2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/net_tstamp.h>
136 #include <linux/static_key.h>
137 #include <net/flow_keys.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 * The list of packet types we will receive (as opposed to discard)
149 * and the routines to invoke.
151 * Why 16. Because with 16 the only overlap we get on a hash of the
152 * low nibble of the protocol value is RARP/SNAP/X.25.
154 * NOTE: That is no longer true with the addition of VLAN tags. Not
155 * sure which should go first, but I bet it won't make much
156 * difference if we are running VLANs. The good news is that
157 * this protocol won't be in the list unless compiled in, so
158 * the average user (w/out VLANs) will not be adversely affected.
175 #define PTYPE_HASH_SIZE (16)
176 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
178 static DEFINE_SPINLOCK(ptype_lock);
179 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
180 static struct list_head ptype_all __read_mostly; /* Taps */
183 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
186 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
188 * Writers must hold the rtnl semaphore while they loop through the
189 * dev_base_head list, and hold dev_base_lock for writing when they do the
190 * actual updates. This allows pure readers to access the list even
191 * while a writer is preparing to update it.
193 * To put it another way, dev_base_lock is held for writing only to
194 * protect against pure readers; the rtnl semaphore provides the
195 * protection against other writers.
197 * See, for example usages, register_netdevice() and
198 * unregister_netdevice(), which must be called with the rtnl
201 DEFINE_RWLOCK(dev_base_lock);
202 EXPORT_SYMBOL(dev_base_lock);
204 static inline void dev_base_seq_inc(struct net *net)
206 while (++net->dev_base_seq == 0);
209 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
211 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
213 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
216 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
218 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
221 static inline void rps_lock(struct softnet_data *sd)
224 spin_lock(&sd->input_pkt_queue.lock);
228 static inline void rps_unlock(struct softnet_data *sd)
231 spin_unlock(&sd->input_pkt_queue.lock);
235 /* Device list insertion */
236 static int list_netdevice(struct net_device *dev)
238 struct net *net = dev_net(dev);
242 write_lock_bh(&dev_base_lock);
243 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
244 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
245 hlist_add_head_rcu(&dev->index_hlist,
246 dev_index_hash(net, dev->ifindex));
247 write_unlock_bh(&dev_base_lock);
249 dev_base_seq_inc(net);
254 /* Device list removal
255 * caller must respect a RCU grace period before freeing/reusing dev
257 static void unlist_netdevice(struct net_device *dev)
261 /* Unlink dev from the device chain */
262 write_lock_bh(&dev_base_lock);
263 list_del_rcu(&dev->dev_list);
264 hlist_del_rcu(&dev->name_hlist);
265 hlist_del_rcu(&dev->index_hlist);
266 write_unlock_bh(&dev_base_lock);
268 dev_base_seq_inc(dev_net(dev));
275 static RAW_NOTIFIER_HEAD(netdev_chain);
278 * Device drivers call our routines to queue packets here. We empty the
279 * queue in the local softnet handler.
282 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
283 EXPORT_PER_CPU_SYMBOL(softnet_data);
285 #ifdef CONFIG_LOCKDEP
287 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
288 * according to dev->type
290 static const unsigned short netdev_lock_type[] =
291 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
292 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
293 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
294 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
295 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
296 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
297 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
298 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
299 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
300 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
301 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
302 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
303 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
304 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
305 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
307 static const char *const netdev_lock_name[] =
308 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
309 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
310 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
311 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
312 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
313 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
314 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
315 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
316 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
317 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
318 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
319 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
320 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
321 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
322 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
324 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
325 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
327 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
331 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
332 if (netdev_lock_type[i] == dev_type)
334 /* the last key is used by default */
335 return ARRAY_SIZE(netdev_lock_type) - 1;
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
343 i = netdev_lock_pos(dev_type);
344 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
345 netdev_lock_name[i]);
348 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
352 i = netdev_lock_pos(dev->type);
353 lockdep_set_class_and_name(&dev->addr_list_lock,
354 &netdev_addr_lock_key[i],
355 netdev_lock_name[i]);
358 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
359 unsigned short dev_type)
362 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
367 /*******************************************************************************
369 Protocol management and registration routines
371 *******************************************************************************/
374 * Add a protocol ID to the list. Now that the input handler is
375 * smarter we can dispense with all the messy stuff that used to be
378 * BEWARE!!! Protocol handlers, mangling input packets,
379 * MUST BE last in hash buckets and checking protocol handlers
380 * MUST start from promiscuous ptype_all chain in net_bh.
381 * It is true now, do not change it.
382 * Explanation follows: if protocol handler, mangling packet, will
383 * be the first on list, it is not able to sense, that packet
384 * is cloned and should be copied-on-write, so that it will
385 * change it and subsequent readers will get broken packet.
389 static inline struct list_head *ptype_head(const struct packet_type *pt)
391 if (pt->type == htons(ETH_P_ALL))
394 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
398 * dev_add_pack - add packet handler
399 * @pt: packet type declaration
401 * Add a protocol handler to the networking stack. The passed &packet_type
402 * is linked into kernel lists and may not be freed until it has been
403 * removed from the kernel lists.
405 * This call does not sleep therefore it can not
406 * guarantee all CPU's that are in middle of receiving packets
407 * will see the new packet type (until the next received packet).
410 void dev_add_pack(struct packet_type *pt)
412 struct list_head *head = ptype_head(pt);
414 spin_lock(&ptype_lock);
415 list_add_rcu(&pt->list, head);
416 spin_unlock(&ptype_lock);
418 EXPORT_SYMBOL(dev_add_pack);
421 * __dev_remove_pack - remove packet handler
422 * @pt: packet type declaration
424 * Remove a protocol handler that was previously added to the kernel
425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
426 * from the kernel lists and can be freed or reused once this function
429 * The packet type might still be in use by receivers
430 * and must not be freed until after all the CPU's have gone
431 * through a quiescent state.
433 void __dev_remove_pack(struct packet_type *pt)
435 struct list_head *head = ptype_head(pt);
436 struct packet_type *pt1;
438 spin_lock(&ptype_lock);
440 list_for_each_entry(pt1, head, list) {
442 list_del_rcu(&pt->list);
447 pr_warn("dev_remove_pack: %p not found\n", pt);
449 spin_unlock(&ptype_lock);
451 EXPORT_SYMBOL(__dev_remove_pack);
454 * dev_remove_pack - remove packet handler
455 * @pt: packet type declaration
457 * Remove a protocol handler that was previously added to the kernel
458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
459 * from the kernel lists and can be freed or reused once this function
462 * This call sleeps to guarantee that no CPU is looking at the packet
465 void dev_remove_pack(struct packet_type *pt)
467 __dev_remove_pack(pt);
471 EXPORT_SYMBOL(dev_remove_pack);
473 /******************************************************************************
475 Device Boot-time Settings Routines
477 *******************************************************************************/
479 /* Boot time configuration table */
480 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
483 * netdev_boot_setup_add - add new setup entry
484 * @name: name of the device
485 * @map: configured settings for the device
487 * Adds new setup entry to the dev_boot_setup list. The function
488 * returns 0 on error and 1 on success. This is a generic routine to
491 static int netdev_boot_setup_add(char *name, struct ifmap *map)
493 struct netdev_boot_setup *s;
497 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
498 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
499 memset(s[i].name, 0, sizeof(s[i].name));
500 strlcpy(s[i].name, name, IFNAMSIZ);
501 memcpy(&s[i].map, map, sizeof(s[i].map));
506 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
510 * netdev_boot_setup_check - check boot time settings
511 * @dev: the netdevice
513 * Check boot time settings for the device.
514 * The found settings are set for the device to be used
515 * later in the device probing.
516 * Returns 0 if no settings found, 1 if they are.
518 int netdev_boot_setup_check(struct net_device *dev)
520 struct netdev_boot_setup *s = dev_boot_setup;
523 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
524 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
525 !strcmp(dev->name, s[i].name)) {
526 dev->irq = s[i].map.irq;
527 dev->base_addr = s[i].map.base_addr;
528 dev->mem_start = s[i].map.mem_start;
529 dev->mem_end = s[i].map.mem_end;
535 EXPORT_SYMBOL(netdev_boot_setup_check);
539 * netdev_boot_base - get address from boot time settings
540 * @prefix: prefix for network device
541 * @unit: id for network device
543 * Check boot time settings for the base address of device.
544 * The found settings are set for the device to be used
545 * later in the device probing.
546 * Returns 0 if no settings found.
548 unsigned long netdev_boot_base(const char *prefix, int unit)
550 const struct netdev_boot_setup *s = dev_boot_setup;
554 sprintf(name, "%s%d", prefix, unit);
557 * If device already registered then return base of 1
558 * to indicate not to probe for this interface
560 if (__dev_get_by_name(&init_net, name))
563 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
564 if (!strcmp(name, s[i].name))
565 return s[i].map.base_addr;
570 * Saves at boot time configured settings for any netdevice.
572 int __init netdev_boot_setup(char *str)
577 str = get_options(str, ARRAY_SIZE(ints), ints);
582 memset(&map, 0, sizeof(map));
586 map.base_addr = ints[2];
588 map.mem_start = ints[3];
590 map.mem_end = ints[4];
592 /* Add new entry to the list */
593 return netdev_boot_setup_add(str, &map);
596 __setup("netdev=", netdev_boot_setup);
598 /*******************************************************************************
600 Device Interface Subroutines
602 *******************************************************************************/
605 * __dev_get_by_name - find a device by its name
606 * @net: the applicable net namespace
607 * @name: name to find
609 * Find an interface by name. Must be called under RTNL semaphore
610 * or @dev_base_lock. If the name is found a pointer to the device
611 * is returned. If the name is not found then %NULL is returned. The
612 * reference counters are not incremented so the caller must be
613 * careful with locks.
616 struct net_device *__dev_get_by_name(struct net *net, const char *name)
618 struct hlist_node *p;
619 struct net_device *dev;
620 struct hlist_head *head = dev_name_hash(net, name);
622 hlist_for_each_entry(dev, p, head, name_hlist)
623 if (!strncmp(dev->name, name, IFNAMSIZ))
628 EXPORT_SYMBOL(__dev_get_by_name);
631 * dev_get_by_name_rcu - find a device by its name
632 * @net: the applicable net namespace
633 * @name: name to find
635 * Find an interface by name.
636 * If the name is found a pointer to the device is returned.
637 * If the name is not found then %NULL is returned.
638 * The reference counters are not incremented so the caller must be
639 * careful with locks. The caller must hold RCU lock.
642 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
644 struct hlist_node *p;
645 struct net_device *dev;
646 struct hlist_head *head = dev_name_hash(net, name);
648 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
649 if (!strncmp(dev->name, name, IFNAMSIZ))
654 EXPORT_SYMBOL(dev_get_by_name_rcu);
657 * dev_get_by_name - find a device by its name
658 * @net: the applicable net namespace
659 * @name: name to find
661 * Find an interface by name. This can be called from any
662 * context and does its own locking. The returned handle has
663 * the usage count incremented and the caller must use dev_put() to
664 * release it when it is no longer needed. %NULL is returned if no
665 * matching device is found.
668 struct net_device *dev_get_by_name(struct net *net, const char *name)
670 struct net_device *dev;
673 dev = dev_get_by_name_rcu(net, name);
679 EXPORT_SYMBOL(dev_get_by_name);
682 * __dev_get_by_index - find a device by its ifindex
683 * @net: the applicable net namespace
684 * @ifindex: index of device
686 * Search for an interface by index. Returns %NULL if the device
687 * is not found or a pointer to the device. The device has not
688 * had its reference counter increased so the caller must be careful
689 * about locking. The caller must hold either the RTNL semaphore
693 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
695 struct hlist_node *p;
696 struct net_device *dev;
697 struct hlist_head *head = dev_index_hash(net, ifindex);
699 hlist_for_each_entry(dev, p, head, index_hlist)
700 if (dev->ifindex == ifindex)
705 EXPORT_SYMBOL(__dev_get_by_index);
708 * dev_get_by_index_rcu - find a device by its ifindex
709 * @net: the applicable net namespace
710 * @ifindex: index of device
712 * Search for an interface by index. Returns %NULL if the device
713 * is not found or a pointer to the device. The device has not
714 * had its reference counter increased so the caller must be careful
715 * about locking. The caller must hold RCU lock.
718 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
720 struct hlist_node *p;
721 struct net_device *dev;
722 struct hlist_head *head = dev_index_hash(net, ifindex);
724 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
725 if (dev->ifindex == ifindex)
730 EXPORT_SYMBOL(dev_get_by_index_rcu);
734 * dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns NULL if the device
739 * is not found or a pointer to the device. The device returned has
740 * had a reference added and the pointer is safe until the user calls
741 * dev_put to indicate they have finished with it.
744 struct net_device *dev_get_by_index(struct net *net, int ifindex)
746 struct net_device *dev;
749 dev = dev_get_by_index_rcu(net, ifindex);
755 EXPORT_SYMBOL(dev_get_by_index);
758 * dev_getbyhwaddr_rcu - find a device by its hardware address
759 * @net: the applicable net namespace
760 * @type: media type of device
761 * @ha: hardware address
763 * Search for an interface by MAC address. Returns NULL if the device
764 * is not found or a pointer to the device.
765 * The caller must hold RCU or RTNL.
766 * The returned device has not had its ref count increased
767 * and the caller must therefore be careful about locking
771 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
774 struct net_device *dev;
776 for_each_netdev_rcu(net, dev)
777 if (dev->type == type &&
778 !memcmp(dev->dev_addr, ha, dev->addr_len))
783 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
785 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
787 struct net_device *dev;
790 for_each_netdev(net, dev)
791 if (dev->type == type)
796 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
798 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
800 struct net_device *dev, *ret = NULL;
803 for_each_netdev_rcu(net, dev)
804 if (dev->type == type) {
812 EXPORT_SYMBOL(dev_getfirstbyhwtype);
815 * dev_get_by_flags_rcu - find any device with given flags
816 * @net: the applicable net namespace
817 * @if_flags: IFF_* values
818 * @mask: bitmask of bits in if_flags to check
820 * Search for any interface with the given flags. Returns NULL if a device
821 * is not found or a pointer to the device. Must be called inside
822 * rcu_read_lock(), and result refcount is unchanged.
825 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
828 struct net_device *dev, *ret;
831 for_each_netdev_rcu(net, dev) {
832 if (((dev->flags ^ if_flags) & mask) == 0) {
839 EXPORT_SYMBOL(dev_get_by_flags_rcu);
842 * dev_valid_name - check if name is okay for network device
845 * Network device names need to be valid file names to
846 * to allow sysfs to work. We also disallow any kind of
849 bool dev_valid_name(const char *name)
853 if (strlen(name) >= IFNAMSIZ)
855 if (!strcmp(name, ".") || !strcmp(name, ".."))
859 if (*name == '/' || isspace(*name))
865 EXPORT_SYMBOL(dev_valid_name);
868 * __dev_alloc_name - allocate a name for a device
869 * @net: network namespace to allocate the device name in
870 * @name: name format string
871 * @buf: scratch buffer and result name string
873 * Passed a format string - eg "lt%d" it will try and find a suitable
874 * id. It scans list of devices to build up a free map, then chooses
875 * the first empty slot. The caller must hold the dev_base or rtnl lock
876 * while allocating the name and adding the device in order to avoid
878 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
879 * Returns the number of the unit assigned or a negative errno code.
882 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
886 const int max_netdevices = 8*PAGE_SIZE;
887 unsigned long *inuse;
888 struct net_device *d;
890 p = strnchr(name, IFNAMSIZ-1, '%');
893 * Verify the string as this thing may have come from
894 * the user. There must be either one "%d" and no other "%"
897 if (p[1] != 'd' || strchr(p + 2, '%'))
900 /* Use one page as a bit array of possible slots */
901 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
905 for_each_netdev(net, d) {
906 if (!sscanf(d->name, name, &i))
908 if (i < 0 || i >= max_netdevices)
911 /* avoid cases where sscanf is not exact inverse of printf */
912 snprintf(buf, IFNAMSIZ, name, i);
913 if (!strncmp(buf, d->name, IFNAMSIZ))
917 i = find_first_zero_bit(inuse, max_netdevices);
918 free_page((unsigned long) inuse);
922 snprintf(buf, IFNAMSIZ, name, i);
923 if (!__dev_get_by_name(net, buf))
926 /* It is possible to run out of possible slots
927 * when the name is long and there isn't enough space left
928 * for the digits, or if all bits are used.
934 * dev_alloc_name - allocate a name for a device
936 * @name: name format string
938 * Passed a format string - eg "lt%d" it will try and find a suitable
939 * id. It scans list of devices to build up a free map, then chooses
940 * the first empty slot. The caller must hold the dev_base or rtnl lock
941 * while allocating the name and adding the device in order to avoid
943 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
944 * Returns the number of the unit assigned or a negative errno code.
