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);
1112 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1114 return call_netdevice_notifiers(event, dev);
1116 EXPORT_SYMBOL(netdev_bonding_change);
1119 * dev_load - load a network module
1120 * @net: the applicable net namespace
1121 * @name: name of interface
1123 * If a network interface is not present and the process has suitable
1124 * privileges this function loads the module. If module loading is not
1125 * available in this kernel then it becomes a nop.
1128 void dev_load(struct net *net, const char *name)
1130 struct net_device *dev;
1134 dev = dev_get_by_name_rcu(net, name);
1138 if (no_module && capable(CAP_NET_ADMIN))
1139 no_module = request_module("netdev-%s", name);
1140 if (no_module && capable(CAP_SYS_MODULE)) {
1141 if (!request_module("%s", name))
1142 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1146 EXPORT_SYMBOL(dev_load);
1148 static int __dev_open(struct net_device *dev)
1150 const struct net_device_ops *ops = dev->netdev_ops;
1155 if (!netif_device_present(dev))
1158 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1159 ret = notifier_to_errno(ret);
1163 set_bit(__LINK_STATE_START, &dev->state);
1165 if (ops->ndo_validate_addr)
1166 ret = ops->ndo_validate_addr(dev);
1168 if (!ret && ops->ndo_open)
1169 ret = ops->ndo_open(dev);
1172 clear_bit(__LINK_STATE_START, &dev->state);
1174 dev->flags |= IFF_UP;
1175 net_dmaengine_get();
1176 dev_set_rx_mode(dev);
1178 add_device_randomness(dev->dev_addr, dev->addr_len);
1185 * dev_open - prepare an interface for use.
1186 * @dev: device to open
1188 * Takes a device from down to up state. The device's private open
1189 * function is invoked and then the multicast lists are loaded. Finally
1190 * the device is moved into the up state and a %NETDEV_UP message is
1191 * sent to the netdev notifier chain.
1193 * Calling this function on an active interface is a nop. On a failure
1194 * a negative errno code is returned.
1196 int dev_open(struct net_device *dev)
1200 if (dev->flags & IFF_UP)
1203 ret = __dev_open(dev);
1207 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1208 call_netdevice_notifiers(NETDEV_UP, dev);
1212 EXPORT_SYMBOL(dev_open);
1214 static int __dev_close_many(struct list_head *head)
1216 struct net_device *dev;
1221 list_for_each_entry(dev, head, unreg_list) {
1222 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1224 clear_bit(__LINK_STATE_START, &dev->state);
1226 /* Synchronize to scheduled poll. We cannot touch poll list, it
1227 * can be even on different cpu. So just clear netif_running().
1229 * dev->stop() will invoke napi_disable() on all of it's
1230 * napi_struct instances on this device.
1232 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1235 dev_deactivate_many(head);
1237 list_for_each_entry(dev, head, unreg_list) {
1238 const struct net_device_ops *ops = dev->netdev_ops;
1241 * Call the device specific close. This cannot fail.
1242 * Only if device is UP
1244 * We allow it to be called even after a DETACH hot-plug
1250 dev->flags &= ~IFF_UP;
1251 net_dmaengine_put();
1257 static int __dev_close(struct net_device *dev)
1262 list_add(&dev->unreg_list, &single);
1263 retval = __dev_close_many(&single);
1268 static int dev_close_many(struct list_head *head)
1270 struct net_device *dev, *tmp;
1271 LIST_HEAD(tmp_list);
1273 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1274 if (!(dev->flags & IFF_UP))
1275 list_move(&dev->unreg_list, &tmp_list);
1277 __dev_close_many(head);
1279 list_for_each_entry(dev, head, unreg_list) {
1280 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1281 call_netdevice_notifiers(NETDEV_DOWN, dev);
1284 /* rollback_registered_many needs the complete original list */
1285 list_splice(&tmp_list, head);
1290 * dev_close - shutdown an interface.
1291 * @dev: device to shutdown
1293 * This function moves an active device into down state. A
1294 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1295 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1298 int dev_close(struct net_device *dev)
1300 if (dev->flags & IFF_UP) {
1303 list_add(&dev->unreg_list, &single);
1304 dev_close_many(&single);
1309 EXPORT_SYMBOL(dev_close);
1313 * dev_disable_lro - disable Large Receive Offload on a device
1316 * Disable Large Receive Offload (LRO) on a net device. Must be
1317 * called under RTNL. This is needed if received packets may be
1318 * forwarded to another interface.
1320 void dev_disable_lro(struct net_device *dev)
1323 * If we're trying to disable lro on a vlan device
1324 * use the underlying physical device instead
1326 if (is_vlan_dev(dev))
1327 dev = vlan_dev_real_dev(dev);
1329 dev->wanted_features &= ~NETIF_F_LRO;
1330 netdev_update_features(dev);
1332 if (unlikely(dev->features & NETIF_F_LRO))
1333 netdev_WARN(dev, "failed to disable LRO!\n");
1335 EXPORT_SYMBOL(dev_disable_lro);
1338 static int dev_boot_phase = 1;
1341 * register_netdevice_notifier - register a network notifier block
1344 * Register a notifier to be called when network device events occur.
1345 * The notifier passed is linked into the kernel structures and must
1346 * not be reused until it has been unregistered. A negative errno code
1347 * is returned on a failure.
1349 * When registered all registration and up events are replayed
1350 * to the new notifier to allow device to have a race free
1351 * view of the network device list.
1354 int register_netdevice_notifier(struct notifier_block *nb)
1356 struct net_device *dev;
1357 struct net_device *last;
1362 err = raw_notifier_chain_register(&netdev_chain, nb);
1368 for_each_netdev(net, dev) {
1369 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1370 err = notifier_to_errno(err);
1374 if (!(dev->flags & IFF_UP))
1377 nb->notifier_call(nb, NETDEV_UP, dev);
1388 for_each_netdev(net, dev) {
1392 if (dev->flags & IFF_UP) {
1393 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1394 nb->notifier_call(nb, NETDEV_DOWN, dev);
1396 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1397 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1402 raw_notifier_chain_unregister(&netdev_chain, nb);
1405 EXPORT_SYMBOL(register_netdevice_notifier);
1408 * unregister_netdevice_notifier - unregister a network notifier block
1411 * Unregister a notifier previously registered by
1412 * register_netdevice_notifier(). The notifier is unlinked into the
1413 * kernel structures and may then be reused. A negative errno code
1414 * is returned on a failure.
1416 * After unregistering unregister and down device events are synthesized
1417 * for all devices on the device list to the removed notifier to remove
1418 * the need for special case cleanup code.
1421 int unregister_netdevice_notifier(struct notifier_block *nb)
1423 struct net_device *dev;
1428 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1433 for_each_netdev(net, dev) {
1434 if (dev->flags & IFF_UP) {
1435 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1436 nb->notifier_call(nb, NETDEV_DOWN, dev);
1438 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1439 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1446 EXPORT_SYMBOL(unregister_netdevice_notifier);
1449 * call_netdevice_notifiers - call all network notifier blocks
1450 * @val: value passed unmodified to notifier function
1451 * @dev: net_device pointer passed unmodified to notifier function
1453 * Call all network notifier blocks. Parameters and return value
1454 * are as for raw_notifier_call_chain().
1457 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1460 return raw_notifier_call_chain(&netdev_chain, val, dev);
1462 EXPORT_SYMBOL(call_netdevice_notifiers);
1464 static struct static_key netstamp_needed __read_mostly;
1465 #ifdef HAVE_JUMP_LABEL
1466 /* We are not allowed to call static_key_slow_dec() from irq context
1467 * If net_disable_timestamp() is called from irq context, defer the
1468 * static_key_slow_dec() calls.
1470 static atomic_t netstamp_needed_deferred;
1473 void net_enable_timestamp(void)
1475 #ifdef HAVE_JUMP_LABEL
1476 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1480 static_key_slow_dec(&netstamp_needed);
1484 WARN_ON(in_interrupt());
1485 static_key_slow_inc(&netstamp_needed);
1487 EXPORT_SYMBOL(net_enable_timestamp);
1489 void net_disable_timestamp(void)
1491 #ifdef HAVE_JUMP_LABEL
1492 if (in_interrupt()) {
1493 atomic_inc(&netstamp_needed_deferred);
1497 static_key_slow_dec(&netstamp_needed);
1499 EXPORT_SYMBOL(net_disable_timestamp);
1501 static inline void net_timestamp_set(struct sk_buff *skb)
1503 skb->tstamp.tv64 = 0;
1504 if (static_key_false(&netstamp_needed))
1505 __net_timestamp(skb);
1508 #define net_timestamp_check(COND, SKB) \
1509 if (static_key_false(&netstamp_needed)) { \
1510 if ((COND) && !(SKB)->tstamp.tv64) \
1511 __net_timestamp(SKB); \
1514 static int net_hwtstamp_validate(struct ifreq *ifr)
1516 struct hwtstamp_config cfg;
1517 enum hwtstamp_tx_types tx_type;
1518 enum hwtstamp_rx_filters rx_filter;
1519 int tx_type_valid = 0;
1520 int rx_filter_valid = 0;
1522 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1525 if (cfg.flags) /* reserved for future extensions */
1528 tx_type = cfg.tx_type;
1529 rx_filter = cfg.rx_filter;
1532 case HWTSTAMP_TX_OFF:
1533 case HWTSTAMP_TX_ON:
1534 case HWTSTAMP_TX_ONESTEP_SYNC:
1539 switch (rx_filter) {
1540 case HWTSTAMP_FILTER_NONE:
1541 case HWTSTAMP_FILTER_ALL:
1542 case HWTSTAMP_FILTER_SOME:
1543 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1544 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1545 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1546 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1547 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1548 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1549 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1550 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1551 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1552 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1553 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1554 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1555 rx_filter_valid = 1;
1559 if (!tx_type_valid || !rx_filter_valid)
1565 static inline bool is_skb_forwardable(struct net_device *dev,
1566 struct sk_buff *skb)
1570 if (!(dev->flags & IFF_UP))
1573 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1574 if (skb->len <= len)
1577 /* if TSO is enabled, we don't care about the length as the packet
1578 * could be forwarded without being segmented before
1580 if (skb_is_gso(skb))
1587 * dev_forward_skb - loopback an skb to another netif
1589 * @dev: destination network device
1590 * @skb: buffer to forward
1593 * NET_RX_SUCCESS (no congestion)
1594 * NET_RX_DROP (packet was dropped, but freed)
1596 * dev_forward_skb can be used for injecting an skb from the
1597 * start_xmit function of one device into the receive queue
1598 * of another device.
1600 * The receiving device may be in another namespace, so
1601 * we have to clear all information in the skb that could
1602 * impact namespace isolation.
