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_alloc_name_ns(struct net *net,
963 struct net_device *dev,
969 ret = __dev_alloc_name(net, name, buf);
971 strlcpy(dev->name, buf, IFNAMSIZ);
975 static int dev_get_valid_name(struct net *net,
976 struct net_device *dev,
981 if (!dev_valid_name(name))
984 if (strchr(name, '%'))
985 return dev_alloc_name_ns(net, dev, name);
986 else if (__dev_get_by_name(net, name))
988 else if (dev->name != name)
989 strlcpy(dev->name, name, IFNAMSIZ);
995 * dev_change_name - change name of a device
997 * @newname: name (or format string) must be at least IFNAMSIZ
999 * Change name of a device, can pass format strings "eth%d".
1002 int dev_change_name(struct net_device *dev, const char *newname)
1004 char oldname[IFNAMSIZ];
1010 BUG_ON(!dev_net(dev));
1013 if (dev->flags & IFF_UP)
1016 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1019 memcpy(oldname, dev->name, IFNAMSIZ);
1021 err = dev_get_valid_name(net, dev, newname);
1026 ret = device_rename(&dev->dev, dev->name);
1028 memcpy(dev->name, oldname, IFNAMSIZ);
1032 write_lock_bh(&dev_base_lock);
1033 hlist_del_rcu(&dev->name_hlist);
1034 write_unlock_bh(&dev_base_lock);
1038 write_lock_bh(&dev_base_lock);
1039 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1040 write_unlock_bh(&dev_base_lock);
1042 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1043 ret = notifier_to_errno(ret);
1046 /* err >= 0 after dev_alloc_name() or stores the first errno */
1049 memcpy(dev->name, oldname, IFNAMSIZ);
1052 pr_err("%s: name change rollback failed: %d\n",
1061 * dev_set_alias - change ifalias of a device
1063 * @alias: name up to IFALIASZ
1064 * @len: limit of bytes to copy from info
1066 * Set ifalias for a device,
1068 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1074 if (len >= IFALIASZ)
1079 kfree(dev->ifalias);
1080 dev->ifalias = NULL;
1085 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1088 dev->ifalias = new_ifalias;
1090 strlcpy(dev->ifalias, alias, len+1);
1096 * netdev_features_change - device changes features
1097 * @dev: device to cause notification
1099 * Called to indicate a device has changed features.
1101 void netdev_features_change(struct net_device *dev)
1103 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1105 EXPORT_SYMBOL(netdev_features_change);
1108 * netdev_state_change - device changes state
1109 * @dev: device to cause notification
1111 * Called to indicate a device has changed state. This function calls
1112 * the notifier chains for netdev_chain and sends a NEWLINK message
1113 * to the routing socket.
1115 void netdev_state_change(struct net_device *dev)
1117 if (dev->flags & IFF_UP) {
1118 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1119 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1122 EXPORT_SYMBOL(netdev_state_change);
1125 * netdev_notify_peers - notify network peers about existence of @dev
1126 * @dev: network device
1128 * Generate traffic such that interested network peers are aware of
1129 * @dev, such as by generating a gratuitous ARP. This may be used when
1130 * a device wants to inform the rest of the network about some sort of
1131 * reconfiguration such as a failover event or virtual machine
1134 void netdev_notify_peers(struct net_device *dev)
1137 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1140 EXPORT_SYMBOL(netdev_notify_peers);
1143 * dev_load - load a network module
1144 * @net: the applicable net namespace
1145 * @name: name of interface
1147 * If a network interface is not present and the process has suitable
1148 * privileges this function loads the module. If module loading is not
1149 * available in this kernel then it becomes a nop.
1152 void dev_load(struct net *net, const char *name)
1154 struct net_device *dev;
1158 dev = dev_get_by_name_rcu(net, name);
1162 if (no_module && capable(CAP_NET_ADMIN))
1163 no_module = request_module("netdev-%s", name);
1164 if (no_module && capable(CAP_SYS_MODULE)) {
1165 if (!request_module("%s", name))
1166 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1170 EXPORT_SYMBOL(dev_load);
1172 static int __dev_open(struct net_device *dev)
1174 const struct net_device_ops *ops = dev->netdev_ops;
1179 if (!netif_device_present(dev))
1182 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1183 ret = notifier_to_errno(ret);
1187 set_bit(__LINK_STATE_START, &dev->state);
1189 if (ops->ndo_validate_addr)
1190 ret = ops->ndo_validate_addr(dev);
1192 if (!ret && ops->ndo_open)
1193 ret = ops->ndo_open(dev);
1196 clear_bit(__LINK_STATE_START, &dev->state);
1198 dev->flags |= IFF_UP;
1199 net_dmaengine_get();
1200 dev_set_rx_mode(dev);
1202 add_device_randomness(dev->dev_addr, dev->addr_len);
1209 * dev_open - prepare an interface for use.
1210 * @dev: device to open
1212 * Takes a device from down to up state. The device's private open
1213 * function is invoked and then the multicast lists are loaded. Finally
1214 * the device is moved into the up state and a %NETDEV_UP message is
1215 * sent to the netdev notifier chain.
1217 * Calling this function on an active interface is a nop. On a failure
1218 * a negative errno code is returned.
1220 int dev_open(struct net_device *dev)
1224 if (dev->flags & IFF_UP)
1227 ret = __dev_open(dev);
1231 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1232 call_netdevice_notifiers(NETDEV_UP, dev);
1236 EXPORT_SYMBOL(dev_open);
1238 static int __dev_close_many(struct list_head *head)
1240 struct net_device *dev;
1245 list_for_each_entry(dev, head, unreg_list) {
1246 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1248 clear_bit(__LINK_STATE_START, &dev->state);
1250 /* Synchronize to scheduled poll. We cannot touch poll list, it
1251 * can be even on different cpu. So just clear netif_running().
1253 * dev->stop() will invoke napi_disable() on all of it's
1254 * napi_struct instances on this device.
1256 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1259 dev_deactivate_many(head);
1261 list_for_each_entry(dev, head, unreg_list) {
1262 const struct net_device_ops *ops = dev->netdev_ops;
1265 * Call the device specific close. This cannot fail.
1266 * Only if device is UP
1268 * We allow it to be called even after a DETACH hot-plug
1274 dev->flags &= ~IFF_UP;
1275 net_dmaengine_put();
1281 static int __dev_close(struct net_device *dev)
1286 list_add(&dev->unreg_list, &single);
1287 retval = __dev_close_many(&single);
1292 static int dev_close_many(struct list_head *head)
1294 struct net_device *dev, *tmp;
1295 LIST_HEAD(tmp_list);
1297 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1298 if (!(dev->flags & IFF_UP))
1299 list_move(&dev->unreg_list, &tmp_list);
1301 __dev_close_many(head);
1303 list_for_each_entry(dev, head, unreg_list) {
1304 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1305 call_netdevice_notifiers(NETDEV_DOWN, dev);
1308 /* rollback_registered_many needs the complete original list */
1309 list_splice(&tmp_list, head);
1314 * dev_close - shutdown an interface.
1315 * @dev: device to shutdown
1317 * This function moves an active device into down state. A
1318 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1319 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1322 int dev_close(struct net_device *dev)
1324 if (dev->flags & IFF_UP) {
1327 list_add(&dev->unreg_list, &single);
1328 dev_close_many(&single);
1333 EXPORT_SYMBOL(dev_close);
1337 * dev_disable_lro - disable Large Receive Offload on a device
1340 * Disable Large Receive Offload (LRO) on a net device. Must be
1341 * called under RTNL. This is needed if received packets may be
1342 * forwarded to another interface.
1344 void dev_disable_lro(struct net_device *dev)
1347 * If we're trying to disable lro on a vlan device
1348 * use the underlying physical device instead
1350 if (is_vlan_dev(dev))
1351 dev = vlan_dev_real_dev(dev);
1353 dev->wanted_features &= ~NETIF_F_LRO;
1354 netdev_update_features(dev);
1356 if (unlikely(dev->features & NETIF_F_LRO))
1357 netdev_WARN(dev, "failed to disable LRO!\n");
1359 EXPORT_SYMBOL(dev_disable_lro);
1362 static int dev_boot_phase = 1;
1365 * register_netdevice_notifier - register a network notifier block
1368 * Register a notifier to be called when network device events occur.
1369 * The notifier passed is linked into the kernel structures and must
1370 * not be reused until it has been unregistered. A negative errno code
1371 * is returned on a failure.
1373 * When registered all registration and up events are replayed
1374 * to the new notifier to allow device to have a race free
1375 * view of the network device list.
1378 int register_netdevice_notifier(struct notifier_block *nb)
1380 struct net_device *dev;
1381 struct net_device *last;
1386 err = raw_notifier_chain_register(&netdev_chain, nb);
1392 for_each_netdev(net, dev) {
1393 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1394 err = notifier_to_errno(err);
1398 if (!(dev->flags & IFF_UP))
1401 nb->notifier_call(nb, NETDEV_UP, dev);
1412 for_each_netdev(net, dev) {
1416 if (dev->flags & IFF_UP) {
1417 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1418 nb->notifier_call(nb, NETDEV_DOWN, dev);
1420 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1425 raw_notifier_chain_unregister(&netdev_chain, nb);
1428 EXPORT_SYMBOL(register_netdevice_notifier);
1431 * unregister_netdevice_notifier - unregister a network notifier block
1434 * Unregister a notifier previously registered by
1435 * register_netdevice_notifier(). The notifier is unlinked into the
1436 * kernel structures and may then be reused. A negative errno code
1437 * is returned on a failure.
1439 * After unregistering unregister and down device events are synthesized
1440 * for all devices on the device list to the removed notifier to remove
1441 * the need for special case cleanup code.
1444 int unregister_netdevice_notifier(struct notifier_block *nb)
1446 struct net_device *dev;
1451 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1456 for_each_netdev(net, dev) {
1457 if (dev->flags & IFF_UP) {
1458 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1459 nb->notifier_call(nb, NETDEV_DOWN, dev);
1461 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1468 EXPORT_SYMBOL(unregister_netdevice_notifier);
1471 * call_netdevice_notifiers - call all network notifier blocks
1472 * @val: value passed unmodified to notifier function
1473 * @dev: net_device pointer passed unmodified to notifier function
1475 * Call all network notifier blocks. Parameters and return value
1476 * are as for raw_notifier_call_chain().
1479 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1482 return raw_notifier_call_chain(&netdev_chain, val, dev);
1484 EXPORT_SYMBOL(call_netdevice_notifiers);
1486 static struct static_key netstamp_needed __read_mostly;
1487 #ifdef HAVE_JUMP_LABEL
1488 /* We are not allowed to call static_key_slow_dec() from irq context
1489 * If net_disable_timestamp() is called from irq context, defer the
1490 * static_key_slow_dec() calls.
1492 static atomic_t netstamp_needed_deferred;
1495 void net_enable_timestamp(void)
1497 #ifdef HAVE_JUMP_LABEL
1498 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1502 static_key_slow_dec(&netstamp_needed);
1506 WARN_ON(in_interrupt());
1507 static_key_slow_inc(&netstamp_needed);
1509 EXPORT_SYMBOL(net_enable_timestamp);
1511 void net_disable_timestamp(void)
1513 #ifdef HAVE_JUMP_LABEL
1514 if (in_interrupt()) {
1515 atomic_inc(&netstamp_needed_deferred);
1519 static_key_slow_dec(&netstamp_needed);
1521 EXPORT_SYMBOL(net_disable_timestamp);
1523 static inline void net_timestamp_set(struct sk_buff *skb)
1525 skb->tstamp.tv64 = 0;
1526 if (static_key_false(&netstamp_needed))
1527 __net_timestamp(skb);
1530 #define net_timestamp_check(COND, SKB) \
1531 if (static_key_false(&netstamp_needed)) { \
1532 if ((COND) && !(SKB)->tstamp.tv64) \
1533 __net_timestamp(SKB); \
1536 static int net_hwtstamp_validate(struct ifreq *ifr)
1538 struct hwtstamp_config cfg;
1539 enum hwtstamp_tx_types tx_type;
1540 enum hwtstamp_rx_filters rx_filter;
1541 int tx_type_valid = 0;
1542 int rx_filter_valid = 0;
1544 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1547 if (cfg.flags) /* reserved for future extensions */
1550 tx_type = cfg.tx_type;
1551 rx_filter = cfg.rx_filter;
1554 case HWTSTAMP_TX_OFF:
1555 case HWTSTAMP_TX_ON:
1556 case HWTSTAMP_TX_ONESTEP_SYNC:
1561 switch (rx_filter) {
1562 case HWTSTAMP_FILTER_NONE:
1563 case HWTSTAMP_FILTER_ALL:
1564 case HWTSTAMP_FILTER_SOME:
1565 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1566 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1567 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1568 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1569 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1570 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1571 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1572 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1573 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1574 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1575 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1576 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1577 rx_filter_valid = 1;
1581 if (!tx_type_valid || !rx_filter_valid)
1587 static inline bool is_skb_forwardable(struct net_device *dev,
1588 struct sk_buff *skb)
1592 if (!(dev->flags & IFF_UP))
1595 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1596 if (skb->len <= len)
1599 /* if TSO is enabled, we don't care about the length as the packet
1600 * could be forwarded without being segmented before
1602 if (skb_is_gso(skb))
1609 * dev_forward_skb - loopback an skb to another netif
1611 * @dev: destination network device
1612 * @skb: buffer to forward
1615 * NET_RX_SUCCESS (no congestion)
1616 * NET_RX_DROP (packet was dropped, but freed)
1618 * dev_forward_skb can be used for injecting an skb from the
1619 * start_xmit function of one device into the receive queue
1620 * of another device.
1622 * The receiving device may be in another namespace, so
1623 * we have to clear all information in the skb that could
1624 * impact namespace isolation.
1626 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1628 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1629 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1630 atomic_long_inc(&dev->rx_dropped);
1639 if (unlikely(!is_skb_forwardable(dev, skb))) {
1640 atomic_long_inc(&dev->rx_dropped);
1647 skb->tstamp.tv64 = 0;
1648 skb->pkt_type = PACKET_HOST;
1649 skb->protocol = eth_type_trans(skb, dev);
1653 return netif_rx(skb);
1655 EXPORT_SYMBOL_GPL(dev_forward_skb);
1657 static inline int deliver_skb(struct sk_buff *skb,
1658 struct packet_type *pt_prev,
1659 struct net_device *orig_dev)
1661 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1663 atomic_inc(&skb->users);
1664 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1667 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1669 if (ptype->af_packet_priv == NULL)
1672 if (ptype->id_match)
1673 return ptype->id_match(ptype, skb->sk);
1674 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1681 * Support routine. Sends outgoing frames to any network
1682 * taps currently in use.
1685 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1687 struct packet_type *ptype;
1688 struct sk_buff *skb2 = NULL;
1689 struct packet_type *pt_prev = NULL;
1692 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1693 /* Never send packets back to the socket
1694 * they originated from - MvS (miquels@drinkel.ow.org)
1696 if ((ptype->dev == dev || !ptype->dev) &&
1697 (!skb_loop_sk(ptype, skb))) {
1699 deliver_skb(skb2, pt_prev, skb->dev);
1704 skb2 = skb_clone(skb, GFP_ATOMIC);
1708 net_timestamp_set(skb2);
1710 /* skb->nh should be correctly
1711 set by sender, so that the second statement is
1712 just protection against buggy protocols.