947 int dev_alloc_name(struct net_device *dev, const char *name)
953 BUG_ON(!dev_net(dev));
955 ret = __dev_alloc_name(net, name, buf);
957 strlcpy(dev->name, buf, IFNAMSIZ);
960 EXPORT_SYMBOL(dev_alloc_name);
962 static int dev_get_valid_name(struct net_device *dev, const char *name)
966 BUG_ON(!dev_net(dev));
969 if (!dev_valid_name(name))
972 if (strchr(name, '%'))
973 return dev_alloc_name(dev, name);
974 else if (__dev_get_by_name(net, name))
976 else if (dev->name != name)
977 strlcpy(dev->name, name, IFNAMSIZ);
983 * dev_change_name - change name of a device
985 * @newname: name (or format string) must be at least IFNAMSIZ
987 * Change name of a device, can pass format strings "eth%d".
990 int dev_change_name(struct net_device *dev, const char *newname)
992 char oldname[IFNAMSIZ];
998 BUG_ON(!dev_net(dev));
1001 if (dev->flags & IFF_UP)
1004 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1007 memcpy(oldname, dev->name, IFNAMSIZ);
1009 err = dev_get_valid_name(dev, newname);
1014 ret = device_rename(&dev->dev, dev->name);
1016 memcpy(dev->name, oldname, IFNAMSIZ);
1020 write_lock_bh(&dev_base_lock);
1021 hlist_del_rcu(&dev->name_hlist);
1022 write_unlock_bh(&dev_base_lock);
1026 write_lock_bh(&dev_base_lock);
1027 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1028 write_unlock_bh(&dev_base_lock);
1030 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1031 ret = notifier_to_errno(ret);
1034 /* err >= 0 after dev_alloc_name() or stores the first errno */
1037 memcpy(dev->name, oldname, IFNAMSIZ);
1040 pr_err("%s: name change rollback failed: %d\n",
1049 * dev_set_alias - change ifalias of a device
1051 * @alias: name up to IFALIASZ
1052 * @len: limit of bytes to copy from info
1054 * Set ifalias for a device,
1056 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1062 if (len >= IFALIASZ)
1067 kfree(dev->ifalias);
1068 dev->ifalias = NULL;
1073 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1076 dev->ifalias = new_ifalias;
1078 strlcpy(dev->ifalias, alias, len+1);
1084 * netdev_features_change - device changes features
1085 * @dev: device to cause notification
1087 * Called to indicate a device has changed features.
1089 void netdev_features_change(struct net_device *dev)
1091 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1093 EXPORT_SYMBOL(netdev_features_change);
1096 * netdev_state_change - device changes state
1097 * @dev: device to cause notification
1099 * Called to indicate a device has changed state. This function calls
1100 * the notifier chains for netdev_chain and sends a NEWLINK message
1101 * to the routing socket.
1103 void netdev_state_change(struct net_device *dev)
1105 if (dev->flags & IFF_UP) {
1106 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1107 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1110 EXPORT_SYMBOL(netdev_state_change);
1113 * netdev_notify_peers - notify network peers about existence of @dev
1114 * @dev: network device
1116 * Generate traffic such that interested network peers are aware of
1117 * @dev, such as by generating a gratuitous ARP. This may be used when
1118 * a device wants to inform the rest of the network about some sort of
1119 * reconfiguration such as a failover event or virtual machine
1122 void netdev_notify_peers(struct net_device *dev)
1125 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1128 EXPORT_SYMBOL(netdev_notify_peers);
1131 * dev_load - load a network module
1132 * @net: the applicable net namespace
1133 * @name: name of interface
1135 * If a network interface is not present and the process has suitable
1136 * privileges this function loads the module. If module loading is not
1137 * available in this kernel then it becomes a nop.
1140 void dev_load(struct net *net, const char *name)
1142 struct net_device *dev;
1146 dev = dev_get_by_name_rcu(net, name);
1150 if (no_module && capable(CAP_NET_ADMIN))
1151 no_module = request_module("netdev-%s", name);
1152 if (no_module && capable(CAP_SYS_MODULE)) {
1153 if (!request_module("%s", name))
1154 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1158 EXPORT_SYMBOL(dev_load);
1160 static int __dev_open(struct net_device *dev)
1162 const struct net_device_ops *ops = dev->netdev_ops;
1167 if (!netif_device_present(dev))
1170 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1171 ret = notifier_to_errno(ret);
1175 set_bit(__LINK_STATE_START, &dev->state);
1177 if (ops->ndo_validate_addr)
1178 ret = ops->ndo_validate_addr(dev);
1180 if (!ret && ops->ndo_open)
1181 ret = ops->ndo_open(dev);
1184 clear_bit(__LINK_STATE_START, &dev->state);
1186 dev->flags |= IFF_UP;
1187 net_dmaengine_get();
1188 dev_set_rx_mode(dev);
1190 add_device_randomness(dev->dev_addr, dev->addr_len);
1197 * dev_open - prepare an interface for use.
1198 * @dev: device to open
1200 * Takes a device from down to up state. The device's private open
1201 * function is invoked and then the multicast lists are loaded. Finally
1202 * the device is moved into the up state and a %NETDEV_UP message is
1203 * sent to the netdev notifier chain.
1205 * Calling this function on an active interface is a nop. On a failure
1206 * a negative errno code is returned.
1208 int dev_open(struct net_device *dev)
1212 if (dev->flags & IFF_UP)
1215 ret = __dev_open(dev);
1219 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1220 call_netdevice_notifiers(NETDEV_UP, dev);
1224 EXPORT_SYMBOL(dev_open);
1226 static int __dev_close_many(struct list_head *head)
1228 struct net_device *dev;
1233 list_for_each_entry(dev, head, unreg_list) {
1234 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1236 clear_bit(__LINK_STATE_START, &dev->state);
1238 /* Synchronize to scheduled poll. We cannot touch poll list, it
1239 * can be even on different cpu. So just clear netif_running().
1241 * dev->stop() will invoke napi_disable() on all of it's
1242 * napi_struct instances on this device.
1244 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1247 dev_deactivate_many(head);
1249 list_for_each_entry(dev, head, unreg_list) {
1250 const struct net_device_ops *ops = dev->netdev_ops;
1253 * Call the device specific close. This cannot fail.
1254 * Only if device is UP
1256 * We allow it to be called even after a DETACH hot-plug
1262 dev->flags &= ~IFF_UP;
1263 net_dmaengine_put();
1269 static int __dev_close(struct net_device *dev)
1274 list_add(&dev->unreg_list, &single);
1275 retval = __dev_close_many(&single);
1280 static int dev_close_many(struct list_head *head)
1282 struct net_device *dev, *tmp;
1283 LIST_HEAD(tmp_list);
1285 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1286 if (!(dev->flags & IFF_UP))
1287 list_move(&dev->unreg_list, &tmp_list);
1289 __dev_close_many(head);
1291 list_for_each_entry(dev, head, unreg_list) {
1292 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1293 call_netdevice_notifiers(NETDEV_DOWN, dev);
1296 /* rollback_registered_many needs the complete original list */
1297 list_splice(&tmp_list, head);
1302 * dev_close - shutdown an interface.
1303 * @dev: device to shutdown
1305 * This function moves an active device into down state. A
1306 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1307 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1310 int dev_close(struct net_device *dev)
1312 if (dev->flags & IFF_UP) {
1315 list_add(&dev->unreg_list, &single);
1316 dev_close_many(&single);
1321 EXPORT_SYMBOL(dev_close);
1325 * dev_disable_lro - disable Large Receive Offload on a device
1328 * Disable Large Receive Offload (LRO) on a net device. Must be
1329 * called under RTNL. This is needed if received packets may be
1330 * forwarded to another interface.
1332 void dev_disable_lro(struct net_device *dev)
1335 * If we're trying to disable lro on a vlan device
1336 * use the underlying physical device instead
1338 if (is_vlan_dev(dev))
1339 dev = vlan_dev_real_dev(dev);
1341 dev->wanted_features &= ~NETIF_F_LRO;
1342 netdev_update_features(dev);
1344 if (unlikely(dev->features & NETIF_F_LRO))
1345 netdev_WARN(dev, "failed to disable LRO!\n");
1347 EXPORT_SYMBOL(dev_disable_lro);
1350 static int dev_boot_phase = 1;
1353 * register_netdevice_notifier - register a network notifier block
1356 * Register a notifier to be called when network device events occur.
1357 * The notifier passed is linked into the kernel structures and must
1358 * not be reused until it has been unregistered. A negative errno code
1359 * is returned on a failure.
1361 * When registered all registration and up events are replayed
1362 * to the new notifier to allow device to have a race free
1363 * view of the network device list.
1366 int register_netdevice_notifier(struct notifier_block *nb)
1368 struct net_device *dev;
1369 struct net_device *last;
1374 err = raw_notifier_chain_register(&netdev_chain, nb);
1380 for_each_netdev(net, dev) {
1381 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1382 err = notifier_to_errno(err);
1386 if (!(dev->flags & IFF_UP))
1389 nb->notifier_call(nb, NETDEV_UP, dev);
1400 for_each_netdev(net, dev) {
1404 if (dev->flags & IFF_UP) {
1405 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1406 nb->notifier_call(nb, NETDEV_DOWN, dev);
1408 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1413 raw_notifier_chain_unregister(&netdev_chain, nb);
1416 EXPORT_SYMBOL(register_netdevice_notifier);
1419 * unregister_netdevice_notifier - unregister a network notifier block
1422 * Unregister a notifier previously registered by
1423 * register_netdevice_notifier(). The notifier is unlinked into the
1424 * kernel structures and may then be reused. A negative errno code
1425 * is returned on a failure.
1427 * After unregistering unregister and down device events are synthesized
1428 * for all devices on the device list to the removed notifier to remove
1429 * the need for special case cleanup code.
1432 int unregister_netdevice_notifier(struct notifier_block *nb)
1434 struct net_device *dev;
1439 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1444 for_each_netdev(net, dev) {
1445 if (dev->flags & IFF_UP) {
1446 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1447 nb->notifier_call(nb, NETDEV_DOWN, dev);
1449 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1456 EXPORT_SYMBOL(unregister_netdevice_notifier);
1459 * call_netdevice_notifiers - call all network notifier blocks
1460 * @val: value passed unmodified to notifier function
1461 * @dev: net_device pointer passed unmodified to notifier function
1463 * Call all network notifier blocks. Parameters and return value
1464 * are as for raw_notifier_call_chain().
1467 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1470 return raw_notifier_call_chain(&netdev_chain, val, dev);
1472 EXPORT_SYMBOL(call_netdevice_notifiers);
1474 static struct static_key netstamp_needed __read_mostly;
1475 #ifdef HAVE_JUMP_LABEL
1476 /* We are not allowed to call static_key_slow_dec() from irq context
1477 * If net_disable_timestamp() is called from irq context, defer the
1478 * static_key_slow_dec() calls.
1480 static atomic_t netstamp_needed_deferred;
1483 void net_enable_timestamp(void)
1485 #ifdef HAVE_JUMP_LABEL
1486 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1490 static_key_slow_dec(&netstamp_needed);
1494 WARN_ON(in_interrupt());
1495 static_key_slow_inc(&netstamp_needed);
1497 EXPORT_SYMBOL(net_enable_timestamp);
1499 void net_disable_timestamp(void)
1501 #ifdef HAVE_JUMP_LABEL
1502 if (in_interrupt()) {
1503 atomic_inc(&netstamp_needed_deferred);
1507 static_key_slow_dec(&netstamp_needed);
1509 EXPORT_SYMBOL(net_disable_timestamp);
1511 static inline void net_timestamp_set(struct sk_buff *skb)
1513 skb->tstamp.tv64 = 0;
1514 if (static_key_false(&netstamp_needed))
1515 __net_timestamp(skb);
1518 #define net_timestamp_check(COND, SKB) \
1519 if (static_key_false(&netstamp_needed)) { \
1520 if ((COND) && !(SKB)->tstamp.tv64) \
1521 __net_timestamp(SKB); \
1524 static int net_hwtstamp_validate(struct ifreq *ifr)
1526 struct hwtstamp_config cfg;
1527 enum hwtstamp_tx_types tx_type;
1528 enum hwtstamp_rx_filters rx_filter;
1529 int tx_type_valid = 0;
1530 int rx_filter_valid = 0;
1532 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1535 if (cfg.flags) /* reserved for future extensions */
1538 tx_type = cfg.tx_type;
1539 rx_filter = cfg.rx_filter;
1542 case HWTSTAMP_TX_OFF:
1543 case HWTSTAMP_TX_ON:
1544 case HWTSTAMP_TX_ONESTEP_SYNC:
1549 switch (rx_filter) {
1550 case HWTSTAMP_FILTER_NONE:
1551 case HWTSTAMP_FILTER_ALL:
1552 case HWTSTAMP_FILTER_SOME:
1553 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1554 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1555 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1556 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1557 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1558 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1559 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1560 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1561 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1562 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1563 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1564 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1565 rx_filter_valid = 1;
1569 if (!tx_type_valid || !rx_filter_valid)
1575 static inline bool is_skb_forwardable(struct net_device *dev,
1576 struct sk_buff *skb)
1580 if (!(dev->flags & IFF_UP))
1583 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1584 if (skb->len <= len)
1587 /* if TSO is enabled, we don't care about the length as the packet
1588 * could be forwarded without being segmented before
1590 if (skb_is_gso(skb))
1597 * dev_forward_skb - loopback an skb to another netif
1599 * @dev: destination network device
1600 * @skb: buffer to forward
1603 * NET_RX_SUCCESS (no congestion)
1604 * NET_RX_DROP (packet was dropped, but freed)
1606 * dev_forward_skb can be used for injecting an skb from the
1607 * start_xmit function of one device into the receive queue
1608 * of another device.
1610 * The receiving device may be in another namespace, so
1611 * we have to clear all information in the skb that could
1612 * impact namespace isolation.
1614 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1616 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1617 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1618 atomic_long_inc(&dev->rx_dropped);
1627 if (unlikely(!is_skb_forwardable(dev, skb))) {
1628 atomic_long_inc(&dev->rx_dropped);
1635 skb->tstamp.tv64 = 0;
1636 skb->pkt_type = PACKET_HOST;
1637 skb->protocol = eth_type_trans(skb, dev);
1641 return netif_rx(skb);
1643 EXPORT_SYMBOL_GPL(dev_forward_skb);
1645 static inline int deliver_skb(struct sk_buff *skb,
1646 struct packet_type *pt_prev,
1647 struct net_device *orig_dev)
1649 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1651 atomic_inc(&skb->users);
1652 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1655 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1657 if (ptype->af_packet_priv == NULL)
1660 if (ptype->id_match)
1661 return ptype->id_match(ptype, skb->sk);
1662 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1669 * Support routine. Sends outgoing frames to any network
1670 * taps currently in use.
1673 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1675 struct packet_type *ptype;
1676 struct sk_buff *skb2 = NULL;
1677 struct packet_type *pt_prev = NULL;
1680 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1681 /* Never send packets back to the socket
1682 * they originated from - MvS (miquels@drinkel.ow.org)
1684 if ((ptype->dev == dev || !ptype->dev) &&
1685 (!skb_loop_sk(ptype, skb))) {
1687 deliver_skb(skb2, pt_prev, skb->dev);
1692 skb2 = skb_clone(skb, GFP_ATOMIC);
1696 net_timestamp_set(skb2);
1698 /* skb->nh should be correctly
1699 set by sender, so that the second statement is
1700 just protection against buggy protocols.
1702 skb_reset_mac_header(skb2);
1704 if (skb_network_header(skb2) < skb2->data ||
1705 skb2->network_header > skb2->tail) {
1706 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1707 ntohs(skb2->protocol),
1709 skb_reset_network_header(skb2);
1712 skb2->transport_header = skb2->network_header;
1713 skb2->pkt_type = PACKET_OUTGOING;
1718 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1723 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1724 * @dev: Network device
1725 * @txq: number of queues available
1727 * If real_num_tx_queues is changed the tc mappings may no longer be
1728 * valid. To resolve this verify the tc mapping remains valid and if
1729 * not NULL the mapping. With no priorities mapping to this
1730 * offset/count pair it will no longer be used. In the worst case TC0
1731 * is invalid nothing can be done so disable priority mappings. If is
1732 * expected that drivers will fix this mapping if they can before
1733 * calling netif_set_real_num_tx_queues.
1735 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1738 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1740 /* If TC0 is invalidated disable TC mapping */
1741 if (tc->offset + tc->count > txq) {
1742 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1747 /* Invalidated prio to tc mappings set to TC0 */
1748 for (i = 1; i < TC_BITMASK + 1; i++) {
1749 int q = netdev_get_prio_tc_map(dev, i);
1751 tc = &dev->tc_to_txq[q];
1752 if (tc->offset + tc->count > txq) {
1753 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1755 netdev_set_prio_tc_map(dev, i, 0);
1761 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1762 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1764 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1768 if (txq < 1 || txq > dev->num_tx_queues)
1771 if (dev->reg_state == NETREG_REGISTERED ||
1772 dev->reg_state == NETREG_UNREGISTERING) {
1775 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1781 netif_setup_tc(dev, txq);
1783 if (txq < dev->real_num_tx_queues)
1784 qdisc_reset_all_tx_gt(dev, txq);
1787 dev->real_num_tx_queues = txq;
1790 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1794 * netif_set_real_num_rx_queues - set actual number of RX queues used
1795 * @dev: Network device
1796 * @rxq: Actual number of RX queues
1798 * This must be called either with the rtnl_lock held or before
1799 * registration of the net device. Returns 0 on success, or a
1800 * negative error code. If called before registration, it always
1803 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1807 if (rxq < 1 || rxq > dev->num_rx_queues)
1810 if (dev->reg_state == NETREG_REGISTERED) {
1813 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1819 dev->real_num_rx_queues = rxq;
1822 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1826 * netif_get_num_default_rss_queues - default number of RSS queues
1828 * This routine should set an upper limit on the number of RSS queues
1829 * used by default by multiqueue devices.