1604 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1606 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1607 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1608 atomic_long_inc(&dev->rx_dropped);
1617 if (unlikely(!is_skb_forwardable(dev, skb))) {
1618 atomic_long_inc(&dev->rx_dropped);
1625 skb->tstamp.tv64 = 0;
1626 skb->pkt_type = PACKET_HOST;
1627 skb->protocol = eth_type_trans(skb, dev);
1631 return netif_rx(skb);
1633 EXPORT_SYMBOL_GPL(dev_forward_skb);
1635 static inline int deliver_skb(struct sk_buff *skb,
1636 struct packet_type *pt_prev,
1637 struct net_device *orig_dev)
1639 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1641 atomic_inc(&skb->users);
1642 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1645 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1647 if (ptype->af_packet_priv == NULL)
1650 if (ptype->id_match)
1651 return ptype->id_match(ptype, skb->sk);
1652 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1659 * Support routine. Sends outgoing frames to any network
1660 * taps currently in use.
1663 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1665 struct packet_type *ptype;
1666 struct sk_buff *skb2 = NULL;
1667 struct packet_type *pt_prev = NULL;
1670 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1671 /* Never send packets back to the socket
1672 * they originated from - MvS (miquels@drinkel.ow.org)
1674 if ((ptype->dev == dev || !ptype->dev) &&
1675 (!skb_loop_sk(ptype, skb))) {
1677 deliver_skb(skb2, pt_prev, skb->dev);
1682 skb2 = skb_clone(skb, GFP_ATOMIC);
1686 net_timestamp_set(skb2);
1688 /* skb->nh should be correctly
1689 set by sender, so that the second statement is
1690 just protection against buggy protocols.
1692 skb_reset_mac_header(skb2);
1694 if (skb_network_header(skb2) < skb2->data ||
1695 skb2->network_header > skb2->tail) {
1696 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1697 ntohs(skb2->protocol),
1699 skb_reset_network_header(skb2);
1702 skb2->transport_header = skb2->network_header;
1703 skb2->pkt_type = PACKET_OUTGOING;
1708 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1713 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1714 * @dev: Network device
1715 * @txq: number of queues available
1717 * If real_num_tx_queues is changed the tc mappings may no longer be
1718 * valid. To resolve this verify the tc mapping remains valid and if
1719 * not NULL the mapping. With no priorities mapping to this
1720 * offset/count pair it will no longer be used. In the worst case TC0
1721 * is invalid nothing can be done so disable priority mappings. If is
1722 * expected that drivers will fix this mapping if they can before
1723 * calling netif_set_real_num_tx_queues.
1725 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1728 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1730 /* If TC0 is invalidated disable TC mapping */
1731 if (tc->offset + tc->count > txq) {
1732 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1737 /* Invalidated prio to tc mappings set to TC0 */
1738 for (i = 1; i < TC_BITMASK + 1; i++) {
1739 int q = netdev_get_prio_tc_map(dev, i);
1741 tc = &dev->tc_to_txq[q];
1742 if (tc->offset + tc->count > txq) {
1743 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1745 netdev_set_prio_tc_map(dev, i, 0);
1751 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1752 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1754 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1758 if (txq < 1 || txq > dev->num_tx_queues)
1761 if (dev->reg_state == NETREG_REGISTERED ||
1762 dev->reg_state == NETREG_UNREGISTERING) {
1765 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1771 netif_setup_tc(dev, txq);
1773 if (txq < dev->real_num_tx_queues)
1774 qdisc_reset_all_tx_gt(dev, txq);
1777 dev->real_num_tx_queues = txq;
1780 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1784 * netif_set_real_num_rx_queues - set actual number of RX queues used
1785 * @dev: Network device
1786 * @rxq: Actual number of RX queues
1788 * This must be called either with the rtnl_lock held or before
1789 * registration of the net device. Returns 0 on success, or a
1790 * negative error code. If called before registration, it always
1793 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1797 if (rxq < 1 || rxq > dev->num_rx_queues)
1800 if (dev->reg_state == NETREG_REGISTERED) {
1803 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1809 dev->real_num_rx_queues = rxq;
1812 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1816 * netif_get_num_default_rss_queues - default number of RSS queues
1818 * This routine should set an upper limit on the number of RSS queues
1819 * used by default by multiqueue devices.
1821 int netif_get_num_default_rss_queues(void)
1823 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
1825 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
1827 static inline void __netif_reschedule(struct Qdisc *q)
1829 struct softnet_data *sd;
1830 unsigned long flags;
1832 local_irq_save(flags);
1833 sd = &__get_cpu_var(softnet_data);
1834 q->next_sched = NULL;
1835 *sd->output_queue_tailp = q;
1836 sd->output_queue_tailp = &q->next_sched;
1837 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1838 local_irq_restore(flags);
1841 void __netif_schedule(struct Qdisc *q)
1843 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1844 __netif_reschedule(q);
1846 EXPORT_SYMBOL(__netif_schedule);
1848 void dev_kfree_skb_irq(struct sk_buff *skb)
1850 if (atomic_dec_and_test(&skb->users)) {
1851 struct softnet_data *sd;
1852 unsigned long flags;
1854 local_irq_save(flags);
1855 sd = &__get_cpu_var(softnet_data);
1856 skb->next = sd->completion_queue;
1857 sd->completion_queue = skb;
1858 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1859 local_irq_restore(flags);
1862 EXPORT_SYMBOL(dev_kfree_skb_irq);
1864 void dev_kfree_skb_any(struct sk_buff *skb)
1866 if (in_irq() || irqs_disabled())
1867 dev_kfree_skb_irq(skb);
1871 EXPORT_SYMBOL(dev_kfree_skb_any);
1875 * netif_device_detach - mark device as removed
1876 * @dev: network device
1878 * Mark device as removed from system and therefore no longer available.
1880 void netif_device_detach(struct net_device *dev)
1882 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1883 netif_running(dev)) {
1884 netif_tx_stop_all_queues(dev);
1887 EXPORT_SYMBOL(netif_device_detach);
1890 * netif_device_attach - mark device as attached
1891 * @dev: network device
1893 * Mark device as attached from system and restart if needed.
1895 void netif_device_attach(struct net_device *dev)
1897 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1898 netif_running(dev)) {
1899 netif_tx_wake_all_queues(dev);
1900 __netdev_watchdog_up(dev);
1903 EXPORT_SYMBOL(netif_device_attach);
1905 static void skb_warn_bad_offload(const struct sk_buff *skb)
1907 static const netdev_features_t null_features = 0;
1908 struct net_device *dev = skb->dev;
1909 const char *driver = "";
1911 if (dev && dev->dev.parent)
1912 driver = dev_driver_string(dev->dev.parent);
1914 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1915 "gso_type=%d ip_summed=%d\n",
1916 driver, dev ? &dev->features : &null_features,
1917 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1918 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1919 skb_shinfo(skb)->gso_type, skb->ip_summed);
1923 * Invalidate hardware checksum when packet is to be mangled, and
1924 * complete checksum manually on outgoing path.
1926 int skb_checksum_help(struct sk_buff *skb)
1929 int ret = 0, offset;
1931 if (skb->ip_summed == CHECKSUM_COMPLETE)
1932 goto out_set_summed;
1934 if (unlikely(skb_shinfo(skb)->gso_size)) {
1935 skb_warn_bad_offload(skb);
1939 offset = skb_checksum_start_offset(skb);
1940 BUG_ON(offset >= skb_headlen(skb));
1941 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1943 offset += skb->csum_offset;
1944 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1946 if (skb_cloned(skb) &&
1947 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1948 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1953 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1955 skb->ip_summed = CHECKSUM_NONE;
1959 EXPORT_SYMBOL(skb_checksum_help);
1962 * skb_gso_segment - Perform segmentation on skb.
1963 * @skb: buffer to segment
1964 * @features: features for the output path (see dev->features)
1966 * This function segments the given skb and returns a list of segments.
1968 * It may return NULL if the skb requires no segmentation. This is
1969 * only possible when GSO is used for verifying header integrity.
1971 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1972 netdev_features_t features)
1974 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1975 struct packet_type *ptype;
1976 __be16 type = skb->protocol;
1977 int vlan_depth = ETH_HLEN;
1980 while (type == htons(ETH_P_8021Q)) {
1981 struct vlan_hdr *vh;
1983 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1984 return ERR_PTR(-EINVAL);
1986 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1987 type = vh->h_vlan_encapsulated_proto;
1988 vlan_depth += VLAN_HLEN;
1991 skb_reset_mac_header(skb);
1992 skb->mac_len = skb->network_header - skb->mac_header;
1993 __skb_pull(skb, skb->mac_len);
1995 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1996 skb_warn_bad_offload(skb);
1998 if (skb_header_cloned(skb) &&
1999 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2000 return ERR_PTR(err);
2004 list_for_each_entry_rcu(ptype,
2005 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2006 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
2007 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2008 err = ptype->gso_send_check(skb);
2009 segs = ERR_PTR(err);
2010 if (err || skb_gso_ok(skb, features))
2012 __skb_push(skb, (skb->data -
2013 skb_network_header(skb)));
2015 segs = ptype->gso_segment(skb, features);
2021 __skb_push(skb, skb->data - skb_mac_header(skb));
2025 EXPORT_SYMBOL(skb_gso_segment);
2027 /* Take action when hardware reception checksum errors are detected. */
2029 void netdev_rx_csum_fault(struct net_device *dev)
2031 if (net_ratelimit()) {
2032 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2036 EXPORT_SYMBOL(netdev_rx_csum_fault);
2039 /* Actually, we should eliminate this check as soon as we know, that:
2040 * 1. IOMMU is present and allows to map all the memory.
2041 * 2. No high memory really exists on this machine.