1714 skb_reset_mac_header(skb2);
1716 if (skb_network_header(skb2) < skb2->data ||
1717 skb2->network_header > skb2->tail) {
1718 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1719 ntohs(skb2->protocol),
1721 skb_reset_network_header(skb2);
1724 skb2->transport_header = skb2->network_header;
1725 skb2->pkt_type = PACKET_OUTGOING;
1730 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1735 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1736 * @dev: Network device
1737 * @txq: number of queues available
1739 * If real_num_tx_queues is changed the tc mappings may no longer be
1740 * valid. To resolve this verify the tc mapping remains valid and if
1741 * not NULL the mapping. With no priorities mapping to this
1742 * offset/count pair it will no longer be used. In the worst case TC0
1743 * is invalid nothing can be done so disable priority mappings. If is
1744 * expected that drivers will fix this mapping if they can before
1745 * calling netif_set_real_num_tx_queues.
1747 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1750 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1752 /* If TC0 is invalidated disable TC mapping */
1753 if (tc->offset + tc->count > txq) {
1754 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1759 /* Invalidated prio to tc mappings set to TC0 */
1760 for (i = 1; i < TC_BITMASK + 1; i++) {
1761 int q = netdev_get_prio_tc_map(dev, i);
1763 tc = &dev->tc_to_txq[q];
1764 if (tc->offset + tc->count > txq) {
1765 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1767 netdev_set_prio_tc_map(dev, i, 0);
1773 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1774 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1776 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1780 if (txq < 1 || txq > dev->num_tx_queues)
1783 if (dev->reg_state == NETREG_REGISTERED ||
1784 dev->reg_state == NETREG_UNREGISTERING) {
1787 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1793 netif_setup_tc(dev, txq);
1795 if (txq < dev->real_num_tx_queues)
1796 qdisc_reset_all_tx_gt(dev, txq);
1799 dev->real_num_tx_queues = txq;
1802 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1806 * netif_set_real_num_rx_queues - set actual number of RX queues used
1807 * @dev: Network device
1808 * @rxq: Actual number of RX queues
1810 * This must be called either with the rtnl_lock held or before
1811 * registration of the net device. Returns 0 on success, or a
1812 * negative error code. If called before registration, it always
1815 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1819 if (rxq < 1 || rxq > dev->num_rx_queues)
1822 if (dev->reg_state == NETREG_REGISTERED) {
1825 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1831 dev->real_num_rx_queues = rxq;
1834 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1838 * netif_get_num_default_rss_queues - default number of RSS queues
1840 * This routine should set an upper limit on the number of RSS queues
1841 * used by default by multiqueue devices.
1843 int netif_get_num_default_rss_queues(void)
1845 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
1847 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
1849 static inline void __netif_reschedule(struct Qdisc *q)
1851 struct softnet_data *sd;
1852 unsigned long flags;
1854 local_irq_save(flags);
1855 sd = &__get_cpu_var(softnet_data);
1856 q->next_sched = NULL;
1857 *sd->output_queue_tailp = q;
1858 sd->output_queue_tailp = &q->next_sched;
1859 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1860 local_irq_restore(flags);
1863 void __netif_schedule(struct Qdisc *q)
1865 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1866 __netif_reschedule(q);
1868 EXPORT_SYMBOL(__netif_schedule);
1870 void dev_kfree_skb_irq(struct sk_buff *skb)
1872 if (atomic_dec_and_test(&skb->users)) {
1873 struct softnet_data *sd;
1874 unsigned long flags;
1876 local_irq_save(flags);
1877 sd = &__get_cpu_var(softnet_data);
1878 skb->next = sd->completion_queue;
1879 sd->completion_queue = skb;
1880 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1881 local_irq_restore(flags);
1884 EXPORT_SYMBOL(dev_kfree_skb_irq);
1886 void dev_kfree_skb_any(struct sk_buff *skb)
1888 if (in_irq() || irqs_disabled())
1889 dev_kfree_skb_irq(skb);
1893 EXPORT_SYMBOL(dev_kfree_skb_any);
1897 * netif_device_detach - mark device as removed
1898 * @dev: network device
1900 * Mark device as removed from system and therefore no longer available.
1902 void netif_device_detach(struct net_device *dev)
1904 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1905 netif_running(dev)) {
1906 netif_tx_stop_all_queues(dev);
1909 EXPORT_SYMBOL(netif_device_detach);
1912 * netif_device_attach - mark device as attached
1913 * @dev: network device
1915 * Mark device as attached from system and restart if needed.
1917 void netif_device_attach(struct net_device *dev)
1919 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1920 netif_running(dev)) {
1921 netif_tx_wake_all_queues(dev);
1922 __netdev_watchdog_up(dev);
1925 EXPORT_SYMBOL(netif_device_attach);
1927 static void skb_warn_bad_offload(const struct sk_buff *skb)
1929 static const netdev_features_t null_features = 0;
1930 struct net_device *dev = skb->dev;
1931 const char *driver = "";
1933 if (dev && dev->dev.parent)
1934 driver = dev_driver_string(dev->dev.parent);
1936 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1937 "gso_type=%d ip_summed=%d\n",
1938 driver, dev ? &dev->features : &null_features,
1939 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1940 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1941 skb_shinfo(skb)->gso_type, skb->ip_summed);
1945 * Invalidate hardware checksum when packet is to be mangled, and
1946 * complete checksum manually on outgoing path.
1948 int skb_checksum_help(struct sk_buff *skb)
1951 int ret = 0, offset;
1953 if (skb->ip_summed == CHECKSUM_COMPLETE)
1954 goto out_set_summed;
1956 if (unlikely(skb_shinfo(skb)->gso_size)) {
1957 skb_warn_bad_offload(skb);
1961 offset = skb_checksum_start_offset(skb);
1962 BUG_ON(offset >= skb_headlen(skb));
1963 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1965 offset += skb->csum_offset;
1966 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1968 if (skb_cloned(skb) &&
1969 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1970 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1975 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1977 skb->ip_summed = CHECKSUM_NONE;
1981 EXPORT_SYMBOL(skb_checksum_help);
1984 * skb_gso_segment - Perform segmentation on skb.
1985 * @skb: buffer to segment
1986 * @features: features for the output path (see dev->features)
1988 * This function segments the given skb and returns a list of segments.
1990 * It may return NULL if the skb requires no segmentation. This is
1991 * only possible when GSO is used for verifying header integrity.
1993 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1994 netdev_features_t features)
1996 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1997 struct packet_type *ptype;
1998 __be16 type = skb->protocol;
1999 int vlan_depth = ETH_HLEN;
2002 while (type == htons(ETH_P_8021Q)) {
2003 struct vlan_hdr *vh;
2005 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2006 return ERR_PTR(-EINVAL);
2008 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2009 type = vh->h_vlan_encapsulated_proto;
2010 vlan_depth += VLAN_HLEN;
2013 skb_reset_mac_header(skb);
2014 skb->mac_len = skb->network_header - skb->mac_header;
2015 __skb_pull(skb, skb->mac_len);
2017 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2018 skb_warn_bad_offload(skb);
2020 if (skb_header_cloned(skb) &&
2021 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2022 return ERR_PTR(err);
2026 list_for_each_entry_rcu(ptype,
2027 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2028 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
2029 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2030 err = ptype->gso_send_check(skb);
2031 segs = ERR_PTR(err);
2032 if (err || skb_gso_ok(skb, features))
2034 __skb_push(skb, (skb->data -
2035 skb_network_header(skb)));
2037 segs = ptype->gso_segment(skb, features);
2043 __skb_push(skb, skb->data - skb_mac_header(skb));
2047 EXPORT_SYMBOL(skb_gso_segment);
2049 /* Take action when hardware reception checksum errors are detected. */
2051 void netdev_rx_csum_fault(struct net_device *dev)
2053 if (net_ratelimit()) {
2054 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2058 EXPORT_SYMBOL(netdev_rx_csum_fault);
2061 /* Actually, we should eliminate this check as soon as we know, that:
2062 * 1. IOMMU is present and allows to map all the memory.
2063 * 2. No high memory really exists on this machine.
2066 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2068 #ifdef CONFIG_HIGHMEM
2070 if (!(dev->features & NETIF_F_HIGHDMA)) {
2071 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2072 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2073 if (PageHighMem(skb_frag_page(frag)))
2078 if (PCI_DMA_BUS_IS_PHYS) {
2079 struct device *pdev = dev->dev.parent;
2083 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2084 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2085 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2086 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2095 void (*destructor)(struct sk_buff *skb);
2098 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2100 static void dev_gso_skb_destructor(struct sk_buff *skb)
2102 struct dev_gso_cb *cb;
2105 struct sk_buff *nskb = skb->next;
2107 skb->next = nskb->next;
2110 } while (skb->next);
2112 cb = DEV_GSO_CB(skb);
2114 cb->destructor(skb);
2118 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2119 * @skb: buffer to segment
2120 * @features: device features as applicable to this skb
2122 * This function segments the given skb and stores the list of segments
2125 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2127 struct sk_buff *segs;
2129 segs = skb_gso_segment(skb, features);
2131 /* Verifying header integrity only. */
2136 return PTR_ERR(segs);
2139 DEV_GSO_CB(skb)->destructor = skb->destructor;
2140 skb->destructor = dev_gso_skb_destructor;
2145 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2147 return ((features & NETIF_F_GEN_CSUM) ||
2148 ((features & NETIF_F_V4_CSUM) &&
2149 protocol == htons(ETH_P_IP)) ||
2150 ((features & NETIF_F_V6_CSUM) &&
2151 protocol == htons(ETH_P_IPV6)) ||
2152 ((features & NETIF_F_FCOE_CRC) &&
2153 protocol == htons(ETH_P_FCOE)));
2156 static netdev_features_t harmonize_features(struct sk_buff *skb,
2157 __be16 protocol, netdev_features_t features)
2159 if (skb->ip_summed != CHECKSUM_NONE &&
2160 !can_checksum_protocol(features, protocol)) {
2161 features &= ~NETIF_F_ALL_CSUM;
2162 features &= ~NETIF_F_SG;
2163 } else if (illegal_highdma(skb->dev, skb)) {
2164 features &= ~NETIF_F_SG;
2170 netdev_features_t netif_skb_features(struct sk_buff *skb)
2172 __be16 protocol = skb->protocol;
2173 netdev_features_t features = skb->dev->features;
2175 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2176 features &= ~NETIF_F_GSO_MASK;
2178 if (protocol == htons(ETH_P_8021Q)) {
2179 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2180 protocol = veh->h_vlan_encapsulated_proto;
2181 } else if (!vlan_tx_tag_present(skb)) {
2182 return harmonize_features(skb, protocol, features);
2185 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2187 if (protocol != htons(ETH_P_8021Q)) {
2188 return harmonize_features(skb, protocol, features);
2190 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2191 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2192 return harmonize_features(skb, protocol, features);
2195 EXPORT_SYMBOL(netif_skb_features);
2198 * Returns true if either:
2199 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2200 * 2. skb is fragmented and the device does not support SG.
2202 static inline int skb_needs_linearize(struct sk_buff *skb,
2205 return skb_is_nonlinear(skb) &&
2206 ((skb_has_frag_list(skb) &&
2207 !(features & NETIF_F_FRAGLIST)) ||
2208 (skb_shinfo(skb)->nr_frags &&
2209 !(features & NETIF_F_SG)));
2212 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2213 struct netdev_queue *txq)
2215 const struct net_device_ops *ops = dev->netdev_ops;
2216 int rc = NETDEV_TX_OK;
2217 unsigned int skb_len;
2219 if (likely(!skb->next)) {
2220 netdev_features_t features;
2223 * If device doesn't need skb->dst, release it right now while
2224 * its hot in this cpu cache
2226 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2229 features = netif_skb_features(skb);
2231 if (vlan_tx_tag_present(skb) &&
2232 !(features & NETIF_F_HW_VLAN_TX)) {
2233 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2240 if (netif_needs_gso(skb, features)) {
2241 if (unlikely(dev_gso_segment(skb, features)))
2246 if (skb_needs_linearize(skb, features) &&
2247 __skb_linearize(skb))
2250 /* If packet is not checksummed and device does not
2251 * support checksumming for this protocol, complete
2252 * checksumming here.
2254 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2255 skb_set_transport_header(skb,
2256 skb_checksum_start_offset(skb));
2257 if (!(features & NETIF_F_ALL_CSUM) &&
2258 skb_checksum_help(skb))
2263 if (!list_empty(&ptype_all))
2264 dev_queue_xmit_nit(skb, dev);
2267 rc = ops->ndo_start_xmit(skb, dev);
2268 trace_net_dev_xmit(skb, rc, dev, skb_len);
2269 if (rc == NETDEV_TX_OK)
2270 txq_trans_update(txq);
2276 struct sk_buff *nskb = skb->next;
2278 skb->next = nskb->next;
2282 * If device doesn't need nskb->dst, release it right now while
2283 * its hot in this cpu cache
2285 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2288 if (!list_empty(&ptype_all))
2289 dev_queue_xmit_nit(nskb, dev);
2291 skb_len = nskb->len;
2292 rc = ops->ndo_start_xmit(nskb, dev);
2293 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2294 if (unlikely(rc != NETDEV_TX_OK)) {
2295 if (rc & ~NETDEV_TX_MASK)
2296 goto out_kfree_gso_skb;
2297 nskb->next = skb->next;
2301 txq_trans_update(txq);
2302 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2303 return NETDEV_TX_BUSY;
2304 } while (skb->next);
2307 if (likely(skb->next == NULL))
2308 skb->destructor = DEV_GSO_CB(skb)->destructor;
2315 static u32 hashrnd __read_mostly;
2318 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2319 * to be used as a distribution range.
