2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/net_tstamp.h>
136 #include <linux/static_key.h>
137 #include <net/flow_keys.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 * The list of packet types we will receive (as opposed to discard)
149 * and the routines to invoke.
151 * Why 16. Because with 16 the only overlap we get on a hash of the
152 * low nibble of the protocol value is RARP/SNAP/X.25.
154 * NOTE: That is no longer true with the addition of VLAN tags. Not
155 * sure which should go first, but I bet it won't make much
156 * difference if we are running VLANs. The good news is that
157 * this protocol won't be in the list unless compiled in, so
158 * the average user (w/out VLANs) will not be adversely affected.
175 #define PTYPE_HASH_SIZE (16)
176 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
178 static DEFINE_SPINLOCK(ptype_lock);
179 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
180 static struct list_head ptype_all __read_mostly; /* Taps */
183 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
186 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
188 * Writers must hold the rtnl semaphore while they loop through the
189 * dev_base_head list, and hold dev_base_lock for writing when they do the
190 * actual updates. This allows pure readers to access the list even
191 * while a writer is preparing to update it.
193 * To put it another way, dev_base_lock is held for writing only to
194 * protect against pure readers; the rtnl semaphore provides the
195 * protection against other writers.
197 * See, for example usages, register_netdevice() and
198 * unregister_netdevice(), which must be called with the rtnl
201 DEFINE_RWLOCK(dev_base_lock);
202 EXPORT_SYMBOL(dev_base_lock);
204 static inline void dev_base_seq_inc(struct net *net)
206 while (++net->dev_base_seq == 0);
209 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
211 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
213 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
216 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
218 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
221 static inline void rps_lock(struct softnet_data *sd)
224 spin_lock(&sd->input_pkt_queue.lock);
228 static inline void rps_unlock(struct softnet_data *sd)
231 spin_unlock(&sd->input_pkt_queue.lock);
235 /* Device list insertion */
236 static int list_netdevice(struct net_device *dev)
238 struct net *net = dev_net(dev);
242 write_lock_bh(&dev_base_lock);
243 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
244 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
245 hlist_add_head_rcu(&dev->index_hlist,
246 dev_index_hash(net, dev->ifindex));
247 write_unlock_bh(&dev_base_lock);
249 dev_base_seq_inc(net);
254 /* Device list removal
255 * caller must respect a RCU grace period before freeing/reusing dev
257 static void unlist_netdevice(struct net_device *dev)
261 /* Unlink dev from the device chain */
262 write_lock_bh(&dev_base_lock);
263 list_del_rcu(&dev->dev_list);
264 hlist_del_rcu(&dev->name_hlist);
265 hlist_del_rcu(&dev->index_hlist);
266 write_unlock_bh(&dev_base_lock);
268 dev_base_seq_inc(dev_net(dev));
275 static RAW_NOTIFIER_HEAD(netdev_chain);
278 * Device drivers call our routines to queue packets here. We empty the
279 * queue in the local softnet handler.
282 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
283 EXPORT_PER_CPU_SYMBOL(softnet_data);
285 #ifdef CONFIG_LOCKDEP
287 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
288 * according to dev->type
290 static const unsigned short netdev_lock_type[] =
291 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
292 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
293 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
294 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
295 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
296 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
297 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
298 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
299 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
300 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
301 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
302 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
303 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
304 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
305 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
307 static const char *const netdev_lock_name[] =
308 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
309 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
310 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
311 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
312 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
313 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
314 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
315 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
316 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
317 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
318 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
319 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
320 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
321 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
322 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
324 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
325 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
327 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
331 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
332 if (netdev_lock_type[i] == dev_type)
334 /* the last key is used by default */
335 return ARRAY_SIZE(netdev_lock_type) - 1;
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
343 i = netdev_lock_pos(dev_type);
344 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
345 netdev_lock_name[i]);
348 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
352 i = netdev_lock_pos(dev->type);
353 lockdep_set_class_and_name(&dev->addr_list_lock,
354 &netdev_addr_lock_key[i],
355 netdev_lock_name[i]);
358 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
359 unsigned short dev_type)
362 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
367 /*******************************************************************************
369 Protocol management and registration routines
371 *******************************************************************************/
374 * Add a protocol ID to the list. Now that the input handler is
375 * smarter we can dispense with all the messy stuff that used to be
378 * BEWARE!!! Protocol handlers, mangling input packets,
379 * MUST BE last in hash buckets and checking protocol handlers
380 * MUST start from promiscuous ptype_all chain in net_bh.
381 * It is true now, do not change it.
382 * Explanation follows: if protocol handler, mangling packet, will
383 * be the first on list, it is not able to sense, that packet
384 * is cloned and should be copied-on-write, so that it will
385 * change it and subsequent readers will get broken packet.
389 static inline struct list_head *ptype_head(const struct packet_type *pt)
391 if (pt->type == htons(ETH_P_ALL))
394 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
398 * dev_add_pack - add packet handler
399 * @pt: packet type declaration
401 * Add a protocol handler to the networking stack. The passed &packet_type
402 * is linked into kernel lists and may not be freed until it has been
403 * removed from the kernel lists.
405 * This call does not sleep therefore it can not
406 * guarantee all CPU's that are in middle of receiving packets
407 * will see the new packet type (until the next received packet).
410 void dev_add_pack(struct packet_type *pt)
412 struct list_head *head = ptype_head(pt);
414 spin_lock(&ptype_lock);
415 list_add_rcu(&pt->list, head);
416 spin_unlock(&ptype_lock);
418 EXPORT_SYMBOL(dev_add_pack);
421 * __dev_remove_pack - remove packet handler
422 * @pt: packet type declaration
424 * Remove a protocol handler that was previously added to the kernel
425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
426 * from the kernel lists and can be freed or reused once this function
429 * The packet type might still be in use by receivers
430 * and must not be freed until after all the CPU's have gone
431 * through a quiescent state.
433 void __dev_remove_pack(struct packet_type *pt)
435 struct list_head *head = ptype_head(pt);
436 struct packet_type *pt1;
438 spin_lock(&ptype_lock);
440 list_for_each_entry(pt1, head, list) {
442 list_del_rcu(&pt->list);
447 pr_warn("dev_remove_pack: %p not found\n", pt);
449 spin_unlock(&ptype_lock);
451 EXPORT_SYMBOL(__dev_remove_pack);
454 * dev_remove_pack - remove packet handler
455 * @pt: packet type declaration
457 * Remove a protocol handler that was previously added to the kernel
458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
459 * from the kernel lists and can be freed or reused once this function
462 * This call sleeps to guarantee that no CPU is looking at the packet
465 void dev_remove_pack(struct packet_type *pt)
467 __dev_remove_pack(pt);
471 EXPORT_SYMBOL(dev_remove_pack);
473 /******************************************************************************
475 Device Boot-time Settings Routines
477 *******************************************************************************/
479 /* Boot time configuration table */
480 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
483 * netdev_boot_setup_add - add new setup entry
484 * @name: name of the device
485 * @map: configured settings for the device
487 * Adds new setup entry to the dev_boot_setup list. The function
488 * returns 0 on error and 1 on success. This is a generic routine to
491 static int netdev_boot_setup_add(char *name, struct ifmap *map)
493 struct netdev_boot_setup *s;
497 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
498 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
499 memset(s[i].name, 0, sizeof(s[i].name));
500 strlcpy(s[i].name, name, IFNAMSIZ);
501 memcpy(&s[i].map, map, sizeof(s[i].map));
506 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
510 * netdev_boot_setup_check - check boot time settings
511 * @dev: the netdevice
513 * Check boot time settings for the device.
514 * The found settings are set for the device to be used
515 * later in the device probing.
516 * Returns 0 if no settings found, 1 if they are.
518 int netdev_boot_setup_check(struct net_device *dev)
520 struct netdev_boot_setup *s = dev_boot_setup;
523 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
524 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
525 !strcmp(dev->name, s[i].name)) {
526 dev->irq = s[i].map.irq;
527 dev->base_addr = s[i].map.base_addr;
528 dev->mem_start = s[i].map.mem_start;
529 dev->mem_end = s[i].map.mem_end;
535 EXPORT_SYMBOL(netdev_boot_setup_check);
539 * netdev_boot_base - get address from boot time settings
540 * @prefix: prefix for network device
541 * @unit: id for network device
543 * Check boot time settings for the base address of device.
544 * The found settings are set for the device to be used
545 * later in the device probing.
546 * Returns 0 if no settings found.
548 unsigned long netdev_boot_base(const char *prefix, int unit)
550 const struct netdev_boot_setup *s = dev_boot_setup;
554 sprintf(name, "%s%d", prefix, unit);
557 * If device already registered then return base of 1
558 * to indicate not to probe for this interface
560 if (__dev_get_by_name(&init_net, name))
563 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
564 if (!strcmp(name, s[i].name))
565 return s[i].map.base_addr;
570 * Saves at boot time configured settings for any netdevice.
572 int __init netdev_boot_setup(char *str)
577 str = get_options(str, ARRAY_SIZE(ints), ints);
582 memset(&map, 0, sizeof(map));
586 map.base_addr = ints[2];
588 map.mem_start = ints[3];
590 map.mem_end = ints[4];
592 /* Add new entry to the list */
593 return netdev_boot_setup_add(str, &map);
596 __setup("netdev=", netdev_boot_setup);
598 /*******************************************************************************
600 Device Interface Subroutines
602 *******************************************************************************/
605 * __dev_get_by_name - find a device by its name
606 * @net: the applicable net namespace
607 * @name: name to find
609 * Find an interface by name. Must be called under RTNL semaphore
610 * or @dev_base_lock. If the name is found a pointer to the device
611 * is returned. If the name is not found then %NULL is returned. The
612 * reference counters are not incremented so the caller must be
613 * careful with locks.
616 struct net_device *__dev_get_by_name(struct net *net, const char *name)
618 struct hlist_node *p;
619 struct net_device *dev;
620 struct hlist_head *head = dev_name_hash(net, name);
622 hlist_for_each_entry(dev, p, head, name_hlist)
623 if (!strncmp(dev->name, name, IFNAMSIZ))
628 EXPORT_SYMBOL(__dev_get_by_name);
631 * dev_get_by_name_rcu - find a device by its name
632 * @net: the applicable net namespace
633 * @name: name to find
635 * Find an interface by name.
636 * If the name is found a pointer to the device is returned.
637 * If the name is not found then %NULL is returned.
638 * The reference counters are not incremented so the caller must be
639 * careful with locks. The caller must hold RCU lock.
642 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
644 struct hlist_node *p;
645 struct net_device *dev;
646 struct hlist_head *head = dev_name_hash(net, name);
648 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
649 if (!strncmp(dev->name, name, IFNAMSIZ))
654 EXPORT_SYMBOL(dev_get_by_name_rcu);
657 * dev_get_by_name - find a device by its name
658 * @net: the applicable net namespace
659 * @name: name to find
661 * Find an interface by name. This can be called from any
662 * context and does its own locking. The returned handle has
663 * the usage count incremented and the caller must use dev_put() to
664 * release it when it is no longer needed. %NULL is returned if no
665 * matching device is found.
668 struct net_device *dev_get_by_name(struct net *net, const char *name)
670 struct net_device *dev;
673 dev = dev_get_by_name_rcu(net, name);
679 EXPORT_SYMBOL(dev_get_by_name);
682 * __dev_get_by_index - find a device by its ifindex
683 * @net: the applicable net namespace
684 * @ifindex: index of device
686 * Search for an interface by index. Returns %NULL if the device
687 * is not found or a pointer to the device. The device has not
688 * had its reference counter increased so the caller must be careful
689 * about locking. The caller must hold either the RTNL semaphore
693 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
695 struct hlist_node *p;
696 struct net_device *dev;
697 struct hlist_head *head = dev_index_hash(net, ifindex);
699 hlist_for_each_entry(dev, p, head, index_hlist)
700 if (dev->ifindex == ifindex)
705 EXPORT_SYMBOL(__dev_get_by_index);
708 * dev_get_by_index_rcu - find a device by its ifindex
709 * @net: the applicable net namespace
710 * @ifindex: index of device
712 * Search for an interface by index. Returns %NULL if the device
713 * is not found or a pointer to the device. The device has not
714 * had its reference counter increased so the caller must be careful
715 * about locking. The caller must hold RCU lock.
718 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
720 struct hlist_node *p;
721 struct net_device *dev;
722 struct hlist_head *head = dev_index_hash(net, ifindex);
724 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
725 if (dev->ifindex == ifindex)
730 EXPORT_SYMBOL(dev_get_by_index_rcu);
734 * dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns NULL if the device
739 * is not found or a pointer to the device. The device returned has
740 * had a reference added and the pointer is safe until the user calls
741 * dev_put to indicate they have finished with it.
744 struct net_device *dev_get_by_index(struct net *net, int ifindex)
746 struct net_device *dev;
749 dev = dev_get_by_index_rcu(net, ifindex);
755 EXPORT_SYMBOL(dev_get_by_index);
758 * dev_getbyhwaddr_rcu - find a device by its hardware address
759 * @net: the applicable net namespace
760 * @type: media type of device
761 * @ha: hardware address
763 * Search for an interface by MAC address. Returns NULL if the device
764 * is not found or a pointer to the device.
765 * The caller must hold RCU or RTNL.
766 * The returned device has not had its ref count increased
767 * and the caller must therefore be careful about locking
771 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
774 struct net_device *dev;
776 for_each_netdev_rcu(net, dev)
777 if (dev->type == type &&
778 !memcmp(dev->dev_addr, ha, dev->addr_len))
783 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
785 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
787 struct net_device *dev;
790 for_each_netdev(net, dev)
791 if (dev->type == type)
796 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
798 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
800 struct net_device *dev, *ret = NULL;
803 for_each_netdev_rcu(net, dev)
804 if (dev->type == type) {
812 EXPORT_SYMBOL(dev_getfirstbyhwtype);
815 * dev_get_by_flags_rcu - find any device with given flags
816 * @net: the applicable net namespace
817 * @if_flags: IFF_* values
818 * @mask: bitmask of bits in if_flags to check
820 * Search for any interface with the given flags. Returns NULL if a device
821 * is not found or a pointer to the device. Must be called inside
822 * rcu_read_lock(), and result refcount is unchanged.
825 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
828 struct net_device *dev, *ret;
831 for_each_netdev_rcu(net, dev) {
832 if (((dev->flags ^ if_flags) & mask) == 0) {
839 EXPORT_SYMBOL(dev_get_by_flags_rcu);
842 * dev_valid_name - check if name is okay for network device
845 * Network device names need to be valid file names to
846 * to allow sysfs to work. We also disallow any kind of
849 bool dev_valid_name(const char *name)
853 if (strlen(name) >= IFNAMSIZ)
855 if (!strcmp(name, ".") || !strcmp(name, ".."))
859 if (*name == '/' || isspace(*name))
865 EXPORT_SYMBOL(dev_valid_name);
868 * __dev_alloc_name - allocate a name for a device
869 * @net: network namespace to allocate the device name in
870 * @name: name format string
871 * @buf: scratch buffer and result name string
873 * Passed a format string - eg "lt%d" it will try and find a suitable
874 * id. It scans list of devices to build up a free map, then chooses
875 * the first empty slot. The caller must hold the dev_base or rtnl lock
876 * while allocating the name and adding the device in order to avoid
878 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
879 * Returns the number of the unit assigned or a negative errno code.
882 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
886 const int max_netdevices = 8*PAGE_SIZE;
887 unsigned long *inuse;
888 struct net_device *d;
890 p = strnchr(name, IFNAMSIZ-1, '%');
893 * Verify the string as this thing may have come from
894 * the user. There must be either one "%d" and no other "%"
897 if (p[1] != 'd' || strchr(p + 2, '%'))
900 /* Use one page as a bit array of possible slots */
901 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
905 for_each_netdev(net, d) {
906 if (!sscanf(d->name, name, &i))
908 if (i < 0 || i >= max_netdevices)
911 /* avoid cases where sscanf is not exact inverse of printf */
912 snprintf(buf, IFNAMSIZ, name, i);
913 if (!strncmp(buf, d->name, IFNAMSIZ))
917 i = find_first_zero_bit(inuse, max_netdevices);
918 free_page((unsigned long) inuse);
922 snprintf(buf, IFNAMSIZ, name, i);
923 if (!__dev_get_by_name(net, buf))
926 /* It is possible to run out of possible slots
927 * when the name is long and there isn't enough space left
928 * for the digits, or if all bits are used.
934 * dev_alloc_name - allocate a name for a device
936 * @name: name format string
938 * Passed a format string - eg "lt%d" it will try and find a suitable
939 * id. It scans list of devices to build up a free map, then chooses
940 * the first empty slot. The caller must hold the dev_base or rtnl lock
941 * while allocating the name and adding the device in order to avoid
943 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
944 * Returns the number of the unit assigned or a negative errno code.