1831 int netif_get_num_default_rss_queues(void)
1833 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
1835 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
1837 static inline void __netif_reschedule(struct Qdisc *q)
1839 struct softnet_data *sd;
1840 unsigned long flags;
1842 local_irq_save(flags);
1843 sd = &__get_cpu_var(softnet_data);
1844 q->next_sched = NULL;
1845 *sd->output_queue_tailp = q;
1846 sd->output_queue_tailp = &q->next_sched;
1847 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1848 local_irq_restore(flags);
1851 void __netif_schedule(struct Qdisc *q)
1853 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1854 __netif_reschedule(q);
1856 EXPORT_SYMBOL(__netif_schedule);
1858 void dev_kfree_skb_irq(struct sk_buff *skb)
1860 if (atomic_dec_and_test(&skb->users)) {
1861 struct softnet_data *sd;
1862 unsigned long flags;
1864 local_irq_save(flags);
1865 sd = &__get_cpu_var(softnet_data);
1866 skb->next = sd->completion_queue;
1867 sd->completion_queue = skb;
1868 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1869 local_irq_restore(flags);
1872 EXPORT_SYMBOL(dev_kfree_skb_irq);
1874 void dev_kfree_skb_any(struct sk_buff *skb)
1876 if (in_irq() || irqs_disabled())
1877 dev_kfree_skb_irq(skb);
1881 EXPORT_SYMBOL(dev_kfree_skb_any);
1885 * netif_device_detach - mark device as removed
1886 * @dev: network device
1888 * Mark device as removed from system and therefore no longer available.
1890 void netif_device_detach(struct net_device *dev)
1892 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1893 netif_running(dev)) {
1894 netif_tx_stop_all_queues(dev);
1897 EXPORT_SYMBOL(netif_device_detach);
1900 * netif_device_attach - mark device as attached
1901 * @dev: network device
1903 * Mark device as attached from system and restart if needed.
1905 void netif_device_attach(struct net_device *dev)
1907 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1908 netif_running(dev)) {
1909 netif_tx_wake_all_queues(dev);
1910 __netdev_watchdog_up(dev);
1913 EXPORT_SYMBOL(netif_device_attach);
1915 static void skb_warn_bad_offload(const struct sk_buff *skb)
1917 static const netdev_features_t null_features = 0;
1918 struct net_device *dev = skb->dev;
1919 const char *driver = "";
1921 if (dev && dev->dev.parent)
1922 driver = dev_driver_string(dev->dev.parent);
1924 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1925 "gso_type=%d ip_summed=%d\n",
1926 driver, dev ? &dev->features : &null_features,
1927 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1928 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1929 skb_shinfo(skb)->gso_type, skb->ip_summed);
1933 * Invalidate hardware checksum when packet is to be mangled, and
1934 * complete checksum manually on outgoing path.
1936 int skb_checksum_help(struct sk_buff *skb)
1939 int ret = 0, offset;
1941 if (skb->ip_summed == CHECKSUM_COMPLETE)
1942 goto out_set_summed;
1944 if (unlikely(skb_shinfo(skb)->gso_size)) {
1945 skb_warn_bad_offload(skb);
1949 offset = skb_checksum_start_offset(skb);
1950 BUG_ON(offset >= skb_headlen(skb));
1951 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1953 offset += skb->csum_offset;
1954 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1956 if (skb_cloned(skb) &&
1957 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1958 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1963 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1965 skb->ip_summed = CHECKSUM_NONE;
1969 EXPORT_SYMBOL(skb_checksum_help);
1972 * skb_gso_segment - Perform segmentation on skb.
1973 * @skb: buffer to segment
1974 * @features: features for the output path (see dev->features)
1976 * This function segments the given skb and returns a list of segments.
1978 * It may return NULL if the skb requires no segmentation. This is
1979 * only possible when GSO is used for verifying header integrity.
1981 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1982 netdev_features_t features)
1984 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1985 struct packet_type *ptype;
1986 __be16 type = skb->protocol;
1987 int vlan_depth = ETH_HLEN;
1990 while (type == htons(ETH_P_8021Q)) {
1991 struct vlan_hdr *vh;
1993 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1994 return ERR_PTR(-EINVAL);
1996 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1997 type = vh->h_vlan_encapsulated_proto;
1998 vlan_depth += VLAN_HLEN;
2001 skb_reset_mac_header(skb);
2002 skb->mac_len = skb->network_header - skb->mac_header;
2003 __skb_pull(skb, skb->mac_len);
2005 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2006 skb_warn_bad_offload(skb);
2008 if (skb_header_cloned(skb) &&
2009 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2010 return ERR_PTR(err);
2014 list_for_each_entry_rcu(ptype,
2015 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2016 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
2017 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2018 err = ptype->gso_send_check(skb);
2019 segs = ERR_PTR(err);
2020 if (err || skb_gso_ok(skb, features))
2022 __skb_push(skb, (skb->data -
2023 skb_network_header(skb)));
2025 segs = ptype->gso_segment(skb, features);
2031 __skb_push(skb, skb->data - skb_mac_header(skb));
2035 EXPORT_SYMBOL(skb_gso_segment);
2037 /* Take action when hardware reception checksum errors are detected. */
2039 void netdev_rx_csum_fault(struct net_device *dev)
2041 if (net_ratelimit()) {
2042 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2046 EXPORT_SYMBOL(netdev_rx_csum_fault);
2049 /* Actually, we should eliminate this check as soon as we know, that:
2050 * 1. IOMMU is present and allows to map all the memory.
2051 * 2. No high memory really exists on this machine.
2054 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2056 #ifdef CONFIG_HIGHMEM
2058 if (!(dev->features & NETIF_F_HIGHDMA)) {
2059 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2060 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2061 if (PageHighMem(skb_frag_page(frag)))
2066 if (PCI_DMA_BUS_IS_PHYS) {
2067 struct device *pdev = dev->dev.parent;
2071 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2072 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2073 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2074 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2083 void (*destructor)(struct sk_buff *skb);
2086 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2088 static void dev_gso_skb_destructor(struct sk_buff *skb)
2090 struct dev_gso_cb *cb;
2093 struct sk_buff *nskb = skb->next;
2095 skb->next = nskb->next;
2098 } while (skb->next);
2100 cb = DEV_GSO_CB(skb);
2102 cb->destructor(skb);
2106 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2107 * @skb: buffer to segment
2108 * @features: device features as applicable to this skb
2110 * This function segments the given skb and stores the list of segments
2113 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2115 struct sk_buff *segs;
2117 segs = skb_gso_segment(skb, features);
2119 /* Verifying header integrity only. */
2124 return PTR_ERR(segs);
2127 DEV_GSO_CB(skb)->destructor = skb->destructor;
2128 skb->destructor = dev_gso_skb_destructor;
2133 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2135 return ((features & NETIF_F_GEN_CSUM) ||
2136 ((features & NETIF_F_V4_CSUM) &&
2137 protocol == htons(ETH_P_IP)) ||
2138 ((features & NETIF_F_V6_CSUM) &&
2139 protocol == htons(ETH_P_IPV6)) ||
2140 ((features & NETIF_F_FCOE_CRC) &&
2141 protocol == htons(ETH_P_FCOE)));
2144 static netdev_features_t harmonize_features(struct sk_buff *skb,
2145 __be16 protocol, netdev_features_t features)
2147 if (!can_checksum_protocol(features, protocol)) {
2148 features &= ~NETIF_F_ALL_CSUM;
2149 features &= ~NETIF_F_SG;
2150 } else if (illegal_highdma(skb->dev, skb)) {
2151 features &= ~NETIF_F_SG;
2157 netdev_features_t netif_skb_features(struct sk_buff *skb)
2159 __be16 protocol = skb->protocol;
2160 netdev_features_t features = skb->dev->features;
2162 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2163 features &= ~NETIF_F_GSO_MASK;
2165 if (protocol == htons(ETH_P_8021Q)) {
2166 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2167 protocol = veh->h_vlan_encapsulated_proto;
2168 } else if (!vlan_tx_tag_present(skb)) {
2169 return harmonize_features(skb, protocol, features);
2172 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2174 if (protocol != htons(ETH_P_8021Q)) {
2175 return harmonize_features(skb, protocol, features);
2177 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2178 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2179 return harmonize_features(skb, protocol, features);
2182 EXPORT_SYMBOL(netif_skb_features);
2185 * Returns true if either:
2186 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2187 * 2. skb is fragmented and the device does not support SG.
2189 static inline int skb_needs_linearize(struct sk_buff *skb,
2192 return skb_is_nonlinear(skb) &&
2193 ((skb_has_frag_list(skb) &&
2194 !(features & NETIF_F_FRAGLIST)) ||
2195 (skb_shinfo(skb)->nr_frags &&
2196 !(features & NETIF_F_SG)));
2199 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2200 struct netdev_queue *txq)
2202 const struct net_device_ops *ops = dev->netdev_ops;
2203 int rc = NETDEV_TX_OK;
2204 unsigned int skb_len;
2206 if (likely(!skb->next)) {
2207 netdev_features_t features;
2210 * If device doesn't need skb->dst, release it right now while
2211 * its hot in this cpu cache
2213 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2216 features = netif_skb_features(skb);
2218 if (vlan_tx_tag_present(skb) &&
2219 !(features & NETIF_F_HW_VLAN_TX)) {
2220 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2227 if (netif_needs_gso(skb, features)) {
2228 if (unlikely(dev_gso_segment(skb, features)))
2233 if (skb_needs_linearize(skb, features) &&
2234 __skb_linearize(skb))
2237 /* If packet is not checksummed and device does not
2238 * support checksumming for this protocol, complete
2239 * checksumming here.
2241 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2242 skb_set_transport_header(skb,
2243 skb_checksum_start_offset(skb));
2244 if (!(features & NETIF_F_ALL_CSUM) &&
2245 skb_checksum_help(skb))
2250 if (!list_empty(&ptype_all))
2251 dev_queue_xmit_nit(skb, dev);
2254 rc = ops->ndo_start_xmit(skb, dev);
2255 trace_net_dev_xmit(skb, rc, dev, skb_len);
2256 if (rc == NETDEV_TX_OK)
2257 txq_trans_update(txq);
2263 struct sk_buff *nskb = skb->next;
2265 skb->next = nskb->next;
2269 * If device doesn't need nskb->dst, release it right now while
2270 * its hot in this cpu cache
2272 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2275 if (!list_empty(&ptype_all))
2276 dev_queue_xmit_nit(nskb, dev);
2278 skb_len = nskb->len;
2279 rc = ops->ndo_start_xmit(nskb, dev);
2280 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2281 if (unlikely(rc != NETDEV_TX_OK)) {
2282 if (rc & ~NETDEV_TX_MASK)
2283 goto out_kfree_gso_skb;
2284 nskb->next = skb->next;
2288 txq_trans_update(txq);
2289 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2290 return NETDEV_TX_BUSY;
2291 } while (skb->next);
2294 if (likely(skb->next == NULL))
2295 skb->destructor = DEV_GSO_CB(skb)->destructor;
2302 static u32 hashrnd __read_mostly;
2305 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2306 * to be used as a distribution range.
2308 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2309 unsigned int num_tx_queues)
2313 u16 qcount = num_tx_queues;
2315 if (skb_rx_queue_recorded(skb)) {
2316 hash = skb_get_rx_queue(skb);
2317 while (unlikely(hash >= num_tx_queues))
2318 hash -= num_tx_queues;
2323 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2324 qoffset = dev->tc_to_txq[tc].offset;
2325 qcount = dev->tc_to_txq[tc].count;
2328 if (skb->sk && skb->sk->sk_hash)
2329 hash = skb->sk->sk_hash;
2331 hash = (__force u16) skb->protocol;
2332 hash = jhash_1word(hash, hashrnd);
2334 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2336 EXPORT_SYMBOL(__skb_tx_hash);
2338 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2340 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2341 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2342 dev->name, queue_index,
2343 dev->real_num_tx_queues);
2349 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2352 struct xps_dev_maps *dev_maps;
2353 struct xps_map *map;
2354 int queue_index = -1;
2357 dev_maps = rcu_dereference(dev->xps_maps);
2359 map = rcu_dereference(
2360 dev_maps->cpu_map[raw_smp_processor_id()]);
2363 queue_index = map->queues[0];
2366 if (skb->sk && skb->sk->sk_hash)
2367 hash = skb->sk->sk_hash;
2369 hash = (__force u16) skb->protocol ^
2371 hash = jhash_1word(hash, hashrnd);
2372 queue_index = map->queues[
2373 ((u64)hash * map->len) >> 32];
2375 if (unlikely(queue_index >= dev->real_num_tx_queues))
2387 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2388 struct sk_buff *skb)
2391 const struct net_device_ops *ops = dev->netdev_ops;
2393 if (dev->real_num_tx_queues == 1)
2395 else if (ops->ndo_select_queue) {
2396 queue_index = ops->ndo_select_queue(dev, skb);
2397 queue_index = dev_cap_txqueue(dev, queue_index);
2399 struct sock *sk = skb->sk;
2400 queue_index = sk_tx_queue_get(sk);
2402 if (queue_index < 0 || skb->ooo_okay ||
2403 queue_index >= dev->real_num_tx_queues) {
2404 int old_index = queue_index;
2406 queue_index = get_xps_queue(dev, skb);
2407 if (queue_index < 0)
2408 queue_index = skb_tx_hash(dev, skb);
2410 if (queue_index != old_index && sk) {
2411 struct dst_entry *dst =
2412 rcu_dereference_check(sk->sk_dst_cache, 1);
2414 if (dst && skb_dst(skb) == dst)
2415 sk_tx_queue_set(sk, queue_index);
2420 skb_set_queue_mapping(skb, queue_index);
2421 return netdev_get_tx_queue(dev, queue_index);
2424 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2425 struct net_device *dev,
2426 struct netdev_queue *txq)
2428 spinlock_t *root_lock = qdisc_lock(q);
2432 qdisc_skb_cb(skb)->pkt_len = skb->len;
2433 qdisc_calculate_pkt_len(skb, q);
2435 * Heuristic to force contended enqueues to serialize on a
2436 * separate lock before trying to get qdisc main lock.
2437 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2438 * and dequeue packets faster.
2440 contended = qdisc_is_running(q);
2441 if (unlikely(contended))
2442 spin_lock(&q->busylock);
2444 spin_lock(root_lock);
2445 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2448 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2449 qdisc_run_begin(q)) {
2451 * This is a work-conserving queue; there are no old skbs
2452 * waiting to be sent out; and the qdisc is not running -
2453 * xmit the skb directly.
2455 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2458 qdisc_bstats_update(q, skb);
2460 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2461 if (unlikely(contended)) {
2462 spin_unlock(&q->busylock);
2469 rc = NET_XMIT_SUCCESS;
2472 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2473 if (qdisc_run_begin(q)) {
2474 if (unlikely(contended)) {
2475 spin_unlock(&q->busylock);
2481 spin_unlock(root_lock);
2482 if (unlikely(contended))
2483 spin_unlock(&q->busylock);
2487 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2488 static void skb_update_prio(struct sk_buff *skb)
2490 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2492 if (!skb->priority && skb->sk && map) {
2493 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2495 if (prioidx < map->priomap_len)
2496 skb->priority = map->priomap[prioidx];
2500 #define skb_update_prio(skb)
2503 static DEFINE_PER_CPU(int, xmit_recursion);
2504 #define RECURSION_LIMIT 10
2507 * dev_loopback_xmit - loop back @skb
2508 * @skb: buffer to transmit
2510 int dev_loopback_xmit(struct sk_buff *skb)
2512 skb_reset_mac_header(skb);
2513 __skb_pull(skb, skb_network_offset(skb));
2514 skb->pkt_type = PACKET_LOOPBACK;
2515 skb->ip_summed = CHECKSUM_UNNECESSARY;
2516 WARN_ON(!skb_dst(skb));
2521 EXPORT_SYMBOL(dev_loopback_xmit);
2524 * dev_queue_xmit - transmit a buffer
2525 * @skb: buffer to transmit
2527 * Queue a buffer for transmission to a network device. The caller must
2528 * have set the device and priority and built the buffer before calling
2529 * this function. The function can be called from an interrupt.
2531 * A negative errno code is returned on a failure. A success does not
2532 * guarantee the frame will be transmitted as it may be dropped due
2533 * to congestion or traffic shaping.
2535 * -----------------------------------------------------------------------------------
2536 * I notice this method can also return errors from the queue disciplines,
2537 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2540 * Regardless of the return value, the skb is consumed, so it is currently
2541 * difficult to retry a send to this method. (You can bump the ref count
2542 * before sending to hold a reference for retry if you are careful.)