2321 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2322 unsigned int num_tx_queues)
2326 u16 qcount = num_tx_queues;
2328 if (skb_rx_queue_recorded(skb)) {
2329 hash = skb_get_rx_queue(skb);
2330 while (unlikely(hash >= num_tx_queues))
2331 hash -= num_tx_queues;
2336 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2337 qoffset = dev->tc_to_txq[tc].offset;
2338 qcount = dev->tc_to_txq[tc].count;
2341 if (skb->sk && skb->sk->sk_hash)
2342 hash = skb->sk->sk_hash;
2344 hash = (__force u16) skb->protocol;
2345 hash = jhash_1word(hash, hashrnd);
2347 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2349 EXPORT_SYMBOL(__skb_tx_hash);
2351 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2353 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2354 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2355 dev->name, queue_index,
2356 dev->real_num_tx_queues);
2362 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2365 struct xps_dev_maps *dev_maps;
2366 struct xps_map *map;
2367 int queue_index = -1;
2370 dev_maps = rcu_dereference(dev->xps_maps);
2372 map = rcu_dereference(
2373 dev_maps->cpu_map[raw_smp_processor_id()]);
2376 queue_index = map->queues[0];
2379 if (skb->sk && skb->sk->sk_hash)
2380 hash = skb->sk->sk_hash;
2382 hash = (__force u16) skb->protocol ^
2384 hash = jhash_1word(hash, hashrnd);
2385 queue_index = map->queues[
2386 ((u64)hash * map->len) >> 32];
2388 if (unlikely(queue_index >= dev->real_num_tx_queues))
2400 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
2401 struct sk_buff *skb)
2404 const struct net_device_ops *ops = dev->netdev_ops;
2406 if (dev->real_num_tx_queues == 1)
2408 else if (ops->ndo_select_queue) {
2409 queue_index = ops->ndo_select_queue(dev, skb);
2410 queue_index = dev_cap_txqueue(dev, queue_index);
2412 struct sock *sk = skb->sk;
2413 queue_index = sk_tx_queue_get(sk);
2415 if (queue_index < 0 || skb->ooo_okay ||
2416 queue_index >= dev->real_num_tx_queues) {
2417 int old_index = queue_index;
2419 queue_index = get_xps_queue(dev, skb);
2420 if (queue_index < 0)
2421 queue_index = skb_tx_hash(dev, skb);
2423 if (queue_index != old_index && sk) {
2424 struct dst_entry *dst =
2425 rcu_dereference_check(sk->sk_dst_cache, 1);
2427 if (dst && skb_dst(skb) == dst)
2428 sk_tx_queue_set(sk, queue_index);
2433 skb_set_queue_mapping(skb, queue_index);
2434 return netdev_get_tx_queue(dev, queue_index);
2437 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2438 struct net_device *dev,
2439 struct netdev_queue *txq)
2441 spinlock_t *root_lock = qdisc_lock(q);
2445 qdisc_skb_cb(skb)->pkt_len = skb->len;
2446 qdisc_calculate_pkt_len(skb, q);
2448 * Heuristic to force contended enqueues to serialize on a
2449 * separate lock before trying to get qdisc main lock.
2450 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2451 * and dequeue packets faster.
2453 contended = qdisc_is_running(q);
2454 if (unlikely(contended))
2455 spin_lock(&q->busylock);
2457 spin_lock(root_lock);
2458 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2461 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2462 qdisc_run_begin(q)) {
2464 * This is a work-conserving queue; there are no old skbs
2465 * waiting to be sent out; and the qdisc is not running -
2466 * xmit the skb directly.
2468 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2471 qdisc_bstats_update(q, skb);
2473 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2474 if (unlikely(contended)) {
2475 spin_unlock(&q->busylock);
2482 rc = NET_XMIT_SUCCESS;
2485 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2486 if (qdisc_run_begin(q)) {
2487 if (unlikely(contended)) {
2488 spin_unlock(&q->busylock);
2494 spin_unlock(root_lock);
2495 if (unlikely(contended))
2496 spin_unlock(&q->busylock);
2500 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2501 static void skb_update_prio(struct sk_buff *skb)
2503 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2505 if (!skb->priority && skb->sk && map) {
2506 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2508 if (prioidx < map->priomap_len)
2509 skb->priority = map->priomap[prioidx];
2513 #define skb_update_prio(skb)
2516 static DEFINE_PER_CPU(int, xmit_recursion);
2517 #define RECURSION_LIMIT 10
2520 * dev_loopback_xmit - loop back @skb
2521 * @skb: buffer to transmit
2523 int dev_loopback_xmit(struct sk_buff *skb)
2525 skb_reset_mac_header(skb);
2526 __skb_pull(skb, skb_network_offset(skb));
2527 skb->pkt_type = PACKET_LOOPBACK;
2528 skb->ip_summed = CHECKSUM_UNNECESSARY;
2529 WARN_ON(!skb_dst(skb));
2534 EXPORT_SYMBOL(dev_loopback_xmit);
2537 * dev_queue_xmit - transmit a buffer
2538 * @skb: buffer to transmit
2540 * Queue a buffer for transmission to a network device. The caller must
2541 * have set the device and priority and built the buffer before calling
2542 * this function. The function can be called from an interrupt.
2544 * A negative errno code is returned on a failure. A success does not
2545 * guarantee the frame will be transmitted as it may be dropped due
2546 * to congestion or traffic shaping.
2548 * -----------------------------------------------------------------------------------
2549 * I notice this method can also return errors from the queue disciplines,
2550 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2553 * Regardless of the return value, the skb is consumed, so it is currently
2554 * difficult to retry a send to this method. (You can bump the ref count
2555 * before sending to hold a reference for retry if you are careful.)
2557 * When calling this method, interrupts MUST be enabled. This is because
2558 * the BH enable code must have IRQs enabled so that it will not deadlock.
2561 int dev_queue_xmit(struct sk_buff *skb)
2563 struct net_device *dev = skb->dev;
2564 struct netdev_queue *txq;
2568 /* Disable soft irqs for various locks below. Also
2569 * stops preemption for RCU.
2573 skb_update_prio(skb);
2575 txq = netdev_pick_tx(dev, skb);
2576 q = rcu_dereference_bh(txq->qdisc);
2578 #ifdef CONFIG_NET_CLS_ACT
2579 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2581 trace_net_dev_queue(skb);
2583 rc = __dev_xmit_skb(skb, q, dev, txq);
2587 /* The device has no queue. Common case for software devices:
2588 loopback, all the sorts of tunnels...
2590 Really, it is unlikely that netif_tx_lock protection is necessary
2591 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2593 However, it is possible, that they rely on protection
2596 Check this and shot the lock. It is not prone from deadlocks.
2597 Either shot noqueue qdisc, it is even simpler 8)
2599 if (dev->flags & IFF_UP) {
2600 int cpu = smp_processor_id(); /* ok because BHs are off */
2602 if (txq->xmit_lock_owner != cpu) {
2604 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2605 goto recursion_alert;
2607 HARD_TX_LOCK(dev, txq, cpu);
2609 if (!netif_xmit_stopped(txq)) {
2610 __this_cpu_inc(xmit_recursion);
2611 rc = dev_hard_start_xmit(skb, dev, txq);
2612 __this_cpu_dec(xmit_recursion);
2613 if (dev_xmit_complete(rc)) {
2614 HARD_TX_UNLOCK(dev, txq);
2618 HARD_TX_UNLOCK(dev, txq);
2619 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2622 /* Recursion is detected! It is possible,
2626 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2632 rcu_read_unlock_bh();
2637 rcu_read_unlock_bh();
2640 EXPORT_SYMBOL(dev_queue_xmit);
2643 /*=======================================================================
2645 =======================================================================*/
2647 int netdev_max_backlog __read_mostly = 1000;
2648 EXPORT_SYMBOL(netdev_max_backlog);
2650 int netdev_tstamp_prequeue __read_mostly = 1;
2651 int netdev_budget __read_mostly = 300;
2652 int weight_p __read_mostly = 64; /* old backlog weight */
2654 /* Called with irq disabled */
2655 static inline void ____napi_schedule(struct softnet_data *sd,
2656 struct napi_struct *napi)
2658 list_add_tail(&napi->poll_list, &sd->poll_list);
2659 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2663 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2664 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2665 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2666 * if hash is a canonical 4-tuple hash over transport ports.
2668 void __skb_get_rxhash(struct sk_buff *skb)
2670 struct flow_keys keys;
2673 if (!skb_flow_dissect(skb, &keys))
2679 /* get a consistent hash (same value on both flow directions) */
2680 if (((__force u32)keys.dst < (__force u32)keys.src) ||
2681 (((__force u32)keys.dst == (__force u32)keys.src) &&
2682 ((__force u16)keys.port16[1] < (__force u16)keys.port16[0]))) {
2683 swap(keys.dst, keys.src);
2684 swap(keys.port16[0], keys.port16[1]);
2687 hash = jhash_3words((__force u32)keys.dst,
2688 (__force u32)keys.src,
2689 (__force u32)keys.ports, hashrnd);
2695 EXPORT_SYMBOL(__skb_get_rxhash);
2699 /* One global table that all flow-based protocols share. */
2700 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2701 EXPORT_SYMBOL(rps_sock_flow_table);
2703 struct static_key rps_needed __read_mostly;
2705 static struct rps_dev_flow *
2706 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2707 struct rps_dev_flow *rflow, u16 next_cpu)
2709 if (next_cpu != RPS_NO_CPU) {
2710 #ifdef CONFIG_RFS_ACCEL
2711 struct netdev_rx_queue *rxqueue;
2712 struct rps_dev_flow_table *flow_table;
2713 struct rps_dev_flow *old_rflow;
2718 /* Should we steer this flow to a different hardware queue? */
2719 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2720 !(dev->features & NETIF_F_NTUPLE))
2722 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2723 if (rxq_index == skb_get_rx_queue(skb))
2726 rxqueue = dev->_rx + rxq_index;
2727 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2730 flow_id = skb->rxhash & flow_table->mask;
2731 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2732 rxq_index, flow_id);
2736 rflow = &flow_table->flows[flow_id];
2738 if (old_rflow->filter == rflow->filter)
2739 old_rflow->filter = RPS_NO_FILTER;
2743 per_cpu(softnet_data, next_cpu).input_queue_head;
2746 rflow->cpu = next_cpu;
2751 * get_rps_cpu is called from netif_receive_skb and returns the target
2752 * CPU from the RPS map of the receiving queue for a given skb.
2753 * rcu_read_lock must be held on entry.
2755 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2756 struct rps_dev_flow **rflowp)
2758 struct netdev_rx_queue *rxqueue;
2759 struct rps_map *map;
2760 struct rps_dev_flow_table *flow_table;
2761 struct rps_sock_flow_table *sock_flow_table;
2765 if (skb_rx_queue_recorded(skb)) {
2766 u16 index = skb_get_rx_queue(skb);
2767 if (unlikely(index >= dev->real_num_rx_queues)) {
2768 WARN_ONCE(dev->real_num_rx_queues > 1,
2769 "%s received packet on queue %u, but number "
2770 "of RX queues is %u\n",
2771 dev->name, index, dev->real_num_rx_queues);
2774 rxqueue = dev->_rx + index;
2778 map = rcu_dereference(rxqueue->rps_map);
2780 if (map->len == 1 &&
2781 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2782 tcpu = map->cpus[0];
2783 if (cpu_online(tcpu))
2787 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2791 skb_reset_network_header(skb);
2792 if (!skb_get_rxhash(skb))
2795 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2796 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2797 if (flow_table && sock_flow_table) {
2799 struct rps_dev_flow *rflow;
2801 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2804 next_cpu = sock_flow_table->ents[skb->rxhash &
2805 sock_flow_table->mask];
2808 * If the desired CPU (where last recvmsg was done) is
2809 * different from current CPU (one in the rx-queue flow
2810 * table entry), switch if one of the following holds:
2811 * - Current CPU is unset (equal to RPS_NO_CPU).
2812 * - Current CPU is offline.
2813 * - The current CPU's queue tail has advanced beyond the
2814 * last packet that was enqueued using this table entry.
2815 * This guarantees that all previous packets for the flow
2816 * have been dequeued, thus preserving in order delivery.
2818 if (unlikely(tcpu != next_cpu) &&
2819 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2820 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2821 rflow->last_qtail)) >= 0))
2822 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2824 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2832 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2834 if (cpu_online(tcpu)) {
2844 #ifdef CONFIG_RFS_ACCEL
2847 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2848 * @dev: Device on which the filter was set
2849 * @rxq_index: RX queue index
2850 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2851 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2853 * Drivers that implement ndo_rx_flow_steer() should periodically call
2854 * this function for each installed filter and remove the filters for
2855 * which it returns %true.
2857 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2858 u32 flow_id, u16 filter_id)
2860 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2861 struct rps_dev_flow_table *flow_table;
2862 struct rps_dev_flow *rflow;
2867 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2868 if (flow_table && flow_id <= flow_table->mask) {
2869 rflow = &flow_table->flows[flow_id];
2870 cpu = ACCESS_ONCE(rflow->cpu);
2871 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2872 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2873 rflow->last_qtail) <
2874 (int)(10 * flow_table->mask)))
2880 EXPORT_SYMBOL(rps_may_expire_flow);
2882 #endif /* CONFIG_RFS_ACCEL */
2884 /* Called from hardirq (IPI) context */
2885 static void rps_trigger_softirq(void *data)
2887 struct softnet_data *sd = data;
2889 ____napi_schedule(sd, &sd->backlog);
2893 #endif /* CONFIG_RPS */
2896 * Check if this softnet_data structure is another cpu one
2897 * If yes, queue it to our IPI list and return 1
2900 static int rps_ipi_queued(struct softnet_data *sd)
2903 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2906 sd->rps_ipi_next = mysd->rps_ipi_list;
2907 mysd->rps_ipi_list = sd;
2909 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2912 #endif /* CONFIG_RPS */
2917 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2918 * queue (may be a remote CPU queue).
2920 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2921 unsigned int *qtail)
2923 struct softnet_data *sd;
2924 unsigned long flags;
2926 sd = &per_cpu(softnet_data, cpu);
2928 local_irq_save(flags);
2931 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2932 if (skb_queue_len(&sd->input_pkt_queue)) {
2934 __skb_queue_tail(&sd->input_pkt_queue, skb);
2935 input_queue_tail_incr_save(sd, qtail);
2937 local_irq_restore(flags);
2938 return NET_RX_SUCCESS;
2941 /* Schedule NAPI for backlog device
2942 * We can use non atomic operation since we own the queue lock
2944 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2945 if (!rps_ipi_queued(sd))
2946 ____napi_schedule(sd, &sd->backlog);
2954 local_irq_restore(flags);
2956 atomic_long_inc(&skb->dev->rx_dropped);
2962 * netif_rx - post buffer to the network code
2963 * @skb: buffer to post
2965 * This function receives a packet from a device driver and queues it for
2966 * the upper (protocol) levels to process. It always succeeds. The buffer
2967 * may be dropped during processing for congestion control or by the
2971 * NET_RX_SUCCESS (no congestion)
2972 * NET_RX_DROP (packet was dropped)
2976 int netif_rx(struct sk_buff *skb)
2980 /* if netpoll wants it, pretend we never saw it */
2981 if (netpoll_rx(skb))
2984 net_timestamp_check(netdev_tstamp_prequeue, skb);
2986 trace_netif_rx(skb);
2988 if (static_key_false(&rps_needed)) {
2989 struct rps_dev_flow voidflow, *rflow = &voidflow;
2995 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2997 cpu = smp_processor_id();
2999 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3007 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3012 EXPORT_SYMBOL(netif_rx);
3014 int netif_rx_ni(struct sk_buff *skb)
3019 err = netif_rx(skb);
3020 if (local_softirq_pending())
3026 EXPORT_SYMBOL(netif_rx_ni);
3028 static void net_tx_action(struct softirq_action *h)
3030 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3032 if (sd->completion_queue) {
3033 struct sk_buff *clist;
3035 local_irq_disable();
3036 clist = sd->completion_queue;
3037 sd->completion_queue = NULL;
3041 struct sk_buff *skb = clist;
3042 clist = clist->next;
3044 WARN_ON(atomic_read(&skb->users));
3045 trace_kfree_skb(skb, net_tx_action);
3050 if (sd->output_queue) {
3053 local_irq_disable();
3054 head = sd->output_queue;
3055 sd->output_queue = NULL;
3056 sd->output_queue_tailp = &sd->output_queue;
3060 struct Qdisc *q = head;
3061 spinlock_t *root_lock;
3063 head = head->next_sched;
3065 root_lock = qdisc_lock(q);
3066 if (spin_trylock(root_lock)) {
3067 smp_mb__before_clear_bit();
3068 clear_bit(__QDISC_STATE_SCHED,
3071 spin_unlock(root_lock);
3073 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3075 __netif_reschedule(q);
3077 smp_mb__before_clear_bit();
3078 clear_bit(__QDISC_STATE_SCHED,
3086 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3087 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3088 /* This hook is defined here for ATM LANE */
3089 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3090 unsigned char *addr) __read_mostly;
3091 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3094 #ifdef CONFIG_NET_CLS_ACT
3095 /* TODO: Maybe we should just force sch_ingress to be compiled in
3096 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3097 * a compare and 2 stores extra right now if we dont have it on
3098 * but have CONFIG_NET_CLS_ACT
3099 * NOTE: This doesn't stop any functionality; if you dont have
3100 * the ingress scheduler, you just can't add policies on ingress.