947 int dev_alloc_name(struct net_device *dev, const char *name)
953 BUG_ON(!dev_net(dev));
955 ret = __dev_alloc_name(net, name, buf);
957 strlcpy(dev->name, buf, IFNAMSIZ);
960 EXPORT_SYMBOL(dev_alloc_name);
962 static int dev_get_valid_name(struct net_device *dev, const char *name)
966 BUG_ON(!dev_net(dev));
969 if (!dev_valid_name(name))
972 if (strchr(name, '%'))
973 return dev_alloc_name(dev, name);
974 else if (__dev_get_by_name(net, name))
976 else if (dev->name != name)
977 strlcpy(dev->name, name, IFNAMSIZ);
983 * dev_change_name - change name of a device
985 * @newname: name (or format string) must be at least IFNAMSIZ
987 * Change name of a device, can pass format strings "eth%d".
990 int dev_change_name(struct net_device *dev, const char *newname)
992 char oldname[IFNAMSIZ];
998 BUG_ON(!dev_net(dev));
1001 if (dev->flags & IFF_UP)
1004 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1007 memcpy(oldname, dev->name, IFNAMSIZ);
1009 err = dev_get_valid_name(dev, newname);
1014 ret = device_rename(&dev->dev, dev->name);
1016 memcpy(dev->name, oldname, IFNAMSIZ);
1020 write_lock_bh(&dev_base_lock);
1021 hlist_del_rcu(&dev->name_hlist);
1022 write_unlock_bh(&dev_base_lock);
1026 write_lock_bh(&dev_base_lock);
1027 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1028 write_unlock_bh(&dev_base_lock);
1030 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1031 ret = notifier_to_errno(ret);
1034 /* err >= 0 after dev_alloc_name() or stores the first errno */
1037 memcpy(dev->name, oldname, IFNAMSIZ);
1040 pr_err("%s: name change rollback failed: %d\n",
1049 * dev_set_alias - change ifalias of a device
1051 * @alias: name up to IFALIASZ
1052 * @len: limit of bytes to copy from info
1054 * Set ifalias for a device,
1056 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1062 if (len >= IFALIASZ)
1067 kfree(dev->ifalias);
1068 dev->ifalias = NULL;
1073 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1076 dev->ifalias = new_ifalias;
1078 strlcpy(dev->ifalias, alias, len+1);
1084 * netdev_features_change - device changes features
1085 * @dev: device to cause notification
1087 * Called to indicate a device has changed features.
1089 void netdev_features_change(struct net_device *dev)
1091 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1093 EXPORT_SYMBOL(netdev_features_change);
1096 * netdev_state_change - device changes state
1097 * @dev: device to cause notification
1099 * Called to indicate a device has changed state. This function calls
1100 * the notifier chains for netdev_chain and sends a NEWLINK message
1101 * to the routing socket.
1103 void netdev_state_change(struct net_device *dev)
1105 if (dev->flags & IFF_UP) {
1106 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1107 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1110 EXPORT_SYMBOL(netdev_state_change);
1113 * netdev_notify_peers - notify network peers about existence of @dev
1114 * @dev: network device
1116 * Generate traffic such that interested network peers are aware of
1117 * @dev, such as by generating a gratuitous ARP. This may be used when
1118 * a device wants to inform the rest of the network about some sort of
1119 * reconfiguration such as a failover event or virtual machine
1122 void netdev_notify_peers(struct net_device *dev)
1125 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1128 EXPORT_SYMBOL(netdev_notify_peers);
1131 * dev_load - load a network module
1132 * @net: the applicable net namespace
1133 * @name: name of interface
1135 * If a network interface is not present and the process has suitable
1136 * privileges this function loads the module. If module loading is not
1137 * available in this kernel then it becomes a nop.
1140 void dev_load(struct net *net, const char *name)
1142 struct net_device *dev;
1146 dev = dev_get_by_name_rcu(net, name);
1150 if (no_module && capable(CAP_NET_ADMIN))
1151 no_module = request_module("netdev-%s", name);
1152 if (no_module && capable(CAP_SYS_MODULE)) {
1153 if (!request_module("%s", name))
1154 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1158 EXPORT_SYMBOL(dev_load);
1160 static int __dev_open(struct net_device *dev)
1162 const struct net_device_ops *ops = dev->netdev_ops;
1167 if (!netif_device_present(dev))
1170 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1171 ret = notifier_to_errno(ret);
1175 set_bit(__LINK_STATE_START, &dev->state);
1177 if (ops->ndo_validate_addr)
1178 ret = ops->ndo_validate_addr(dev);
1180 if (!ret && ops->ndo_open)
1181 ret = ops->ndo_open(dev);
1184 clear_bit(__LINK_STATE_START, &dev->state);
1186 dev->flags |= IFF_UP;
1187 net_dmaengine_get();
1188 dev_set_rx_mode(dev);
1190 add_device_randomness(dev->dev_addr, dev->addr_len);
1197 * dev_open - prepare an interface for use.
1198 * @dev: device to open
1200 * Takes a device from down to up state. The device's private open
1201 * function is invoked and then the multicast lists are loaded. Finally
1202 * the device is moved into the up state and a %NETDEV_UP message is
1203 * sent to the netdev notifier chain.
1205 * Calling this function on an active interface is a nop. On a failure
1206 * a negative errno code is returned.
1208 int dev_open(struct net_device *dev)
1212 if (dev->flags & IFF_UP)
1215 ret = __dev_open(dev);
1219 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1220 call_netdevice_notifiers(NETDEV_UP, dev);
1224 EXPORT_SYMBOL(dev_open);
1226 static int __dev_close_many(struct list_head *head)
1228 struct net_device *dev;
1233 list_for_each_entry(dev, head, unreg_list) {
1234 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1236 clear_bit(__LINK_STATE_START, &dev->state);
1238 /* Synchronize to scheduled poll. We cannot touch poll list, it
1239 * can be even on different cpu. So just clear netif_running().
1241 * dev->stop() will invoke napi_disable() on all of it's
1242 * napi_struct instances on this device.
1244 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1247 dev_deactivate_many(head);
1249 list_for_each_entry(dev, head, unreg_list) {
1250 const struct net_device_ops *ops = dev->netdev_ops;
1253 * Call the device specific close. This cannot fail.
1254 * Only if device is UP
1256 * We allow it to be called even after a DETACH hot-plug
1262 dev->flags &= ~IFF_UP;
1263 net_dmaengine_put();
1269 static int __dev_close(struct net_device *dev)
1274 list_add(&dev->unreg_list, &single);
1275 retval = __dev_close_many(&single);
1280 static int dev_close_many(struct list_head *head)
1282 struct net_device *dev, *tmp;
1283 LIST_HEAD(tmp_list);
1285 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1286 if (!(dev->flags & IFF_UP))
1287 list_move(&dev->unreg_list, &tmp_list);
1289 __dev_close_many(head);
1291 list_for_each_entry(dev, head, unreg_list) {
1292 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1293 call_netdevice_notifiers(NETDEV_DOWN, dev);
1296 /* rollback_registered_many needs the complete original list */
1297 list_splice(&tmp_list, head);
1302 * dev_close - shutdown an interface.
1303 * @dev: device to shutdown
1305 * This function moves an active device into down state. A
1306 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1307 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1310 int dev_close(struct net_device *dev)
1312 if (dev->flags & IFF_UP) {
1315 list_add(&dev->unreg_list, &single);
1316 dev_close_many(&single);
1321 EXPORT_SYMBOL(dev_close);
1325 * dev_disable_lro - disable Large Receive Offload on a device
1328 * Disable Large Receive Offload (LRO) on a net device. Must be
1329 * called under RTNL. This is needed if received packets may be
1330 * forwarded to another interface.
1332 void dev_disable_lro(struct net_device *dev)
1335 * If we're trying to disable lro on a vlan device
1336 * use the underlying physical device instead
1338 if (is_vlan_dev(dev))
1339 dev = vlan_dev_real_dev(dev);
1341 dev->wanted_features &= ~NETIF_F_LRO;
1342 netdev_update_features(dev);
1344 if (unlikely(dev->features & NETIF_F_LRO))
1345 netdev_WARN(dev, "failed to disable LRO!\n");
1347 EXPORT_SYMBOL(dev_disable_lro);
1350 static int dev_boot_phase = 1;
1353 * register_netdevice_notifier - register a network notifier block
1356 * Register a notifier to be called when network device events occur.
1357 * The notifier passed is linked into the kernel structures and must
1358 * not be reused until it has been unregistered. A negative errno code
1359 * is returned on a failure.
1361 * When registered all registration and up events are replayed
1362 * to the new notifier to allow device to have a race free
1363 * view of the network device list.
1366 int register_netdevice_notifier(struct notifier_block *nb)
1368 struct net_device *dev;
1369 struct net_device *last;
1374 err = raw_notifier_chain_register(&netdev_chain, nb);
1380 for_each_netdev(net, dev) {
1381 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1382 err = notifier_to_errno(err);
1386 if (!(dev->flags & IFF_UP))
1389 nb->notifier_call(nb, NETDEV_UP, dev);
1400 for_each_netdev(net, dev) {
1404 if (dev->flags & IFF_UP) {
1405 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1406 nb->notifier_call(nb, NETDEV_DOWN, dev);
1408 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1413 raw_notifier_chain_unregister(&netdev_chain, nb);
1416 EXPORT_SYMBOL(register_netdevice_notifier);
1419 * unregister_netdevice_notifier - unregister a network notifier block
1422 * Unregister a notifier previously registered by
1423 * register_netdevice_notifier(). The notifier is unlinked into the
1424 * kernel structures and may then be reused. A negative errno code
1425 * is returned on a failure.
1427 * After unregistering unregister and down device events are synthesized
1428 * for all devices on the device list to the removed notifier to remove
1429 * the need for special case cleanup code.
1432 int unregister_netdevice_notifier(struct notifier_block *nb)
1434 struct net_device *dev;
1439 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1444 for_each_netdev(net, dev) {
1445 if (dev->flags & IFF_UP) {
1446 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1447 nb->notifier_call(nb, NETDEV_DOWN, dev);
1449 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1456 EXPORT_SYMBOL(unregister_netdevice_notifier);
1459 * call_netdevice_notifiers - call all network notifier blocks
1460 * @val: value passed unmodified to notifier function
1461 * @dev: net_device pointer passed unmodified to notifier function
1463 * Call all network notifier blocks. Parameters and return value
1464 * are as for raw_notifier_call_chain().
1467 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1470 return raw_notifier_call_chain(&netdev_chain, val, dev);
1472 EXPORT_SYMBOL(call_netdevice_notifiers);
1474 static struct static_key netstamp_needed __read_mostly;
1475 #ifdef HAVE_JUMP_LABEL
1476 /* We are not allowed to call static_key_slow_dec() from irq context
1477 * If net_disable_timestamp() is called from irq context, defer the
1478 * static_key_slow_dec() calls.
1480 static atomic_t netstamp_needed_deferred;
1483 void net_enable_timestamp(void)
1485 #ifdef HAVE_JUMP_LABEL
1486 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1490 static_key_slow_dec(&netstamp_needed);
1494 WARN_ON(in_interrupt());
1495 static_key_slow_inc(&netstamp_needed);
1497 EXPORT_SYMBOL(net_enable_timestamp);
1499 void net_disable_timestamp(void)
1501 #ifdef HAVE_JUMP_LABEL
1502 if (in_interrupt()) {
1503 atomic_inc(&netstamp_needed_deferred);
1507 static_key_slow_dec(&netstamp_needed);
1509 EXPORT_SYMBOL(net_disable_timestamp);
1511 static inline void net_timestamp_set(struct sk_buff *skb)
1513 skb->tstamp.tv64 = 0;
1514 if (static_key_false(&netstamp_needed))
1515 __net_timestamp(skb);
1518 #define net_timestamp_check(COND, SKB) \
1519 if (static_key_false(&netstamp_needed)) { \
1520 if ((COND) && !(SKB)->tstamp.tv64) \
1521 __net_timestamp(SKB); \
1524 static int net_hwtstamp_validate(struct ifreq *ifr)
1526 struct hwtstamp_config cfg;
1527 enum hwtstamp_tx_types tx_type;
1528 enum hwtstamp_rx_filters rx_filter;
1529 int tx_type_valid = 0;
1530 int rx_filter_valid = 0;
1532 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1535 if (cfg.flags) /* reserved for future extensions */
1538 tx_type = cfg.tx_type;
1539 rx_filter = cfg.rx_filter;
1542 case HWTSTAMP_TX_OFF:
1543 case HWTSTAMP_TX_ON:
1544 case HWTSTAMP_TX_ONESTEP_SYNC:
1549 switch (rx_filter) {
1550 case HWTSTAMP_FILTER_NONE:
1551 case HWTSTAMP_FILTER_ALL:
1552 case HWTSTAMP_FILTER_SOME:
1553 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1554 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1555 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1556 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1557 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1558 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1559 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1560 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1561 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1562 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1563 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1564 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1565 rx_filter_valid = 1;
1569 if (!tx_type_valid || !rx_filter_valid)
1575 static inline bool is_skb_forwardable(struct net_device *dev,
1576 struct sk_buff *skb)
1580 if (!(dev->flags & IFF_UP))
1583 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1584 if (skb->len <= len)
1587 /* if TSO is enabled, we don't care about the length as the packet
1588 * could be forwarded without being segmented before
1590 if (skb_is_gso(skb))
1597 * dev_forward_skb - loopback an skb to another netif
1599 * @dev: destination network device
1600 * @skb: buffer to forward
1603 * NET_RX_SUCCESS (no congestion)
1604 * NET_RX_DROP (packet was dropped, but freed)
1606 * dev_forward_skb can be used for injecting an skb from the
1607 * start_xmit function of one device into the receive queue
1608 * of another device.
1610 * The receiving device may be in another namespace, so
1611 * we have to clear all information in the skb that could
1612 * impact namespace isolation.
1614 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1616 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1617 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1618 atomic_long_inc(&dev->rx_dropped);
1627 if (unlikely(!is_skb_forwardable(dev, skb))) {
1628 atomic_long_inc(&dev->rx_dropped);
1635 skb->tstamp.tv64 = 0;
1636 skb->pkt_type = PACKET_HOST;
1637 skb->protocol = eth_type_trans(skb, dev);
1641 return netif_rx(skb);
1643 EXPORT_SYMBOL_GPL(dev_forward_skb);
1645 static inline int deliver_skb(struct sk_buff *skb,
1646 struct packet_type *pt_prev,
1647 struct net_device *orig_dev)
1649 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1651 atomic_inc(&skb->users);
1652 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1655 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1657 if (ptype->af_packet_priv == NULL)
1660 if (ptype->id_match)
1661 return ptype->id_match(ptype, skb->sk);
1662 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1669 * Support routine. Sends outgoing frames to any network
1670 * taps currently in use.
1673 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1675 struct packet_type *ptype;
1676 struct sk_buff *skb2 = NULL;
1677 struct packet_type *pt_prev = NULL;
1680 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1681 /* Never send packets back to the socket
1682 * they originated from - MvS (miquels@drinkel.ow.org)
1684 if ((ptype->dev == dev || !ptype->dev) &&
1685 (!skb_loop_sk(ptype, skb))) {
1687 deliver_skb(skb2, pt_prev, skb->dev);
1692 skb2 = skb_clone(skb, GFP_ATOMIC);
1696 net_timestamp_set(skb2);
1698 /* skb->nh should be correctly
1699 set by sender, so that the second statement is
1700 just protection against buggy protocols.
1702 skb_reset_mac_header(skb2);
1704 if (skb_network_header(skb2) < skb2->data ||
1705 skb2->network_header > skb2->tail) {
1706 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1707 ntohs(skb2->protocol),
1709 skb_reset_network_header(skb2);
1712 skb2->transport_header = skb2->network_header;
1713 skb2->pkt_type = PACKET_OUTGOING;
1718 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1723 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1724 * @dev: Network device
1725 * @txq: number of queues available
1727 * If real_num_tx_queues is changed the tc mappings may no longer be
1728 * valid. To resolve this verify the tc mapping remains valid and if
1729 * not NULL the mapping. With no priorities mapping to this
1730 * offset/count pair it will no longer be used. In the worst case TC0
1731 * is invalid nothing can be done so disable priority mappings. If is
1732 * expected that drivers will fix this mapping if they can before
1733 * calling netif_set_real_num_tx_queues.
1735 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1738 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1740 /* If TC0 is invalidated disable TC mapping */
1741 if (tc->offset + tc->count > txq) {
1742 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1747 /* Invalidated prio to tc mappings set to TC0 */
1748 for (i = 1; i < TC_BITMASK + 1; i++) {
1749 int q = netdev_get_prio_tc_map(dev, i);
1751 tc = &dev->tc_to_txq[q];
1752 if (tc->offset + tc->count > txq) {
1753 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1755 netdev_set_prio_tc_map(dev, i, 0);
1761 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1762 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1764 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1768 if (txq < 1 || txq > dev->num_tx_queues)
1771 if (dev->reg_state == NETREG_REGISTERED ||
1772 dev->reg_state == NETREG_UNREGISTERING) {
1775 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1781 netif_setup_tc(dev, txq);
1783 if (txq < dev->real_num_tx_queues)
1784 qdisc_reset_all_tx_gt(dev, txq);
1787 dev->real_num_tx_queues = txq;
1790 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1794 * netif_set_real_num_rx_queues - set actual number of RX queues used
1795 * @dev: Network device
1796 * @rxq: Actual number of RX queues
1798 * This must be called either with the rtnl_lock held or before
1799 * registration of the net device. Returns 0 on success, or a
1800 * negative error code. If called before registration, it always
1803 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1807 if (rxq < 1 || rxq > dev->num_rx_queues)
1810 if (dev->reg_state == NETREG_REGISTERED) {
1813 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1819 dev->real_num_rx_queues = rxq;
1822 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1826 * netif_get_num_default_rss_queues - default number of RSS queues
1828 * This routine should set an upper limit on the number of RSS queues
1829 * used by default by multiqueue devices.
