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 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1862 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1864 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1868 if (txq < 1 || txq > dev->num_tx_queues)
1871 if (dev->reg_state == NETREG_REGISTERED ||
1872 dev->reg_state == NETREG_UNREGISTERING) {
1875 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1881 netif_setup_tc(dev, txq);
1883 if (txq < dev->real_num_tx_queues)
1884 qdisc_reset_all_tx_gt(dev, txq);
1887 dev->real_num_tx_queues = txq;
1890 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1894 * netif_set_real_num_rx_queues - set actual number of RX queues used
1895 * @dev: Network device
1896 * @rxq: Actual number of RX queues
1898 * This must be called either with the rtnl_lock held or before
1899 * registration of the net device. Returns 0 on success, or a
1900 * negative error code. If called before registration, it always
1903 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1907 if (rxq < 1 || rxq > dev->num_rx_queues)
1910 if (dev->reg_state == NETREG_REGISTERED) {
1913 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1919 dev->real_num_rx_queues = rxq;
1922 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1926 * netif_get_num_default_rss_queues - default number of RSS queues
1928 * This routine should set an upper limit on the number of RSS queues
1929 * used by default by multiqueue devices.
1931 int netif_get_num_default_rss_queues(void)
1933 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
1935 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
1937 static inline void __netif_reschedule(struct Qdisc *q)
1939 struct softnet_data *sd;
1940 unsigned long flags;
1942 local_irq_save(flags);
1943 sd = &__get_cpu_var(softnet_data);
1944 q->next_sched = NULL;
1945 *sd->output_queue_tailp = q;
1946 sd->output_queue_tailp = &q->next_sched;
1947 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1948 local_irq_restore(flags);
1951 void __netif_schedule(struct Qdisc *q)
1953 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1954 __netif_reschedule(q);
1956 EXPORT_SYMBOL(__netif_schedule);
1958 void dev_kfree_skb_irq(struct sk_buff *skb)
1960 if (atomic_dec_and_test(&skb->users)) {
1961 struct softnet_data *sd;
1962 unsigned long flags;
1964 local_irq_save(flags);
1965 sd = &__get_cpu_var(softnet_data);
1966 skb->next = sd->completion_queue;
1967 sd->completion_queue = skb;
1968 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1969 local_irq_restore(flags);
1972 EXPORT_SYMBOL(dev_kfree_skb_irq);
1974 void dev_kfree_skb_any(struct sk_buff *skb)
1976 if (in_irq() || irqs_disabled())
1977 dev_kfree_skb_irq(skb);
1981 EXPORT_SYMBOL(dev_kfree_skb_any);
1985 * netif_device_detach - mark device as removed
1986 * @dev: network device
1988 * Mark device as removed from system and therefore no longer available.
1990 void netif_device_detach(struct net_device *dev)
1992 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1993 netif_running(dev)) {
1994 netif_tx_stop_all_queues(dev);
1997 EXPORT_SYMBOL(netif_device_detach);
2000 * netif_device_attach - mark device as attached
2001 * @dev: network device
2003 * Mark device as attached from system and restart if needed.
2005 void netif_device_attach(struct net_device *dev)
2007 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2008 netif_running(dev)) {
2009 netif_tx_wake_all_queues(dev);
2010 __netdev_watchdog_up(dev);
2013 EXPORT_SYMBOL(netif_device_attach);
2015 static void skb_warn_bad_offload(const struct sk_buff *skb)
2017 static const netdev_features_t null_features = 0;
2018 struct net_device *dev = skb->dev;
2019 const char *driver = "";
2021 if (dev && dev->dev.parent)
2022 driver = dev_driver_string(dev->dev.parent);
2024 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2025 "gso_type=%d ip_summed=%d\n",
2026 driver, dev ? &dev->features : &null_features,
2027 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2028 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2029 skb_shinfo(skb)->gso_type, skb->ip_summed);
2033 * Invalidate hardware checksum when packet is to be mangled, and
2034 * complete checksum manually on outgoing path.
2036 int skb_checksum_help(struct sk_buff *skb)
2039 int ret = 0, offset;
2041 if (skb->ip_summed == CHECKSUM_COMPLETE)
2042 goto out_set_summed;
2044 if (unlikely(skb_shinfo(skb)->gso_size)) {
2045 skb_warn_bad_offload(skb);
2049 offset = skb_checksum_start_offset(skb);
2050 BUG_ON(offset >= skb_headlen(skb));
2051 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2053 offset += skb->csum_offset;
2054 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2056 if (skb_cloned(skb) &&
2057 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2058 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2063 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2065 skb->ip_summed = CHECKSUM_NONE;
2069 EXPORT_SYMBOL(skb_checksum_help);
2072 * skb_gso_segment - Perform segmentation on skb.
2073 * @skb: buffer to segment
2074 * @features: features for the output path (see dev->features)
2076 * This function segments the given skb and returns a list of segments.
2078 * It may return NULL if the skb requires no segmentation. This is
2079 * only possible when GSO is used for verifying header integrity.
2081 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
2082 netdev_features_t features)
2084 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2085 struct packet_offload *ptype;
2086 __be16 type = skb->protocol;
2087 int vlan_depth = ETH_HLEN;
2090 while (type == htons(ETH_P_8021Q)) {
2091 struct vlan_hdr *vh;
2093 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2094 return ERR_PTR(-EINVAL);
2096 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2097 type = vh->h_vlan_encapsulated_proto;
2098 vlan_depth += VLAN_HLEN;
2101 skb_reset_mac_header(skb);
2102 skb->mac_len = skb->network_header - skb->mac_header;
2103 __skb_pull(skb, skb->mac_len);
2105 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2106 skb_warn_bad_offload(skb);
2108 if (skb_header_cloned(skb) &&
2109 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2110 return ERR_PTR(err);
2114 list_for_each_entry_rcu(ptype, &offload_base, list) {
2115 if (ptype->type == type && ptype->callbacks.gso_segment) {
2116 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2117 err = ptype->callbacks.gso_send_check(skb);
2118 segs = ERR_PTR(err);
2119 if (err || skb_gso_ok(skb, features))
2121 __skb_push(skb, (skb->data -
2122 skb_network_header(skb)));
2124 segs = ptype->callbacks.gso_segment(skb, features);
2130 __skb_push(skb, skb->data - skb_mac_header(skb));
2134 EXPORT_SYMBOL(skb_gso_segment);
2136 /* Take action when hardware reception checksum errors are detected. */
2138 void netdev_rx_csum_fault(struct net_device *dev)
2140 if (net_ratelimit()) {
2141 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2145 EXPORT_SYMBOL(netdev_rx_csum_fault);
2148 /* Actually, we should eliminate this check as soon as we know, that:
2149 * 1. IOMMU is present and allows to map all the memory.
2150 * 2. No high memory really exists on this machine.
2153 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2155 #ifdef CONFIG_HIGHMEM
2157 if (!(dev->features & NETIF_F_HIGHDMA)) {
2158 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2159 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2160 if (PageHighMem(skb_frag_page(frag)))
2165 if (PCI_DMA_BUS_IS_PHYS) {
2166 struct device *pdev = dev->dev.parent;
2170 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2171 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2172 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2173 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2182 void (*destructor)(struct sk_buff *skb);
2185 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2187 static void dev_gso_skb_destructor(struct sk_buff *skb)
2189 struct dev_gso_cb *cb;
2192 struct sk_buff *nskb = skb->next;
2194 skb->next = nskb->next;
2197 } while (skb->next);
2199 cb = DEV_GSO_CB(skb);
2201 cb->destructor(skb);
2205 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2206 * @skb: buffer to segment
2207 * @features: device features as applicable to this skb
2209 * This function segments the given skb and stores the list of segments
2212 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2214 struct sk_buff *segs;
2216 segs = skb_gso_segment(skb, features);
2218 /* Verifying header integrity only. */
2223 return PTR_ERR(segs);
2226 DEV_GSO_CB(skb)->destructor = skb->destructor;
2227 skb->destructor = dev_gso_skb_destructor;
2232 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2234 return ((features & NETIF_F_GEN_CSUM) ||
2235 ((features & NETIF_F_V4_CSUM) &&
2236 protocol == htons(ETH_P_IP)) ||
2237 ((features & NETIF_F_V6_CSUM) &&
2238 protocol == htons(ETH_P_IPV6)) ||
2239 ((features & NETIF_F_FCOE_CRC) &&
2240 protocol == htons(ETH_P_FCOE)));
2243 static netdev_features_t harmonize_features(struct sk_buff *skb,
2244 __be16 protocol, netdev_features_t features)
2246 if (skb->ip_summed != CHECKSUM_NONE &&
2247 !can_checksum_protocol(features, protocol)) {
2248 features &= ~NETIF_F_ALL_CSUM;
2249 features &= ~NETIF_F_SG;
2250 } else if (illegal_highdma(skb->dev, skb)) {
2251 features &= ~NETIF_F_SG;
2257 netdev_features_t netif_skb_features(struct sk_buff *skb)
2259 __be16 protocol = skb->protocol;
2260 netdev_features_t features = skb->dev->features;
2262 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2263 features &= ~NETIF_F_GSO_MASK;
2265 if (protocol == htons(ETH_P_8021Q)) {
2266 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2267 protocol = veh->h_vlan_encapsulated_proto;
2268 } else if (!vlan_tx_tag_present(skb)) {
2269 return harmonize_features(skb, protocol, features);
2272 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2274 if (protocol != htons(ETH_P_8021Q)) {
2275 return harmonize_features(skb, protocol, features);
2277 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2278 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2279 return harmonize_features(skb, protocol, features);
2282 EXPORT_SYMBOL(netif_skb_features);
2285 * Returns true if either:
2286 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2287 * 2. skb is fragmented and the device does not support SG.
2289 static inline int skb_needs_linearize(struct sk_buff *skb,
2292 return skb_is_nonlinear(skb) &&
2293 ((skb_has_frag_list(skb) &&
2294 !(features & NETIF_F_FRAGLIST)) ||
2295 (skb_shinfo(skb)->nr_frags &&
2296 !(features & NETIF_F_SG)));
2299 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2300 struct netdev_queue *txq)
2302 const struct net_device_ops *ops = dev->netdev_ops;
2303 int rc = NETDEV_TX_OK;
2304 unsigned int skb_len;
2306 if (likely(!skb->next)) {
2307 netdev_features_t features;
2310 * If device doesn't need skb->dst, release it right now while
2311 * its hot in this cpu cache
2313 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2316 features = netif_skb_features(skb);
2318 if (vlan_tx_tag_present(skb) &&
2319 !(features & NETIF_F_HW_VLAN_TX)) {
2320 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2327 /* If encapsulation offload request, verify we are testing
2328 * hardware encapsulation features instead of standard
2329 * features for the netdev
2331 if (skb->encapsulation)
2332 features &= dev->hw_enc_features;
2334 if (netif_needs_gso(skb, features)) {
2335 if (unlikely(dev_gso_segment(skb, features)))
2340 if (skb_needs_linearize(skb, features) &&
2341 __skb_linearize(skb))
2344 /* If packet is not checksummed and device does not
2345 * support checksumming for this protocol, complete
2346 * checksumming here.
2348 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2349 if (skb->encapsulation)
2350 skb_set_inner_transport_header(skb,
2351 skb_checksum_start_offset(skb));
2353 skb_set_transport_header(skb,
2354 skb_checksum_start_offset(skb));
2355 if (!(features & NETIF_F_ALL_CSUM) &&
2356 skb_checksum_help(skb))
2361 if (!list_empty(&ptype_all))
2362 dev_queue_xmit_nit(skb, dev);
2365 rc = ops->ndo_start_xmit(skb, dev);
2366 trace_net_dev_xmit(skb, rc, dev, skb_len);
2367 if (rc == NETDEV_TX_OK)
2368 txq_trans_update(txq);
2374 struct sk_buff *nskb = skb->next;
2376 skb->next = nskb->next;
2380 * If device doesn't need nskb->dst, release it right now while
2381 * its hot in this cpu cache
2383 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2386 if (!list_empty(&ptype_all))
2387 dev_queue_xmit_nit(nskb, dev);
2389 skb_len = nskb->len;
2390 rc = ops->ndo_start_xmit(nskb, dev);
2391 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2392 if (unlikely(rc != NETDEV_TX_OK)) {
2393 if (rc & ~NETDEV_TX_MASK)
2394 goto out_kfree_gso_skb;
2395 nskb->next = skb->next;
2399 txq_trans_update(txq);
2400 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2401 return NETDEV_TX_BUSY;
2402 } while (skb->next);
2405 if (likely(skb->next == NULL))
2406 skb->destructor = DEV_GSO_CB(skb)->destructor;
2413 static u32 hashrnd __read_mostly;
2416 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2417 * to be used as a distribution range.
