chains
5924 if (dev->netdev_ops->ndo_uninit)
5925 dev->netdev_ops->ndo_uninit(dev);
5927 /* Notifier chain MUST detach us all upper devices. */
5928 WARN_ON(netdev_has_any_upper_dev(dev));
5930 /* Remove entries from kobject tree */
5931 netdev_unregister_kobject(dev);
5933 /* Remove XPS queueing entries */
5934 netif_reset_xps_queues_gt(dev, 0);
5940 list_for_each_entry(dev, head, unreg_list)
5944 static void rollback_registered(struct net_device *dev)
5948 list_add(&dev->unreg_list, &single);
5949 rollback_registered_many(&single);
5953 static netdev_features_t netdev_fix_features(struct net_device *dev,
5954 netdev_features_t features)
5956 /* Fix illegal checksum combinations */
5957 if ((features & NETIF_F_HW_CSUM) &&
5958 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5959 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5960 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5963 /* Fix illegal SG+CSUM combinations. */
5964 if ((features & NETIF_F_SG) &&
5965 !(features & NETIF_F_ALL_CSUM)) {
5967 "Dropping NETIF_F_SG since no checksum feature.\n");
5968 features &= ~NETIF_F_SG;
5971 /* TSO requires that SG is present as well. */
5972 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5973 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5974 features &= ~NETIF_F_ALL_TSO;
5977 /* TSO ECN requires that TSO is present as well. */
5978 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5979 features &= ~NETIF_F_TSO_ECN;
5981 /* Software GSO depends on SG. */
5982 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5983 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5984 features &= ~NETIF_F_GSO;
5987 /* UFO needs SG and checksumming */
5988 if (features & NETIF_F_UFO) {
5989 /* maybe split UFO into V4 and V6? */
5990 if (!((features & NETIF_F_GEN_CSUM) ||
5991 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5992 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5994 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5995 features &= ~NETIF_F_UFO;
5998 if (!(features & NETIF_F_SG)) {
6000 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6001 features &= ~NETIF_F_UFO;
6008 int __netdev_update_features(struct net_device *dev)
6010 netdev_features_t features;
6015 features = netdev_get_wanted_features(dev);
6017 if (dev->netdev_ops->ndo_fix_features)
6018 features = dev->netdev_ops->ndo_fix_features(dev, features);
6020 /* driver might be less strict about feature dependencies */
6021 features = netdev_fix_features(dev, features);
6023 if (dev->features == features)
6026 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6027 &dev->features, &features);
6029 if (dev->netdev_ops->ndo_set_features)
6030 err = dev->netdev_ops->ndo_set_features(dev, features);
6032 if (unlikely(err < 0)) {
6034 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6035 err, &features, &dev->features);
6040 dev->features = features;
6046 * netdev_update_features - recalculate device features
6047 * @dev: the device to check
6049 * Recalculate dev->features set and send notifications if it
6050 * has changed. Should be called after driver or hardware dependent
6051 * conditions might have changed that influence the features.
6053 void netdev_update_features(struct net_device *dev)
6055 if (__netdev_update_features(dev))
6056 netdev_features_change(dev);
6058 EXPORT_SYMBOL(netdev_update_features);
6061 * netdev_change_features - recalculate device features
6062 * @dev: the device to check
6064 * Recalculate dev->features set and send notifications even
6065 * if they have not changed. Should be called instead of
6066 * netdev_update_features() if also dev->vlan_features might
6067 * have changed to allow the changes to be propagated to stacked
6070 void netdev_change_features(struct net_device *dev)
6072 __netdev_update_features(dev);
6073 netdev_features_change(dev);
6075 EXPORT_SYMBOL(netdev_change_features);
6078 * netif_stacked_transfer_operstate - transfer operstate
6079 * @rootdev: the root or lower level device to transfer state from
6080 * @dev: the device to transfer operstate to
6082 * Transfer operational state from root to device. This is normally
6083 * called when a stacking relationship exists between the root
6084 * device and the device(a leaf device).