1831 int netif_get_num_default_rss_queues(void)
1833 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
1835 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
1837 static inline void __netif_reschedule(struct Qdisc *q)
1839 struct softnet_data *sd;
1840 unsigned long flags;
1842 local_irq_save(flags);
1843 sd = &__get_cpu_var(softnet_data);
1844 q->next_sched = NULL;
1845 *sd->output_queue_tailp = q;
1846 sd->output_queue_tailp = &q->next_sched;
1847 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1848 local_irq_restore(flags);
1851 void __netif_schedule(struct Qdisc *q)
1853 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1854 __netif_reschedule(q);
1856 EXPORT_SYMBOL(__netif_schedule);
1858 void dev_kfree_skb_irq(struct sk_buff *skb)
1860 if (atomic_dec_and_test(&skb->users)) {
1861 struct softnet_data *sd;
1862 unsigned long flags;
1864 local_irq_save(flags);
1865 sd = &__get_cpu_var(softnet_data);
1866 skb->next = sd->completion_queue;
1867 sd->completion_queue = skb;
1868 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1869 local_irq_restore(flags);
1872 EXPORT_SYMBOL(dev_kfree_skb_irq);
1874 void dev_kfree_skb_any(struct sk_buff *skb)
1876 if (in_irq() || irqs_disabled())
1877 dev_kfree_skb_irq(skb);
1881 EXPORT_SYMBOL(dev_kfree_skb_any);
1885 * netif_device_detach - mark device as removed
1886 * @dev: network device
1888 * Mark device as removed from system and therefore no longer available.
1890 void netif_device_detach(struct net_device *dev)
1892 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1893 netif_running(dev)) {
1894 netif_tx_stop_all_queues(dev);
1897 EXPORT_SYMBOL(netif_device_detach);
1900 * netif_device_attach - mark device as attached
1901 * @dev: network device
1903 * Mark device as attached from system and restart if needed.
1905 void netif_device_attach(struct net_device *dev)
1907 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1908 netif_running(dev)) {
1909 netif_tx_wake_all_queues(dev);
1910 __netdev_watchdog_up(dev);
1913 EXPORT_SYMBOL(netif_device_attach);
1915 static void skb_warn_bad_offload(const struct sk_buff *skb)
1917 static const netdev_features_t null_features = 0;
1918 struct net_device *dev = skb->dev;
1919 const char *driver = "";
1921 if (dev && dev->dev.parent)
1922 driver = dev_driver_string(dev->dev.parent);
1924 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1925 "gso_type=%d ip_summed=%d\n",
1926 driver, dev ? &dev->features : &null_features,
1927 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1928 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1929 skb_shinfo(skb)->gso_type, skb->ip_summed);
1933 * Invalidate hardware checksum when packet is to be mangled, and
1934 * complete checksum manually on outgoing path.
1936 int skb_checksum_help(struct sk_buff *skb)
1939 int ret = 0, offset;
1941 if (skb->ip_summed == CHECKSUM_COMPLETE)
1942 goto out_set_summed;
1944 if (unlikely(skb_shinfo(skb)->gso_size)) {
1945 skb_warn_bad_offload(skb);
1949 offset = skb_checksum_start_offset(skb);
1950 BUG_ON(offset >= skb_headlen(skb));
1951 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1953 offset += skb->csum_offset;
1954 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1956 if (skb_cloned(skb) &&
1957 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1958 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1963 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1965 skb->ip_summed = CHECKSUM_NONE;
1969 EXPORT_SYMBOL(skb_checksum_help);
1972 * skb_gso_segment - Perform segmentation on skb.
1973 * @skb: buffer to segment
1974 * @features: features for the output path (see dev->features)
1976 * This function segments the given skb and returns a list of segments.
1978 * It may return NULL if the skb requires no segmentation. This is
1979 * only possible when GSO is used for verifying header integrity.
1981 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1982 netdev_features_t features)
1984 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1985 struct packet_type *ptype;
1986 __be16 type = skb->protocol;
1987 int vlan_depth = ETH_HLEN;
1990 while (type == htons(ETH_P_8021Q)) {
1991 struct vlan_hdr *vh;
1993 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1994 return ERR_PTR(-EINVAL);
1996 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1997 type = vh->h_vlan_encapsulated_proto;
1998 vlan_depth += VLAN_HLEN;
2001 skb_reset_mac_header(skb);
2002 skb->mac_len = skb->network_header - skb->mac_header;
2003 __skb_pull(skb, skb->mac_len);
2005 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2006 skb_warn_bad_offload(skb);
2008 if (skb_header_cloned(skb) &&
2009 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2010 return ERR_PTR(err);
2014 list_for_each_entry_rcu(ptype,
2015 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2016 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
2017 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2018 err = ptype->gso_send_check(skb);
2019 segs = ERR_PTR(err);
2020 if (err || skb_gso_ok(skb, features))
2022 __skb_push(skb, (skb->data -
2023 skb_network_header(skb)));
2025 segs = ptype->gso_segment(skb, features);
2031 __skb_push(skb, skb->data - skb_mac_header(skb));
2035 EXPORT_SYMBOL(skb_gso_segment);
2037 /* Take action when hardware reception checksum errors are detected. */
2039 void netdev_rx_csum_fault(struct net_device *dev)
2041 if (net_ratelimit()) {
2042 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2046 EXPORT_SYMBOL(netdev_rx_csum_fault);
2049 /* Actually, we should eliminate this check as soon as we know, that:
2050 * 1. IOMMU is present and allows to map all the memory.
2051 * 2. No high memory really exists on this machine.
2054 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2056 #ifdef CONFIG_HIGHMEM
2058 if (!(dev->features & NETIF_F_HIGHDMA)) {
2059 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2060 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2061 if (PageHighMem(skb_frag_page(frag)))
2066 if (PCI_DMA_BUS_IS_PHYS) {
2067 struct device *pdev = dev->dev.parent;
2071 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2072 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2073 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2074 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2083 void (*destructor)(struct sk_buff *skb);
2086 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2088 static void dev_gso_skb_destructor(struct sk_buff *skb)
2090 struct dev_gso_cb *cb;
2093 struct sk_buff *nskb = skb->next;
2095 skb->next = nskb->next;
2098 } while (skb->next);
2100 cb = DEV_GSO_CB(skb);
2102 cb->destructor(skb);
2106 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2107 * @skb: buffer to segment
2108 * @features: device features as applicable to this skb
2110 * This function segments the given skb and stores the list of segments
2113 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2115 struct sk_buff *segs;
2117 segs = skb_gso_segment(skb, features);
2119 /* Verifying header integrity only. */
2124 return PTR_ERR(segs);
2127 DEV_GSO_CB(skb)->destructor = skb->destructor;
2128 skb->destructor = dev_gso_skb_destructor;
2133 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2135 return ((features & NETIF_F_GEN_CSUM) ||
2136 ((features & NETIF_F_V4_CSUM) &&
2137 protocol == htons(ETH_P_IP)) ||
2138 ((features & NETIF_F_V6_CSUM) &&
2139 protocol == htons(ETH_P_IPV6)) ||
2140 ((features & NETIF_F_FCOE_CRC) &&
2141 protocol == htons(ETH_P_FCOE)));
2144 static netdev_features_t harmonize_features(struct sk_buff *skb,
2145 __be16 protocol, netdev_features_t features)
2147 if (!can_checksum_protocol(features, protocol)) {
2148 features &= ~NETIF_F_ALL_CSUM;
2149 features &= ~NETIF_F_SG;
2150 } else if (illegal_highdma(skb->dev, skb)) {
2151 features &= ~NETIF_F_SG;
2157 netdev_features_t netif_skb_features(struct sk_buff *skb)
2159 __be16 protocol = skb->protocol;
2160 netdev_features_t features = skb->dev->features;
2162 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2163 features &= ~NETIF_F_GSO_MASK;
2165 if (protocol == htons(ETH_P_8021Q)) {
2166 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2167 protocol = veh->h_vlan_encapsulated_proto;
2168 } else if (!vlan_tx_tag_present(skb)) {
2169 return harmonize_features(skb, protocol, features);
2172 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2174 if (protocol != htons(ETH_P_8021Q)) {
2175 return harmonize_features(skb, protocol, features);
2177 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2178 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2179 return harmonize_features(skb, protocol, features);
2182 EXPORT_SYMBOL(netif_skb_features);
2185 * Returns true if either:
2186 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2187 * 2. skb is fragmented and the device does not support SG, or if
2188 * at least one of fragments is in highmem and device does not
2189 * support DMA from it.
2191 static inline int skb_needs_linearize(struct sk_buff *skb,
2194 return skb_is_nonlinear(skb) &&
2195 ((skb_has_frag_list(skb) &&
2196 !(features & NETIF_F_FRAGLIST)) ||
2197 (skb_shinfo(skb)->nr_frags &&
2198 !(features & NETIF_F_SG)));
2201 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2202 struct netdev_queue *txq)
2204 const struct net_device_ops *ops = dev->netdev_ops;
2205 int rc = NETDEV_TX_OK;
2206 unsigned int skb_len;
2208 if (likely(!skb->next)) {
2209 netdev_features_t features;
2212 * If device doesn't need skb->dst, release it right now while
2213 * its hot in this cpu cache
2215 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2218 if (!list_empty(&ptype_all))
2219 dev_queue_xmit_nit(skb, dev);
2221 features = netif_skb_features(skb);
2223 if (vlan_tx_tag_present(skb) &&
2224 !(features & NETIF_F_HW_VLAN_TX)) {
2225 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2232 if (netif_needs_gso(skb, features)) {
2233 if (unlikely(dev_gso_segment(skb, features)))
2238 if (skb_needs_linearize(skb, features) &&
2239 __skb_linearize(skb))
2242 /* If packet is not checksummed and device does not
2243 * support checksumming for this protocol, complete
2244 * checksumming here.
2246 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2247 skb_set_transport_header(skb,
2248 skb_checksum_start_offset(skb));
2249 if (!(features & NETIF_F_ALL_CSUM) &&
2250 skb_checksum_help(skb))
2256 rc = ops->ndo_start_xmit(skb, dev);
2257 trace_net_dev_xmit(skb, rc, dev, skb_len);
2258 if (rc == NETDEV_TX_OK)
2259 txq_trans_update(txq);
2265 struct sk_buff *nskb = skb->next;
2267 skb->next = nskb->next;
2271 * If device doesn't need nskb->dst, release it right now while
2272 * its hot in this cpu cache
2274 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2277 skb_len = nskb->len;
2278 rc = ops->ndo_start_xmit(nskb, dev);
2279 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2280 if (unlikely(rc != NETDEV_TX_OK)) {
2281 if (rc & ~NETDEV_TX_MASK)
2282 goto out_kfree_gso_skb;
2283 nskb->next = skb->next;
2287 txq_trans_update(txq);
2288 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2289 return NETDEV_TX_BUSY;
2290 } while (skb->next);
2293 if (likely(skb->next == NULL))
2294 skb->destructor = DEV_GSO_CB(skb)->destructor;
2301 static u32 hashrnd __read_mostly;
2304 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2305 * to be used as a distribution range.
2307 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2308 unsigned int num_tx_queues)
2312 u16 qcount = num_tx_queues;
2314 if (skb_rx_queue_recorded(skb)) {
2315 hash = skb_get_rx_queue(skb);
2316 while (unlikely(hash >= num_tx_queues))
2317 hash -= num_tx_queues;
2322 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2323 qoffset = dev->tc_to_txq[tc].offset;
2324 qcount = dev->tc_to_txq[tc].count;
2327 if (skb->sk && skb->sk->sk_hash)
2328 hash = skb->sk->sk_hash;
2330 hash = (__force u16) skb->protocol;
2331 hash = jhash_1word(hash, hashrnd);
2333 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2335 EXPORT_SYMBOL(__skb_tx_hash);
2337 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2339 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2340 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2341 dev->name, queue_index,
2342 dev->real_num_tx_queues);
2348 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2351 struct xps_dev_maps *dev_maps;
2352 struct xps_map *map;
2353 int queue_index = -1;
2356 dev_maps = rcu_dereference(dev->xps_maps);
2358 map = rcu_dereference(
2359 dev_maps->cpu_map[raw_smp_processor_id()]);
2362 queue_index = map->queues[0];
2365 if (skb->sk && skb->sk->sk_hash)
2366 hash = skb->sk->sk_hash;
2368 hash = (__force u16) skb->protocol ^
2370 hash = jhash_1word(hash, hashrnd);
2371 queue_index = map->queues[
2372 ((u64)hash * map->len) >> 32];
2374 if (unlikely(queue_index >= dev->real_num_tx_queues))
2386 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2387 struct sk_buff *skb)
2390 const struct net_device_ops *ops = dev->netdev_ops;
2392 if (dev->real_num_tx_queues == 1)
2394 else if (ops->ndo_select_queue) {
2395 queue_index = ops->ndo_select_queue(dev, skb);
2396 queue_index = dev_cap_txqueue(dev, queue_index);
2398 struct sock *sk = skb->sk;
2399 queue_index = sk_tx_queue_get(sk);
2401 if (queue_index < 0 || skb->ooo_okay ||
2402 queue_index >= dev->real_num_tx_queues) {
2403 int old_index = queue_index;
2405 queue_index = get_xps_queue(dev, skb);
2406 if (queue_index < 0)
2407 queue_index = skb_tx_hash(dev, skb);
2409 if (queue_index != old_index && sk) {
2410 struct dst_entry *dst =
2411 rcu_dereference_check(sk->sk_dst_cache, 1);
2413 if (dst && skb_dst(skb) == dst)
2414 sk_tx_queue_set(sk, queue_index);
2419 skb_set_queue_mapping(skb, queue_index);
2420 return netdev_get_tx_queue(dev, queue_index);
2423 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2424 struct net_device *dev,
2425 struct netdev_queue *txq)
2427 spinlock_t *root_lock = qdisc_lock(q);
2431 qdisc_skb_cb(skb)->pkt_len = skb->len;
2432 qdisc_calculate_pkt_len(skb, q);
2434 * Heuristic to force contended enqueues to serialize on a
2435 * separate lock before trying to get qdisc main lock.
2436 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2437 * and dequeue packets faster.
2439 contended = qdisc_is_running(q);
2440 if (unlikely(contended))
2441 spin_lock(&q->busylock);
2443 spin_lock(root_lock);
2444 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2447 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2448 qdisc_run_begin(q)) {
2450 * This is a work-conserving queue; there are no old skbs
2451 * waiting to be sent out; and the qdisc is not running -
2452 * xmit the skb directly.
2454 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2457 qdisc_bstats_update(q, skb);
2459 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2460 if (unlikely(contended)) {
2461 spin_unlock(&q->busylock);
2468 rc = NET_XMIT_SUCCESS;
2471 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2472 if (qdisc_run_begin(q)) {
2473 if (unlikely(contended)) {
2474 spin_unlock(&q->busylock);
2480 spin_unlock(root_lock);
2481 if (unlikely(contended))
2482 spin_unlock(&q->busylock);
2486 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2487 static void skb_update_prio(struct sk_buff *skb)
2489 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2491 if (!skb->priority && skb->sk && map) {
2492 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2494 if (prioidx < map->priomap_len)
2495 skb->priority = map->priomap[prioidx];
2499 #define skb_update_prio(skb)
2502 static DEFINE_PER_CPU(int, xmit_recursion);
2503 #define RECURSION_LIMIT 10
2506 * dev_loopback_xmit - loop back @skb
2507 * @skb: buffer to transmit
2509 int dev_loopback_xmit(struct sk_buff *skb)
2511 skb_reset_mac_header(skb);
2512 __skb_pull(skb, skb_network_offset(skb));
2513 skb->pkt_type = PACKET_LOOPBACK;
2514 skb->ip_summed = CHECKSUM_UNNECESSARY;
2515 WARN_ON(!skb_dst(skb));
2520 EXPORT_SYMBOL(dev_loopback_xmit);
2523 * dev_queue_xmit - transmit a buffer
2524 * @skb: buffer to transmit
2526 * Queue a buffer for transmission to a network device. The caller must
2527 * have set the device and priority and built the buffer before calling
2528 * this function. The function can be called from an interrupt.
2530 * A negative errno code is returned on a failure. A success does not
2531 * guarantee the frame will be transmitted as it may be dropped due
2532 * to congestion or traffic shaping.