2419 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2420 unsigned int num_tx_queues)
2424 u16 qcount = num_tx_queues;
2426 if (skb_rx_queue_recorded(skb)) {
2427 hash = skb_get_rx_queue(skb);
2428 while (unlikely(hash >= num_tx_queues))
2429 hash -= num_tx_queues;
2434 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2435 qoffset = dev->tc_to_txq[tc].offset;
2436 qcount = dev->tc_to_txq[tc].count;
2439 if (skb->sk && skb->sk->sk_hash)
2440 hash = skb->sk->sk_hash;
2442 hash = (__force u16) skb->protocol;
2443 hash = jhash_1word(hash, hashrnd);
2445 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2447 EXPORT_SYMBOL(__skb_tx_hash);
2449 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2451 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2452 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2453 dev->name, queue_index,
2454 dev->real_num_tx_queues);
2460 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2463 struct xps_dev_maps *dev_maps;
2464 struct xps_map *map;
2465 int queue_index = -1;
2468 dev_maps = rcu_dereference(dev->xps_maps);
2470 map = rcu_dereference(
2471 dev_maps->cpu_map[raw_smp_processor_id()]);
2474 queue_index = map->queues[0];
2477 if (skb->sk && skb->sk->sk_hash)
2478 hash = skb->sk->sk_hash;
2480 hash = (__force u16) skb->protocol ^
2482 hash = jhash_1word(hash, hashrnd);
2483 queue_index = map->queues[
2484 ((u64)hash * map->len) >> 32];
2486 if (unlikely(queue_index >= dev->real_num_tx_queues))
2498 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
2499 struct sk_buff *skb)
2502 const struct net_device_ops *ops = dev->netdev_ops;
2504 if (dev->real_num_tx_queues == 1)
2506 else if (ops->ndo_select_queue) {
2507 queue_index = ops->ndo_select_queue(dev, skb);
2508 queue_index = dev_cap_txqueue(dev, queue_index);
2510 struct sock *sk = skb->sk;
2511 queue_index = sk_tx_queue_get(sk);
2513 if (queue_index < 0 || skb->ooo_okay ||
2514 queue_index >= dev->real_num_tx_queues) {
2515 int old_index = queue_index;
2517 queue_index = get_xps_queue(dev, skb);
2518 if (queue_index < 0)
2519 queue_index = skb_tx_hash(dev, skb);
2521 if (queue_index != old_index && sk) {
2522 struct dst_entry *dst =
2523 rcu_dereference_check(sk->sk_dst_cache, 1);
2525 if (dst && skb_dst(skb) == dst)
2526 sk_tx_queue_set(sk, queue_index);
2531 skb_set_queue_mapping(skb, queue_index);
2532 return netdev_get_tx_queue(dev, queue_index);
2535 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2536 struct net_device *dev,
2537 struct netdev_queue *txq)
2539 spinlock_t *root_lock = qdisc_lock(q);
2543 qdisc_skb_cb(skb)->pkt_len = skb->len;
2544 qdisc_calculate_pkt_len(skb, q);
2546 * Heuristic to force contended enqueues to serialize on a
2547 * separate lock before trying to get qdisc main lock.
2548 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2549 * and dequeue packets faster.
2551 contended = qdisc_is_running(q);
2552 if (unlikely(contended))
2553 spin_lock(&q->busylock);
2555 spin_lock(root_lock);
2556 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2559 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2560 qdisc_run_begin(q)) {
2562 * This is a work-conserving queue; there are no old skbs
2563 * waiting to be sent out; and the qdisc is not running -
2564 * xmit the skb directly.
2566 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2569 qdisc_bstats_update(q, skb);
2571 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2572 if (unlikely(contended)) {
2573 spin_unlock(&q->busylock);
2580 rc = NET_XMIT_SUCCESS;
2583 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2584 if (qdisc_run_begin(q)) {
2585 if (unlikely(contended)) {
2586 spin_unlock(&q->busylock);
2592 spin_unlock(root_lock);
2593 if (unlikely(contended))
2594 spin_unlock(&q->busylock);
2598 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2599 static void skb_update_prio(struct sk_buff *skb)
2601 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2603 if (!skb->priority && skb->sk && map) {
2604 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2606 if (prioidx < map->priomap_len)
2607 skb->priority = map->priomap[prioidx];
2611 #define skb_update_prio(skb)
2614 static DEFINE_PER_CPU(int, xmit_recursion);
2615 #define RECURSION_LIMIT 10
2618 * dev_loopback_xmit - loop back @skb
2619 * @skb: buffer to transmit
2621 int dev_loopback_xmit(struct sk_buff *skb)
2623 skb_reset_mac_header(skb);
2624 __skb_pull(skb, skb_network_offset(skb));
2625 skb->pkt_type = PACKET_LOOPBACK;
2626 skb->ip_summed = CHECKSUM_UNNECESSARY;
2627 WARN_ON(!skb_dst(skb));
2632 EXPORT_SYMBOL(dev_loopback_xmit);
2635 * dev_queue_xmit - transmit a buffer
2636 * @skb: buffer to transmit
2638 * Queue a buffer for transmission to a network device. The caller must
2639 * have set the device and priority and built the buffer before calling
2640 * this function. The function can be called from an interrupt.
2642 * A negative errno code is returned on a failure. A success does not
2643 * guarantee the frame will be transmitted as it may be dropped due
2644 * to congestion or traffic shaping.
2646 * -----------------------------------------------------------------------------------
2647 * I notice this method can also return errors from the queue disciplines,
2648 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2651 * Regardless of the return value, the skb is consumed, so it is currently
2652 * difficult to retry a send to this method. (You can bump the ref count
2653 * before sending to hold a reference for retry if you are careful.)
2655 * When calling this method, interrupts MUST be enabled. This is because
2656 * the BH enable code must have IRQs enabled so that it will not deadlock.
2659 int dev_queue_xmit(struct sk_buff *skb)
2661 struct net_device *dev = skb->dev;
2662 struct netdev_queue *txq;
2666 /* Disable soft irqs for various locks below. Also
2667 * stops preemption for RCU.
2671 skb_update_prio(skb);
2673 txq = netdev_pick_tx(dev, skb);
2674 q = rcu_dereference_bh(txq->qdisc);
2676 #ifdef CONFIG_NET_CLS_ACT
2677 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2679 trace_net_dev_queue(skb);
2681 rc = __dev_xmit_skb(skb, q, dev, txq);
2685 /* The device has no queue. Common case for software devices:
2686 loopback, all the sorts of tunnels...
2688 Really, it is unlikely that netif_tx_lock protection is necessary
2689 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2691 However, it is possible, that they rely on protection
2694 Check this and shot the lock. It is not prone from deadlocks.
2695 Either shot noqueue qdisc, it is even simpler 8)
2697 if (dev->flags & IFF_UP) {
2698 int cpu = smp_processor_id(); /* ok because BHs are off */
2700 if (txq->xmit_lock_owner != cpu) {
2702 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2703 goto recursion_alert;
2705 HARD_TX_LOCK(dev, txq, cpu);
2707 if (!netif_xmit_stopped(txq)) {
2708 __this_cpu_inc(xmit_recursion);
2709 rc = dev_hard_start_xmit(skb, dev, txq);
2710 __this_cpu_dec(xmit_recursion);
2711 if (dev_xmit_complete(rc)) {
2712 HARD_TX_UNLOCK(dev, txq);
2716 HARD_TX_UNLOCK(dev, txq);
2717 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2720 /* Recursion is detected! It is possible,
2724 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2730 rcu_read_unlock_bh();
2735 rcu_read_unlock_bh();
2738 EXPORT_SYMBOL(dev_queue_xmit);
2741 /*=======================================================================
2743 =======================================================================*/
2745 int netdev_max_backlog __read_mostly = 1000;
2746 EXPORT_SYMBOL(netdev_max_backlog);
2748 int netdev_tstamp_prequeue __read_mostly = 1;
2749 int netdev_budget __read_mostly = 300;
2750 int weight_p __read_mostly = 64; /* old backlog weight */
2752 /* Called with irq disabled */
2753 static inline void ____napi_schedule(struct softnet_data *sd,
2754 struct napi_struct *napi)
2756 list_add_tail(&napi->poll_list, &sd->poll_list);
2757 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2761 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2762 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2763 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2764 * if hash is a canonical 4-tuple hash over transport ports.
2766 void __skb_get_rxhash(struct sk_buff *skb)
2768 struct flow_keys keys;
2771 if (!skb_flow_dissect(skb, &keys))
2777 /* get a consistent hash (same value on both flow directions) */
2778 if (((__force u32)keys.dst < (__force u32)keys.src) ||
2779 (((__force u32)keys.dst == (__force u32)keys.src) &&
2780 ((__force u16)keys.port16[1] < (__force u16)keys.port16[0]))) {
2781 swap(keys.dst, keys.src);
2782 swap(keys.port16[0], keys.port16[1]);
2785 hash = jhash_3words((__force u32)keys.dst,
2786 (__force u32)keys.src,
2787 (__force u32)keys.ports, hashrnd);
2793 EXPORT_SYMBOL(__skb_get_rxhash);
2797 /* One global table that all flow-based protocols share. */
2798 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2799 EXPORT_SYMBOL(rps_sock_flow_table);
2801 struct static_key rps_needed __read_mostly;
2803 static struct rps_dev_flow *
2804 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2805 struct rps_dev_flow *rflow, u16 next_cpu)
2807 if (next_cpu != RPS_NO_CPU) {
2808 #ifdef CONFIG_RFS_ACCEL
2809 struct netdev_rx_queue *rxqueue;
2810 struct rps_dev_flow_table *flow_table;
2811 struct rps_dev_flow *old_rflow;
2816 /* Should we steer this flow to a different hardware queue? */
2817 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2818 !(dev->features & NETIF_F_NTUPLE))
2820 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2821 if (rxq_index == skb_get_rx_queue(skb))
2824 rxqueue = dev->_rx + rxq_index;
2825 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2828 flow_id = skb->rxhash & flow_table->mask;
2829 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2830 rxq_index, flow_id);
2834 rflow = &flow_table->flows[flow_id];
2836 if (old_rflow->filter == rflow->filter)
2837 old_rflow->filter = RPS_NO_FILTER;
2841 per_cpu(softnet_data, next_cpu).input_queue_head;
2844 rflow->cpu = next_cpu;
2849 * get_rps_cpu is called from netif_receive_skb and returns the target
2850 * CPU from the RPS map of the receiving queue for a given skb.
2851 * rcu_read_lock must be held on entry.
2853 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2854 struct rps_dev_flow **rflowp)
2856 struct netdev_rx_queue *rxqueue;
2857 struct rps_map *map;
2858 struct rps_dev_flow_table *flow_table;
2859 struct rps_sock_flow_table *sock_flow_table;
2863 if (skb_rx_queue_recorded(skb)) {
2864 u16 index = skb_get_rx_queue(skb);
2865 if (unlikely(index >= dev->real_num_rx_queues)) {
2866 WARN_ONCE(dev->real_num_rx_queues > 1,
2867 "%s received packet on queue %u, but number "
2868 "of RX queues is %u\n",
2869 dev->name, index, dev->real_num_rx_queues);
2872 rxqueue = dev->_rx + index;
2876 map = rcu_dereference(rxqueue->rps_map);
2878 if (map->len == 1 &&
2879 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2880 tcpu = map->cpus[0];
2881 if (cpu_online(tcpu))
2885 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2889 skb_reset_network_header(skb);
2890 if (!skb_get_rxhash(skb))
2893 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2894 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2895 if (flow_table && sock_flow_table) {
2897 struct rps_dev_flow *rflow;
2899 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2902 next_cpu = sock_flow_table->ents[skb->rxhash &
2903 sock_flow_table->mask];
2906 * If the desired CPU (where last recvmsg was done) is
2907 * different from current CPU (one in the rx-queue flow
2908 * table entry), switch if one of the following holds:
2909 * - Current CPU is unset (equal to RPS_NO_CPU).
2910 * - Current CPU is offline.
2911 * - The current CPU's queue tail has advanced beyond the
2912 * last packet that was enqueued using this table entry.
2913 * This guarantees that all previous packets for the flow
2914 * have been dequeued, thus preserving in order delivery.
2916 if (unlikely(tcpu != next_cpu) &&
2917 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2918 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2919 rflow->last_qtail)) >= 0)) {
2921 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2924 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2932 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2934 if (cpu_online(tcpu)) {
2944 #ifdef CONFIG_RFS_ACCEL
2947 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2948 * @dev: Device on which the filter was set
2949 * @rxq_index: RX queue index
2950 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2951 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2953 * Drivers that implement ndo_rx_flow_steer() should periodically call
2954 * this function for each installed filter and remove the filters for
2955 * which it returns %true.
2957 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2958 u32 flow_id, u16 filter_id)
2960 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2961 struct rps_dev_flow_table *flow_table;
2962 struct rps_dev_flow *rflow;
2967 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2968 if (flow_table && flow_id <= flow_table->mask) {
2969 rflow = &flow_table->flows[flow_id];
2970 cpu = ACCESS_ONCE(rflow->cpu);
2971 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2972 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2973 rflow->last_qtail) <
2974 (int)(10 * flow_table->mask)))
2980 EXPORT_SYMBOL(rps_may_expire_flow);
2982 #endif /* CONFIG_RFS_ACCEL */
2984 /* Called from hardirq (IPI) context */
2985 static void rps_trigger_softirq(void *data)
2987 struct softnet_data *sd = data;
2989 ____napi_schedule(sd, &sd->backlog);
2993 #endif /* CONFIG_RPS */
2996 * Check if this softnet_data structure is another cpu one
2997 * If yes, queue it to our IPI list and return 1
3000 static int rps_ipi_queued(struct softnet_data *sd)
3003 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3006 sd->rps_ipi_next = mysd->rps_ipi_list;
3007 mysd->rps_ipi_list = sd;
3009 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3012 #endif /* CONFIG_RPS */
3017 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3018 * queue (may be a remote CPU queue).