6086 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6087 struct net_device *dev)
6089 if (rootdev->operstate == IF_OPER_DORMANT)
6090 netif_dormant_on(dev);
6092 netif_dormant_off(dev);
6094 if (netif_carrier_ok(rootdev)) {
6095 if (!netif_carrier_ok(dev))
6096 netif_carrier_on(dev);
6098 if (netif_carrier_ok(dev))
6099 netif_carrier_off(dev);
6102 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6105 static int netif_alloc_rx_queues(struct net_device *dev)
6107 unsigned int i, count = dev->num_rx_queues;
6108 struct netdev_rx_queue *rx;
6112 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
6114 pr_err("netdev: Unable to allocate %u rx queues\n", count);
6119 for (i = 0; i < count; i++)
6125 static void netdev_init_one_queue(struct net_device *dev,
6126 struct netdev_queue *queue, void *_unused)
6128 /* Initialize queue lock */
6129 spin_lock_init(&queue->_xmit_lock);
6130 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6131 queue->xmit_lock_owner = -1;
6132 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6135 dql_init(&queue->dql, HZ);
6139 static int netif_alloc_netdev_queues(struct net_device *dev)
6141 unsigned int count = dev->num_tx_queues;
6142 struct netdev_queue *tx;
6146 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
6148 pr_err("netdev: Unable to allocate %u tx queues\n", count);
6153 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6154 spin_lock_init(&dev->tx_global_lock);
6160 * register_netdevice - register a network device
6161 * @dev: device to register
6163 * Take a completed network device structure and add it to the kernel
6164 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6165 * chain. 0 is returned on success. A negative errno code is returned
6166 * on a failure to set up the device, or if the name is a duplicate.
6168 * Callers must hold the rtnl semaphore. You may want
6169 * register_netdev() instead of this.
6172 * The locking appears insufficient to guarantee two parallel registers
6173 * will not get the same name.
6176 int register_netdevice(struct net_device *dev)
6179 struct net *net = dev_net(dev);
6181 BUG_ON(dev_boot_phase);
6186 /* When net_device's are persistent, this will be fatal. */
6187 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6190 spin_lock_init(&dev->addr_list_lock);
6191 netdev_set_addr_lockdep_class(dev);
6195 ret = dev_get_valid_name(net, dev, dev->name);
6199 /* Init, if this function is available */
6200 if (dev->netdev_ops->ndo_init) {
6201 ret = dev->netdev_ops->ndo_init(dev);
6211 dev->ifindex = dev_new_index(net);
6212 else if (__dev_get_by_index(net, dev->ifindex))
6215 if (dev->iflink == -1)
6216 dev->iflink = dev->ifindex;
6218 /* Transfer changeable features to wanted_features and enable
6219 * software offloads (GSO and GRO).
6221 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6222 dev->features |= NETIF_F_SOFT_FEATURES;
6223 dev->wanted_features = dev->features & dev->hw_features;
6225 /* Turn on no cache copy if HW is doing checksum */
6226 if (!(dev->flags & IFF_LOOPBACK)) {
6227 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6228 if (dev->features & NETIF_F_ALL_CSUM) {
6229 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
6230 dev->features |= NETIF_F_NOCACHE_COPY;
6234 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6236 dev->vlan_features |= NETIF_F_HIGHDMA;
6238 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6239 ret = notifier_to_errno(ret);
6243 ret = netdev_register_kobject(dev);
6246 dev->reg_state = NETREG_REGISTERED;
6248 __netdev_update_features(dev);
6251 * Default initial state at registry is that the
6252 * device is present.
6255 set_bit(__LINK_STATE_PRESENT, &dev->state);
6257 linkwatch_init_dev(dev);
6259 dev_init_scheduler(dev);
6261 list_netdevice(dev);
6262 add_device_randomness(dev->dev_addr, dev->addr_len);
6264 /* If the device has permanent device address, driver should
6265 * set dev_addr and also addr_assign_type should be set to
6266 * NET_ADDR_PERM (default value).