2534 * -----------------------------------------------------------------------------------
2535 * I notice this method can also return errors from the queue disciplines,
2536 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2539 * Regardless of the return value, the skb is consumed, so it is currently
2540 * difficult to retry a send to this method. (You can bump the ref count
2541 * before sending to hold a reference for retry if you are careful.)
2543 * When calling this method, interrupts MUST be enabled. This is because
2544 * the BH enable code must have IRQs enabled so that it will not deadlock.
2547 int dev_queue_xmit(struct sk_buff *skb)
2549 struct net_device *dev = skb->dev;
2550 struct netdev_queue *txq;
2554 /* Disable soft irqs for various locks below. Also
2555 * stops preemption for RCU.
2559 skb_update_prio(skb);
2561 txq = dev_pick_tx(dev, skb);
2562 q = rcu_dereference_bh(txq->qdisc);
2564 #ifdef CONFIG_NET_CLS_ACT
2565 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2567 trace_net_dev_queue(skb);
2569 rc = __dev_xmit_skb(skb, q, dev, txq);
2573 /* The device has no queue. Common case for software devices:
2574 loopback, all the sorts of tunnels...
2576 Really, it is unlikely that netif_tx_lock protection is necessary
2577 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2579 However, it is possible, that they rely on protection
2582 Check this and shot the lock. It is not prone from deadlocks.
2583 Either shot noqueue qdisc, it is even simpler 8)
2585 if (dev->flags & IFF_UP) {
2586 int cpu = smp_processor_id(); /* ok because BHs are off */
2588 if (txq->xmit_lock_owner != cpu) {
2590 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2591 goto recursion_alert;
2593 HARD_TX_LOCK(dev, txq, cpu);
2595 if (!netif_xmit_stopped(txq)) {
2596 __this_cpu_inc(xmit_recursion);
2597 rc = dev_hard_start_xmit(skb, dev, txq);
2598 __this_cpu_dec(xmit_recursion);
2599 if (dev_xmit_complete(rc)) {
2600 HARD_TX_UNLOCK(dev, txq);
2604 HARD_TX_UNLOCK(dev, txq);
2605 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2608 /* Recursion is detected! It is possible,
2612 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2618 rcu_read_unlock_bh();
2623 rcu_read_unlock_bh();
2626 EXPORT_SYMBOL(dev_queue_xmit);
2629 /*=======================================================================
2631 =======================================================================*/
2633 int netdev_max_backlog __read_mostly = 1000;
2634 int netdev_tstamp_prequeue __read_mostly = 1;
2635 int netdev_budget __read_mostly = 300;
2636 int weight_p __read_mostly = 64; /* old backlog weight */
2638 /* Called with irq disabled */
2639 static inline void ____napi_schedule(struct softnet_data *sd,
2640 struct napi_struct *napi)
2642 list_add_tail(&napi->poll_list, &sd->poll_list);
2643 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2647 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2648 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2649 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2650 * if hash is a canonical 4-tuple hash over transport ports.
2652 void __skb_get_rxhash(struct sk_buff *skb)
2654 struct flow_keys keys;
2657 if (!skb_flow_dissect(skb, &keys))
2661 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2662 swap(keys.port16[0], keys.port16[1]);
2666 /* get a consistent hash (same value on both flow directions) */
2667 if ((__force u32)keys.dst < (__force u32)keys.src)
2668 swap(keys.dst, keys.src);
2670 hash = jhash_3words((__force u32)keys.dst,
2671 (__force u32)keys.src,
2672 (__force u32)keys.ports, hashrnd);
2678 EXPORT_SYMBOL(__skb_get_rxhash);
2682 /* One global table that all flow-based protocols share. */
2683 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2684 EXPORT_SYMBOL(rps_sock_flow_table);
2686 struct static_key rps_needed __read_mostly;
2688 static struct rps_dev_flow *
2689 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2690 struct rps_dev_flow *rflow, u16 next_cpu)
2692 if (next_cpu != RPS_NO_CPU) {
2693 #ifdef CONFIG_RFS_ACCEL
2694 struct netdev_rx_queue *rxqueue;
2695 struct rps_dev_flow_table *flow_table;
2696 struct rps_dev_flow *old_rflow;
2701 /* Should we steer this flow to a different hardware queue? */
2702 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2703 !(dev->features & NETIF_F_NTUPLE))
2705 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2706 if (rxq_index == skb_get_rx_queue(skb))
2709 rxqueue = dev->_rx + rxq_index;
2710 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2713 flow_id = skb->rxhash & flow_table->mask;
2714 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2715 rxq_index, flow_id);
2719 rflow = &flow_table->flows[flow_id];
2721 if (old_rflow->filter == rflow->filter)
2722 old_rflow->filter = RPS_NO_FILTER;
2726 per_cpu(softnet_data, next_cpu).input_queue_head;
2729 rflow->cpu = next_cpu;
2734 * get_rps_cpu is called from netif_receive_skb and returns the target
2735 * CPU from the RPS map of the receiving queue for a given skb.
2736 * rcu_read_lock must be held on entry.
2738 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2739 struct rps_dev_flow **rflowp)
2741 struct netdev_rx_queue *rxqueue;
2742 struct rps_map *map;
2743 struct rps_dev_flow_table *flow_table;
2744 struct rps_sock_flow_table *sock_flow_table;
2748 if (skb_rx_queue_recorded(skb)) {
2749 u16 index = skb_get_rx_queue(skb);
2750 if (unlikely(index >= dev->real_num_rx_queues)) {
2751 WARN_ONCE(dev->real_num_rx_queues > 1,
2752 "%s received packet on queue %u, but number "
2753 "of RX queues is %u\n",
2754 dev->name, index, dev->real_num_rx_queues);
2757 rxqueue = dev->_rx + index;
2761 map = rcu_dereference(rxqueue->rps_map);
2763 if (map->len == 1 &&
2764 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2765 tcpu = map->cpus[0];
2766 if (cpu_online(tcpu))
2770 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2774 skb_reset_network_header(skb);
2775 if (!skb_get_rxhash(skb))
2778 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2779 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2780 if (flow_table && sock_flow_table) {
2782 struct rps_dev_flow *rflow;
2784 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2787 next_cpu = sock_flow_table->ents[skb->rxhash &
2788 sock_flow_table->mask];
2791 * If the desired CPU (where last recvmsg was done) is
2792 * different from current CPU (one in the rx-queue flow
2793 * table entry), switch if one of the following holds:
2794 * - Current CPU is unset (equal to RPS_NO_CPU).
2795 * - Current CPU is offline.
2796 * - The current CPU's queue tail has advanced beyond the
2797 * last packet that was enqueued using this table entry.
2798 * This guarantees that all previous packets for the flow
2799 * have been dequeued, thus preserving in order delivery.
2801 if (unlikely(tcpu != next_cpu) &&
2802 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2803 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2804 rflow->last_qtail)) >= 0))
2805 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2807 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2815 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2817 if (cpu_online(tcpu)) {
2827 #ifdef CONFIG_RFS_ACCEL
2830 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2831 * @dev: Device on which the filter was set
2832 * @rxq_index: RX queue index
2833 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2834 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2836 * Drivers that implement ndo_rx_flow_steer() should periodically call
2837 * this function for each installed filter and remove the filters for
2838 * which it returns %true.
2840 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2841 u32 flow_id, u16 filter_id)
2843 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2844 struct rps_dev_flow_table *flow_table;
2845 struct rps_dev_flow *rflow;
2850 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2851 if (flow_table && flow_id <= flow_table->mask) {
2852 rflow = &flow_table->flows[flow_id];
2853 cpu = ACCESS_ONCE(rflow->cpu);
2854 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2855 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2856 rflow->last_qtail) <
2857 (int)(10 * flow_table->mask)))
2863 EXPORT_SYMBOL(rps_may_expire_flow);
2865 #endif /* CONFIG_RFS_ACCEL */
2867 /* Called from hardirq (IPI) context */
2868 static void rps_trigger_softirq(void *data)
2870 struct softnet_data *sd = data;
2872 ____napi_schedule(sd, &sd->backlog);
2876 #endif /* CONFIG_RPS */
2879 * Check if this softnet_data structure is another cpu one
2880 * If yes, queue it to our IPI list and return 1
2883 static int rps_ipi_queued(struct softnet_data *sd)
2886 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2889 sd->rps_ipi_next = mysd->rps_ipi_list;
2890 mysd->rps_ipi_list = sd;
2892 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2895 #endif /* CONFIG_RPS */
2900 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2901 * queue (may be a remote CPU queue).
2903 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2904 unsigned int *qtail)
2906 struct softnet_data *sd;
2907 unsigned long flags;
2909 sd = &per_cpu(softnet_data, cpu);
2911 local_irq_save(flags);
2914 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2915 if (skb_queue_len(&sd->input_pkt_queue)) {
2917 __skb_queue_tail(&sd->input_pkt_queue, skb);
2918 input_queue_tail_incr_save(sd, qtail);
2920 local_irq_restore(flags);
2921 return NET_RX_SUCCESS;
2924 /* Schedule NAPI for backlog device
2925 * We can use non atomic operation since we own the queue lock
2927 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2928 if (!rps_ipi_queued(sd))
2929 ____napi_schedule(sd, &sd->backlog);
2937 local_irq_restore(flags);
2939 atomic_long_inc(&skb->dev->rx_dropped);
2945 * netif_rx - post buffer to the network code
2946 * @skb: buffer to post
2948 * This function receives a packet from a device driver and queues it for
2949 * the upper (protocol) levels to process. It always succeeds. The buffer
2950 * may be dropped during processing for congestion control or by the
2954 * NET_RX_SUCCESS (no congestion)
2955 * NET_RX_DROP (packet was dropped)
2959 int netif_rx(struct sk_buff *skb)
2963 /* if netpoll wants it, pretend we never saw it */
2964 if (netpoll_rx(skb))
2967 net_timestamp_check(netdev_tstamp_prequeue, skb);
2969 trace_netif_rx(skb);
2971 if (static_key_false(&rps_needed)) {
2972 struct rps_dev_flow voidflow, *rflow = &voidflow;
2978 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2980 cpu = smp_processor_id();
2982 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2990 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2995 EXPORT_SYMBOL(netif_rx);
2997 int netif_rx_ni(struct sk_buff *skb)
3002 err = netif_rx(skb);
3003 if (local_softirq_pending())
3009 EXPORT_SYMBOL(netif_rx_ni);
3011 static void net_tx_action(struct softirq_action *h)
3013 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3015 if (sd->completion_queue) {
3016 struct sk_buff *clist;
3018 local_irq_disable();
3019 clist = sd->completion_queue;
3020 sd->completion_queue = NULL;
3024 struct sk_buff *skb = clist;
3025 clist = clist->next;
3027 WARN_ON(atomic_read(&skb->users));
3028 trace_kfree_skb(skb, net_tx_action);
3033 if (sd->output_queue) {
3036 local_irq_disable();
3037 head = sd->output_queue;
3038 sd->output_queue = NULL;
3039 sd->output_queue_tailp = &sd->output_queue;
3043 struct Qdisc *q = head;
3044 spinlock_t *root_lock;
3046 head = head->next_sched;
3048 root_lock = qdisc_lock(q);
3049 if (spin_trylock(root_lock)) {
3050 smp_mb__before_clear_bit();
3051 clear_bit(__QDISC_STATE_SCHED,
3054 spin_unlock(root_lock);
3056 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3058 __netif_reschedule(q);
3060 smp_mb__before_clear_bit();
3061 clear_bit(__QDISC_STATE_SCHED,
3069 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3070 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3071 /* This hook is defined here for ATM LANE */
3072 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3073 unsigned char *addr) __read_mostly;
3074 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3077 #ifdef CONFIG_NET_CLS_ACT
3078 /* TODO: Maybe we should just force sch_ingress to be compiled in
3079 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3080 * a compare and 2 stores extra right now if we dont have it on
3081 * but have CONFIG_NET_CLS_ACT
3082 * NOTE: This doesn't stop any functionality; if you dont have
3083 * the ingress scheduler, you just can't add policies on ingress.
3086 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3088 struct net_device *dev = skb->dev;
3089 u32 ttl = G_TC_RTTL(skb->tc_verd);
3090 int result = TC_ACT_OK;
3093 if (unlikely(MAX_RED_LOOP < ttl++)) {
3094 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3095 skb->skb_iif, dev->ifindex);
3099 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3100 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3103 if (q != &noop_qdisc) {
3104 spin_lock(qdisc_lock(q));
3105 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3106 result = qdisc_enqueue_root(skb, q);
3107 spin_unlock(qdisc_lock(q));
3113 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3114 struct packet_type **pt_prev,
3115 int *ret, struct net_device *orig_dev)
3117 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3119 if (!rxq || rxq->qdisc == &noop_qdisc)
3123 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3127 switch (ing_filter(skb, rxq)) {
3141 * netdev_rx_handler_register - register receive handler
3142 * @dev: device to register a handler for
3143 * @rx_handler: receive handler to register
3144 * @rx_handler_data: data pointer that is used by rx handler
3146 * Register a receive hander for a device. This handler will then be
3147 * called from __netif_receive_skb. A negative errno code is returned
3150 * The caller must hold the rtnl_mutex.
3152 * For a general description of rx_handler, see enum rx_handler_result.
3154 int netdev_rx_handler_register(struct net_device *dev,
3155 rx_handler_func_t *rx_handler,
3156 void *rx_handler_data)
3160 if (dev->rx_handler)
3163 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3164 rcu_assign_pointer(dev->rx_handler, rx_handler);
3168 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3171 * netdev_rx_handler_unregister - unregister receive handler
3172 * @dev: device to unregister a handler from
3174 * Unregister a receive hander from a device.
3176 * The caller must hold the rtnl_mutex.
3178 void netdev_rx_handler_unregister(struct net_device *dev)
3182 RCU_INIT_POINTER(dev->rx_handler, NULL);
3183 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3185 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3188 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3189 * the special handling of PFMEMALLOC skbs.
3191 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3193 switch (skb->protocol) {
3194 case __constant_htons(ETH_P_ARP):
3195 case __constant_htons(ETH_P_IP):
3196 case __constant_htons(ETH_P_IPV6):
3197 case __constant_htons(ETH_P_8021Q):
3204 static int __netif_receive_skb(struct sk_buff *skb)
3206 struct packet_type *ptype, *pt_prev;
3207 rx_handler_func_t *rx_handler;
3208 struct net_device *orig_dev;
3209 struct net_device *null_or_dev;
3210 bool deliver_exact = false;
3211 int ret = NET_RX_DROP;
3213 unsigned long pflags = current->flags;
3215 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3217 trace_netif_receive_skb(skb);
3220 * PFMEMALLOC skbs are special, they should
3221 * - be delivered to SOCK_MEMALLOC sockets only
3222 * - stay away from userspace
3223 * - have bounded memory usage
3225 * Use PF_MEMALLOC as this saves us from propagating the allocation
3226 * context down to all allocation sites.
3228 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3229 current->flags |= PF_MEMALLOC;
3231 /* if we've gotten here through NAPI, check netpoll */
3232 if (netpoll_receive_skb(skb))
3235 orig_dev = skb->dev;
3237 skb_reset_network_header(skb);
3238 skb_reset_transport_header(skb);
3239 skb_reset_mac_len(skb);
3246 skb->skb_iif = skb->dev->ifindex;
3248 __this_cpu_inc(softnet_data.processed);
3250 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3251 skb = vlan_untag(skb);
3256 #ifdef CONFIG_NET_CLS_ACT
3257 if (skb->tc_verd & TC_NCLS) {
3258 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3263 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3266 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3267 if (!ptype->dev || ptype->dev == skb->dev) {
3269 ret = deliver_skb(skb, pt_prev, orig_dev);
3275 #ifdef CONFIG_NET_CLS_ACT
3276 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3282 if (sk_memalloc_socks() && skb_pfmemalloc(skb)
3283 && !skb_pfmemalloc_protocol(skb))
3286 rx_handler = rcu_dereference(skb->dev->rx_handler);
3287 if (vlan_tx_tag_present(skb)) {
3289 ret = deliver_skb(skb, pt_prev, orig_dev);
3292 if (vlan_do_receive(&skb, !rx_handler))
3294 else if (unlikely(!skb))
3300 ret = deliver_skb(skb, pt_prev, orig_dev);
3303 switch (rx_handler(&skb)) {
3304 case RX_HANDLER_CONSUMED:
3306 case RX_HANDLER_ANOTHER:
3308 case RX_HANDLER_EXACT:
3309 deliver_exact = true;
3310 case RX_HANDLER_PASS:
3317 /* deliver only exact match when indicated */
3318 null_or_dev = deliver_exact ? skb->dev : NULL;
3320 type = skb->protocol;
3321 list_for_each_entry_rcu(ptype,
3322 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3323 if (ptype->type == type &&
3324 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3325 ptype->dev == orig_dev)) {
3327 ret = deliver_skb(skb, pt_prev, orig_dev);
3333 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3336 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3339 atomic_long_inc(&skb->dev->rx_dropped);
3341 /* Jamal, now you will not able to escape explaining
3342 * me how you were going to use this. :-)
3350 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3355 * netif_receive_skb - process receive buffer from network
3356 * @skb: buffer to process
3358 * netif_receive_skb() is the main receive data processing function.