3103 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3105 struct net_device *dev = skb->dev;
3106 u32 ttl = G_TC_RTTL(skb->tc_verd);
3107 int result = TC_ACT_OK;
3110 if (unlikely(MAX_RED_LOOP < ttl++)) {
3111 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3112 skb->skb_iif, dev->ifindex);
3116 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3117 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3120 if (q != &noop_qdisc) {
3121 spin_lock(qdisc_lock(q));
3122 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3123 result = qdisc_enqueue_root(skb, q);
3124 spin_unlock(qdisc_lock(q));
3130 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3131 struct packet_type **pt_prev,
3132 int *ret, struct net_device *orig_dev)
3134 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3136 if (!rxq || rxq->qdisc == &noop_qdisc)
3140 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3144 switch (ing_filter(skb, rxq)) {
3158 * netdev_rx_handler_register - register receive handler
3159 * @dev: device to register a handler for
3160 * @rx_handler: receive handler to register
3161 * @rx_handler_data: data pointer that is used by rx handler
3163 * Register a receive hander for a device. This handler will then be
3164 * called from __netif_receive_skb. A negative errno code is returned
3167 * The caller must hold the rtnl_mutex.
3169 * For a general description of rx_handler, see enum rx_handler_result.
3171 int netdev_rx_handler_register(struct net_device *dev,
3172 rx_handler_func_t *rx_handler,
3173 void *rx_handler_data)
3177 if (dev->rx_handler)
3180 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3181 rcu_assign_pointer(dev->rx_handler, rx_handler);
3185 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3188 * netdev_rx_handler_unregister - unregister receive handler
3189 * @dev: device to unregister a handler from
3191 * Unregister a receive hander from a device.
3193 * The caller must hold the rtnl_mutex.
3195 void netdev_rx_handler_unregister(struct net_device *dev)
3199 RCU_INIT_POINTER(dev->rx_handler, NULL);
3200 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3202 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3205 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3206 * the special handling of PFMEMALLOC skbs.
3208 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3210 switch (skb->protocol) {
3211 case __constant_htons(ETH_P_ARP):
3212 case __constant_htons(ETH_P_IP):
3213 case __constant_htons(ETH_P_IPV6):
3214 case __constant_htons(ETH_P_8021Q):
3221 static int __netif_receive_skb(struct sk_buff *skb)
3223 struct packet_type *ptype, *pt_prev;
3224 rx_handler_func_t *rx_handler;
3225 struct net_device *orig_dev;
3226 struct net_device *null_or_dev;
3227 bool deliver_exact = false;
3228 int ret = NET_RX_DROP;
3230 unsigned long pflags = current->flags;
3232 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3234 trace_netif_receive_skb(skb);
3237 * PFMEMALLOC skbs are special, they should
3238 * - be delivered to SOCK_MEMALLOC sockets only
3239 * - stay away from userspace
3240 * - have bounded memory usage
3242 * Use PF_MEMALLOC as this saves us from propagating the allocation
3243 * context down to all allocation sites.
3245 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3246 current->flags |= PF_MEMALLOC;
3248 /* if we've gotten here through NAPI, check netpoll */
3249 if (netpoll_receive_skb(skb))
3252 orig_dev = skb->dev;
3254 skb_reset_network_header(skb);
3255 skb_reset_transport_header(skb);
3256 skb_reset_mac_len(skb);
3263 skb->skb_iif = skb->dev->ifindex;
3265 __this_cpu_inc(softnet_data.processed);
3267 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3268 skb = vlan_untag(skb);
3273 #ifdef CONFIG_NET_CLS_ACT
3274 if (skb->tc_verd & TC_NCLS) {
3275 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3280 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3283 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3284 if (!ptype->dev || ptype->dev == skb->dev) {
3286 ret = deliver_skb(skb, pt_prev, orig_dev);
3292 #ifdef CONFIG_NET_CLS_ACT
3293 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3299 if (sk_memalloc_socks() && skb_pfmemalloc(skb)
3300 && !skb_pfmemalloc_protocol(skb))
3303 rx_handler = rcu_dereference(skb->dev->rx_handler);
3304 if (vlan_tx_tag_present(skb)) {
3306 ret = deliver_skb(skb, pt_prev, orig_dev);
3309 if (vlan_do_receive(&skb, !rx_handler))
3311 else if (unlikely(!skb))
3317 ret = deliver_skb(skb, pt_prev, orig_dev);
3320 switch (rx_handler(&skb)) {
3321 case RX_HANDLER_CONSUMED:
3323 case RX_HANDLER_ANOTHER:
3325 case RX_HANDLER_EXACT:
3326 deliver_exact = true;
3327 case RX_HANDLER_PASS:
3334 /* deliver only exact match when indicated */
3335 null_or_dev = deliver_exact ? skb->dev : NULL;
3337 type = skb->protocol;
3338 list_for_each_entry_rcu(ptype,
3339 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3340 if (ptype->type == type &&
3341 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3342 ptype->dev == orig_dev)) {
3344 ret = deliver_skb(skb, pt_prev, orig_dev);
3350 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3353 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3356 atomic_long_inc(&skb->dev->rx_dropped);
3358 /* Jamal, now you will not able to escape explaining
3359 * me how you were going to use this. :-)
3367 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3372 * netif_receive_skb - process receive buffer from network
3373 * @skb: buffer to process
3375 * netif_receive_skb() is the main receive data processing function.
3376 * It always succeeds. The buffer may be dropped during processing
3377 * for congestion control or by the protocol layers.
3379 * This function may only be called from softirq context and interrupts
3380 * should be enabled.
3382 * Return values (usually ignored):
3383 * NET_RX_SUCCESS: no congestion
3384 * NET_RX_DROP: packet was dropped
3386 int netif_receive_skb(struct sk_buff *skb)
3388 net_timestamp_check(netdev_tstamp_prequeue, skb);
3390 if (skb_defer_rx_timestamp(skb))
3391 return NET_RX_SUCCESS;
3394 if (static_key_false(&rps_needed)) {
3395 struct rps_dev_flow voidflow, *rflow = &voidflow;
3400 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3403 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3410 return __netif_receive_skb(skb);
3412 EXPORT_SYMBOL(netif_receive_skb);
3414 /* Network device is going away, flush any packets still pending
3415 * Called with irqs disabled.
3417 static void flush_backlog(void *arg)
3419 struct net_device *dev = arg;
3420 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3421 struct sk_buff *skb, *tmp;
3424 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3425 if (skb->dev == dev) {
3426 __skb_unlink(skb, &sd->input_pkt_queue);
3428 input_queue_head_incr(sd);
3433 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3434 if (skb->dev == dev) {
3435 __skb_unlink(skb, &sd->process_queue);
3437 input_queue_head_incr(sd);
3442 static int napi_gro_complete(struct sk_buff *skb)
3444 struct packet_type *ptype;
3445 __be16 type = skb->protocol;
3446 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3449 if (NAPI_GRO_CB(skb)->count == 1) {
3450 skb_shinfo(skb)->gso_size = 0;
3455 list_for_each_entry_rcu(ptype, head, list) {
3456 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3459 err = ptype->gro_complete(skb);
3465 WARN_ON(&ptype->list == head);
3467 return NET_RX_SUCCESS;
3471 return netif_receive_skb(skb);
3474 inline void napi_gro_flush(struct napi_struct *napi)
3476 struct sk_buff *skb, *next;
3478 for (skb = napi->gro_list; skb; skb = next) {
3481 napi_gro_complete(skb);
3484 napi->gro_count = 0;
3485 napi->gro_list = NULL;
3487 EXPORT_SYMBOL(napi_gro_flush);
3489 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3491 struct sk_buff **pp = NULL;
3492 struct packet_type *ptype;
3493 __be16 type = skb->protocol;
3494 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3497 enum gro_result ret;
3499 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3502 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3506 list_for_each_entry_rcu(ptype, head, list) {
3507 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3510 skb_set_network_header(skb, skb_gro_offset(skb));
3511 mac_len = skb->network_header - skb->mac_header;
3512 skb->mac_len = mac_len;
3513 NAPI_GRO_CB(skb)->same_flow = 0;
3514 NAPI_GRO_CB(skb)->flush = 0;
3515 NAPI_GRO_CB(skb)->free = 0;
3517 pp = ptype->gro_receive(&napi->gro_list, skb);
3522 if (&ptype->list == head)
3525 same_flow = NAPI_GRO_CB(skb)->same_flow;
3526 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3529 struct sk_buff *nskb = *pp;
3533 napi_gro_complete(nskb);
3540 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3544 NAPI_GRO_CB(skb)->count = 1;
3545 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3546 skb->next = napi->gro_list;
3547 napi->gro_list = skb;
3551 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3552 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3554 BUG_ON(skb->end - skb->tail < grow);
3556 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3559 skb->data_len -= grow;
3561 skb_shinfo(skb)->frags[0].page_offset += grow;
3562 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3564 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3565 skb_frag_unref(skb, 0);
3566 memmove(skb_shinfo(skb)->frags,
3567 skb_shinfo(skb)->frags + 1,
3568 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3579 EXPORT_SYMBOL(dev_gro_receive);
3581 static inline gro_result_t
3582 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3585 unsigned int maclen = skb->dev->hard_header_len;
3587 for (p = napi->gro_list; p; p = p->next) {
3588 unsigned long diffs;
3590 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3591 diffs |= p->vlan_tci ^ skb->vlan_tci;
3592 if (maclen == ETH_HLEN)
3593 diffs |= compare_ether_header(skb_mac_header(p),
3594 skb_gro_mac_header(skb));
3596 diffs = memcmp(skb_mac_header(p),
3597 skb_gro_mac_header(skb),
3599 NAPI_GRO_CB(p)->same_flow = !diffs;
3600 NAPI_GRO_CB(p)->flush = 0;
3603 return dev_gro_receive(napi, skb);
3606 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3610 if (netif_receive_skb(skb))
3618 case GRO_MERGED_FREE:
3619 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3620 kmem_cache_free(skbuff_head_cache, skb);
3632 EXPORT_SYMBOL(napi_skb_finish);
3634 static void skb_gro_reset_offset(struct sk_buff *skb)
3636 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3637 const skb_frag_t *frag0 = &pinfo->frags[0];
3639 NAPI_GRO_CB(skb)->data_offset = 0;
3640 NAPI_GRO_CB(skb)->frag0 = NULL;
3641 NAPI_GRO_CB(skb)->frag0_len = 0;
3643 if (skb->mac_header == skb->tail &&
3645 !PageHighMem(skb_frag_page(frag0))) {
3646 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3647 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3651 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3653 skb_gro_reset_offset(skb);
3655 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3657 EXPORT_SYMBOL(napi_gro_receive);
3659 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3661 __skb_pull(skb, skb_headlen(skb));
3662 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3663 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3665 skb->dev = napi->dev;
3671 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3673 struct sk_buff *skb = napi->skb;
3676 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3682 EXPORT_SYMBOL(napi_get_frags);
3684 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3690 skb->protocol = eth_type_trans(skb, skb->dev);
3692 if (ret == GRO_HELD)
3693 skb_gro_pull(skb, -ETH_HLEN);
3694 else if (netif_receive_skb(skb))
3699 case GRO_MERGED_FREE:
3700 napi_reuse_skb(napi, skb);
3709 EXPORT_SYMBOL(napi_frags_finish);
3711 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3713 struct sk_buff *skb = napi->skb;
3720 skb_reset_mac_header(skb);
3721 skb_gro_reset_offset(skb);
3723 off = skb_gro_offset(skb);
3724 hlen = off + sizeof(*eth);
3725 eth = skb_gro_header_fast(skb, off);
3726 if (skb_gro_header_hard(skb, hlen)) {
3727 eth = skb_gro_header_slow(skb, hlen, off);
3728 if (unlikely(!eth)) {
3729 napi_reuse_skb(napi, skb);
3735 skb_gro_pull(skb, sizeof(*eth));
3738 * This works because the only protocols we care about don't require
3739 * special handling. We'll fix it up properly at the end.
3741 skb->protocol = eth->h_proto;
3747 gro_result_t napi_gro_frags(struct napi_struct *napi)
3749 struct sk_buff *skb = napi_frags_skb(napi);
3754 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3756 EXPORT_SYMBOL(napi_gro_frags);
3759 * net_rps_action sends any pending IPI's for rps.
3760 * Note: called with local irq disabled, but exits with local irq enabled.
3762 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3765 struct softnet_data *remsd = sd->rps_ipi_list;
3768 sd->rps_ipi_list = NULL;
3772 /* Send pending IPI's to kick RPS processing on remote cpus. */
3774 struct softnet_data *next = remsd->rps_ipi_next;
3776 if (cpu_online(remsd->cpu))
3777 __smp_call_function_single(remsd->cpu,
3786 static int process_backlog(struct napi_struct *napi, int quota)
3789 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3792 /* Check if we have pending ipi, its better to send them now,
3793 * not waiting net_rx_action() end.
3795 if (sd->rps_ipi_list) {
3796 local_irq_disable();
3797 net_rps_action_and_irq_enable(sd);
3800 napi->weight = weight_p;
3801 local_irq_disable();
3802 while (work < quota) {
3803 struct sk_buff *skb;
3806 while ((skb = __skb_dequeue(&sd->process_queue))) {
3808 __netif_receive_skb(skb);
3809 local_irq_disable();
3810 input_queue_head_incr(sd);
3811 if (++work >= quota) {
3818 qlen = skb_queue_len(&sd->input_pkt_queue);
3820 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3821 &sd->process_queue);
3823 if (qlen < quota - work) {
3825 * Inline a custom version of __napi_complete().
3826 * only current cpu owns and manipulates this napi,
3827 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3828 * we can use a plain write instead of clear_bit(),
3829 * and we dont need an smp_mb() memory barrier.