2044 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2046 #ifdef CONFIG_HIGHMEM
2048 if (!(dev->features & NETIF_F_HIGHDMA)) {
2049 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2050 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2051 if (PageHighMem(skb_frag_page(frag)))
2056 if (PCI_DMA_BUS_IS_PHYS) {
2057 struct device *pdev = dev->dev.parent;
2061 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2062 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2063 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2064 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2073 void (*destructor)(struct sk_buff *skb);
2076 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2078 static void dev_gso_skb_destructor(struct sk_buff *skb)
2080 struct dev_gso_cb *cb;
2083 struct sk_buff *nskb = skb->next;
2085 skb->next = nskb->next;
2088 } while (skb->next);
2090 cb = DEV_GSO_CB(skb);
2092 cb->destructor(skb);
2096 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2097 * @skb: buffer to segment
2098 * @features: device features as applicable to this skb
2100 * This function segments the given skb and stores the list of segments
2103 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2105 struct sk_buff *segs;
2107 segs = skb_gso_segment(skb, features);
2109 /* Verifying header integrity only. */
2114 return PTR_ERR(segs);
2117 DEV_GSO_CB(skb)->destructor = skb->destructor;
2118 skb->destructor = dev_gso_skb_destructor;
2123 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2125 return ((features & NETIF_F_GEN_CSUM) ||
2126 ((features & NETIF_F_V4_CSUM) &&
2127 protocol == htons(ETH_P_IP)) ||
2128 ((features & NETIF_F_V6_CSUM) &&
2129 protocol == htons(ETH_P_IPV6)) ||
2130 ((features & NETIF_F_FCOE_CRC) &&
2131 protocol == htons(ETH_P_FCOE)));
2134 static netdev_features_t harmonize_features(struct sk_buff *skb,
2135 __be16 protocol, netdev_features_t features)
2137 if (!can_checksum_protocol(features, protocol)) {
2138 features &= ~NETIF_F_ALL_CSUM;
2139 features &= ~NETIF_F_SG;
2140 } else if (illegal_highdma(skb->dev, skb)) {
2141 features &= ~NETIF_F_SG;
2147 netdev_features_t netif_skb_features(struct sk_buff *skb)
2149 __be16 protocol = skb->protocol;
2150 netdev_features_t features = skb->dev->features;
2152 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2153 features &= ~NETIF_F_GSO_MASK;
2155 if (protocol == htons(ETH_P_8021Q)) {
2156 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2157 protocol = veh->h_vlan_encapsulated_proto;
2158 } else if (!vlan_tx_tag_present(skb)) {
2159 return harmonize_features(skb, protocol, features);
2162 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2164 if (protocol != htons(ETH_P_8021Q)) {
2165 return harmonize_features(skb, protocol, features);
2167 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2168 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2169 return harmonize_features(skb, protocol, features);
2172 EXPORT_SYMBOL(netif_skb_features);
2175 * Returns true if either:
2176 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2177 * 2. skb is fragmented and the device does not support SG, or if
2178 * at least one of fragments is in highmem and device does not
2179 * support DMA from it.
2181 static inline int skb_needs_linearize(struct sk_buff *skb,
2184 return skb_is_nonlinear(skb) &&
2185 ((skb_has_frag_list(skb) &&
2186 !(features & NETIF_F_FRAGLIST)) ||
2187 (skb_shinfo(skb)->nr_frags &&
2188 !(features & NETIF_F_SG)));
2191 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2192 struct netdev_queue *txq)
2194 const struct net_device_ops *ops = dev->netdev_ops;
2195 int rc = NETDEV_TX_OK;
2196 unsigned int skb_len;
2198 if (likely(!skb->next)) {
2199 netdev_features_t features;
2202 * If device doesn't need skb->dst, release it right now while
2203 * its hot in this cpu cache
2205 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2208 if (!list_empty(&ptype_all))
2209 dev_queue_xmit_nit(skb, dev);
2211 features = netif_skb_features(skb);
2213 if (vlan_tx_tag_present(skb) &&
2214 !(features & NETIF_F_HW_VLAN_TX)) {
2215 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2222 if (netif_needs_gso(skb, features)) {
2223 if (unlikely(dev_gso_segment(skb, features)))
2228 if (skb_needs_linearize(skb, features) &&
2229 __skb_linearize(skb))
2232 /* If packet is not checksummed and device does not
2233 * support checksumming for this protocol, complete
2234 * checksumming here.
2236 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2237 skb_set_transport_header(skb,
2238 skb_checksum_start_offset(skb));
2239 if (!(features & NETIF_F_ALL_CSUM) &&
2240 skb_checksum_help(skb))
2246 rc = ops->ndo_start_xmit(skb, dev);
2247 trace_net_dev_xmit(skb, rc, dev, skb_len);
2248 if (rc == NETDEV_TX_OK)
2249 txq_trans_update(txq);
2255 struct sk_buff *nskb = skb->next;
2257 skb->next = nskb->next;
2261 * If device doesn't need nskb->dst, release it right now while
2262 * its hot in this cpu cache
2264 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2267 skb_len = nskb->len;
2268 rc = ops->ndo_start_xmit(nskb, dev);
2269 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2270 if (unlikely(rc != NETDEV_TX_OK)) {
2271 if (rc & ~NETDEV_TX_MASK)
2272 goto out_kfree_gso_skb;
2273 nskb->next = skb->next;
2277 txq_trans_update(txq);
2278 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2279 return NETDEV_TX_BUSY;
2280 } while (skb->next);
2283 if (likely(skb->next == NULL))
2284 skb->destructor = DEV_GSO_CB(skb)->destructor;
2291 static u32 hashrnd __read_mostly;
2294 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2295 * to be used as a distribution range.
2297 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2298 unsigned int num_tx_queues)
2302 u16 qcount = num_tx_queues;
2304 if (skb_rx_queue_recorded(skb)) {
2305 hash = skb_get_rx_queue(skb);
2306 while (unlikely(hash >= num_tx_queues))
2307 hash -= num_tx_queues;
2312 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2313 qoffset = dev->tc_to_txq[tc].offset;
2314 qcount = dev->tc_to_txq[tc].count;
2317 if (skb->sk && skb->sk->sk_hash)
2318 hash = skb->sk->sk_hash;
2320 hash = (__force u16) skb->protocol;
2321 hash = jhash_1word(hash, hashrnd);
2323 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2325 EXPORT_SYMBOL(__skb_tx_hash);
2327 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2329 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2330 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2331 dev->name, queue_index,
2332 dev->real_num_tx_queues);
2338 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2341 struct xps_dev_maps *dev_maps;
2342 struct xps_map *map;
2343 int queue_index = -1;
2346 dev_maps = rcu_dereference(dev->xps_maps);
2348 map = rcu_dereference(
2349 dev_maps->cpu_map[raw_smp_processor_id()]);
2352 queue_index = map->queues[0];
2355 if (skb->sk && skb->sk->sk_hash)
2356 hash = skb->sk->sk_hash;
2358 hash = (__force u16) skb->protocol ^
2360 hash = jhash_1word(hash, hashrnd);
2361 queue_index = map->queues[
2362 ((u64)hash * map->len) >> 32];
2364 if (unlikely(queue_index >= dev->real_num_tx_queues))
2376 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2377 struct sk_buff *skb)
2380 const struct net_device_ops *ops = dev->netdev_ops;
2382 if (dev->real_num_tx_queues == 1)
2384 else if (ops->ndo_select_queue) {
2385 queue_index = ops->ndo_select_queue(dev, skb);
2386 queue_index = dev_cap_txqueue(dev, queue_index);
2388 struct sock *sk = skb->sk;
2389 queue_index = sk_tx_queue_get(sk);
2391 if (queue_index < 0 || skb->ooo_okay ||
2392 queue_index >= dev->real_num_tx_queues) {
2393 int old_index = queue_index;
2395 queue_index = get_xps_queue(dev, skb);
2396 if (queue_index < 0)
2397 queue_index = skb_tx_hash(dev, skb);
2399 if (queue_index != old_index && sk) {
2400 struct dst_entry *dst =
2401 rcu_dereference_check(sk->sk_dst_cache, 1);
2403 if (dst && skb_dst(skb) == dst)
2404 sk_tx_queue_set(sk, queue_index);
2409 skb_set_queue_mapping(skb, queue_index);
2410 return netdev_get_tx_queue(dev, queue_index);
2413 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2414 struct net_device *dev,
2415 struct netdev_queue *txq)
2417 spinlock_t *root_lock = qdisc_lock(q);
2421 qdisc_skb_cb(skb)->pkt_len = skb->len;
2422 qdisc_calculate_pkt_len(skb, q);
2424 * Heuristic to force contended enqueues to serialize on a
2425 * separate lock before trying to get qdisc main lock.
2426 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2427 * and dequeue packets faster.
2429 contended = qdisc_is_running(q);
2430 if (unlikely(contended))
2431 spin_lock(&q->busylock);
2433 spin_lock(root_lock);
2434 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2437 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2438 qdisc_run_begin(q)) {
2440 * This is a work-conserving queue; there are no old skbs
2441 * waiting to be sent out; and the qdisc is not running -
2442 * xmit the skb directly.
2444 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2447 qdisc_bstats_update(q, skb);
2449 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2450 if (unlikely(contended)) {
2451 spin_unlock(&q->busylock);
2458 rc = NET_XMIT_SUCCESS;
2461 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2462 if (qdisc_run_begin(q)) {
2463 if (unlikely(contended)) {
2464 spin_unlock(&q->busylock);
2470 spin_unlock(root_lock);
2471 if (unlikely(contended))
2472 spin_unlock(&q->busylock);
2476 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2477 static void skb_update_prio(struct sk_buff *skb)
2479 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2481 if (!skb->priority && skb->sk && map) {
2482 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2484 if (prioidx < map->priomap_len)
2485 skb->priority = map->priomap[prioidx];
2489 #define skb_update_prio(skb)
2492 static DEFINE_PER_CPU(int, xmit_recursion);
2493 #define RECURSION_LIMIT 10
2496 * dev_loopback_xmit - loop back @skb
2497 * @skb: buffer to transmit
2499 int dev_loopback_xmit(struct sk_buff *skb)
2501 skb_reset_mac_header(skb);
2502 __skb_pull(skb, skb_network_offset(skb));
2503 skb->pkt_type = PACKET_LOOPBACK;
2504 skb->ip_summed = CHECKSUM_UNNECESSARY;
2505 WARN_ON(!skb_dst(skb));
2510 EXPORT_SYMBOL(dev_loopback_xmit);
2513 * dev_queue_xmit - transmit a buffer
2514 * @skb: buffer to transmit
2516 * Queue a buffer for transmission to a network device. The caller must
2517 * have set the device and priority and built the buffer before calling
2518 * this function. The function can be called from an interrupt.
2520 * A negative errno code is returned on a failure. A success does not
2521 * guarantee the frame will be transmitted as it may be dropped due
2522 * to congestion or traffic shaping.
2524 * -----------------------------------------------------------------------------------
2525 * I notice this method can also return errors from the queue disciplines,
2526 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2529 * Regardless of the return value, the skb is consumed, so it is currently
2530 * difficult to retry a send to this method. (You can bump the ref count
2531 * before sending to hold a reference for retry if you are careful.)
2533 * When calling this method, interrupts MUST be enabled. This is because
2534 * the BH enable code must have IRQs enabled so that it will not deadlock.
2537 int dev_queue_xmit(struct sk_buff *skb)
2539 struct net_device *dev = skb->dev;
2540 struct netdev_queue *txq;
2544 /* Disable soft irqs for various locks below. Also
2545 * stops preemption for RCU.
2549 skb_update_prio(skb);
2551 txq = dev_pick_tx(dev, skb);
2552 q = rcu_dereference_bh(txq->qdisc);
2554 #ifdef CONFIG_NET_CLS_ACT
2555 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2557 trace_net_dev_queue(skb);
2559 rc = __dev_xmit_skb(skb, q, dev, txq);
2563 /* The device has no queue. Common case for software devices:
2564 loopback, all the sorts of tunnels...