3359 * It always succeeds. The buffer may be dropped during processing
3360 * for congestion control or by the protocol layers.
3362 * This function may only be called from softirq context and interrupts
3363 * should be enabled.
3365 * Return values (usually ignored):
3366 * NET_RX_SUCCESS: no congestion
3367 * NET_RX_DROP: packet was dropped
3369 int netif_receive_skb(struct sk_buff *skb)
3371 net_timestamp_check(netdev_tstamp_prequeue, skb);
3373 if (skb_defer_rx_timestamp(skb))
3374 return NET_RX_SUCCESS;
3377 if (static_key_false(&rps_needed)) {
3378 struct rps_dev_flow voidflow, *rflow = &voidflow;
3383 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3386 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3393 return __netif_receive_skb(skb);
3395 EXPORT_SYMBOL(netif_receive_skb);
3397 /* Network device is going away, flush any packets still pending
3398 * Called with irqs disabled.
3400 static void flush_backlog(void *arg)
3402 struct net_device *dev = arg;
3403 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3404 struct sk_buff *skb, *tmp;
3407 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3408 if (skb->dev == dev) {
3409 __skb_unlink(skb, &sd->input_pkt_queue);
3411 input_queue_head_incr(sd);
3416 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3417 if (skb->dev == dev) {
3418 __skb_unlink(skb, &sd->process_queue);
3420 input_queue_head_incr(sd);
3425 static int napi_gro_complete(struct sk_buff *skb)
3427 struct packet_type *ptype;
3428 __be16 type = skb->protocol;
3429 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3432 if (NAPI_GRO_CB(skb)->count == 1) {
3433 skb_shinfo(skb)->gso_size = 0;
3438 list_for_each_entry_rcu(ptype, head, list) {
3439 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3442 err = ptype->gro_complete(skb);
3448 WARN_ON(&ptype->list == head);
3450 return NET_RX_SUCCESS;
3454 return netif_receive_skb(skb);
3457 inline void napi_gro_flush(struct napi_struct *napi)
3459 struct sk_buff *skb, *next;
3461 for (skb = napi->gro_list; skb; skb = next) {
3464 napi_gro_complete(skb);
3467 napi->gro_count = 0;
3468 napi->gro_list = NULL;
3470 EXPORT_SYMBOL(napi_gro_flush);
3472 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3474 struct sk_buff **pp = NULL;
3475 struct packet_type *ptype;
3476 __be16 type = skb->protocol;
3477 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3480 enum gro_result ret;
3482 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3485 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3489 list_for_each_entry_rcu(ptype, head, list) {
3490 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3493 skb_set_network_header(skb, skb_gro_offset(skb));
3494 mac_len = skb->network_header - skb->mac_header;
3495 skb->mac_len = mac_len;
3496 NAPI_GRO_CB(skb)->same_flow = 0;
3497 NAPI_GRO_CB(skb)->flush = 0;
3498 NAPI_GRO_CB(skb)->free = 0;
3500 pp = ptype->gro_receive(&napi->gro_list, skb);
3505 if (&ptype->list == head)
3508 same_flow = NAPI_GRO_CB(skb)->same_flow;
3509 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3512 struct sk_buff *nskb = *pp;
3516 napi_gro_complete(nskb);
3523 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3527 NAPI_GRO_CB(skb)->count = 1;
3528 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3529 skb->next = napi->gro_list;
3530 napi->gro_list = skb;
3534 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3535 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3537 BUG_ON(skb->end - skb->tail < grow);
3539 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3542 skb->data_len -= grow;
3544 skb_shinfo(skb)->frags[0].page_offset += grow;
3545 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3547 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3548 skb_frag_unref(skb, 0);
3549 memmove(skb_shinfo(skb)->frags,
3550 skb_shinfo(skb)->frags + 1,
3551 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3562 EXPORT_SYMBOL(dev_gro_receive);
3564 static inline gro_result_t
3565 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3568 unsigned int maclen = skb->dev->hard_header_len;
3570 for (p = napi->gro_list; p; p = p->next) {
3571 unsigned long diffs;
3573 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3574 diffs |= p->vlan_tci ^ skb->vlan_tci;
3575 if (maclen == ETH_HLEN)
3576 diffs |= compare_ether_header(skb_mac_header(p),
3577 skb_gro_mac_header(skb));
3579 diffs = memcmp(skb_mac_header(p),
3580 skb_gro_mac_header(skb),
3582 NAPI_GRO_CB(p)->same_flow = !diffs;
3583 NAPI_GRO_CB(p)->flush = 0;
3586 return dev_gro_receive(napi, skb);
3589 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3593 if (netif_receive_skb(skb))
3601 case GRO_MERGED_FREE:
3602 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3603 kmem_cache_free(skbuff_head_cache, skb);
3615 EXPORT_SYMBOL(napi_skb_finish);
3617 void skb_gro_reset_offset(struct sk_buff *skb)
3619 NAPI_GRO_CB(skb)->data_offset = 0;
3620 NAPI_GRO_CB(skb)->frag0 = NULL;
3621 NAPI_GRO_CB(skb)->frag0_len = 0;
3623 if (skb->mac_header == skb->tail &&
3624 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3625 NAPI_GRO_CB(skb)->frag0 =
3626 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3627 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3630 EXPORT_SYMBOL(skb_gro_reset_offset);
3632 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3634 skb_gro_reset_offset(skb);
3636 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3638 EXPORT_SYMBOL(napi_gro_receive);
3640 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3642 __skb_pull(skb, skb_headlen(skb));
3643 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3644 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3646 skb->dev = napi->dev;
3652 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3654 struct sk_buff *skb = napi->skb;
3657 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3663 EXPORT_SYMBOL(napi_get_frags);
3665 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3671 skb->protocol = eth_type_trans(skb, skb->dev);
3673 if (ret == GRO_HELD)
3674 skb_gro_pull(skb, -ETH_HLEN);
3675 else if (netif_receive_skb(skb))
3680 case GRO_MERGED_FREE:
3681 napi_reuse_skb(napi, skb);
3690 EXPORT_SYMBOL(napi_frags_finish);
3692 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3694 struct sk_buff *skb = napi->skb;
3701 skb_reset_mac_header(skb);
3702 skb_gro_reset_offset(skb);
3704 off = skb_gro_offset(skb);
3705 hlen = off + sizeof(*eth);
3706 eth = skb_gro_header_fast(skb, off);
3707 if (skb_gro_header_hard(skb, hlen)) {
3708 eth = skb_gro_header_slow(skb, hlen, off);
3709 if (unlikely(!eth)) {
3710 napi_reuse_skb(napi, skb);
3716 skb_gro_pull(skb, sizeof(*eth));
3719 * This works because the only protocols we care about don't require
3720 * special handling. We'll fix it up properly at the end.
3722 skb->protocol = eth->h_proto;
3728 gro_result_t napi_gro_frags(struct napi_struct *napi)
3730 struct sk_buff *skb = napi_frags_skb(napi);
3735 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3737 EXPORT_SYMBOL(napi_gro_frags);
3740 * net_rps_action sends any pending IPI's for rps.
3741 * Note: called with local irq disabled, but exits with local irq enabled.
3743 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3746 struct softnet_data *remsd = sd->rps_ipi_list;
3749 sd->rps_ipi_list = NULL;
3753 /* Send pending IPI's to kick RPS processing on remote cpus. */
3755 struct softnet_data *next = remsd->rps_ipi_next;
3757 if (cpu_online(remsd->cpu))
3758 __smp_call_function_single(remsd->cpu,
3767 static int process_backlog(struct napi_struct *napi, int quota)
3770 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3773 /* Check if we have pending ipi, its better to send them now,
3774 * not waiting net_rx_action() end.
3776 if (sd->rps_ipi_list) {
3777 local_irq_disable();
3778 net_rps_action_and_irq_enable(sd);
3781 napi->weight = weight_p;
3782 local_irq_disable();
3783 while (work < quota) {
3784 struct sk_buff *skb;
3787 while ((skb = __skb_dequeue(&sd->process_queue))) {
3789 __netif_receive_skb(skb);
3790 local_irq_disable();
3791 input_queue_head_incr(sd);
3792 if (++work >= quota) {
3799 qlen = skb_queue_len(&sd->input_pkt_queue);
3801 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3802 &sd->process_queue);
3804 if (qlen < quota - work) {
3806 * Inline a custom version of __napi_complete().
3807 * only current cpu owns and manipulates this napi,
3808 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3809 * we can use a plain write instead of clear_bit(),
3810 * and we dont need an smp_mb() memory barrier.
3812 list_del(&napi->poll_list);
3815 quota = work + qlen;
3825 * __napi_schedule - schedule for receive
3826 * @n: entry to schedule
3828 * The entry's receive function will be scheduled to run
3830 void __napi_schedule(struct napi_struct *n)
3832 unsigned long flags;
3834 local_irq_save(flags);
3835 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3836 local_irq_restore(flags);
3838 EXPORT_SYMBOL(__napi_schedule);
3840 void __napi_complete(struct napi_struct *n)
3842 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3843 BUG_ON(n->gro_list);
3845 list_del(&n->poll_list);
3846 smp_mb__before_clear_bit();
3847 clear_bit(NAPI_STATE_SCHED, &n->state);
3849 EXPORT_SYMBOL(__napi_complete);
3851 void napi_complete(struct napi_struct *n)
3853 unsigned long flags;
3856 * don't let napi dequeue from the cpu poll list
3857 * just in case its running on a different cpu
3859 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3863 local_irq_save(flags);
3865 local_irq_restore(flags);
3867 EXPORT_SYMBOL(napi_complete);
3869 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3870 int (*poll)(struct napi_struct *, int), int weight)
3872 INIT_LIST_HEAD(&napi->poll_list);
3873 napi->gro_count = 0;
3874 napi->gro_list = NULL;
3877 napi->weight = weight;
3878 list_add(&napi->dev_list, &dev->napi_list);
3880 #ifdef CONFIG_NETPOLL
3881 spin_lock_init(&napi->poll_lock);
3882 napi->poll_owner = -1;
3884 set_bit(NAPI_STATE_SCHED, &napi->state);
3886 EXPORT_SYMBOL(netif_napi_add);
3888 void netif_napi_del(struct napi_struct *napi)
3890 struct sk_buff *skb, *next;
3892 list_del_init(&napi->dev_list);
3893 napi_free_frags(napi);
3895 for (skb = napi->gro_list; skb; skb = next) {
3901 napi->gro_list = NULL;
3902 napi->gro_count = 0;
3904 EXPORT_SYMBOL(netif_napi_del);
3906 static void net_rx_action(struct softirq_action *h)
3908 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3909 unsigned long time_limit = jiffies + 2;
3910 int budget = netdev_budget;
3913 local_irq_disable();
3915 while (!list_empty(&sd->poll_list)) {
3916 struct napi_struct *n;
3919 /* If softirq window is exhuasted then punt.
3920 * Allow this to run for 2 jiffies since which will allow
3921 * an average latency of 1.5/HZ.
3923 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3928 /* Even though interrupts have been re-enabled, this
3929 * access is safe because interrupts can only add new
3930 * entries to the tail of this list, and only ->poll()
3931 * calls can remove this head entry from the list.
3933 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3935 have = netpoll_poll_lock(n);
3939 /* This NAPI_STATE_SCHED test is for avoiding a race
3940 * with netpoll's poll_napi(). Only the entity which
3941 * obtains the lock and sees NAPI_STATE_SCHED set will
3942 * actually make the ->poll() call. Therefore we avoid
3943 * accidentally calling ->poll() when NAPI is not scheduled.
3946 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3947 work = n->poll(n, weight);
3951 WARN_ON_ONCE(work > weight);
3955 local_irq_disable();
3957 /* Drivers must not modify the NAPI state if they
3958 * consume the entire weight. In such cases this code
3959 * still "owns" the NAPI instance and therefore can
3960 * move the instance around on the list at-will.
3962 if (unlikely(work == weight)) {
3963 if (unlikely(napi_disable_pending(n))) {
3966 local_irq_disable();
3968 list_move_tail(&n->poll_list, &sd->poll_list);
3971 netpoll_poll_unlock(have);
3974 net_rps_action_and_irq_enable(sd);
3976 #ifdef CONFIG_NET_DMA
3978 * There may not be any more sk_buffs coming right now, so push
3979 * any pending DMA copies to hardware
3981 dma_issue_pending_all();
3988 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3992 static gifconf_func_t *gifconf_list[NPROTO];
3995 * register_gifconf - register a SIOCGIF handler
3996 * @family: Address family
3997 * @gifconf: Function handler
3999 * Register protocol dependent address dumping routines. The handler
4000 * that is passed must not be freed or reused until it has been replaced
4001 * by another handler.
4003 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4005 if (family >= NPROTO)
4007 gifconf_list[family] = gifconf;
4010 EXPORT_SYMBOL(register_gifconf);
4014 * Map an interface index to its name (SIOCGIFNAME)
4018 * We need this ioctl for efficient implementation of the
4019 * if_indextoname() function required by the IPv6 API. Without
4020 * it, we would have to search all the interfaces to find a
4024 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4026 struct net_device *dev;
4030 * Fetch the caller's info block.
4033 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4037 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4043 strcpy(ifr.ifr_name, dev->name);
4046 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4052 * Perform a SIOCGIFCONF call. This structure will change
4053 * size eventually, and there is nothing I can do about it.
4054 * Thus we will need a 'compatibility mode'.
4057 static int dev_ifconf(struct net *net, char __user *arg)
4060 struct net_device *dev;
4067 * Fetch the caller's info block.
4070 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4077 * Loop over the interfaces, and write an info block for each.
4081 for_each_netdev(net, dev) {
4082 for (i = 0; i < NPROTO; i++) {
4083 if (gifconf_list[i]) {
4086 done = gifconf_list[i](dev, NULL, 0);
4088 done = gifconf_list[i](dev, pos + total,
4098 * All done. Write the updated control block back to the caller.
4100 ifc.ifc_len = total;
4103 * Both BSD and Solaris return 0 here, so we do too.