3831 list_del(&napi->poll_list);
3834 quota = work + qlen;
3844 * __napi_schedule - schedule for receive
3845 * @n: entry to schedule
3847 * The entry's receive function will be scheduled to run
3849 void __napi_schedule(struct napi_struct *n)
3851 unsigned long flags;
3853 local_irq_save(flags);
3854 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3855 local_irq_restore(flags);
3857 EXPORT_SYMBOL(__napi_schedule);
3859 void __napi_complete(struct napi_struct *n)
3861 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3862 BUG_ON(n->gro_list);
3864 list_del(&n->poll_list);
3865 smp_mb__before_clear_bit();
3866 clear_bit(NAPI_STATE_SCHED, &n->state);
3868 EXPORT_SYMBOL(__napi_complete);
3870 void napi_complete(struct napi_struct *n)
3872 unsigned long flags;
3875 * don't let napi dequeue from the cpu poll list
3876 * just in case its running on a different cpu
3878 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3882 local_irq_save(flags);
3884 local_irq_restore(flags);
3886 EXPORT_SYMBOL(napi_complete);
3888 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3889 int (*poll)(struct napi_struct *, int), int weight)
3891 INIT_LIST_HEAD(&napi->poll_list);
3892 napi->gro_count = 0;
3893 napi->gro_list = NULL;
3896 napi->weight = weight;
3897 list_add(&napi->dev_list, &dev->napi_list);
3899 #ifdef CONFIG_NETPOLL
3900 spin_lock_init(&napi->poll_lock);
3901 napi->poll_owner = -1;
3903 set_bit(NAPI_STATE_SCHED, &napi->state);
3905 EXPORT_SYMBOL(netif_napi_add);
3907 void netif_napi_del(struct napi_struct *napi)
3909 struct sk_buff *skb, *next;
3911 list_del_init(&napi->dev_list);
3912 napi_free_frags(napi);
3914 for (skb = napi->gro_list; skb; skb = next) {
3920 napi->gro_list = NULL;
3921 napi->gro_count = 0;
3923 EXPORT_SYMBOL(netif_napi_del);
3925 static void net_rx_action(struct softirq_action *h)
3927 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3928 unsigned long time_limit = jiffies + 2;
3929 int budget = netdev_budget;
3932 local_irq_disable();
3934 while (!list_empty(&sd->poll_list)) {
3935 struct napi_struct *n;
3938 /* If softirq window is exhuasted then punt.
3939 * Allow this to run for 2 jiffies since which will allow
3940 * an average latency of 1.5/HZ.
3942 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3947 /* Even though interrupts have been re-enabled, this
3948 * access is safe because interrupts can only add new
3949 * entries to the tail of this list, and only ->poll()
3950 * calls can remove this head entry from the list.
3952 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3954 have = netpoll_poll_lock(n);
3958 /* This NAPI_STATE_SCHED test is for avoiding a race
3959 * with netpoll's poll_napi(). Only the entity which
3960 * obtains the lock and sees NAPI_STATE_SCHED set will
3961 * actually make the ->poll() call. Therefore we avoid
3962 * accidentally calling ->poll() when NAPI is not scheduled.
3965 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3966 work = n->poll(n, weight);
3970 WARN_ON_ONCE(work > weight);
3974 local_irq_disable();
3976 /* Drivers must not modify the NAPI state if they
3977 * consume the entire weight. In such cases this code
3978 * still "owns" the NAPI instance and therefore can
3979 * move the instance around on the list at-will.
3981 if (unlikely(work == weight)) {
3982 if (unlikely(napi_disable_pending(n))) {
3985 local_irq_disable();
3987 list_move_tail(&n->poll_list, &sd->poll_list);
3990 netpoll_poll_unlock(have);
3993 net_rps_action_and_irq_enable(sd);
3995 #ifdef CONFIG_NET_DMA
3997 * There may not be any more sk_buffs coming right now, so push
3998 * any pending DMA copies to hardware
4000 dma_issue_pending_all();
4007 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4011 static gifconf_func_t *gifconf_list[NPROTO];
4014 * register_gifconf - register a SIOCGIF handler
4015 * @family: Address family
4016 * @gifconf: Function handler
4018 * Register protocol dependent address dumping routines. The handler
4019 * that is passed must not be freed or reused until it has been replaced
4020 * by another handler.
4022 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4024 if (family >= NPROTO)
4026 gifconf_list[family] = gifconf;
4029 EXPORT_SYMBOL(register_gifconf);
4033 * Map an interface index to its name (SIOCGIFNAME)
4037 * We need this ioctl for efficient implementation of the
4038 * if_indextoname() function required by the IPv6 API. Without
4039 * it, we would have to search all the interfaces to find a
4043 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4045 struct net_device *dev;
4049 * Fetch the caller's info block.
4052 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4056 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4062 strcpy(ifr.ifr_name, dev->name);
4065 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4071 * Perform a SIOCGIFCONF call. This structure will change
4072 * size eventually, and there is nothing I can do about it.
4073 * Thus we will need a 'compatibility mode'.
4076 static int dev_ifconf(struct net *net, char __user *arg)
4079 struct net_device *dev;
4086 * Fetch the caller's info block.
4089 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4096 * Loop over the interfaces, and write an info block for each.
4100 for_each_netdev(net, dev) {
4101 for (i = 0; i < NPROTO; i++) {
4102 if (gifconf_list[i]) {
4105 done = gifconf_list[i](dev, NULL, 0);
4107 done = gifconf_list[i](dev, pos + total,
4117 * All done. Write the updated control block back to the caller.
4119 ifc.ifc_len = total;
4122 * Both BSD and Solaris return 0 here, so we do too.
4124 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4127 #ifdef CONFIG_PROC_FS
4129 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4131 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4132 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4133 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4135 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4137 struct net *net = seq_file_net(seq);
4138 struct net_device *dev;
4139 struct hlist_node *p;
4140 struct hlist_head *h;
4141 unsigned int count = 0, offset = get_offset(*pos);
4143 h = &net->dev_name_head[get_bucket(*pos)];
4144 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4145 if (++count == offset)
4152 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4154 struct net_device *dev;
4155 unsigned int bucket;
4158 dev = dev_from_same_bucket(seq, pos);
4162 bucket = get_bucket(*pos) + 1;
4163 *pos = set_bucket_offset(bucket, 1);
4164 } while (bucket < NETDEV_HASHENTRIES);
4170 * This is invoked by the /proc filesystem handler to display a device
4173 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4178 return SEQ_START_TOKEN;
4180 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4183 return dev_from_bucket(seq, pos);
4186 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4189 return dev_from_bucket(seq, pos);
4192 void dev_seq_stop(struct seq_file *seq, void *v)
4198 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4200 struct rtnl_link_stats64 temp;
4201 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4203 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4204 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4205 dev->name, stats->rx_bytes, stats->rx_packets,
4207 stats->rx_dropped + stats->rx_missed_errors,
4208 stats->rx_fifo_errors,
4209 stats->rx_length_errors + stats->rx_over_errors +
4210 stats->rx_crc_errors + stats->rx_frame_errors,
4211 stats->rx_compressed, stats->multicast,
4212 stats->tx_bytes, stats->tx_packets,
4213 stats->tx_errors, stats->tx_dropped,
4214 stats->tx_fifo_errors, stats->collisions,
4215 stats->tx_carrier_errors +
4216 stats->tx_aborted_errors +
4217 stats->tx_window_errors +
4218 stats->tx_heartbeat_errors,
4219 stats->tx_compressed);
4223 * Called from the PROCfs module. This now uses the new arbitrary sized
4224 * /proc/net interface to create /proc/net/dev
4226 static int dev_seq_show(struct seq_file *seq, void *v)
4228 if (v == SEQ_START_TOKEN)
4229 seq_puts(seq, "Inter-| Receive "
4231 " face |bytes packets errs drop fifo frame "
4232 "compressed multicast|bytes packets errs "
4233 "drop fifo colls carrier compressed\n");
4235 dev_seq_printf_stats(seq, v);
4239 static struct softnet_data *softnet_get_online(loff_t *pos)
4241 struct softnet_data *sd = NULL;
4243 while (*pos < nr_cpu_ids)
4244 if (cpu_online(*pos)) {
4245 sd = &per_cpu(softnet_data, *pos);
4252 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4254 return softnet_get_online(pos);
4257 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4260 return softnet_get_online(pos);
4263 static void softnet_seq_stop(struct seq_file *seq, void *v)
4267 static int softnet_seq_show(struct seq_file *seq, void *v)
4269 struct softnet_data *sd = v;
4271 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4272 sd->processed, sd->dropped, sd->time_squeeze, 0,
4273 0, 0, 0, 0, /* was fastroute */
4274 sd->cpu_collision, sd->received_rps);
4278 static const struct seq_operations dev_seq_ops = {
4279 .start = dev_seq_start,
4280 .next = dev_seq_next,
4281 .stop = dev_seq_stop,
4282 .show = dev_seq_show,
4285 static int dev_seq_open(struct inode *inode, struct file *file)
4287 return seq_open_net(inode, file, &dev_seq_ops,
4288 sizeof(struct seq_net_private));
4291 static const struct file_operations dev_seq_fops = {
4292 .owner = THIS_MODULE,
4293 .open = dev_seq_open,
4295 .llseek = seq_lseek,
4296 .release = seq_release_net,
4299 static const struct seq_operations softnet_seq_ops = {
4300 .start = softnet_seq_start,
4301 .next = softnet_seq_next,
4302 .stop = softnet_seq_stop,
4303 .show = softnet_seq_show,
4306 static int softnet_seq_open(struct inode *inode, struct file *file)
4308 return seq_open(file, &softnet_seq_ops);
4311 static const struct file_operations softnet_seq_fops = {
4312 .owner = THIS_MODULE,
4313 .open = softnet_seq_open,
4315 .llseek = seq_lseek,
4316 .release = seq_release,
4319 static void *ptype_get_idx(loff_t pos)
4321 struct packet_type *pt = NULL;
4325 list_for_each_entry_rcu(pt, &ptype_all, list) {
4331 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4332 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4341 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4345 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4348 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4350 struct packet_type *pt;
4351 struct list_head *nxt;
4355 if (v == SEQ_START_TOKEN)
4356 return ptype_get_idx(0);
4359 nxt = pt->list.next;
4360 if (pt->type == htons(ETH_P_ALL)) {
4361 if (nxt != &ptype_all)
4364 nxt = ptype_base[0].next;
4366 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4368 while (nxt == &ptype_base[hash]) {
4369 if (++hash >= PTYPE_HASH_SIZE)
4371 nxt = ptype_base[hash].next;
4374 return list_entry(nxt, struct packet_type, list);
4377 static void ptype_seq_stop(struct seq_file *seq, void *v)
4383 static int ptype_seq_show(struct seq_file *seq, void *v)
4385 struct packet_type *pt = v;
4387 if (v == SEQ_START_TOKEN)
4388 seq_puts(seq, "Type Device Function\n");
4389 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4390 if (pt->type == htons(ETH_P_ALL))
4391 seq_puts(seq, "ALL ");
4393 seq_printf(seq, "%04x", ntohs(pt->type));
4395 seq_printf(seq, " %-8s %pF\n",
4396 pt->dev ? pt->dev->name : "", pt->func);
4402 static const struct seq_operations ptype_seq_ops = {
4403 .start = ptype_seq_start,
4404 .next = ptype_seq_next,
4405 .stop = ptype_seq_stop,
4406 .show = ptype_seq_show,
4409 static int ptype_seq_open(struct inode *inode, struct file *file)
4411 return seq_open_net(inode, file, &ptype_seq_ops,
4412 sizeof(struct seq_net_private));
4415 static const struct file_operations ptype_seq_fops = {
4416 .owner = THIS_MODULE,
4417 .open = ptype_seq_open,
4419 .llseek = seq_lseek,
4420 .release = seq_release_net,
4424 static int __net_init dev_proc_net_init(struct net *net)
4428 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4430 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4432 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4435 if (wext_proc_init(net))
4441 proc_net_remove(net, "ptype");
4443 proc_net_remove(net, "softnet_stat");
4445 proc_net_remove(net, "dev");
4449 static void __net_exit dev_proc_net_exit(struct net *net)
4451 wext_proc_exit(net);
4453 proc_net_remove(net, "ptype");
4454 proc_net_remove(net, "softnet_stat");
4455 proc_net_remove(net, "dev");
4458 static struct pernet_operations __net_initdata dev_proc_ops = {
4459 .init = dev_proc_net_init,
4460 .exit = dev_proc_net_exit,
4463 static int __init dev_proc_init(void)
4465 return register_pernet_subsys(&dev_proc_ops);
4468 #define dev_proc_init() 0
4469 #endif /* CONFIG_PROC_FS */
4473 * netdev_set_master - set up master pointer
4474 * @slave: slave device
4475 * @master: new master device
4477 * Changes the master device of the slave. Pass %NULL to break the
4478 * bonding. The caller must hold the RTNL semaphore. On a failure
4479 * a negative errno code is returned. On success the reference counts
4480 * are adjusted and the function returns zero.
4482 int netdev_set_master(struct net_device *slave, struct net_device *master)
4484 struct net_device *old = slave->master;
4494 slave->master = master;
4500 EXPORT_SYMBOL(netdev_set_master);
4503 * netdev_set_bond_master - set up bonding master/slave pair
4504 * @slave: slave device
4505 * @master: new master device
4507 * Changes the master device of the slave. Pass %NULL to break the
4508 * bonding. The caller must hold the RTNL semaphore. On a failure
4509 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4510 * to the routing socket and the function returns zero.
4512 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4518 err = netdev_set_master(slave, master);
4522 slave->flags |= IFF_SLAVE;
4524 slave->flags &= ~IFF_SLAVE;
4526 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4529 EXPORT_SYMBOL(netdev_set_bond_master);
4531 static void dev_change_rx_flags(struct net_device *dev, int flags)
4533 const struct net_device_ops *ops = dev->netdev_ops;
4535 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4536 ops->ndo_change_rx_flags(dev, flags);
4539 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4541 unsigned int old_flags = dev->flags;
4547 dev->flags |= IFF_PROMISC;
4548 dev->promiscuity += inc;
4549 if (dev->promiscuity == 0) {
4552 * If inc causes overflow, untouch promisc and return error.
4555 dev->flags &= ~IFF_PROMISC;
4557 dev->promiscuity -= inc;
4558 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4563 if (dev->flags != old_flags) {
4564 pr_info("device %s %s promiscuous mode\n",
4566 dev->flags & IFF_PROMISC ? "entered" : "left");
4567 if (audit_enabled) {
4568 current_uid_gid(&uid, &gid);
4569 audit_log(current->audit_context, GFP_ATOMIC,
4570 AUDIT_ANOM_PROMISCUOUS,
4571 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4572 dev->name, (dev->flags & IFF_PROMISC),
4573 (old_flags & IFF_PROMISC),
4574 from_kuid(&init_user_ns, audit_get_loginuid(current)),
4575 from_kuid(&init_user_ns, uid),
4576 from_kgid(&init_user_ns, gid),
4577 audit_get_sessionid(current));
4580 dev_change_rx_flags(dev, IFF_PROMISC);
4586 * dev_set_promiscuity - update promiscuity count on a device
4590 * Add or remove promiscuity from a device. While the count in the device
4591 * remains above zero the interface remains promiscuous. Once it hits zero
4592 * the device reverts back to normal filtering operation. A negative inc
4593 * value is used to drop promiscuity on the device.