6268 if (dev->addr_assign_type == NET_ADDR_PERM)
6269 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6271 /* Notify protocols, that a new device appeared. */
6272 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6273 ret = notifier_to_errno(ret);
6275 rollback_registered(dev);
6276 dev->reg_state = NETREG_UNREGISTERED;
6279 * Prevent userspace races by waiting until the network
6280 * device is fully setup before sending notifications.
6282 if (!dev->rtnl_link_ops ||
6283 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6284 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6290 if (dev->netdev_ops->ndo_uninit)
6291 dev->netdev_ops->ndo_uninit(dev);
6294 EXPORT_SYMBOL(register_netdevice);
6297 * init_dummy_netdev - init a dummy network device for NAPI
6298 * @dev: device to init
6300 * This takes a network device structure and initialize the minimum
6301 * amount of fields so it can be used to schedule NAPI polls without
6302 * registering a full blown interface. This is to be used by drivers
6303 * that need to tie several hardware interfaces to a single NAPI
6304 * poll scheduler due to HW limitations.
6306 int init_dummy_netdev(struct net_device *dev)
6308 /* Clear everything. Note we don't initialize spinlocks
6309 * are they aren't supposed to be taken by any of the
6310 * NAPI code and this dummy netdev is supposed to be
6311 * only ever used for NAPI polls
6313 memset(dev, 0, sizeof(struct net_device));
6315 /* make sure we BUG if trying to hit standard
6316 * register/unregister code path
6318 dev->reg_state = NETREG_DUMMY;
6320 /* NAPI wants this */
6321 INIT_LIST_HEAD(&dev->napi_list);
6323 /* a dummy interface is started by default */
6324 set_bit(__LINK_STATE_PRESENT, &dev->state);
6325 set_bit(__LINK_STATE_START, &dev->state);
6327 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6328 * because users of this 'device' dont need to change
6334 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6338 * register_netdev - register a network device
6339 * @dev: device to register
6341 * Take a completed network device structure and add it to the kernel
6342 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6343 * chain. 0 is returned on success. A negative errno code is returned
6344 * on a failure to set up the device, or if the name is a duplicate.
6346 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6347 * and expands the device name if you passed a format string to
6350 int register_netdev(struct net_device *dev)
6355 err = register_netdevice(dev);
6359 EXPORT_SYMBOL(register_netdev);
6361 int netdev_refcnt_read(const struct net_device *dev)
6365 for_each_possible_cpu(i)
6366 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6369 EXPORT_SYMBOL(netdev_refcnt_read);
6372 * netdev_wait_allrefs - wait until all references are gone.
6373 * @dev: target net_device
6375 * This is called when unregistering network devices.
6377 * Any protocol or device that holds a reference should register
6378 * for netdevice notification, and cleanup and put back the
6379 * reference if they receive an UNREGISTER event.
6380 * We can get stuck here if buggy protocols don't correctly
6383 static void netdev_wait_allrefs(struct net_device *dev)
6385 unsigned long rebroadcast_time, warning_time;
6388 linkwatch_forget_dev(dev);
6390 rebroadcast_time = warning_time = jiffies;
6391 refcnt = netdev_refcnt_read(dev);
6393 while (refcnt != 0) {
6394 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6397 /* Rebroadcast unregister notification */
6398 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6404 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6405 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6407 /* We must not have linkwatch events
6408 * pending on unregister. If this
6409 * happens, we simply run the queue
6410 * unscheduled, resulting in a noop
6413 linkwatch_run_queue();
6418 rebroadcast_time = jiffies;
6423 refcnt = netdev_refcnt_read(dev);
6425 if (time_after(jiffies, warning_time + 10 * HZ)) {
6426 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6428 warning_time = jiffies;
6437 * register_netdevice(x1);
6438 * register_netdevice(x2);
6440 * unregister_netdevice(y1);
6441 * unregister_netdevice(y2);
6447 * We are invoked by rtnl_unlock().