2544 * When calling this method, interrupts MUST be enabled. This is because
2545 * the BH enable code must have IRQs enabled so that it will not deadlock.
2548 int dev_queue_xmit(struct sk_buff *skb)
2550 struct net_device *dev = skb->dev;
2551 struct netdev_queue *txq;
2555 /* Disable soft irqs for various locks below. Also
2556 * stops preemption for RCU.
2560 skb_update_prio(skb);
2562 txq = dev_pick_tx(dev, skb);
2563 q = rcu_dereference_bh(txq->qdisc);
2565 #ifdef CONFIG_NET_CLS_ACT
2566 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2568 trace_net_dev_queue(skb);
2570 rc = __dev_xmit_skb(skb, q, dev, txq);
2574 /* The device has no queue. Common case for software devices:
2575 loopback, all the sorts of tunnels...
2577 Really, it is unlikely that netif_tx_lock protection is necessary
2578 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2580 However, it is possible, that they rely on protection
2583 Check this and shot the lock. It is not prone from deadlocks.
2584 Either shot noqueue qdisc, it is even simpler 8)
2586 if (dev->flags & IFF_UP) {
2587 int cpu = smp_processor_id(); /* ok because BHs are off */
2589 if (txq->xmit_lock_owner != cpu) {
2591 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2592 goto recursion_alert;
2594 HARD_TX_LOCK(dev, txq, cpu);
2596 if (!netif_xmit_stopped(txq)) {
2597 __this_cpu_inc(xmit_recursion);
2598 rc = dev_hard_start_xmit(skb, dev, txq);
2599 __this_cpu_dec(xmit_recursion);
2600 if (dev_xmit_complete(rc)) {
2601 HARD_TX_UNLOCK(dev, txq);
2605 HARD_TX_UNLOCK(dev, txq);
2606 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2609 /* Recursion is detected! It is possible,
2613 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2619 rcu_read_unlock_bh();
2624 rcu_read_unlock_bh();
2627 EXPORT_SYMBOL(dev_queue_xmit);
2630 /*=======================================================================
2632 =======================================================================*/
2634 int netdev_max_backlog __read_mostly = 1000;
2635 int netdev_tstamp_prequeue __read_mostly = 1;
2636 int netdev_budget __read_mostly = 300;
2637 int weight_p __read_mostly = 64; /* old backlog weight */
2639 /* Called with irq disabled */
2640 static inline void ____napi_schedule(struct softnet_data *sd,
2641 struct napi_struct *napi)
2643 list_add_tail(&napi->poll_list, &sd->poll_list);
2644 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2648 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2649 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2650 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2651 * if hash is a canonical 4-tuple hash over transport ports.
2653 void __skb_get_rxhash(struct sk_buff *skb)
2655 struct flow_keys keys;
2658 if (!skb_flow_dissect(skb, &keys))
2664 /* get a consistent hash (same value on both flow directions) */
2665 if (((__force u32)keys.dst < (__force u32)keys.src) ||
2666 (((__force u32)keys.dst == (__force u32)keys.src) &&
2667 ((__force u16)keys.port16[1] < (__force u16)keys.port16[0]))) {
2668 swap(keys.dst, keys.src);
2669 swap(keys.port16[0], keys.port16[1]);
2672 hash = jhash_3words((__force u32)keys.dst,
2673 (__force u32)keys.src,
2674 (__force u32)keys.ports, hashrnd);
2680 EXPORT_SYMBOL(__skb_get_rxhash);
2684 /* One global table that all flow-based protocols share. */
2685 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2686 EXPORT_SYMBOL(rps_sock_flow_table);
2688 struct static_key rps_needed __read_mostly;
2690 static struct rps_dev_flow *
2691 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2692 struct rps_dev_flow *rflow, u16 next_cpu)
2694 if (next_cpu != RPS_NO_CPU) {
2695 #ifdef CONFIG_RFS_ACCEL
2696 struct netdev_rx_queue *rxqueue;
2697 struct rps_dev_flow_table *flow_table;
2698 struct rps_dev_flow *old_rflow;
2703 /* Should we steer this flow to a different hardware queue? */
2704 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2705 !(dev->features & NETIF_F_NTUPLE))
2707 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2708 if (rxq_index == skb_get_rx_queue(skb))
2711 rxqueue = dev->_rx + rxq_index;
2712 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2715 flow_id = skb->rxhash & flow_table->mask;
2716 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2717 rxq_index, flow_id);
2721 rflow = &flow_table->flows[flow_id];
2723 if (old_rflow->filter == rflow->filter)
2724 old_rflow->filter = RPS_NO_FILTER;
2728 per_cpu(softnet_data, next_cpu).input_queue_head;
2731 rflow->cpu = next_cpu;
2736 * get_rps_cpu is called from netif_receive_skb and returns the target
2737 * CPU from the RPS map of the receiving queue for a given skb.
2738 * rcu_read_lock must be held on entry.
2740 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2741 struct rps_dev_flow **rflowp)
2743 struct netdev_rx_queue *rxqueue;
2744 struct rps_map *map;
2745 struct rps_dev_flow_table *flow_table;
2746 struct rps_sock_flow_table *sock_flow_table;
2750 if (skb_rx_queue_recorded(skb)) {
2751 u16 index = skb_get_rx_queue(skb);
2752 if (unlikely(index >= dev->real_num_rx_queues)) {
2753 WARN_ONCE(dev->real_num_rx_queues > 1,
2754 "%s received packet on queue %u, but number "
2755 "of RX queues is %u\n",
2756 dev->name, index, dev->real_num_rx_queues);
2759 rxqueue = dev->_rx + index;
2763 map = rcu_dereference(rxqueue->rps_map);
2765 if (map->len == 1 &&
2766 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2767 tcpu = map->cpus[0];
2768 if (cpu_online(tcpu))
2772 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2776 skb_reset_network_header(skb);
2777 if (!skb_get_rxhash(skb))
2780 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2781 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2782 if (flow_table && sock_flow_table) {
2784 struct rps_dev_flow *rflow;
2786 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2789 next_cpu = sock_flow_table->ents[skb->rxhash &
2790 sock_flow_table->mask];
2793 * If the desired CPU (where last recvmsg was done) is
2794 * different from current CPU (one in the rx-queue flow
2795 * table entry), switch if one of the following holds:
2796 * - Current CPU is unset (equal to RPS_NO_CPU).
2797 * - Current CPU is offline.
2798 * - The current CPU's queue tail has advanced beyond the
2799 * last packet that was enqueued using this table entry.
2800 * This guarantees that all previous packets for the flow
2801 * have been dequeued, thus preserving in order delivery.
2803 if (unlikely(tcpu != next_cpu) &&
2804 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2805 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2806 rflow->last_qtail)) >= 0))
2807 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2809 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2817 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2819 if (cpu_online(tcpu)) {
2829 #ifdef CONFIG_RFS_ACCEL
2832 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2833 * @dev: Device on which the filter was set
2834 * @rxq_index: RX queue index
2835 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2836 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2838 * Drivers that implement ndo_rx_flow_steer() should periodically call
2839 * this function for each installed filter and remove the filters for
2840 * which it returns %true.
2842 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2843 u32 flow_id, u16 filter_id)
2845 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2846 struct rps_dev_flow_table *flow_table;
2847 struct rps_dev_flow *rflow;
2852 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2853 if (flow_table && flow_id <= flow_table->mask) {
2854 rflow = &flow_table->flows[flow_id];
2855 cpu = ACCESS_ONCE(rflow->cpu);
2856 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2857 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2858 rflow->last_qtail) <
2859 (int)(10 * flow_table->mask)))
2865 EXPORT_SYMBOL(rps_may_expire_flow);
2867 #endif /* CONFIG_RFS_ACCEL */
2869 /* Called from hardirq (IPI) context */
2870 static void rps_trigger_softirq(void *data)
2872 struct softnet_data *sd = data;
2874 ____napi_schedule(sd, &sd->backlog);
2878 #endif /* CONFIG_RPS */
2881 * Check if this softnet_data structure is another cpu one
2882 * If yes, queue it to our IPI list and return 1
2885 static int rps_ipi_queued(struct softnet_data *sd)
2888 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2891 sd->rps_ipi_next = mysd->rps_ipi_list;
2892 mysd->rps_ipi_list = sd;
2894 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2897 #endif /* CONFIG_RPS */
2902 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2903 * queue (may be a remote CPU queue).
2905 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2906 unsigned int *qtail)
2908 struct softnet_data *sd;
2909 unsigned long flags;
2911 sd = &per_cpu(softnet_data, cpu);
2913 local_irq_save(flags);
2916 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2917 if (skb_queue_len(&sd->input_pkt_queue)) {
2919 __skb_queue_tail(&sd->input_pkt_queue, skb);
2920 input_queue_tail_incr_save(sd, qtail);
2922 local_irq_restore(flags);
2923 return NET_RX_SUCCESS;
2926 /* Schedule NAPI for backlog device
2927 * We can use non atomic operation since we own the queue lock
2929 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2930 if (!rps_ipi_queued(sd))
2931 ____napi_schedule(sd, &sd->backlog);
2939 local_irq_restore(flags);
2941 atomic_long_inc(&skb->dev->rx_dropped);
2947 * netif_rx - post buffer to the network code
2948 * @skb: buffer to post
2950 * This function receives a packet from a device driver and queues it for
2951 * the upper (protocol) levels to process. It always succeeds. The buffer
2952 * may be dropped during processing for congestion control or by the
2956 * NET_RX_SUCCESS (no congestion)
2957 * NET_RX_DROP (packet was dropped)
2961 int netif_rx(struct sk_buff *skb)
2965 /* if netpoll wants it, pretend we never saw it */
2966 if (netpoll_rx(skb))
2969 net_timestamp_check(netdev_tstamp_prequeue, skb);
2971 trace_netif_rx(skb);
2973 if (static_key_false(&rps_needed)) {
2974 struct rps_dev_flow voidflow, *rflow = &voidflow;
2980 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2982 cpu = smp_processor_id();
2984 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2992 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2997 EXPORT_SYMBOL(netif_rx);
2999 int netif_rx_ni(struct sk_buff *skb)
3004 err = netif_rx(skb);
3005 if (local_softirq_pending())
3011 EXPORT_SYMBOL(netif_rx_ni);
3013 static void net_tx_action(struct softirq_action *h)
3015 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3017 if (sd->completion_queue) {
3018 struct sk_buff *clist;
3020 local_irq_disable();
3021 clist = sd->completion_queue;
3022 sd->completion_queue = NULL;
3026 struct sk_buff *skb = clist;
3027 clist = clist->next;
3029 WARN_ON(atomic_read(&skb->users));
3030 trace_kfree_skb(skb, net_tx_action);
3035 if (sd->output_queue) {
3038 local_irq_disable();
3039 head = sd->output_queue;
3040 sd->output_queue = NULL;
3041 sd->output_queue_tailp = &sd->output_queue;
3045 struct Qdisc *q = head;
3046 spinlock_t *root_lock;
3048 head = head->next_sched;
3050 root_lock = qdisc_lock(q);
3051 if (spin_trylock(root_lock)) {
3052 smp_mb__before_clear_bit();
3053 clear_bit(__QDISC_STATE_SCHED,
3056 spin_unlock(root_lock);
3058 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3060 __netif_reschedule(q);
3062 smp_mb__before_clear_bit();
3063 clear_bit(__QDISC_STATE_SCHED,
3071 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3072 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3073 /* This hook is defined here for ATM LANE */
3074 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3075 unsigned char *addr) __read_mostly;
3076 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3079 #ifdef CONFIG_NET_CLS_ACT
3080 /* TODO: Maybe we should just force sch_ingress to be compiled in
3081 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3082 * a compare and 2 stores extra right now if we dont have it on
3083 * but have CONFIG_NET_CLS_ACT
3084 * NOTE: This doesn't stop any functionality; if you dont have
3085 * the ingress scheduler, you just can't add policies on ingress.
3088 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3090 struct net_device *dev = skb->dev;
3091 u32 ttl = G_TC_RTTL(skb->tc_verd);
3092 int result = TC_ACT_OK;
3095 if (unlikely(MAX_RED_LOOP < ttl++)) {
3096 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3097 skb->skb_iif, dev->ifindex);
3101 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3102 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3105 if (q != &noop_qdisc) {
3106 spin_lock(qdisc_lock(q));
3107 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3108 result = qdisc_enqueue_root(skb, q);
3109 spin_unlock(qdisc_lock(q));
3115 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3116 struct packet_type **pt_prev,
3117 int *ret, struct net_device *orig_dev)
3119 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3121 if (!rxq || rxq->qdisc == &noop_qdisc)
3125 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3129 switch (ing_filter(skb, rxq)) {
3143 * netdev_rx_handler_register - register receive handler
3144 * @dev: device to register a handler for
3145 * @rx_handler: receive handler to register
3146 * @rx_handler_data: data pointer that is used by rx handler
3148 * Register a receive hander for a device. This handler will then be
3149 * called from __netif_receive_skb. A negative errno code is returned
3152 * The caller must hold the rtnl_mutex.
3154 * For a general description of rx_handler, see enum rx_handler_result.
3156 int netdev_rx_handler_register(struct net_device *dev,
3157 rx_handler_func_t *rx_handler,
3158 void *rx_handler_data)
3162 if (dev->rx_handler)
3165 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3166 rcu_assign_pointer(dev->rx_handler, rx_handler);
3170 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3173 * netdev_rx_handler_unregister - unregister receive handler
3174 * @dev: device to unregister a handler from
3176 * Unregister a receive hander from a device.
3178 * The caller must hold the rtnl_mutex.
3180 void netdev_rx_handler_unregister(struct net_device *dev)
3184 RCU_INIT_POINTER(dev->rx_handler, NULL);
3185 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3187 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3190 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3191 * the special handling of PFMEMALLOC skbs.
3193 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3195 switch (skb->protocol) {
3196 case __constant_htons(ETH_P_ARP):
3197 case __constant_htons(ETH_P_IP):
3198 case __constant_htons(ETH_P_IPV6):
3199 case __constant_htons(ETH_P_8021Q):
3206 static int __netif_receive_skb(struct sk_buff *skb)
3208 struct packet_type *ptype, *pt_prev;
3209 rx_handler_func_t *rx_handler;
3210 struct net_device *orig_dev;
3211 struct net_device *null_or_dev;
3212 bool deliver_exact = false;
3213 int ret = NET_RX_DROP;
3215 unsigned long pflags = current->flags;
3217 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3219 trace_netif_receive_skb(skb);
3222 * PFMEMALLOC skbs are special, they should
3223 * - be delivered to SOCK_MEMALLOC sockets only
3224 * - stay away from userspace
3225 * - have bounded memory usage
3227 * Use PF_MEMALLOC as this saves us from propagating the allocation
3228 * context down to all allocation sites.
3230 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3231 current->flags |= PF_MEMALLOC;
3233 /* if we've gotten here through NAPI, check netpoll */
3234 if (netpoll_receive_skb(skb))
3237 orig_dev = skb->dev;
3239 skb_reset_network_header(skb);
3240 skb_reset_transport_header(skb);
3241 skb_reset_mac_len(skb);
3248 skb->skb_iif = skb->dev->ifindex;
3250 __this_cpu_inc(softnet_data.processed);
3252 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3253 skb = vlan_untag(skb);
3258 #ifdef CONFIG_NET_CLS_ACT
3259 if (skb->tc_verd & TC_NCLS) {
3260 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3265 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3268 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3269 if (!ptype->dev || ptype->dev == skb->dev) {
3271 ret = deliver_skb(skb, pt_prev, orig_dev);
3277 #ifdef CONFIG_NET_CLS_ACT
3278 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3284 if (sk_memalloc_socks() && skb_pfmemalloc(skb)
3285 && !skb_pfmemalloc_protocol(skb))
3288 rx_handler = rcu_dereference(skb->dev->rx_handler);
3289 if (vlan_tx_tag_present(skb)) {
3291 ret = deliver_skb(skb, pt_prev, orig_dev);
3294 if (vlan_do_receive(&skb, !rx_handler))
3296 else if (unlikely(!skb))
3302 ret = deliver_skb(skb, pt_prev, orig_dev);
3305 switch (rx_handler(&skb)) {
3306 case RX_HANDLER_CONSUMED:
3308 case RX_HANDLER_ANOTHER:
3310 case RX_HANDLER_EXACT:
3311 deliver_exact = true;
3312 case RX_HANDLER_PASS:
3319 /* deliver only exact match when indicated */
3320 null_or_dev = deliver_exact ? skb->dev : NULL;
3322 type = skb->protocol;
3323 list_for_each_entry_rcu(ptype,
3324 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3325 if (ptype->type == type &&
3326 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3327 ptype->dev == orig_dev)) {
3329 ret = deliver_skb(skb, pt_prev, orig_dev);
3335 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3338 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3341 atomic_long_inc(&skb->dev->rx_dropped);
3343 /* Jamal, now you will not able to escape explaining
3344 * me how you were going to use this. :-)
3352 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3357 * netif_receive_skb - process receive buffer from network
3358 * @skb: buffer to process
3360 * netif_receive_skb() is the main receive data processing function.