4594 * Return 0 if successful or a negative errno code on error.
4596 int dev_set_promiscuity(struct net_device *dev, int inc)
4598 unsigned int old_flags = dev->flags;
4601 err = __dev_set_promiscuity(dev, inc);
4604 if (dev->flags != old_flags)
4605 dev_set_rx_mode(dev);
4608 EXPORT_SYMBOL(dev_set_promiscuity);
4611 * dev_set_allmulti - update allmulti count on a device
4615 * Add or remove reception of all multicast frames to a device. While the
4616 * count in the device remains above zero the interface remains listening
4617 * to all interfaces. Once it hits zero the device reverts back to normal
4618 * filtering operation. A negative @inc value is used to drop the counter
4619 * when releasing a resource needing all multicasts.
4620 * Return 0 if successful or a negative errno code on error.
4623 int dev_set_allmulti(struct net_device *dev, int inc)
4625 unsigned int old_flags = dev->flags;
4629 dev->flags |= IFF_ALLMULTI;
4630 dev->allmulti += inc;
4631 if (dev->allmulti == 0) {
4634 * If inc causes overflow, untouch allmulti and return error.
4637 dev->flags &= ~IFF_ALLMULTI;
4639 dev->allmulti -= inc;
4640 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4645 if (dev->flags ^ old_flags) {
4646 dev_change_rx_flags(dev, IFF_ALLMULTI);
4647 dev_set_rx_mode(dev);
4651 EXPORT_SYMBOL(dev_set_allmulti);
4654 * Upload unicast and multicast address lists to device and
4655 * configure RX filtering. When the device doesn't support unicast
4656 * filtering it is put in promiscuous mode while unicast addresses
4659 void __dev_set_rx_mode(struct net_device *dev)
4661 const struct net_device_ops *ops = dev->netdev_ops;
4663 /* dev_open will call this function so the list will stay sane. */
4664 if (!(dev->flags&IFF_UP))
4667 if (!netif_device_present(dev))
4670 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4671 /* Unicast addresses changes may only happen under the rtnl,
4672 * therefore calling __dev_set_promiscuity here is safe.
4674 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4675 __dev_set_promiscuity(dev, 1);
4676 dev->uc_promisc = true;
4677 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4678 __dev_set_promiscuity(dev, -1);
4679 dev->uc_promisc = false;
4683 if (ops->ndo_set_rx_mode)
4684 ops->ndo_set_rx_mode(dev);
4687 void dev_set_rx_mode(struct net_device *dev)
4689 netif_addr_lock_bh(dev);
4690 __dev_set_rx_mode(dev);
4691 netif_addr_unlock_bh(dev);
4695 * dev_get_flags - get flags reported to userspace
4698 * Get the combination of flag bits exported through APIs to userspace.
4700 unsigned int dev_get_flags(const struct net_device *dev)
4704 flags = (dev->flags & ~(IFF_PROMISC |
4709 (dev->gflags & (IFF_PROMISC |
4712 if (netif_running(dev)) {
4713 if (netif_oper_up(dev))
4714 flags |= IFF_RUNNING;
4715 if (netif_carrier_ok(dev))
4716 flags |= IFF_LOWER_UP;
4717 if (netif_dormant(dev))
4718 flags |= IFF_DORMANT;
4723 EXPORT_SYMBOL(dev_get_flags);
4725 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4727 unsigned int old_flags = dev->flags;
4733 * Set the flags on our device.
4736 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4737 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4739 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4743 * Load in the correct multicast list now the flags have changed.
4746 if ((old_flags ^ flags) & IFF_MULTICAST)
4747 dev_change_rx_flags(dev, IFF_MULTICAST);
4749 dev_set_rx_mode(dev);
4752 * Have we downed the interface. We handle IFF_UP ourselves
4753 * according to user attempts to set it, rather than blindly
4758 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4759 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4762 dev_set_rx_mode(dev);
4765 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4766 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4768 dev->gflags ^= IFF_PROMISC;
4769 dev_set_promiscuity(dev, inc);
4772 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4773 is important. Some (broken) drivers set IFF_PROMISC, when
4774 IFF_ALLMULTI is requested not asking us and not reporting.
4776 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4777 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4779 dev->gflags ^= IFF_ALLMULTI;
4780 dev_set_allmulti(dev, inc);
4786 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4788 unsigned int changes = dev->flags ^ old_flags;
4790 if (changes & IFF_UP) {
4791 if (dev->flags & IFF_UP)
4792 call_netdevice_notifiers(NETDEV_UP, dev);
4794 call_netdevice_notifiers(NETDEV_DOWN, dev);
4797 if (dev->flags & IFF_UP &&
4798 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4799 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4803 * dev_change_flags - change device settings
4805 * @flags: device state flags
4807 * Change settings on device based state flags. The flags are
4808 * in the userspace exported format.
4810 int dev_change_flags(struct net_device *dev, unsigned int flags)
4813 unsigned int changes, old_flags = dev->flags;
4815 ret = __dev_change_flags(dev, flags);
4819 changes = old_flags ^ dev->flags;
4821 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4823 __dev_notify_flags(dev, old_flags);
4826 EXPORT_SYMBOL(dev_change_flags);
4829 * dev_set_mtu - Change maximum transfer unit
4831 * @new_mtu: new transfer unit
4833 * Change the maximum transfer size of the network device.
4835 int dev_set_mtu(struct net_device *dev, int new_mtu)
4837 const struct net_device_ops *ops = dev->netdev_ops;
4840 if (new_mtu == dev->mtu)
4843 /* MTU must be positive. */
4847 if (!netif_device_present(dev))
4851 if (ops->ndo_change_mtu)
4852 err = ops->ndo_change_mtu(dev, new_mtu);
4856 if (!err && dev->flags & IFF_UP)
4857 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4860 EXPORT_SYMBOL(dev_set_mtu);
4863 * dev_set_group - Change group this device belongs to
4865 * @new_group: group this device should belong to
4867 void dev_set_group(struct net_device *dev, int new_group)
4869 dev->group = new_group;
4871 EXPORT_SYMBOL(dev_set_group);
4874 * dev_set_mac_address - Change Media Access Control Address
4878 * Change the hardware (MAC) address of the device
4880 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4882 const struct net_device_ops *ops = dev->netdev_ops;
4885 if (!ops->ndo_set_mac_address)
4887 if (sa->sa_family != dev->type)
4889 if (!netif_device_present(dev))
4891 err = ops->ndo_set_mac_address(dev, sa);
4893 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4894 add_device_randomness(dev->dev_addr, dev->addr_len);
4897 EXPORT_SYMBOL(dev_set_mac_address);
4900 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4902 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4905 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4911 case SIOCGIFFLAGS: /* Get interface flags */
4912 ifr->ifr_flags = (short) dev_get_flags(dev);
4915 case SIOCGIFMETRIC: /* Get the metric on the interface
4916 (currently unused) */
4917 ifr->ifr_metric = 0;
4920 case SIOCGIFMTU: /* Get the MTU of a device */
4921 ifr->ifr_mtu = dev->mtu;
4926 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4928 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4929 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4930 ifr->ifr_hwaddr.sa_family = dev->type;
4938 ifr->ifr_map.mem_start = dev->mem_start;
4939 ifr->ifr_map.mem_end = dev->mem_end;
4940 ifr->ifr_map.base_addr = dev->base_addr;
4941 ifr->ifr_map.irq = dev->irq;
4942 ifr->ifr_map.dma = dev->dma;
4943 ifr->ifr_map.port = dev->if_port;
4947 ifr->ifr_ifindex = dev->ifindex;
4951 ifr->ifr_qlen = dev->tx_queue_len;
4955 /* dev_ioctl() should ensure this case
4967 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4969 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4972 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4973 const struct net_device_ops *ops;
4978 ops = dev->netdev_ops;
4981 case SIOCSIFFLAGS: /* Set interface flags */
4982 return dev_change_flags(dev, ifr->ifr_flags);
4984 case SIOCSIFMETRIC: /* Set the metric on the interface
4985 (currently unused) */
4988 case SIOCSIFMTU: /* Set the MTU of a device */
4989 return dev_set_mtu(dev, ifr->ifr_mtu);
4992 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4994 case SIOCSIFHWBROADCAST:
4995 if (ifr->ifr_hwaddr.sa_family != dev->type)
4997 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4998 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4999 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5003 if (ops->ndo_set_config) {
5004 if (!netif_device_present(dev))
5006 return ops->ndo_set_config(dev, &ifr->ifr_map);
5011 if (!ops->ndo_set_rx_mode ||
5012 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5014 if (!netif_device_present(dev))
5016 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5019 if (!ops->ndo_set_rx_mode ||
5020 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5022 if (!netif_device_present(dev))
5024 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5027 if (ifr->ifr_qlen < 0)
5029 dev->tx_queue_len = ifr->ifr_qlen;
5033 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5034 return dev_change_name(dev, ifr->ifr_newname);
5037 err = net_hwtstamp_validate(ifr);
5043 * Unknown or private ioctl
5046 if ((cmd >= SIOCDEVPRIVATE &&
5047 cmd <= SIOCDEVPRIVATE + 15) ||
5048 cmd == SIOCBONDENSLAVE ||
5049 cmd == SIOCBONDRELEASE ||
5050 cmd == SIOCBONDSETHWADDR ||
5051 cmd == SIOCBONDSLAVEINFOQUERY ||
5052 cmd == SIOCBONDINFOQUERY ||
5053 cmd == SIOCBONDCHANGEACTIVE ||
5054 cmd == SIOCGMIIPHY ||
5055 cmd == SIOCGMIIREG ||
5056 cmd == SIOCSMIIREG ||
5057 cmd == SIOCBRADDIF ||
5058 cmd == SIOCBRDELIF ||
5059 cmd == SIOCSHWTSTAMP ||
5060 cmd == SIOCWANDEV) {
5062 if (ops->ndo_do_ioctl) {
5063 if (netif_device_present(dev))
5064 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5076 * This function handles all "interface"-type I/O control requests. The actual
5077 * 'doing' part of this is dev_ifsioc above.
5081 * dev_ioctl - network device ioctl
5082 * @net: the applicable net namespace
5083 * @cmd: command to issue
5084 * @arg: pointer to a struct ifreq in user space
5086 * Issue ioctl functions to devices. This is normally called by the
5087 * user space syscall interfaces but can sometimes be useful for
5088 * other purposes. The return value is the return from the syscall if
5089 * positive or a negative errno code on error.
5092 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5098 /* One special case: SIOCGIFCONF takes ifconf argument
5099 and requires shared lock, because it sleeps writing
5103 if (cmd == SIOCGIFCONF) {
5105 ret = dev_ifconf(net, (char __user *) arg);
5109 if (cmd == SIOCGIFNAME)
5110 return dev_ifname(net, (struct ifreq __user *)arg);
5112 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5115 ifr.ifr_name[IFNAMSIZ-1] = 0;
5117 colon = strchr(ifr.ifr_name, ':');
5122 * See which interface the caller is talking about.
5127 * These ioctl calls:
5128 * - can be done by all.
5129 * - atomic and do not require locking.
5140 dev_load(net, ifr.ifr_name);
5142 ret = dev_ifsioc_locked(net, &ifr, cmd);
5147 if (copy_to_user(arg, &ifr,
5148 sizeof(struct ifreq)))
5154 dev_load(net, ifr.ifr_name);
5156 ret = dev_ethtool(net, &ifr);
5161 if (copy_to_user(arg, &ifr,
5162 sizeof(struct ifreq)))
5168 * These ioctl calls:
5169 * - require superuser power.
5170 * - require strict serialization.
5176 if (!capable(CAP_NET_ADMIN))
5178 dev_load(net, ifr.ifr_name);
5180 ret = dev_ifsioc(net, &ifr, cmd);
5185 if (copy_to_user(arg, &ifr,
5186 sizeof(struct ifreq)))
5192 * These ioctl calls:
5193 * - require superuser power.
5194 * - require strict serialization.
5195 * - do not return a value
5205 case SIOCSIFHWBROADCAST:
5208 case SIOCBONDENSLAVE:
5209 case SIOCBONDRELEASE:
5210 case SIOCBONDSETHWADDR:
5211 case SIOCBONDCHANGEACTIVE:
5215 if (!capable(CAP_NET_ADMIN))
5218 case SIOCBONDSLAVEINFOQUERY:
5219 case SIOCBONDINFOQUERY:
5220 dev_load(net, ifr.ifr_name);
5222 ret = dev_ifsioc(net, &ifr, cmd);
5227 /* Get the per device memory space. We can add this but
5228 * currently do not support it */
5230 /* Set the per device memory buffer space.
5231 * Not applicable in our case */
5236 * Unknown or private ioctl.
5239 if (cmd == SIOCWANDEV ||
5240 (cmd >= SIOCDEVPRIVATE &&
5241 cmd <= SIOCDEVPRIVATE + 15)) {
5242 dev_load(net, ifr.ifr_name);
5244 ret = dev_ifsioc(net, &ifr, cmd);
5246 if (!ret && copy_to_user(arg, &ifr,
5247 sizeof(struct ifreq)))
5251 /* Take care of Wireless Extensions */
5252 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5253 return wext_handle_ioctl(net, &ifr, cmd, arg);
5260 * dev_new_index - allocate an ifindex
5261 * @net: the applicable net namespace
5263 * Returns a suitable unique value for a new device interface
5264 * number. The caller must hold the rtnl semaphore or the
5265 * dev_base_lock to be sure it remains unique.
5267 static int dev_new_index(struct net *net)
5269 int ifindex = net->ifindex;
5273 if (!__dev_get_by_index(net, ifindex))
5274 return net->ifindex = ifindex;
5278 /* Delayed registration/unregisteration */
5279 static LIST_HEAD(net_todo_list);
5281 static void net_set_todo(struct net_device *dev)
5283 list_add_tail(&dev->todo_list, &net_todo_list);
5286 static void rollback_registered_many(struct list_head *head)
5288 struct net_device *dev, *tmp;
5290 BUG_ON(dev_boot_phase);
5293 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5294 /* Some devices call without registering
5295 * for initialization unwind. Remove those
5296 * devices and proceed with the remaining.
5298 if (dev->reg_state == NETREG_UNINITIALIZED) {
5299 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5303 list_del(&dev->unreg_list);
5306 dev->dismantle = true;
5307 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5310 /* If device is running, close it first. */
5311 dev_close_many(head);
5313 list_for_each_entry(dev, head, unreg_list) {
5314 /* And unlink it from device chain. */
5315 unlist_netdevice(dev);
5317 dev->reg_state = NETREG_UNREGISTERING;
5322 list_for_each_entry(dev, head, unreg_list) {
5323 /* Shutdown queueing discipline. */
5327 /* Notify protocols, that we are about to destroy
5328 this device. They should clean all the things.