6448 * This allows us to deal with problems:
6449 * 1) We can delete sysfs objects which invoke hotplug
6450 * without deadlocking with linkwatch via keventd.
6451 * 2) Since we run with the RTNL semaphore not held, we can sleep
6452 * safely in order to wait for the netdev refcnt to drop to zero.
6454 * We must not return until all unregister events added during
6455 * the interval the lock was held have been completed.
6457 void netdev_run_todo(void)
6459 struct list_head list;
6461 /* Snapshot list, allow later requests */
6462 list_replace_init(&net_todo_list, &list);
6467 /* Wait for rcu callbacks to finish before next phase */
6468 if (!list_empty(&list))
6471 while (!list_empty(&list)) {
6472 struct net_device *dev
6473 = list_first_entry(&list, struct net_device, todo_list);
6474 list_del(&dev->todo_list);
6477 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6480 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6481 pr_err("network todo '%s' but state %d\n",
6482 dev->name, dev->reg_state);
6487 dev->reg_state = NETREG_UNREGISTERED;
6489 on_each_cpu(flush_backlog, dev, 1);
6491 netdev_wait_allrefs(dev);
6494 BUG_ON(netdev_refcnt_read(dev));
6495 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6496 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6497 WARN_ON(dev->dn_ptr);
6499 if (dev->destructor)
6500 dev->destructor(dev);
6502 /* Free network device */
6503 kobject_put(&dev->dev.kobj);
6507 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6508 * fields in the same order, with only the type differing.
6510 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6511 const struct net_device_stats *netdev_stats)
6513 #if BITS_PER_LONG == 64
6514 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6515 memcpy(stats64, netdev_stats, sizeof(*stats64));
6517 size_t i, n = sizeof(*stats64) / sizeof(u64);
6518 const unsigned long *src = (const unsigned long *)netdev_stats;
6519 u64 *dst = (u64 *)stats64;
6521 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6522 sizeof(*stats64) / sizeof(u64));
6523 for (i = 0; i < n; i++)
6527 EXPORT_SYMBOL(netdev_stats_to_stats64);
6530 * dev_get_stats - get network device statistics
6531 * @dev: device to get statistics from
6532 * @storage: place to store stats
6534 * Get network statistics from device. Return @storage.
6535 * The device driver may provide its own method by setting
6536 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6537 * otherwise the internal statistics structure is used.
6539 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6540 struct rtnl_link_stats64 *storage)
6542 const struct net_device_ops *ops = dev->netdev_ops;
6544 if (ops->ndo_get_stats64) {
6545 memset(storage, 0, sizeof(*storage));
6546 ops->ndo_get_stats64(dev, storage);
6547 } else if (ops->ndo_get_stats) {
6548 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6550 netdev_stats_to_stats64(storage, &dev->stats);
6552 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6555 EXPORT_SYMBOL(dev_get_stats);
6557 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6559 struct netdev_queue *queue = dev_ingress_queue(dev);
6561 #ifdef CONFIG_NET_CLS_ACT
6564 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6567 netdev_init_one_queue(dev, queue, NULL);
6568 queue->qdisc = &noop_qdisc;
6569 queue->qdisc_sleeping = &noop_qdisc;
6570 rcu_assign_pointer(dev->ingress_queue, queue);
6575 static const struct ethtool_ops default_ethtool_ops;
6578 * alloc_netdev_mqs - allocate network device
6579 * @sizeof_priv: size of private data to allocate space for
6580 * @name: device name format string
6581 * @setup: callback to initialize device
6582 * @txqs: the number of TX subqueues to allocate
6583 * @rxqs: the number of RX subqueues to allocate
6585 * Allocates a struct net_device with private data area for driver use
6586 * and performs basic initialization. Also allocates subquue structs
6587 * for each queue on the device.
