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)
1060 if (len >= IFALIASZ)
1065 kfree(dev->ifalias);
1066 dev->ifalias = NULL;
1071 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1075 strlcpy(dev->ifalias, alias, len+1);
1081 * netdev_features_change - device changes features
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed features.
1086 void netdev_features_change(struct net_device *dev)
1088 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1090 EXPORT_SYMBOL(netdev_features_change);
1093 * netdev_state_change - device changes state
1094 * @dev: device to cause notification
1096 * Called to indicate a device has changed state. This function calls
1097 * the notifier chains for netdev_chain and sends a NEWLINK message
1098 * to the routing socket.
1100 void netdev_state_change(struct net_device *dev)
1102 if (dev->flags & IFF_UP) {
1103 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1104 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1107 EXPORT_SYMBOL(netdev_state_change);
1109 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1111 return call_netdevice_notifiers(event, dev);
1113 EXPORT_SYMBOL(netdev_bonding_change);
1116 * dev_load - load a network module
1117 * @net: the applicable net namespace
1118 * @name: name of interface
1120 * If a network interface is not present and the process has suitable
1121 * privileges this function loads the module. If module loading is not
1122 * available in this kernel then it becomes a nop.
1125 void dev_load(struct net *net, const char *name)
1127 struct net_device *dev;
1131 dev = dev_get_by_name_rcu(net, name);
1135 if (no_module && capable(CAP_NET_ADMIN))
1136 no_module = request_module("netdev-%s", name);
1137 if (no_module && capable(CAP_SYS_MODULE)) {
1138 if (!request_module("%s", name))
1139 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1143 EXPORT_SYMBOL(dev_load);
1145 static int __dev_open(struct net_device *dev)
1147 const struct net_device_ops *ops = dev->netdev_ops;
1152 if (!netif_device_present(dev))
1155 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1156 ret = notifier_to_errno(ret);
1160 set_bit(__LINK_STATE_START, &dev->state);
1162 if (ops->ndo_validate_addr)
1163 ret = ops->ndo_validate_addr(dev);
1165 if (!ret && ops->ndo_open)
1166 ret = ops->ndo_open(dev);
1169 clear_bit(__LINK_STATE_START, &dev->state);
1171 dev->flags |= IFF_UP;
1172 net_dmaengine_get();
1173 dev_set_rx_mode(dev);
1175 add_device_randomness(dev->dev_addr, dev->addr_len);
1182 * dev_open - prepare an interface for use.
1183 * @dev: device to open
1185 * Takes a device from down to up state. The device's private open
1186 * function is invoked and then the multicast lists are loaded. Finally
1187 * the device is moved into the up state and a %NETDEV_UP message is
1188 * sent to the netdev notifier chain.
1190 * Calling this function on an active interface is a nop. On a failure
1191 * a negative errno code is returned.
1193 int dev_open(struct net_device *dev)
1197 if (dev->flags & IFF_UP)
1200 ret = __dev_open(dev);
1204 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1205 call_netdevice_notifiers(NETDEV_UP, dev);
1209 EXPORT_SYMBOL(dev_open);
1211 static int __dev_close_many(struct list_head *head)
1213 struct net_device *dev;
1218 list_for_each_entry(dev, head, unreg_list) {
1219 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1221 clear_bit(__LINK_STATE_START, &dev->state);
1223 /* Synchronize to scheduled poll. We cannot touch poll list, it
1224 * can be even on different cpu. So just clear netif_running().
1226 * dev->stop() will invoke napi_disable() on all of it's
1227 * napi_struct instances on this device.
1229 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1232 dev_deactivate_many(head);
1234 list_for_each_entry(dev, head, unreg_list) {
1235 const struct net_device_ops *ops = dev->netdev_ops;
1238 * Call the device specific close. This cannot fail.
1239 * Only if device is UP
1241 * We allow it to be called even after a DETACH hot-plug
1247 dev->flags &= ~IFF_UP;
1248 net_dmaengine_put();
1254 static int __dev_close(struct net_device *dev)
1259 list_add(&dev->unreg_list, &single);
1260 retval = __dev_close_many(&single);
1265 static int dev_close_many(struct list_head *head)
1267 struct net_device *dev, *tmp;
1268 LIST_HEAD(tmp_list);
1270 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1271 if (!(dev->flags & IFF_UP))
1272 list_move(&dev->unreg_list, &tmp_list);
1274 __dev_close_many(head);
1276 list_for_each_entry(dev, head, unreg_list) {
1277 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1278 call_netdevice_notifiers(NETDEV_DOWN, dev);
1281 /* rollback_registered_many needs the complete original list */
1282 list_splice(&tmp_list, head);
1287 * dev_close - shutdown an interface.
1288 * @dev: device to shutdown
1290 * This function moves an active device into down state. A
1291 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1292 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1295 int dev_close(struct net_device *dev)
1297 if (dev->flags & IFF_UP) {
1300 list_add(&dev->unreg_list, &single);
1301 dev_close_many(&single);
1306 EXPORT_SYMBOL(dev_close);
1310 * dev_disable_lro - disable Large Receive Offload on a device
1313 * Disable Large Receive Offload (LRO) on a net device. Must be
1314 * called under RTNL. This is needed if received packets may be
1315 * forwarded to another interface.
1317 void dev_disable_lro(struct net_device *dev)
1320 * If we're trying to disable lro on a vlan device
1321 * use the underlying physical device instead
1323 if (is_vlan_dev(dev))
1324 dev = vlan_dev_real_dev(dev);
1326 dev->wanted_features &= ~NETIF_F_LRO;
1327 netdev_update_features(dev);
1329 if (unlikely(dev->features & NETIF_F_LRO))
1330 netdev_WARN(dev, "failed to disable LRO!\n");
1332 EXPORT_SYMBOL(dev_disable_lro);
1335 static int dev_boot_phase = 1;
1338 * register_netdevice_notifier - register a network notifier block
1341 * Register a notifier to be called when network device events occur.
1342 * The notifier passed is linked into the kernel structures and must
1343 * not be reused until it has been unregistered. A negative errno code
1344 * is returned on a failure.
1346 * When registered all registration and up events are replayed
1347 * to the new notifier to allow device to have a race free
1348 * view of the network device list.
1351 int register_netdevice_notifier(struct notifier_block *nb)
1353 struct net_device *dev;
1354 struct net_device *last;
1359 err = raw_notifier_chain_register(&netdev_chain, nb);
1365 for_each_netdev(net, dev) {
1366 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1367 err = notifier_to_errno(err);
1371 if (!(dev->flags & IFF_UP))
1374 nb->notifier_call(nb, NETDEV_UP, dev);
1385 for_each_netdev(net, dev) {
1389 if (dev->flags & IFF_UP) {
1390 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1391 nb->notifier_call(nb, NETDEV_DOWN, dev);
1393 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1394 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1399 raw_notifier_chain_unregister(&netdev_chain, nb);
1402 EXPORT_SYMBOL(register_netdevice_notifier);
1405 * unregister_netdevice_notifier - unregister a network notifier block
1408 * Unregister a notifier previously registered by
1409 * register_netdevice_notifier(). The notifier is unlinked into the
1410 * kernel structures and may then be reused. A negative errno code
1411 * is returned on a failure.
1413 * After unregistering unregister and down device events are synthesized
1414 * for all devices on the device list to the removed notifier to remove
1415 * the need for special case cleanup code.
1418 int unregister_netdevice_notifier(struct notifier_block *nb)
1420 struct net_device *dev;
1425 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1430 for_each_netdev(net, dev) {
1431 if (dev->flags & IFF_UP) {
1432 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1433 nb->notifier_call(nb, NETDEV_DOWN, dev);
1435 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1436 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1443 EXPORT_SYMBOL(unregister_netdevice_notifier);
1446 * call_netdevice_notifiers - call all network notifier blocks
1447 * @val: value passed unmodified to notifier function
1448 * @dev: net_device pointer passed unmodified to notifier function
1450 * Call all network notifier blocks. Parameters and return value
1451 * are as for raw_notifier_call_chain().
1454 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1457 return raw_notifier_call_chain(&netdev_chain, val, dev);
1459 EXPORT_SYMBOL(call_netdevice_notifiers);
1461 static struct static_key netstamp_needed __read_mostly;
1462 #ifdef HAVE_JUMP_LABEL
1463 /* We are not allowed to call static_key_slow_dec() from irq context
1464 * If net_disable_timestamp() is called from irq context, defer the
1465 * static_key_slow_dec() calls.
1467 static atomic_t netstamp_needed_deferred;
1470 void net_enable_timestamp(void)
1472 #ifdef HAVE_JUMP_LABEL
1473 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1477 static_key_slow_dec(&netstamp_needed);
1481 WARN_ON(in_interrupt());
1482 static_key_slow_inc(&netstamp_needed);
1484 EXPORT_SYMBOL(net_enable_timestamp);
1486 void net_disable_timestamp(void)
1488 #ifdef HAVE_JUMP_LABEL
1489 if (in_interrupt()) {
1490 atomic_inc(&netstamp_needed_deferred);
1494 static_key_slow_dec(&netstamp_needed);
1496 EXPORT_SYMBOL(net_disable_timestamp);
1498 static inline void net_timestamp_set(struct sk_buff *skb)
1500 skb->tstamp.tv64 = 0;
1501 if (static_key_false(&netstamp_needed))
1502 __net_timestamp(skb);
1505 #define net_timestamp_check(COND, SKB) \
1506 if (static_key_false(&netstamp_needed)) { \
1507 if ((COND) && !(SKB)->tstamp.tv64) \
1508 __net_timestamp(SKB); \
1511 static int net_hwtstamp_validate(struct ifreq *ifr)
1513 struct hwtstamp_config cfg;
1514 enum hwtstamp_tx_types tx_type;
1515 enum hwtstamp_rx_filters rx_filter;
1516 int tx_type_valid = 0;
1517 int rx_filter_valid = 0;
1519 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1522 if (cfg.flags) /* reserved for future extensions */
1525 tx_type = cfg.tx_type;
1526 rx_filter = cfg.rx_filter;
1529 case HWTSTAMP_TX_OFF:
1530 case HWTSTAMP_TX_ON:
1531 case HWTSTAMP_TX_ONESTEP_SYNC:
1536 switch (rx_filter) {
1537 case HWTSTAMP_FILTER_NONE:
1538 case HWTSTAMP_FILTER_ALL:
1539 case HWTSTAMP_FILTER_SOME:
1540 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1542 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1543 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1545 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1546 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1548 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1549 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1550 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1551 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1552 rx_filter_valid = 1;
1556 if (!tx_type_valid || !rx_filter_valid)
1562 static inline bool is_skb_forwardable(struct net_device *dev,
1563 struct sk_buff *skb)
1567 if (!(dev->flags & IFF_UP))
1570 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1571 if (skb->len <= len)
1574 /* if TSO is enabled, we don't care about the length as the packet
1575 * could be forwarded without being segmented before
1577 if (skb_is_gso(skb))
1584 * dev_forward_skb - loopback an skb to another netif
1586 * @dev: destination network device
1587 * @skb: buffer to forward
1590 * NET_RX_SUCCESS (no congestion)
1591 * NET_RX_DROP (packet was dropped, but freed)
1593 * dev_forward_skb can be used for injecting an skb from the
1594 * start_xmit function of one device into the receive queue
1595 * of another device.
1597 * The receiving device may be in another namespace, so
1598 * we have to clear all information in the skb that could
1599 * impact namespace isolation.
1601 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1603 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1604 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1605 atomic_long_inc(&dev->rx_dropped);
1614 if (unlikely(!is_skb_forwardable(dev, skb))) {
1615 atomic_long_inc(&dev->rx_dropped);
1622 skb->tstamp.tv64 = 0;
1623 skb->pkt_type = PACKET_HOST;
1624 skb->protocol = eth_type_trans(skb, dev);
1628 return netif_rx(skb);
1630 EXPORT_SYMBOL_GPL(dev_forward_skb);
1632 static inline int deliver_skb(struct sk_buff *skb,
1633 struct packet_type *pt_prev,
1634 struct net_device *orig_dev)
1636 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1638 atomic_inc(&skb->users);
1639 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1643 * Support routine. Sends outgoing frames to any network
1644 * taps currently in use.
1647 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1649 struct packet_type *ptype;
1650 struct sk_buff *skb2 = NULL;
1651 struct packet_type *pt_prev = NULL;
1654 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1655 /* Never send packets back to the socket
1656 * they originated from - MvS (miquels@drinkel.ow.org)
1658 if ((ptype->dev == dev || !ptype->dev) &&
1659 (ptype->af_packet_priv == NULL ||
1660 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1662 deliver_skb(skb2, pt_prev, skb->dev);
1667 skb2 = skb_clone(skb, GFP_ATOMIC);
1671 net_timestamp_set(skb2);
1673 /* skb->nh should be correctly
1674 set by sender, so that the second statement is
1675 just protection against buggy protocols.
1677 skb_reset_mac_header(skb2);
1679 if (skb_network_header(skb2) < skb2->data ||
1680 skb2->network_header > skb2->tail) {
1681 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1682 ntohs(skb2->protocol),
1684 skb_reset_network_header(skb2);
1687 skb2->transport_header = skb2->network_header;
1688 skb2->pkt_type = PACKET_OUTGOING;
1693 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1698 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1699 * @dev: Network device
1700 * @txq: number of queues available
1702 * If real_num_tx_queues is changed the tc mappings may no longer be
1703 * valid. To resolve this verify the tc mapping remains valid and if
1704 * not NULL the mapping. With no priorities mapping to this
1705 * offset/count pair it will no longer be used. In the worst case TC0
1706 * is invalid nothing can be done so disable priority mappings. If is
1707 * expected that drivers will fix this mapping if they can before
1708 * calling netif_set_real_num_tx_queues.
1710 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1713 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1715 /* If TC0 is invalidated disable TC mapping */
1716 if (tc->offset + tc->count > txq) {
1717 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1722 /* Invalidated prio to tc mappings set to TC0 */
1723 for (i = 1; i < TC_BITMASK + 1; i++) {
1724 int q = netdev_get_prio_tc_map(dev, i);
1726 tc = &dev->tc_to_txq[q];
1727 if (tc->offset + tc->count > txq) {
1728 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1730 netdev_set_prio_tc_map(dev, i, 0);
1736 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1737 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1739 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1743 if (txq < 1 || txq > dev->num_tx_queues)
1746 if (dev->reg_state == NETREG_REGISTERED ||
1747 dev->reg_state == NETREG_UNREGISTERING) {
1750 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1756 netif_setup_tc(dev, txq);
1758 if (txq < dev->real_num_tx_queues)
1759 qdisc_reset_all_tx_gt(dev, txq);
1762 dev->real_num_tx_queues = txq;
1765 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1769 * netif_set_real_num_rx_queues - set actual number of RX queues used
1770 * @dev: Network device
1771 * @rxq: Actual number of RX queues
1773 * This must be called either with the rtnl_lock held or before
1774 * registration of the net device. Returns 0 on success, or a
1775 * negative error code. If called before registration, it always
1778 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1782 if (rxq < 1 || rxq > dev->num_rx_queues)
1785 if (dev->reg_state == NETREG_REGISTERED) {
1788 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1794 dev->real_num_rx_queues = rxq;
1797 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1801 * netif_get_num_default_rss_queues - default number of RSS queues
1803 * This routine should set an upper limit on the number of RSS queues
1804 * used by default by multiqueue devices.