3361 * It always succeeds. The buffer may be dropped during processing
3362 * for congestion control or by the protocol layers.
3364 * This function may only be called from softirq context and interrupts
3365 * should be enabled.
3367 * Return values (usually ignored):
3368 * NET_RX_SUCCESS: no congestion
3369 * NET_RX_DROP: packet was dropped
3371 int netif_receive_skb(struct sk_buff *skb)
3373 net_timestamp_check(netdev_tstamp_prequeue, skb);
3375 if (skb_defer_rx_timestamp(skb))
3376 return NET_RX_SUCCESS;
3379 if (static_key_false(&rps_needed)) {
3380 struct rps_dev_flow voidflow, *rflow = &voidflow;
3385 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3388 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3395 return __netif_receive_skb(skb);
3397 EXPORT_SYMBOL(netif_receive_skb);
3399 /* Network device is going away, flush any packets still pending
3400 * Called with irqs disabled.
3402 static void flush_backlog(void *arg)
3404 struct net_device *dev = arg;
3405 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3406 struct sk_buff *skb, *tmp;
3409 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3410 if (skb->dev == dev) {
3411 __skb_unlink(skb, &sd->input_pkt_queue);
3413 input_queue_head_incr(sd);
3418 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3419 if (skb->dev == dev) {
3420 __skb_unlink(skb, &sd->process_queue);
3422 input_queue_head_incr(sd);
3427 static int napi_gro_complete(struct sk_buff *skb)
3429 struct packet_type *ptype;
3430 __be16 type = skb->protocol;
3431 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3434 if (NAPI_GRO_CB(skb)->count == 1) {
3435 skb_shinfo(skb)->gso_size = 0;
3440 list_for_each_entry_rcu(ptype, head, list) {
3441 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3444 err = ptype->gro_complete(skb);
3450 WARN_ON(&ptype->list == head);
3452 return NET_RX_SUCCESS;
3456 return netif_receive_skb(skb);
3459 inline void napi_gro_flush(struct napi_struct *napi)
3461 struct sk_buff *skb, *next;
3463 for (skb = napi->gro_list; skb; skb = next) {
3466 napi_gro_complete(skb);
3469 napi->gro_count = 0;
3470 napi->gro_list = NULL;
3472 EXPORT_SYMBOL(napi_gro_flush);
3474 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3476 struct sk_buff **pp = NULL;
3477 struct packet_type *ptype;
3478 __be16 type = skb->protocol;
3479 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3482 enum gro_result ret;
3484 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3487 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3491 list_for_each_entry_rcu(ptype, head, list) {
3492 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3495 skb_set_network_header(skb, skb_gro_offset(skb));
3496 mac_len = skb->network_header - skb->mac_header;
3497 skb->mac_len = mac_len;
3498 NAPI_GRO_CB(skb)->same_flow = 0;
3499 NAPI_GRO_CB(skb)->flush = 0;
3500 NAPI_GRO_CB(skb)->free = 0;
3502 pp = ptype->gro_receive(&napi->gro_list, skb);
3507 if (&ptype->list == head)
3510 same_flow = NAPI_GRO_CB(skb)->same_flow;
3511 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3514 struct sk_buff *nskb = *pp;
3518 napi_gro_complete(nskb);
3525 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3529 NAPI_GRO_CB(skb)->count = 1;
3530 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3531 skb->next = napi->gro_list;
3532 napi->gro_list = skb;
3536 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3537 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3539 BUG_ON(skb->end - skb->tail < grow);
3541 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3544 skb->data_len -= grow;
3546 skb_shinfo(skb)->frags[0].page_offset += grow;
3547 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3549 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3550 skb_frag_unref(skb, 0);
3551 memmove(skb_shinfo(skb)->frags,
3552 skb_shinfo(skb)->frags + 1,
3553 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3564 EXPORT_SYMBOL(dev_gro_receive);
3566 static inline gro_result_t
3567 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3570 unsigned int maclen = skb->dev->hard_header_len;
3572 for (p = napi->gro_list; p; p = p->next) {
3573 unsigned long diffs;
3575 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3576 diffs |= p->vlan_tci ^ skb->vlan_tci;
3577 if (maclen == ETH_HLEN)
3578 diffs |= compare_ether_header(skb_mac_header(p),
3579 skb_gro_mac_header(skb));
3581 diffs = memcmp(skb_mac_header(p),
3582 skb_gro_mac_header(skb),
3584 NAPI_GRO_CB(p)->same_flow = !diffs;
3585 NAPI_GRO_CB(p)->flush = 0;
3588 return dev_gro_receive(napi, skb);
3591 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3595 if (netif_receive_skb(skb))
3603 case GRO_MERGED_FREE:
3604 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3605 kmem_cache_free(skbuff_head_cache, skb);
3617 EXPORT_SYMBOL(napi_skb_finish);
3619 void skb_gro_reset_offset(struct sk_buff *skb)
3621 NAPI_GRO_CB(skb)->data_offset = 0;
3622 NAPI_GRO_CB(skb)->frag0 = NULL;
3623 NAPI_GRO_CB(skb)->frag0_len = 0;
3625 if (skb->mac_header == skb->tail &&
3626 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3627 NAPI_GRO_CB(skb)->frag0 =
3628 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3629 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3632 EXPORT_SYMBOL(skb_gro_reset_offset);
3634 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3636 skb_gro_reset_offset(skb);
3638 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3640 EXPORT_SYMBOL(napi_gro_receive);
3642 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3644 __skb_pull(skb, skb_headlen(skb));
3645 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3646 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3648 skb->dev = napi->dev;
3654 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3656 struct sk_buff *skb = napi->skb;
3659 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3665 EXPORT_SYMBOL(napi_get_frags);
3667 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3673 skb->protocol = eth_type_trans(skb, skb->dev);
3675 if (ret == GRO_HELD)
3676 skb_gro_pull(skb, -ETH_HLEN);
3677 else if (netif_receive_skb(skb))
3682 case GRO_MERGED_FREE:
3683 napi_reuse_skb(napi, skb);
3692 EXPORT_SYMBOL(napi_frags_finish);
3694 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3696 struct sk_buff *skb = napi->skb;
3703 skb_reset_mac_header(skb);
3704 skb_gro_reset_offset(skb);
3706 off = skb_gro_offset(skb);
3707 hlen = off + sizeof(*eth);
3708 eth = skb_gro_header_fast(skb, off);
3709 if (skb_gro_header_hard(skb, hlen)) {
3710 eth = skb_gro_header_slow(skb, hlen, off);
3711 if (unlikely(!eth)) {
3712 napi_reuse_skb(napi, skb);
3718 skb_gro_pull(skb, sizeof(*eth));
3721 * This works because the only protocols we care about don't require
3722 * special handling. We'll fix it up properly at the end.
3724 skb->protocol = eth->h_proto;
3730 gro_result_t napi_gro_frags(struct napi_struct *napi)
3732 struct sk_buff *skb = napi_frags_skb(napi);
3737 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3739 EXPORT_SYMBOL(napi_gro_frags);
3742 * net_rps_action sends any pending IPI's for rps.
3743 * Note: called with local irq disabled, but exits with local irq enabled.
3745 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3748 struct softnet_data *remsd = sd->rps_ipi_list;
3751 sd->rps_ipi_list = NULL;
3755 /* Send pending IPI's to kick RPS processing on remote cpus. */
3757 struct softnet_data *next = remsd->rps_ipi_next;
3759 if (cpu_online(remsd->cpu))
3760 __smp_call_function_single(remsd->cpu,
3769 static int process_backlog(struct napi_struct *napi, int quota)
3772 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3775 /* Check if we have pending ipi, its better to send them now,
3776 * not waiting net_rx_action() end.
3778 if (sd->rps_ipi_list) {
3779 local_irq_disable();
3780 net_rps_action_and_irq_enable(sd);
3783 napi->weight = weight_p;
3784 local_irq_disable();
3785 while (work < quota) {
3786 struct sk_buff *skb;
3789 while ((skb = __skb_dequeue(&sd->process_queue))) {
3791 __netif_receive_skb(skb);
3792 local_irq_disable();
3793 input_queue_head_incr(sd);
3794 if (++work >= quota) {
3801 qlen = skb_queue_len(&sd->input_pkt_queue);
3803 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3804 &sd->process_queue);
3806 if (qlen < quota - work) {
3808 * Inline a custom version of __napi_complete().
3809 * only current cpu owns and manipulates this napi,
3810 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3811 * we can use a plain write instead of clear_bit(),
3812 * and we dont need an smp_mb() memory barrier.
3814 list_del(&napi->poll_list);
3817 quota = work + qlen;
3827 * __napi_schedule - schedule for receive
3828 * @n: entry to schedule
3830 * The entry's receive function will be scheduled to run
3832 void __napi_schedule(struct napi_struct *n)
3834 unsigned long flags;
3836 local_irq_save(flags);
3837 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3838 local_irq_restore(flags);
3840 EXPORT_SYMBOL(__napi_schedule);
3842 void __napi_complete(struct napi_struct *n)
3844 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3845 BUG_ON(n->gro_list);
3847 list_del(&n->poll_list);
3848 smp_mb__before_clear_bit();
3849 clear_bit(NAPI_STATE_SCHED, &n->state);
3851 EXPORT_SYMBOL(__napi_complete);
3853 void napi_complete(struct napi_struct *n)
3855 unsigned long flags;
3858 * don't let napi dequeue from the cpu poll list
3859 * just in case its running on a different cpu
3861 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3865 local_irq_save(flags);
3867 local_irq_restore(flags);
3869 EXPORT_SYMBOL(napi_complete);
3871 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3872 int (*poll)(struct napi_struct *, int), int weight)
3874 INIT_LIST_HEAD(&napi->poll_list);
3875 napi->gro_count = 0;
3876 napi->gro_list = NULL;
3879 napi->weight = weight;
3880 list_add(&napi->dev_list, &dev->napi_list);
3882 #ifdef CONFIG_NETPOLL
3883 spin_lock_init(&napi->poll_lock);
3884 napi->poll_owner = -1;
3886 set_bit(NAPI_STATE_SCHED, &napi->state);
3888 EXPORT_SYMBOL(netif_napi_add);
3890 void netif_napi_del(struct napi_struct *napi)
3892 struct sk_buff *skb, *next;
3894 list_del_init(&napi->dev_list);
3895 napi_free_frags(napi);
3897 for (skb = napi->gro_list; skb; skb = next) {
3903 napi->gro_list = NULL;
3904 napi->gro_count = 0;
3906 EXPORT_SYMBOL(netif_napi_del);
3908 static void net_rx_action(struct softirq_action *h)
3910 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3911 unsigned long time_limit = jiffies + 2;
3912 int budget = netdev_budget;
3915 local_irq_disable();
3917 while (!list_empty(&sd->poll_list)) {
3918 struct napi_struct *n;
3921 /* If softirq window is exhuasted then punt.
3922 * Allow this to run for 2 jiffies since which will allow
3923 * an average latency of 1.5/HZ.
3925 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3930 /* Even though interrupts have been re-enabled, this
3931 * access is safe because interrupts can only add new
3932 * entries to the tail of this list, and only ->poll()
3933 * calls can remove this head entry from the list.
3935 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3937 have = netpoll_poll_lock(n);
3941 /* This NAPI_STATE_SCHED test is for avoiding a race
3942 * with netpoll's poll_napi(). Only the entity which
3943 * obtains the lock and sees NAPI_STATE_SCHED set will
3944 * actually make the ->poll() call. Therefore we avoid
3945 * accidentally calling ->poll() when NAPI is not scheduled.
3948 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3949 work = n->poll(n, weight);
3953 WARN_ON_ONCE(work > weight);
3957 local_irq_disable();
3959 /* Drivers must not modify the NAPI state if they
3960 * consume the entire weight. In such cases this code
3961 * still "owns" the NAPI instance and therefore can
3962 * move the instance around on the list at-will.
3964 if (unlikely(work == weight)) {
3965 if (unlikely(napi_disable_pending(n))) {
3968 local_irq_disable();
3970 list_move_tail(&n->poll_list, &sd->poll_list);
3973 netpoll_poll_unlock(have);
3976 net_rps_action_and_irq_enable(sd);
3978 #ifdef CONFIG_NET_DMA
3980 * There may not be any more sk_buffs coming right now, so push
3981 * any pending DMA copies to hardware
3983 dma_issue_pending_all();
3990 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3994 static gifconf_func_t *gifconf_list[NPROTO];
3997 * register_gifconf - register a SIOCGIF handler
3998 * @family: Address family
3999 * @gifconf: Function handler
4001 * Register protocol dependent address dumping routines. The handler
4002 * that is passed must not be freed or reused until it has been replaced
4003 * by another handler.
4005 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4007 if (family >= NPROTO)
4009 gifconf_list[family] = gifconf;
4012 EXPORT_SYMBOL(register_gifconf);
4016 * Map an interface index to its name (SIOCGIFNAME)
4020 * We need this ioctl for efficient implementation of the
4021 * if_indextoname() function required by the IPv6 API. Without
4022 * it, we would have to search all the interfaces to find a
4026 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4028 struct net_device *dev;
4032 * Fetch the caller's info block.
4035 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4039 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4045 strcpy(ifr.ifr_name, dev->name);
4048 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4054 * Perform a SIOCGIFCONF call. This structure will change
4055 * size eventually, and there is nothing I can do about it.
4056 * Thus we will need a 'compatibility mode'.
4059 static int dev_ifconf(struct net *net, char __user *arg)
4062 struct net_device *dev;
4069 * Fetch the caller's info block.
4072 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4079 * Loop over the interfaces, and write an info block for each.
4083 for_each_netdev(net, dev) {
4084 for (i = 0; i < NPROTO; i++) {
4085 if (gifconf_list[i]) {
4088 done = gifconf_list[i](dev, NULL, 0);
4090 done = gifconf_list[i](dev, pos + total,
4100 * All done. Write the updated control block back to the caller.
4102 ifc.ifc_len = total;
4105 * Both BSD and Solaris return 0 here, so we do too.