6589 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6590 void (*setup)(struct net_device *),
6591 unsigned int txqs, unsigned int rxqs)
6593 struct net_device *dev;
6595 struct net_device *p;
6597 BUG_ON(strlen(name) >= sizeof(dev->name));
6600 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6606 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6611 alloc_size = sizeof(struct net_device);
6613 /* ensure 32-byte alignment of private area */
6614 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6615 alloc_size += sizeof_priv;
6617 /* ensure 32-byte alignment of whole construct */
6618 alloc_size += NETDEV_ALIGN - 1;
6620 p = kzalloc(alloc_size, GFP_KERNEL);
6622 pr_err("alloc_netdev: Unable to allocate device\n");
6626 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6627 dev->padded = (char *)dev - (char *)p;
6629 dev->pcpu_refcnt = alloc_percpu(int);
6630 if (!dev->pcpu_refcnt)
6633 if (dev_addr_init(dev))
6639 dev_net_set(dev, &init_net);
6641 dev->gso_max_size = GSO_MAX_SIZE;
6642 dev->gso_max_segs = GSO_MAX_SEGS;
6644 INIT_LIST_HEAD(&dev->napi_list);
6645 INIT_LIST_HEAD(&dev->unreg_list);
6646 INIT_LIST_HEAD(&dev->link_watch_list);
6647 INIT_LIST_HEAD(&dev->upper_dev_list);
6648 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6651 dev->num_tx_queues = txqs;
6652 dev->real_num_tx_queues = txqs;
6653 if (netif_alloc_netdev_queues(dev))
6657 dev->num_rx_queues = rxqs;
6658 dev->real_num_rx_queues = rxqs;
6659 if (netif_alloc_rx_queues(dev))
6663 strcpy(dev->name, name);
6664 dev->group = INIT_NETDEV_GROUP;
6665 if (!dev->ethtool_ops)
6666 dev->ethtool_ops = &default_ethtool_ops;
6674 free_percpu(dev->pcpu_refcnt);
6684 EXPORT_SYMBOL(alloc_netdev_mqs);
6687 * free_netdev - free network device
6690 * This function does the last stage of destroying an allocated device
6691 * interface. The reference to the device object is released.
6692 * If this is the last reference then it will be freed.
6694 void free_netdev(struct net_device *dev)
6696 struct napi_struct *p, *n;
6698 release_net(dev_net(dev));
6705 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6707 /* Flush device addresses */
6708 dev_addr_flush(dev);
6710 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6713 free_percpu(dev->pcpu_refcnt);
6714 dev->pcpu_refcnt = NULL;
6716 /* Compatibility with error handling in drivers */
6717 if (dev->reg_state == NETREG_UNINITIALIZED) {
6718 kfree((char *)dev - dev->padded);
6722 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6723 dev->reg_state = NETREG_RELEASED;
6725 /* will free via device release */
6726 put_device(&dev->dev);
6728 EXPORT_SYMBOL(free_netdev);
6731 * synchronize_net - Synchronize with packet receive processing
6733 * Wait for packets currently being received to be done.
6734 * Does not block later packets from starting.
6736 void synchronize_net(void)
6739 if (rtnl_is_locked())
6740 synchronize_rcu_expedited();
6744 EXPORT_SYMBOL(synchronize_net);
6747 * unregister_netdevice_queue - remove device from the kernel
6751 * This function shuts down a device interface and removes it
6752 * from the kernel tables.
6753 * If head not NULL, device is queued to be unregistered later.
6755 * Callers must hold the rtnl semaphore. You may want
6756 * unregister_netdev() instead of this.
6759 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6764 list_move_tail(&dev->unreg_list, head);
6766 rollback_registered(dev);
6767 /* Finish processing unregister after unlock */
6771 EXPORT_SYMBOL(unregister_netdevice_queue);
6774 * unregister_netdevice_many - unregister many devices
6775 * @head: list of devices
6777 void unregister_netdevice_many(struct list_head *head)
6779 struct net_device *dev;
6781 if (!list_empty(head)) {
6782 rollback_registered_many(head);
6783 list_for_each_entry(dev, head, unreg_list)
6787 EXPORT_SYMBOL(unregister_netdevice_many);
6790 * unregister_netdev - remove device from the kernel
6793 * This function shuts down a device interface and removes it
6794 * from the kernel tables.