3020 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3021 unsigned int *qtail)
3023 struct softnet_data *sd;
3024 unsigned long flags;
3026 sd = &per_cpu(softnet_data, cpu);
3028 local_irq_save(flags);
3031 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
3032 if (skb_queue_len(&sd->input_pkt_queue)) {
3034 __skb_queue_tail(&sd->input_pkt_queue, skb);
3035 input_queue_tail_incr_save(sd, qtail);
3037 local_irq_restore(flags);
3038 return NET_RX_SUCCESS;
3041 /* Schedule NAPI for backlog device
3042 * We can use non atomic operation since we own the queue lock
3044 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3045 if (!rps_ipi_queued(sd))
3046 ____napi_schedule(sd, &sd->backlog);
3054 local_irq_restore(flags);
3056 atomic_long_inc(&skb->dev->rx_dropped);
3062 * netif_rx - post buffer to the network code
3063 * @skb: buffer to post
3065 * This function receives a packet from a device driver and queues it for
3066 * the upper (protocol) levels to process. It always succeeds. The buffer
3067 * may be dropped during processing for congestion control or by the
3071 * NET_RX_SUCCESS (no congestion)
3072 * NET_RX_DROP (packet was dropped)
3076 int netif_rx(struct sk_buff *skb)
3080 /* if netpoll wants it, pretend we never saw it */
3081 if (netpoll_rx(skb))
3084 net_timestamp_check(netdev_tstamp_prequeue, skb);
3086 trace_netif_rx(skb);
3088 if (static_key_false(&rps_needed)) {
3089 struct rps_dev_flow voidflow, *rflow = &voidflow;
3095 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3097 cpu = smp_processor_id();
3099 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3107 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3112 EXPORT_SYMBOL(netif_rx);
3114 int netif_rx_ni(struct sk_buff *skb)
3119 err = netif_rx(skb);
3120 if (local_softirq_pending())
3126 EXPORT_SYMBOL(netif_rx_ni);
3128 static void net_tx_action(struct softirq_action *h)
3130 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3132 if (sd->completion_queue) {
3133 struct sk_buff *clist;
3135 local_irq_disable();
3136 clist = sd->completion_queue;
3137 sd->completion_queue = NULL;
3141 struct sk_buff *skb = clist;
3142 clist = clist->next;
3144 WARN_ON(atomic_read(&skb->users));
3145 trace_kfree_skb(skb, net_tx_action);
3150 if (sd->output_queue) {
3153 local_irq_disable();
3154 head = sd->output_queue;
3155 sd->output_queue = NULL;
3156 sd->output_queue_tailp = &sd->output_queue;
3160 struct Qdisc *q = head;
3161 spinlock_t *root_lock;
3163 head = head->next_sched;
3165 root_lock = qdisc_lock(q);
3166 if (spin_trylock(root_lock)) {
3167 smp_mb__before_clear_bit();
3168 clear_bit(__QDISC_STATE_SCHED,
3171 spin_unlock(root_lock);
3173 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3175 __netif_reschedule(q);
3177 smp_mb__before_clear_bit();
3178 clear_bit(__QDISC_STATE_SCHED,
3186 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3187 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3188 /* This hook is defined here for ATM LANE */
3189 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3190 unsigned char *addr) __read_mostly;
3191 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3194 #ifdef CONFIG_NET_CLS_ACT
3195 /* TODO: Maybe we should just force sch_ingress to be compiled in
3196 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3197 * a compare and 2 stores extra right now if we dont have it on
3198 * but have CONFIG_NET_CLS_ACT
3199 * NOTE: This doesn't stop any functionality; if you dont have
3200 * the ingress scheduler, you just can't add policies on ingress.
3203 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3205 struct net_device *dev = skb->dev;
3206 u32 ttl = G_TC_RTTL(skb->tc_verd);
3207 int result = TC_ACT_OK;
3210 if (unlikely(MAX_RED_LOOP < ttl++)) {
3211 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3212 skb->skb_iif, dev->ifindex);
3216 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3217 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3220 if (q != &noop_qdisc) {
3221 spin_lock(qdisc_lock(q));
3222 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3223 result = qdisc_enqueue_root(skb, q);
3224 spin_unlock(qdisc_lock(q));
3230 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3231 struct packet_type **pt_prev,
3232 int *ret, struct net_device *orig_dev)
3234 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3236 if (!rxq || rxq->qdisc == &noop_qdisc)
3240 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3244 switch (ing_filter(skb, rxq)) {
3258 * netdev_rx_handler_register - register receive handler
3259 * @dev: device to register a handler for
3260 * @rx_handler: receive handler to register
3261 * @rx_handler_data: data pointer that is used by rx handler
3263 * Register a receive hander for a device. This handler will then be
3264 * called from __netif_receive_skb. A negative errno code is returned
3267 * The caller must hold the rtnl_mutex.
3269 * For a general description of rx_handler, see enum rx_handler_result.
3271 int netdev_rx_handler_register(struct net_device *dev,
3272 rx_handler_func_t *rx_handler,
3273 void *rx_handler_data)
3277 if (dev->rx_handler)
3280 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3281 rcu_assign_pointer(dev->rx_handler, rx_handler);
3285 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3288 * netdev_rx_handler_unregister - unregister receive handler
3289 * @dev: device to unregister a handler from
3291 * Unregister a receive hander from a device.
3293 * The caller must hold the rtnl_mutex.
3295 void netdev_rx_handler_unregister(struct net_device *dev)
3299 RCU_INIT_POINTER(dev->rx_handler, NULL);
3300 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3302 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3305 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3306 * the special handling of PFMEMALLOC skbs.
3308 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3310 switch (skb->protocol) {
3311 case __constant_htons(ETH_P_ARP):
3312 case __constant_htons(ETH_P_IP):
3313 case __constant_htons(ETH_P_IPV6):
3314 case __constant_htons(ETH_P_8021Q):
3321 static int __netif_receive_skb(struct sk_buff *skb)
3323 struct packet_type *ptype, *pt_prev;
3324 rx_handler_func_t *rx_handler;
3325 struct net_device *orig_dev;
3326 struct net_device *null_or_dev;
3327 bool deliver_exact = false;
3328 int ret = NET_RX_DROP;
3330 unsigned long pflags = current->flags;
3332 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3334 trace_netif_receive_skb(skb);
3337 * PFMEMALLOC skbs are special, they should
3338 * - be delivered to SOCK_MEMALLOC sockets only
3339 * - stay away from userspace
3340 * - have bounded memory usage
3342 * Use PF_MEMALLOC as this saves us from propagating the allocation
3343 * context down to all allocation sites.
3345 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3346 current->flags |= PF_MEMALLOC;
3348 /* if we've gotten here through NAPI, check netpoll */
3349 if (netpoll_receive_skb(skb))
3352 orig_dev = skb->dev;
3354 skb_reset_network_header(skb);
3355 skb_reset_transport_header(skb);
3356 skb_reset_mac_len(skb);
3363 skb->skb_iif = skb->dev->ifindex;
3365 __this_cpu_inc(softnet_data.processed);
3367 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3368 skb = vlan_untag(skb);
3373 #ifdef CONFIG_NET_CLS_ACT
3374 if (skb->tc_verd & TC_NCLS) {
3375 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3380 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3383 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3384 if (!ptype->dev || ptype->dev == skb->dev) {
3386 ret = deliver_skb(skb, pt_prev, orig_dev);
3392 #ifdef CONFIG_NET_CLS_ACT
3393 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3399 if (sk_memalloc_socks() && skb_pfmemalloc(skb)
3400 && !skb_pfmemalloc_protocol(skb))
3403 if (vlan_tx_tag_present(skb)) {
3405 ret = deliver_skb(skb, pt_prev, orig_dev);
3408 if (vlan_do_receive(&skb))
3410 else if (unlikely(!skb))
3414 rx_handler = rcu_dereference(skb->dev->rx_handler);
3417 ret = deliver_skb(skb, pt_prev, orig_dev);
3420 switch (rx_handler(&skb)) {
3421 case RX_HANDLER_CONSUMED:
3423 case RX_HANDLER_ANOTHER:
3425 case RX_HANDLER_EXACT:
3426 deliver_exact = true;
3427 case RX_HANDLER_PASS:
3434 if (vlan_tx_nonzero_tag_present(skb))
3435 skb->pkt_type = PACKET_OTHERHOST;
3437 /* deliver only exact match when indicated */
3438 null_or_dev = deliver_exact ? skb->dev : NULL;
3440 type = skb->protocol;
3441 list_for_each_entry_rcu(ptype,
3442 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3443 if (ptype->type == type &&
3444 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3445 ptype->dev == orig_dev)) {
3447 ret = deliver_skb(skb, pt_prev, orig_dev);
3453 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3456 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3459 atomic_long_inc(&skb->dev->rx_dropped);
3461 /* Jamal, now you will not able to escape explaining
3462 * me how you were going to use this. :-)
3470 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3475 * netif_receive_skb - process receive buffer from network
3476 * @skb: buffer to process
3478 * netif_receive_skb() is the main receive data processing function.
3479 * It always succeeds. The buffer may be dropped during processing
3480 * for congestion control or by the protocol layers.
3482 * This function may only be called from softirq context and interrupts
3483 * should be enabled.
3485 * Return values (usually ignored):
3486 * NET_RX_SUCCESS: no congestion
3487 * NET_RX_DROP: packet was dropped
3489 int netif_receive_skb(struct sk_buff *skb)
3491 net_timestamp_check(netdev_tstamp_prequeue, skb);
3493 if (skb_defer_rx_timestamp(skb))
3494 return NET_RX_SUCCESS;
3497 if (static_key_false(&rps_needed)) {
3498 struct rps_dev_flow voidflow, *rflow = &voidflow;
3503 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3506 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3513 return __netif_receive_skb(skb);
3515 EXPORT_SYMBOL(netif_receive_skb);
3517 /* Network device is going away, flush any packets still pending
3518 * Called with irqs disabled.
3520 static void flush_backlog(void *arg)
3522 struct net_device *dev = arg;
3523 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3524 struct sk_buff *skb, *tmp;
3527 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3528 if (skb->dev == dev) {
3529 __skb_unlink(skb, &sd->input_pkt_queue);
3531 input_queue_head_incr(sd);
3536 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3537 if (skb->dev == dev) {
3538 __skb_unlink(skb, &sd->process_queue);
3540 input_queue_head_incr(sd);
3545 static int napi_gro_complete(struct sk_buff *skb)
3547 struct packet_offload *ptype;
3548 __be16 type = skb->protocol;
3549 struct list_head *head = &offload_base;
3552 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3554 if (NAPI_GRO_CB(skb)->count == 1) {
3555 skb_shinfo(skb)->gso_size = 0;
3560 list_for_each_entry_rcu(ptype, head, list) {
3561 if (ptype->type != type || !ptype->callbacks.gro_complete)
3564 err = ptype->callbacks.gro_complete(skb);
3570 WARN_ON(&ptype->list == head);
3572 return NET_RX_SUCCESS;
3576 return netif_receive_skb(skb);
3579 /* napi->gro_list contains packets ordered by age.
3580 * youngest packets at the head of it.
3581 * Complete skbs in reverse order to reduce latencies.