1806 int netif_get_num_default_rss_queues(void)
1808 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
1810 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
1812 static inline void __netif_reschedule(struct Qdisc *q)
1814 struct softnet_data *sd;
1815 unsigned long flags;
1817 local_irq_save(flags);
1818 sd = &__get_cpu_var(softnet_data);
1819 q->next_sched = NULL;
1820 *sd->output_queue_tailp = q;
1821 sd->output_queue_tailp = &q->next_sched;
1822 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1823 local_irq_restore(flags);
1826 void __netif_schedule(struct Qdisc *q)
1828 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1829 __netif_reschedule(q);
1831 EXPORT_SYMBOL(__netif_schedule);
1833 void dev_kfree_skb_irq(struct sk_buff *skb)
1835 if (atomic_dec_and_test(&skb->users)) {
1836 struct softnet_data *sd;
1837 unsigned long flags;
1839 local_irq_save(flags);
1840 sd = &__get_cpu_var(softnet_data);
1841 skb->next = sd->completion_queue;
1842 sd->completion_queue = skb;
1843 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1844 local_irq_restore(flags);
1847 EXPORT_SYMBOL(dev_kfree_skb_irq);
1849 void dev_kfree_skb_any(struct sk_buff *skb)
1851 if (in_irq() || irqs_disabled())
1852 dev_kfree_skb_irq(skb);
1856 EXPORT_SYMBOL(dev_kfree_skb_any);
1860 * netif_device_detach - mark device as removed
1861 * @dev: network device
1863 * Mark device as removed from system and therefore no longer available.
1865 void netif_device_detach(struct net_device *dev)
1867 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1868 netif_running(dev)) {
1869 netif_tx_stop_all_queues(dev);
1872 EXPORT_SYMBOL(netif_device_detach);
1875 * netif_device_attach - mark device as attached
1876 * @dev: network device
1878 * Mark device as attached from system and restart if needed.
1880 void netif_device_attach(struct net_device *dev)
1882 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1883 netif_running(dev)) {
1884 netif_tx_wake_all_queues(dev);
1885 __netdev_watchdog_up(dev);
1888 EXPORT_SYMBOL(netif_device_attach);
1890 static void skb_warn_bad_offload(const struct sk_buff *skb)
1892 static const netdev_features_t null_features = 0;
1893 struct net_device *dev = skb->dev;
1894 const char *driver = "";
1896 if (dev && dev->dev.parent)
1897 driver = dev_driver_string(dev->dev.parent);
1899 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1900 "gso_type=%d ip_summed=%d\n",
1901 driver, dev ? &dev->features : &null_features,
1902 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1903 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1904 skb_shinfo(skb)->gso_type, skb->ip_summed);
1908 * Invalidate hardware checksum when packet is to be mangled, and
1909 * complete checksum manually on outgoing path.
1911 int skb_checksum_help(struct sk_buff *skb)
1914 int ret = 0, offset;
1916 if (skb->ip_summed == CHECKSUM_COMPLETE)
1917 goto out_set_summed;
1919 if (unlikely(skb_shinfo(skb)->gso_size)) {
1920 skb_warn_bad_offload(skb);
1924 offset = skb_checksum_start_offset(skb);
1925 BUG_ON(offset >= skb_headlen(skb));
1926 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1928 offset += skb->csum_offset;
1929 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1931 if (skb_cloned(skb) &&
1932 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1933 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1938 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1940 skb->ip_summed = CHECKSUM_NONE;
1944 EXPORT_SYMBOL(skb_checksum_help);
1947 * skb_gso_segment - Perform segmentation on skb.
1948 * @skb: buffer to segment
1949 * @features: features for the output path (see dev->features)
1951 * This function segments the given skb and returns a list of segments.
1953 * It may return NULL if the skb requires no segmentation. This is
1954 * only possible when GSO is used for verifying header integrity.
1956 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1957 netdev_features_t features)
1959 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1960 struct packet_type *ptype;
1961 __be16 type = skb->protocol;
1962 int vlan_depth = ETH_HLEN;
1965 while (type == htons(ETH_P_8021Q)) {
1966 struct vlan_hdr *vh;
1968 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1969 return ERR_PTR(-EINVAL);
1971 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1972 type = vh->h_vlan_encapsulated_proto;
1973 vlan_depth += VLAN_HLEN;
1976 skb_reset_mac_header(skb);
1977 skb->mac_len = skb->network_header - skb->mac_header;
1978 __skb_pull(skb, skb->mac_len);
1980 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1981 skb_warn_bad_offload(skb);
1983 if (skb_header_cloned(skb) &&
1984 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1985 return ERR_PTR(err);
1989 list_for_each_entry_rcu(ptype,
1990 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1991 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1992 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1993 err = ptype->gso_send_check(skb);
1994 segs = ERR_PTR(err);
1995 if (err || skb_gso_ok(skb, features))
1997 __skb_push(skb, (skb->data -
1998 skb_network_header(skb)));
2000 segs = ptype->gso_segment(skb, features);
2006 __skb_push(skb, skb->data - skb_mac_header(skb));
2010 EXPORT_SYMBOL(skb_gso_segment);
2012 /* Take action when hardware reception checksum errors are detected. */
2014 void netdev_rx_csum_fault(struct net_device *dev)
2016 if (net_ratelimit()) {
2017 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2021 EXPORT_SYMBOL(netdev_rx_csum_fault);
2024 /* Actually, we should eliminate this check as soon as we know, that:
2025 * 1. IOMMU is present and allows to map all the memory.
2026 * 2. No high memory really exists on this machine.
2029 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2031 #ifdef CONFIG_HIGHMEM
2033 if (!(dev->features & NETIF_F_HIGHDMA)) {
2034 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2035 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2036 if (PageHighMem(skb_frag_page(frag)))
2041 if (PCI_DMA_BUS_IS_PHYS) {
2042 struct device *pdev = dev->dev.parent;
2046 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2047 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2048 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2049 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2058 void (*destructor)(struct sk_buff *skb);
2061 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2063 static void dev_gso_skb_destructor(struct sk_buff *skb)
2065 struct dev_gso_cb *cb;
2068 struct sk_buff *nskb = skb->next;
2070 skb->next = nskb->next;
2073 } while (skb->next);
2075 cb = DEV_GSO_CB(skb);
2077 cb->destructor(skb);
2081 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2082 * @skb: buffer to segment
2083 * @features: device features as applicable to this skb
2085 * This function segments the given skb and stores the list of segments
2088 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2090 struct sk_buff *segs;
2092 segs = skb_gso_segment(skb, features);
2094 /* Verifying header integrity only. */
2099 return PTR_ERR(segs);
2102 DEV_GSO_CB(skb)->destructor = skb->destructor;
2103 skb->destructor = dev_gso_skb_destructor;
2108 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2110 return ((features & NETIF_F_GEN_CSUM) ||
2111 ((features & NETIF_F_V4_CSUM) &&
2112 protocol == htons(ETH_P_IP)) ||
2113 ((features & NETIF_F_V6_CSUM) &&
2114 protocol == htons(ETH_P_IPV6)) ||
2115 ((features & NETIF_F_FCOE_CRC) &&
2116 protocol == htons(ETH_P_FCOE)));
2119 static netdev_features_t harmonize_features(struct sk_buff *skb,
2120 __be16 protocol, netdev_features_t features)
2122 if (!can_checksum_protocol(features, protocol)) {
2123 features &= ~NETIF_F_ALL_CSUM;
2124 features &= ~NETIF_F_SG;
2125 } else if (illegal_highdma(skb->dev, skb)) {
2126 features &= ~NETIF_F_SG;
2132 netdev_features_t netif_skb_features(struct sk_buff *skb)
2134 __be16 protocol = skb->protocol;
2135 netdev_features_t features = skb->dev->features;
2137 if (protocol == htons(ETH_P_8021Q)) {
2138 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2139 protocol = veh->h_vlan_encapsulated_proto;
2140 } else if (!vlan_tx_tag_present(skb)) {
2141 return harmonize_features(skb, protocol, features);
2144 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2146 if (protocol != htons(ETH_P_8021Q)) {
2147 return harmonize_features(skb, protocol, features);
2149 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2150 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2151 return harmonize_features(skb, protocol, features);
2154 EXPORT_SYMBOL(netif_skb_features);
2157 * Returns true if either:
2158 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2159 * 2. skb is fragmented and the device does not support SG, or if
2160 * at least one of fragments is in highmem and device does not
2161 * support DMA from it.
2163 static inline int skb_needs_linearize(struct sk_buff *skb,
2166 return skb_is_nonlinear(skb) &&
2167 ((skb_has_frag_list(skb) &&
2168 !(features & NETIF_F_FRAGLIST)) ||
2169 (skb_shinfo(skb)->nr_frags &&
2170 !(features & NETIF_F_SG)));
2173 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2174 struct netdev_queue *txq)
2176 const struct net_device_ops *ops = dev->netdev_ops;
2177 int rc = NETDEV_TX_OK;
2178 unsigned int skb_len;
2180 if (likely(!skb->next)) {
2181 netdev_features_t features;
2184 * If device doesn't need skb->dst, release it right now while
2185 * its hot in this cpu cache
2187 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2190 if (!list_empty(&ptype_all))
2191 dev_queue_xmit_nit(skb, dev);
2193 features = netif_skb_features(skb);
2195 if (vlan_tx_tag_present(skb) &&
2196 !(features & NETIF_F_HW_VLAN_TX)) {
2197 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2204 if (netif_needs_gso(skb, features)) {
2205 if (unlikely(dev_gso_segment(skb, features)))
2210 if (skb_needs_linearize(skb, features) &&
2211 __skb_linearize(skb))
2214 /* If packet is not checksummed and device does not
2215 * support checksumming for this protocol, complete
2216 * checksumming here.
2218 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2219 skb_set_transport_header(skb,
2220 skb_checksum_start_offset(skb));
2221 if (!(features & NETIF_F_ALL_CSUM) &&
2222 skb_checksum_help(skb))
2228 rc = ops->ndo_start_xmit(skb, dev);
2229 trace_net_dev_xmit(skb, rc, dev, skb_len);
2230 if (rc == NETDEV_TX_OK)
2231 txq_trans_update(txq);
2237 struct sk_buff *nskb = skb->next;
2239 skb->next = nskb->next;
2243 * If device doesn't need nskb->dst, release it right now while
2244 * its hot in this cpu cache
2246 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2249 skb_len = nskb->len;
2250 rc = ops->ndo_start_xmit(nskb, dev);
2251 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2252 if (unlikely(rc != NETDEV_TX_OK)) {
2253 if (rc & ~NETDEV_TX_MASK)
2254 goto out_kfree_gso_skb;
2255 nskb->next = skb->next;
2259 txq_trans_update(txq);
2260 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2261 return NETDEV_TX_BUSY;
2262 } while (skb->next);
2265 if (likely(skb->next == NULL))
2266 skb->destructor = DEV_GSO_CB(skb)->destructor;
2273 static u32 hashrnd __read_mostly;
2276 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2277 * to be used as a distribution range.
2279 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2280 unsigned int num_tx_queues)
2284 u16 qcount = num_tx_queues;
2286 if (skb_rx_queue_recorded(skb)) {
2287 hash = skb_get_rx_queue(skb);
2288 while (unlikely(hash >= num_tx_queues))
2289 hash -= num_tx_queues;
2294 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2295 qoffset = dev->tc_to_txq[tc].offset;
2296 qcount = dev->tc_to_txq[tc].count;
2299 if (skb->sk && skb->sk->sk_hash)
2300 hash = skb->sk->sk_hash;
2302 hash = (__force u16) skb->protocol;
2303 hash = jhash_1word(hash, hashrnd);
2305 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2307 EXPORT_SYMBOL(__skb_tx_hash);
2309 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2311 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2312 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2313 dev->name, queue_index,
2314 dev->real_num_tx_queues);
2320 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2323 struct xps_dev_maps *dev_maps;
2324 struct xps_map *map;
2325 int queue_index = -1;
2328 dev_maps = rcu_dereference(dev->xps_maps);
2330 map = rcu_dereference(
2331 dev_maps->cpu_map[raw_smp_processor_id()]);
2334 queue_index = map->queues[0];
2337 if (skb->sk && skb->sk->sk_hash)
2338 hash = skb->sk->sk_hash;
2340 hash = (__force u16) skb->protocol ^
2342 hash = jhash_1word(hash, hashrnd);
2343 queue_index = map->queues[
2344 ((u64)hash * map->len) >> 32];
2346 if (unlikely(queue_index >= dev->real_num_tx_queues))
2358 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2359 struct sk_buff *skb)
2362 const struct net_device_ops *ops = dev->netdev_ops;
2364 if (dev->real_num_tx_queues == 1)
2366 else if (ops->ndo_select_queue) {
2367 queue_index = ops->ndo_select_queue(dev, skb);
2368 queue_index = dev_cap_txqueue(dev, queue_index);
2370 struct sock *sk = skb->sk;
2371 queue_index = sk_tx_queue_get(sk);
2373 if (queue_index < 0 || skb->ooo_okay ||
2374 queue_index >= dev->real_num_tx_queues) {
2375 int old_index = queue_index;
2377 queue_index = get_xps_queue(dev, skb);
2378 if (queue_index < 0)
2379 queue_index = skb_tx_hash(dev, skb);
2381 if (queue_index != old_index && sk) {
2382 struct dst_entry *dst =
2383 rcu_dereference_check(sk->sk_dst_cache, 1);
2385 if (dst && skb_dst(skb) == dst)
2386 sk_tx_queue_set(sk, queue_index);
2391 skb_set_queue_mapping(skb, queue_index);
2392 return netdev_get_tx_queue(dev, queue_index);
2395 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2396 struct net_device *dev,
2397 struct netdev_queue *txq)
2399 spinlock_t *root_lock = qdisc_lock(q);
2403 qdisc_skb_cb(skb)->pkt_len = skb->len;
2404 qdisc_calculate_pkt_len(skb, q);
2406 * Heuristic to force contended enqueues to serialize on a
2407 * separate lock before trying to get qdisc main lock.