6796 * This is just a wrapper for unregister_netdevice that takes
6797 * the rtnl semaphore. In general you want to use this and not
6798 * unregister_netdevice.
6800 void unregister_netdev(struct net_device *dev)
6803 unregister_netdevice(dev);
6806 EXPORT_SYMBOL(unregister_netdev);
6809 * dev_change_net_namespace - move device to different nethost namespace
6811 * @net: network namespace
6812 * @pat: If not NULL name pattern to try if the current device name
6813 * is already taken in the destination network namespace.
6815 * This function shuts down a device interface and moves it
6816 * to a new network namespace. On success 0 is returned, on
6817 * a failure a netagive errno code is returned.
6819 * Callers must hold the rtnl semaphore.
6822 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6828 /* Don't allow namespace local devices to be moved. */
6830 if (dev->features & NETIF_F_NETNS_LOCAL)
6833 /* Ensure the device has been registrered */
6834 if (dev->reg_state != NETREG_REGISTERED)
6837 /* Get out if there is nothing todo */
6839 if (net_eq(dev_net(dev), net))
6842 /* Pick the destination device name, and ensure
6843 * we can use it in the destination network namespace.
6846 if (__dev_get_by_name(net, dev->name)) {
6847 /* We get here if we can't use the current device name */
6850 if (dev_get_valid_name(net, dev, pat) < 0)
6855 * And now a mini version of register_netdevice unregister_netdevice.
6858 /* If device is running close it first. */
6861 /* And unlink it from device chain */
6863 unlist_netdevice(dev);
6867 /* Shutdown queueing discipline. */
6870 /* Notify protocols, that we are about to destroy
6871 this device. They should clean all the things.
6873 Note that dev->reg_state stays at NETREG_REGISTERED.
6874 This is wanted because this way 8021q and macvlan know
6875 the device is just moving and can keep their slaves up.
6877 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6879 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6880 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6883 * Flush the unicast and multicast chains
6888 /* Send a netdev-removed uevent to the old namespace */
6889 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6891 /* Actually switch the network namespace */
6892 dev_net_set(dev, net);
6894 /* If there is an ifindex conflict assign a new one */
6895 if (__dev_get_by_index(net, dev->ifindex)) {
6896 int iflink = (dev->iflink == dev->ifindex);
6897 dev->ifindex = dev_new_index(net);
6899 dev->iflink = dev->ifindex;
6902 /* Send a netdev-add uevent to the new namespace */
6903 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6905 /* Fixup kobjects */
6906 err = device_rename(&dev->dev, dev->name);
6909 /* Add the device back in the hashes */
6910 list_netdevice(dev);
6912 /* Notify protocols, that a new device appeared. */
6913 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6916 * Prevent userspace races by waiting until the network
6917 * device is fully setup before sending notifications.
6919 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6926 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6928 static int dev_cpu_callback(struct notifier_block *nfb,
6929 unsigned long action,
6932 struct sk_buff **list_skb;
6933 struct sk_buff *skb;
6934 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6935 struct softnet_data *sd, *oldsd;
6937 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6940 local_irq_disable();
6941 cpu = smp_processor_id();
6942 sd = &per_cpu(softnet_data, cpu);
6943 oldsd = &per_cpu(softnet_data, oldcpu);
6945 /* Find end of our completion_queue. */
6946 list_skb = &sd->completion_queue;
6948 list_skb = &(*list_skb)->next;
6949 /* Append completion queue from offline CPU. */
6950 *list_skb = oldsd->completion_queue;
6951 oldsd->completion_queue = NULL;
6953 /* Append output queue from offline CPU. */
6954 if (oldsd->output_queue) {
6955 *sd->output_queue_tailp = oldsd->output_queue;
6956 sd->output_queue_tailp = oldsd->output_queue_tailp;
6957 oldsd->output_queue = NULL;
6958 oldsd->output_queue_tailp = &oldsd->output_queue;
6960 /* Append NAPI poll list from offline CPU. */
6961 if (!list_empty(&oldsd->poll_list)) {
6962 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6963 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6966 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6969 /* Process offline CPU's input_pkt_queue */
6970 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6972 input_queue_head_incr(oldsd);
6974 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6976 input_queue_head_incr(oldsd);
6984 * netdev_increment_features - increment feature set by one
6985 * @all: current feature set
6986 * @one: new feature set
6987 * @mask: mask feature set
6989 * Computes a new feature set after adding a device with feature set
6990 * @one to the master device with current feature set @all. Will not
6991 * enable anything that is off in @mask. Returns the new feature set.