2566 Really, it is unlikely that netif_tx_lock protection is necessary
2567 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2569 However, it is possible, that they rely on protection
2572 Check this and shot the lock. It is not prone from deadlocks.
2573 Either shot noqueue qdisc, it is even simpler 8)
2575 if (dev->flags & IFF_UP) {
2576 int cpu = smp_processor_id(); /* ok because BHs are off */
2578 if (txq->xmit_lock_owner != cpu) {
2580 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2581 goto recursion_alert;
2583 HARD_TX_LOCK(dev, txq, cpu);
2585 if (!netif_xmit_stopped(txq)) {
2586 __this_cpu_inc(xmit_recursion);
2587 rc = dev_hard_start_xmit(skb, dev, txq);
2588 __this_cpu_dec(xmit_recursion);
2589 if (dev_xmit_complete(rc)) {
2590 HARD_TX_UNLOCK(dev, txq);
2594 HARD_TX_UNLOCK(dev, txq);
2595 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2598 /* Recursion is detected! It is possible,
2602 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2608 rcu_read_unlock_bh();
2613 rcu_read_unlock_bh();
2616 EXPORT_SYMBOL(dev_queue_xmit);
2619 /*=======================================================================
2621 =======================================================================*/
2623 int netdev_max_backlog __read_mostly = 1000;
2624 int netdev_tstamp_prequeue __read_mostly = 1;
2625 int netdev_budget __read_mostly = 300;
2626 int weight_p __read_mostly = 64; /* old backlog weight */
2628 /* Called with irq disabled */
2629 static inline void ____napi_schedule(struct softnet_data *sd,
2630 struct napi_struct *napi)
2632 list_add_tail(&napi->poll_list, &sd->poll_list);
2633 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2637 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2638 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2639 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2640 * if hash is a canonical 4-tuple hash over transport ports.
2642 void __skb_get_rxhash(struct sk_buff *skb)
2644 struct flow_keys keys;
2647 if (!skb_flow_dissect(skb, &keys))
2653 /* get a consistent hash (same value on both flow directions) */
2654 if (((__force u32)keys.dst < (__force u32)keys.src) ||
2655 (((__force u32)keys.dst == (__force u32)keys.src) &&
2656 ((__force u16)keys.port16[1] < (__force u16)keys.port16[0]))) {
2657 swap(keys.dst, keys.src);
2658 swap(keys.port16[0], keys.port16[1]);
2661 hash = jhash_3words((__force u32)keys.dst,
2662 (__force u32)keys.src,
2663 (__force u32)keys.ports, hashrnd);
2669 EXPORT_SYMBOL(__skb_get_rxhash);
2673 /* One global table that all flow-based protocols share. */
2674 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2675 EXPORT_SYMBOL(rps_sock_flow_table);
2677 struct static_key rps_needed __read_mostly;
2679 static struct rps_dev_flow *
2680 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2681 struct rps_dev_flow *rflow, u16 next_cpu)
2683 if (next_cpu != RPS_NO_CPU) {
2684 #ifdef CONFIG_RFS_ACCEL
2685 struct netdev_rx_queue *rxqueue;
2686 struct rps_dev_flow_table *flow_table;
2687 struct rps_dev_flow *old_rflow;
2692 /* Should we steer this flow to a different hardware queue? */
2693 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2694 !(dev->features & NETIF_F_NTUPLE))
2696 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2697 if (rxq_index == skb_get_rx_queue(skb))
2700 rxqueue = dev->_rx + rxq_index;
2701 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2704 flow_id = skb->rxhash & flow_table->mask;
2705 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2706 rxq_index, flow_id);
2710 rflow = &flow_table->flows[flow_id];
2712 if (old_rflow->filter == rflow->filter)
2713 old_rflow->filter = RPS_NO_FILTER;
2717 per_cpu(softnet_data, next_cpu).input_queue_head;
2720 rflow->cpu = next_cpu;
2725 * get_rps_cpu is called from netif_receive_skb and returns the target
2726 * CPU from the RPS map of the receiving queue for a given skb.
2727 * rcu_read_lock must be held on entry.
2729 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2730 struct rps_dev_flow **rflowp)
2732 struct netdev_rx_queue *rxqueue;
2733 struct rps_map *map;
2734 struct rps_dev_flow_table *flow_table;
2735 struct rps_sock_flow_table *sock_flow_table;
2739 if (skb_rx_queue_recorded(skb)) {
2740 u16 index = skb_get_rx_queue(skb);
2741 if (unlikely(index >= dev->real_num_rx_queues)) {
2742 WARN_ONCE(dev->real_num_rx_queues > 1,
2743 "%s received packet on queue %u, but number "
2744 "of RX queues is %u\n",
2745 dev->name, index, dev->real_num_rx_queues);
2748 rxqueue = dev->_rx + index;
2752 map = rcu_dereference(rxqueue->rps_map);
2754 if (map->len == 1 &&
2755 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2756 tcpu = map->cpus[0];
2757 if (cpu_online(tcpu))
2761 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2765 skb_reset_network_header(skb);
2766 if (!skb_get_rxhash(skb))
2769 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2770 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2771 if (flow_table && sock_flow_table) {
2773 struct rps_dev_flow *rflow;
2775 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2778 next_cpu = sock_flow_table->ents[skb->rxhash &
2779 sock_flow_table->mask];
2782 * If the desired CPU (where last recvmsg was done) is
2783 * different from current CPU (one in the rx-queue flow
2784 * table entry), switch if one of the following holds:
2785 * - Current CPU is unset (equal to RPS_NO_CPU).
2786 * - Current CPU is offline.
2787 * - The current CPU's queue tail has advanced beyond the
2788 * last packet that was enqueued using this table entry.
2789 * This guarantees that all previous packets for the flow
2790 * have been dequeued, thus preserving in order delivery.
2792 if (unlikely(tcpu != next_cpu) &&
2793 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2794 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2795 rflow->last_qtail)) >= 0))
2796 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2798 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2806 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2808 if (cpu_online(tcpu)) {
2818 #ifdef CONFIG_RFS_ACCEL
2821 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2822 * @dev: Device on which the filter was set
2823 * @rxq_index: RX queue index
2824 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2825 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2827 * Drivers that implement ndo_rx_flow_steer() should periodically call
2828 * this function for each installed filter and remove the filters for
2829 * which it returns %true.
2831 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2832 u32 flow_id, u16 filter_id)
2834 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2835 struct rps_dev_flow_table *flow_table;
2836 struct rps_dev_flow *rflow;
2841 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2842 if (flow_table && flow_id <= flow_table->mask) {
2843 rflow = &flow_table->flows[flow_id];
2844 cpu = ACCESS_ONCE(rflow->cpu);
2845 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2846 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2847 rflow->last_qtail) <
2848 (int)(10 * flow_table->mask)))
2854 EXPORT_SYMBOL(rps_may_expire_flow);
2856 #endif /* CONFIG_RFS_ACCEL */
2858 /* Called from hardirq (IPI) context */
2859 static void rps_trigger_softirq(void *data)
2861 struct softnet_data *sd = data;
2863 ____napi_schedule(sd, &sd->backlog);
2867 #endif /* CONFIG_RPS */
2870 * Check if this softnet_data structure is another cpu one
2871 * If yes, queue it to our IPI list and return 1
2874 static int rps_ipi_queued(struct softnet_data *sd)
2877 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2880 sd->rps_ipi_next = mysd->rps_ipi_list;
2881 mysd->rps_ipi_list = sd;
2883 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2886 #endif /* CONFIG_RPS */
2891 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2892 * queue (may be a remote CPU queue).
2894 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2895 unsigned int *qtail)
2897 struct softnet_data *sd;
2898 unsigned long flags;
2900 sd = &per_cpu(softnet_data, cpu);
2902 local_irq_save(flags);
2905 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2906 if (skb_queue_len(&sd->input_pkt_queue)) {
2908 __skb_queue_tail(&sd->input_pkt_queue, skb);
2909 input_queue_tail_incr_save(sd, qtail);
2911 local_irq_restore(flags);
2912 return NET_RX_SUCCESS;
2915 /* Schedule NAPI for backlog device
2916 * We can use non atomic operation since we own the queue lock
2918 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2919 if (!rps_ipi_queued(sd))
2920 ____napi_schedule(sd, &sd->backlog);
2928 local_irq_restore(flags);
2930 atomic_long_inc(&skb->dev->rx_dropped);
2936 * netif_rx - post buffer to the network code
2937 * @skb: buffer to post
2939 * This function receives a packet from a device driver and queues it for
2940 * the upper (protocol) levels to process. It always succeeds. The buffer
2941 * may be dropped during processing for congestion control or by the
2945 * NET_RX_SUCCESS (no congestion)
2946 * NET_RX_DROP (packet was dropped)
2950 int netif_rx(struct sk_buff *skb)
2954 /* if netpoll wants it, pretend we never saw it */
2955 if (netpoll_rx(skb))
2958 net_timestamp_check(netdev_tstamp_prequeue, skb);
2960 trace_netif_rx(skb);
2962 if (static_key_false(&rps_needed)) {
2963 struct rps_dev_flow voidflow, *rflow = &voidflow;
2969 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2971 cpu = smp_processor_id();
2973 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2981 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2986 EXPORT_SYMBOL(netif_rx);
2988 int netif_rx_ni(struct sk_buff *skb)
2993 err = netif_rx(skb);
2994 if (local_softirq_pending())
3000 EXPORT_SYMBOL(netif_rx_ni);
3002 static void net_tx_action(struct softirq_action *h)
3004 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3006 if (sd->completion_queue) {
3007 struct sk_buff *clist;
3009 local_irq_disable();
3010 clist = sd->completion_queue;
3011 sd->completion_queue = NULL;
3015 struct sk_buff *skb = clist;
3016 clist = clist->next;
3018 WARN_ON(atomic_read(&skb->users));
3019 trace_kfree_skb(skb, net_tx_action);
3024 if (sd->output_queue) {
3027 local_irq_disable();
3028 head = sd->output_queue;
3029 sd->output_queue = NULL;
3030 sd->output_queue_tailp = &sd->output_queue;
3034 struct Qdisc *q = head;
3035 spinlock_t *root_lock;
3037 head = head->next_sched;
3039 root_lock = qdisc_lock(q);
3040 if (spin_trylock(root_lock)) {
3041 smp_mb__before_clear_bit();
3042 clear_bit(__QDISC_STATE_SCHED,
3045 spin_unlock(root_lock);
3047 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3049 __netif_reschedule(q);
3051 smp_mb__before_clear_bit();
3052 clear_bit(__QDISC_STATE_SCHED,
3060 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3061 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3062 /* This hook is defined here for ATM LANE */
3063 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3064 unsigned char *addr) __read_mostly;
3065 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3068 #ifdef CONFIG_NET_CLS_ACT
3069 /* TODO: Maybe we should just force sch_ingress to be compiled in
3070 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3071 * a compare and 2 stores extra right now if we dont have it on
3072 * but have CONFIG_NET_CLS_ACT
3073 * NOTE: This doesn't stop any functionality; if you dont have
3074 * the ingress scheduler, you just can't add policies on ingress.