2408 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2409 * and dequeue packets faster.
2411 contended = qdisc_is_running(q);
2412 if (unlikely(contended))
2413 spin_lock(&q->busylock);
2415 spin_lock(root_lock);
2416 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2419 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2420 qdisc_run_begin(q)) {
2422 * This is a work-conserving queue; there are no old skbs
2423 * waiting to be sent out; and the qdisc is not running -
2424 * xmit the skb directly.
2426 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2429 qdisc_bstats_update(q, skb);
2431 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2432 if (unlikely(contended)) {
2433 spin_unlock(&q->busylock);
2440 rc = NET_XMIT_SUCCESS;
2443 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2444 if (qdisc_run_begin(q)) {
2445 if (unlikely(contended)) {
2446 spin_unlock(&q->busylock);
2452 spin_unlock(root_lock);
2453 if (unlikely(contended))
2454 spin_unlock(&q->busylock);
2458 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2459 static void skb_update_prio(struct sk_buff *skb)
2461 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2463 if (!skb->priority && skb->sk && map) {
2464 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2466 if (prioidx < map->priomap_len)
2467 skb->priority = map->priomap[prioidx];
2471 #define skb_update_prio(skb)
2474 static DEFINE_PER_CPU(int, xmit_recursion);
2475 #define RECURSION_LIMIT 10
2478 * dev_loopback_xmit - loop back @skb
2479 * @skb: buffer to transmit
2481 int dev_loopback_xmit(struct sk_buff *skb)
2483 skb_reset_mac_header(skb);
2484 __skb_pull(skb, skb_network_offset(skb));
2485 skb->pkt_type = PACKET_LOOPBACK;
2486 skb->ip_summed = CHECKSUM_UNNECESSARY;
2487 WARN_ON(!skb_dst(skb));
2492 EXPORT_SYMBOL(dev_loopback_xmit);
2495 * dev_queue_xmit - transmit a buffer
2496 * @skb: buffer to transmit
2498 * Queue a buffer for transmission to a network device. The caller must
2499 * have set the device and priority and built the buffer before calling
2500 * this function. The function can be called from an interrupt.
2502 * A negative errno code is returned on a failure. A success does not
2503 * guarantee the frame will be transmitted as it may be dropped due
2504 * to congestion or traffic shaping.
2506 * -----------------------------------------------------------------------------------
2507 * I notice this method can also return errors from the queue disciplines,
2508 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2511 * Regardless of the return value, the skb is consumed, so it is currently
2512 * difficult to retry a send to this method. (You can bump the ref count
2513 * before sending to hold a reference for retry if you are careful.)
2515 * When calling this method, interrupts MUST be enabled. This is because
2516 * the BH enable code must have IRQs enabled so that it will not deadlock.
2519 int dev_queue_xmit(struct sk_buff *skb)
2521 struct net_device *dev = skb->dev;
2522 struct netdev_queue *txq;
2526 /* Disable soft irqs for various locks below. Also
2527 * stops preemption for RCU.
2531 skb_update_prio(skb);
2533 txq = dev_pick_tx(dev, skb);
2534 q = rcu_dereference_bh(txq->qdisc);
2536 #ifdef CONFIG_NET_CLS_ACT
2537 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2539 trace_net_dev_queue(skb);
2541 rc = __dev_xmit_skb(skb, q, dev, txq);
2545 /* The device has no queue. Common case for software devices:
2546 loopback, all the sorts of tunnels...
2548 Really, it is unlikely that netif_tx_lock protection is necessary
2549 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2551 However, it is possible, that they rely on protection
2554 Check this and shot the lock. It is not prone from deadlocks.
2555 Either shot noqueue qdisc, it is even simpler 8)
2557 if (dev->flags & IFF_UP) {
2558 int cpu = smp_processor_id(); /* ok because BHs are off */
2560 if (txq->xmit_lock_owner != cpu) {
2562 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2563 goto recursion_alert;
2565 HARD_TX_LOCK(dev, txq, cpu);
2567 if (!netif_xmit_stopped(txq)) {
2568 __this_cpu_inc(xmit_recursion);
2569 rc = dev_hard_start_xmit(skb, dev, txq);
2570 __this_cpu_dec(xmit_recursion);
2571 if (dev_xmit_complete(rc)) {
2572 HARD_TX_UNLOCK(dev, txq);
2576 HARD_TX_UNLOCK(dev, txq);
2577 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2580 /* Recursion is detected! It is possible,
2584 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2590 rcu_read_unlock_bh();
2595 rcu_read_unlock_bh();
2598 EXPORT_SYMBOL(dev_queue_xmit);
2601 /*=======================================================================
2603 =======================================================================*/
2605 int netdev_max_backlog __read_mostly = 1000;
2606 int netdev_tstamp_prequeue __read_mostly = 1;
2607 int netdev_budget __read_mostly = 300;
2608 int weight_p __read_mostly = 64; /* old backlog weight */
2610 /* Called with irq disabled */
2611 static inline void ____napi_schedule(struct softnet_data *sd,
2612 struct napi_struct *napi)
2614 list_add_tail(&napi->poll_list, &sd->poll_list);
2615 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2619 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2620 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2621 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2622 * if hash is a canonical 4-tuple hash over transport ports.
2624 void __skb_get_rxhash(struct sk_buff *skb)
2626 struct flow_keys keys;
2629 if (!skb_flow_dissect(skb, &keys))
2633 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2634 swap(keys.port16[0], keys.port16[1]);
2638 /* get a consistent hash (same value on both flow directions) */
2639 if ((__force u32)keys.dst < (__force u32)keys.src)
2640 swap(keys.dst, keys.src);
2642 hash = jhash_3words((__force u32)keys.dst,
2643 (__force u32)keys.src,
2644 (__force u32)keys.ports, hashrnd);
2650 EXPORT_SYMBOL(__skb_get_rxhash);
2654 /* One global table that all flow-based protocols share. */
2655 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2656 EXPORT_SYMBOL(rps_sock_flow_table);
2658 struct static_key rps_needed __read_mostly;
2660 static struct rps_dev_flow *
2661 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2662 struct rps_dev_flow *rflow, u16 next_cpu)
2664 if (next_cpu != RPS_NO_CPU) {
2665 #ifdef CONFIG_RFS_ACCEL
2666 struct netdev_rx_queue *rxqueue;
2667 struct rps_dev_flow_table *flow_table;
2668 struct rps_dev_flow *old_rflow;
2673 /* Should we steer this flow to a different hardware queue? */
2674 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2675 !(dev->features & NETIF_F_NTUPLE))
2677 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2678 if (rxq_index == skb_get_rx_queue(skb))
2681 rxqueue = dev->_rx + rxq_index;
2682 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2685 flow_id = skb->rxhash & flow_table->mask;
2686 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2687 rxq_index, flow_id);
2691 rflow = &flow_table->flows[flow_id];
2693 if (old_rflow->filter == rflow->filter)
2694 old_rflow->filter = RPS_NO_FILTER;
2698 per_cpu(softnet_data, next_cpu).input_queue_head;
2701 rflow->cpu = next_cpu;
2706 * get_rps_cpu is called from netif_receive_skb and returns the target
2707 * CPU from the RPS map of the receiving queue for a given skb.
2708 * rcu_read_lock must be held on entry.
2710 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2711 struct rps_dev_flow **rflowp)
2713 struct netdev_rx_queue *rxqueue;
2714 struct rps_map *map;
2715 struct rps_dev_flow_table *flow_table;
2716 struct rps_sock_flow_table *sock_flow_table;
2720 if (skb_rx_queue_recorded(skb)) {
2721 u16 index = skb_get_rx_queue(skb);
2722 if (unlikely(index >= dev->real_num_rx_queues)) {
2723 WARN_ONCE(dev->real_num_rx_queues > 1,
2724 "%s received packet on queue %u, but number "
2725 "of RX queues is %u\n",
2726 dev->name, index, dev->real_num_rx_queues);
2729 rxqueue = dev->_rx + index;
2733 map = rcu_dereference(rxqueue->rps_map);
2735 if (map->len == 1 &&
2736 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2737 tcpu = map->cpus[0];
2738 if (cpu_online(tcpu))
2742 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2746 skb_reset_network_header(skb);
2747 if (!skb_get_rxhash(skb))
2750 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2751 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2752 if (flow_table && sock_flow_table) {
2754 struct rps_dev_flow *rflow;
2756 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2759 next_cpu = sock_flow_table->ents[skb->rxhash &
2760 sock_flow_table->mask];
2763 * If the desired CPU (where last recvmsg was done) is
2764 * different from current CPU (one in the rx-queue flow
2765 * table entry), switch if one of the following holds:
2766 * - Current CPU is unset (equal to RPS_NO_CPU).
2767 * - Current CPU is offline.
2768 * - The current CPU's queue tail has advanced beyond the
2769 * last packet that was enqueued using this table entry.
2770 * This guarantees that all previous packets for the flow
2771 * have been dequeued, thus preserving in order delivery.
2773 if (unlikely(tcpu != next_cpu) &&
2774 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2775 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2776 rflow->last_qtail)) >= 0))
2777 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2779 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2787 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2789 if (cpu_online(tcpu)) {
2799 #ifdef CONFIG_RFS_ACCEL
2802 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2803 * @dev: Device on which the filter was set
2804 * @rxq_index: RX queue index
2805 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2806 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2808 * Drivers that implement ndo_rx_flow_steer() should periodically call
2809 * this function for each installed filter and remove the filters for
2810 * which it returns %true.
2812 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2813 u32 flow_id, u16 filter_id)
2815 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2816 struct rps_dev_flow_table *flow_table;
2817 struct rps_dev_flow *rflow;
2822 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2823 if (flow_table && flow_id <= flow_table->mask) {
2824 rflow = &flow_table->flows[flow_id];
2825 cpu = ACCESS_ONCE(rflow->cpu);
2826 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2827 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2828 rflow->last_qtail) <
2829 (int)(10 * flow_table->mask)))
2835 EXPORT_SYMBOL(rps_may_expire_flow);
2837 #endif /* CONFIG_RFS_ACCEL */
2839 /* Called from hardirq (IPI) context */
2840 static void rps_trigger_softirq(void *data)
2842 struct softnet_data *sd = data;
2844 ____napi_schedule(sd, &sd->backlog);
2848 #endif /* CONFIG_RPS */
2851 * Check if this softnet_data structure is another cpu one
2852 * If yes, queue it to our IPI list and return 1
2855 static int rps_ipi_queued(struct softnet_data *sd)
2858 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2861 sd->rps_ipi_next = mysd->rps_ipi_list;
2862 mysd->rps_ipi_list = sd;
2864 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2867 #endif /* CONFIG_RPS */
2872 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2873 * queue (may be a remote CPU queue).
2875 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2876 unsigned int *qtail)
2878 struct softnet_data *sd;
2879 unsigned long flags;
2881 sd = &per_cpu(softnet_data, cpu);
2883 local_irq_save(flags);
2886 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2887 if (skb_queue_len(&sd->input_pkt_queue)) {
2889 __skb_queue_tail(&sd->input_pkt_queue, skb);
2890 input_queue_tail_incr_save(sd, qtail);
2892 local_irq_restore(flags);
2893 return NET_RX_SUCCESS;
2896 /* Schedule NAPI for backlog device
2897 * We can use non atomic operation since we own the queue lock
2899 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2900 if (!rps_ipi_queued(sd))
2901 ____napi_schedule(sd, &sd->backlog);
2909 local_irq_restore(flags);
2911 atomic_long_inc(&skb->dev->rx_dropped);
2917 * netif_rx - post buffer to the network code
2918 * @skb: buffer to post
2920 * This function receives a packet from a device driver and queues it for
2921 * the upper (protocol) levels to process. It always succeeds. The buffer
2922 * may be dropped during processing for congestion control or by the
2926 * NET_RX_SUCCESS (no congestion)
2927 * NET_RX_DROP (packet was dropped)
2931 int netif_rx(struct sk_buff *skb)
2935 /* if netpoll wants it, pretend we never saw it */
2936 if (netpoll_rx(skb))
2939 net_timestamp_check(netdev_tstamp_prequeue, skb);
2941 trace_netif_rx(skb);
2943 if (static_key_false(&rps_needed)) {
2944 struct rps_dev_flow voidflow, *rflow = &voidflow;
2950 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2952 cpu = smp_processor_id();
2954 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2962 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2967 EXPORT_SYMBOL(netif_rx);
2969 int netif_rx_ni(struct sk_buff *skb)
2974 err = netif_rx(skb);
2975 if (local_softirq_pending())
2981 EXPORT_SYMBOL(netif_rx_ni);
2983 static void net_tx_action(struct softirq_action *h)
2985 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2987 if (sd->completion_queue) {
2988 struct sk_buff *clist;
2990 local_irq_disable();
2991 clist = sd->completion_queue;
2992 sd->completion_queue = NULL;
2996 struct sk_buff *skb = clist;
2997 clist = clist->next;
2999 WARN_ON(atomic_read(&skb->users));
3000 trace_kfree_skb(skb, net_tx_action);
3005 if (sd->output_queue) {
3008 local_irq_disable();
3009 head = sd->output_queue;
3010 sd->output_queue = NULL;
3011 sd->output_queue_tailp = &sd->output_queue;
3015 struct Qdisc *q = head;
3016 spinlock_t *root_lock;
3018 head = head->next_sched;
3020 root_lock = qdisc_lock(q);
3021 if (spin_trylock(root_lock)) {
3022 smp_mb__before_clear_bit();
3023 clear_bit(__QDISC_STATE_SCHED,
3026 spin_unlock(root_lock);
3028 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3030 __netif_reschedule(q);
3032 smp_mb__before_clear_bit();
3033 clear_bit(__QDISC_STATE_SCHED,
3041 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3042 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3043 /* This hook is defined here for ATM LANE */
3044 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3045 unsigned char *addr) __read_mostly;
3046 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3049 #ifdef CONFIG_NET_CLS_ACT
3050 /* TODO: Maybe we should just force sch_ingress to be compiled in
3051 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3052 * a compare and 2 stores extra right now if we dont have it on
3053 * but have CONFIG_NET_CLS_ACT
3054 * NOTE: This doesn't stop any functionality; if you dont have
3055 * the ingress scheduler, you just can't add policies on ingress.