6993 netdev_features_t netdev_increment_features(netdev_features_t all,
6994 netdev_features_t one, netdev_features_t mask)
6996 if (mask & NETIF_F_GEN_CSUM)
6997 mask |= NETIF_F_ALL_CSUM;
6998 mask |= NETIF_F_VLAN_CHALLENGED;
7000 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7001 all &= one | ~NETIF_F_ALL_FOR_ALL;
7003 /* If one device supports hw checksumming, set for all. */
7004 if (all & NETIF_F_GEN_CSUM)
7005 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7009 EXPORT_SYMBOL(netdev_increment_features);
7011 static struct hlist_head *netdev_create_hash(void)
7014 struct hlist_head *hash;
7016 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7018 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7019 INIT_HLIST_HEAD(&hash[i]);
7024 /* Initialize per network namespace state */
7025 static int __net_init netdev_init(struct net *net)
7027 if (net != &init_net)
7028 INIT_LIST_HEAD(&net->dev_base_head);
7030 net->dev_name_head = netdev_create_hash();
7031 if (net->dev_name_head == NULL)
7034 net->dev_index_head = netdev_create_hash();
7035 if (net->dev_index_head == NULL)
7041 kfree(net->dev_name_head);
7047 * netdev_drivername - network driver for the device
7048 * @dev: network device
7050 * Determine network driver for device.
7052 const char *netdev_drivername(const struct net_device *dev)
7054 const struct device_driver *driver;
7055 const struct device *parent;
7056 const char *empty = "";
7058 parent = dev->dev.parent;
7062 driver = parent->driver;
7063 if (driver && driver->name)
7064 return driver->name;
7068 static int __netdev_printk(const char *level, const struct net_device *dev,
7069 struct va_format *vaf)
7073 if (dev && dev->dev.parent) {
7074 r = dev_printk_emit(level[1] - '0',
7077 dev_driver_string(dev->dev.parent),
7078 dev_name(dev->dev.parent),
7079 netdev_name(dev), vaf);
7081 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
7083 r = printk("%s(NULL net_device): %pV", level, vaf);
7089 int netdev_printk(const char *level, const struct net_device *dev,
7090 const char *format, ...)
7092 struct va_format vaf;
7096 va_start(args, format);
7101 r = __netdev_printk(level, dev, &vaf);
7107 EXPORT_SYMBOL(netdev_printk);
7109 #define define_netdev_printk_level(func, level) \
7110 int func(const struct net_device *dev, const char *fmt, ...) \
7113 struct va_format vaf; \
7116 va_start(args, fmt); \
7121 r = __netdev_printk(level, dev, &vaf); \
7127 EXPORT_SYMBOL(func);
7129 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7130 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7131 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7132 define_netdev_printk_level(netdev_err, KERN_ERR);
7133 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7134 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7135 define_netdev_printk_level(netdev_info, KERN_INFO);
7137 static void __net_exit netdev_exit(struct net *net)
7139 kfree(net->dev_name_head);
7140 kfree(net->dev_index_head);
7143 static struct pernet_operations __net_initdata netdev_net_ops = {
7144 .init = netdev_init,
7145 .exit = netdev_exit,
7148 static void __net_exit default_device_exit(struct net *net)
7150 struct net_device *dev, *aux;
7152 * Push all migratable network devices back to the
7153 * initial network namespace
7156 for_each_netdev_safe(net, dev, aux) {
7158 char fb_name[IFNAMSIZ];
7160 /* Ignore unmoveable devices (i.e. loopback) */
7161 if (dev->features & NETIF_F_NETNS_LOCAL)
7164 /* Leave virtual devices for the generic cleanup */
7165 if (dev->rtnl_link_ops)
7168 /* Push remaining network devices to init_net */
7169 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7170 err = dev_change_net_namespace(dev, &init_net, fb_name);
7172 pr_emerg("%s: failed to move %s to init_net: %d\n",
7173 __func__, dev->name, err);
7180 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7182 /* At exit all network devices most be removed from a network
7183 * namespace. Do this in the reverse order of registration.