3583 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3585 struct sk_buff *skb, *prev = NULL;
3587 /* scan list and build reverse chain */
3588 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3593 for (skb = prev; skb; skb = prev) {
3596 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3600 napi_gro_complete(skb);
3604 napi->gro_list = NULL;
3606 EXPORT_SYMBOL(napi_gro_flush);
3608 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3611 unsigned int maclen = skb->dev->hard_header_len;
3613 for (p = napi->gro_list; p; p = p->next) {
3614 unsigned long diffs;
3616 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3617 diffs |= p->vlan_tci ^ skb->vlan_tci;
3618 if (maclen == ETH_HLEN)
3619 diffs |= compare_ether_header(skb_mac_header(p),
3620 skb_gro_mac_header(skb));
3622 diffs = memcmp(skb_mac_header(p),
3623 skb_gro_mac_header(skb),
3625 NAPI_GRO_CB(p)->same_flow = !diffs;
3626 NAPI_GRO_CB(p)->flush = 0;
3630 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3632 struct sk_buff **pp = NULL;
3633 struct packet_offload *ptype;
3634 __be16 type = skb->protocol;
3635 struct list_head *head = &offload_base;
3638 enum gro_result ret;
3640 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3643 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3646 gro_list_prepare(napi, skb);
3649 list_for_each_entry_rcu(ptype, head, list) {
3650 if (ptype->type != type || !ptype->callbacks.gro_receive)
3653 skb_set_network_header(skb, skb_gro_offset(skb));
3654 mac_len = skb->network_header - skb->mac_header;
3655 skb->mac_len = mac_len;
3656 NAPI_GRO_CB(skb)->same_flow = 0;
3657 NAPI_GRO_CB(skb)->flush = 0;
3658 NAPI_GRO_CB(skb)->free = 0;
3660 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3665 if (&ptype->list == head)
3668 same_flow = NAPI_GRO_CB(skb)->same_flow;
3669 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3672 struct sk_buff *nskb = *pp;
3676 napi_gro_complete(nskb);
3683 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3687 NAPI_GRO_CB(skb)->count = 1;
3688 NAPI_GRO_CB(skb)->age = jiffies;
3689 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3690 skb->next = napi->gro_list;
3691 napi->gro_list = skb;
3695 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3696 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3698 BUG_ON(skb->end - skb->tail < grow);
3700 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3703 skb->data_len -= grow;
3705 skb_shinfo(skb)->frags[0].page_offset += grow;
3706 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3708 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3709 skb_frag_unref(skb, 0);
3710 memmove(skb_shinfo(skb)->frags,
3711 skb_shinfo(skb)->frags + 1,
3712 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3725 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3729 if (netif_receive_skb(skb))
3737 case GRO_MERGED_FREE:
3738 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3739 kmem_cache_free(skbuff_head_cache, skb);
3752 static void skb_gro_reset_offset(struct sk_buff *skb)
3754 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3755 const skb_frag_t *frag0 = &pinfo->frags[0];
3757 NAPI_GRO_CB(skb)->data_offset = 0;
3758 NAPI_GRO_CB(skb)->frag0 = NULL;
3759 NAPI_GRO_CB(skb)->frag0_len = 0;
3761 if (skb->mac_header == skb->tail &&
3763 !PageHighMem(skb_frag_page(frag0))) {
3764 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3765 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3769 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3771 skb_gro_reset_offset(skb);
3773 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3775 EXPORT_SYMBOL(napi_gro_receive);
3777 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3779 __skb_pull(skb, skb_headlen(skb));
3780 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3781 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3783 skb->dev = napi->dev;
3789 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3791 struct sk_buff *skb = napi->skb;
3794 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3800 EXPORT_SYMBOL(napi_get_frags);
3802 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3808 skb->protocol = eth_type_trans(skb, skb->dev);
3810 if (ret == GRO_HELD)
3811 skb_gro_pull(skb, -ETH_HLEN);
3812 else if (netif_receive_skb(skb))
3817 case GRO_MERGED_FREE:
3818 napi_reuse_skb(napi, skb);
3828 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3830 struct sk_buff *skb = napi->skb;
3837 skb_reset_mac_header(skb);
3838 skb_gro_reset_offset(skb);
3840 off = skb_gro_offset(skb);
3841 hlen = off + sizeof(*eth);
3842 eth = skb_gro_header_fast(skb, off);
3843 if (skb_gro_header_hard(skb, hlen)) {
3844 eth = skb_gro_header_slow(skb, hlen, off);
3845 if (unlikely(!eth)) {
3846 napi_reuse_skb(napi, skb);
3852 skb_gro_pull(skb, sizeof(*eth));
3855 * This works because the only protocols we care about don't require
3856 * special handling. We'll fix it up properly at the end.
3858 skb->protocol = eth->h_proto;
3864 gro_result_t napi_gro_frags(struct napi_struct *napi)
3866 struct sk_buff *skb = napi_frags_skb(napi);
3871 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
3873 EXPORT_SYMBOL(napi_gro_frags);
3876 * net_rps_action sends any pending IPI's for rps.
3877 * Note: called with local irq disabled, but exits with local irq enabled.
3879 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3882 struct softnet_data *remsd = sd->rps_ipi_list;
3885 sd->rps_ipi_list = NULL;
3889 /* Send pending IPI's to kick RPS processing on remote cpus. */
3891 struct softnet_data *next = remsd->rps_ipi_next;
3893 if (cpu_online(remsd->cpu))
3894 __smp_call_function_single(remsd->cpu,
3903 static int process_backlog(struct napi_struct *napi, int quota)
3906 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3909 /* Check if we have pending ipi, its better to send them now,
3910 * not waiting net_rx_action() end.
3912 if (sd->rps_ipi_list) {
3913 local_irq_disable();
3914 net_rps_action_and_irq_enable(sd);
3917 napi->weight = weight_p;
3918 local_irq_disable();
3919 while (work < quota) {
3920 struct sk_buff *skb;
3923 while ((skb = __skb_dequeue(&sd->process_queue))) {
3925 __netif_receive_skb(skb);
3926 local_irq_disable();
3927 input_queue_head_incr(sd);
3928 if (++work >= quota) {
3935 qlen = skb_queue_len(&sd->input_pkt_queue);
3937 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3938 &sd->process_queue);
3940 if (qlen < quota - work) {
3942 * Inline a custom version of __napi_complete().
3943 * only current cpu owns and manipulates this napi,
3944 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3945 * we can use a plain write instead of clear_bit(),
3946 * and we dont need an smp_mb() memory barrier.
3948 list_del(&napi->poll_list);
3951 quota = work + qlen;
3961 * __napi_schedule - schedule for receive
3962 * @n: entry to schedule
3964 * The entry's receive function will be scheduled to run
3966 void __napi_schedule(struct napi_struct *n)
3968 unsigned long flags;
3970 local_irq_save(flags);
3971 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3972 local_irq_restore(flags);
3974 EXPORT_SYMBOL(__napi_schedule);
3976 void __napi_complete(struct napi_struct *n)
3978 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3979 BUG_ON(n->gro_list);
3981 list_del(&n->poll_list);
3982 smp_mb__before_clear_bit();
3983 clear_bit(NAPI_STATE_SCHED, &n->state);
3985 EXPORT_SYMBOL(__napi_complete);
3987 void napi_complete(struct napi_struct *n)
3989 unsigned long flags;
3992 * don't let napi dequeue from the cpu poll list
3993 * just in case its running on a different cpu
3995 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3998 napi_gro_flush(n, false);
3999 local_irq_save(flags);
4001 local_irq_restore(flags);
4003 EXPORT_SYMBOL(napi_complete);
4005 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4006 int (*poll)(struct napi_struct *, int), int weight)
4008 INIT_LIST_HEAD(&napi->poll_list);
4009 napi->gro_count = 0;
4010 napi->gro_list = NULL;
4013 napi->weight = weight;
4014 list_add(&napi->dev_list, &dev->napi_list);
4016 #ifdef CONFIG_NETPOLL
4017 spin_lock_init(&napi->poll_lock);
4018 napi->poll_owner = -1;
4020 set_bit(NAPI_STATE_SCHED, &napi->state);
4022 EXPORT_SYMBOL(netif_napi_add);
4024 void netif_napi_del(struct napi_struct *napi)
4026 struct sk_buff *skb, *next;
4028 list_del_init(&napi->dev_list);
4029 napi_free_frags(napi);
4031 for (skb = napi->gro_list; skb; skb = next) {
4037 napi->gro_list = NULL;
4038 napi->gro_count = 0;
4040 EXPORT_SYMBOL(netif_napi_del);
4042 static void net_rx_action(struct softirq_action *h)
4044 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4045 unsigned long time_limit = jiffies + 2;
4046 int budget = netdev_budget;
4049 local_irq_disable();
4051 while (!list_empty(&sd->poll_list)) {
4052 struct napi_struct *n;
4055 /* If softirq window is exhuasted then punt.
4056 * Allow this to run for 2 jiffies since which will allow
4057 * an average latency of 1.5/HZ.
4059 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
4064 /* Even though interrupts have been re-enabled, this
4065 * access is safe because interrupts can only add new
4066 * entries to the tail of this list, and only ->poll()
4067 * calls can remove this head entry from the list.
4069 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4071 have = netpoll_poll_lock(n);
4075 /* This NAPI_STATE_SCHED test is for avoiding a race
4076 * with netpoll's poll_napi(). Only the entity which
4077 * obtains the lock and sees NAPI_STATE_SCHED set will
4078 * actually make the ->poll() call. Therefore we avoid
4079 * accidentally calling ->poll() when NAPI is not scheduled.
4082 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4083 work = n->poll(n, weight);
4087 WARN_ON_ONCE(work > weight);
4091 local_irq_disable();
4093 /* Drivers must not modify the NAPI state if they
4094 * consume the entire weight. In such cases this code
4095 * still "owns" the NAPI instance and therefore can
4096 * move the instance around on the list at-will.
4098 if (unlikely(work == weight)) {
4099 if (unlikely(napi_disable_pending(n))) {
4102 local_irq_disable();
4105 /* flush too old packets
4106 * If HZ < 1000, flush all packets.
4109 napi_gro_flush(n, HZ >= 1000);
4110 local_irq_disable();
4112 list_move_tail(&n->poll_list, &sd->poll_list);
4116 netpoll_poll_unlock(have);
4119 net_rps_action_and_irq_enable(sd);
4121 #ifdef CONFIG_NET_DMA
4123 * There may not be any more sk_buffs coming right now, so push
4124 * any pending DMA copies to hardware
4126 dma_issue_pending_all();
4133 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4137 static gifconf_func_t *gifconf_list[NPROTO];
4140 * register_gifconf - register a SIOCGIF handler
4141 * @family: Address family
4142 * @gifconf: Function handler
4144 * Register protocol dependent address dumping routines. The handler
4145 * that is passed must not be freed or reused until it has been replaced
4146 * by another handler.
4148 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4150 if (family >= NPROTO)
4152 gifconf_list[family] = gifconf;
4155 EXPORT_SYMBOL(register_gifconf);
4159 * Map an interface index to its name (SIOCGIFNAME)
4163 * We need this ioctl for efficient implementation of the
4164 * if_indextoname() function required by the IPv6 API. Without
4165 * it, we would have to search all the interfaces to find a
4169 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4171 struct net_device *dev;
4176 * Fetch the caller's info block.
4179 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4183 seq = read_seqcount_begin(&devnet_rename_seq);
4185 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4191 strcpy(ifr.ifr_name, dev->name);
4193 if (read_seqcount_retry(&devnet_rename_seq, seq))
4196 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4202 * Perform a SIOCGIFCONF call. This structure will change
4203 * size eventually, and there is nothing I can do about it.
4204 * Thus we will need a 'compatibility mode'.
4207 static int dev_ifconf(struct net *net, char __user *arg)
4210 struct net_device *dev;
4217 * Fetch the caller's info block.
4220 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4227 * Loop over the interfaces, and write an info block for each.
4231 for_each_netdev(net, dev) {
4232 for (i = 0; i < NPROTO; i++) {
4233 if (gifconf_list[i]) {
4236 done = gifconf_list[i](dev, NULL, 0);
4238 done = gifconf_list[i](dev, pos + total,
4248 * All done. Write the updated control block back to the caller.
4250 ifc.ifc_len = total;
4253 * Both BSD and Solaris return 0 here, so we do too.