5330 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5332 if (!dev->rtnl_link_ops ||
5333 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5334 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5337 * Flush the unicast and multicast chains
5342 if (dev->netdev_ops->ndo_uninit)
5343 dev->netdev_ops->ndo_uninit(dev);
5345 /* Notifier chain MUST detach us from master device. */
5346 WARN_ON(dev->master);
5348 /* Remove entries from kobject tree */
5349 netdev_unregister_kobject(dev);
5354 list_for_each_entry(dev, head, unreg_list)
5358 static void rollback_registered(struct net_device *dev)
5362 list_add(&dev->unreg_list, &single);
5363 rollback_registered_many(&single);
5367 static netdev_features_t netdev_fix_features(struct net_device *dev,
5368 netdev_features_t features)
5370 /* Fix illegal checksum combinations */
5371 if ((features & NETIF_F_HW_CSUM) &&
5372 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5373 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5374 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5377 /* Fix illegal SG+CSUM combinations. */
5378 if ((features & NETIF_F_SG) &&
5379 !(features & NETIF_F_ALL_CSUM)) {
5381 "Dropping NETIF_F_SG since no checksum feature.\n");
5382 features &= ~NETIF_F_SG;
5385 /* TSO requires that SG is present as well. */
5386 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5387 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5388 features &= ~NETIF_F_ALL_TSO;
5391 /* TSO ECN requires that TSO is present as well. */
5392 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5393 features &= ~NETIF_F_TSO_ECN;
5395 /* Software GSO depends on SG. */
5396 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5397 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5398 features &= ~NETIF_F_GSO;
5401 /* UFO needs SG and checksumming */
5402 if (features & NETIF_F_UFO) {
5403 /* maybe split UFO into V4 and V6? */
5404 if (!((features & NETIF_F_GEN_CSUM) ||
5405 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5406 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5408 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5409 features &= ~NETIF_F_UFO;
5412 if (!(features & NETIF_F_SG)) {
5414 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5415 features &= ~NETIF_F_UFO;
5422 int __netdev_update_features(struct net_device *dev)
5424 netdev_features_t features;
5429 features = netdev_get_wanted_features(dev);
5431 if (dev->netdev_ops->ndo_fix_features)
5432 features = dev->netdev_ops->ndo_fix_features(dev, features);
5434 /* driver might be less strict about feature dependencies */
5435 features = netdev_fix_features(dev, features);
5437 if (dev->features == features)
5440 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5441 &dev->features, &features);
5443 if (dev->netdev_ops->ndo_set_features)
5444 err = dev->netdev_ops->ndo_set_features(dev, features);
5446 if (unlikely(err < 0)) {
5448 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5449 err, &features, &dev->features);
5454 dev->features = features;
5460 * netdev_update_features - recalculate device features
5461 * @dev: the device to check
5463 * Recalculate dev->features set and send notifications if it
5464 * has changed. Should be called after driver or hardware dependent
5465 * conditions might have changed that influence the features.
5467 void netdev_update_features(struct net_device *dev)
5469 if (__netdev_update_features(dev))
5470 netdev_features_change(dev);
5472 EXPORT_SYMBOL(netdev_update_features);
5475 * netdev_change_features - recalculate device features
5476 * @dev: the device to check
5478 * Recalculate dev->features set and send notifications even
5479 * if they have not changed. Should be called instead of
5480 * netdev_update_features() if also dev->vlan_features might
5481 * have changed to allow the changes to be propagated to stacked
5484 void netdev_change_features(struct net_device *dev)
5486 __netdev_update_features(dev);
5487 netdev_features_change(dev);
5489 EXPORT_SYMBOL(netdev_change_features);
5492 * netif_stacked_transfer_operstate - transfer operstate
5493 * @rootdev: the root or lower level device to transfer state from
5494 * @dev: the device to transfer operstate to
5496 * Transfer operational state from root to device. This is normally
5497 * called when a stacking relationship exists between the root
5498 * device and the device(a leaf device).
5500 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5501 struct net_device *dev)
5503 if (rootdev->operstate == IF_OPER_DORMANT)
5504 netif_dormant_on(dev);
5506 netif_dormant_off(dev);
5508 if (netif_carrier_ok(rootdev)) {
5509 if (!netif_carrier_ok(dev))
5510 netif_carrier_on(dev);
5512 if (netif_carrier_ok(dev))
5513 netif_carrier_off(dev);
5516 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5519 static int netif_alloc_rx_queues(struct net_device *dev)
5521 unsigned int i, count = dev->num_rx_queues;
5522 struct netdev_rx_queue *rx;
5526 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5528 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5533 for (i = 0; i < count; i++)
5539 static void netdev_init_one_queue(struct net_device *dev,
5540 struct netdev_queue *queue, void *_unused)
5542 /* Initialize queue lock */
5543 spin_lock_init(&queue->_xmit_lock);
5544 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5545 queue->xmit_lock_owner = -1;
5546 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5549 dql_init(&queue->dql, HZ);
5553 static int netif_alloc_netdev_queues(struct net_device *dev)
5555 unsigned int count = dev->num_tx_queues;
5556 struct netdev_queue *tx;
5560 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5562 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5567 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5568 spin_lock_init(&dev->tx_global_lock);
5574 * register_netdevice - register a network device
5575 * @dev: device to register
5577 * Take a completed network device structure and add it to the kernel
5578 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5579 * chain. 0 is returned on success. A negative errno code is returned
5580 * on a failure to set up the device, or if the name is a duplicate.
5582 * Callers must hold the rtnl semaphore. You may want
5583 * register_netdev() instead of this.
5586 * The locking appears insufficient to guarantee two parallel registers
5587 * will not get the same name.
5590 int register_netdevice(struct net_device *dev)
5593 struct net *net = dev_net(dev);
5595 BUG_ON(dev_boot_phase);
5600 /* When net_device's are persistent, this will be fatal. */
5601 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5604 spin_lock_init(&dev->addr_list_lock);
5605 netdev_set_addr_lockdep_class(dev);
5609 ret = dev_get_valid_name(net, dev, dev->name);
5613 /* Init, if this function is available */
5614 if (dev->netdev_ops->ndo_init) {
5615 ret = dev->netdev_ops->ndo_init(dev);
5625 dev->ifindex = dev_new_index(net);
5626 else if (__dev_get_by_index(net, dev->ifindex))
5629 if (dev->iflink == -1)
5630 dev->iflink = dev->ifindex;
5632 /* Transfer changeable features to wanted_features and enable
5633 * software offloads (GSO and GRO).
5635 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5636 dev->features |= NETIF_F_SOFT_FEATURES;
5637 dev->wanted_features = dev->features & dev->hw_features;
5639 /* Turn on no cache copy if HW is doing checksum */
5640 if (!(dev->flags & IFF_LOOPBACK)) {
5641 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5642 if (dev->features & NETIF_F_ALL_CSUM) {
5643 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5644 dev->features |= NETIF_F_NOCACHE_COPY;
5648 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5650 dev->vlan_features |= NETIF_F_HIGHDMA;
5652 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5653 ret = notifier_to_errno(ret);
5657 ret = netdev_register_kobject(dev);
5660 dev->reg_state = NETREG_REGISTERED;
5662 __netdev_update_features(dev);
5665 * Default initial state at registry is that the
5666 * device is present.
5669 set_bit(__LINK_STATE_PRESENT, &dev->state);
5671 linkwatch_init_dev(dev);
5673 dev_init_scheduler(dev);
5675 list_netdevice(dev);
5676 add_device_randomness(dev->dev_addr, dev->addr_len);
5678 /* Notify protocols, that a new device appeared. */
5679 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5680 ret = notifier_to_errno(ret);
5682 rollback_registered(dev);
5683 dev->reg_state = NETREG_UNREGISTERED;
5686 * Prevent userspace races by waiting until the network
5687 * device is fully setup before sending notifications.
5689 if (!dev->rtnl_link_ops ||
5690 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5691 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5697 if (dev->netdev_ops->ndo_uninit)
5698 dev->netdev_ops->ndo_uninit(dev);
5701 EXPORT_SYMBOL(register_netdevice);
5704 * init_dummy_netdev - init a dummy network device for NAPI
5705 * @dev: device to init
5707 * This takes a network device structure and initialize the minimum
5708 * amount of fields so it can be used to schedule NAPI polls without
5709 * registering a full blown interface. This is to be used by drivers
5710 * that need to tie several hardware interfaces to a single NAPI
5711 * poll scheduler due to HW limitations.
5713 int init_dummy_netdev(struct net_device *dev)
5715 /* Clear everything. Note we don't initialize spinlocks
5716 * are they aren't supposed to be taken by any of the
5717 * NAPI code and this dummy netdev is supposed to be
5718 * only ever used for NAPI polls
5720 memset(dev, 0, sizeof(struct net_device));
5722 /* make sure we BUG if trying to hit standard
5723 * register/unregister code path
5725 dev->reg_state = NETREG_DUMMY;
5727 /* NAPI wants this */
5728 INIT_LIST_HEAD(&dev->napi_list);
5730 /* a dummy interface is started by default */
5731 set_bit(__LINK_STATE_PRESENT, &dev->state);
5732 set_bit(__LINK_STATE_START, &dev->state);
5734 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5735 * because users of this 'device' dont need to change
5741 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5745 * register_netdev - register a network device
5746 * @dev: device to register
5748 * Take a completed network device structure and add it to the kernel
5749 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5750 * chain. 0 is returned on success. A negative errno code is returned
5751 * on a failure to set up the device, or if the name is a duplicate.
5753 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5754 * and expands the device name if you passed a format string to
5757 int register_netdev(struct net_device *dev)
5762 err = register_netdevice(dev);
5766 EXPORT_SYMBOL(register_netdev);
5768 int netdev_refcnt_read(const struct net_device *dev)
5772 for_each_possible_cpu(i)
5773 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5776 EXPORT_SYMBOL(netdev_refcnt_read);
5779 * netdev_wait_allrefs - wait until all references are gone.
5780 * @dev: target net_device
5782 * This is called when unregistering network devices.
5784 * Any protocol or device that holds a reference should register
5785 * for netdevice notification, and cleanup and put back the
5786 * reference if they receive an UNREGISTER event.
5787 * We can get stuck here if buggy protocols don't correctly
5790 static void netdev_wait_allrefs(struct net_device *dev)
5792 unsigned long rebroadcast_time, warning_time;
5795 linkwatch_forget_dev(dev);
5797 rebroadcast_time = warning_time = jiffies;
5798 refcnt = netdev_refcnt_read(dev);
5800 while (refcnt != 0) {
5801 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5804 /* Rebroadcast unregister notification */
5805 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5811 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5812 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5814 /* We must not have linkwatch events
5815 * pending on unregister. If this
5816 * happens, we simply run the queue
5817 * unscheduled, resulting in a noop
5820 linkwatch_run_queue();
5825 rebroadcast_time = jiffies;
5830 refcnt = netdev_refcnt_read(dev);
5832 if (time_after(jiffies, warning_time + 10 * HZ)) {
5833 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5835 warning_time = jiffies;
5844 * register_netdevice(x1);
5845 * register_netdevice(x2);
5847 * unregister_netdevice(y1);
5848 * unregister_netdevice(y2);
5854 * We are invoked by rtnl_unlock().
5855 * This allows us to deal with problems:
5856 * 1) We can delete sysfs objects which invoke hotplug
5857 * without deadlocking with linkwatch via keventd.
5858 * 2) Since we run with the RTNL semaphore not held, we can sleep
5859 * safely in order to wait for the netdev refcnt to drop to zero.
5861 * We must not return until all unregister events added during
5862 * the interval the lock was held have been completed.
5864 void netdev_run_todo(void)
5866 struct list_head list;
5868 /* Snapshot list, allow later requests */
5869 list_replace_init(&net_todo_list, &list);
5874 /* Wait for rcu callbacks to finish before next phase */
5875 if (!list_empty(&list))
5878 while (!list_empty(&list)) {
5879 struct net_device *dev
5880 = list_first_entry(&list, struct net_device, todo_list);
5881 list_del(&dev->todo_list);
5884 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5887 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5888 pr_err("network todo '%s' but state %d\n",
5889 dev->name, dev->reg_state);
5894 dev->reg_state = NETREG_UNREGISTERED;
5896 on_each_cpu(flush_backlog, dev, 1);
5898 netdev_wait_allrefs(dev);
5901 BUG_ON(netdev_refcnt_read(dev));
5902 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5903 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5904 WARN_ON(dev->dn_ptr);
5906 if (dev->destructor)
5907 dev->destructor(dev);
5909 /* Free network device */
5910 kobject_put(&dev->dev.kobj);
5914 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5915 * fields in the same order, with only the type differing.
5917 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5918 const struct net_device_stats *netdev_stats)
5920 #if BITS_PER_LONG == 64
5921 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5922 memcpy(stats64, netdev_stats, sizeof(*stats64));
5924 size_t i, n = sizeof(*stats64) / sizeof(u64);
5925 const unsigned long *src = (const unsigned long *)netdev_stats;
5926 u64 *dst = (u64 *)stats64;
5928 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5929 sizeof(*stats64) / sizeof(u64));
5930 for (i = 0; i < n; i++)
5934 EXPORT_SYMBOL(netdev_stats_to_stats64);
5937 * dev_get_stats - get network device statistics
5938 * @dev: device to get statistics from
5939 * @storage: place to store stats
5941 * Get network statistics from device. Return @storage.
5942 * The device driver may provide its own method by setting
5943 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5944 * otherwise the internal statistics structure is used.
5946 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5947 struct rtnl_link_stats64 *storage)
5949 const struct net_device_ops *ops = dev->netdev_ops;
5951 if (ops->ndo_get_stats64) {
5952 memset(storage, 0, sizeof(*storage));
5953 ops->ndo_get_stats64(dev, storage);
5954 } else if (ops->ndo_get_stats) {
5955 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5957 netdev_stats_to_stats64(storage, &dev->stats);
5959 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5962 EXPORT_SYMBOL(dev_get_stats);
5964 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5966 struct netdev_queue *queue = dev_ingress_queue(dev);
5968 #ifdef CONFIG_NET_CLS_ACT
5971 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5974 netdev_init_one_queue(dev, queue, NULL);
5975 queue->qdisc = &noop_qdisc;
5976 queue->qdisc_sleeping = &noop_qdisc;
5977 rcu_assign_pointer(dev->ingress_queue, queue);
5982 static const struct ethtool_ops default_ethtool_ops;
5985 * alloc_netdev_mqs - allocate network device
5986 * @sizeof_priv: size of private data to allocate space for
5987 * @name: device name format string
5988 * @setup: callback to initialize device
5989 * @txqs: the number of TX subqueues to allocate
5990 * @rxqs: the number of RX subqueues to allocate
5992 * Allocates a struct net_device with private data area for driver use
5993 * and performs basic initialization. Also allocates subquue structs
5994 * for each queue on the device.