7184 * Do this across as many network namespaces as possible to
7185 * improve batching efficiency.
7187 struct net_device *dev;
7189 LIST_HEAD(dev_kill_list);
7192 list_for_each_entry(net, net_list, exit_list) {
7193 for_each_netdev_reverse(net, dev) {
7194 if (dev->rtnl_link_ops)
7195 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7197 unregister_netdevice_queue(dev, &dev_kill_list);
7200 unregister_netdevice_many(&dev_kill_list);
7201 list_del(&dev_kill_list);
7205 static struct pernet_operations __net_initdata default_device_ops = {
7206 .exit = default_device_exit,
7207 .exit_batch = default_device_exit_batch,
7211 * Initialize the DEV module. At boot time this walks the device list and
7212 * unhooks any devices that fail to initialise (normally hardware not
7213 * present) and leaves us with a valid list of present and active devices.
7218 * This is called single threaded during boot, so no need
7219 * to take the rtnl semaphore.
7221 static int __init net_dev_init(void)
7223 int i, rc = -ENOMEM;
7225 BUG_ON(!dev_boot_phase);
7227 if (dev_proc_init())
7230 if (netdev_kobject_init())
7233 INIT_LIST_HEAD(&ptype_all);
7234 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7235 INIT_LIST_HEAD(&ptype_base[i]);
7237 INIT_LIST_HEAD(&offload_base);
7239 if (register_pernet_subsys(&netdev_net_ops))
7243 * Initialise the packet receive queues.
7246 for_each_possible_cpu(i) {
7247 struct softnet_data *sd = &per_cpu(softnet_data, i);
7249 memset(sd, 0, sizeof(*sd));
7250 skb_queue_head_init(&sd->input_pkt_queue);
7251 skb_queue_head_init(&sd->process_queue);
7252 sd->completion_queue = NULL;
7253 INIT_LIST_HEAD(&sd->poll_list);
7254 sd->output_queue = NULL;
7255 sd->output_queue_tailp = &sd->output_queue;
7257 sd->csd.func = rps_trigger_softirq;
7263 sd->backlog.poll = process_backlog;
7264 sd->backlog.weight = weight_p;
7265 sd->backlog.gro_list = NULL;
7266 sd->backlog.gro_count = 0;
7271 /* The loopback device is special if any other network devices
7272 * is present in a network namespace the loopback device must
7273 * be present. Since we now dynamically allocate and free the
7274 * loopback device ensure this invariant is maintained by
7275 * keeping the loopback device as the first device on the
7276 * list of network devices. Ensuring the loopback devices
7277 * is the first device that appears and the last network device
7280 if (register_pernet_device(&loopback_net_ops))
7283 if (register_pernet_device(&default_device_ops))
7286 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7287 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7289 hotcpu_notifier(dev_cpu_callback, 0);
7297 subsys_initcall(net_dev_init);
7299 static int __init initialize_hashrnd(void)
7301 get_random_bytes(&hashrnd, sizeof(hashrnd));
7305 late_initcall_sync(initialize_hashrnd);