4255 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4258 #ifdef CONFIG_PROC_FS
4260 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4262 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4263 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4264 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4266 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4268 struct net *net = seq_file_net(seq);
4269 struct net_device *dev;
4270 struct hlist_node *p;
4271 struct hlist_head *h;
4272 unsigned int count = 0, offset = get_offset(*pos);
4274 h = &net->dev_name_head[get_bucket(*pos)];
4275 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4276 if (++count == offset)
4283 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4285 struct net_device *dev;
4286 unsigned int bucket;
4289 dev = dev_from_same_bucket(seq, pos);
4293 bucket = get_bucket(*pos) + 1;
4294 *pos = set_bucket_offset(bucket, 1);
4295 } while (bucket < NETDEV_HASHENTRIES);
4301 * This is invoked by the /proc filesystem handler to display a device
4304 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4309 return SEQ_START_TOKEN;
4311 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4314 return dev_from_bucket(seq, pos);
4317 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4320 return dev_from_bucket(seq, pos);
4323 void dev_seq_stop(struct seq_file *seq, void *v)
4329 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4331 struct rtnl_link_stats64 temp;
4332 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4334 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4335 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4336 dev->name, stats->rx_bytes, stats->rx_packets,
4338 stats->rx_dropped + stats->rx_missed_errors,
4339 stats->rx_fifo_errors,
4340 stats->rx_length_errors + stats->rx_over_errors +
4341 stats->rx_crc_errors + stats->rx_frame_errors,
4342 stats->rx_compressed, stats->multicast,
4343 stats->tx_bytes, stats->tx_packets,
4344 stats->tx_errors, stats->tx_dropped,
4345 stats->tx_fifo_errors, stats->collisions,
4346 stats->tx_carrier_errors +
4347 stats->tx_aborted_errors +
4348 stats->tx_window_errors +
4349 stats->tx_heartbeat_errors,
4350 stats->tx_compressed);
4354 * Called from the PROCfs module. This now uses the new arbitrary sized
4355 * /proc/net interface to create /proc/net/dev
4357 static int dev_seq_show(struct seq_file *seq, void *v)
4359 if (v == SEQ_START_TOKEN)
4360 seq_puts(seq, "Inter-| Receive "
4362 " face |bytes packets errs drop fifo frame "
4363 "compressed multicast|bytes packets errs "
4364 "drop fifo colls carrier compressed\n");
4366 dev_seq_printf_stats(seq, v);
4370 static struct softnet_data *softnet_get_online(loff_t *pos)
4372 struct softnet_data *sd = NULL;
4374 while (*pos < nr_cpu_ids)
4375 if (cpu_online(*pos)) {
4376 sd = &per_cpu(softnet_data, *pos);
4383 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4385 return softnet_get_online(pos);
4388 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4391 return softnet_get_online(pos);
4394 static void softnet_seq_stop(struct seq_file *seq, void *v)
4398 static int softnet_seq_show(struct seq_file *seq, void *v)
4400 struct softnet_data *sd = v;
4402 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4403 sd->processed, sd->dropped, sd->time_squeeze, 0,
4404 0, 0, 0, 0, /* was fastroute */
4405 sd->cpu_collision, sd->received_rps);
4409 static const struct seq_operations dev_seq_ops = {
4410 .start = dev_seq_start,
4411 .next = dev_seq_next,
4412 .stop = dev_seq_stop,
4413 .show = dev_seq_show,
4416 static int dev_seq_open(struct inode *inode, struct file *file)
4418 return seq_open_net(inode, file, &dev_seq_ops,
4419 sizeof(struct seq_net_private));
4422 static const struct file_operations dev_seq_fops = {
4423 .owner = THIS_MODULE,
4424 .open = dev_seq_open,
4426 .llseek = seq_lseek,
4427 .release = seq_release_net,
4430 static const struct seq_operations softnet_seq_ops = {
4431 .start = softnet_seq_start,
4432 .next = softnet_seq_next,
4433 .stop = softnet_seq_stop,
4434 .show = softnet_seq_show,
4437 static int softnet_seq_open(struct inode *inode, struct file *file)
4439 return seq_open(file, &softnet_seq_ops);
4442 static const struct file_operations softnet_seq_fops = {
4443 .owner = THIS_MODULE,
4444 .open = softnet_seq_open,
4446 .llseek = seq_lseek,
4447 .release = seq_release,
4450 static void *ptype_get_idx(loff_t pos)
4452 struct packet_type *pt = NULL;
4456 list_for_each_entry_rcu(pt, &ptype_all, list) {
4462 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4463 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4472 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4476 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4479 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4481 struct packet_type *pt;
4482 struct list_head *nxt;
4486 if (v == SEQ_START_TOKEN)
4487 return ptype_get_idx(0);
4490 nxt = pt->list.next;
4491 if (pt->type == htons(ETH_P_ALL)) {
4492 if (nxt != &ptype_all)
4495 nxt = ptype_base[0].next;
4497 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4499 while (nxt == &ptype_base[hash]) {
4500 if (++hash >= PTYPE_HASH_SIZE)
4502 nxt = ptype_base[hash].next;
4505 return list_entry(nxt, struct packet_type, list);
4508 static void ptype_seq_stop(struct seq_file *seq, void *v)
4514 static int ptype_seq_show(struct seq_file *seq, void *v)
4516 struct packet_type *pt = v;
4518 if (v == SEQ_START_TOKEN)
4519 seq_puts(seq, "Type Device Function\n");
4520 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4521 if (pt->type == htons(ETH_P_ALL))
4522 seq_puts(seq, "ALL ");
4524 seq_printf(seq, "%04x", ntohs(pt->type));
4526 seq_printf(seq, " %-8s %pF\n",
4527 pt->dev ? pt->dev->name : "", pt->func);
4533 static const struct seq_operations ptype_seq_ops = {
4534 .start = ptype_seq_start,
4535 .next = ptype_seq_next,
4536 .stop = ptype_seq_stop,
4537 .show = ptype_seq_show,
4540 static int ptype_seq_open(struct inode *inode, struct file *file)
4542 return seq_open_net(inode, file, &ptype_seq_ops,
4543 sizeof(struct seq_net_private));
4546 static const struct file_operations ptype_seq_fops = {
4547 .owner = THIS_MODULE,
4548 .open = ptype_seq_open,
4550 .llseek = seq_lseek,
4551 .release = seq_release_net,
4555 static int __net_init dev_proc_net_init(struct net *net)
4559 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4561 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4563 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4566 if (wext_proc_init(net))
4572 proc_net_remove(net, "ptype");
4574 proc_net_remove(net, "softnet_stat");
4576 proc_net_remove(net, "dev");
4580 static void __net_exit dev_proc_net_exit(struct net *net)
4582 wext_proc_exit(net);
4584 proc_net_remove(net, "ptype");
4585 proc_net_remove(net, "softnet_stat");
4586 proc_net_remove(net, "dev");
4589 static struct pernet_operations __net_initdata dev_proc_ops = {
4590 .init = dev_proc_net_init,
4591 .exit = dev_proc_net_exit,
4594 static int __init dev_proc_init(void)
4596 return register_pernet_subsys(&dev_proc_ops);
4599 #define dev_proc_init() 0
4600 #endif /* CONFIG_PROC_FS */
4604 * netdev_set_master - set up master pointer
4605 * @slave: slave device
4606 * @master: new master device
4608 * Changes the master device of the slave. Pass %NULL to break the
4609 * bonding. The caller must hold the RTNL semaphore. On a failure
4610 * a negative errno code is returned. On success the reference counts
4611 * are adjusted and the function returns zero.
4613 int netdev_set_master(struct net_device *slave, struct net_device *master)
4615 struct net_device *old = slave->master;
4625 slave->master = master;
4631 EXPORT_SYMBOL(netdev_set_master);
4634 * netdev_set_bond_master - set up bonding master/slave pair
4635 * @slave: slave device
4636 * @master: new master device
4638 * Changes the master device of the slave. Pass %NULL to break the
4639 * bonding. The caller must hold the RTNL semaphore. On a failure
4640 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4641 * to the routing socket and the function returns zero.
4643 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4649 err = netdev_set_master(slave, master);
4653 slave->flags |= IFF_SLAVE;
4655 slave->flags &= ~IFF_SLAVE;
4657 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4660 EXPORT_SYMBOL(netdev_set_bond_master);
4662 static void dev_change_rx_flags(struct net_device *dev, int flags)
4664 const struct net_device_ops *ops = dev->netdev_ops;
4666 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4667 ops->ndo_change_rx_flags(dev, flags);
4670 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4672 unsigned int old_flags = dev->flags;
4678 dev->flags |= IFF_PROMISC;
4679 dev->promiscuity += inc;
4680 if (dev->promiscuity == 0) {
4683 * If inc causes overflow, untouch promisc and return error.
4686 dev->flags &= ~IFF_PROMISC;
4688 dev->promiscuity -= inc;
4689 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4694 if (dev->flags != old_flags) {
4695 pr_info("device %s %s promiscuous mode\n",
4697 dev->flags & IFF_PROMISC ? "entered" : "left");
4698 if (audit_enabled) {
4699 current_uid_gid(&uid, &gid);
4700 audit_log(current->audit_context, GFP_ATOMIC,
4701 AUDIT_ANOM_PROMISCUOUS,
4702 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4703 dev->name, (dev->flags & IFF_PROMISC),
4704 (old_flags & IFF_PROMISC),
4705 from_kuid(&init_user_ns, audit_get_loginuid(current)),
4706 from_kuid(&init_user_ns, uid),
4707 from_kgid(&init_user_ns, gid),
4708 audit_get_sessionid(current));
4711 dev_change_rx_flags(dev, IFF_PROMISC);
4717 * dev_set_promiscuity - update promiscuity count on a device
4721 * Add or remove promiscuity from a device. While the count in the device
4722 * remains above zero the interface remains promiscuous. Once it hits zero
4723 * the device reverts back to normal filtering operation. A negative inc
4724 * value is used to drop promiscuity on the device.
4725 * Return 0 if successful or a negative errno code on error.
4727 int dev_set_promiscuity(struct net_device *dev, int inc)
4729 unsigned int old_flags = dev->flags;
4732 err = __dev_set_promiscuity(dev, inc);
4735 if (dev->flags != old_flags)
4736 dev_set_rx_mode(dev);
4739 EXPORT_SYMBOL(dev_set_promiscuity);
4742 * dev_set_allmulti - update allmulti count on a device
4746 * Add or remove reception of all multicast frames to a device. While the
4747 * count in the device remains above zero the interface remains listening
4748 * to all interfaces. Once it hits zero the device reverts back to normal
4749 * filtering operation. A negative @inc value is used to drop the counter
4750 * when releasing a resource needing all multicasts.
4751 * Return 0 if successful or a negative errno code on error.
4754 int dev_set_allmulti(struct net_device *dev, int inc)
4756 unsigned int old_flags = dev->flags;
4760 dev->flags |= IFF_ALLMULTI;
4761 dev->allmulti += inc;
4762 if (dev->allmulti == 0) {
4765 * If inc causes overflow, untouch allmulti and return error.
4768 dev->flags &= ~IFF_ALLMULTI;
4770 dev->allmulti -= inc;
4771 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4776 if (dev->flags ^ old_flags) {
4777 dev_change_rx_flags(dev, IFF_ALLMULTI);
4778 dev_set_rx_mode(dev);
4782 EXPORT_SYMBOL(dev_set_allmulti);
4785 * Upload unicast and multicast address lists to device and
4786 * configure RX filtering. When the device doesn't support unicast
4787 * filtering it is put in promiscuous mode while unicast addresses
4790 void __dev_set_rx_mode(struct net_device *dev)
4792 const struct net_device_ops *ops = dev->netdev_ops;
4794 /* dev_open will call this function so the list will stay sane. */
4795 if (!(dev->flags&IFF_UP))
4798 if (!netif_device_present(dev))
4801 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4802 /* Unicast addresses changes may only happen under the rtnl,
4803 * therefore calling __dev_set_promiscuity here is safe.
4805 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4806 __dev_set_promiscuity(dev, 1);
4807 dev->uc_promisc = true;
4808 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4809 __dev_set_promiscuity(dev, -1);
4810 dev->uc_promisc = false;
4814 if (ops->ndo_set_rx_mode)
4815 ops->ndo_set_rx_mode(dev);
4818 void dev_set_rx_mode(struct net_device *dev)
4820 netif_addr_lock_bh(dev);
4821 __dev_set_rx_mode(dev);
4822 netif_addr_unlock_bh(dev);
4826 * dev_get_flags - get flags reported to userspace
4829 * Get the combination of flag bits exported through APIs to userspace.
4831 unsigned int dev_get_flags(const struct net_device *dev)
4835 flags = (dev->flags & ~(IFF_PROMISC |
4840 (dev->gflags & (IFF_PROMISC |
4843 if (netif_running(dev)) {
4844 if (netif_oper_up(dev))
4845 flags |= IFF_RUNNING;
4846 if (netif_carrier_ok(dev))
4847 flags |= IFF_LOWER_UP;
4848 if (netif_dormant(dev))
4849 flags |= IFF_DORMANT;
4854 EXPORT_SYMBOL(dev_get_flags);
4856 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4858 unsigned int old_flags = dev->flags;
4864 * Set the flags on our device.
4867 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4868 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4870 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4874 * Load in the correct multicast list now the flags have changed.
4877 if ((old_flags ^ flags) & IFF_MULTICAST)
4878 dev_change_rx_flags(dev, IFF_MULTICAST);
4880 dev_set_rx_mode(dev);
4883 * Have we downed the interface. We handle IFF_UP ourselves
4884 * according to user attempts to set it, rather than blindly
4889 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4890 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4893 dev_set_rx_mode(dev);
4896 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4897 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4899 dev->gflags ^= IFF_PROMISC;
4900 dev_set_promiscuity(dev, inc);
4903 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4904 is important. Some (broken) drivers set IFF_PROMISC, when
4905 IFF_ALLMULTI is requested not asking us and not reporting.
4907 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4908 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4910 dev->gflags ^= IFF_ALLMULTI;
4911 dev_set_allmulti(dev, inc);
4917 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4919 unsigned int changes = dev->flags ^ old_flags;
4921 if (changes & IFF_UP) {
4922 if (dev->flags & IFF_UP)
4923 call_netdevice_notifiers(NETDEV_UP, dev);
4925 call_netdevice_notifiers(NETDEV_DOWN, dev);
4928 if (dev->flags & IFF_UP &&
4929 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4930 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4934 * dev_change_flags - change device settings
4936 * @flags: device state flags
4938 * Change settings on device based state flags. The flags are
4939 * in the userspace exported format.
4941 int dev_change_flags(struct net_device *dev, unsigned int flags)
4944 unsigned int changes, old_flags = dev->flags;
4946 ret = __dev_change_flags(dev, flags);
4950 changes = old_flags ^ dev->flags;
4952 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4954 __dev_notify_flags(dev, old_flags);
4957 EXPORT_SYMBOL(dev_change_flags);
4960 * dev_set_mtu - Change maximum transfer unit
4962 * @new_mtu: new transfer unit
4964 * Change the maximum transfer size of the network device.