4105 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4108 #ifdef CONFIG_PROC_FS
4110 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4112 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4113 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4114 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4116 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4118 struct net *net = seq_file_net(seq);
4119 struct net_device *dev;
4120 struct hlist_node *p;
4121 struct hlist_head *h;
4122 unsigned int count = 0, offset = get_offset(*pos);
4124 h = &net->dev_name_head[get_bucket(*pos)];
4125 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4126 if (++count == offset)
4133 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4135 struct net_device *dev;
4136 unsigned int bucket;
4139 dev = dev_from_same_bucket(seq, pos);
4143 bucket = get_bucket(*pos) + 1;
4144 *pos = set_bucket_offset(bucket, 1);
4145 } while (bucket < NETDEV_HASHENTRIES);
4151 * This is invoked by the /proc filesystem handler to display a device
4154 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4159 return SEQ_START_TOKEN;
4161 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4164 return dev_from_bucket(seq, pos);
4167 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4170 return dev_from_bucket(seq, pos);
4173 void dev_seq_stop(struct seq_file *seq, void *v)
4179 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4181 struct rtnl_link_stats64 temp;
4182 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4184 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4185 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4186 dev->name, stats->rx_bytes, stats->rx_packets,
4188 stats->rx_dropped + stats->rx_missed_errors,
4189 stats->rx_fifo_errors,
4190 stats->rx_length_errors + stats->rx_over_errors +
4191 stats->rx_crc_errors + stats->rx_frame_errors,
4192 stats->rx_compressed, stats->multicast,
4193 stats->tx_bytes, stats->tx_packets,
4194 stats->tx_errors, stats->tx_dropped,
4195 stats->tx_fifo_errors, stats->collisions,
4196 stats->tx_carrier_errors +
4197 stats->tx_aborted_errors +
4198 stats->tx_window_errors +
4199 stats->tx_heartbeat_errors,
4200 stats->tx_compressed);
4204 * Called from the PROCfs module. This now uses the new arbitrary sized
4205 * /proc/net interface to create /proc/net/dev
4207 static int dev_seq_show(struct seq_file *seq, void *v)
4209 if (v == SEQ_START_TOKEN)
4210 seq_puts(seq, "Inter-| Receive "
4212 " face |bytes packets errs drop fifo frame "
4213 "compressed multicast|bytes packets errs "
4214 "drop fifo colls carrier compressed\n");
4216 dev_seq_printf_stats(seq, v);
4220 static struct softnet_data *softnet_get_online(loff_t *pos)
4222 struct softnet_data *sd = NULL;
4224 while (*pos < nr_cpu_ids)
4225 if (cpu_online(*pos)) {
4226 sd = &per_cpu(softnet_data, *pos);
4233 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4235 return softnet_get_online(pos);
4238 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4241 return softnet_get_online(pos);
4244 static void softnet_seq_stop(struct seq_file *seq, void *v)
4248 static int softnet_seq_show(struct seq_file *seq, void *v)
4250 struct softnet_data *sd = v;
4252 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4253 sd->processed, sd->dropped, sd->time_squeeze, 0,
4254 0, 0, 0, 0, /* was fastroute */
4255 sd->cpu_collision, sd->received_rps);
4259 static const struct seq_operations dev_seq_ops = {
4260 .start = dev_seq_start,
4261 .next = dev_seq_next,
4262 .stop = dev_seq_stop,
4263 .show = dev_seq_show,
4266 static int dev_seq_open(struct inode *inode, struct file *file)
4268 return seq_open_net(inode, file, &dev_seq_ops,
4269 sizeof(struct seq_net_private));
4272 static const struct file_operations dev_seq_fops = {
4273 .owner = THIS_MODULE,
4274 .open = dev_seq_open,
4276 .llseek = seq_lseek,
4277 .release = seq_release_net,
4280 static const struct seq_operations softnet_seq_ops = {
4281 .start = softnet_seq_start,
4282 .next = softnet_seq_next,
4283 .stop = softnet_seq_stop,
4284 .show = softnet_seq_show,
4287 static int softnet_seq_open(struct inode *inode, struct file *file)
4289 return seq_open(file, &softnet_seq_ops);
4292 static const struct file_operations softnet_seq_fops = {
4293 .owner = THIS_MODULE,
4294 .open = softnet_seq_open,
4296 .llseek = seq_lseek,
4297 .release = seq_release,
4300 static void *ptype_get_idx(loff_t pos)
4302 struct packet_type *pt = NULL;
4306 list_for_each_entry_rcu(pt, &ptype_all, list) {
4312 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4313 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4322 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4326 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4329 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4331 struct packet_type *pt;
4332 struct list_head *nxt;
4336 if (v == SEQ_START_TOKEN)
4337 return ptype_get_idx(0);
4340 nxt = pt->list.next;
4341 if (pt->type == htons(ETH_P_ALL)) {
4342 if (nxt != &ptype_all)
4345 nxt = ptype_base[0].next;
4347 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4349 while (nxt == &ptype_base[hash]) {
4350 if (++hash >= PTYPE_HASH_SIZE)
4352 nxt = ptype_base[hash].next;
4355 return list_entry(nxt, struct packet_type, list);
4358 static void ptype_seq_stop(struct seq_file *seq, void *v)
4364 static int ptype_seq_show(struct seq_file *seq, void *v)
4366 struct packet_type *pt = v;
4368 if (v == SEQ_START_TOKEN)
4369 seq_puts(seq, "Type Device Function\n");
4370 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4371 if (pt->type == htons(ETH_P_ALL))
4372 seq_puts(seq, "ALL ");
4374 seq_printf(seq, "%04x", ntohs(pt->type));
4376 seq_printf(seq, " %-8s %pF\n",
4377 pt->dev ? pt->dev->name : "", pt->func);
4383 static const struct seq_operations ptype_seq_ops = {
4384 .start = ptype_seq_start,
4385 .next = ptype_seq_next,
4386 .stop = ptype_seq_stop,
4387 .show = ptype_seq_show,
4390 static int ptype_seq_open(struct inode *inode, struct file *file)
4392 return seq_open_net(inode, file, &ptype_seq_ops,
4393 sizeof(struct seq_net_private));
4396 static const struct file_operations ptype_seq_fops = {
4397 .owner = THIS_MODULE,
4398 .open = ptype_seq_open,
4400 .llseek = seq_lseek,
4401 .release = seq_release_net,
4405 static int __net_init dev_proc_net_init(struct net *net)
4409 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4411 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4413 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4416 if (wext_proc_init(net))
4422 proc_net_remove(net, "ptype");
4424 proc_net_remove(net, "softnet_stat");
4426 proc_net_remove(net, "dev");
4430 static void __net_exit dev_proc_net_exit(struct net *net)
4432 wext_proc_exit(net);
4434 proc_net_remove(net, "ptype");
4435 proc_net_remove(net, "softnet_stat");
4436 proc_net_remove(net, "dev");
4439 static struct pernet_operations __net_initdata dev_proc_ops = {
4440 .init = dev_proc_net_init,
4441 .exit = dev_proc_net_exit,
4444 static int __init dev_proc_init(void)
4446 return register_pernet_subsys(&dev_proc_ops);
4449 #define dev_proc_init() 0
4450 #endif /* CONFIG_PROC_FS */
4454 * netdev_set_master - set up master pointer
4455 * @slave: slave device
4456 * @master: new master device
4458 * Changes the master device of the slave. Pass %NULL to break the
4459 * bonding. The caller must hold the RTNL semaphore. On a failure
4460 * a negative errno code is returned. On success the reference counts
4461 * are adjusted and the function returns zero.
4463 int netdev_set_master(struct net_device *slave, struct net_device *master)
4465 struct net_device *old = slave->master;
4475 slave->master = master;
4481 EXPORT_SYMBOL(netdev_set_master);
4484 * netdev_set_bond_master - set up bonding master/slave pair
4485 * @slave: slave device
4486 * @master: new master device
4488 * Changes the master device of the slave. Pass %NULL to break the
4489 * bonding. The caller must hold the RTNL semaphore. On a failure
4490 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4491 * to the routing socket and the function returns zero.
4493 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4499 err = netdev_set_master(slave, master);
4503 slave->flags |= IFF_SLAVE;
4505 slave->flags &= ~IFF_SLAVE;
4507 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4510 EXPORT_SYMBOL(netdev_set_bond_master);
4512 static void dev_change_rx_flags(struct net_device *dev, int flags)
4514 const struct net_device_ops *ops = dev->netdev_ops;
4516 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4517 ops->ndo_change_rx_flags(dev, flags);
4520 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4522 unsigned int old_flags = dev->flags;
4528 dev->flags |= IFF_PROMISC;
4529 dev->promiscuity += inc;
4530 if (dev->promiscuity == 0) {
4533 * If inc causes overflow, untouch promisc and return error.
4536 dev->flags &= ~IFF_PROMISC;
4538 dev->promiscuity -= inc;
4539 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4544 if (dev->flags != old_flags) {
4545 pr_info("device %s %s promiscuous mode\n",
4547 dev->flags & IFF_PROMISC ? "entered" : "left");
4548 if (audit_enabled) {
4549 current_uid_gid(&uid, &gid);
4550 audit_log(current->audit_context, GFP_ATOMIC,
4551 AUDIT_ANOM_PROMISCUOUS,
4552 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4553 dev->name, (dev->flags & IFF_PROMISC),
4554 (old_flags & IFF_PROMISC),
4555 audit_get_loginuid(current),
4557 audit_get_sessionid(current));
4560 dev_change_rx_flags(dev, IFF_PROMISC);
4566 * dev_set_promiscuity - update promiscuity count on a device
4570 * Add or remove promiscuity from a device. While the count in the device
4571 * remains above zero the interface remains promiscuous. Once it hits zero
4572 * the device reverts back to normal filtering operation. A negative inc
4573 * value is used to drop promiscuity on the device.
4574 * Return 0 if successful or a negative errno code on error.
4576 int dev_set_promiscuity(struct net_device *dev, int inc)
4578 unsigned int old_flags = dev->flags;
4581 err = __dev_set_promiscuity(dev, inc);
4584 if (dev->flags != old_flags)
4585 dev_set_rx_mode(dev);
4588 EXPORT_SYMBOL(dev_set_promiscuity);
4591 * dev_set_allmulti - update allmulti count on a device
4595 * Add or remove reception of all multicast frames to a device. While the
4596 * count in the device remains above zero the interface remains listening
4597 * to all interfaces. Once it hits zero the device reverts back to normal
4598 * filtering operation. A negative @inc value is used to drop the counter
4599 * when releasing a resource needing all multicasts.
4600 * Return 0 if successful or a negative errno code on error.
4603 int dev_set_allmulti(struct net_device *dev, int inc)
4605 unsigned int old_flags = dev->flags;
4609 dev->flags |= IFF_ALLMULTI;
4610 dev->allmulti += inc;
4611 if (dev->allmulti == 0) {
4614 * If inc causes overflow, untouch allmulti and return error.
4617 dev->flags &= ~IFF_ALLMULTI;
4619 dev->allmulti -= inc;
4620 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4625 if (dev->flags ^ old_flags) {
4626 dev_change_rx_flags(dev, IFF_ALLMULTI);
4627 dev_set_rx_mode(dev);
4631 EXPORT_SYMBOL(dev_set_allmulti);
4634 * Upload unicast and multicast address lists to device and
4635 * configure RX filtering. When the device doesn't support unicast
4636 * filtering it is put in promiscuous mode while unicast addresses
4639 void __dev_set_rx_mode(struct net_device *dev)
4641 const struct net_device_ops *ops = dev->netdev_ops;
4643 /* dev_open will call this function so the list will stay sane. */
4644 if (!(dev->flags&IFF_UP))
4647 if (!netif_device_present(dev))
4650 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4651 /* Unicast addresses changes may only happen under the rtnl,
4652 * therefore calling __dev_set_promiscuity here is safe.
4654 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4655 __dev_set_promiscuity(dev, 1);
4656 dev->uc_promisc = true;
4657 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4658 __dev_set_promiscuity(dev, -1);
4659 dev->uc_promisc = false;
4663 if (ops->ndo_set_rx_mode)
4664 ops->ndo_set_rx_mode(dev);
4667 void dev_set_rx_mode(struct net_device *dev)
4669 netif_addr_lock_bh(dev);
4670 __dev_set_rx_mode(dev);
4671 netif_addr_unlock_bh(dev);
4675 * dev_get_flags - get flags reported to userspace
4678 * Get the combination of flag bits exported through APIs to userspace.
4680 unsigned int dev_get_flags(const struct net_device *dev)
4684 flags = (dev->flags & ~(IFF_PROMISC |
4689 (dev->gflags & (IFF_PROMISC |
4692 if (netif_running(dev)) {
4693 if (netif_oper_up(dev))
4694 flags |= IFF_RUNNING;
4695 if (netif_carrier_ok(dev))
4696 flags |= IFF_LOWER_UP;
4697 if (netif_dormant(dev))
4698 flags |= IFF_DORMANT;
4703 EXPORT_SYMBOL(dev_get_flags);
4705 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4707 unsigned int old_flags = dev->flags;
4713 * Set the flags on our device.
4716 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4717 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4719 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4723 * Load in the correct multicast list now the flags have changed.
4726 if ((old_flags ^ flags) & IFF_MULTICAST)
4727 dev_change_rx_flags(dev, IFF_MULTICAST);
4729 dev_set_rx_mode(dev);
4732 * Have we downed the interface. We handle IFF_UP ourselves
4733 * according to user attempts to set it, rather than blindly
4738 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4739 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4742 dev_set_rx_mode(dev);
4745 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4746 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4748 dev->gflags ^= IFF_PROMISC;
4749 dev_set_promiscuity(dev, inc);
4752 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4753 is important. Some (broken) drivers set IFF_PROMISC, when
4754 IFF_ALLMULTI is requested not asking us and not reporting.
4756 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4757 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4759 dev->gflags ^= IFF_ALLMULTI;
4760 dev_set_allmulti(dev, inc);
4766 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4768 unsigned int changes = dev->flags ^ old_flags;
4770 if (changes & IFF_UP) {
4771 if (dev->flags & IFF_UP)
4772 call_netdevice_notifiers(NETDEV_UP, dev);
4774 call_netdevice_notifiers(NETDEV_DOWN, dev);
4777 if (dev->flags & IFF_UP &&
4778 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4779 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4783 * dev_change_flags - change device settings
4785 * @flags: device state flags
4787 * Change settings on device based state flags. The flags are
4788 * in the userspace exported format.
4790 int dev_change_flags(struct net_device *dev, unsigned int flags)
4793 unsigned int changes, old_flags = dev->flags;
4795 ret = __dev_change_flags(dev, flags);
4799 changes = old_flags ^ dev->flags;
4801 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4803 __dev_notify_flags(dev, old_flags);
4806 EXPORT_SYMBOL(dev_change_flags);
4809 * dev_set_mtu - Change maximum transfer unit
4811 * @new_mtu: new transfer unit
4813 * Change the maximum transfer size of the network device.
4815 int dev_set_mtu(struct net_device *dev, int new_mtu)
4817 const struct net_device_ops *ops = dev->netdev_ops;
4820 if (new_mtu == dev->mtu)
4823 /* MTU must be positive. */
4827 if (!netif_device_present(dev))
4831 if (ops->ndo_change_mtu)
4832 err = ops->ndo_change_mtu(dev, new_mtu);
4836 if (!err && dev->flags & IFF_UP)
4837 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4840 EXPORT_SYMBOL(dev_set_mtu);
4843 * dev_set_group - Change group this device belongs to
4845 * @new_group: group this device should belong to
4847 void dev_set_group(struct net_device *dev, int new_group)
4849 dev->group = new_group;
4851 EXPORT_SYMBOL(dev_set_group);
4854 * dev_set_mac_address - Change Media Access Control Address
4858 * Change the hardware (MAC) address of the device
4860 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4862 const struct net_device_ops *ops = dev->netdev_ops;
4865 if (!ops->ndo_set_mac_address)
4867 if (sa->sa_family != dev->type)
4869 if (!netif_device_present(dev))
4871 err = ops->ndo_set_mac_address(dev, sa);
4873 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4874 add_device_randomness(dev->dev_addr, dev->addr_len);
4877 EXPORT_SYMBOL(dev_set_mac_address);
4880 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4882 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4885 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4891 case SIOCGIFFLAGS: /* Get interface flags */
4892 ifr->ifr_flags = (short) dev_get_flags(dev);
4895 case SIOCGIFMETRIC: /* Get the metric on the interface
4896 (currently unused) */
4897 ifr->ifr_metric = 0;
4900 case SIOCGIFMTU: /* Get the MTU of a device */
4901 ifr->ifr_mtu = dev->mtu;
4906 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4908 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4909 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4910 ifr->ifr_hwaddr.sa_family = dev->type;
4918 ifr->ifr_map.mem_start = dev->mem_start;
4919 ifr->ifr_map.mem_end = dev->mem_end;
4920 ifr->ifr_map.base_addr = dev->base_addr;
4921 ifr->ifr_map.irq = dev->irq;
4922 ifr->ifr_map.dma = dev->dma;
4923 ifr->ifr_map.port = dev->if_port;
4927 ifr->ifr_ifindex = dev->ifindex;
4931 ifr->ifr_qlen = dev->tx_queue_len;
4935 /* dev_ioctl() should ensure this case
4947 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4949 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4952 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4953 const struct net_device_ops *ops;
4958 ops = dev->netdev_ops;
4961 case SIOCSIFFLAGS: /* Set interface flags */
4962 return dev_change_flags(dev, ifr->ifr_flags);
4964 case SIOCSIFMETRIC: /* Set the metric on the interface
4965 (currently unused) */
4968 case SIOCSIFMTU: /* Set the MTU of a device */
4969 return dev_set_mtu(dev, ifr->ifr_mtu);
4972 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4974 case SIOCSIFHWBROADCAST:
4975 if (ifr->ifr_hwaddr.sa_family != dev->type)
4977 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4978 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4979 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4983 if (ops->ndo_set_config) {
4984 if (!netif_device_present(dev))
4986 return ops->ndo_set_config(dev, &ifr->ifr_map);
4991 if (!ops->ndo_set_rx_mode ||
4992 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4994 if (!netif_device_present(dev))
4996 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4999 if (!ops->ndo_set_rx_mode ||
5000 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5002 if (!netif_device_present(dev))
5004 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5007 if (ifr->ifr_qlen < 0)
5009 dev->tx_queue_len = ifr->ifr_qlen;
5013 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5014 return dev_change_name(dev, ifr->ifr_newname);
5017 err = net_hwtstamp_validate(ifr);
5023 * Unknown or private ioctl
5026 if ((cmd >= SIOCDEVPRIVATE &&
5027 cmd <= SIOCDEVPRIVATE + 15) ||
5028 cmd == SIOCBONDENSLAVE ||
5029 cmd == SIOCBONDRELEASE ||
5030 cmd == SIOCBONDSETHWADDR ||
5031 cmd == SIOCBONDSLAVEINFOQUERY ||
5032 cmd == SIOCBONDINFOQUERY ||
5033 cmd == SIOCBONDCHANGEACTIVE ||
5034 cmd == SIOCGMIIPHY ||
5035 cmd == SIOCGMIIREG ||
5036 cmd == SIOCSMIIREG ||
5037 cmd == SIOCBRADDIF ||
5038 cmd == SIOCBRDELIF ||
5039 cmd == SIOCSHWTSTAMP ||
5040 cmd == SIOCWANDEV) {
5042 if (ops->ndo_do_ioctl) {
5043 if (netif_device_present(dev))
5044 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5056 * This function handles all "interface"-type I/O control requests. The actual
5057 * 'doing' part of this is dev_ifsioc above.
5061 * dev_ioctl - network device ioctl
5062 * @net: the applicable net namespace
5063 * @cmd: command to issue
5064 * @arg: pointer to a struct ifreq in user space
5066 * Issue ioctl functions to devices. This is normally called by the
5067 * user space syscall interfaces but can sometimes be useful for
5068 * other purposes. The return value is the return from the syscall if
5069 * positive or a negative errno code on error.