3058 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3060 struct net_device *dev = skb->dev;
3061 u32 ttl = G_TC_RTTL(skb->tc_verd);
3062 int result = TC_ACT_OK;
3065 if (unlikely(MAX_RED_LOOP < ttl++)) {
3066 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3067 skb->skb_iif, dev->ifindex);
3071 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3072 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3075 if (q != &noop_qdisc) {
3076 spin_lock(qdisc_lock(q));
3077 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3078 result = qdisc_enqueue_root(skb, q);
3079 spin_unlock(qdisc_lock(q));
3085 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3086 struct packet_type **pt_prev,
3087 int *ret, struct net_device *orig_dev)
3089 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3091 if (!rxq || rxq->qdisc == &noop_qdisc)
3095 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3099 switch (ing_filter(skb, rxq)) {
3113 * netdev_rx_handler_register - register receive handler
3114 * @dev: device to register a handler for
3115 * @rx_handler: receive handler to register
3116 * @rx_handler_data: data pointer that is used by rx handler
3118 * Register a receive hander for a device. This handler will then be
3119 * called from __netif_receive_skb. A negative errno code is returned
3122 * The caller must hold the rtnl_mutex.
3124 * For a general description of rx_handler, see enum rx_handler_result.
3126 int netdev_rx_handler_register(struct net_device *dev,
3127 rx_handler_func_t *rx_handler,
3128 void *rx_handler_data)
3132 if (dev->rx_handler)
3135 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3136 rcu_assign_pointer(dev->rx_handler, rx_handler);
3140 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3143 * netdev_rx_handler_unregister - unregister receive handler
3144 * @dev: device to unregister a handler from
3146 * Unregister a receive hander from a device.
3148 * The caller must hold the rtnl_mutex.
3150 void netdev_rx_handler_unregister(struct net_device *dev)
3154 RCU_INIT_POINTER(dev->rx_handler, NULL);
3155 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3157 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3160 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3161 * the special handling of PFMEMALLOC skbs.
3163 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3165 switch (skb->protocol) {
3166 case __constant_htons(ETH_P_ARP):
3167 case __constant_htons(ETH_P_IP):
3168 case __constant_htons(ETH_P_IPV6):
3169 case __constant_htons(ETH_P_8021Q):
3176 static int __netif_receive_skb(struct sk_buff *skb)
3178 struct packet_type *ptype, *pt_prev;
3179 rx_handler_func_t *rx_handler;
3180 struct net_device *orig_dev;
3181 struct net_device *null_or_dev;
3182 bool deliver_exact = false;
3183 int ret = NET_RX_DROP;
3185 unsigned long pflags = current->flags;
3187 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3189 trace_netif_receive_skb(skb);
3192 * PFMEMALLOC skbs are special, they should
3193 * - be delivered to SOCK_MEMALLOC sockets only
3194 * - stay away from userspace
3195 * - have bounded memory usage
3197 * Use PF_MEMALLOC as this saves us from propagating the allocation
3198 * context down to all allocation sites.
3200 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3201 current->flags |= PF_MEMALLOC;
3203 /* if we've gotten here through NAPI, check netpoll */
3204 if (netpoll_receive_skb(skb))
3207 orig_dev = skb->dev;
3209 skb_reset_network_header(skb);
3210 skb_reset_transport_header(skb);
3211 skb_reset_mac_len(skb);
3218 skb->skb_iif = skb->dev->ifindex;
3220 __this_cpu_inc(softnet_data.processed);
3222 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3223 skb = vlan_untag(skb);
3228 #ifdef CONFIG_NET_CLS_ACT
3229 if (skb->tc_verd & TC_NCLS) {
3230 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3235 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3238 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3239 if (!ptype->dev || ptype->dev == skb->dev) {
3241 ret = deliver_skb(skb, pt_prev, orig_dev);
3247 #ifdef CONFIG_NET_CLS_ACT
3248 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3254 if (sk_memalloc_socks() && skb_pfmemalloc(skb)
3255 && !skb_pfmemalloc_protocol(skb))
3258 rx_handler = rcu_dereference(skb->dev->rx_handler);
3259 if (vlan_tx_tag_present(skb)) {
3261 ret = deliver_skb(skb, pt_prev, orig_dev);
3264 if (vlan_do_receive(&skb, !rx_handler))
3266 else if (unlikely(!skb))
3272 ret = deliver_skb(skb, pt_prev, orig_dev);
3275 switch (rx_handler(&skb)) {
3276 case RX_HANDLER_CONSUMED:
3278 case RX_HANDLER_ANOTHER:
3280 case RX_HANDLER_EXACT:
3281 deliver_exact = true;
3282 case RX_HANDLER_PASS:
3289 /* deliver only exact match when indicated */
3290 null_or_dev = deliver_exact ? skb->dev : NULL;
3292 type = skb->protocol;
3293 list_for_each_entry_rcu(ptype,
3294 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3295 if (ptype->type == type &&
3296 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3297 ptype->dev == orig_dev)) {
3299 ret = deliver_skb(skb, pt_prev, orig_dev);
3305 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3308 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3311 atomic_long_inc(&skb->dev->rx_dropped);
3313 /* Jamal, now you will not able to escape explaining
3314 * me how you were going to use this. :-)
3322 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3327 * netif_receive_skb - process receive buffer from network
3328 * @skb: buffer to process
3330 * netif_receive_skb() is the main receive data processing function.
3331 * It always succeeds. The buffer may be dropped during processing
3332 * for congestion control or by the protocol layers.
3334 * This function may only be called from softirq context and interrupts
3335 * should be enabled.
3337 * Return values (usually ignored):
3338 * NET_RX_SUCCESS: no congestion
3339 * NET_RX_DROP: packet was dropped
3341 int netif_receive_skb(struct sk_buff *skb)
3343 net_timestamp_check(netdev_tstamp_prequeue, skb);
3345 if (skb_defer_rx_timestamp(skb))
3346 return NET_RX_SUCCESS;
3349 if (static_key_false(&rps_needed)) {
3350 struct rps_dev_flow voidflow, *rflow = &voidflow;
3355 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3358 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3365 return __netif_receive_skb(skb);
3367 EXPORT_SYMBOL(netif_receive_skb);
3369 /* Network device is going away, flush any packets still pending
3370 * Called with irqs disabled.
3372 static void flush_backlog(void *arg)
3374 struct net_device *dev = arg;
3375 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3376 struct sk_buff *skb, *tmp;
3379 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3380 if (skb->dev == dev) {
3381 __skb_unlink(skb, &sd->input_pkt_queue);
3383 input_queue_head_incr(sd);
3388 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3389 if (skb->dev == dev) {
3390 __skb_unlink(skb, &sd->process_queue);
3392 input_queue_head_incr(sd);
3397 static int napi_gro_complete(struct sk_buff *skb)
3399 struct packet_type *ptype;
3400 __be16 type = skb->protocol;
3401 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3404 if (NAPI_GRO_CB(skb)->count == 1) {
3405 skb_shinfo(skb)->gso_size = 0;
3410 list_for_each_entry_rcu(ptype, head, list) {
3411 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3414 err = ptype->gro_complete(skb);
3420 WARN_ON(&ptype->list == head);
3422 return NET_RX_SUCCESS;
3426 return netif_receive_skb(skb);
3429 inline void napi_gro_flush(struct napi_struct *napi)
3431 struct sk_buff *skb, *next;
3433 for (skb = napi->gro_list; skb; skb = next) {
3436 napi_gro_complete(skb);
3439 napi->gro_count = 0;
3440 napi->gro_list = NULL;
3442 EXPORT_SYMBOL(napi_gro_flush);
3444 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3446 struct sk_buff **pp = NULL;
3447 struct packet_type *ptype;
3448 __be16 type = skb->protocol;
3449 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3452 enum gro_result ret;
3454 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3457 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3461 list_for_each_entry_rcu(ptype, head, list) {
3462 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3465 skb_set_network_header(skb, skb_gro_offset(skb));
3466 mac_len = skb->network_header - skb->mac_header;
3467 skb->mac_len = mac_len;
3468 NAPI_GRO_CB(skb)->same_flow = 0;
3469 NAPI_GRO_CB(skb)->flush = 0;
3470 NAPI_GRO_CB(skb)->free = 0;
3472 pp = ptype->gro_receive(&napi->gro_list, skb);
3477 if (&ptype->list == head)
3480 same_flow = NAPI_GRO_CB(skb)->same_flow;
3481 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3484 struct sk_buff *nskb = *pp;
3488 napi_gro_complete(nskb);
3495 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3499 NAPI_GRO_CB(skb)->count = 1;
3500 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3501 skb->next = napi->gro_list;
3502 napi->gro_list = skb;
3506 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3507 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3509 BUG_ON(skb->end - skb->tail < grow);
3511 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3514 skb->data_len -= grow;
3516 skb_shinfo(skb)->frags[0].page_offset += grow;
3517 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3519 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3520 skb_frag_unref(skb, 0);
3521 memmove(skb_shinfo(skb)->frags,
3522 skb_shinfo(skb)->frags + 1,
3523 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3534 EXPORT_SYMBOL(dev_gro_receive);
3536 static inline gro_result_t
3537 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3540 unsigned int maclen = skb->dev->hard_header_len;
3542 for (p = napi->gro_list; p; p = p->next) {
3543 unsigned long diffs;
3545 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3546 diffs |= p->vlan_tci ^ skb->vlan_tci;
3547 if (maclen == ETH_HLEN)
3548 diffs |= compare_ether_header(skb_mac_header(p),
3549 skb_gro_mac_header(skb));
3551 diffs = memcmp(skb_mac_header(p),
3552 skb_gro_mac_header(skb),
3554 NAPI_GRO_CB(p)->same_flow = !diffs;
3555 NAPI_GRO_CB(p)->flush = 0;
3558 return dev_gro_receive(napi, skb);
3561 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3565 if (netif_receive_skb(skb))
3573 case GRO_MERGED_FREE:
3574 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3575 kmem_cache_free(skbuff_head_cache, skb);
3587 EXPORT_SYMBOL(napi_skb_finish);
3589 void skb_gro_reset_offset(struct sk_buff *skb)
3591 NAPI_GRO_CB(skb)->data_offset = 0;
3592 NAPI_GRO_CB(skb)->frag0 = NULL;
3593 NAPI_GRO_CB(skb)->frag0_len = 0;
3595 if (skb->mac_header == skb->tail &&
3596 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3597 NAPI_GRO_CB(skb)->frag0 =
3598 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3599 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3602 EXPORT_SYMBOL(skb_gro_reset_offset);
3604 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3606 skb_gro_reset_offset(skb);
3608 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3610 EXPORT_SYMBOL(napi_gro_receive);
3612 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3614 __skb_pull(skb, skb_headlen(skb));
3615 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3616 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3618 skb->dev = napi->dev;
3624 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3626 struct sk_buff *skb = napi->skb;
3629 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3635 EXPORT_SYMBOL(napi_get_frags);
3637 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3643 skb->protocol = eth_type_trans(skb, skb->dev);
3645 if (ret == GRO_HELD)
3646 skb_gro_pull(skb, -ETH_HLEN);
3647 else if (netif_receive_skb(skb))
3652 case GRO_MERGED_FREE:
3653 napi_reuse_skb(napi, skb);
3662 EXPORT_SYMBOL(napi_frags_finish);
3664 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3666 struct sk_buff *skb = napi->skb;
3673 skb_reset_mac_header(skb);
3674 skb_gro_reset_offset(skb);
3676 off = skb_gro_offset(skb);
3677 hlen = off + sizeof(*eth);
3678 eth = skb_gro_header_fast(skb, off);
3679 if (skb_gro_header_hard(skb, hlen)) {
3680 eth = skb_gro_header_slow(skb, hlen, off);
3681 if (unlikely(!eth)) {
3682 napi_reuse_skb(napi, skb);
3688 skb_gro_pull(skb, sizeof(*eth));
3691 * This works because the only protocols we care about don't require
3692 * special handling. We'll fix it up properly at the end.
3694 skb->protocol = eth->h_proto;
3700 gro_result_t napi_gro_frags(struct napi_struct *napi)
3702 struct sk_buff *skb = napi_frags_skb(napi);
3707 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3709 EXPORT_SYMBOL(napi_gro_frags);
3712 * net_rps_action sends any pending IPI's for rps.
3713 * Note: called with local irq disabled, but exits with local irq enabled.
3715 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3718 struct softnet_data *remsd = sd->rps_ipi_list;
3721 sd->rps_ipi_list = NULL;
3725 /* Send pending IPI's to kick RPS processing on remote cpus. */
3727 struct softnet_data *next = remsd->rps_ipi_next;
3729 if (cpu_online(remsd->cpu))
3730 __smp_call_function_single(remsd->cpu,
3739 static int process_backlog(struct napi_struct *napi, int quota)
3742 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3745 /* Check if we have pending ipi, its better to send them now,
3746 * not waiting net_rx_action() end.
3748 if (sd->rps_ipi_list) {
3749 local_irq_disable();
3750 net_rps_action_and_irq_enable(sd);
3753 napi->weight = weight_p;
3754 local_irq_disable();
3755 while (work < quota) {
3756 struct sk_buff *skb;
3759 while ((skb = __skb_dequeue(&sd->process_queue))) {
3761 __netif_receive_skb(skb);
3762 local_irq_disable();
3763 input_queue_head_incr(sd);
3764 if (++work >= quota) {
3771 qlen = skb_queue_len(&sd->input_pkt_queue);
3773 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3774 &sd->process_queue);
3776 if (qlen < quota - work) {
3778 * Inline a custom version of __napi_complete().
3779 * only current cpu owns and manipulates this napi,
3780 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3781 * we can use a plain write instead of clear_bit(),
3782 * and we dont need an smp_mb() memory barrier.