3077 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3079 struct net_device *dev = skb->dev;
3080 u32 ttl = G_TC_RTTL(skb->tc_verd);
3081 int result = TC_ACT_OK;
3084 if (unlikely(MAX_RED_LOOP < ttl++)) {
3085 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3086 skb->skb_iif, dev->ifindex);
3090 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3091 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3094 if (q != &noop_qdisc) {
3095 spin_lock(qdisc_lock(q));
3096 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3097 result = qdisc_enqueue_root(skb, q);
3098 spin_unlock(qdisc_lock(q));
3104 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3105 struct packet_type **pt_prev,
3106 int *ret, struct net_device *orig_dev)
3108 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3110 if (!rxq || rxq->qdisc == &noop_qdisc)
3114 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3118 switch (ing_filter(skb, rxq)) {
3132 * netdev_rx_handler_register - register receive handler
3133 * @dev: device to register a handler for
3134 * @rx_handler: receive handler to register
3135 * @rx_handler_data: data pointer that is used by rx handler
3137 * Register a receive hander for a device. This handler will then be
3138 * called from __netif_receive_skb. A negative errno code is returned
3141 * The caller must hold the rtnl_mutex.
3143 * For a general description of rx_handler, see enum rx_handler_result.
3145 int netdev_rx_handler_register(struct net_device *dev,
3146 rx_handler_func_t *rx_handler,
3147 void *rx_handler_data)
3151 if (dev->rx_handler)
3154 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3155 rcu_assign_pointer(dev->rx_handler, rx_handler);
3159 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3162 * netdev_rx_handler_unregister - unregister receive handler
3163 * @dev: device to unregister a handler from
3165 * Unregister a receive hander from a device.
3167 * The caller must hold the rtnl_mutex.
3169 void netdev_rx_handler_unregister(struct net_device *dev)
3173 RCU_INIT_POINTER(dev->rx_handler, NULL);
3174 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3176 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3179 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3180 * the special handling of PFMEMALLOC skbs.
3182 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3184 switch (skb->protocol) {
3185 case __constant_htons(ETH_P_ARP):
3186 case __constant_htons(ETH_P_IP):
3187 case __constant_htons(ETH_P_IPV6):
3188 case __constant_htons(ETH_P_8021Q):
3195 static int __netif_receive_skb(struct sk_buff *skb)
3197 struct packet_type *ptype, *pt_prev;
3198 rx_handler_func_t *rx_handler;
3199 struct net_device *orig_dev;
3200 struct net_device *null_or_dev;
3201 bool deliver_exact = false;
3202 int ret = NET_RX_DROP;
3204 unsigned long pflags = current->flags;
3206 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3208 trace_netif_receive_skb(skb);
3211 * PFMEMALLOC skbs are special, they should
3212 * - be delivered to SOCK_MEMALLOC sockets only
3213 * - stay away from userspace
3214 * - have bounded memory usage
3216 * Use PF_MEMALLOC as this saves us from propagating the allocation
3217 * context down to all allocation sites.
3219 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3220 current->flags |= PF_MEMALLOC;
3222 /* if we've gotten here through NAPI, check netpoll */
3223 if (netpoll_receive_skb(skb))
3226 orig_dev = skb->dev;
3228 skb_reset_network_header(skb);
3229 skb_reset_transport_header(skb);
3230 skb_reset_mac_len(skb);
3237 skb->skb_iif = skb->dev->ifindex;
3239 __this_cpu_inc(softnet_data.processed);
3241 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3242 skb = vlan_untag(skb);
3247 #ifdef CONFIG_NET_CLS_ACT
3248 if (skb->tc_verd & TC_NCLS) {
3249 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3254 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
3257 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3258 if (!ptype->dev || ptype->dev == skb->dev) {
3260 ret = deliver_skb(skb, pt_prev, orig_dev);
3266 #ifdef CONFIG_NET_CLS_ACT
3267 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3273 if (sk_memalloc_socks() && skb_pfmemalloc(skb)
3274 && !skb_pfmemalloc_protocol(skb))
3277 rx_handler = rcu_dereference(skb->dev->rx_handler);
3278 if (vlan_tx_tag_present(skb)) {
3280 ret = deliver_skb(skb, pt_prev, orig_dev);
3283 if (vlan_do_receive(&skb, !rx_handler))
3285 else if (unlikely(!skb))
3291 ret = deliver_skb(skb, pt_prev, orig_dev);
3294 switch (rx_handler(&skb)) {
3295 case RX_HANDLER_CONSUMED:
3297 case RX_HANDLER_ANOTHER:
3299 case RX_HANDLER_EXACT:
3300 deliver_exact = true;
3301 case RX_HANDLER_PASS:
3308 /* deliver only exact match when indicated */
3309 null_or_dev = deliver_exact ? skb->dev : NULL;
3311 type = skb->protocol;
3312 list_for_each_entry_rcu(ptype,
3313 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3314 if (ptype->type == type &&
3315 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3316 ptype->dev == orig_dev)) {
3318 ret = deliver_skb(skb, pt_prev, orig_dev);
3324 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3327 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3330 atomic_long_inc(&skb->dev->rx_dropped);
3332 /* Jamal, now you will not able to escape explaining
3333 * me how you were going to use this. :-)
3341 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3346 * netif_receive_skb - process receive buffer from network
3347 * @skb: buffer to process
3349 * netif_receive_skb() is the main receive data processing function.
3350 * It always succeeds. The buffer may be dropped during processing
3351 * for congestion control or by the protocol layers.
3353 * This function may only be called from softirq context and interrupts
3354 * should be enabled.
3356 * Return values (usually ignored):
3357 * NET_RX_SUCCESS: no congestion
3358 * NET_RX_DROP: packet was dropped
3360 int netif_receive_skb(struct sk_buff *skb)
3362 net_timestamp_check(netdev_tstamp_prequeue, skb);
3364 if (skb_defer_rx_timestamp(skb))
3365 return NET_RX_SUCCESS;
3368 if (static_key_false(&rps_needed)) {
3369 struct rps_dev_flow voidflow, *rflow = &voidflow;
3374 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3377 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3384 return __netif_receive_skb(skb);
3386 EXPORT_SYMBOL(netif_receive_skb);
3388 /* Network device is going away, flush any packets still pending
3389 * Called with irqs disabled.
3391 static void flush_backlog(void *arg)
3393 struct net_device *dev = arg;
3394 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3395 struct sk_buff *skb, *tmp;
3398 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3399 if (skb->dev == dev) {
3400 __skb_unlink(skb, &sd->input_pkt_queue);
3402 input_queue_head_incr(sd);
3407 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3408 if (skb->dev == dev) {
3409 __skb_unlink(skb, &sd->process_queue);
3411 input_queue_head_incr(sd);
3416 static int napi_gro_complete(struct sk_buff *skb)
3418 struct packet_type *ptype;
3419 __be16 type = skb->protocol;
3420 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3423 if (NAPI_GRO_CB(skb)->count == 1) {
3424 skb_shinfo(skb)->gso_size = 0;
3429 list_for_each_entry_rcu(ptype, head, list) {
3430 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3433 err = ptype->gro_complete(skb);
3439 WARN_ON(&ptype->list == head);
3441 return NET_RX_SUCCESS;
3445 return netif_receive_skb(skb);
3448 inline void napi_gro_flush(struct napi_struct *napi)
3450 struct sk_buff *skb, *next;
3452 for (skb = napi->gro_list; skb; skb = next) {
3455 napi_gro_complete(skb);
3458 napi->gro_count = 0;
3459 napi->gro_list = NULL;
3461 EXPORT_SYMBOL(napi_gro_flush);
3463 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3465 struct sk_buff **pp = NULL;
3466 struct packet_type *ptype;
3467 __be16 type = skb->protocol;
3468 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3471 enum gro_result ret;
3473 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3476 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3480 list_for_each_entry_rcu(ptype, head, list) {
3481 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3484 skb_set_network_header(skb, skb_gro_offset(skb));
3485 mac_len = skb->network_header - skb->mac_header;
3486 skb->mac_len = mac_len;
3487 NAPI_GRO_CB(skb)->same_flow = 0;
3488 NAPI_GRO_CB(skb)->flush = 0;
3489 NAPI_GRO_CB(skb)->free = 0;
3491 pp = ptype->gro_receive(&napi->gro_list, skb);
3496 if (&ptype->list == head)
3499 same_flow = NAPI_GRO_CB(skb)->same_flow;
3500 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3503 struct sk_buff *nskb = *pp;
3507 napi_gro_complete(nskb);
3514 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3518 NAPI_GRO_CB(skb)->count = 1;
3519 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3520 skb->next = napi->gro_list;
3521 napi->gro_list = skb;
3525 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3526 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3528 BUG_ON(skb->end - skb->tail < grow);
3530 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3533 skb->data_len -= grow;
3535 skb_shinfo(skb)->frags[0].page_offset += grow;
3536 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3538 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3539 skb_frag_unref(skb, 0);
3540 memmove(skb_shinfo(skb)->frags,
3541 skb_shinfo(skb)->frags + 1,
3542 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3553 EXPORT_SYMBOL(dev_gro_receive);
3555 static inline gro_result_t
3556 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3559 unsigned int maclen = skb->dev->hard_header_len;
3561 for (p = napi->gro_list; p; p = p->next) {
3562 unsigned long diffs;
3564 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3565 diffs |= p->vlan_tci ^ skb->vlan_tci;
3566 if (maclen == ETH_HLEN)
3567 diffs |= compare_ether_header(skb_mac_header(p),
3568 skb_gro_mac_header(skb));
3570 diffs = memcmp(skb_mac_header(p),
3571 skb_gro_mac_header(skb),
3573 NAPI_GRO_CB(p)->same_flow = !diffs;
3574 NAPI_GRO_CB(p)->flush = 0;
3577 return dev_gro_receive(napi, skb);
3580 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3584 if (netif_receive_skb(skb))
3592 case GRO_MERGED_FREE:
3593 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3594 kmem_cache_free(skbuff_head_cache, skb);
3606 EXPORT_SYMBOL(napi_skb_finish);
3608 void skb_gro_reset_offset(struct sk_buff *skb)
3610 NAPI_GRO_CB(skb)->data_offset = 0;
3611 NAPI_GRO_CB(skb)->frag0 = NULL;
3612 NAPI_GRO_CB(skb)->frag0_len = 0;
3614 if (skb->mac_header == skb->tail &&
3615 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3616 NAPI_GRO_CB(skb)->frag0 =
3617 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3618 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3621 EXPORT_SYMBOL(skb_gro_reset_offset);
3623 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3625 skb_gro_reset_offset(skb);
3627 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3629 EXPORT_SYMBOL(napi_gro_receive);
3631 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3633 __skb_pull(skb, skb_headlen(skb));
3634 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3635 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3637 skb->dev = napi->dev;
3643 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3645 struct sk_buff *skb = napi->skb;
3648 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3654 EXPORT_SYMBOL(napi_get_frags);
3656 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3662 skb->protocol = eth_type_trans(skb, skb->dev);
3664 if (ret == GRO_HELD)
3665 skb_gro_pull(skb, -ETH_HLEN);
3666 else if (netif_receive_skb(skb))
3671 case GRO_MERGED_FREE:
3672 napi_reuse_skb(napi, skb);
3681 EXPORT_SYMBOL(napi_frags_finish);
3683 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3685 struct sk_buff *skb = napi->skb;
3692 skb_reset_mac_header(skb);
3693 skb_gro_reset_offset(skb);
3695 off = skb_gro_offset(skb);
3696 hlen = off + sizeof(*eth);
3697 eth = skb_gro_header_fast(skb, off);
3698 if (skb_gro_header_hard(skb, hlen)) {
3699 eth = skb_gro_header_slow(skb, hlen, off);
3700 if (unlikely(!eth)) {
3701 napi_reuse_skb(napi, skb);
3707 skb_gro_pull(skb, sizeof(*eth));
3710 * This works because the only protocols we care about don't require
3711 * special handling. We'll fix it up properly at the end.