4107 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4110 #ifdef CONFIG_PROC_FS
4112 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4114 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4115 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4116 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4118 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4120 struct net *net = seq_file_net(seq);
4121 struct net_device *dev;
4122 struct hlist_node *p;
4123 struct hlist_head *h;
4124 unsigned int count = 0, offset = get_offset(*pos);
4126 h = &net->dev_name_head[get_bucket(*pos)];
4127 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4128 if (++count == offset)
4135 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4137 struct net_device *dev;
4138 unsigned int bucket;
4141 dev = dev_from_same_bucket(seq, pos);
4145 bucket = get_bucket(*pos) + 1;
4146 *pos = set_bucket_offset(bucket, 1);
4147 } while (bucket < NETDEV_HASHENTRIES);
4153 * This is invoked by the /proc filesystem handler to display a device
4156 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4161 return SEQ_START_TOKEN;
4163 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4166 return dev_from_bucket(seq, pos);
4169 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4172 return dev_from_bucket(seq, pos);
4175 void dev_seq_stop(struct seq_file *seq, void *v)
4181 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4183 struct rtnl_link_stats64 temp;
4184 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4186 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4187 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4188 dev->name, stats->rx_bytes, stats->rx_packets,
4190 stats->rx_dropped + stats->rx_missed_errors,
4191 stats->rx_fifo_errors,
4192 stats->rx_length_errors + stats->rx_over_errors +
4193 stats->rx_crc_errors + stats->rx_frame_errors,
4194 stats->rx_compressed, stats->multicast,
4195 stats->tx_bytes, stats->tx_packets,
4196 stats->tx_errors, stats->tx_dropped,
4197 stats->tx_fifo_errors, stats->collisions,
4198 stats->tx_carrier_errors +
4199 stats->tx_aborted_errors +
4200 stats->tx_window_errors +
4201 stats->tx_heartbeat_errors,
4202 stats->tx_compressed);
4206 * Called from the PROCfs module. This now uses the new arbitrary sized
4207 * /proc/net interface to create /proc/net/dev
4209 static int dev_seq_show(struct seq_file *seq, void *v)
4211 if (v == SEQ_START_TOKEN)
4212 seq_puts(seq, "Inter-| Receive "
4214 " face |bytes packets errs drop fifo frame "
4215 "compressed multicast|bytes packets errs "
4216 "drop fifo colls carrier compressed\n");
4218 dev_seq_printf_stats(seq, v);
4222 static struct softnet_data *softnet_get_online(loff_t *pos)
4224 struct softnet_data *sd = NULL;
4226 while (*pos < nr_cpu_ids)
4227 if (cpu_online(*pos)) {
4228 sd = &per_cpu(softnet_data, *pos);
4235 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4237 return softnet_get_online(pos);
4240 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4243 return softnet_get_online(pos);
4246 static void softnet_seq_stop(struct seq_file *seq, void *v)
4250 static int softnet_seq_show(struct seq_file *seq, void *v)
4252 struct softnet_data *sd = v;
4254 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4255 sd->processed, sd->dropped, sd->time_squeeze, 0,
4256 0, 0, 0, 0, /* was fastroute */
4257 sd->cpu_collision, sd->received_rps);
4261 static const struct seq_operations dev_seq_ops = {
4262 .start = dev_seq_start,
4263 .next = dev_seq_next,
4264 .stop = dev_seq_stop,
4265 .show = dev_seq_show,
4268 static int dev_seq_open(struct inode *inode, struct file *file)
4270 return seq_open_net(inode, file, &dev_seq_ops,
4271 sizeof(struct seq_net_private));
4274 static const struct file_operations dev_seq_fops = {
4275 .owner = THIS_MODULE,
4276 .open = dev_seq_open,
4278 .llseek = seq_lseek,
4279 .release = seq_release_net,
4282 static const struct seq_operations softnet_seq_ops = {
4283 .start = softnet_seq_start,
4284 .next = softnet_seq_next,
4285 .stop = softnet_seq_stop,
4286 .show = softnet_seq_show,
4289 static int softnet_seq_open(struct inode *inode, struct file *file)
4291 return seq_open(file, &softnet_seq_ops);
4294 static const struct file_operations softnet_seq_fops = {
4295 .owner = THIS_MODULE,
4296 .open = softnet_seq_open,
4298 .llseek = seq_lseek,
4299 .release = seq_release,
4302 static void *ptype_get_idx(loff_t pos)
4304 struct packet_type *pt = NULL;
4308 list_for_each_entry_rcu(pt, &ptype_all, list) {
4314 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4315 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4324 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4328 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4331 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4333 struct packet_type *pt;
4334 struct list_head *nxt;
4338 if (v == SEQ_START_TOKEN)
4339 return ptype_get_idx(0);
4342 nxt = pt->list.next;
4343 if (pt->type == htons(ETH_P_ALL)) {
4344 if (nxt != &ptype_all)
4347 nxt = ptype_base[0].next;
4349 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4351 while (nxt == &ptype_base[hash]) {
4352 if (++hash >= PTYPE_HASH_SIZE)
4354 nxt = ptype_base[hash].next;
4357 return list_entry(nxt, struct packet_type, list);
4360 static void ptype_seq_stop(struct seq_file *seq, void *v)
4366 static int ptype_seq_show(struct seq_file *seq, void *v)
4368 struct packet_type *pt = v;
4370 if (v == SEQ_START_TOKEN)
4371 seq_puts(seq, "Type Device Function\n");
4372 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4373 if (pt->type == htons(ETH_P_ALL))
4374 seq_puts(seq, "ALL ");
4376 seq_printf(seq, "%04x", ntohs(pt->type));
4378 seq_printf(seq, " %-8s %pF\n",
4379 pt->dev ? pt->dev->name : "", pt->func);
4385 static const struct seq_operations ptype_seq_ops = {
4386 .start = ptype_seq_start,
4387 .next = ptype_seq_next,
4388 .stop = ptype_seq_stop,
4389 .show = ptype_seq_show,
4392 static int ptype_seq_open(struct inode *inode, struct file *file)
4394 return seq_open_net(inode, file, &ptype_seq_ops,
4395 sizeof(struct seq_net_private));
4398 static const struct file_operations ptype_seq_fops = {
4399 .owner = THIS_MODULE,
4400 .open = ptype_seq_open,
4402 .llseek = seq_lseek,
4403 .release = seq_release_net,
4407 static int __net_init dev_proc_net_init(struct net *net)
4411 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4413 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4415 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4418 if (wext_proc_init(net))
4424 proc_net_remove(net, "ptype");
4426 proc_net_remove(net, "softnet_stat");
4428 proc_net_remove(net, "dev");
4432 static void __net_exit dev_proc_net_exit(struct net *net)
4434 wext_proc_exit(net);
4436 proc_net_remove(net, "ptype");
4437 proc_net_remove(net, "softnet_stat");
4438 proc_net_remove(net, "dev");
4441 static struct pernet_operations __net_initdata dev_proc_ops = {
4442 .init = dev_proc_net_init,
4443 .exit = dev_proc_net_exit,
4446 static int __init dev_proc_init(void)
4448 return register_pernet_subsys(&dev_proc_ops);
4451 #define dev_proc_init() 0
4452 #endif /* CONFIG_PROC_FS */
4456 * netdev_set_master - set up master pointer
4457 * @slave: slave device
4458 * @master: new master device
4460 * Changes the master device of the slave. Pass %NULL to break the
4461 * bonding. The caller must hold the RTNL semaphore. On a failure
4462 * a negative errno code is returned. On success the reference counts
4463 * are adjusted and the function returns zero.
4465 int netdev_set_master(struct net_device *slave, struct net_device *master)
4467 struct net_device *old = slave->master;
4477 slave->master = master;
4483 EXPORT_SYMBOL(netdev_set_master);
4486 * netdev_set_bond_master - set up bonding master/slave pair
4487 * @slave: slave device
4488 * @master: new master device
4490 * Changes the master device of the slave. Pass %NULL to break the
4491 * bonding. The caller must hold the RTNL semaphore. On a failure
4492 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4493 * to the routing socket and the function returns zero.
4495 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4501 err = netdev_set_master(slave, master);
4505 slave->flags |= IFF_SLAVE;
4507 slave->flags &= ~IFF_SLAVE;
4509 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4512 EXPORT_SYMBOL(netdev_set_bond_master);
4514 static void dev_change_rx_flags(struct net_device *dev, int flags)
4516 const struct net_device_ops *ops = dev->netdev_ops;
4518 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4519 ops->ndo_change_rx_flags(dev, flags);
4522 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4524 unsigned int old_flags = dev->flags;
4530 dev->flags |= IFF_PROMISC;
4531 dev->promiscuity += inc;
4532 if (dev->promiscuity == 0) {
4535 * If inc causes overflow, untouch promisc and return error.
4538 dev->flags &= ~IFF_PROMISC;
4540 dev->promiscuity -= inc;
4541 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4546 if (dev->flags != old_flags) {
4547 pr_info("device %s %s promiscuous mode\n",
4549 dev->flags & IFF_PROMISC ? "entered" : "left");
4550 if (audit_enabled) {
4551 current_uid_gid(&uid, &gid);
4552 audit_log(current->audit_context, GFP_ATOMIC,
4553 AUDIT_ANOM_PROMISCUOUS,
4554 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4555 dev->name, (dev->flags & IFF_PROMISC),
4556 (old_flags & IFF_PROMISC),
4557 audit_get_loginuid(current),
4558 from_kuid(&init_user_ns, uid),
4559 from_kgid(&init_user_ns, gid),
4560 audit_get_sessionid(current));
4563 dev_change_rx_flags(dev, IFF_PROMISC);
4569 * dev_set_promiscuity - update promiscuity count on a device
4573 * Add or remove promiscuity from a device. While the count in the device
4574 * remains above zero the interface remains promiscuous. Once it hits zero
4575 * the device reverts back to normal filtering operation. A negative inc
4576 * value is used to drop promiscuity on the device.
4577 * Return 0 if successful or a negative errno code on error.
4579 int dev_set_promiscuity(struct net_device *dev, int inc)
4581 unsigned int old_flags = dev->flags;
4584 err = __dev_set_promiscuity(dev, inc);
4587 if (dev->flags != old_flags)
4588 dev_set_rx_mode(dev);
4591 EXPORT_SYMBOL(dev_set_promiscuity);
4594 * dev_set_allmulti - update allmulti count on a device
4598 * Add or remove reception of all multicast frames to a device. While the
4599 * count in the device remains above zero the interface remains listening
4600 * to all interfaces. Once it hits zero the device reverts back to normal
4601 * filtering operation. A negative @inc value is used to drop the counter
4602 * when releasing a resource needing all multicasts.
4603 * Return 0 if successful or a negative errno code on error.
4606 int dev_set_allmulti(struct net_device *dev, int inc)
4608 unsigned int old_flags = dev->flags;
4612 dev->flags |= IFF_ALLMULTI;
4613 dev->allmulti += inc;
4614 if (dev->allmulti == 0) {
4617 * If inc causes overflow, untouch allmulti and return error.
4620 dev->flags &= ~IFF_ALLMULTI;
4622 dev->allmulti -= inc;
4623 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4628 if (dev->flags ^ old_flags) {
4629 dev_change_rx_flags(dev, IFF_ALLMULTI);
4630 dev_set_rx_mode(dev);
4634 EXPORT_SYMBOL(dev_set_allmulti);
4637 * Upload unicast and multicast address lists to device and
4638 * configure RX filtering. When the device doesn't support unicast
4639 * filtering it is put in promiscuous mode while unicast addresses
4642 void __dev_set_rx_mode(struct net_device *dev)
4644 const struct net_device_ops *ops = dev->netdev_ops;
4646 /* dev_open will call this function so the list will stay sane. */
4647 if (!(dev->flags&IFF_UP))
4650 if (!netif_device_present(dev))
4653 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4654 /* Unicast addresses changes may only happen under the rtnl,
4655 * therefore calling __dev_set_promiscuity here is safe.
4657 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4658 __dev_set_promiscuity(dev, 1);
4659 dev->uc_promisc = true;
4660 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4661 __dev_set_promiscuity(dev, -1);
4662 dev->uc_promisc = false;
4666 if (ops->ndo_set_rx_mode)
4667 ops->ndo_set_rx_mode(dev);
4670 void dev_set_rx_mode(struct net_device *dev)
4672 netif_addr_lock_bh(dev);
4673 __dev_set_rx_mode(dev);
4674 netif_addr_unlock_bh(dev);
4678 * dev_get_flags - get flags reported to userspace
4681 * Get the combination of flag bits exported through APIs to userspace.
4683 unsigned int dev_get_flags(const struct net_device *dev)
4687 flags = (dev->flags & ~(IFF_PROMISC |
4692 (dev->gflags & (IFF_PROMISC |
4695 if (netif_running(dev)) {
4696 if (netif_oper_up(dev))
4697 flags |= IFF_RUNNING;
4698 if (netif_carrier_ok(dev))
4699 flags |= IFF_LOWER_UP;
4700 if (netif_dormant(dev))
4701 flags |= IFF_DORMANT;
4706 EXPORT_SYMBOL(dev_get_flags);
4708 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4710 unsigned int old_flags = dev->flags;
4716 * Set the flags on our device.
4719 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4720 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4722 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4726 * Load in the correct multicast list now the flags have changed.
4729 if ((old_flags ^ flags) & IFF_MULTICAST)
4730 dev_change_rx_flags(dev, IFF_MULTICAST);
4732 dev_set_rx_mode(dev);
4735 * Have we downed the interface. We handle IFF_UP ourselves
4736 * according to user attempts to set it, rather than blindly
4741 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4742 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4745 dev_set_rx_mode(dev);
4748 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4749 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4751 dev->gflags ^= IFF_PROMISC;
4752 dev_set_promiscuity(dev, inc);
4755 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4756 is important. Some (broken) drivers set IFF_PROMISC, when
4757 IFF_ALLMULTI is requested not asking us and not reporting.
4759 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4760 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4762 dev->gflags ^= IFF_ALLMULTI;
4763 dev_set_allmulti(dev, inc);
4769 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4771 unsigned int changes = dev->flags ^ old_flags;
4773 if (changes & IFF_UP) {
4774 if (dev->flags & IFF_UP)
4775 call_netdevice_notifiers(NETDEV_UP, dev);
4777 call_netdevice_notifiers(NETDEV_DOWN, dev);
4780 if (dev->flags & IFF_UP &&
4781 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4782 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4786 * dev_change_flags - change device settings
4788 * @flags: device state flags
4790 * Change settings on device based state flags. The flags are
4791 * in the userspace exported format.
4793 int dev_change_flags(struct net_device *dev, unsigned int flags)
4796 unsigned int changes, old_flags = dev->flags;
4798 ret = __dev_change_flags(dev, flags);
4802 changes = old_flags ^ dev->flags;
4804 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4806 __dev_notify_flags(dev, old_flags);
4809 EXPORT_SYMBOL(dev_change_flags);
4812 * dev_set_mtu - Change maximum transfer unit
4814 * @new_mtu: new transfer unit
4816 * Change the maximum transfer size of the network device.
4818 int dev_set_mtu(struct net_device *dev, int new_mtu)
4820 const struct net_device_ops *ops = dev->netdev_ops;
4823 if (new_mtu == dev->mtu)
4826 /* MTU must be positive. */
4830 if (!netif_device_present(dev))
4834 if (ops->ndo_change_mtu)
4835 err = ops->ndo_change_mtu(dev, new_mtu);
4839 if (!err && dev->flags & IFF_UP)
4840 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4843 EXPORT_SYMBOL(dev_set_mtu);
4846 * dev_set_group - Change group this device belongs to
4848 * @new_group: group this device should belong to
4850 void dev_set_group(struct net_device *dev, int new_group)
4852 dev->group = new_group;
4854 EXPORT_SYMBOL(dev_set_group);
4857 * dev_set_mac_address - Change Media Access Control Address
4861 * Change the hardware (MAC) address of the device
4863 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4865 const struct net_device_ops *ops = dev->netdev_ops;
4868 if (!ops->ndo_set_mac_address)
4870 if (sa->sa_family != dev->type)
4872 if (!netif_device_present(dev))
4874 err = ops->ndo_set_mac_address(dev, sa);
4876 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4877 add_device_randomness(dev->dev_addr, dev->addr_len);
4880 EXPORT_SYMBOL(dev_set_mac_address);
4883 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4885 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4888 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4894 case SIOCGIFFLAGS: /* Get interface flags */
4895 ifr->ifr_flags = (short) dev_get_flags(dev);
4898 case SIOCGIFMETRIC: /* Get the metric on the interface
4899 (currently unused) */
4900 ifr->ifr_metric = 0;
4903 case SIOCGIFMTU: /* Get the MTU of a device */
4904 ifr->ifr_mtu = dev->mtu;
4909 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4911 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4912 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4913 ifr->ifr_hwaddr.sa_family = dev->type;
4921 ifr->ifr_map.mem_start = dev->mem_start;
4922 ifr->ifr_map.mem_end = dev->mem_end;
4923 ifr->ifr_map.base_addr = dev->base_addr;
4924 ifr->ifr_map.irq = dev->irq;
4925 ifr->ifr_map.dma = dev->dma;
4926 ifr->ifr_map.port = dev->if_port;
4930 ifr->ifr_ifindex = dev->ifindex;
4934 ifr->ifr_qlen = dev->tx_queue_len;
4938 /* dev_ioctl() should ensure this case
4950 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4952 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4955 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4956 const struct net_device_ops *ops;
4961 ops = dev->netdev_ops;
4964 case SIOCSIFFLAGS: /* Set interface flags */
4965 return dev_change_flags(dev, ifr->ifr_flags);
4967 case SIOCSIFMETRIC: /* Set the metric on the interface
4968 (currently unused) */
4971 case SIOCSIFMTU: /* Set the MTU of a device */
4972 return dev_set_mtu(dev, ifr->ifr_mtu);
4975 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4977 case SIOCSIFHWBROADCAST:
4978 if (ifr->ifr_hwaddr.sa_family != dev->type)
4980 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4981 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4982 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4986 if (ops->ndo_set_config) {
4987 if (!netif_device_present(dev))
4989 return ops->ndo_set_config(dev, &ifr->ifr_map);
4994 if (!ops->ndo_set_rx_mode ||
4995 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4997 if (!netif_device_present(dev))
4999 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5002 if (!ops->ndo_set_rx_mode ||
5003 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5005 if (!netif_device_present(dev))
5007 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5010 if (ifr->ifr_qlen < 0)
5012 dev->tx_queue_len = ifr->ifr_qlen;
5016 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5017 return dev_change_name(dev, ifr->ifr_newname);
5020 err = net_hwtstamp_validate(ifr);
5026 * Unknown or private ioctl
5029 if ((cmd >= SIOCDEVPRIVATE &&
5030 cmd <= SIOCDEVPRIVATE + 15) ||
5031 cmd == SIOCBONDENSLAVE ||
5032 cmd == SIOCBONDRELEASE ||
5033 cmd == SIOCBONDSETHWADDR ||
5034 cmd == SIOCBONDSLAVEINFOQUERY ||
5035 cmd == SIOCBONDINFOQUERY ||
5036 cmd == SIOCBONDCHANGEACTIVE ||
5037 cmd == SIOCGMIIPHY ||
5038 cmd == SIOCGMIIREG ||
5039 cmd == SIOCSMIIREG ||
5040 cmd == SIOCBRADDIF ||
5041 cmd == SIOCBRDELIF ||
5042 cmd == SIOCSHWTSTAMP ||
5043 cmd == SIOCWANDEV) {
5045 if (ops->ndo_do_ioctl) {
5046 if (netif_device_present(dev))
5047 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5059 * This function handles all "interface"-type I/O control requests. The actual
5060 * 'doing' part of this is dev_ifsioc above.
5064 * dev_ioctl - network device ioctl
5065 * @net: the applicable net namespace
5066 * @cmd: command to issue
5067 * @arg: pointer to a struct ifreq in user space
5069 * Issue ioctl functions to devices. This is normally called by the
5070 * user space syscall interfaces but can sometimes be useful for
5071 * other purposes. The return value is the return from the syscall if
5072 * positive or a negative errno code on error.