5072 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5078 /* One special case: SIOCGIFCONF takes ifconf argument
5079 and requires shared lock, because it sleeps writing
5083 if (cmd == SIOCGIFCONF) {
5085 ret = dev_ifconf(net, (char __user *) arg);
5089 if (cmd == SIOCGIFNAME)
5090 return dev_ifname(net, (struct ifreq __user *)arg);
5092 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5095 ifr.ifr_name[IFNAMSIZ-1] = 0;
5097 colon = strchr(ifr.ifr_name, ':');
5102 * See which interface the caller is talking about.
5107 * These ioctl calls:
5108 * - can be done by all.
5109 * - atomic and do not require locking.
5120 dev_load(net, ifr.ifr_name);
5122 ret = dev_ifsioc_locked(net, &ifr, cmd);
5127 if (copy_to_user(arg, &ifr,
5128 sizeof(struct ifreq)))
5134 dev_load(net, ifr.ifr_name);
5136 ret = dev_ethtool(net, &ifr);
5141 if (copy_to_user(arg, &ifr,
5142 sizeof(struct ifreq)))
5148 * These ioctl calls:
5149 * - require superuser power.
5150 * - require strict serialization.
5156 if (!capable(CAP_NET_ADMIN))
5158 dev_load(net, ifr.ifr_name);
5160 ret = dev_ifsioc(net, &ifr, cmd);
5165 if (copy_to_user(arg, &ifr,
5166 sizeof(struct ifreq)))
5172 * These ioctl calls:
5173 * - require superuser power.
5174 * - require strict serialization.
5175 * - do not return a value
5185 case SIOCSIFHWBROADCAST:
5188 case SIOCBONDENSLAVE:
5189 case SIOCBONDRELEASE:
5190 case SIOCBONDSETHWADDR:
5191 case SIOCBONDCHANGEACTIVE:
5195 if (!capable(CAP_NET_ADMIN))
5198 case SIOCBONDSLAVEINFOQUERY:
5199 case SIOCBONDINFOQUERY:
5200 dev_load(net, ifr.ifr_name);
5202 ret = dev_ifsioc(net, &ifr, cmd);
5207 /* Get the per device memory space. We can add this but
5208 * currently do not support it */
5210 /* Set the per device memory buffer space.
5211 * Not applicable in our case */
5216 * Unknown or private ioctl.
5219 if (cmd == SIOCWANDEV ||
5220 (cmd >= SIOCDEVPRIVATE &&
5221 cmd <= SIOCDEVPRIVATE + 15)) {
5222 dev_load(net, ifr.ifr_name);
5224 ret = dev_ifsioc(net, &ifr, cmd);
5226 if (!ret && copy_to_user(arg, &ifr,
5227 sizeof(struct ifreq)))
5231 /* Take care of Wireless Extensions */
5232 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5233 return wext_handle_ioctl(net, &ifr, cmd, arg);
5240 * dev_new_index - allocate an ifindex
5241 * @net: the applicable net namespace
5243 * Returns a suitable unique value for a new device interface
5244 * number. The caller must hold the rtnl semaphore or the
5245 * dev_base_lock to be sure it remains unique.
5247 static int dev_new_index(struct net *net)
5249 int ifindex = net->ifindex;
5253 if (!__dev_get_by_index(net, ifindex))
5254 return net->ifindex = ifindex;
5258 /* Delayed registration/unregisteration */
5259 static LIST_HEAD(net_todo_list);
5261 static void net_set_todo(struct net_device *dev)
5263 list_add_tail(&dev->todo_list, &net_todo_list);
5266 static void rollback_registered_many(struct list_head *head)
5268 struct net_device *dev, *tmp;
5270 BUG_ON(dev_boot_phase);
5273 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5274 /* Some devices call without registering
5275 * for initialization unwind. Remove those
5276 * devices and proceed with the remaining.
5278 if (dev->reg_state == NETREG_UNINITIALIZED) {
5279 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5283 list_del(&dev->unreg_list);
5286 dev->dismantle = true;
5287 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5290 /* If device is running, close it first. */
5291 dev_close_many(head);
5293 list_for_each_entry(dev, head, unreg_list) {
5294 /* And unlink it from device chain. */
5295 unlist_netdevice(dev);
5297 dev->reg_state = NETREG_UNREGISTERING;
5302 list_for_each_entry(dev, head, unreg_list) {
5303 /* Shutdown queueing discipline. */
5307 /* Notify protocols, that we are about to destroy
5308 this device. They should clean all the things.
5310 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5312 if (!dev->rtnl_link_ops ||
5313 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5314 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5317 * Flush the unicast and multicast chains
5322 if (dev->netdev_ops->ndo_uninit)
5323 dev->netdev_ops->ndo_uninit(dev);
5325 /* Notifier chain MUST detach us from master device. */
5326 WARN_ON(dev->master);
5328 /* Remove entries from kobject tree */
5329 netdev_unregister_kobject(dev);
5334 list_for_each_entry(dev, head, unreg_list)
5338 static void rollback_registered(struct net_device *dev)
5342 list_add(&dev->unreg_list, &single);
5343 rollback_registered_many(&single);
5347 static netdev_features_t netdev_fix_features(struct net_device *dev,
5348 netdev_features_t features)
5350 /* Fix illegal checksum combinations */
5351 if ((features & NETIF_F_HW_CSUM) &&
5352 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5353 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5354 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5357 /* Fix illegal SG+CSUM combinations. */
5358 if ((features & NETIF_F_SG) &&
5359 !(features & NETIF_F_ALL_CSUM)) {
5361 "Dropping NETIF_F_SG since no checksum feature.\n");
5362 features &= ~NETIF_F_SG;
5365 /* TSO requires that SG is present as well. */
5366 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5367 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5368 features &= ~NETIF_F_ALL_TSO;
5371 /* TSO ECN requires that TSO is present as well. */
5372 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5373 features &= ~NETIF_F_TSO_ECN;
5375 /* Software GSO depends on SG. */
5376 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5377 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5378 features &= ~NETIF_F_GSO;
5381 /* UFO needs SG and checksumming */
5382 if (features & NETIF_F_UFO) {
5383 /* maybe split UFO into V4 and V6? */
5384 if (!((features & NETIF_F_GEN_CSUM) ||
5385 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5386 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5388 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5389 features &= ~NETIF_F_UFO;
5392 if (!(features & NETIF_F_SG)) {
5394 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5395 features &= ~NETIF_F_UFO;
5402 int __netdev_update_features(struct net_device *dev)
5404 netdev_features_t features;
5409 features = netdev_get_wanted_features(dev);
5411 if (dev->netdev_ops->ndo_fix_features)
5412 features = dev->netdev_ops->ndo_fix_features(dev, features);
5414 /* driver might be less strict about feature dependencies */
5415 features = netdev_fix_features(dev, features);
5417 if (dev->features == features)
5420 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5421 &dev->features, &features);
5423 if (dev->netdev_ops->ndo_set_features)
5424 err = dev->netdev_ops->ndo_set_features(dev, features);
5426 if (unlikely(err < 0)) {
5428 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5429 err, &features, &dev->features);
5434 dev->features = features;
5440 * netdev_update_features - recalculate device features
5441 * @dev: the device to check
5443 * Recalculate dev->features set and send notifications if it
5444 * has changed. Should be called after driver or hardware dependent
5445 * conditions might have changed that influence the features.
5447 void netdev_update_features(struct net_device *dev)
5449 if (__netdev_update_features(dev))
5450 netdev_features_change(dev);
5452 EXPORT_SYMBOL(netdev_update_features);
5455 * netdev_change_features - recalculate device features
5456 * @dev: the device to check
5458 * Recalculate dev->features set and send notifications even
5459 * if they have not changed. Should be called instead of
5460 * netdev_update_features() if also dev->vlan_features might
5461 * have changed to allow the changes to be propagated to stacked
5464 void netdev_change_features(struct net_device *dev)
5466 __netdev_update_features(dev);
5467 netdev_features_change(dev);
5469 EXPORT_SYMBOL(netdev_change_features);
5472 * netif_stacked_transfer_operstate - transfer operstate
5473 * @rootdev: the root or lower level device to transfer state from
5474 * @dev: the device to transfer operstate to
5476 * Transfer operational state from root to device. This is normally
5477 * called when a stacking relationship exists between the root
5478 * device and the device(a leaf device).
5480 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5481 struct net_device *dev)
5483 if (rootdev->operstate == IF_OPER_DORMANT)
5484 netif_dormant_on(dev);
5486 netif_dormant_off(dev);
5488 if (netif_carrier_ok(rootdev)) {
5489 if (!netif_carrier_ok(dev))
5490 netif_carrier_on(dev);
5492 if (netif_carrier_ok(dev))
5493 netif_carrier_off(dev);
5496 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5499 static int netif_alloc_rx_queues(struct net_device *dev)
5501 unsigned int i, count = dev->num_rx_queues;
5502 struct netdev_rx_queue *rx;
5506 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5508 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5513 for (i = 0; i < count; i++)
5519 static void netdev_init_one_queue(struct net_device *dev,
5520 struct netdev_queue *queue, void *_unused)
5522 /* Initialize queue lock */
5523 spin_lock_init(&queue->_xmit_lock);
5524 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5525 queue->xmit_lock_owner = -1;
5526 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5529 dql_init(&queue->dql, HZ);
5533 static int netif_alloc_netdev_queues(struct net_device *dev)
5535 unsigned int count = dev->num_tx_queues;
5536 struct netdev_queue *tx;
5540 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5542 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5547 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5548 spin_lock_init(&dev->tx_global_lock);
5554 * register_netdevice - register a network device
5555 * @dev: device to register
5557 * Take a completed network device structure and add it to the kernel
5558 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5559 * chain. 0 is returned on success. A negative errno code is returned
5560 * on a failure to set up the device, or if the name is a duplicate.
5562 * Callers must hold the rtnl semaphore. You may want
5563 * register_netdev() instead of this.
5566 * The locking appears insufficient to guarantee two parallel registers
5567 * will not get the same name.
5570 int register_netdevice(struct net_device *dev)
5573 struct net *net = dev_net(dev);
5575 BUG_ON(dev_boot_phase);
5580 /* When net_device's are persistent, this will be fatal. */
5581 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5584 spin_lock_init(&dev->addr_list_lock);
5585 netdev_set_addr_lockdep_class(dev);
5589 ret = dev_get_valid_name(dev, dev->name);
5593 /* Init, if this function is available */
5594 if (dev->netdev_ops->ndo_init) {
5595 ret = dev->netdev_ops->ndo_init(dev);
5605 dev->ifindex = dev_new_index(net);
5606 else if (__dev_get_by_index(net, dev->ifindex))
5609 if (dev->iflink == -1)
5610 dev->iflink = dev->ifindex;
5612 /* Transfer changeable features to wanted_features and enable
5613 * software offloads (GSO and GRO).
5615 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5616 dev->features |= NETIF_F_SOFT_FEATURES;
5617 dev->wanted_features = dev->features & dev->hw_features;
5619 /* Turn on no cache copy if HW is doing checksum */
5620 if (!(dev->flags & IFF_LOOPBACK)) {
5621 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5622 if (dev->features & NETIF_F_ALL_CSUM) {
5623 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5624 dev->features |= NETIF_F_NOCACHE_COPY;
5628 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5630 dev->vlan_features |= NETIF_F_HIGHDMA;
5632 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5633 ret = notifier_to_errno(ret);
5637 ret = netdev_register_kobject(dev);
5640 dev->reg_state = NETREG_REGISTERED;
5642 __netdev_update_features(dev);
5645 * Default initial state at registry is that the
5646 * device is present.
5649 set_bit(__LINK_STATE_PRESENT, &dev->state);
5651 dev_init_scheduler(dev);
5653 list_netdevice(dev);
5654 add_device_randomness(dev->dev_addr, dev->addr_len);
5656 /* Notify protocols, that a new device appeared. */
5657 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5658 ret = notifier_to_errno(ret);
5660 rollback_registered(dev);
5661 dev->reg_state = NETREG_UNREGISTERED;
5664 * Prevent userspace races by waiting until the network
5665 * device is fully setup before sending notifications.
5667 if (!dev->rtnl_link_ops ||
5668 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5669 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5675 if (dev->netdev_ops->ndo_uninit)
5676 dev->netdev_ops->ndo_uninit(dev);
5679 EXPORT_SYMBOL(register_netdevice);
5682 * init_dummy_netdev - init a dummy network device for NAPI
5683 * @dev: device to init
5685 * This takes a network device structure and initialize the minimum
5686 * amount of fields so it can be used to schedule NAPI polls without
5687 * registering a full blown interface. This is to be used by drivers
5688 * that need to tie several hardware interfaces to a single NAPI
5689 * poll scheduler due to HW limitations.
5691 int init_dummy_netdev(struct net_device *dev)
5693 /* Clear everything. Note we don't initialize spinlocks
5694 * are they aren't supposed to be taken by any of the
5695 * NAPI code and this dummy netdev is supposed to be
5696 * only ever used for NAPI polls
5698 memset(dev, 0, sizeof(struct net_device));
5700 /* make sure we BUG if trying to hit standard
5701 * register/unregister code path
5703 dev->reg_state = NETREG_DUMMY;
5705 /* NAPI wants this */
5706 INIT_LIST_HEAD(&dev->napi_list);
5708 /* a dummy interface is started by default */
5709 set_bit(__LINK_STATE_PRESENT, &dev->state);
5710 set_bit(__LINK_STATE_START, &dev->state);
5712 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5713 * because users of this 'device' dont need to change
5719 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5723 * register_netdev - register a network device
5724 * @dev: device to register
5726 * Take a completed network device structure and add it to the kernel
5727 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5728 * chain. 0 is returned on success. A negative errno code is returned
5729 * on a failure to set up the device, or if the name is a duplicate.
5731 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5732 * and expands the device name if you passed a format string to
5735 int register_netdev(struct net_device *dev)
5740 err = register_netdevice(dev);
5744 EXPORT_SYMBOL(register_netdev);
5746 int netdev_refcnt_read(const struct net_device *dev)
5750 for_each_possible_cpu(i)
5751 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5754 EXPORT_SYMBOL(netdev_refcnt_read);
5757 * netdev_wait_allrefs - wait until all references are gone.
5758 * @dev: target net_device
5760 * This is called when unregistering network devices.
5762 * Any protocol or device that holds a reference should register
5763 * for netdevice notification, and cleanup and put back the
5764 * reference if they receive an UNREGISTER event.
5765 * We can get stuck here if buggy protocols don't correctly
5768 static void netdev_wait_allrefs(struct net_device *dev)
5770 unsigned long rebroadcast_time, warning_time;
5773 linkwatch_forget_dev(dev);
5775 rebroadcast_time = warning_time = jiffies;
5776 refcnt = netdev_refcnt_read(dev);
5778 while (refcnt != 0) {
5779 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5782 /* Rebroadcast unregister notification */
5783 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5789 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5790 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5792 /* We must not have linkwatch events
5793 * pending on unregister. If this
5794 * happens, we simply run the queue
5795 * unscheduled, resulting in a noop
5798 linkwatch_run_queue();
5803 rebroadcast_time = jiffies;
5808 refcnt = netdev_refcnt_read(dev);
5810 if (time_after(jiffies, warning_time + 10 * HZ)) {
5811 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5813 warning_time = jiffies;
5822 * register_netdevice(x1);
5823 * register_netdevice(x2);
5825 * unregister_netdevice(y1);
5826 * unregister_netdevice(y2);
5832 * We are invoked by rtnl_unlock().
5833 * This allows us to deal with problems:
5834 * 1) We can delete sysfs objects which invoke hotplug
5835 * without deadlocking with linkwatch via keventd.
5836 * 2) Since we run with the RTNL semaphore not held, we can sleep
5837 * safely in order to wait for the netdev refcnt to drop to zero.
5839 * We must not return until all unregister events added during
5840 * the interval the lock was held have been completed.
5842 void netdev_run_todo(void)
5844 struct list_head list;
5846 /* Snapshot list, allow later requests */
5847 list_replace_init(&net_todo_list, &list);
5852 /* Wait for rcu callbacks to finish before next phase */
5853 if (!list_empty(&list))
5856 while (!list_empty(&list)) {
5857 struct net_device *dev
5858 = list_first_entry(&list, struct net_device, todo_list);
5859 list_del(&dev->todo_list);
5862 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5865 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5866 pr_err("network todo '%s' but state %d\n",
5867 dev->name, dev->reg_state);
5872 dev->reg_state = NETREG_UNREGISTERED;
5874 on_each_cpu(flush_backlog, dev, 1);
5876 netdev_wait_allrefs(dev);
5879 BUG_ON(netdev_refcnt_read(dev));
5880 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5881 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5882 WARN_ON(dev->dn_ptr);
5884 if (dev->destructor)
5885 dev->destructor(dev);
5887 /* Free network device */
5888 kobject_put(&dev->dev.kobj);
5892 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5893 * fields in the same order, with only the type differing.