3713 skb->protocol = eth->h_proto;
3719 gro_result_t napi_gro_frags(struct napi_struct *napi)
3721 struct sk_buff *skb = napi_frags_skb(napi);
3726 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3728 EXPORT_SYMBOL(napi_gro_frags);
3731 * net_rps_action sends any pending IPI's for rps.
3732 * Note: called with local irq disabled, but exits with local irq enabled.
3734 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3737 struct softnet_data *remsd = sd->rps_ipi_list;
3740 sd->rps_ipi_list = NULL;
3744 /* Send pending IPI's to kick RPS processing on remote cpus. */
3746 struct softnet_data *next = remsd->rps_ipi_next;
3748 if (cpu_online(remsd->cpu))
3749 __smp_call_function_single(remsd->cpu,
3758 static int process_backlog(struct napi_struct *napi, int quota)
3761 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3764 /* Check if we have pending ipi, its better to send them now,
3765 * not waiting net_rx_action() end.
3767 if (sd->rps_ipi_list) {
3768 local_irq_disable();
3769 net_rps_action_and_irq_enable(sd);
3772 napi->weight = weight_p;
3773 local_irq_disable();
3774 while (work < quota) {
3775 struct sk_buff *skb;
3778 while ((skb = __skb_dequeue(&sd->process_queue))) {
3780 __netif_receive_skb(skb);
3781 local_irq_disable();
3782 input_queue_head_incr(sd);
3783 if (++work >= quota) {
3790 qlen = skb_queue_len(&sd->input_pkt_queue);
3792 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3793 &sd->process_queue);
3795 if (qlen < quota - work) {
3797 * Inline a custom version of __napi_complete().
3798 * only current cpu owns and manipulates this napi,
3799 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3800 * we can use a plain write instead of clear_bit(),
3801 * and we dont need an smp_mb() memory barrier.
3803 list_del(&napi->poll_list);
3806 quota = work + qlen;
3816 * __napi_schedule - schedule for receive
3817 * @n: entry to schedule
3819 * The entry's receive function will be scheduled to run
3821 void __napi_schedule(struct napi_struct *n)
3823 unsigned long flags;
3825 local_irq_save(flags);
3826 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3827 local_irq_restore(flags);
3829 EXPORT_SYMBOL(__napi_schedule);
3831 void __napi_complete(struct napi_struct *n)
3833 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3834 BUG_ON(n->gro_list);
3836 list_del(&n->poll_list);
3837 smp_mb__before_clear_bit();
3838 clear_bit(NAPI_STATE_SCHED, &n->state);
3840 EXPORT_SYMBOL(__napi_complete);
3842 void napi_complete(struct napi_struct *n)
3844 unsigned long flags;
3847 * don't let napi dequeue from the cpu poll list
3848 * just in case its running on a different cpu
3850 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3854 local_irq_save(flags);
3856 local_irq_restore(flags);
3858 EXPORT_SYMBOL(napi_complete);
3860 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3861 int (*poll)(struct napi_struct *, int), int weight)
3863 INIT_LIST_HEAD(&napi->poll_list);
3864 napi->gro_count = 0;
3865 napi->gro_list = NULL;
3868 napi->weight = weight;
3869 list_add(&napi->dev_list, &dev->napi_list);
3871 #ifdef CONFIG_NETPOLL
3872 spin_lock_init(&napi->poll_lock);
3873 napi->poll_owner = -1;
3875 set_bit(NAPI_STATE_SCHED, &napi->state);
3877 EXPORT_SYMBOL(netif_napi_add);
3879 void netif_napi_del(struct napi_struct *napi)
3881 struct sk_buff *skb, *next;
3883 list_del_init(&napi->dev_list);
3884 napi_free_frags(napi);
3886 for (skb = napi->gro_list; skb; skb = next) {
3892 napi->gro_list = NULL;
3893 napi->gro_count = 0;
3895 EXPORT_SYMBOL(netif_napi_del);
3897 static void net_rx_action(struct softirq_action *h)
3899 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3900 unsigned long time_limit = jiffies + 2;
3901 int budget = netdev_budget;
3904 local_irq_disable();
3906 while (!list_empty(&sd->poll_list)) {
3907 struct napi_struct *n;
3910 /* If softirq window is exhuasted then punt.
3911 * Allow this to run for 2 jiffies since which will allow
3912 * an average latency of 1.5/HZ.
3914 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3919 /* Even though interrupts have been re-enabled, this
3920 * access is safe because interrupts can only add new
3921 * entries to the tail of this list, and only ->poll()
3922 * calls can remove this head entry from the list.
3924 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3926 have = netpoll_poll_lock(n);
3930 /* This NAPI_STATE_SCHED test is for avoiding a race
3931 * with netpoll's poll_napi(). Only the entity which
3932 * obtains the lock and sees NAPI_STATE_SCHED set will
3933 * actually make the ->poll() call. Therefore we avoid
3934 * accidentally calling ->poll() when NAPI is not scheduled.
3937 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3938 work = n->poll(n, weight);
3942 WARN_ON_ONCE(work > weight);
3946 local_irq_disable();
3948 /* Drivers must not modify the NAPI state if they
3949 * consume the entire weight. In such cases this code
3950 * still "owns" the NAPI instance and therefore can
3951 * move the instance around on the list at-will.
3953 if (unlikely(work == weight)) {
3954 if (unlikely(napi_disable_pending(n))) {
3957 local_irq_disable();
3959 list_move_tail(&n->poll_list, &sd->poll_list);
3962 netpoll_poll_unlock(have);
3965 net_rps_action_and_irq_enable(sd);
3967 #ifdef CONFIG_NET_DMA
3969 * There may not be any more sk_buffs coming right now, so push
3970 * any pending DMA copies to hardware
3972 dma_issue_pending_all();
3979 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3983 static gifconf_func_t *gifconf_list[NPROTO];
3986 * register_gifconf - register a SIOCGIF handler
3987 * @family: Address family
3988 * @gifconf: Function handler
3990 * Register protocol dependent address dumping routines. The handler
3991 * that is passed must not be freed or reused until it has been replaced
3992 * by another handler.
3994 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3996 if (family >= NPROTO)
3998 gifconf_list[family] = gifconf;
4001 EXPORT_SYMBOL(register_gifconf);
4005 * Map an interface index to its name (SIOCGIFNAME)
4009 * We need this ioctl for efficient implementation of the
4010 * if_indextoname() function required by the IPv6 API. Without
4011 * it, we would have to search all the interfaces to find a
4015 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4017 struct net_device *dev;
4021 * Fetch the caller's info block.
4024 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4028 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4034 strcpy(ifr.ifr_name, dev->name);
4037 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4043 * Perform a SIOCGIFCONF call. This structure will change
4044 * size eventually, and there is nothing I can do about it.
4045 * Thus we will need a 'compatibility mode'.
4048 static int dev_ifconf(struct net *net, char __user *arg)
4051 struct net_device *dev;
4058 * Fetch the caller's info block.
4061 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4068 * Loop over the interfaces, and write an info block for each.
4072 for_each_netdev(net, dev) {
4073 for (i = 0; i < NPROTO; i++) {
4074 if (gifconf_list[i]) {
4077 done = gifconf_list[i](dev, NULL, 0);
4079 done = gifconf_list[i](dev, pos + total,
4089 * All done. Write the updated control block back to the caller.
4091 ifc.ifc_len = total;
4094 * Both BSD and Solaris return 0 here, so we do too.