5075 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5081 /* One special case: SIOCGIFCONF takes ifconf argument
5082 and requires shared lock, because it sleeps writing
5086 if (cmd == SIOCGIFCONF) {
5088 ret = dev_ifconf(net, (char __user *) arg);
5092 if (cmd == SIOCGIFNAME)
5093 return dev_ifname(net, (struct ifreq __user *)arg);
5095 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5098 ifr.ifr_name[IFNAMSIZ-1] = 0;
5100 colon = strchr(ifr.ifr_name, ':');
5105 * See which interface the caller is talking about.
5110 * These ioctl calls:
5111 * - can be done by all.
5112 * - atomic and do not require locking.
5123 dev_load(net, ifr.ifr_name);
5125 ret = dev_ifsioc_locked(net, &ifr, cmd);
5130 if (copy_to_user(arg, &ifr,
5131 sizeof(struct ifreq)))
5137 dev_load(net, ifr.ifr_name);
5139 ret = dev_ethtool(net, &ifr);
5144 if (copy_to_user(arg, &ifr,
5145 sizeof(struct ifreq)))
5151 * These ioctl calls:
5152 * - require superuser power.
5153 * - require strict serialization.
5159 if (!capable(CAP_NET_ADMIN))
5161 dev_load(net, ifr.ifr_name);
5163 ret = dev_ifsioc(net, &ifr, cmd);
5168 if (copy_to_user(arg, &ifr,
5169 sizeof(struct ifreq)))
5175 * These ioctl calls:
5176 * - require superuser power.
5177 * - require strict serialization.
5178 * - do not return a value
5188 case SIOCSIFHWBROADCAST:
5191 case SIOCBONDENSLAVE:
5192 case SIOCBONDRELEASE:
5193 case SIOCBONDSETHWADDR:
5194 case SIOCBONDCHANGEACTIVE:
5198 if (!capable(CAP_NET_ADMIN))
5201 case SIOCBONDSLAVEINFOQUERY:
5202 case SIOCBONDINFOQUERY:
5203 dev_load(net, ifr.ifr_name);
5205 ret = dev_ifsioc(net, &ifr, cmd);
5210 /* Get the per device memory space. We can add this but
5211 * currently do not support it */
5213 /* Set the per device memory buffer space.
5214 * Not applicable in our case */
5219 * Unknown or private ioctl.
5222 if (cmd == SIOCWANDEV ||
5223 (cmd >= SIOCDEVPRIVATE &&
5224 cmd <= SIOCDEVPRIVATE + 15)) {
5225 dev_load(net, ifr.ifr_name);
5227 ret = dev_ifsioc(net, &ifr, cmd);
5229 if (!ret && copy_to_user(arg, &ifr,
5230 sizeof(struct ifreq)))
5234 /* Take care of Wireless Extensions */
5235 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5236 return wext_handle_ioctl(net, &ifr, cmd, arg);
5243 * dev_new_index - allocate an ifindex
5244 * @net: the applicable net namespace
5246 * Returns a suitable unique value for a new device interface
5247 * number. The caller must hold the rtnl semaphore or the
5248 * dev_base_lock to be sure it remains unique.
5250 static int dev_new_index(struct net *net)
5252 int ifindex = net->ifindex;
5256 if (!__dev_get_by_index(net, ifindex))
5257 return net->ifindex = ifindex;
5261 /* Delayed registration/unregisteration */
5262 static LIST_HEAD(net_todo_list);
5264 static void net_set_todo(struct net_device *dev)
5266 list_add_tail(&dev->todo_list, &net_todo_list);
5269 static void rollback_registered_many(struct list_head *head)
5271 struct net_device *dev, *tmp;
5273 BUG_ON(dev_boot_phase);
5276 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5277 /* Some devices call without registering
5278 * for initialization unwind. Remove those
5279 * devices and proceed with the remaining.
5281 if (dev->reg_state == NETREG_UNINITIALIZED) {
5282 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5286 list_del(&dev->unreg_list);
5289 dev->dismantle = true;
5290 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5293 /* If device is running, close it first. */
5294 dev_close_many(head);
5296 list_for_each_entry(dev, head, unreg_list) {
5297 /* And unlink it from device chain. */
5298 unlist_netdevice(dev);
5300 dev->reg_state = NETREG_UNREGISTERING;
5305 list_for_each_entry(dev, head, unreg_list) {
5306 /* Shutdown queueing discipline. */
5310 /* Notify protocols, that we are about to destroy
5311 this device. They should clean all the things.
5313 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5315 if (!dev->rtnl_link_ops ||
5316 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5317 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5320 * Flush the unicast and multicast chains
5325 if (dev->netdev_ops->ndo_uninit)
5326 dev->netdev_ops->ndo_uninit(dev);
5328 /* Notifier chain MUST detach us from master device. */
5329 WARN_ON(dev->master);
5331 /* Remove entries from kobject tree */
5332 netdev_unregister_kobject(dev);
5337 list_for_each_entry(dev, head, unreg_list)
5341 static void rollback_registered(struct net_device *dev)
5345 list_add(&dev->unreg_list, &single);
5346 rollback_registered_many(&single);
5350 static netdev_features_t netdev_fix_features(struct net_device *dev,
5351 netdev_features_t features)
5353 /* Fix illegal checksum combinations */
5354 if ((features & NETIF_F_HW_CSUM) &&
5355 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5356 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5357 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5360 /* Fix illegal SG+CSUM combinations. */
5361 if ((features & NETIF_F_SG) &&
5362 !(features & NETIF_F_ALL_CSUM)) {
5364 "Dropping NETIF_F_SG since no checksum feature.\n");
5365 features &= ~NETIF_F_SG;
5368 /* TSO requires that SG is present as well. */
5369 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5370 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5371 features &= ~NETIF_F_ALL_TSO;
5374 /* TSO ECN requires that TSO is present as well. */
5375 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5376 features &= ~NETIF_F_TSO_ECN;
5378 /* Software GSO depends on SG. */
5379 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5380 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5381 features &= ~NETIF_F_GSO;
5384 /* UFO needs SG and checksumming */
5385 if (features & NETIF_F_UFO) {
5386 /* maybe split UFO into V4 and V6? */
5387 if (!((features & NETIF_F_GEN_CSUM) ||
5388 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5389 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5391 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5392 features &= ~NETIF_F_UFO;
5395 if (!(features & NETIF_F_SG)) {
5397 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5398 features &= ~NETIF_F_UFO;
5405 int __netdev_update_features(struct net_device *dev)
5407 netdev_features_t features;
5412 features = netdev_get_wanted_features(dev);
5414 if (dev->netdev_ops->ndo_fix_features)
5415 features = dev->netdev_ops->ndo_fix_features(dev, features);
5417 /* driver might be less strict about feature dependencies */
5418 features = netdev_fix_features(dev, features);
5420 if (dev->features == features)
5423 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5424 &dev->features, &features);
5426 if (dev->netdev_ops->ndo_set_features)
5427 err = dev->netdev_ops->ndo_set_features(dev, features);
5429 if (unlikely(err < 0)) {
5431 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5432 err, &features, &dev->features);
5437 dev->features = features;
5443 * netdev_update_features - recalculate device features
5444 * @dev: the device to check
5446 * Recalculate dev->features set and send notifications if it
5447 * has changed. Should be called after driver or hardware dependent
5448 * conditions might have changed that influence the features.
5450 void netdev_update_features(struct net_device *dev)
5452 if (__netdev_update_features(dev))
5453 netdev_features_change(dev);
5455 EXPORT_SYMBOL(netdev_update_features);
5458 * netdev_change_features - recalculate device features
5459 * @dev: the device to check
5461 * Recalculate dev->features set and send notifications even
5462 * if they have not changed. Should be called instead of
5463 * netdev_update_features() if also dev->vlan_features might
5464 * have changed to allow the changes to be propagated to stacked
5467 void netdev_change_features(struct net_device *dev)
5469 __netdev_update_features(dev);
5470 netdev_features_change(dev);
5472 EXPORT_SYMBOL(netdev_change_features);
5475 * netif_stacked_transfer_operstate - transfer operstate
5476 * @rootdev: the root or lower level device to transfer state from
5477 * @dev: the device to transfer operstate to
5479 * Transfer operational state from root to device. This is normally
5480 * called when a stacking relationship exists between the root
5481 * device and the device(a leaf device).
5483 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5484 struct net_device *dev)
5486 if (rootdev->operstate == IF_OPER_DORMANT)
5487 netif_dormant_on(dev);
5489 netif_dormant_off(dev);
5491 if (netif_carrier_ok(rootdev)) {
5492 if (!netif_carrier_ok(dev))
5493 netif_carrier_on(dev);
5495 if (netif_carrier_ok(dev))
5496 netif_carrier_off(dev);
5499 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5502 static int netif_alloc_rx_queues(struct net_device *dev)
5504 unsigned int i, count = dev->num_rx_queues;
5505 struct netdev_rx_queue *rx;
5509 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5511 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5516 for (i = 0; i < count; i++)
5522 static void netdev_init_one_queue(struct net_device *dev,
5523 struct netdev_queue *queue, void *_unused)
5525 /* Initialize queue lock */
5526 spin_lock_init(&queue->_xmit_lock);
5527 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5528 queue->xmit_lock_owner = -1;
5529 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5532 dql_init(&queue->dql, HZ);
5536 static int netif_alloc_netdev_queues(struct net_device *dev)
5538 unsigned int count = dev->num_tx_queues;
5539 struct netdev_queue *tx;
5543 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5545 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5550 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5551 spin_lock_init(&dev->tx_global_lock);
5557 * register_netdevice - register a network device
5558 * @dev: device to register
5560 * Take a completed network device structure and add it to the kernel
5561 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5562 * chain. 0 is returned on success. A negative errno code is returned
5563 * on a failure to set up the device, or if the name is a duplicate.
5565 * Callers must hold the rtnl semaphore. You may want
5566 * register_netdev() instead of this.
5569 * The locking appears insufficient to guarantee two parallel registers
5570 * will not get the same name.
5573 int register_netdevice(struct net_device *dev)
5576 struct net *net = dev_net(dev);
5578 BUG_ON(dev_boot_phase);
5583 /* When net_device's are persistent, this will be fatal. */
5584 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5587 spin_lock_init(&dev->addr_list_lock);
5588 netdev_set_addr_lockdep_class(dev);
5592 ret = dev_get_valid_name(dev, dev->name);
5596 /* Init, if this function is available */
5597 if (dev->netdev_ops->ndo_init) {
5598 ret = dev->netdev_ops->ndo_init(dev);
5608 dev->ifindex = dev_new_index(net);
5609 else if (__dev_get_by_index(net, dev->ifindex))
5612 if (dev->iflink == -1)
5613 dev->iflink = dev->ifindex;
5615 /* Transfer changeable features to wanted_features and enable
5616 * software offloads (GSO and GRO).
5618 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5619 dev->features |= NETIF_F_SOFT_FEATURES;
5620 dev->wanted_features = dev->features & dev->hw_features;
5622 /* Turn on no cache copy if HW is doing checksum */
5623 if (!(dev->flags & IFF_LOOPBACK)) {
5624 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5625 if (dev->features & NETIF_F_ALL_CSUM) {
5626 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5627 dev->features |= NETIF_F_NOCACHE_COPY;
5631 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5633 dev->vlan_features |= NETIF_F_HIGHDMA;
5635 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5636 ret = notifier_to_errno(ret);
5640 ret = netdev_register_kobject(dev);
5643 dev->reg_state = NETREG_REGISTERED;
5645 __netdev_update_features(dev);
5648 * Default initial state at registry is that the
5649 * device is present.
5652 set_bit(__LINK_STATE_PRESENT, &dev->state);
5654 linkwatch_init_dev(dev);
5656 dev_init_scheduler(dev);
5658 list_netdevice(dev);
5659 add_device_randomness(dev->dev_addr, dev->addr_len);
5661 /* Notify protocols, that a new device appeared. */
5662 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5663 ret = notifier_to_errno(ret);
5665 rollback_registered(dev);
5666 dev->reg_state = NETREG_UNREGISTERED;
5669 * Prevent userspace races by waiting until the network
5670 * device is fully setup before sending notifications.
5672 if (!dev->rtnl_link_ops ||
5673 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5674 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5680 if (dev->netdev_ops->ndo_uninit)
5681 dev->netdev_ops->ndo_uninit(dev);
5684 EXPORT_SYMBOL(register_netdevice);
5687 * init_dummy_netdev - init a dummy network device for NAPI
5688 * @dev: device to init
5690 * This takes a network device structure and initialize the minimum
5691 * amount of fields so it can be used to schedule NAPI polls without
5692 * registering a full blown interface. This is to be used by drivers
5693 * that need to tie several hardware interfaces to a single NAPI
5694 * poll scheduler due to HW limitations.
5696 int init_dummy_netdev(struct net_device *dev)
5698 /* Clear everything. Note we don't initialize spinlocks
5699 * are they aren't supposed to be taken by any of the
5700 * NAPI code and this dummy netdev is supposed to be
5701 * only ever used for NAPI polls
5703 memset(dev, 0, sizeof(struct net_device));
5705 /* make sure we BUG if trying to hit standard
5706 * register/unregister code path
5708 dev->reg_state = NETREG_DUMMY;
5710 /* NAPI wants this */
5711 INIT_LIST_HEAD(&dev->napi_list);
5713 /* a dummy interface is started by default */
5714 set_bit(__LINK_STATE_PRESENT, &dev->state);
5715 set_bit(__LINK_STATE_START, &dev->state);
5717 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5718 * because users of this 'device' dont need to change
5724 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5728 * register_netdev - register a network device
5729 * @dev: device to register
5731 * Take a completed network device structure and add it to the kernel
5732 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5733 * chain. 0 is returned on success. A negative errno code is returned
5734 * on a failure to set up the device, or if the name is a duplicate.
5736 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5737 * and expands the device name if you passed a format string to
5740 int register_netdev(struct net_device *dev)
5745 err = register_netdevice(dev);
5749 EXPORT_SYMBOL(register_netdev);
5751 int netdev_refcnt_read(const struct net_device *dev)
5755 for_each_possible_cpu(i)
5756 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5759 EXPORT_SYMBOL(netdev_refcnt_read);
5762 * netdev_wait_allrefs - wait until all references are gone.
5763 * @dev: target net_device
5765 * This is called when unregistering network devices.
5767 * Any protocol or device that holds a reference should register
5768 * for netdevice notification, and cleanup and put back the
5769 * reference if they receive an UNREGISTER event.
5770 * We can get stuck here if buggy protocols don't correctly
5773 static void netdev_wait_allrefs(struct net_device *dev)
5775 unsigned long rebroadcast_time, warning_time;
5778 linkwatch_forget_dev(dev);
5780 rebroadcast_time = warning_time = jiffies;
5781 refcnt = netdev_refcnt_read(dev);
5783 while (refcnt != 0) {
5784 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5787 /* Rebroadcast unregister notification */
5788 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5794 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5795 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5797 /* We must not have linkwatch events
5798 * pending on unregister. If this
5799 * happens, we simply run the queue
5800 * unscheduled, resulting in a noop
5803 linkwatch_run_queue();
5808 rebroadcast_time = jiffies;
5813 refcnt = netdev_refcnt_read(dev);
5815 if (time_after(jiffies, warning_time + 10 * HZ)) {
5816 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5818 warning_time = jiffies;
5827 * register_netdevice(x1);
5828 * register_netdevice(x2);
5830 * unregister_netdevice(y1);
5831 * unregister_netdevice(y2);
5837 * We are invoked by rtnl_unlock().
5838 * This allows us to deal with problems:
5839 * 1) We can delete sysfs objects which invoke hotplug
5840 * without deadlocking with linkwatch via keventd.
5841 * 2) Since we run with the RTNL semaphore not held, we can sleep
5842 * safely in order to wait for the netdev refcnt to drop to zero.
5844 * We must not return until all unregister events added during
5845 * the interval the lock was held have been completed.
5847 void netdev_run_todo(void)
5849 struct list_head list;
5851 /* Snapshot list, allow later requests */
5852 list_replace_init(&net_todo_list, &list);
5857 /* Wait for rcu callbacks to finish before next phase */
5858 if (!list_empty(&list))
5861 while (!list_empty(&list)) {
5862 struct net_device *dev
5863 = list_first_entry(&list, struct net_device, todo_list);
5864 list_del(&dev->todo_list);
5867 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5870 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5871 pr_err("network todo '%s' but state %d\n",
5872 dev->name, dev->reg_state);
5877 dev->reg_state = NETREG_UNREGISTERED;
5879 on_each_cpu(flush_backlog, dev, 1);
5881 netdev_wait_allrefs(dev);
5884 BUG_ON(netdev_refcnt_read(dev));
5885 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5886 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5887 WARN_ON(dev->dn_ptr);
5889 if (dev->destructor)
5890 dev->destructor(dev);
5892 /* Free network device */
5893 kobject_put(&dev->dev.kobj);
5897 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5898 * fields in the same order, with only the type differing.