3784 list_del(&napi->poll_list);
3787 quota = work + qlen;
3797 * __napi_schedule - schedule for receive
3798 * @n: entry to schedule
3800 * The entry's receive function will be scheduled to run
3802 void __napi_schedule(struct napi_struct *n)
3804 unsigned long flags;
3806 local_irq_save(flags);
3807 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3808 local_irq_restore(flags);
3810 EXPORT_SYMBOL(__napi_schedule);
3812 void __napi_complete(struct napi_struct *n)
3814 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3815 BUG_ON(n->gro_list);
3817 list_del(&n->poll_list);
3818 smp_mb__before_clear_bit();
3819 clear_bit(NAPI_STATE_SCHED, &n->state);
3821 EXPORT_SYMBOL(__napi_complete);
3823 void napi_complete(struct napi_struct *n)
3825 unsigned long flags;
3828 * don't let napi dequeue from the cpu poll list
3829 * just in case its running on a different cpu
3831 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3835 local_irq_save(flags);
3837 local_irq_restore(flags);
3839 EXPORT_SYMBOL(napi_complete);
3841 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3842 int (*poll)(struct napi_struct *, int), int weight)
3844 INIT_LIST_HEAD(&napi->poll_list);
3845 napi->gro_count = 0;
3846 napi->gro_list = NULL;
3849 napi->weight = weight;
3850 list_add(&napi->dev_list, &dev->napi_list);
3852 #ifdef CONFIG_NETPOLL
3853 spin_lock_init(&napi->poll_lock);
3854 napi->poll_owner = -1;
3856 set_bit(NAPI_STATE_SCHED, &napi->state);
3858 EXPORT_SYMBOL(netif_napi_add);
3860 void netif_napi_del(struct napi_struct *napi)
3862 struct sk_buff *skb, *next;
3864 list_del_init(&napi->dev_list);
3865 napi_free_frags(napi);
3867 for (skb = napi->gro_list; skb; skb = next) {
3873 napi->gro_list = NULL;
3874 napi->gro_count = 0;
3876 EXPORT_SYMBOL(netif_napi_del);
3878 static void net_rx_action(struct softirq_action *h)
3880 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3881 unsigned long time_limit = jiffies + 2;
3882 int budget = netdev_budget;
3885 local_irq_disable();
3887 while (!list_empty(&sd->poll_list)) {
3888 struct napi_struct *n;
3891 /* If softirq window is exhuasted then punt.
3892 * Allow this to run for 2 jiffies since which will allow
3893 * an average latency of 1.5/HZ.
3895 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3900 /* Even though interrupts have been re-enabled, this
3901 * access is safe because interrupts can only add new
3902 * entries to the tail of this list, and only ->poll()
3903 * calls can remove this head entry from the list.
3905 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3907 have = netpoll_poll_lock(n);
3911 /* This NAPI_STATE_SCHED test is for avoiding a race
3912 * with netpoll's poll_napi(). Only the entity which
3913 * obtains the lock and sees NAPI_STATE_SCHED set will
3914 * actually make the ->poll() call. Therefore we avoid
3915 * accidentally calling ->poll() when NAPI is not scheduled.
3918 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3919 work = n->poll(n, weight);
3923 WARN_ON_ONCE(work > weight);
3927 local_irq_disable();
3929 /* Drivers must not modify the NAPI state if they
3930 * consume the entire weight. In such cases this code
3931 * still "owns" the NAPI instance and therefore can
3932 * move the instance around on the list at-will.
3934 if (unlikely(work == weight)) {
3935 if (unlikely(napi_disable_pending(n))) {
3938 local_irq_disable();
3940 list_move_tail(&n->poll_list, &sd->poll_list);
3943 netpoll_poll_unlock(have);
3946 net_rps_action_and_irq_enable(sd);
3948 #ifdef CONFIG_NET_DMA
3950 * There may not be any more sk_buffs coming right now, so push
3951 * any pending DMA copies to hardware
3953 dma_issue_pending_all();
3960 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3964 static gifconf_func_t *gifconf_list[NPROTO];
3967 * register_gifconf - register a SIOCGIF handler
3968 * @family: Address family
3969 * @gifconf: Function handler
3971 * Register protocol dependent address dumping routines. The handler
3972 * that is passed must not be freed or reused until it has been replaced
3973 * by another handler.
3975 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3977 if (family >= NPROTO)
3979 gifconf_list[family] = gifconf;
3982 EXPORT_SYMBOL(register_gifconf);
3986 * Map an interface index to its name (SIOCGIFNAME)
3990 * We need this ioctl for efficient implementation of the
3991 * if_indextoname() function required by the IPv6 API. Without
3992 * it, we would have to search all the interfaces to find a
3996 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3998 struct net_device *dev;
4002 * Fetch the caller's info block.
4005 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4009 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4015 strcpy(ifr.ifr_name, dev->name);
4018 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4024 * Perform a SIOCGIFCONF call. This structure will change
4025 * size eventually, and there is nothing I can do about it.
4026 * Thus we will need a 'compatibility mode'.
4029 static int dev_ifconf(struct net *net, char __user *arg)
4032 struct net_device *dev;
4039 * Fetch the caller's info block.
4042 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4049 * Loop over the interfaces, and write an info block for each.
4053 for_each_netdev(net, dev) {
4054 for (i = 0; i < NPROTO; i++) {
4055 if (gifconf_list[i]) {
4058 done = gifconf_list[i](dev, NULL, 0);
4060 done = gifconf_list[i](dev, pos + total,
4070 * All done. Write the updated control block back to the caller.
4072 ifc.ifc_len = total;
4075 * Both BSD and Solaris return 0 here, so we do too.
4077 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4080 #ifdef CONFIG_PROC_FS
4082 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4084 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4085 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4086 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4088 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4090 struct net *net = seq_file_net(seq);
4091 struct net_device *dev;
4092 struct hlist_node *p;
4093 struct hlist_head *h;
4094 unsigned int count = 0, offset = get_offset(*pos);
4096 h = &net->dev_name_head[get_bucket(*pos)];
4097 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4098 if (++count == offset)
4105 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4107 struct net_device *dev;
4108 unsigned int bucket;
4111 dev = dev_from_same_bucket(seq, pos);
4115 bucket = get_bucket(*pos) + 1;
4116 *pos = set_bucket_offset(bucket, 1);
4117 } while (bucket < NETDEV_HASHENTRIES);
4123 * This is invoked by the /proc filesystem handler to display a device
4126 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4131 return SEQ_START_TOKEN;
4133 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4136 return dev_from_bucket(seq, pos);
4139 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4142 return dev_from_bucket(seq, pos);
4145 void dev_seq_stop(struct seq_file *seq, void *v)
4151 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4153 struct rtnl_link_stats64 temp;
4154 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4156 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4157 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4158 dev->name, stats->rx_bytes, stats->rx_packets,
4160 stats->rx_dropped + stats->rx_missed_errors,
4161 stats->rx_fifo_errors,
4162 stats->rx_length_errors + stats->rx_over_errors +
4163 stats->rx_crc_errors + stats->rx_frame_errors,
4164 stats->rx_compressed, stats->multicast,
4165 stats->tx_bytes, stats->tx_packets,
4166 stats->tx_errors, stats->tx_dropped,
4167 stats->tx_fifo_errors, stats->collisions,
4168 stats->tx_carrier_errors +
4169 stats->tx_aborted_errors +
4170 stats->tx_window_errors +
4171 stats->tx_heartbeat_errors,
4172 stats->tx_compressed);
4176 * Called from the PROCfs module. This now uses the new arbitrary sized
4177 * /proc/net interface to create /proc/net/dev
4179 static int dev_seq_show(struct seq_file *seq, void *v)
4181 if (v == SEQ_START_TOKEN)
4182 seq_puts(seq, "Inter-| Receive "
4184 " face |bytes packets errs drop fifo frame "
4185 "compressed multicast|bytes packets errs "
4186 "drop fifo colls carrier compressed\n");
4188 dev_seq_printf_stats(seq, v);
4192 static struct softnet_data *softnet_get_online(loff_t *pos)
4194 struct softnet_data *sd = NULL;
4196 while (*pos < nr_cpu_ids)
4197 if (cpu_online(*pos)) {
4198 sd = &per_cpu(softnet_data, *pos);
4205 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4207 return softnet_get_online(pos);
4210 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4213 return softnet_get_online(pos);
4216 static void softnet_seq_stop(struct seq_file *seq, void *v)
4220 static int softnet_seq_show(struct seq_file *seq, void *v)
4222 struct softnet_data *sd = v;
4224 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4225 sd->processed, sd->dropped, sd->time_squeeze, 0,
4226 0, 0, 0, 0, /* was fastroute */
4227 sd->cpu_collision, sd->received_rps);
4231 static const struct seq_operations dev_seq_ops = {
4232 .start = dev_seq_start,
4233 .next = dev_seq_next,
4234 .stop = dev_seq_stop,
4235 .show = dev_seq_show,
4238 static int dev_seq_open(struct inode *inode, struct file *file)
4240 return seq_open_net(inode, file, &dev_seq_ops,
4241 sizeof(struct seq_net_private));
4244 static const struct file_operations dev_seq_fops = {
4245 .owner = THIS_MODULE,
4246 .open = dev_seq_open,
4248 .llseek = seq_lseek,
4249 .release = seq_release_net,
4252 static const struct seq_operations softnet_seq_ops = {
4253 .start = softnet_seq_start,
4254 .next = softnet_seq_next,
4255 .stop = softnet_seq_stop,
4256 .show = softnet_seq_show,
4259 static int softnet_seq_open(struct inode *inode, struct file *file)
4261 return seq_open(file, &softnet_seq_ops);
4264 static const struct file_operations softnet_seq_fops = {
4265 .owner = THIS_MODULE,
4266 .open = softnet_seq_open,
4268 .llseek = seq_lseek,
4269 .release = seq_release,
4272 static void *ptype_get_idx(loff_t pos)
4274 struct packet_type *pt = NULL;
4278 list_for_each_entry_rcu(pt, &ptype_all, list) {
4284 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4285 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4294 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4298 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4301 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4303 struct packet_type *pt;
4304 struct list_head *nxt;
4308 if (v == SEQ_START_TOKEN)
4309 return ptype_get_idx(0);
4312 nxt = pt->list.next;
4313 if (pt->type == htons(ETH_P_ALL)) {
4314 if (nxt != &ptype_all)
4317 nxt = ptype_base[0].next;
4319 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4321 while (nxt == &ptype_base[hash]) {
4322 if (++hash >= PTYPE_HASH_SIZE)
4324 nxt = ptype_base[hash].next;
4327 return list_entry(nxt, struct packet_type, list);
4330 static void ptype_seq_stop(struct seq_file *seq, void *v)
4336 static int ptype_seq_show(struct seq_file *seq, void *v)
4338 struct packet_type *pt = v;
4340 if (v == SEQ_START_TOKEN)
4341 seq_puts(seq, "Type Device Function\n");
4342 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4343 if (pt->type == htons(ETH_P_ALL))
4344 seq_puts(seq, "ALL ");
4346 seq_printf(seq, "%04x", ntohs(pt->type));
4348 seq_printf(seq, " %-8s %pF\n",
4349 pt->dev ? pt->dev->name : "", pt->func);
4355 static const struct seq_operations ptype_seq_ops = {
4356 .start = ptype_seq_start,
4357 .next = ptype_seq_next,
4358 .stop = ptype_seq_stop,
4359 .show = ptype_seq_show,
4362 static int ptype_seq_open(struct inode *inode, struct file *file)
4364 return seq_open_net(inode, file, &ptype_seq_ops,
4365 sizeof(struct seq_net_private));
4368 static const struct file_operations ptype_seq_fops = {
4369 .owner = THIS_MODULE,
4370 .open = ptype_seq_open,
4372 .llseek = seq_lseek,
4373 .release = seq_release_net,
4377 static int __net_init dev_proc_net_init(struct net *net)
4381 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4383 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4385 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4388 if (wext_proc_init(net))
4394 proc_net_remove(net, "ptype");
4396 proc_net_remove(net, "softnet_stat");
4398 proc_net_remove(net, "dev");
4402 static void __net_exit dev_proc_net_exit(struct net *net)
4404 wext_proc_exit(net);
4406 proc_net_remove(net, "ptype");
4407 proc_net_remove(net, "softnet_stat");
4408 proc_net_remove(net, "dev");
4411 static struct pernet_operations __net_initdata dev_proc_ops = {
4412 .init = dev_proc_net_init,
4413 .exit = dev_proc_net_exit,
4416 static int __init dev_proc_init(void)
4418 return register_pernet_subsys(&dev_proc_ops);
4421 #define dev_proc_init() 0
4422 #endif /* CONFIG_PROC_FS */
4426 * netdev_set_master - set up master pointer
4427 * @slave: slave device
4428 * @master: new master device
4430 * Changes the master device of the slave. Pass %NULL to break the
4431 * bonding. The caller must hold the RTNL semaphore. On a failure
4432 * a negative errno code is returned. On success the reference counts
4433 * are adjusted and the function returns zero.
4435 int netdev_set_master(struct net_device *slave, struct net_device *master)
4437 struct net_device *old = slave->master;
4447 slave->master = master;
4453 EXPORT_SYMBOL(netdev_set_master);
4456 * netdev_set_bond_master - set up bonding master/slave pair
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 %RTM_NEWLINK is sent
4463 * to the routing socket and the function returns zero.
4465 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4471 err = netdev_set_master(slave, master);
4475 slave->flags |= IFF_SLAVE;
4477 slave->flags &= ~IFF_SLAVE;
4479 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4482 EXPORT_SYMBOL(netdev_set_bond_master);
4484 static void dev_change_rx_flags(struct net_device *dev, int flags)
4486 const struct net_device_ops *ops = dev->netdev_ops;
4488 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4489 ops->ndo_change_rx_flags(dev, flags);
4492 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4494 unsigned int old_flags = dev->flags;
4500 dev->flags |= IFF_PROMISC;
4501 dev->promiscuity += inc;
4502 if (dev->promiscuity == 0) {
4505 * If inc causes overflow, untouch promisc and return error.
4508 dev->flags &= ~IFF_PROMISC;
4510 dev->promiscuity -= inc;
4511 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4516 if (dev->flags != old_flags) {
4517 pr_info("device %s %s promiscuous mode\n",
4519 dev->flags & IFF_PROMISC ? "entered" : "left");
4520 if (audit_enabled) {
4521 current_uid_gid(&uid, &gid);
4522 audit_log(current->audit_context, GFP_ATOMIC,
4523 AUDIT_ANOM_PROMISCUOUS,
4524 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4525 dev->name, (dev->flags & IFF_PROMISC),
4526 (old_flags & IFF_PROMISC),
4527 audit_get_loginuid(current),
4529 audit_get_sessionid(current));
4532 dev_change_rx_flags(dev, IFF_PROMISC);
4538 * dev_set_promiscuity - update promiscuity count on a device
4542 * Add or remove promiscuity from a device. While the count in the device
4543 * remains above zero the interface remains promiscuous. Once it hits zero
4544 * the device reverts back to normal filtering operation. A negative inc
4545 * value is used to drop promiscuity on the device.
4546 * Return 0 if successful or a negative errno code on error.
4548 int dev_set_promiscuity(struct net_device *dev, int inc)
4550 unsigned int old_flags = dev->flags;
4553 err = __dev_set_promiscuity(dev, inc);
4556 if (dev->flags != old_flags)
4557 dev_set_rx_mode(dev);
4560 EXPORT_SYMBOL(dev_set_promiscuity);
4563 * dev_set_allmulti - update allmulti count on a device
4567 * Add or remove reception of all multicast frames to a device. While the
4568 * count in the device remains above zero the interface remains listening
4569 * to all interfaces. Once it hits zero the device reverts back to normal
4570 * filtering operation. A negative @inc value is used to drop the counter
4571 * when releasing a resource needing all multicasts.