4096 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4099 #ifdef CONFIG_PROC_FS
4101 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4103 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4104 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4105 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4107 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4109 struct net *net = seq_file_net(seq);
4110 struct net_device *dev;
4111 struct hlist_node *p;
4112 struct hlist_head *h;
4113 unsigned int count = 0, offset = get_offset(*pos);
4115 h = &net->dev_name_head[get_bucket(*pos)];
4116 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4117 if (++count == offset)
4124 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4126 struct net_device *dev;
4127 unsigned int bucket;
4130 dev = dev_from_same_bucket(seq, pos);
4134 bucket = get_bucket(*pos) + 1;
4135 *pos = set_bucket_offset(bucket, 1);
4136 } while (bucket < NETDEV_HASHENTRIES);
4142 * This is invoked by the /proc filesystem handler to display a device
4145 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4150 return SEQ_START_TOKEN;
4152 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4155 return dev_from_bucket(seq, pos);
4158 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4161 return dev_from_bucket(seq, pos);
4164 void dev_seq_stop(struct seq_file *seq, void *v)
4170 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4172 struct rtnl_link_stats64 temp;
4173 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4175 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4176 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4177 dev->name, stats->rx_bytes, stats->rx_packets,
4179 stats->rx_dropped + stats->rx_missed_errors,
4180 stats->rx_fifo_errors,
4181 stats->rx_length_errors + stats->rx_over_errors +
4182 stats->rx_crc_errors + stats->rx_frame_errors,
4183 stats->rx_compressed, stats->multicast,
4184 stats->tx_bytes, stats->tx_packets,
4185 stats->tx_errors, stats->tx_dropped,
4186 stats->tx_fifo_errors, stats->collisions,
4187 stats->tx_carrier_errors +
4188 stats->tx_aborted_errors +
4189 stats->tx_window_errors +
4190 stats->tx_heartbeat_errors,
4191 stats->tx_compressed);
4195 * Called from the PROCfs module. This now uses the new arbitrary sized
4196 * /proc/net interface to create /proc/net/dev
4198 static int dev_seq_show(struct seq_file *seq, void *v)
4200 if (v == SEQ_START_TOKEN)
4201 seq_puts(seq, "Inter-| Receive "
4203 " face |bytes packets errs drop fifo frame "
4204 "compressed multicast|bytes packets errs "
4205 "drop fifo colls carrier compressed\n");
4207 dev_seq_printf_stats(seq, v);
4211 static struct softnet_data *softnet_get_online(loff_t *pos)
4213 struct softnet_data *sd = NULL;
4215 while (*pos < nr_cpu_ids)
4216 if (cpu_online(*pos)) {
4217 sd = &per_cpu(softnet_data, *pos);
4224 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4226 return softnet_get_online(pos);
4229 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4232 return softnet_get_online(pos);
4235 static void softnet_seq_stop(struct seq_file *seq, void *v)
4239 static int softnet_seq_show(struct seq_file *seq, void *v)
4241 struct softnet_data *sd = v;
4243 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4244 sd->processed, sd->dropped, sd->time_squeeze, 0,
4245 0, 0, 0, 0, /* was fastroute */
4246 sd->cpu_collision, sd->received_rps);
4250 static const struct seq_operations dev_seq_ops = {
4251 .start = dev_seq_start,
4252 .next = dev_seq_next,
4253 .stop = dev_seq_stop,
4254 .show = dev_seq_show,
4257 static int dev_seq_open(struct inode *inode, struct file *file)
4259 return seq_open_net(inode, file, &dev_seq_ops,
4260 sizeof(struct seq_net_private));
4263 static const struct file_operations dev_seq_fops = {
4264 .owner = THIS_MODULE,
4265 .open = dev_seq_open,
4267 .llseek = seq_lseek,
4268 .release = seq_release_net,
4271 static const struct seq_operations softnet_seq_ops = {
4272 .start = softnet_seq_start,
4273 .next = softnet_seq_next,
4274 .stop = softnet_seq_stop,
4275 .show = softnet_seq_show,
4278 static int softnet_seq_open(struct inode *inode, struct file *file)
4280 return seq_open(file, &softnet_seq_ops);
4283 static const struct file_operations softnet_seq_fops = {
4284 .owner = THIS_MODULE,
4285 .open = softnet_seq_open,
4287 .llseek = seq_lseek,
4288 .release = seq_release,
4291 static void *ptype_get_idx(loff_t pos)
4293 struct packet_type *pt = NULL;
4297 list_for_each_entry_rcu(pt, &ptype_all, list) {
4303 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4304 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4313 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4317 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4320 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4322 struct packet_type *pt;
4323 struct list_head *nxt;
4327 if (v == SEQ_START_TOKEN)
4328 return ptype_get_idx(0);
4331 nxt = pt->list.next;
4332 if (pt->type == htons(ETH_P_ALL)) {
4333 if (nxt != &ptype_all)
4336 nxt = ptype_base[0].next;
4338 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4340 while (nxt == &ptype_base[hash]) {
4341 if (++hash >= PTYPE_HASH_SIZE)
4343 nxt = ptype_base[hash].next;
4346 return list_entry(nxt, struct packet_type, list);
4349 static void ptype_seq_stop(struct seq_file *seq, void *v)
4355 static int ptype_seq_show(struct seq_file *seq, void *v)
4357 struct packet_type *pt = v;
4359 if (v == SEQ_START_TOKEN)
4360 seq_puts(seq, "Type Device Function\n");
4361 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4362 if (pt->type == htons(ETH_P_ALL))
4363 seq_puts(seq, "ALL ");
4365 seq_printf(seq, "%04x", ntohs(pt->type));
4367 seq_printf(seq, " %-8s %pF\n",
4368 pt->dev ? pt->dev->name : "", pt->func);
4374 static const struct seq_operations ptype_seq_ops = {
4375 .start = ptype_seq_start,
4376 .next = ptype_seq_next,
4377 .stop = ptype_seq_stop,
4378 .show = ptype_seq_show,
4381 static int ptype_seq_open(struct inode *inode, struct file *file)
4383 return seq_open_net(inode, file, &ptype_seq_ops,
4384 sizeof(struct seq_net_private));
4387 static const struct file_operations ptype_seq_fops = {
4388 .owner = THIS_MODULE,
4389 .open = ptype_seq_open,
4391 .llseek = seq_lseek,
4392 .release = seq_release_net,
4396 static int __net_init dev_proc_net_init(struct net *net)
4400 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4402 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4404 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4407 if (wext_proc_init(net))
4413 proc_net_remove(net, "ptype");
4415 proc_net_remove(net, "softnet_stat");
4417 proc_net_remove(net, "dev");
4421 static void __net_exit dev_proc_net_exit(struct net *net)
4423 wext_proc_exit(net);
4425 proc_net_remove(net, "ptype");
4426 proc_net_remove(net, "softnet_stat");
4427 proc_net_remove(net, "dev");
4430 static struct pernet_operations __net_initdata dev_proc_ops = {
4431 .init = dev_proc_net_init,
4432 .exit = dev_proc_net_exit,
4435 static int __init dev_proc_init(void)
4437 return register_pernet_subsys(&dev_proc_ops);
4440 #define dev_proc_init() 0
4441 #endif /* CONFIG_PROC_FS */
4445 * netdev_set_master - set up master pointer
4446 * @slave: slave device
4447 * @master: new master device
4449 * Changes the master device of the slave. Pass %NULL to break the
4450 * bonding. The caller must hold the RTNL semaphore. On a failure
4451 * a negative errno code is returned. On success the reference counts
4452 * are adjusted and the function returns zero.
4454 int netdev_set_master(struct net_device *slave, struct net_device *master)
4456 struct net_device *old = slave->master;
4466 slave->master = master;
4472 EXPORT_SYMBOL(netdev_set_master);
4475 * netdev_set_bond_master - set up bonding master/slave pair
4476 * @slave: slave device
4477 * @master: new master device
4479 * Changes the master device of the slave. Pass %NULL to break the
4480 * bonding. The caller must hold the RTNL semaphore. On a failure
4481 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4482 * to the routing socket and the function returns zero.
4484 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4490 err = netdev_set_master(slave, master);
4494 slave->flags |= IFF_SLAVE;
4496 slave->flags &= ~IFF_SLAVE;
4498 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4501 EXPORT_SYMBOL(netdev_set_bond_master);
4503 static void dev_change_rx_flags(struct net_device *dev, int flags)
4505 const struct net_device_ops *ops = dev->netdev_ops;
4507 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4508 ops->ndo_change_rx_flags(dev, flags);
4511 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4513 unsigned int old_flags = dev->flags;
4519 dev->flags |= IFF_PROMISC;
4520 dev->promiscuity += inc;
4521 if (dev->promiscuity == 0) {
4524 * If inc causes overflow, untouch promisc and return error.
4527 dev->flags &= ~IFF_PROMISC;
4529 dev->promiscuity -= inc;
4530 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4535 if (dev->flags != old_flags) {
4536 pr_info("device %s %s promiscuous mode\n",
4538 dev->flags & IFF_PROMISC ? "entered" : "left");
4539 if (audit_enabled) {
4540 current_uid_gid(&uid, &gid);
4541 audit_log(current->audit_context, GFP_ATOMIC,
4542 AUDIT_ANOM_PROMISCUOUS,
4543 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4544 dev->name, (dev->flags & IFF_PROMISC),
4545 (old_flags & IFF_PROMISC),
4546 audit_get_loginuid(current),
4548 audit_get_sessionid(current));
4551 dev_change_rx_flags(dev, IFF_PROMISC);
4557 * dev_set_promiscuity - update promiscuity count on a device
4561 * Add or remove promiscuity from a device. While the count in the device
4562 * remains above zero the interface remains promiscuous. Once it hits zero
4563 * the device reverts back to normal filtering operation. A negative inc
4564 * value is used to drop promiscuity on the device.
4565 * Return 0 if successful or a negative errno code on error.
4567 int dev_set_promiscuity(struct net_device *dev, int inc)
4569 unsigned int old_flags = dev->flags;
4572 err = __dev_set_promiscuity(dev, inc);
4575 if (dev->flags != old_flags)
4576 dev_set_rx_mode(dev);
4579 EXPORT_SYMBOL(dev_set_promiscuity);
4582 * dev_set_allmulti - update allmulti count on a device
4586 * Add or remove reception of all multicast frames to a device. While the
4587 * count in the device remains above zero the interface remains listening
4588 * to all interfaces. Once it hits zero the device reverts back to normal
4589 * filtering operation. A negative @inc value is used to drop the counter
4590 * when releasing a resource needing all multicasts.
4591 * Return 0 if successful or a negative errno code on error.
4594 int dev_set_allmulti(struct net_device *dev, int inc)
4596 unsigned int old_flags = dev->flags;
4600 dev->flags |= IFF_ALLMULTI;
4601 dev->allmulti += inc;
4602 if (dev->allmulti == 0) {
4605 * If inc causes overflow, untouch allmulti and return error.
4608 dev->flags &= ~IFF_ALLMULTI;
4610 dev->allmulti -= inc;
4611 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4616 if (dev->flags ^ old_flags) {
4617 dev_change_rx_flags(dev, IFF_ALLMULTI);
4618 dev_set_rx_mode(dev);
4622 EXPORT_SYMBOL(dev_set_allmulti);
4625 * Upload unicast and multicast address lists to device and
4626 * configure RX filtering. When the device doesn't support unicast
4627 * filtering it is put in promiscuous mode while unicast addresses
4630 void __dev_set_rx_mode(struct net_device *dev)
4632 const struct net_device_ops *ops = dev->netdev_ops;
4634 /* dev_open will call this function so the list will stay sane. */
4635 if (!(dev->flags&IFF_UP))
4638 if (!netif_device_present(dev))
4641 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4642 /* Unicast addresses changes may only happen under the rtnl,
4643 * therefore calling __dev_set_promiscuity here is safe.
4645 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4646 __dev_set_promiscuity(dev, 1);
4647 dev->uc_promisc = true;
4648 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4649 __dev_set_promiscuity(dev, -1);
4650 dev->uc_promisc = false;
4654 if (ops->ndo_set_rx_mode)
4655 ops->ndo_set_rx_mode(dev);
4658 void dev_set_rx_mode(struct net_device *dev)
4660 netif_addr_lock_bh(dev);
4661 __dev_set_rx_mode(dev);
4662 netif_addr_unlock_bh(dev);
4666 * dev_get_flags - get flags reported to userspace
4669 * Get the combination of flag bits exported through APIs to userspace.
4671 unsigned int dev_get_flags(const struct net_device *dev)
4675 flags = (dev->flags & ~(IFF_PROMISC |
4680 (dev->gflags & (IFF_PROMISC |
4683 if (netif_running(dev)) {
4684 if (netif_oper_up(dev))
4685 flags |= IFF_RUNNING;
4686 if (netif_carrier_ok(dev))
4687 flags |= IFF_LOWER_UP;
4688 if (netif_dormant(dev))
4689 flags |= IFF_DORMANT;
4694 EXPORT_SYMBOL(dev_get_flags);
4696 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4698 unsigned int old_flags = dev->flags;
4704 * Set the flags on our device.
4707 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4708 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4710 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4714 * Load in the correct multicast list now the flags have changed.