4966 int dev_set_mtu(struct net_device *dev, int new_mtu)
4968 const struct net_device_ops *ops = dev->netdev_ops;
4971 if (new_mtu == dev->mtu)
4974 /* MTU must be positive. */
4978 if (!netif_device_present(dev))
4982 if (ops->ndo_change_mtu)
4983 err = ops->ndo_change_mtu(dev, new_mtu);
4988 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4991 EXPORT_SYMBOL(dev_set_mtu);
4994 * dev_set_group - Change group this device belongs to
4996 * @new_group: group this device should belong to
4998 void dev_set_group(struct net_device *dev, int new_group)
5000 dev->group = new_group;
5002 EXPORT_SYMBOL(dev_set_group);
5005 * dev_set_mac_address - Change Media Access Control Address
5009 * Change the hardware (MAC) address of the device
5011 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5013 const struct net_device_ops *ops = dev->netdev_ops;
5016 if (!ops->ndo_set_mac_address)
5018 if (sa->sa_family != dev->type)
5020 if (!netif_device_present(dev))
5022 err = ops->ndo_set_mac_address(dev, sa);
5024 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5025 add_device_randomness(dev->dev_addr, dev->addr_len);
5028 EXPORT_SYMBOL(dev_set_mac_address);
5031 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
5033 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
5036 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
5042 case SIOCGIFFLAGS: /* Get interface flags */
5043 ifr->ifr_flags = (short) dev_get_flags(dev);
5046 case SIOCGIFMETRIC: /* Get the metric on the interface
5047 (currently unused) */
5048 ifr->ifr_metric = 0;
5051 case SIOCGIFMTU: /* Get the MTU of a device */
5052 ifr->ifr_mtu = dev->mtu;
5057 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
5059 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
5060 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
5061 ifr->ifr_hwaddr.sa_family = dev->type;
5069 ifr->ifr_map.mem_start = dev->mem_start;
5070 ifr->ifr_map.mem_end = dev->mem_end;
5071 ifr->ifr_map.base_addr = dev->base_addr;
5072 ifr->ifr_map.irq = dev->irq;
5073 ifr->ifr_map.dma = dev->dma;
5074 ifr->ifr_map.port = dev->if_port;
5078 ifr->ifr_ifindex = dev->ifindex;
5082 ifr->ifr_qlen = dev->tx_queue_len;
5086 /* dev_ioctl() should ensure this case
5098 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
5100 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
5103 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
5104 const struct net_device_ops *ops;
5109 ops = dev->netdev_ops;
5112 case SIOCSIFFLAGS: /* Set interface flags */
5113 return dev_change_flags(dev, ifr->ifr_flags);
5115 case SIOCSIFMETRIC: /* Set the metric on the interface
5116 (currently unused) */
5119 case SIOCSIFMTU: /* Set the MTU of a device */
5120 return dev_set_mtu(dev, ifr->ifr_mtu);
5123 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
5125 case SIOCSIFHWBROADCAST:
5126 if (ifr->ifr_hwaddr.sa_family != dev->type)
5128 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
5129 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
5130 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5134 if (ops->ndo_set_config) {
5135 if (!netif_device_present(dev))
5137 return ops->ndo_set_config(dev, &ifr->ifr_map);
5142 if (!ops->ndo_set_rx_mode ||
5143 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5145 if (!netif_device_present(dev))
5147 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5150 if (!ops->ndo_set_rx_mode ||
5151 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5153 if (!netif_device_present(dev))
5155 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5158 if (ifr->ifr_qlen < 0)
5160 dev->tx_queue_len = ifr->ifr_qlen;
5164 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5165 return dev_change_name(dev, ifr->ifr_newname);
5168 err = net_hwtstamp_validate(ifr);
5174 * Unknown or private ioctl
5177 if ((cmd >= SIOCDEVPRIVATE &&
5178 cmd <= SIOCDEVPRIVATE + 15) ||
5179 cmd == SIOCBONDENSLAVE ||
5180 cmd == SIOCBONDRELEASE ||
5181 cmd == SIOCBONDSETHWADDR ||
5182 cmd == SIOCBONDSLAVEINFOQUERY ||
5183 cmd == SIOCBONDINFOQUERY ||
5184 cmd == SIOCBONDCHANGEACTIVE ||
5185 cmd == SIOCGMIIPHY ||
5186 cmd == SIOCGMIIREG ||
5187 cmd == SIOCSMIIREG ||
5188 cmd == SIOCBRADDIF ||
5189 cmd == SIOCBRDELIF ||
5190 cmd == SIOCSHWTSTAMP ||
5191 cmd == SIOCWANDEV) {
5193 if (ops->ndo_do_ioctl) {
5194 if (netif_device_present(dev))
5195 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5207 * This function handles all "interface"-type I/O control requests. The actual
5208 * 'doing' part of this is dev_ifsioc above.
5212 * dev_ioctl - network device ioctl
5213 * @net: the applicable net namespace
5214 * @cmd: command to issue
5215 * @arg: pointer to a struct ifreq in user space
5217 * Issue ioctl functions to devices. This is normally called by the
5218 * user space syscall interfaces but can sometimes be useful for
5219 * other purposes. The return value is the return from the syscall if
5220 * positive or a negative errno code on error.
5223 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5229 /* One special case: SIOCGIFCONF takes ifconf argument
5230 and requires shared lock, because it sleeps writing
5234 if (cmd == SIOCGIFCONF) {
5236 ret = dev_ifconf(net, (char __user *) arg);
5240 if (cmd == SIOCGIFNAME)
5241 return dev_ifname(net, (struct ifreq __user *)arg);
5243 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5246 ifr.ifr_name[IFNAMSIZ-1] = 0;
5248 colon = strchr(ifr.ifr_name, ':');
5253 * See which interface the caller is talking about.
5258 * These ioctl calls:
5259 * - can be done by all.
5260 * - atomic and do not require locking.
5271 dev_load(net, ifr.ifr_name);
5273 ret = dev_ifsioc_locked(net, &ifr, cmd);
5278 if (copy_to_user(arg, &ifr,
5279 sizeof(struct ifreq)))
5285 dev_load(net, ifr.ifr_name);
5287 ret = dev_ethtool(net, &ifr);
5292 if (copy_to_user(arg, &ifr,
5293 sizeof(struct ifreq)))
5299 * These ioctl calls:
5300 * - require superuser power.
5301 * - require strict serialization.
5307 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
5309 dev_load(net, ifr.ifr_name);
5311 ret = dev_ifsioc(net, &ifr, cmd);
5316 if (copy_to_user(arg, &ifr,
5317 sizeof(struct ifreq)))
5323 * These ioctl calls:
5324 * - require superuser power.
5325 * - require strict serialization.
5326 * - do not return a value
5330 if (!capable(CAP_NET_ADMIN))
5334 * These ioctl calls:
5335 * - require local superuser power.
5336 * - require strict serialization.
5337 * - do not return a value
5346 case SIOCSIFHWBROADCAST:
5348 case SIOCBONDENSLAVE:
5349 case SIOCBONDRELEASE:
5350 case SIOCBONDSETHWADDR:
5351 case SIOCBONDCHANGEACTIVE:
5355 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
5358 case SIOCBONDSLAVEINFOQUERY:
5359 case SIOCBONDINFOQUERY:
5360 dev_load(net, ifr.ifr_name);
5362 ret = dev_ifsioc(net, &ifr, cmd);
5367 /* Get the per device memory space. We can add this but
5368 * currently do not support it */
5370 /* Set the per device memory buffer space.
5371 * Not applicable in our case */
5376 * Unknown or private ioctl.
5379 if (cmd == SIOCWANDEV ||
5380 (cmd >= SIOCDEVPRIVATE &&
5381 cmd <= SIOCDEVPRIVATE + 15)) {
5382 dev_load(net, ifr.ifr_name);
5384 ret = dev_ifsioc(net, &ifr, cmd);
5386 if (!ret && copy_to_user(arg, &ifr,
5387 sizeof(struct ifreq)))
5391 /* Take care of Wireless Extensions */
5392 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5393 return wext_handle_ioctl(net, &ifr, cmd, arg);
5400 * dev_new_index - allocate an ifindex
5401 * @net: the applicable net namespace
5403 * Returns a suitable unique value for a new device interface
5404 * number. The caller must hold the rtnl semaphore or the
5405 * dev_base_lock to be sure it remains unique.
5407 static int dev_new_index(struct net *net)
5409 int ifindex = net->ifindex;
5413 if (!__dev_get_by_index(net, ifindex))
5414 return net->ifindex = ifindex;
5418 /* Delayed registration/unregisteration */
5419 static LIST_HEAD(net_todo_list);
5421 static void net_set_todo(struct net_device *dev)
5423 list_add_tail(&dev->todo_list, &net_todo_list);
5426 static void rollback_registered_many(struct list_head *head)
5428 struct net_device *dev, *tmp;
5430 BUG_ON(dev_boot_phase);
5433 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5434 /* Some devices call without registering
5435 * for initialization unwind. Remove those
5436 * devices and proceed with the remaining.
5438 if (dev->reg_state == NETREG_UNINITIALIZED) {
5439 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5443 list_del(&dev->unreg_list);
5446 dev->dismantle = true;
5447 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5450 /* If device is running, close it first. */
5451 dev_close_many(head);
5453 list_for_each_entry(dev, head, unreg_list) {
5454 /* And unlink it from device chain. */
5455 unlist_netdevice(dev);
5457 dev->reg_state = NETREG_UNREGISTERING;
5462 list_for_each_entry(dev, head, unreg_list) {
5463 /* Shutdown queueing discipline. */
5467 /* Notify protocols, that we are about to destroy
5468 this device. They should clean all the things.
5470 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5472 if (!dev->rtnl_link_ops ||
5473 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5474 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5477 * Flush the unicast and multicast chains
5482 if (dev->netdev_ops->ndo_uninit)
5483 dev->netdev_ops->ndo_uninit(dev);
5485 /* Notifier chain MUST detach us from master device. */
5486 WARN_ON(dev->master);
5488 /* Remove entries from kobject tree */
5489 netdev_unregister_kobject(dev);
5494 list_for_each_entry(dev, head, unreg_list)
5498 static void rollback_registered(struct net_device *dev)
5502 list_add(&dev->unreg_list, &single);
5503 rollback_registered_many(&single);
5507 static netdev_features_t netdev_fix_features(struct net_device *dev,
5508 netdev_features_t features)
5510 /* Fix illegal checksum combinations */
5511 if ((features & NETIF_F_HW_CSUM) &&
5512 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5513 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5514 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5517 /* Fix illegal SG+CSUM combinations. */
5518 if ((features & NETIF_F_SG) &&
5519 !(features & NETIF_F_ALL_CSUM)) {
5521 "Dropping NETIF_F_SG since no checksum feature.\n");
5522 features &= ~NETIF_F_SG;
5525 /* TSO requires that SG is present as well. */
5526 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5527 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5528 features &= ~NETIF_F_ALL_TSO;
5531 /* TSO ECN requires that TSO is present as well. */
5532 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5533 features &= ~NETIF_F_TSO_ECN;
5535 /* Software GSO depends on SG. */
5536 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5537 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5538 features &= ~NETIF_F_GSO;
5541 /* UFO needs SG and checksumming */
5542 if (features & NETIF_F_UFO) {
5543 /* maybe split UFO into V4 and V6? */
5544 if (!((features & NETIF_F_GEN_CSUM) ||
5545 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5546 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5548 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5549 features &= ~NETIF_F_UFO;
5552 if (!(features & NETIF_F_SG)) {
5554 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5555 features &= ~NETIF_F_UFO;
5562 int __netdev_update_features(struct net_device *dev)
5564 netdev_features_t features;
5569 features = netdev_get_wanted_features(dev);
5571 if (dev->netdev_ops->ndo_fix_features)
5572 features = dev->netdev_ops->ndo_fix_features(dev, features);
5574 /* driver might be less strict about feature dependencies */
5575 features = netdev_fix_features(dev, features);
5577 if (dev->features == features)
5580 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5581 &dev->features, &features);
5583 if (dev->netdev_ops->ndo_set_features)
5584 err = dev->netdev_ops->ndo_set_features(dev, features);
5586 if (unlikely(err < 0)) {
5588 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5589 err, &features, &dev->features);
5594 dev->features = features;
5600 * netdev_update_features - recalculate device features
5601 * @dev: the device to check
5603 * Recalculate dev->features set and send notifications if it
5604 * has changed. Should be called after driver or hardware dependent
5605 * conditions might have changed that influence the features.
5607 void netdev_update_features(struct net_device *dev)
5609 if (__netdev_update_features(dev))
5610 netdev_features_change(dev);
5612 EXPORT_SYMBOL(netdev_update_features);
5615 * netdev_change_features - recalculate device features
5616 * @dev: the device to check
5618 * Recalculate dev->features set and send notifications even
5619 * if they have not changed. Should be called instead of
5620 * netdev_update_features() if also dev->vlan_features might
5621 * have changed to allow the changes to be propagated to stacked
5624 void netdev_change_features(struct net_device *dev)
5626 __netdev_update_features(dev);
5627 netdev_features_change(dev);
5629 EXPORT_SYMBOL(netdev_change_features);
5632 * netif_stacked_transfer_operstate - transfer operstate
5633 * @rootdev: the root or lower level device to transfer state from
5634 * @dev: the device to transfer operstate to
5636 * Transfer operational state from root to device. This is normally
5637 * called when a stacking relationship exists between the root
5638 * device and the device(a leaf device).