4572 * Return 0 if successful or a negative errno code on error.
4575 int dev_set_allmulti(struct net_device *dev, int inc)
4577 unsigned int old_flags = dev->flags;
4581 dev->flags |= IFF_ALLMULTI;
4582 dev->allmulti += inc;
4583 if (dev->allmulti == 0) {
4586 * If inc causes overflow, untouch allmulti and return error.
4589 dev->flags &= ~IFF_ALLMULTI;
4591 dev->allmulti -= inc;
4592 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4597 if (dev->flags ^ old_flags) {
4598 dev_change_rx_flags(dev, IFF_ALLMULTI);
4599 dev_set_rx_mode(dev);
4603 EXPORT_SYMBOL(dev_set_allmulti);
4606 * Upload unicast and multicast address lists to device and
4607 * configure RX filtering. When the device doesn't support unicast
4608 * filtering it is put in promiscuous mode while unicast addresses
4611 void __dev_set_rx_mode(struct net_device *dev)
4613 const struct net_device_ops *ops = dev->netdev_ops;
4615 /* dev_open will call this function so the list will stay sane. */
4616 if (!(dev->flags&IFF_UP))
4619 if (!netif_device_present(dev))
4622 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4623 /* Unicast addresses changes may only happen under the rtnl,
4624 * therefore calling __dev_set_promiscuity here is safe.
4626 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4627 __dev_set_promiscuity(dev, 1);
4628 dev->uc_promisc = true;
4629 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4630 __dev_set_promiscuity(dev, -1);
4631 dev->uc_promisc = false;
4635 if (ops->ndo_set_rx_mode)
4636 ops->ndo_set_rx_mode(dev);
4639 void dev_set_rx_mode(struct net_device *dev)
4641 netif_addr_lock_bh(dev);
4642 __dev_set_rx_mode(dev);
4643 netif_addr_unlock_bh(dev);
4647 * dev_get_flags - get flags reported to userspace
4650 * Get the combination of flag bits exported through APIs to userspace.
4652 unsigned int dev_get_flags(const struct net_device *dev)
4656 flags = (dev->flags & ~(IFF_PROMISC |
4661 (dev->gflags & (IFF_PROMISC |
4664 if (netif_running(dev)) {
4665 if (netif_oper_up(dev))
4666 flags |= IFF_RUNNING;
4667 if (netif_carrier_ok(dev))
4668 flags |= IFF_LOWER_UP;
4669 if (netif_dormant(dev))
4670 flags |= IFF_DORMANT;
4675 EXPORT_SYMBOL(dev_get_flags);
4677 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4679 unsigned int old_flags = dev->flags;
4685 * Set the flags on our device.
4688 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4689 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4691 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4695 * Load in the correct multicast list now the flags have changed.
4698 if ((old_flags ^ flags) & IFF_MULTICAST)
4699 dev_change_rx_flags(dev, IFF_MULTICAST);
4701 dev_set_rx_mode(dev);
4704 * Have we downed the interface. We handle IFF_UP ourselves
4705 * according to user attempts to set it, rather than blindly
4710 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4711 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4714 dev_set_rx_mode(dev);
4717 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4718 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4720 dev->gflags ^= IFF_PROMISC;
4721 dev_set_promiscuity(dev, inc);
4724 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4725 is important. Some (broken) drivers set IFF_PROMISC, when
4726 IFF_ALLMULTI is requested not asking us and not reporting.
4728 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4729 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4731 dev->gflags ^= IFF_ALLMULTI;
4732 dev_set_allmulti(dev, inc);
4738 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4740 unsigned int changes = dev->flags ^ old_flags;
4742 if (changes & IFF_UP) {
4743 if (dev->flags & IFF_UP)
4744 call_netdevice_notifiers(NETDEV_UP, dev);
4746 call_netdevice_notifiers(NETDEV_DOWN, dev);
4749 if (dev->flags & IFF_UP &&
4750 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4751 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4755 * dev_change_flags - change device settings
4757 * @flags: device state flags
4759 * Change settings on device based state flags. The flags are
4760 * in the userspace exported format.
4762 int dev_change_flags(struct net_device *dev, unsigned int flags)
4765 unsigned int changes, old_flags = dev->flags;
4767 ret = __dev_change_flags(dev, flags);
4771 changes = old_flags ^ dev->flags;
4773 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4775 __dev_notify_flags(dev, old_flags);
4778 EXPORT_SYMBOL(dev_change_flags);
4781 * dev_set_mtu - Change maximum transfer unit
4783 * @new_mtu: new transfer unit
4785 * Change the maximum transfer size of the network device.
4787 int dev_set_mtu(struct net_device *dev, int new_mtu)
4789 const struct net_device_ops *ops = dev->netdev_ops;
4792 if (new_mtu == dev->mtu)
4795 /* MTU must be positive. */
4799 if (!netif_device_present(dev))
4803 if (ops->ndo_change_mtu)
4804 err = ops->ndo_change_mtu(dev, new_mtu);
4808 if (!err && dev->flags & IFF_UP)
4809 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4812 EXPORT_SYMBOL(dev_set_mtu);
4815 * dev_set_group - Change group this device belongs to
4817 * @new_group: group this device should belong to
4819 void dev_set_group(struct net_device *dev, int new_group)
4821 dev->group = new_group;
4823 EXPORT_SYMBOL(dev_set_group);
4826 * dev_set_mac_address - Change Media Access Control Address
4830 * Change the hardware (MAC) address of the device
4832 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4834 const struct net_device_ops *ops = dev->netdev_ops;
4837 if (!ops->ndo_set_mac_address)
4839 if (sa->sa_family != dev->type)
4841 if (!netif_device_present(dev))
4843 err = ops->ndo_set_mac_address(dev, sa);
4845 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4846 add_device_randomness(dev->dev_addr, dev->addr_len);
4849 EXPORT_SYMBOL(dev_set_mac_address);
4852 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4854 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4857 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4863 case SIOCGIFFLAGS: /* Get interface flags */
4864 ifr->ifr_flags = (short) dev_get_flags(dev);
4867 case SIOCGIFMETRIC: /* Get the metric on the interface
4868 (currently unused) */
4869 ifr->ifr_metric = 0;
4872 case SIOCGIFMTU: /* Get the MTU of a device */
4873 ifr->ifr_mtu = dev->mtu;
4878 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4880 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4881 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4882 ifr->ifr_hwaddr.sa_family = dev->type;
4890 ifr->ifr_map.mem_start = dev->mem_start;
4891 ifr->ifr_map.mem_end = dev->mem_end;
4892 ifr->ifr_map.base_addr = dev->base_addr;
4893 ifr->ifr_map.irq = dev->irq;
4894 ifr->ifr_map.dma = dev->dma;
4895 ifr->ifr_map.port = dev->if_port;
4899 ifr->ifr_ifindex = dev->ifindex;
4903 ifr->ifr_qlen = dev->tx_queue_len;
4907 /* dev_ioctl() should ensure this case
4919 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4921 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4924 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4925 const struct net_device_ops *ops;
4930 ops = dev->netdev_ops;
4933 case SIOCSIFFLAGS: /* Set interface flags */
4934 return dev_change_flags(dev, ifr->ifr_flags);
4936 case SIOCSIFMETRIC: /* Set the metric on the interface
4937 (currently unused) */
4940 case SIOCSIFMTU: /* Set the MTU of a device */
4941 return dev_set_mtu(dev, ifr->ifr_mtu);
4944 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4946 case SIOCSIFHWBROADCAST:
4947 if (ifr->ifr_hwaddr.sa_family != dev->type)
4949 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4950 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4951 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4955 if (ops->ndo_set_config) {
4956 if (!netif_device_present(dev))
4958 return ops->ndo_set_config(dev, &ifr->ifr_map);
4963 if (!ops->ndo_set_rx_mode ||
4964 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4966 if (!netif_device_present(dev))
4968 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4971 if (!ops->ndo_set_rx_mode ||
4972 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4974 if (!netif_device_present(dev))
4976 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4979 if (ifr->ifr_qlen < 0)
4981 dev->tx_queue_len = ifr->ifr_qlen;
4985 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4986 return dev_change_name(dev, ifr->ifr_newname);
4989 err = net_hwtstamp_validate(ifr);
4995 * Unknown or private ioctl
4998 if ((cmd >= SIOCDEVPRIVATE &&
4999 cmd <= SIOCDEVPRIVATE + 15) ||
5000 cmd == SIOCBONDENSLAVE ||
5001 cmd == SIOCBONDRELEASE ||
5002 cmd == SIOCBONDSETHWADDR ||
5003 cmd == SIOCBONDSLAVEINFOQUERY ||
5004 cmd == SIOCBONDINFOQUERY ||
5005 cmd == SIOCBONDCHANGEACTIVE ||
5006 cmd == SIOCGMIIPHY ||
5007 cmd == SIOCGMIIREG ||
5008 cmd == SIOCSMIIREG ||
5009 cmd == SIOCBRADDIF ||
5010 cmd == SIOCBRDELIF ||
5011 cmd == SIOCSHWTSTAMP ||
5012 cmd == SIOCWANDEV) {
5014 if (ops->ndo_do_ioctl) {
5015 if (netif_device_present(dev))
5016 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5028 * This function handles all "interface"-type I/O control requests. The actual
5029 * 'doing' part of this is dev_ifsioc above.
5033 * dev_ioctl - network device ioctl
5034 * @net: the applicable net namespace
5035 * @cmd: command to issue
5036 * @arg: pointer to a struct ifreq in user space
5038 * Issue ioctl functions to devices. This is normally called by the
5039 * user space syscall interfaces but can sometimes be useful for
5040 * other purposes. The return value is the return from the syscall if
5041 * positive or a negative errno code on error.
5044 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5050 /* One special case: SIOCGIFCONF takes ifconf argument
5051 and requires shared lock, because it sleeps writing
5055 if (cmd == SIOCGIFCONF) {
5057 ret = dev_ifconf(net, (char __user *) arg);
5061 if (cmd == SIOCGIFNAME)
5062 return dev_ifname(net, (struct ifreq __user *)arg);
5064 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5067 ifr.ifr_name[IFNAMSIZ-1] = 0;
5069 colon = strchr(ifr.ifr_name, ':');
5074 * See which interface the caller is talking about.
5079 * These ioctl calls:
5080 * - can be done by all.
5081 * - atomic and do not require locking.
5092 dev_load(net, ifr.ifr_name);
5094 ret = dev_ifsioc_locked(net, &ifr, cmd);
5099 if (copy_to_user(arg, &ifr,
5100 sizeof(struct ifreq)))
5106 dev_load(net, ifr.ifr_name);
5108 ret = dev_ethtool(net, &ifr);
5113 if (copy_to_user(arg, &ifr,
5114 sizeof(struct ifreq)))
5120 * These ioctl calls:
5121 * - require superuser power.
5122 * - require strict serialization.
5128 if (!capable(CAP_NET_ADMIN))
5130 dev_load(net, ifr.ifr_name);
5132 ret = dev_ifsioc(net, &ifr, cmd);
5137 if (copy_to_user(arg, &ifr,
5138 sizeof(struct ifreq)))
5144 * These ioctl calls:
5145 * - require superuser power.
5146 * - require strict serialization.
5147 * - do not return a value
5157 case SIOCSIFHWBROADCAST:
5160 case SIOCBONDENSLAVE:
5161 case SIOCBONDRELEASE:
5162 case SIOCBONDSETHWADDR:
5163 case SIOCBONDCHANGEACTIVE:
5167 if (!capable(CAP_NET_ADMIN))
5170 case SIOCBONDSLAVEINFOQUERY:
5171 case SIOCBONDINFOQUERY:
5172 dev_load(net, ifr.ifr_name);
5174 ret = dev_ifsioc(net, &ifr, cmd);
5179 /* Get the per device memory space. We can add this but
5180 * currently do not support it */
5182 /* Set the per device memory buffer space.
5183 * Not applicable in our case */
5188 * Unknown or private ioctl.
5191 if (cmd == SIOCWANDEV ||
5192 (cmd >= SIOCDEVPRIVATE &&
5193 cmd <= SIOCDEVPRIVATE + 15)) {
5194 dev_load(net, ifr.ifr_name);
5196 ret = dev_ifsioc(net, &ifr, cmd);
5198 if (!ret && copy_to_user(arg, &ifr,
5199 sizeof(struct ifreq)))
5203 /* Take care of Wireless Extensions */
5204 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5205 return wext_handle_ioctl(net, &ifr, cmd, arg);
5212 * dev_new_index - allocate an ifindex
5213 * @net: the applicable net namespace
5215 * Returns a suitable unique value for a new device interface
5216 * number. The caller must hold the rtnl semaphore or the
5217 * dev_base_lock to be sure it remains unique.
5219 static int dev_new_index(struct net *net)
5225 if (!__dev_get_by_index(net, ifindex))
5230 /* Delayed registration/unregisteration */
5231 static LIST_HEAD(net_todo_list);
5233 static void net_set_todo(struct net_device *dev)
5235 list_add_tail(&dev->todo_list, &net_todo_list);
5238 static void rollback_registered_many(struct list_head *head)
5240 struct net_device *dev, *tmp;
5242 BUG_ON(dev_boot_phase);
5245 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5246 /* Some devices call without registering
5247 * for initialization unwind. Remove those
5248 * devices and proceed with the remaining.
5250 if (dev->reg_state == NETREG_UNINITIALIZED) {
5251 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5255 list_del(&dev->unreg_list);
5258 dev->dismantle = true;
5259 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5262 /* If device is running, close it first. */
5263 dev_close_many(head);
5265 list_for_each_entry(dev, head, unreg_list) {
5266 /* And unlink it from device chain. */
5267 unlist_netdevice(dev);
5269 dev->reg_state = NETREG_UNREGISTERING;
5274 list_for_each_entry(dev, head, unreg_list) {
5275 /* Shutdown queueing discipline. */
5279 /* Notify protocols, that we are about to destroy
5280 this device. They should clean all the things.