5900 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5901 const struct net_device_stats *netdev_stats)
5903 #if BITS_PER_LONG == 64
5904 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5905 memcpy(stats64, netdev_stats, sizeof(*stats64));
5907 size_t i, n = sizeof(*stats64) / sizeof(u64);
5908 const unsigned long *src = (const unsigned long *)netdev_stats;
5909 u64 *dst = (u64 *)stats64;
5911 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5912 sizeof(*stats64) / sizeof(u64));
5913 for (i = 0; i < n; i++)
5917 EXPORT_SYMBOL(netdev_stats_to_stats64);
5920 * dev_get_stats - get network device statistics
5921 * @dev: device to get statistics from
5922 * @storage: place to store stats
5924 * Get network statistics from device. Return @storage.
5925 * The device driver may provide its own method by setting
5926 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5927 * otherwise the internal statistics structure is used.
5929 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5930 struct rtnl_link_stats64 *storage)
5932 const struct net_device_ops *ops = dev->netdev_ops;
5934 if (ops->ndo_get_stats64) {
5935 memset(storage, 0, sizeof(*storage));
5936 ops->ndo_get_stats64(dev, storage);
5937 } else if (ops->ndo_get_stats) {
5938 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5940 netdev_stats_to_stats64(storage, &dev->stats);
5942 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5945 EXPORT_SYMBOL(dev_get_stats);
5947 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5949 struct netdev_queue *queue = dev_ingress_queue(dev);
5951 #ifdef CONFIG_NET_CLS_ACT
5954 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5957 netdev_init_one_queue(dev, queue, NULL);
5958 queue->qdisc = &noop_qdisc;
5959 queue->qdisc_sleeping = &noop_qdisc;
5960 rcu_assign_pointer(dev->ingress_queue, queue);
5966 * alloc_netdev_mqs - allocate network device
5967 * @sizeof_priv: size of private data to allocate space for
5968 * @name: device name format string
5969 * @setup: callback to initialize device
5970 * @txqs: the number of TX subqueues to allocate
5971 * @rxqs: the number of RX subqueues to allocate
5973 * Allocates a struct net_device with private data area for driver use
5974 * and performs basic initialization. Also allocates subquue structs
5975 * for each queue on the device.
5977 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5978 void (*setup)(struct net_device *),
5979 unsigned int txqs, unsigned int rxqs)
5981 struct net_device *dev;
5983 struct net_device *p;
5985 BUG_ON(strlen(name) >= sizeof(dev->name));
5988 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5994 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5999 alloc_size = sizeof(struct net_device);
6001 /* ensure 32-byte alignment of private area */
6002 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6003 alloc_size += sizeof_priv;
6005 /* ensure 32-byte alignment of whole construct */
6006 alloc_size += NETDEV_ALIGN - 1;
6008 p = kzalloc(alloc_size, GFP_KERNEL);
6010 pr_err("alloc_netdev: Unable to allocate device\n");
6014 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6015 dev->padded = (char *)dev - (char *)p;
6017 dev->pcpu_refcnt = alloc_percpu(int);
6018 if (!dev->pcpu_refcnt)
6021 if (dev_addr_init(dev))
6027 dev_net_set(dev, &init_net);
6029 dev->gso_max_size = GSO_MAX_SIZE;
6030 dev->gso_max_segs = GSO_MAX_SEGS;
6032 INIT_LIST_HEAD(&dev->napi_list);
6033 INIT_LIST_HEAD(&dev->unreg_list);
6034 INIT_LIST_HEAD(&dev->link_watch_list);
6035 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6038 dev->num_tx_queues = txqs;
6039 dev->real_num_tx_queues = txqs;
6040 if (netif_alloc_netdev_queues(dev))
6044 dev->num_rx_queues = rxqs;
6045 dev->real_num_rx_queues = rxqs;
6046 if (netif_alloc_rx_queues(dev))
6050 strcpy(dev->name, name);
6051 dev->group = INIT_NETDEV_GROUP;
6059 free_percpu(dev->pcpu_refcnt);
6069 EXPORT_SYMBOL(alloc_netdev_mqs);
6072 * free_netdev - free network device
6075 * This function does the last stage of destroying an allocated device
6076 * interface. The reference to the device object is released.
6077 * If this is the last reference then it will be freed.
6079 void free_netdev(struct net_device *dev)
6081 struct napi_struct *p, *n;
6083 release_net(dev_net(dev));
6090 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6092 /* Flush device addresses */
6093 dev_addr_flush(dev);
6095 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6098 free_percpu(dev->pcpu_refcnt);
6099 dev->pcpu_refcnt = NULL;
6101 /* Compatibility with error handling in drivers */
6102 if (dev->reg_state == NETREG_UNINITIALIZED) {
6103 kfree((char *)dev - dev->padded);
6107 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6108 dev->reg_state = NETREG_RELEASED;
6110 /* will free via device release */
6111 put_device(&dev->dev);
6113 EXPORT_SYMBOL(free_netdev);
6116 * synchronize_net - Synchronize with packet receive processing
6118 * Wait for packets currently being received to be done.
6119 * Does not block later packets from starting.
6121 void synchronize_net(void)
6124 if (rtnl_is_locked())
6125 synchronize_rcu_expedited();
6129 EXPORT_SYMBOL(synchronize_net);
6132 * unregister_netdevice_queue - remove device from the kernel
6136 * This function shuts down a device interface and removes it
6137 * from the kernel tables.
6138 * If head not NULL, device is queued to be unregistered later.
6140 * Callers must hold the rtnl semaphore. You may want
6141 * unregister_netdev() instead of this.
6144 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6149 list_move_tail(&dev->unreg_list, head);
6151 rollback_registered(dev);
6152 /* Finish processing unregister after unlock */
6156 EXPORT_SYMBOL(unregister_netdevice_queue);
6159 * unregister_netdevice_many - unregister many devices
6160 * @head: list of devices
6162 void unregister_netdevice_many(struct list_head *head)
6164 struct net_device *dev;
6166 if (!list_empty(head)) {
6167 rollback_registered_many(head);
6168 list_for_each_entry(dev, head, unreg_list)
6172 EXPORT_SYMBOL(unregister_netdevice_many);
6175 * unregister_netdev - remove device from the kernel
6178 * This function shuts down a device interface and removes it
6179 * from the kernel tables.
6181 * This is just a wrapper for unregister_netdevice that takes
6182 * the rtnl semaphore. In general you want to use this and not
6183 * unregister_netdevice.
6185 void unregister_netdev(struct net_device *dev)
6188 unregister_netdevice(dev);
6191 EXPORT_SYMBOL(unregister_netdev);
6194 * dev_change_net_namespace - move device to different nethost namespace
6196 * @net: network namespace
6197 * @pat: If not NULL name pattern to try if the current device name
6198 * is already taken in the destination network namespace.
6200 * This function shuts down a device interface and moves it
6201 * to a new network namespace. On success 0 is returned, on
6202 * a failure a netagive errno code is returned.
6204 * Callers must hold the rtnl semaphore.
6207 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6213 /* Don't allow namespace local devices to be moved. */
6215 if (dev->features & NETIF_F_NETNS_LOCAL)
6218 /* Ensure the device has been registrered */
6220 if (dev->reg_state != NETREG_REGISTERED)
6223 /* Get out if there is nothing todo */
6225 if (net_eq(dev_net(dev), net))
6228 /* Pick the destination device name, and ensure
6229 * we can use it in the destination network namespace.
6232 if (__dev_get_by_name(net, dev->name)) {
6233 /* We get here if we can't use the current device name */
6236 if (dev_get_valid_name(dev, pat) < 0)
6241 * And now a mini version of register_netdevice unregister_netdevice.
6244 /* If device is running close it first. */
6247 /* And unlink it from device chain */
6249 unlist_netdevice(dev);
6253 /* Shutdown queueing discipline. */
6256 /* Notify protocols, that we are about to destroy
6257 this device. They should clean all the things.
6259 Note that dev->reg_state stays at NETREG_REGISTERED.
6260 This is wanted because this way 8021q and macvlan know
6261 the device is just moving and can keep their slaves up.
6263 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6265 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6266 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6269 * Flush the unicast and multicast chains
6274 /* Actually switch the network namespace */
6275 dev_net_set(dev, net);
6277 /* If there is an ifindex conflict assign a new one */
6278 if (__dev_get_by_index(net, dev->ifindex)) {
6279 int iflink = (dev->iflink == dev->ifindex);
6280 dev->ifindex = dev_new_index(net);
6282 dev->iflink = dev->ifindex;
6285 /* Fixup kobjects */
6286 err = device_rename(&dev->dev, dev->name);
6289 /* Add the device back in the hashes */
6290 list_netdevice(dev);
6292 /* Notify protocols, that a new device appeared. */
6293 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6296 * Prevent userspace races by waiting until the network
6297 * device is fully setup before sending notifications.
6299 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6306 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6308 static int dev_cpu_callback(struct notifier_block *nfb,
6309 unsigned long action,
6312 struct sk_buff **list_skb;
6313 struct sk_buff *skb;
6314 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6315 struct softnet_data *sd, *oldsd;
6317 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6320 local_irq_disable();
6321 cpu = smp_processor_id();
6322 sd = &per_cpu(softnet_data, cpu);
6323 oldsd = &per_cpu(softnet_data, oldcpu);
6325 /* Find end of our completion_queue. */
6326 list_skb = &sd->completion_queue;
6328 list_skb = &(*list_skb)->next;
6329 /* Append completion queue from offline CPU. */
6330 *list_skb = oldsd->completion_queue;
6331 oldsd->completion_queue = NULL;
6333 /* Append output queue from offline CPU. */
6334 if (oldsd->output_queue) {
6335 *sd->output_queue_tailp = oldsd->output_queue;
6336 sd->output_queue_tailp = oldsd->output_queue_tailp;
6337 oldsd->output_queue = NULL;
6338 oldsd->output_queue_tailp = &oldsd->output_queue;
6340 /* Append NAPI poll list from offline CPU. */
6341 if (!list_empty(&oldsd->poll_list)) {
6342 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6343 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6346 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6349 /* Process offline CPU's input_pkt_queue */
6350 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6352 input_queue_head_incr(oldsd);
6354 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6356 input_queue_head_incr(oldsd);
6364 * netdev_increment_features - increment feature set by one
6365 * @all: current feature set
6366 * @one: new feature set
6367 * @mask: mask feature set
6369 * Computes a new feature set after adding a device with feature set
6370 * @one to the master device with current feature set @all. Will not
6371 * enable anything that is off in @mask. Returns the new feature set.
6373 netdev_features_t netdev_increment_features(netdev_features_t all,
6374 netdev_features_t one, netdev_features_t mask)
6376 if (mask & NETIF_F_GEN_CSUM)
6377 mask |= NETIF_F_ALL_CSUM;
6378 mask |= NETIF_F_VLAN_CHALLENGED;
6380 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6381 all &= one | ~NETIF_F_ALL_FOR_ALL;
6383 /* If one device supports hw checksumming, set for all. */
6384 if (all & NETIF_F_GEN_CSUM)
6385 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6389 EXPORT_SYMBOL(netdev_increment_features);
6391 static struct hlist_head *netdev_create_hash(void)
6394 struct hlist_head *hash;
6396 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6398 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6399 INIT_HLIST_HEAD(&hash[i]);
6404 /* Initialize per network namespace state */
6405 static int __net_init netdev_init(struct net *net)
6407 if (net != &init_net)
6408 INIT_LIST_HEAD(&net->dev_base_head);
6410 net->dev_name_head = netdev_create_hash();
6411 if (net->dev_name_head == NULL)
6414 net->dev_index_head = netdev_create_hash();
6415 if (net->dev_index_head == NULL)
6421 kfree(net->dev_name_head);
6427 * netdev_drivername - network driver for the device
6428 * @dev: network device
6430 * Determine network driver for device.
6432 const char *netdev_drivername(const struct net_device *dev)
6434 const struct device_driver *driver;
6435 const struct device *parent;
6436 const char *empty = "";
6438 parent = dev->dev.parent;
6442 driver = parent->driver;
6443 if (driver && driver->name)
6444 return driver->name;
6448 int __netdev_printk(const char *level, const struct net_device *dev,
6449 struct va_format *vaf)
6453 if (dev && dev->dev.parent)
6454 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6455 netdev_name(dev), vaf);
6457 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6459 r = printk("%s(NULL net_device): %pV", level, vaf);
6463 EXPORT_SYMBOL(__netdev_printk);
6465 int netdev_printk(const char *level, const struct net_device *dev,
6466 const char *format, ...)
6468 struct va_format vaf;
6472 va_start(args, format);
6477 r = __netdev_printk(level, dev, &vaf);
6482 EXPORT_SYMBOL(netdev_printk);
6484 #define define_netdev_printk_level(func, level) \
6485 int func(const struct net_device *dev, const char *fmt, ...) \
6488 struct va_format vaf; \
6491 va_start(args, fmt); \
6496 r = __netdev_printk(level, dev, &vaf); \
6501 EXPORT_SYMBOL(func);
6503 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6504 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6505 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6506 define_netdev_printk_level(netdev_err, KERN_ERR);
6507 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6508 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6509 define_netdev_printk_level(netdev_info, KERN_INFO);
6511 static void __net_exit netdev_exit(struct net *net)
6513 kfree(net->dev_name_head);
6514 kfree(net->dev_index_head);
6517 static struct pernet_operations __net_initdata netdev_net_ops = {
6518 .init = netdev_init,
6519 .exit = netdev_exit,
6522 static void __net_exit default_device_exit(struct net *net)
6524 struct net_device *dev, *aux;
6526 * Push all migratable network devices back to the
6527 * initial network namespace
6530 for_each_netdev_safe(net, dev, aux) {
6532 char fb_name[IFNAMSIZ];
6534 /* Ignore unmoveable devices (i.e. loopback) */
6535 if (dev->features & NETIF_F_NETNS_LOCAL)
6538 /* Leave virtual devices for the generic cleanup */
6539 if (dev->rtnl_link_ops)
6542 /* Push remaining network devices to init_net */
6543 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6544 err = dev_change_net_namespace(dev, &init_net, fb_name);
6546 pr_emerg("%s: failed to move %s to init_net: %d\n",
6547 __func__, dev->name, err);
6554 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6556 /* At exit all network devices most be removed from a network
6557 * namespace. Do this in the reverse order of registration.
6558 * Do this across as many network namespaces as possible to
6559 * improve batching efficiency.
6561 struct net_device *dev;
6563 LIST_HEAD(dev_kill_list);
6566 list_for_each_entry(net, net_list, exit_list) {
6567 for_each_netdev_reverse(net, dev) {
6568 if (dev->rtnl_link_ops)
6569 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6571 unregister_netdevice_queue(dev, &dev_kill_list);
6574 unregister_netdevice_many(&dev_kill_list);
6575 list_del(&dev_kill_list);
6579 static struct pernet_operations __net_initdata default_device_ops = {
6580 .exit = default_device_exit,
6581 .exit_batch = default_device_exit_batch,
6585 * Initialize the DEV module. At boot time this walks the device list and
6586 * unhooks any devices that fail to initialise (normally hardware not
6587 * present) and leaves us with a valid list of present and active devices.
6592 * This is called single threaded during boot, so no need
6593 * to take the rtnl semaphore.
6595 static int __init net_dev_init(void)
6597 int i, rc = -ENOMEM;
6599 BUG_ON(!dev_boot_phase);
6601 if (dev_proc_init())
6604 if (netdev_kobject_init())
6607 INIT_LIST_HEAD(&ptype_all);
6608 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6609 INIT_LIST_HEAD(&ptype_base[i]);
6611 if (register_pernet_subsys(&netdev_net_ops))
6615 * Initialise the packet receive queues.
6618 for_each_possible_cpu(i) {
6619 struct softnet_data *sd = &per_cpu(softnet_data, i);
6621 memset(sd, 0, sizeof(*sd));
6622 skb_queue_head_init(&sd->input_pkt_queue);
6623 skb_queue_head_init(&sd->process_queue);
6624 sd->completion_queue = NULL;
6625 INIT_LIST_HEAD(&sd->poll_list);
6626 sd->output_queue = NULL;
6627 sd->output_queue_tailp = &sd->output_queue;
6629 sd->csd.func = rps_trigger_softirq;
6635 sd->backlog.poll = process_backlog;
6636 sd->backlog.weight = weight_p;
6637 sd->backlog.gro_list = NULL;
6638 sd->backlog.gro_count = 0;
6643 /* The loopback device is special if any other network devices
6644 * is present in a network namespace the loopback device must
6645 * be present. Since we now dynamically allocate and free the
6646 * loopback device ensure this invariant is maintained by
6647 * keeping the loopback device as the first device on the
6648 * list of network devices. Ensuring the loopback devices
6649 * is the first device that appears and the last network device
6652 if (register_pernet_device(&loopback_net_ops))
6655 if (register_pernet_device(&default_device_ops))
6658 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6659 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6661 hotcpu_notifier(dev_cpu_callback, 0);
6669 subsys_initcall(net_dev_init);
6671 static int __init initialize_hashrnd(void)
6673 get_random_bytes(&hashrnd, sizeof(hashrnd));
6677 late_initcall_sync(initialize_hashrnd);