5996 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5997 void (*setup)(struct net_device *),
5998 unsigned int txqs, unsigned int rxqs)
6000 struct net_device *dev;
6002 struct net_device *p;
6004 BUG_ON(strlen(name) >= sizeof(dev->name));
6007 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6013 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6018 alloc_size = sizeof(struct net_device);
6020 /* ensure 32-byte alignment of private area */
6021 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6022 alloc_size += sizeof_priv;
6024 /* ensure 32-byte alignment of whole construct */
6025 alloc_size += NETDEV_ALIGN - 1;
6027 p = kzalloc(alloc_size, GFP_KERNEL);
6029 pr_err("alloc_netdev: Unable to allocate device\n");
6033 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6034 dev->padded = (char *)dev - (char *)p;
6036 dev->pcpu_refcnt = alloc_percpu(int);
6037 if (!dev->pcpu_refcnt)
6040 if (dev_addr_init(dev))
6046 dev_net_set(dev, &init_net);
6048 dev->gso_max_size = GSO_MAX_SIZE;
6049 dev->gso_max_segs = GSO_MAX_SEGS;
6051 INIT_LIST_HEAD(&dev->napi_list);
6052 INIT_LIST_HEAD(&dev->unreg_list);
6053 INIT_LIST_HEAD(&dev->link_watch_list);
6054 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6057 dev->num_tx_queues = txqs;
6058 dev->real_num_tx_queues = txqs;
6059 if (netif_alloc_netdev_queues(dev))
6063 dev->num_rx_queues = rxqs;
6064 dev->real_num_rx_queues = rxqs;
6065 if (netif_alloc_rx_queues(dev))
6069 strcpy(dev->name, name);
6070 dev->group = INIT_NETDEV_GROUP;
6071 if (!dev->ethtool_ops)
6072 dev->ethtool_ops = &default_ethtool_ops;
6080 free_percpu(dev->pcpu_refcnt);
6090 EXPORT_SYMBOL(alloc_netdev_mqs);
6093 * free_netdev - free network device
6096 * This function does the last stage of destroying an allocated device
6097 * interface. The reference to the device object is released.
6098 * If this is the last reference then it will be freed.
6100 void free_netdev(struct net_device *dev)
6102 struct napi_struct *p, *n;
6104 release_net(dev_net(dev));
6111 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6113 /* Flush device addresses */
6114 dev_addr_flush(dev);
6116 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6119 free_percpu(dev->pcpu_refcnt);
6120 dev->pcpu_refcnt = NULL;
6122 /* Compatibility with error handling in drivers */
6123 if (dev->reg_state == NETREG_UNINITIALIZED) {
6124 kfree((char *)dev - dev->padded);
6128 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6129 dev->reg_state = NETREG_RELEASED;
6131 /* will free via device release */
6132 put_device(&dev->dev);
6134 EXPORT_SYMBOL(free_netdev);
6137 * synchronize_net - Synchronize with packet receive processing
6139 * Wait for packets currently being received to be done.
6140 * Does not block later packets from starting.
6142 void synchronize_net(void)
6145 if (rtnl_is_locked())
6146 synchronize_rcu_expedited();
6150 EXPORT_SYMBOL(synchronize_net);
6153 * unregister_netdevice_queue - remove device from the kernel
6157 * This function shuts down a device interface and removes it
6158 * from the kernel tables.
6159 * If head not NULL, device is queued to be unregistered later.
6161 * Callers must hold the rtnl semaphore. You may want
6162 * unregister_netdev() instead of this.
6165 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6170 list_move_tail(&dev->unreg_list, head);
6172 rollback_registered(dev);
6173 /* Finish processing unregister after unlock */
6177 EXPORT_SYMBOL(unregister_netdevice_queue);
6180 * unregister_netdevice_many - unregister many devices
6181 * @head: list of devices
6183 void unregister_netdevice_many(struct list_head *head)
6185 struct net_device *dev;
6187 if (!list_empty(head)) {
6188 rollback_registered_many(head);
6189 list_for_each_entry(dev, head, unreg_list)
6193 EXPORT_SYMBOL(unregister_netdevice_many);
6196 * unregister_netdev - remove device from the kernel
6199 * This function shuts down a device interface and removes it
6200 * from the kernel tables.
6202 * This is just a wrapper for unregister_netdevice that takes
6203 * the rtnl semaphore. In general you want to use this and not
6204 * unregister_netdevice.
6206 void unregister_netdev(struct net_device *dev)
6209 unregister_netdevice(dev);
6212 EXPORT_SYMBOL(unregister_netdev);
6215 * dev_change_net_namespace - move device to different nethost namespace
6217 * @net: network namespace
6218 * @pat: If not NULL name pattern to try if the current device name
6219 * is already taken in the destination network namespace.
6221 * This function shuts down a device interface and moves it
6222 * to a new network namespace. On success 0 is returned, on
6223 * a failure a netagive errno code is returned.
6225 * Callers must hold the rtnl semaphore.
6228 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6234 /* Don't allow namespace local devices to be moved. */
6236 if (dev->features & NETIF_F_NETNS_LOCAL)
6239 /* Ensure the device has been registrered */
6241 if (dev->reg_state != NETREG_REGISTERED)
6244 /* Get out if there is nothing todo */
6246 if (net_eq(dev_net(dev), net))
6249 /* Pick the destination device name, and ensure
6250 * we can use it in the destination network namespace.
6253 if (__dev_get_by_name(net, dev->name)) {
6254 /* We get here if we can't use the current device name */
6257 if (dev_get_valid_name(net, dev, pat) < 0)
6262 * And now a mini version of register_netdevice unregister_netdevice.
6265 /* If device is running close it first. */
6268 /* And unlink it from device chain */
6270 unlist_netdevice(dev);
6274 /* Shutdown queueing discipline. */
6277 /* Notify protocols, that we are about to destroy
6278 this device. They should clean all the things.
6280 Note that dev->reg_state stays at NETREG_REGISTERED.
6281 This is wanted because this way 8021q and macvlan know
6282 the device is just moving and can keep their slaves up.
6284 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6286 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6287 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6290 * Flush the unicast and multicast chains
6295 /* Actually switch the network namespace */
6296 dev_net_set(dev, net);
6298 /* If there is an ifindex conflict assign a new one */
6299 if (__dev_get_by_index(net, dev->ifindex)) {
6300 int iflink = (dev->iflink == dev->ifindex);
6301 dev->ifindex = dev_new_index(net);
6303 dev->iflink = dev->ifindex;
6306 /* Fixup kobjects */
6307 err = device_rename(&dev->dev, dev->name);
6310 /* Add the device back in the hashes */
6311 list_netdevice(dev);
6313 /* Notify protocols, that a new device appeared. */
6314 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6317 * Prevent userspace races by waiting until the network
6318 * device is fully setup before sending notifications.
6320 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6327 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6329 static int dev_cpu_callback(struct notifier_block *nfb,
6330 unsigned long action,
6333 struct sk_buff **list_skb;
6334 struct sk_buff *skb;
6335 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6336 struct softnet_data *sd, *oldsd;
6338 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6341 local_irq_disable();
6342 cpu = smp_processor_id();
6343 sd = &per_cpu(softnet_data, cpu);
6344 oldsd = &per_cpu(softnet_data, oldcpu);
6346 /* Find end of our completion_queue. */
6347 list_skb = &sd->completion_queue;
6349 list_skb = &(*list_skb)->next;
6350 /* Append completion queue from offline CPU. */
6351 *list_skb = oldsd->completion_queue;
6352 oldsd->completion_queue = NULL;
6354 /* Append output queue from offline CPU. */
6355 if (oldsd->output_queue) {
6356 *sd->output_queue_tailp = oldsd->output_queue;
6357 sd->output_queue_tailp = oldsd->output_queue_tailp;
6358 oldsd->output_queue = NULL;
6359 oldsd->output_queue_tailp = &oldsd->output_queue;
6361 /* Append NAPI poll list from offline CPU. */
6362 if (!list_empty(&oldsd->poll_list)) {
6363 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6364 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6367 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6370 /* Process offline CPU's input_pkt_queue */
6371 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6373 input_queue_head_incr(oldsd);
6375 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6377 input_queue_head_incr(oldsd);
6385 * netdev_increment_features - increment feature set by one
6386 * @all: current feature set
6387 * @one: new feature set
6388 * @mask: mask feature set
6390 * Computes a new feature set after adding a device with feature set
6391 * @one to the master device with current feature set @all. Will not
6392 * enable anything that is off in @mask. Returns the new feature set.
6394 netdev_features_t netdev_increment_features(netdev_features_t all,
6395 netdev_features_t one, netdev_features_t mask)
6397 if (mask & NETIF_F_GEN_CSUM)
6398 mask |= NETIF_F_ALL_CSUM;
6399 mask |= NETIF_F_VLAN_CHALLENGED;
6401 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6402 all &= one | ~NETIF_F_ALL_FOR_ALL;
6404 /* If one device supports hw checksumming, set for all. */
6405 if (all & NETIF_F_GEN_CSUM)
6406 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6410 EXPORT_SYMBOL(netdev_increment_features);
6412 static struct hlist_head *netdev_create_hash(void)
6415 struct hlist_head *hash;
6417 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6419 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6420 INIT_HLIST_HEAD(&hash[i]);
6425 /* Initialize per network namespace state */
6426 static int __net_init netdev_init(struct net *net)
6428 if (net != &init_net)
6429 INIT_LIST_HEAD(&net->dev_base_head);
6431 net->dev_name_head = netdev_create_hash();
6432 if (net->dev_name_head == NULL)
6435 net->dev_index_head = netdev_create_hash();
6436 if (net->dev_index_head == NULL)
6442 kfree(net->dev_name_head);
6448 * netdev_drivername - network driver for the device
6449 * @dev: network device
6451 * Determine network driver for device.
6453 const char *netdev_drivername(const struct net_device *dev)
6455 const struct device_driver *driver;
6456 const struct device *parent;
6457 const char *empty = "";
6459 parent = dev->dev.parent;
6463 driver = parent->driver;
6464 if (driver && driver->name)
6465 return driver->name;
6469 static int __netdev_printk(const char *level, const struct net_device *dev,
6470 struct va_format *vaf)
6474 if (dev && dev->dev.parent) {
6475 r = dev_printk_emit(level[1] - '0',
6478 dev_driver_string(dev->dev.parent),
6479 dev_name(dev->dev.parent),
6480 netdev_name(dev), vaf);
6482 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6484 r = printk("%s(NULL net_device): %pV", level, vaf);
6490 int netdev_printk(const char *level, const struct net_device *dev,
6491 const char *format, ...)
6493 struct va_format vaf;
6497 va_start(args, format);
6502 r = __netdev_printk(level, dev, &vaf);
6508 EXPORT_SYMBOL(netdev_printk);
6510 #define define_netdev_printk_level(func, level) \
6511 int func(const struct net_device *dev, const char *fmt, ...) \
6514 struct va_format vaf; \
6517 va_start(args, fmt); \
6522 r = __netdev_printk(level, dev, &vaf); \
6528 EXPORT_SYMBOL(func);
6530 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6531 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6532 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6533 define_netdev_printk_level(netdev_err, KERN_ERR);
6534 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6535 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6536 define_netdev_printk_level(netdev_info, KERN_INFO);
6538 static void __net_exit netdev_exit(struct net *net)
6540 kfree(net->dev_name_head);
6541 kfree(net->dev_index_head);
6544 static struct pernet_operations __net_initdata netdev_net_ops = {
6545 .init = netdev_init,
6546 .exit = netdev_exit,
6549 static void __net_exit default_device_exit(struct net *net)
6551 struct net_device *dev, *aux;
6553 * Push all migratable network devices back to the
6554 * initial network namespace
6557 for_each_netdev_safe(net, dev, aux) {
6559 char fb_name[IFNAMSIZ];
6561 /* Ignore unmoveable devices (i.e. loopback) */
6562 if (dev->features & NETIF_F_NETNS_LOCAL)
6565 /* Leave virtual devices for the generic cleanup */
6566 if (dev->rtnl_link_ops)
6569 /* Push remaining network devices to init_net */
6570 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6571 err = dev_change_net_namespace(dev, &init_net, fb_name);
6573 pr_emerg("%s: failed to move %s to init_net: %d\n",
6574 __func__, dev->name, err);
6581 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6583 /* At exit all network devices most be removed from a network
6584 * namespace. Do this in the reverse order of registration.
6585 * Do this across as many network namespaces as possible to
6586 * improve batching efficiency.
6588 struct net_device *dev;
6590 LIST_HEAD(dev_kill_list);
6593 list_for_each_entry(net, net_list, exit_list) {
6594 for_each_netdev_reverse(net, dev) {
6595 if (dev->rtnl_link_ops)
6596 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6598 unregister_netdevice_queue(dev, &dev_kill_list);
6601 unregister_netdevice_many(&dev_kill_list);
6602 list_del(&dev_kill_list);
6606 static struct pernet_operations __net_initdata default_device_ops = {
6607 .exit = default_device_exit,
6608 .exit_batch = default_device_exit_batch,
6612 * Initialize the DEV module. At boot time this walks the device list and
6613 * unhooks any devices that fail to initialise (normally hardware not
6614 * present) and leaves us with a valid list of present and active devices.
6619 * This is called single threaded during boot, so no need
6620 * to take the rtnl semaphore.
6622 static int __init net_dev_init(void)
6624 int i, rc = -ENOMEM;
6626 BUG_ON(!dev_boot_phase);
6628 if (dev_proc_init())
6631 if (netdev_kobject_init())
6634 INIT_LIST_HEAD(&ptype_all);
6635 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6636 INIT_LIST_HEAD(&ptype_base[i]);
6638 if (register_pernet_subsys(&netdev_net_ops))
6642 * Initialise the packet receive queues.
6645 for_each_possible_cpu(i) {
6646 struct softnet_data *sd = &per_cpu(softnet_data, i);
6648 memset(sd, 0, sizeof(*sd));
6649 skb_queue_head_init(&sd->input_pkt_queue);
6650 skb_queue_head_init(&sd->process_queue);
6651 sd->completion_queue = NULL;
6652 INIT_LIST_HEAD(&sd->poll_list);
6653 sd->output_queue = NULL;
6654 sd->output_queue_tailp = &sd->output_queue;
6656 sd->csd.func = rps_trigger_softirq;
6662 sd->backlog.poll = process_backlog;
6663 sd->backlog.weight = weight_p;
6664 sd->backlog.gro_list = NULL;
6665 sd->backlog.gro_count = 0;
6670 /* The loopback device is special if any other network devices
6671 * is present in a network namespace the loopback device must
6672 * be present. Since we now dynamically allocate and free the
6673 * loopback device ensure this invariant is maintained by
6674 * keeping the loopback device as the first device on the
6675 * list of network devices. Ensuring the loopback devices
6676 * is the first device that appears and the last network device
6679 if (register_pernet_device(&loopback_net_ops))
6682 if (register_pernet_device(&default_device_ops))
6685 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6686 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6688 hotcpu_notifier(dev_cpu_callback, 0);
6696 subsys_initcall(net_dev_init);
6698 static int __init initialize_hashrnd(void)
6700 get_random_bytes(&hashrnd, sizeof(hashrnd));
6704 late_initcall_sync(initialize_hashrnd);