5640 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5641 struct net_device *dev)
5643 if (rootdev->operstate == IF_OPER_DORMANT)
5644 netif_dormant_on(dev);
5646 netif_dormant_off(dev);
5648 if (netif_carrier_ok(rootdev)) {
5649 if (!netif_carrier_ok(dev))
5650 netif_carrier_on(dev);
5652 if (netif_carrier_ok(dev))
5653 netif_carrier_off(dev);
5656 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5659 static int netif_alloc_rx_queues(struct net_device *dev)
5661 unsigned int i, count = dev->num_rx_queues;
5662 struct netdev_rx_queue *rx;
5666 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5668 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5673 for (i = 0; i < count; i++)
5679 static void netdev_init_one_queue(struct net_device *dev,
5680 struct netdev_queue *queue, void *_unused)
5682 /* Initialize queue lock */
5683 spin_lock_init(&queue->_xmit_lock);
5684 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5685 queue->xmit_lock_owner = -1;
5686 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5689 dql_init(&queue->dql, HZ);
5693 static int netif_alloc_netdev_queues(struct net_device *dev)
5695 unsigned int count = dev->num_tx_queues;
5696 struct netdev_queue *tx;
5700 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5702 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5707 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5708 spin_lock_init(&dev->tx_global_lock);
5714 * register_netdevice - register a network device
5715 * @dev: device to register
5717 * Take a completed network device structure and add it to the kernel
5718 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5719 * chain. 0 is returned on success. A negative errno code is returned
5720 * on a failure to set up the device, or if the name is a duplicate.
5722 * Callers must hold the rtnl semaphore. You may want
5723 * register_netdev() instead of this.
5726 * The locking appears insufficient to guarantee two parallel registers
5727 * will not get the same name.
5730 int register_netdevice(struct net_device *dev)
5733 struct net *net = dev_net(dev);
5735 BUG_ON(dev_boot_phase);
5740 /* When net_device's are persistent, this will be fatal. */
5741 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5744 spin_lock_init(&dev->addr_list_lock);
5745 netdev_set_addr_lockdep_class(dev);
5749 ret = dev_get_valid_name(net, dev, dev->name);
5753 /* Init, if this function is available */
5754 if (dev->netdev_ops->ndo_init) {
5755 ret = dev->netdev_ops->ndo_init(dev);
5765 dev->ifindex = dev_new_index(net);
5766 else if (__dev_get_by_index(net, dev->ifindex))
5769 if (dev->iflink == -1)
5770 dev->iflink = dev->ifindex;
5772 /* Transfer changeable features to wanted_features and enable
5773 * software offloads (GSO and GRO).
5775 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5776 dev->features |= NETIF_F_SOFT_FEATURES;
5777 dev->wanted_features = dev->features & dev->hw_features;
5779 /* Turn on no cache copy if HW is doing checksum */
5780 if (!(dev->flags & IFF_LOOPBACK)) {
5781 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5782 if (dev->features & NETIF_F_ALL_CSUM) {
5783 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5784 dev->features |= NETIF_F_NOCACHE_COPY;
5788 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5790 dev->vlan_features |= NETIF_F_HIGHDMA;
5792 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5793 ret = notifier_to_errno(ret);
5797 ret = netdev_register_kobject(dev);
5800 dev->reg_state = NETREG_REGISTERED;
5802 __netdev_update_features(dev);
5805 * Default initial state at registry is that the
5806 * device is present.
5809 set_bit(__LINK_STATE_PRESENT, &dev->state);
5811 linkwatch_init_dev(dev);
5813 dev_init_scheduler(dev);
5815 list_netdevice(dev);
5816 add_device_randomness(dev->dev_addr, dev->addr_len);
5818 /* Notify protocols, that a new device appeared. */
5819 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5820 ret = notifier_to_errno(ret);
5822 rollback_registered(dev);
5823 dev->reg_state = NETREG_UNREGISTERED;
5826 * Prevent userspace races by waiting until the network
5827 * device is fully setup before sending notifications.
5829 if (!dev->rtnl_link_ops ||
5830 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5831 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5837 if (dev->netdev_ops->ndo_uninit)
5838 dev->netdev_ops->ndo_uninit(dev);
5841 EXPORT_SYMBOL(register_netdevice);
5844 * init_dummy_netdev - init a dummy network device for NAPI
5845 * @dev: device to init
5847 * This takes a network device structure and initialize the minimum
5848 * amount of fields so it can be used to schedule NAPI polls without
5849 * registering a full blown interface. This is to be used by drivers
5850 * that need to tie several hardware interfaces to a single NAPI
5851 * poll scheduler due to HW limitations.
5853 int init_dummy_netdev(struct net_device *dev)
5855 /* Clear everything. Note we don't initialize spinlocks
5856 * are they aren't supposed to be taken by any of the
5857 * NAPI code and this dummy netdev is supposed to be
5858 * only ever used for NAPI polls
5860 memset(dev, 0, sizeof(struct net_device));
5862 /* make sure we BUG if trying to hit standard
5863 * register/unregister code path
5865 dev->reg_state = NETREG_DUMMY;
5867 /* NAPI wants this */
5868 INIT_LIST_HEAD(&dev->napi_list);
5870 /* a dummy interface is started by default */
5871 set_bit(__LINK_STATE_PRESENT, &dev->state);
5872 set_bit(__LINK_STATE_START, &dev->state);
5874 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5875 * because users of this 'device' dont need to change
5881 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5885 * register_netdev - register a network device
5886 * @dev: device to register
5888 * Take a completed network device structure and add it to the kernel
5889 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5890 * chain. 0 is returned on success. A negative errno code is returned
5891 * on a failure to set up the device, or if the name is a duplicate.
5893 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5894 * and expands the device name if you passed a format string to
5897 int register_netdev(struct net_device *dev)
5902 err = register_netdevice(dev);
5906 EXPORT_SYMBOL(register_netdev);
5908 int netdev_refcnt_read(const struct net_device *dev)
5912 for_each_possible_cpu(i)
5913 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5916 EXPORT_SYMBOL(netdev_refcnt_read);
5919 * netdev_wait_allrefs - wait until all references are gone.
5920 * @dev: target net_device
5922 * This is called when unregistering network devices.
5924 * Any protocol or device that holds a reference should register
5925 * for netdevice notification, and cleanup and put back the
5926 * reference if they receive an UNREGISTER event.
5927 * We can get stuck here if buggy protocols don't correctly
5930 static void netdev_wait_allrefs(struct net_device *dev)
5932 unsigned long rebroadcast_time, warning_time;
5935 linkwatch_forget_dev(dev);
5937 rebroadcast_time = warning_time = jiffies;
5938 refcnt = netdev_refcnt_read(dev);
5940 while (refcnt != 0) {
5941 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5944 /* Rebroadcast unregister notification */
5945 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5951 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5952 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5954 /* We must not have linkwatch events
5955 * pending on unregister. If this
5956 * happens, we simply run the queue
5957 * unscheduled, resulting in a noop
5960 linkwatch_run_queue();
5965 rebroadcast_time = jiffies;
5970 refcnt = netdev_refcnt_read(dev);
5972 if (time_after(jiffies, warning_time + 10 * HZ)) {
5973 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5975 warning_time = jiffies;
5984 * register_netdevice(x1);
5985 * register_netdevice(x2);
5987 * unregister_netdevice(y1);
5988 * unregister_netdevice(y2);
5994 * We are invoked by rtnl_unlock().
5995 * This allows us to deal with problems:
5996 * 1) We can delete sysfs objects which invoke hotplug
5997 * without deadlocking with linkwatch via keventd.
5998 * 2) Since we run with the RTNL semaphore not held, we can sleep
5999 * safely in order to wait for the netdev refcnt to drop to zero.
6001 * We must not return until all unregister events added during
6002 * the interval the lock was held have been completed.
6004 void netdev_run_todo(void)
6006 struct list_head list;
6008 /* Snapshot list, allow later requests */
6009 list_replace_init(&net_todo_list, &list);
6014 /* Wait for rcu callbacks to finish before next phase */
6015 if (!list_empty(&list))
6018 while (!list_empty(&list)) {
6019 struct net_device *dev
6020 = list_first_entry(&list, struct net_device, todo_list);
6021 list_del(&dev->todo_list);
6024 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6027 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6028 pr_err("network todo '%s' but state %d\n",
6029 dev->name, dev->reg_state);
6034 dev->reg_state = NETREG_UNREGISTERED;
6036 on_each_cpu(flush_backlog, dev, 1);
6038 netdev_wait_allrefs(dev);
6041 BUG_ON(netdev_refcnt_read(dev));
6042 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6043 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6044 WARN_ON(dev->dn_ptr);
6046 if (dev->destructor)
6047 dev->destructor(dev);
6049 /* Free network device */
6050 kobject_put(&dev->dev.kobj);
6054 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6055 * fields in the same order, with only the type differing.
6057 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6058 const struct net_device_stats *netdev_stats)
6060 #if BITS_PER_LONG == 64
6061 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6062 memcpy(stats64, netdev_stats, sizeof(*stats64));
6064 size_t i, n = sizeof(*stats64) / sizeof(u64);
6065 const unsigned long *src = (const unsigned long *)netdev_stats;
6066 u64 *dst = (u64 *)stats64;
6068 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6069 sizeof(*stats64) / sizeof(u64));
6070 for (i = 0; i < n; i++)
6074 EXPORT_SYMBOL(netdev_stats_to_stats64);
6077 * dev_get_stats - get network device statistics
6078 * @dev: device to get statistics from
6079 * @storage: place to store stats
6081 * Get network statistics from device. Return @storage.
6082 * The device driver may provide its own method by setting
6083 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6084 * otherwise the internal statistics structure is used.
6086 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6087 struct rtnl_link_stats64 *storage)
6089 const struct net_device_ops *ops = dev->netdev_ops;
6091 if (ops->ndo_get_stats64) {
6092 memset(storage, 0, sizeof(*storage));
6093 ops->ndo_get_stats64(dev, storage);
6094 } else if (ops->ndo_get_stats) {
6095 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6097 netdev_stats_to_stats64(storage, &dev->stats);
6099 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6102 EXPORT_SYMBOL(dev_get_stats);
6104 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6106 struct netdev_queue *queue = dev_ingress_queue(dev);
6108 #ifdef CONFIG_NET_CLS_ACT
6111 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6114 netdev_init_one_queue(dev, queue, NULL);
6115 queue->qdisc = &noop_qdisc;
6116 queue->qdisc_sleeping = &noop_qdisc;
6117 rcu_assign_pointer(dev->ingress_queue, queue);
6122 static const struct ethtool_ops default_ethtool_ops;
6124 void netdev_set_default_ethtool_ops(struct net_device *dev,
6125 const struct ethtool_ops *ops)
6127 if (dev->ethtool_ops == &default_ethtool_ops)
6128 dev->ethtool_ops = ops;
6130 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6133 * alloc_netdev_mqs - allocate network device
6134 * @sizeof_priv: size of private data to allocate space for
6135 * @name: device name format string
6136 * @setup: callback to initialize device
6137 * @txqs: the number of TX subqueues to allocate
6138 * @rxqs: the number of RX subqueues to allocate
6140 * Allocates a struct net_device with private data area for driver use
6141 * and performs basic initialization. Also allocates subquue structs
6142 * for each queue on the device.
6144 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6145 void (*setup)(struct net_device *),
6146 unsigned int txqs, unsigned int rxqs)
6148 struct net_device *dev;
6150 struct net_device *p;
6152 BUG_ON(strlen(name) >= sizeof(dev->name));
6155 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6161 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6166 alloc_size = sizeof(struct net_device);
6168 /* ensure 32-byte alignment of private area */
6169 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6170 alloc_size += sizeof_priv;
6172 /* ensure 32-byte alignment of whole construct */
6173 alloc_size += NETDEV_ALIGN - 1;
6175 p = kzalloc(alloc_size, GFP_KERNEL);
6177 pr_err("alloc_netdev: Unable to allocate device\n");
6181 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6182 dev->padded = (char *)dev - (char *)p;
6184 dev->pcpu_refcnt = alloc_percpu(int);
6185 if (!dev->pcpu_refcnt)
6188 if (dev_addr_init(dev))
6194 dev_net_set(dev, &init_net);
6196 dev->gso_max_size = GSO_MAX_SIZE;
6197 dev->gso_max_segs = GSO_MAX_SEGS;
6199 INIT_LIST_HEAD(&dev->napi_list);
6200 INIT_LIST_HEAD(&dev->unreg_list);
6201 INIT_LIST_HEAD(&dev->link_watch_list);
6202 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6205 dev->num_tx_queues = txqs;
6206 dev->real_num_tx_queues = txqs;
6207 if (netif_alloc_netdev_queues(dev))
6211 dev->num_rx_queues = rxqs;
6212 dev->real_num_rx_queues = rxqs;
6213 if (netif_alloc_rx_queues(dev))
6217 strcpy(dev->name, name);
6218 dev->group = INIT_NETDEV_GROUP;
6219 if (!dev->ethtool_ops)
6220 dev->ethtool_ops = &default_ethtool_ops;
6228 free_percpu(dev->pcpu_refcnt);
6238 EXPORT_SYMBOL(alloc_netdev_mqs);
6241 * free_netdev - free network device
6244 * This function does the last stage of destroying an allocated device
6245 * interface. The reference to the device object is released.
6246 * If this is the last reference then it will be freed.