5282 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5284 if (!dev->rtnl_link_ops ||
5285 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5286 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5289 * Flush the unicast and multicast chains
5294 if (dev->netdev_ops->ndo_uninit)
5295 dev->netdev_ops->ndo_uninit(dev);
5297 /* Notifier chain MUST detach us from master device. */
5298 WARN_ON(dev->master);
5300 /* Remove entries from kobject tree */
5301 netdev_unregister_kobject(dev);
5304 /* Process any work delayed until the end of the batch */
5305 dev = list_first_entry(head, struct net_device, unreg_list);
5306 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5310 list_for_each_entry(dev, head, unreg_list)
5314 static void rollback_registered(struct net_device *dev)
5318 list_add(&dev->unreg_list, &single);
5319 rollback_registered_many(&single);
5323 static netdev_features_t netdev_fix_features(struct net_device *dev,
5324 netdev_features_t features)
5326 /* Fix illegal checksum combinations */
5327 if ((features & NETIF_F_HW_CSUM) &&
5328 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5329 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5330 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5333 /* Fix illegal SG+CSUM combinations. */
5334 if ((features & NETIF_F_SG) &&
5335 !(features & NETIF_F_ALL_CSUM)) {
5337 "Dropping NETIF_F_SG since no checksum feature.\n");
5338 features &= ~NETIF_F_SG;
5341 /* TSO requires that SG is present as well. */
5342 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5343 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5344 features &= ~NETIF_F_ALL_TSO;
5347 /* TSO ECN requires that TSO is present as well. */
5348 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5349 features &= ~NETIF_F_TSO_ECN;
5351 /* Software GSO depends on SG. */
5352 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5353 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5354 features &= ~NETIF_F_GSO;
5357 /* UFO needs SG and checksumming */
5358 if (features & NETIF_F_UFO) {
5359 /* maybe split UFO into V4 and V6? */
5360 if (!((features & NETIF_F_GEN_CSUM) ||
5361 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5362 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5364 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5365 features &= ~NETIF_F_UFO;
5368 if (!(features & NETIF_F_SG)) {
5370 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5371 features &= ~NETIF_F_UFO;
5378 int __netdev_update_features(struct net_device *dev)
5380 netdev_features_t features;
5385 features = netdev_get_wanted_features(dev);
5387 if (dev->netdev_ops->ndo_fix_features)
5388 features = dev->netdev_ops->ndo_fix_features(dev, features);
5390 /* driver might be less strict about feature dependencies */
5391 features = netdev_fix_features(dev, features);
5393 if (dev->features == features)
5396 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5397 &dev->features, &features);
5399 if (dev->netdev_ops->ndo_set_features)
5400 err = dev->netdev_ops->ndo_set_features(dev, features);
5402 if (unlikely(err < 0)) {
5404 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5405 err, &features, &dev->features);
5410 dev->features = features;
5416 * netdev_update_features - recalculate device features
5417 * @dev: the device to check
5419 * Recalculate dev->features set and send notifications if it
5420 * has changed. Should be called after driver or hardware dependent
5421 * conditions might have changed that influence the features.
5423 void netdev_update_features(struct net_device *dev)
5425 if (__netdev_update_features(dev))
5426 netdev_features_change(dev);
5428 EXPORT_SYMBOL(netdev_update_features);
5431 * netdev_change_features - recalculate device features
5432 * @dev: the device to check
5434 * Recalculate dev->features set and send notifications even
5435 * if they have not changed. Should be called instead of
5436 * netdev_update_features() if also dev->vlan_features might
5437 * have changed to allow the changes to be propagated to stacked
5440 void netdev_change_features(struct net_device *dev)
5442 __netdev_update_features(dev);
5443 netdev_features_change(dev);
5445 EXPORT_SYMBOL(netdev_change_features);
5448 * netif_stacked_transfer_operstate - transfer operstate
5449 * @rootdev: the root or lower level device to transfer state from
5450 * @dev: the device to transfer operstate to
5452 * Transfer operational state from root to device. This is normally
5453 * called when a stacking relationship exists between the root
5454 * device and the device(a leaf device).
5456 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5457 struct net_device *dev)
5459 if (rootdev->operstate == IF_OPER_DORMANT)
5460 netif_dormant_on(dev);
5462 netif_dormant_off(dev);
5464 if (netif_carrier_ok(rootdev)) {
5465 if (!netif_carrier_ok(dev))
5466 netif_carrier_on(dev);
5468 if (netif_carrier_ok(dev))
5469 netif_carrier_off(dev);
5472 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5475 static int netif_alloc_rx_queues(struct net_device *dev)
5477 unsigned int i, count = dev->num_rx_queues;
5478 struct netdev_rx_queue *rx;
5482 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5484 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5489 for (i = 0; i < count; i++)
5495 static void netdev_init_one_queue(struct net_device *dev,
5496 struct netdev_queue *queue, void *_unused)
5498 /* Initialize queue lock */
5499 spin_lock_init(&queue->_xmit_lock);
5500 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5501 queue->xmit_lock_owner = -1;
5502 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5505 dql_init(&queue->dql, HZ);
5509 static int netif_alloc_netdev_queues(struct net_device *dev)
5511 unsigned int count = dev->num_tx_queues;
5512 struct netdev_queue *tx;
5516 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5518 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5523 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5524 spin_lock_init(&dev->tx_global_lock);
5530 * register_netdevice - register a network device
5531 * @dev: device to register
5533 * Take a completed network device structure and add it to the kernel
5534 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5535 * chain. 0 is returned on success. A negative errno code is returned
5536 * on a failure to set up the device, or if the name is a duplicate.
5538 * Callers must hold the rtnl semaphore. You may want
5539 * register_netdev() instead of this.
5542 * The locking appears insufficient to guarantee two parallel registers
5543 * will not get the same name.
5546 int register_netdevice(struct net_device *dev)
5549 struct net *net = dev_net(dev);
5551 BUG_ON(dev_boot_phase);
5556 /* When net_device's are persistent, this will be fatal. */
5557 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5560 spin_lock_init(&dev->addr_list_lock);
5561 netdev_set_addr_lockdep_class(dev);
5565 ret = dev_get_valid_name(dev, dev->name);
5569 /* Init, if this function is available */
5570 if (dev->netdev_ops->ndo_init) {
5571 ret = dev->netdev_ops->ndo_init(dev);
5579 dev->ifindex = dev_new_index(net);
5580 if (dev->iflink == -1)
5581 dev->iflink = dev->ifindex;
5583 /* Transfer changeable features to wanted_features and enable
5584 * software offloads (GSO and GRO).
5586 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5587 dev->features |= NETIF_F_SOFT_FEATURES;
5588 dev->wanted_features = dev->features & dev->hw_features;
5590 /* Turn on no cache copy if HW is doing checksum */
5591 if (!(dev->flags & IFF_LOOPBACK)) {
5592 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5593 if (dev->features & NETIF_F_ALL_CSUM) {
5594 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5595 dev->features |= NETIF_F_NOCACHE_COPY;
5599 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5601 dev->vlan_features |= NETIF_F_HIGHDMA;
5603 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5604 ret = notifier_to_errno(ret);
5608 ret = netdev_register_kobject(dev);
5611 dev->reg_state = NETREG_REGISTERED;
5613 __netdev_update_features(dev);
5616 * Default initial state at registry is that the
5617 * device is present.
5620 set_bit(__LINK_STATE_PRESENT, &dev->state);
5622 dev_init_scheduler(dev);
5624 list_netdevice(dev);
5625 add_device_randomness(dev->dev_addr, dev->addr_len);
5627 /* Notify protocols, that a new device appeared. */
5628 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5629 ret = notifier_to_errno(ret);
5631 rollback_registered(dev);
5632 dev->reg_state = NETREG_UNREGISTERED;
5635 * Prevent userspace races by waiting until the network
5636 * device is fully setup before sending notifications.
5638 if (!dev->rtnl_link_ops ||
5639 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5640 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5646 if (dev->netdev_ops->ndo_uninit)
5647 dev->netdev_ops->ndo_uninit(dev);
5650 EXPORT_SYMBOL(register_netdevice);
5653 * init_dummy_netdev - init a dummy network device for NAPI
5654 * @dev: device to init
5656 * This takes a network device structure and initialize the minimum
5657 * amount of fields so it can be used to schedule NAPI polls without
5658 * registering a full blown interface. This is to be used by drivers
5659 * that need to tie several hardware interfaces to a single NAPI
5660 * poll scheduler due to HW limitations.
5662 int init_dummy_netdev(struct net_device *dev)
5664 /* Clear everything. Note we don't initialize spinlocks
5665 * are they aren't supposed to be taken by any of the
5666 * NAPI code and this dummy netdev is supposed to be
5667 * only ever used for NAPI polls
5669 memset(dev, 0, sizeof(struct net_device));
5671 /* make sure we BUG if trying to hit standard
5672 * register/unregister code path
5674 dev->reg_state = NETREG_DUMMY;
5676 /* NAPI wants this */
5677 INIT_LIST_HEAD(&dev->napi_list);
5679 /* a dummy interface is started by default */
5680 set_bit(__LINK_STATE_PRESENT, &dev->state);
5681 set_bit(__LINK_STATE_START, &dev->state);
5683 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5684 * because users of this 'device' dont need to change
5690 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5694 * register_netdev - register a network device
5695 * @dev: device to register
5697 * Take a completed network device structure and add it to the kernel
5698 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5699 * chain. 0 is returned on success. A negative errno code is returned
5700 * on a failure to set up the device, or if the name is a duplicate.
5702 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5703 * and expands the device name if you passed a format string to
5706 int register_netdev(struct net_device *dev)
5711 err = register_netdevice(dev);
5715 EXPORT_SYMBOL(register_netdev);
5717 int netdev_refcnt_read(const struct net_device *dev)
5721 for_each_possible_cpu(i)
5722 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5725 EXPORT_SYMBOL(netdev_refcnt_read);
5728 * netdev_wait_allrefs - wait until all references are gone.
5730 * This is called when unregistering network devices.
5732 * Any protocol or device that holds a reference should register
5733 * for netdevice notification, and cleanup and put back the
5734 * reference if they receive an UNREGISTER event.
5735 * We can get stuck here if buggy protocols don't correctly
5738 static void netdev_wait_allrefs(struct net_device *dev)
5740 unsigned long rebroadcast_time, warning_time;
5743 linkwatch_forget_dev(dev);
5745 rebroadcast_time = warning_time = jiffies;
5746 refcnt = netdev_refcnt_read(dev);
5748 while (refcnt != 0) {
5749 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5752 /* Rebroadcast unregister notification */
5753 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5754 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5755 * should have already handle it the first time */
5757 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5759 /* We must not have linkwatch events
5760 * pending on unregister. If this
5761 * happens, we simply run the queue
5762 * unscheduled, resulting in a noop
5765 linkwatch_run_queue();
5770 rebroadcast_time = jiffies;
5775 refcnt = netdev_refcnt_read(dev);
5777 if (time_after(jiffies, warning_time + 10 * HZ)) {
5778 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5780 warning_time = jiffies;
5789 * register_netdevice(x1);
5790 * register_netdevice(x2);
5792 * unregister_netdevice(y1);
5793 * unregister_netdevice(y2);
5799 * We are invoked by rtnl_unlock().
5800 * This allows us to deal with problems:
5801 * 1) We can delete sysfs objects which invoke hotplug
5802 * without deadlocking with linkwatch via keventd.
5803 * 2) Since we run with the RTNL semaphore not held, we can sleep
5804 * safely in order to wait for the netdev refcnt to drop to zero.
5806 * We must not return until all unregister events added during
5807 * the interval the lock was held have been completed.
5809 void netdev_run_todo(void)
5811 struct list_head list;
5813 /* Snapshot list, allow later requests */
5814 list_replace_init(&net_todo_list, &list);
5818 /* Wait for rcu callbacks to finish before attempting to drain
5819 * the device list. This usually avoids a 250ms wait.
5821 if (!list_empty(&list))
5824 while (!list_empty(&list)) {
5825 struct net_device *dev
5826 = list_first_entry(&list, struct net_device, todo_list);
5827 list_del(&dev->todo_list);
5829 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5830 pr_err("network todo '%s' but state %d\n",
5831 dev->name, dev->reg_state);
5836 dev->reg_state = NETREG_UNREGISTERED;
5838 on_each_cpu(flush_backlog, dev, 1);
5840 netdev_wait_allrefs(dev);
5843 BUG_ON(netdev_refcnt_read(dev));
5844 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5845 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5846 WARN_ON(dev->dn_ptr);
5848 if (dev->destructor)
5849 dev->destructor(dev);
5851 /* Free network device */
5852 kobject_put(&dev->dev.kobj);
5856 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5857 * fields in the same order, with only the type differing.
5859 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5860 const struct net_device_stats *netdev_stats)
5862 #if BITS_PER_LONG == 64
5863 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5864 memcpy(stats64, netdev_stats, sizeof(*stats64));
5866 size_t i, n = sizeof(*stats64) / sizeof(u64);
5867 const unsigned long *src = (const unsigned long *)netdev_stats;
5868 u64 *dst = (u64 *)stats64;
5870 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5871 sizeof(*stats64) / sizeof(u64));
5872 for (i = 0; i < n; i++)
5876 EXPORT_SYMBOL(netdev_stats_to_stats64);
5879 * dev_get_stats - get network device statistics
5880 * @dev: device to get statistics from
5881 * @storage: place to store stats
5883 * Get network statistics from device. Return @storage.
5884 * The device driver may provide its own method by setting
5885 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5886 * otherwise the internal statistics structure is used.
5888 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5889 struct rtnl_link_stats64 *storage)
5891 const struct net_device_ops *ops = dev->netdev_ops;
5893 if (ops->ndo_get_stats64) {
5894 memset(storage, 0, sizeof(*storage));
5895 ops->ndo_get_stats64(dev, storage);
5896 } else if (ops->ndo_get_stats) {
5897 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5899 netdev_stats_to_stats64(storage, &dev->stats);
5901 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5904 EXPORT_SYMBOL(dev_get_stats);
5906 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5908 struct netdev_queue *queue = dev_ingress_queue(dev);
5910 #ifdef CONFIG_NET_CLS_ACT
5913 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5916 netdev_init_one_queue(dev, queue, NULL);
5917 queue->qdisc = &noop_qdisc;
5918 queue->qdisc_sleeping = &noop_qdisc;
5919 rcu_assign_pointer(dev->ingress_queue, queue);
5925 * alloc_netdev_mqs - allocate network device
5926 * @sizeof_priv: size of private data to allocate space for
5927 * @name: device name format string
5928 * @setup: callback to initialize device
5929 * @txqs: the number of TX subqueues to allocate
5930 * @rxqs: the number of RX subqueues to allocate
5932 * Allocates a struct net_device with private data area for driver use
5933 * and performs basic initialization. Also allocates subquue structs
5934 * for each queue on the device.