5895 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5896 const struct net_device_stats *netdev_stats)
5898 #if BITS_PER_LONG == 64
5899 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5900 memcpy(stats64, netdev_stats, sizeof(*stats64));
5902 size_t i, n = sizeof(*stats64) / sizeof(u64);
5903 const unsigned long *src = (const unsigned long *)netdev_stats;
5904 u64 *dst = (u64 *)stats64;
5906 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5907 sizeof(*stats64) / sizeof(u64));
5908 for (i = 0; i < n; i++)
5912 EXPORT_SYMBOL(netdev_stats_to_stats64);
5915 * dev_get_stats - get network device statistics
5916 * @dev: device to get statistics from
5917 * @storage: place to store stats
5919 * Get network statistics from device. Return @storage.
5920 * The device driver may provide its own method by setting
5921 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5922 * otherwise the internal statistics structure is used.
5924 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5925 struct rtnl_link_stats64 *storage)
5927 const struct net_device_ops *ops = dev->netdev_ops;
5929 if (ops->ndo_get_stats64) {
5930 memset(storage, 0, sizeof(*storage));
5931 ops->ndo_get_stats64(dev, storage);
5932 } else if (ops->ndo_get_stats) {
5933 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5935 netdev_stats_to_stats64(storage, &dev->stats);
5937 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5940 EXPORT_SYMBOL(dev_get_stats);
5942 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5944 struct netdev_queue *queue = dev_ingress_queue(dev);
5946 #ifdef CONFIG_NET_CLS_ACT
5949 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5952 netdev_init_one_queue(dev, queue, NULL);
5953 queue->qdisc = &noop_qdisc;
5954 queue->qdisc_sleeping = &noop_qdisc;
5955 rcu_assign_pointer(dev->ingress_queue, queue);
5961 * alloc_netdev_mqs - allocate network device
5962 * @sizeof_priv: size of private data to allocate space for
5963 * @name: device name format string
5964 * @setup: callback to initialize device
5965 * @txqs: the number of TX subqueues to allocate
5966 * @rxqs: the number of RX subqueues to allocate
5968 * Allocates a struct net_device with private data area for driver use
5969 * and performs basic initialization. Also allocates subquue structs
5970 * for each queue on the device.
5972 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5973 void (*setup)(struct net_device *),
5974 unsigned int txqs, unsigned int rxqs)
5976 struct net_device *dev;
5978 struct net_device *p;
5980 BUG_ON(strlen(name) >= sizeof(dev->name));
5983 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5989 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5994 alloc_size = sizeof(struct net_device);
5996 /* ensure 32-byte alignment of private area */
5997 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5998 alloc_size += sizeof_priv;
6000 /* ensure 32-byte alignment of whole construct */
6001 alloc_size += NETDEV_ALIGN - 1;
6003 p = kzalloc(alloc_size, GFP_KERNEL);
6005 pr_err("alloc_netdev: Unable to allocate device\n");
6009 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6010 dev->padded = (char *)dev - (char *)p;
6012 dev->pcpu_refcnt = alloc_percpu(int);
6013 if (!dev->pcpu_refcnt)
6016 if (dev_addr_init(dev))
6022 dev_net_set(dev, &init_net);
6024 dev->gso_max_size = GSO_MAX_SIZE;
6025 dev->gso_max_segs = GSO_MAX_SEGS;
6027 INIT_LIST_HEAD(&dev->napi_list);
6028 INIT_LIST_HEAD(&dev->unreg_list);
6029 INIT_LIST_HEAD(&dev->link_watch_list);
6030 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6033 dev->num_tx_queues = txqs;
6034 dev->real_num_tx_queues = txqs;
6035 if (netif_alloc_netdev_queues(dev))
6039 dev->num_rx_queues = rxqs;
6040 dev->real_num_rx_queues = rxqs;
6041 if (netif_alloc_rx_queues(dev))
6045 strcpy(dev->name, name);
6046 dev->group = INIT_NETDEV_GROUP;
6054 free_percpu(dev->pcpu_refcnt);
6064 EXPORT_SYMBOL(alloc_netdev_mqs);
6067 * free_netdev - free network device
6070 * This function does the last stage of destroying an allocated device
6071 * interface. The reference to the device object is released.
6072 * If this is the last reference then it will be freed.
6074 void free_netdev(struct net_device *dev)
6076 struct napi_struct *p, *n;
6078 release_net(dev_net(dev));
6085 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6087 /* Flush device addresses */
6088 dev_addr_flush(dev);
6090 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6093 free_percpu(dev->pcpu_refcnt);
6094 dev->pcpu_refcnt = NULL;
6096 /* Compatibility with error handling in drivers */
6097 if (dev->reg_state == NETREG_UNINITIALIZED) {
6098 kfree((char *)dev - dev->padded);
6102 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6103 dev->reg_state = NETREG_RELEASED;
6105 /* will free via device release */
6106 put_device(&dev->dev);
6108 EXPORT_SYMBOL(free_netdev);
6111 * synchronize_net - Synchronize with packet receive processing
6113 * Wait for packets currently being received to be done.
6114 * Does not block later packets from starting.
6116 void synchronize_net(void)
6119 if (rtnl_is_locked())
6120 synchronize_rcu_expedited();
6124 EXPORT_SYMBOL(synchronize_net);
6127 * unregister_netdevice_queue - remove device from the kernel
6131 * This function shuts down a device interface and removes it
6132 * from the kernel tables.
6133 * If head not NULL, device is queued to be unregistered later.
6135 * Callers must hold the rtnl semaphore. You may want
6136 * unregister_netdev() instead of this.
6139 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6144 list_move_tail(&dev->unreg_list, head);
6146 rollback_registered(dev);
6147 /* Finish processing unregister after unlock */
6151 EXPORT_SYMBOL(unregister_netdevice_queue);
6154 * unregister_netdevice_many - unregister many devices
6155 * @head: list of devices
6157 void unregister_netdevice_many(struct list_head *head)
6159 struct net_device *dev;
6161 if (!list_empty(head)) {
6162 rollback_registered_many(head);
6163 list_for_each_entry(dev, head, unreg_list)
6167 EXPORT_SYMBOL(unregister_netdevice_many);
6170 * unregister_netdev - remove device from the kernel
6173 * This function shuts down a device interface and removes it
6174 * from the kernel tables.
6176 * This is just a wrapper for unregister_netdevice that takes
6177 * the rtnl semaphore. In general you want to use this and not
6178 * unregister_netdevice.
6180 void unregister_netdev(struct net_device *dev)
6183 unregister_netdevice(dev);
6186 EXPORT_SYMBOL(unregister_netdev);
6189 * dev_change_net_namespace - move device to different nethost namespace
6191 * @net: network namespace
6192 * @pat: If not NULL name pattern to try if the current device name
6193 * is already taken in the destination network namespace.
6195 * This function shuts down a device interface and moves it
6196 * to a new network namespace. On success 0 is returned, on
6197 * a failure a netagive errno code is returned.
6199 * Callers must hold the rtnl semaphore.
6202 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6208 /* Don't allow namespace local devices to be moved. */
6210 if (dev->features & NETIF_F_NETNS_LOCAL)
6213 /* Ensure the device has been registrered */
6215 if (dev->reg_state != NETREG_REGISTERED)
6218 /* Get out if there is nothing todo */
6220 if (net_eq(dev_net(dev), net))
6223 /* Pick the destination device name, and ensure
6224 * we can use it in the destination network namespace.
6227 if (__dev_get_by_name(net, dev->name)) {
6228 /* We get here if we can't use the current device name */
6231 if (dev_get_valid_name(dev, pat) < 0)
6236 * And now a mini version of register_netdevice unregister_netdevice.
6239 /* If device is running close it first. */
6242 /* And unlink it from device chain */
6244 unlist_netdevice(dev);
6248 /* Shutdown queueing discipline. */
6251 /* Notify protocols, that we are about to destroy
6252 this device. They should clean all the things.
6254 Note that dev->reg_state stays at NETREG_REGISTERED.
6255 This is wanted because this way 8021q and macvlan know
6256 the device is just moving and can keep their slaves up.
6258 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6259 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6262 * Flush the unicast and multicast chains
6267 /* Actually switch the network namespace */
6268 dev_net_set(dev, net);
6270 /* If there is an ifindex conflict assign a new one */
6271 if (__dev_get_by_index(net, dev->ifindex)) {
6272 int iflink = (dev->iflink == dev->ifindex);
6273 dev->ifindex = dev_new_index(net);
6275 dev->iflink = dev->ifindex;
6278 /* Fixup kobjects */
6279 err = device_rename(&dev->dev, dev->name);
6282 /* Add the device back in the hashes */
6283 list_netdevice(dev);
6285 /* Notify protocols, that a new device appeared. */
6286 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6289 * Prevent userspace races by waiting until the network
6290 * device is fully setup before sending notifications.
6292 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6299 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6301 static int dev_cpu_callback(struct notifier_block *nfb,
6302 unsigned long action,
6305 struct sk_buff **list_skb;
6306 struct sk_buff *skb;
6307 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6308 struct softnet_data *sd, *oldsd;
6310 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6313 local_irq_disable();
6314 cpu = smp_processor_id();
6315 sd = &per_cpu(softnet_data, cpu);
6316 oldsd = &per_cpu(softnet_data, oldcpu);
6318 /* Find end of our completion_queue. */
6319 list_skb = &sd->completion_queue;
6321 list_skb = &(*list_skb)->next;
6322 /* Append completion queue from offline CPU. */
6323 *list_skb = oldsd->completion_queue;
6324 oldsd->completion_queue = NULL;
6326 /* Append output queue from offline CPU. */
6327 if (oldsd->output_queue) {
6328 *sd->output_queue_tailp = oldsd->output_queue;
6329 sd->output_queue_tailp = oldsd->output_queue_tailp;
6330 oldsd->output_queue = NULL;
6331 oldsd->output_queue_tailp = &oldsd->output_queue;
6333 /* Append NAPI poll list from offline CPU. */
6334 if (!list_empty(&oldsd->poll_list)) {
6335 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6336 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6339 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6342 /* Process offline CPU's input_pkt_queue */
6343 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6345 input_queue_head_incr(oldsd);
6347 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6349 input_queue_head_incr(oldsd);
6357 * netdev_increment_features - increment feature set by one
6358 * @all: current feature set
6359 * @one: new feature set
6360 * @mask: mask feature set
6362 * Computes a new feature set after adding a device with feature set
6363 * @one to the master device with current feature set @all. Will not
6364 * enable anything that is off in @mask. Returns the new feature set.
6366 netdev_features_t netdev_increment_features(netdev_features_t all,
6367 netdev_features_t one, netdev_features_t mask)
6369 if (mask & NETIF_F_GEN_CSUM)
6370 mask |= NETIF_F_ALL_CSUM;
6371 mask |= NETIF_F_VLAN_CHALLENGED;
6373 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6374 all &= one | ~NETIF_F_ALL_FOR_ALL;
6376 /* If one device supports hw checksumming, set for all. */
6377 if (all & NETIF_F_GEN_CSUM)
6378 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6382 EXPORT_SYMBOL(netdev_increment_features);
6384 static struct hlist_head *netdev_create_hash(void)
6387 struct hlist_head *hash;
6389 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6391 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6392 INIT_HLIST_HEAD(&hash[i]);
6397 /* Initialize per network namespace state */
6398 static int __net_init netdev_init(struct net *net)
6400 if (net != &init_net)
6401 INIT_LIST_HEAD(&net->dev_base_head);
6403 net->dev_name_head = netdev_create_hash();
6404 if (net->dev_name_head == NULL)
6407 net->dev_index_head = netdev_create_hash();
6408 if (net->dev_index_head == NULL)
6414 kfree(net->dev_name_head);
6420 * netdev_drivername - network driver for the device
6421 * @dev: network device
6423 * Determine network driver for device.
6425 const char *netdev_drivername(const struct net_device *dev)
6427 const struct device_driver *driver;
6428 const struct device *parent;
6429 const char *empty = "";
6431 parent = dev->dev.parent;
6435 driver = parent->driver;
6436 if (driver && driver->name)
6437 return driver->name;
6441 int __netdev_printk(const char *level, const struct net_device *dev,
6442 struct va_format *vaf)
6446 if (dev && dev->dev.parent)
6447 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6448 netdev_name(dev), vaf);
6450 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6452 r = printk("%s(NULL net_device): %pV", level, vaf);
6456 EXPORT_SYMBOL(__netdev_printk);
6458 int netdev_printk(const char *level, const struct net_device *dev,
6459 const char *format, ...)
6461 struct va_format vaf;
6465 va_start(args, format);
6470 r = __netdev_printk(level, dev, &vaf);
6475 EXPORT_SYMBOL(netdev_printk);
6477 #define define_netdev_printk_level(func, level) \
6478 int func(const struct net_device *dev, const char *fmt, ...) \
6481 struct va_format vaf; \
6484 va_start(args, fmt); \
6489 r = __netdev_printk(level, dev, &vaf); \
6494 EXPORT_SYMBOL(func);
6496 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6497 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6498 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6499 define_netdev_printk_level(netdev_err, KERN_ERR);
6500 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6501 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6502 define_netdev_printk_level(netdev_info, KERN_INFO);
6504 static void __net_exit netdev_exit(struct net *net)
6506 kfree(net->dev_name_head);
6507 kfree(net->dev_index_head);
6510 static struct pernet_operations __net_initdata netdev_net_ops = {
6511 .init = netdev_init,
6512 .exit = netdev_exit,
6515 static void __net_exit default_device_exit(struct net *net)
6517 struct net_device *dev, *aux;
6519 * Push all migratable network devices back to the
6520 * initial network namespace
6523 for_each_netdev_safe(net, dev, aux) {
6525 char fb_name[IFNAMSIZ];
6527 /* Ignore unmoveable devices (i.e. loopback) */
6528 if (dev->features & NETIF_F_NETNS_LOCAL)
6531 /* Leave virtual devices for the generic cleanup */
6532 if (dev->rtnl_link_ops)
6535 /* Push remaining network devices to init_net */
6536 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6537 err = dev_change_net_namespace(dev, &init_net, fb_name);
6539 pr_emerg("%s: failed to move %s to init_net: %d\n",
6540 __func__, dev->name, err);
6547 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6549 /* At exit all network devices most be removed from a network
6550 * namespace. Do this in the reverse order of registration.
6551 * Do this across as many network namespaces as possible to
6552 * improve batching efficiency.
6554 struct net_device *dev;
6556 LIST_HEAD(dev_kill_list);
6559 list_for_each_entry(net, net_list, exit_list) {
6560 for_each_netdev_reverse(net, dev) {
6561 if (dev->rtnl_link_ops)
6562 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6564 unregister_netdevice_queue(dev, &dev_kill_list);
6567 unregister_netdevice_many(&dev_kill_list);
6568 list_del(&dev_kill_list);
6572 static struct pernet_operations __net_initdata default_device_ops = {
6573 .exit = default_device_exit,
6574 .exit_batch = default_device_exit_batch,
6578 * Initialize the DEV module. At boot time this walks the device list and
6579 * unhooks any devices that fail to initialise (normally hardware not
6580 * present) and leaves us with a valid list of present and active devices.
6585 * This is called single threaded during boot, so no need
6586 * to take the rtnl semaphore.
6588 static int __init net_dev_init(void)
6590 int i, rc = -ENOMEM;
6592 BUG_ON(!dev_boot_phase);
6594 if (dev_proc_init())
6597 if (netdev_kobject_init())
6600 INIT_LIST_HEAD(&ptype_all);
6601 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6602 INIT_LIST_HEAD(&ptype_base[i]);
6604 if (register_pernet_subsys(&netdev_net_ops))
6608 * Initialise the packet receive queues.
6611 for_each_possible_cpu(i) {
6612 struct softnet_data *sd = &per_cpu(softnet_data, i);
6614 memset(sd, 0, sizeof(*sd));
6615 skb_queue_head_init(&sd->input_pkt_queue);
6616 skb_queue_head_init(&sd->process_queue);
6617 sd->completion_queue = NULL;
6618 INIT_LIST_HEAD(&sd->poll_list);
6619 sd->output_queue = NULL;
6620 sd->output_queue_tailp = &sd->output_queue;
6622 sd->csd.func = rps_trigger_softirq;
6628 sd->backlog.poll = process_backlog;
6629 sd->backlog.weight = weight_p;
6630 sd->backlog.gro_list = NULL;
6631 sd->backlog.gro_count = 0;
6636 /* The loopback device is special if any other network devices
6637 * is present in a network namespace the loopback device must
6638 * be present. Since we now dynamically allocate and free the
6639 * loopback device ensure this invariant is maintained by
6640 * keeping the loopback device as the first device on the
6641 * list of network devices. Ensuring the loopback devices
6642 * is the first device that appears and the last network device
6645 if (register_pernet_device(&loopback_net_ops))
6648 if (register_pernet_device(&default_device_ops))
6651 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6652 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6654 hotcpu_notifier(dev_cpu_callback, 0);
6662 subsys_initcall(net_dev_init);
6664 static int __init initialize_hashrnd(void)
6666 get_random_bytes(&hashrnd, sizeof(hashrnd));
6670 late_initcall_sync(initialize_hashrnd);