6248 void free_netdev(struct net_device *dev)
6250 struct napi_struct *p, *n;
6252 release_net(dev_net(dev));
6259 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6261 /* Flush device addresses */
6262 dev_addr_flush(dev);
6264 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6267 free_percpu(dev->pcpu_refcnt);
6268 dev->pcpu_refcnt = NULL;
6270 /* Compatibility with error handling in drivers */
6271 if (dev->reg_state == NETREG_UNINITIALIZED) {
6272 kfree((char *)dev - dev->padded);
6276 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6277 dev->reg_state = NETREG_RELEASED;
6279 /* will free via device release */
6280 put_device(&dev->dev);
6282 EXPORT_SYMBOL(free_netdev);
6285 * synchronize_net - Synchronize with packet receive processing
6287 * Wait for packets currently being received to be done.
6288 * Does not block later packets from starting.
6290 void synchronize_net(void)
6293 if (rtnl_is_locked())
6294 synchronize_rcu_expedited();
6298 EXPORT_SYMBOL(synchronize_net);
6301 * unregister_netdevice_queue - remove device from the kernel
6305 * This function shuts down a device interface and removes it
6306 * from the kernel tables.
6307 * If head not NULL, device is queued to be unregistered later.
6309 * Callers must hold the rtnl semaphore. You may want
6310 * unregister_netdev() instead of this.
6313 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6318 list_move_tail(&dev->unreg_list, head);
6320 rollback_registered(dev);
6321 /* Finish processing unregister after unlock */
6325 EXPORT_SYMBOL(unregister_netdevice_queue);
6328 * unregister_netdevice_many - unregister many devices
6329 * @head: list of devices
6331 void unregister_netdevice_many(struct list_head *head)
6333 struct net_device *dev;
6335 if (!list_empty(head)) {
6336 rollback_registered_many(head);
6337 list_for_each_entry(dev, head, unreg_list)
6341 EXPORT_SYMBOL(unregister_netdevice_many);
6344 * unregister_netdev - remove device from the kernel
6347 * This function shuts down a device interface and removes it
6348 * from the kernel tables.
6350 * This is just a wrapper for unregister_netdevice that takes
6351 * the rtnl semaphore. In general you want to use this and not
6352 * unregister_netdevice.
6354 void unregister_netdev(struct net_device *dev)
6357 unregister_netdevice(dev);
6360 EXPORT_SYMBOL(unregister_netdev);
6363 * dev_change_net_namespace - move device to different nethost namespace
6365 * @net: network namespace
6366 * @pat: If not NULL name pattern to try if the current device name
6367 * is already taken in the destination network namespace.
6369 * This function shuts down a device interface and moves it
6370 * to a new network namespace. On success 0 is returned, on
6371 * a failure a netagive errno code is returned.
6373 * Callers must hold the rtnl semaphore.
6376 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6382 /* Don't allow namespace local devices to be moved. */
6384 if (dev->features & NETIF_F_NETNS_LOCAL)
6387 /* Ensure the device has been registrered */
6388 if (dev->reg_state != NETREG_REGISTERED)
6391 /* Get out if there is nothing todo */
6393 if (net_eq(dev_net(dev), net))
6396 /* Pick the destination device name, and ensure
6397 * we can use it in the destination network namespace.
6400 if (__dev_get_by_name(net, dev->name)) {
6401 /* We get here if we can't use the current device name */
6404 if (dev_get_valid_name(net, dev, pat) < 0)
6409 * And now a mini version of register_netdevice unregister_netdevice.
6412 /* If device is running close it first. */
6415 /* And unlink it from device chain */
6417 unlist_netdevice(dev);
6421 /* Shutdown queueing discipline. */
6424 /* Notify protocols, that we are about to destroy
6425 this device. They should clean all the things.
6427 Note that dev->reg_state stays at NETREG_REGISTERED.
6428 This is wanted because this way 8021q and macvlan know
6429 the device is just moving and can keep their slaves up.
6431 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6433 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6434 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6437 * Flush the unicast and multicast chains
6442 /* Send a netdev-removed uevent to the old namespace */
6443 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6445 /* Actually switch the network namespace */
6446 dev_net_set(dev, net);
6448 /* If there is an ifindex conflict assign a new one */
6449 if (__dev_get_by_index(net, dev->ifindex)) {
6450 int iflink = (dev->iflink == dev->ifindex);
6451 dev->ifindex = dev_new_index(net);
6453 dev->iflink = dev->ifindex;
6456 /* Send a netdev-add uevent to the new namespace */
6457 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6459 /* Fixup kobjects */
6460 err = device_rename(&dev->dev, dev->name);
6463 /* Add the device back in the hashes */
6464 list_netdevice(dev);
6466 /* Notify protocols, that a new device appeared. */
6467 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6470 * Prevent userspace races by waiting until the network
6471 * device is fully setup before sending notifications.
6473 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6480 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6482 static int dev_cpu_callback(struct notifier_block *nfb,
6483 unsigned long action,
6486 struct sk_buff **list_skb;
6487 struct sk_buff *skb;
6488 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6489 struct softnet_data *sd, *oldsd;
6491 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6494 local_irq_disable();
6495 cpu = smp_processor_id();
6496 sd = &per_cpu(softnet_data, cpu);
6497 oldsd = &per_cpu(softnet_data, oldcpu);
6499 /* Find end of our completion_queue. */
6500 list_skb = &sd->completion_queue;
6502 list_skb = &(*list_skb)->next;
6503 /* Append completion queue from offline CPU. */
6504 *list_skb = oldsd->completion_queue;
6505 oldsd->completion_queue = NULL;
6507 /* Append output queue from offline CPU. */
6508 if (oldsd->output_queue) {
6509 *sd->output_queue_tailp = oldsd->output_queue;
6510 sd->output_queue_tailp = oldsd->output_queue_tailp;
6511 oldsd->output_queue = NULL;
6512 oldsd->output_queue_tailp = &oldsd->output_queue;
6514 /* Append NAPI poll list from offline CPU. */
6515 if (!list_empty(&oldsd->poll_list)) {
6516 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6517 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6520 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6523 /* Process offline CPU's input_pkt_queue */
6524 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6526 input_queue_head_incr(oldsd);
6528 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6530 input_queue_head_incr(oldsd);
6538 * netdev_increment_features - increment feature set by one
6539 * @all: current feature set
6540 * @one: new feature set
6541 * @mask: mask feature set
6543 * Computes a new feature set after adding a device with feature set
6544 * @one to the master device with current feature set @all. Will not
6545 * enable anything that is off in @mask. Returns the new feature set.
6547 netdev_features_t netdev_increment_features(netdev_features_t all,
6548 netdev_features_t one, netdev_features_t mask)
6550 if (mask & NETIF_F_GEN_CSUM)
6551 mask |= NETIF_F_ALL_CSUM;
6552 mask |= NETIF_F_VLAN_CHALLENGED;
6554 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6555 all &= one | ~NETIF_F_ALL_FOR_ALL;
6557 /* If one device supports hw checksumming, set for all. */
6558 if (all & NETIF_F_GEN_CSUM)
6559 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6563 EXPORT_SYMBOL(netdev_increment_features);
6565 static struct hlist_head *netdev_create_hash(void)
6568 struct hlist_head *hash;
6570 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6572 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6573 INIT_HLIST_HEAD(&hash[i]);
6578 /* Initialize per network namespace state */
6579 static int __net_init netdev_init(struct net *net)
6581 if (net != &init_net)
6582 INIT_LIST_HEAD(&net->dev_base_head);
6584 net->dev_name_head = netdev_create_hash();
6585 if (net->dev_name_head == NULL)
6588 net->dev_index_head = netdev_create_hash();
6589 if (net->dev_index_head == NULL)
6595 kfree(net->dev_name_head);
6601 * netdev_drivername - network driver for the device
6602 * @dev: network device
6604 * Determine network driver for device.
6606 const char *netdev_drivername(const struct net_device *dev)
6608 const struct device_driver *driver;
6609 const struct device *parent;
6610 const char *empty = "";
6612 parent = dev->dev.parent;
6616 driver = parent->driver;
6617 if (driver && driver->name)
6618 return driver->name;
6622 static int __netdev_printk(const char *level, const struct net_device *dev,
6623 struct va_format *vaf)
6627 if (dev && dev->dev.parent) {
6628 r = dev_printk_emit(level[1] - '0',
6631 dev_driver_string(dev->dev.parent),
6632 dev_name(dev->dev.parent),
6633 netdev_name(dev), vaf);
6635 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6637 r = printk("%s(NULL net_device): %pV", level, vaf);
6643 int netdev_printk(const char *level, const struct net_device *dev,
6644 const char *format, ...)
6646 struct va_format vaf;
6650 va_start(args, format);
6655 r = __netdev_printk(level, dev, &vaf);
6661 EXPORT_SYMBOL(netdev_printk);
6663 #define define_netdev_printk_level(func, level) \
6664 int func(const struct net_device *dev, const char *fmt, ...) \
6667 struct va_format vaf; \
6670 va_start(args, fmt); \
6675 r = __netdev_printk(level, dev, &vaf); \
6681 EXPORT_SYMBOL(func);
6683 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6684 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6685 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6686 define_netdev_printk_level(netdev_err, KERN_ERR);
6687 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6688 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6689 define_netdev_printk_level(netdev_info, KERN_INFO);
6691 static void __net_exit netdev_exit(struct net *net)
6693 kfree(net->dev_name_head);
6694 kfree(net->dev_index_head);
6697 static struct pernet_operations __net_initdata netdev_net_ops = {
6698 .init = netdev_init,
6699 .exit = netdev_exit,
6702 static void __net_exit default_device_exit(struct net *net)
6704 struct net_device *dev, *aux;
6706 * Push all migratable network devices back to the
6707 * initial network namespace
6710 for_each_netdev_safe(net, dev, aux) {
6712 char fb_name[IFNAMSIZ];
6714 /* Ignore unmoveable devices (i.e. loopback) */
6715 if (dev->features & NETIF_F_NETNS_LOCAL)
6718 /* Leave virtual devices for the generic cleanup */
6719 if (dev->rtnl_link_ops)
6722 /* Push remaining network devices to init_net */
6723 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6724 err = dev_change_net_namespace(dev, &init_net, fb_name);
6726 pr_emerg("%s: failed to move %s to init_net: %d\n",
6727 __func__, dev->name, err);
6734 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6736 /* At exit all network devices most be removed from a network
6737 * namespace. Do this in the reverse order of registration.
6738 * Do this across as many network namespaces as possible to
6739 * improve batching efficiency.
6741 struct net_device *dev;
6743 LIST_HEAD(dev_kill_list);
6746 list_for_each_entry(net, net_list, exit_list) {
6747 for_each_netdev_reverse(net, dev) {
6748 if (dev->rtnl_link_ops)
6749 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6751 unregister_netdevice_queue(dev, &dev_kill_list);
6754 unregister_netdevice_many(&dev_kill_list);
6755 list_del(&dev_kill_list);
6759 static struct pernet_operations __net_initdata default_device_ops = {
6760 .exit = default_device_exit,
6761 .exit_batch = default_device_exit_batch,
6765 * Initialize the DEV module. At boot time this walks the device list and
6766 * unhooks any devices that fail to initialise (normally hardware not
6767 * present) and leaves us with a valid list of present and active devices.
6772 * This is called single threaded during boot, so no need
6773 * to take the rtnl semaphore.
6775 static int __init net_dev_init(void)
6777 int i, rc = -ENOMEM;
6779 BUG_ON(!dev_boot_phase);
6781 if (dev_proc_init())
6784 if (netdev_kobject_init())
6787 INIT_LIST_HEAD(&ptype_all);
6788 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6789 INIT_LIST_HEAD(&ptype_base[i]);
6791 INIT_LIST_HEAD(&offload_base);
6793 if (register_pernet_subsys(&netdev_net_ops))
6797 * Initialise the packet receive queues.
6800 for_each_possible_cpu(i) {
6801 struct softnet_data *sd = &per_cpu(softnet_data, i);
6803 memset(sd, 0, sizeof(*sd));
6804 skb_queue_head_init(&sd->input_pkt_queue);
6805 skb_queue_head_init(&sd->process_queue);
6806 sd->completion_queue = NULL;
6807 INIT_LIST_HEAD(&sd->poll_list);
6808 sd->output_queue = NULL;
6809 sd->output_queue_tailp = &sd->output_queue;
6811 sd->csd.func = rps_trigger_softirq;
6817 sd->backlog.poll = process_backlog;
6818 sd->backlog.weight = weight_p;
6819 sd->backlog.gro_list = NULL;
6820 sd->backlog.gro_count = 0;
6825 /* The loopback device is special if any other network devices
6826 * is present in a network namespace the loopback device must
6827 * be present. Since we now dynamically allocate and free the
6828 * loopback device ensure this invariant is maintained by
6829 * keeping the loopback device as the first device on the
6830 * list of network devices. Ensuring the loopback devices
6831 * is the first device that appears and the last network device
6834 if (register_pernet_device(&loopback_net_ops))
6837 if (register_pernet_device(&default_device_ops))
6840 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6841 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6843 hotcpu_notifier(dev_cpu_callback, 0);
6851 subsys_initcall(net_dev_init);
6853 static int __init initialize_hashrnd(void)
6855 get_random_bytes(&hashrnd, sizeof(hashrnd));
6859 late_initcall_sync(initialize_hashrnd);