5936 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5937 void (*setup)(struct net_device *),
5938 unsigned int txqs, unsigned int rxqs)
5940 struct net_device *dev;
5942 struct net_device *p;
5944 BUG_ON(strlen(name) >= sizeof(dev->name));
5947 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5953 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5958 alloc_size = sizeof(struct net_device);
5960 /* ensure 32-byte alignment of private area */
5961 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5962 alloc_size += sizeof_priv;
5964 /* ensure 32-byte alignment of whole construct */
5965 alloc_size += NETDEV_ALIGN - 1;
5967 p = kzalloc(alloc_size, GFP_KERNEL);
5969 pr_err("alloc_netdev: Unable to allocate device\n");
5973 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5974 dev->padded = (char *)dev - (char *)p;
5976 dev->pcpu_refcnt = alloc_percpu(int);
5977 if (!dev->pcpu_refcnt)
5980 if (dev_addr_init(dev))
5986 dev_net_set(dev, &init_net);
5988 dev->gso_max_size = GSO_MAX_SIZE;
5990 INIT_LIST_HEAD(&dev->napi_list);
5991 INIT_LIST_HEAD(&dev->unreg_list);
5992 INIT_LIST_HEAD(&dev->link_watch_list);
5993 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5996 dev->num_tx_queues = txqs;
5997 dev->real_num_tx_queues = txqs;
5998 if (netif_alloc_netdev_queues(dev))
6002 dev->num_rx_queues = rxqs;
6003 dev->real_num_rx_queues = rxqs;
6004 if (netif_alloc_rx_queues(dev))
6008 strcpy(dev->name, name);
6009 dev->group = INIT_NETDEV_GROUP;
6017 free_percpu(dev->pcpu_refcnt);
6027 EXPORT_SYMBOL(alloc_netdev_mqs);
6030 * free_netdev - free network device
6033 * This function does the last stage of destroying an allocated device
6034 * interface. The reference to the device object is released.
6035 * If this is the last reference then it will be freed.
6037 void free_netdev(struct net_device *dev)
6039 struct napi_struct *p, *n;
6041 release_net(dev_net(dev));
6048 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6050 /* Flush device addresses */
6051 dev_addr_flush(dev);
6053 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6056 free_percpu(dev->pcpu_refcnt);
6057 dev->pcpu_refcnt = NULL;
6059 /* Compatibility with error handling in drivers */
6060 if (dev->reg_state == NETREG_UNINITIALIZED) {
6061 kfree((char *)dev - dev->padded);
6065 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6066 dev->reg_state = NETREG_RELEASED;
6068 /* will free via device release */
6069 put_device(&dev->dev);
6071 EXPORT_SYMBOL(free_netdev);
6074 * synchronize_net - Synchronize with packet receive processing
6076 * Wait for packets currently being received to be done.
6077 * Does not block later packets from starting.
6079 void synchronize_net(void)
6082 if (rtnl_is_locked())
6083 synchronize_rcu_expedited();
6087 EXPORT_SYMBOL(synchronize_net);
6090 * unregister_netdevice_queue - remove device from the kernel
6094 * This function shuts down a device interface and removes it
6095 * from the kernel tables.
6096 * If head not NULL, device is queued to be unregistered later.
6098 * Callers must hold the rtnl semaphore. You may want
6099 * unregister_netdev() instead of this.
6102 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6107 list_move_tail(&dev->unreg_list, head);
6109 rollback_registered(dev);
6110 /* Finish processing unregister after unlock */
6114 EXPORT_SYMBOL(unregister_netdevice_queue);
6117 * unregister_netdevice_many - unregister many devices
6118 * @head: list of devices
6120 void unregister_netdevice_many(struct list_head *head)
6122 struct net_device *dev;
6124 if (!list_empty(head)) {
6125 rollback_registered_many(head);
6126 list_for_each_entry(dev, head, unreg_list)
6130 EXPORT_SYMBOL(unregister_netdevice_many);
6133 * unregister_netdev - remove device from the kernel
6136 * This function shuts down a device interface and removes it
6137 * from the kernel tables.
6139 * This is just a wrapper for unregister_netdevice that takes
6140 * the rtnl semaphore. In general you want to use this and not
6141 * unregister_netdevice.
6143 void unregister_netdev(struct net_device *dev)
6146 unregister_netdevice(dev);
6149 EXPORT_SYMBOL(unregister_netdev);
6152 * dev_change_net_namespace - move device to different nethost namespace
6154 * @net: network namespace
6155 * @pat: If not NULL name pattern to try if the current device name
6156 * is already taken in the destination network namespace.
6158 * This function shuts down a device interface and moves it
6159 * to a new network namespace. On success 0 is returned, on
6160 * a failure a netagive errno code is returned.
6162 * Callers must hold the rtnl semaphore.
6165 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6171 /* Don't allow namespace local devices to be moved. */
6173 if (dev->features & NETIF_F_NETNS_LOCAL)
6176 /* Ensure the device has been registrered */
6178 if (dev->reg_state != NETREG_REGISTERED)
6181 /* Get out if there is nothing todo */
6183 if (net_eq(dev_net(dev), net))
6186 /* Pick the destination device name, and ensure
6187 * we can use it in the destination network namespace.
6190 if (__dev_get_by_name(net, dev->name)) {
6191 /* We get here if we can't use the current device name */
6194 if (dev_get_valid_name(dev, pat) < 0)
6199 * And now a mini version of register_netdevice unregister_netdevice.
6202 /* If device is running close it first. */
6205 /* And unlink it from device chain */
6207 unlist_netdevice(dev);
6211 /* Shutdown queueing discipline. */
6214 /* Notify protocols, that we are about to destroy
6215 this device. They should clean all the things.
6217 Note that dev->reg_state stays at NETREG_REGISTERED.
6218 This is wanted because this way 8021q and macvlan know
6219 the device is just moving and can keep their slaves up.
6221 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6222 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6223 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6226 * Flush the unicast and multicast chains
6231 /* Actually switch the network namespace */
6232 dev_net_set(dev, net);
6234 /* If there is an ifindex conflict assign a new one */
6235 if (__dev_get_by_index(net, dev->ifindex)) {
6236 int iflink = (dev->iflink == dev->ifindex);
6237 dev->ifindex = dev_new_index(net);
6239 dev->iflink = dev->ifindex;
6242 /* Fixup kobjects */
6243 err = device_rename(&dev->dev, dev->name);
6246 /* Add the device back in the hashes */
6247 list_netdevice(dev);
6249 /* Notify protocols, that a new device appeared. */
6250 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6253 * Prevent userspace races by waiting until the network
6254 * device is fully setup before sending notifications.
6256 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6263 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6265 static int dev_cpu_callback(struct notifier_block *nfb,
6266 unsigned long action,
6269 struct sk_buff **list_skb;
6270 struct sk_buff *skb;
6271 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6272 struct softnet_data *sd, *oldsd;
6274 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6277 local_irq_disable();
6278 cpu = smp_processor_id();
6279 sd = &per_cpu(softnet_data, cpu);
6280 oldsd = &per_cpu(softnet_data, oldcpu);
6282 /* Find end of our completion_queue. */
6283 list_skb = &sd->completion_queue;
6285 list_skb = &(*list_skb)->next;
6286 /* Append completion queue from offline CPU. */
6287 *list_skb = oldsd->completion_queue;
6288 oldsd->completion_queue = NULL;
6290 /* Append output queue from offline CPU. */
6291 if (oldsd->output_queue) {
6292 *sd->output_queue_tailp = oldsd->output_queue;
6293 sd->output_queue_tailp = oldsd->output_queue_tailp;
6294 oldsd->output_queue = NULL;
6295 oldsd->output_queue_tailp = &oldsd->output_queue;
6297 /* Append NAPI poll list from offline CPU. */
6298 if (!list_empty(&oldsd->poll_list)) {
6299 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6300 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6303 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6306 /* Process offline CPU's input_pkt_queue */
6307 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6309 input_queue_head_incr(oldsd);
6311 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6313 input_queue_head_incr(oldsd);
6321 * netdev_increment_features - increment feature set by one
6322 * @all: current feature set
6323 * @one: new feature set
6324 * @mask: mask feature set
6326 * Computes a new feature set after adding a device with feature set
6327 * @one to the master device with current feature set @all. Will not
6328 * enable anything that is off in @mask. Returns the new feature set.
6330 netdev_features_t netdev_increment_features(netdev_features_t all,
6331 netdev_features_t one, netdev_features_t mask)
6333 if (mask & NETIF_F_GEN_CSUM)
6334 mask |= NETIF_F_ALL_CSUM;
6335 mask |= NETIF_F_VLAN_CHALLENGED;
6337 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6338 all &= one | ~NETIF_F_ALL_FOR_ALL;
6340 /* If one device supports hw checksumming, set for all. */
6341 if (all & NETIF_F_GEN_CSUM)
6342 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6346 EXPORT_SYMBOL(netdev_increment_features);
6348 static struct hlist_head *netdev_create_hash(void)
6351 struct hlist_head *hash;
6353 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6355 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6356 INIT_HLIST_HEAD(&hash[i]);
6361 /* Initialize per network namespace state */
6362 static int __net_init netdev_init(struct net *net)
6364 if (net != &init_net)
6365 INIT_LIST_HEAD(&net->dev_base_head);
6367 net->dev_name_head = netdev_create_hash();
6368 if (net->dev_name_head == NULL)
6371 net->dev_index_head = netdev_create_hash();
6372 if (net->dev_index_head == NULL)
6378 kfree(net->dev_name_head);
6384 * netdev_drivername - network driver for the device
6385 * @dev: network device
6387 * Determine network driver for device.
6389 const char *netdev_drivername(const struct net_device *dev)
6391 const struct device_driver *driver;
6392 const struct device *parent;
6393 const char *empty = "";
6395 parent = dev->dev.parent;
6399 driver = parent->driver;
6400 if (driver && driver->name)
6401 return driver->name;
6405 int __netdev_printk(const char *level, const struct net_device *dev,
6406 struct va_format *vaf)
6410 if (dev && dev->dev.parent)
6411 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6412 netdev_name(dev), vaf);
6414 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6416 r = printk("%s(NULL net_device): %pV", level, vaf);
6420 EXPORT_SYMBOL(__netdev_printk);
6422 int netdev_printk(const char *level, const struct net_device *dev,
6423 const char *format, ...)
6425 struct va_format vaf;
6429 va_start(args, format);
6434 r = __netdev_printk(level, dev, &vaf);
6439 EXPORT_SYMBOL(netdev_printk);
6441 #define define_netdev_printk_level(func, level) \
6442 int func(const struct net_device *dev, const char *fmt, ...) \
6445 struct va_format vaf; \
6448 va_start(args, fmt); \
6453 r = __netdev_printk(level, dev, &vaf); \
6458 EXPORT_SYMBOL(func);
6460 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6461 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6462 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6463 define_netdev_printk_level(netdev_err, KERN_ERR);
6464 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6465 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6466 define_netdev_printk_level(netdev_info, KERN_INFO);
6468 static void __net_exit netdev_exit(struct net *net)
6470 kfree(net->dev_name_head);
6471 kfree(net->dev_index_head);
6474 static struct pernet_operations __net_initdata netdev_net_ops = {
6475 .init = netdev_init,
6476 .exit = netdev_exit,
6479 static void __net_exit default_device_exit(struct net *net)
6481 struct net_device *dev, *aux;
6483 * Push all migratable network devices back to the
6484 * initial network namespace
6487 for_each_netdev_safe(net, dev, aux) {
6489 char fb_name[IFNAMSIZ];
6491 /* Ignore unmoveable devices (i.e. loopback) */
6492 if (dev->features & NETIF_F_NETNS_LOCAL)
6495 /* Leave virtual devices for the generic cleanup */
6496 if (dev->rtnl_link_ops)
6499 /* Push remaining network devices to init_net */
6500 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6501 err = dev_change_net_namespace(dev, &init_net, fb_name);
6503 pr_emerg("%s: failed to move %s to init_net: %d\n",
6504 __func__, dev->name, err);
6511 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6513 /* At exit all network devices most be removed from a network
6514 * namespace. Do this in the reverse order of registration.
6515 * Do this across as many network namespaces as possible to
6516 * improve batching efficiency.
6518 struct net_device *dev;
6520 LIST_HEAD(dev_kill_list);
6523 list_for_each_entry(net, net_list, exit_list) {
6524 for_each_netdev_reverse(net, dev) {
6525 if (dev->rtnl_link_ops)
6526 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6528 unregister_netdevice_queue(dev, &dev_kill_list);
6531 unregister_netdevice_many(&dev_kill_list);
6532 list_del(&dev_kill_list);
6536 static struct pernet_operations __net_initdata default_device_ops = {
6537 .exit = default_device_exit,
6538 .exit_batch = default_device_exit_batch,
6542 * Initialize the DEV module. At boot time this walks the device list and
6543 * unhooks any devices that fail to initialise (normally hardware not
6544 * present) and leaves us with a valid list of present and active devices.
6549 * This is called single threaded during boot, so no need
6550 * to take the rtnl semaphore.
6552 static int __init net_dev_init(void)
6554 int i, rc = -ENOMEM;
6556 BUG_ON(!dev_boot_phase);
6558 if (dev_proc_init())
6561 if (netdev_kobject_init())
6564 INIT_LIST_HEAD(&ptype_all);
6565 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6566 INIT_LIST_HEAD(&ptype_base[i]);
6568 if (register_pernet_subsys(&netdev_net_ops))
6572 * Initialise the packet receive queues.
6575 for_each_possible_cpu(i) {
6576 struct softnet_data *sd = &per_cpu(softnet_data, i);
6578 memset(sd, 0, sizeof(*sd));
6579 skb_queue_head_init(&sd->input_pkt_queue);
6580 skb_queue_head_init(&sd->process_queue);
6581 sd->completion_queue = NULL;
6582 INIT_LIST_HEAD(&sd->poll_list);
6583 sd->output_queue = NULL;
6584 sd->output_queue_tailp = &sd->output_queue;
6586 sd->csd.func = rps_trigger_softirq;
6592 sd->backlog.poll = process_backlog;
6593 sd->backlog.weight = weight_p;
6594 sd->backlog.gro_list = NULL;
6595 sd->backlog.gro_count = 0;
6600 /* The loopback device is special if any other network devices
6601 * is present in a network namespace the loopback device must
6602 * be present. Since we now dynamically allocate and free the
6603 * loopback device ensure this invariant is maintained by
6604 * keeping the loopback device as the first device on the
6605 * list of network devices. Ensuring the loopback devices
6606 * is the first device that appears and the last network device
6609 if (register_pernet_device(&loopback_net_ops))
6612 if (register_pernet_device(&default_device_ops))
6615 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6616 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6618 hotcpu_notifier(dev_cpu_callback, 0);
6626 subsys_initcall(net_dev_init);
6628 static int __init initialize_hashrnd(void)
6630 get_random_bytes(&hashrnd, sizeof(hashrnd));
6634 late_initcall_sync(initialize_hashrnd);