4717 if ((old_flags ^ flags) & IFF_MULTICAST)
4718 dev_change_rx_flags(dev, IFF_MULTICAST);
4720 dev_set_rx_mode(dev);
4723 * Have we downed the interface. We handle IFF_UP ourselves
4724 * according to user attempts to set it, rather than blindly
4729 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4730 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4733 dev_set_rx_mode(dev);
4736 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4737 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4739 dev->gflags ^= IFF_PROMISC;
4740 dev_set_promiscuity(dev, inc);
4743 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4744 is important. Some (broken) drivers set IFF_PROMISC, when
4745 IFF_ALLMULTI is requested not asking us and not reporting.
4747 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4748 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4750 dev->gflags ^= IFF_ALLMULTI;
4751 dev_set_allmulti(dev, inc);
4757 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4759 unsigned int changes = dev->flags ^ old_flags;
4761 if (changes & IFF_UP) {
4762 if (dev->flags & IFF_UP)
4763 call_netdevice_notifiers(NETDEV_UP, dev);
4765 call_netdevice_notifiers(NETDEV_DOWN, dev);
4768 if (dev->flags & IFF_UP &&
4769 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4770 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4774 * dev_change_flags - change device settings
4776 * @flags: device state flags
4778 * Change settings on device based state flags. The flags are
4779 * in the userspace exported format.
4781 int dev_change_flags(struct net_device *dev, unsigned int flags)
4784 unsigned int changes, old_flags = dev->flags;
4786 ret = __dev_change_flags(dev, flags);
4790 changes = old_flags ^ dev->flags;
4792 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4794 __dev_notify_flags(dev, old_flags);
4797 EXPORT_SYMBOL(dev_change_flags);
4800 * dev_set_mtu - Change maximum transfer unit
4802 * @new_mtu: new transfer unit
4804 * Change the maximum transfer size of the network device.
4806 int dev_set_mtu(struct net_device *dev, int new_mtu)
4808 const struct net_device_ops *ops = dev->netdev_ops;
4811 if (new_mtu == dev->mtu)
4814 /* MTU must be positive. */
4818 if (!netif_device_present(dev))
4822 if (ops->ndo_change_mtu)
4823 err = ops->ndo_change_mtu(dev, new_mtu);
4827 if (!err && dev->flags & IFF_UP)
4828 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4831 EXPORT_SYMBOL(dev_set_mtu);
4834 * dev_set_group - Change group this device belongs to
4836 * @new_group: group this device should belong to
4838 void dev_set_group(struct net_device *dev, int new_group)
4840 dev->group = new_group;
4842 EXPORT_SYMBOL(dev_set_group);
4845 * dev_set_mac_address - Change Media Access Control Address
4849 * Change the hardware (MAC) address of the device
4851 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4853 const struct net_device_ops *ops = dev->netdev_ops;
4856 if (!ops->ndo_set_mac_address)
4858 if (sa->sa_family != dev->type)
4860 if (!netif_device_present(dev))
4862 err = ops->ndo_set_mac_address(dev, sa);
4864 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4865 add_device_randomness(dev->dev_addr, dev->addr_len);
4868 EXPORT_SYMBOL(dev_set_mac_address);
4871 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4873 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4876 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4882 case SIOCGIFFLAGS: /* Get interface flags */
4883 ifr->ifr_flags = (short) dev_get_flags(dev);
4886 case SIOCGIFMETRIC: /* Get the metric on the interface
4887 (currently unused) */
4888 ifr->ifr_metric = 0;
4891 case SIOCGIFMTU: /* Get the MTU of a device */
4892 ifr->ifr_mtu = dev->mtu;
4897 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4899 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4900 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4901 ifr->ifr_hwaddr.sa_family = dev->type;
4909 ifr->ifr_map.mem_start = dev->mem_start;
4910 ifr->ifr_map.mem_end = dev->mem_end;
4911 ifr->ifr_map.base_addr = dev->base_addr;
4912 ifr->ifr_map.irq = dev->irq;
4913 ifr->ifr_map.dma = dev->dma;
4914 ifr->ifr_map.port = dev->if_port;
4918 ifr->ifr_ifindex = dev->ifindex;
4922 ifr->ifr_qlen = dev->tx_queue_len;
4926 /* dev_ioctl() should ensure this case
4938 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4940 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4943 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4944 const struct net_device_ops *ops;
4949 ops = dev->netdev_ops;
4952 case SIOCSIFFLAGS: /* Set interface flags */
4953 return dev_change_flags(dev, ifr->ifr_flags);
4955 case SIOCSIFMETRIC: /* Set the metric on the interface
4956 (currently unused) */
4959 case SIOCSIFMTU: /* Set the MTU of a device */
4960 return dev_set_mtu(dev, ifr->ifr_mtu);
4963 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4965 case SIOCSIFHWBROADCAST:
4966 if (ifr->ifr_hwaddr.sa_family != dev->type)
4968 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4969 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4970 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4974 if (ops->ndo_set_config) {
4975 if (!netif_device_present(dev))
4977 return ops->ndo_set_config(dev, &ifr->ifr_map);
4982 if (!ops->ndo_set_rx_mode ||
4983 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4985 if (!netif_device_present(dev))
4987 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4990 if (!ops->ndo_set_rx_mode ||
4991 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4993 if (!netif_device_present(dev))
4995 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4998 if (ifr->ifr_qlen < 0)
5000 dev->tx_queue_len = ifr->ifr_qlen;
5004 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5005 return dev_change_name(dev, ifr->ifr_newname);
5008 err = net_hwtstamp_validate(ifr);
5014 * Unknown or private ioctl
5017 if ((cmd >= SIOCDEVPRIVATE &&
5018 cmd <= SIOCDEVPRIVATE + 15) ||
5019 cmd == SIOCBONDENSLAVE ||
5020 cmd == SIOCBONDRELEASE ||
5021 cmd == SIOCBONDSETHWADDR ||
5022 cmd == SIOCBONDSLAVEINFOQUERY ||
5023 cmd == SIOCBONDINFOQUERY ||
5024 cmd == SIOCBONDCHANGEACTIVE ||
5025 cmd == SIOCGMIIPHY ||
5026 cmd == SIOCGMIIREG ||
5027 cmd == SIOCSMIIREG ||
5028 cmd == SIOCBRADDIF ||
5029 cmd == SIOCBRDELIF ||
5030 cmd == SIOCSHWTSTAMP ||
5031 cmd == SIOCWANDEV) {
5033 if (ops->ndo_do_ioctl) {
5034 if (netif_device_present(dev))
5035 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5047 * This function handles all "interface"-type I/O control requests. The actual
5048 * 'doing' part of this is dev_ifsioc above.
5052 * dev_ioctl - network device ioctl
5053 * @net: the applicable net namespace
5054 * @cmd: command to issue
5055 * @arg: pointer to a struct ifreq in user space
5057 * Issue ioctl functions to devices. This is normally called by the
5058 * user space syscall interfaces but can sometimes be useful for
5059 * other purposes. The return value is the return from the syscall if
5060 * positive or a negative errno code on error.
5063 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5069 /* One special case: SIOCGIFCONF takes ifconf argument
5070 and requires shared lock, because it sleeps writing
5074 if (cmd == SIOCGIFCONF) {
5076 ret = dev_ifconf(net, (char __user *) arg);
5080 if (cmd == SIOCGIFNAME)
5081 return dev_ifname(net, (struct ifreq __user *)arg);
5083 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5086 ifr.ifr_name[IFNAMSIZ-1] = 0;
5088 colon = strchr(ifr.ifr_name, ':');
5093 * See which interface the caller is talking about.
5098 * These ioctl calls:
5099 * - can be done by all.
5100 * - atomic and do not require locking.
5111 dev_load(net, ifr.ifr_name);
5113 ret = dev_ifsioc_locked(net, &ifr, cmd);
5118 if (copy_to_user(arg, &ifr,
5119 sizeof(struct ifreq)))
5125 dev_load(net, ifr.ifr_name);
5127 ret = dev_ethtool(net, &ifr);
5132 if (copy_to_user(arg, &ifr,
5133 sizeof(struct ifreq)))
5139 * These ioctl calls:
5140 * - require superuser power.
5141 * - require strict serialization.
5147 if (!capable(CAP_NET_ADMIN))
5149 dev_load(net, ifr.ifr_name);
5151 ret = dev_ifsioc(net, &ifr, cmd);
5156 if (copy_to_user(arg, &ifr,
5157 sizeof(struct ifreq)))
5163 * These ioctl calls:
5164 * - require superuser power.
5165 * - require strict serialization.
5166 * - do not return a value
5176 case SIOCSIFHWBROADCAST:
5179 case SIOCBONDENSLAVE:
5180 case SIOCBONDRELEASE:
5181 case SIOCBONDSETHWADDR:
5182 case SIOCBONDCHANGEACTIVE:
5186 if (!capable(CAP_NET_ADMIN))
5189 case SIOCBONDSLAVEINFOQUERY:
5190 case SIOCBONDINFOQUERY:
5191 dev_load(net, ifr.ifr_name);
5193 ret = dev_ifsioc(net, &ifr, cmd);
5198 /* Get the per device memory space. We can add this but
5199 * currently do not support it */
5201 /* Set the per device memory buffer space.
5202 * Not applicable in our case */
5207 * Unknown or private ioctl.
5210 if (cmd == SIOCWANDEV ||
5211 (cmd >= SIOCDEVPRIVATE &&
5212 cmd <= SIOCDEVPRIVATE + 15)) {
5213 dev_load(net, ifr.ifr_name);
5215 ret = dev_ifsioc(net, &ifr, cmd);
5217 if (!ret && copy_to_user(arg, &ifr,
5218 sizeof(struct ifreq)))
5222 /* Take care of Wireless Extensions */
5223 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5224 return wext_handle_ioctl(net, &ifr, cmd, arg);
5231 * dev_new_index - allocate an ifindex
5232 * @net: the applicable net namespace
5234 * Returns a suitable unique value for a new device interface
5235 * number. The caller must hold the rtnl semaphore or the
5236 * dev_base_lock to be sure it remains unique.
5238 static int dev_new_index(struct net *net)
5244 if (!__dev_get_by_index(net, ifindex))
5249 /* Delayed registration/unregisteration */
5250 static LIST_HEAD(net_todo_list);
5252 static void net_set_todo(struct net_device *dev)
5254 list_add_tail(&dev->todo_list, &net_todo_list);
5257 static void rollback_registered_many(struct list_head *head)
5259 struct net_device *dev, *tmp;
5261 BUG_ON(dev_boot_phase);
5264 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5265 /* Some devices call without registering
5266 * for initialization unwind. Remove those
5267 * devices and proceed with the remaining.
5269 if (dev->reg_state == NETREG_UNINITIALIZED) {
5270 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5274 list_del(&dev->unreg_list);
5277 dev->dismantle = true;
5278 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5281 /* If device is running, close it first. */
5282 dev_close_many(head);
5284 list_for_each_entry(dev, head, unreg_list) {
5285 /* And unlink it from device chain. */
5286 unlist_netdevice(dev);
5288 dev->reg_state = NETREG_UNREGISTERING;
5293 list_for_each_entry(dev, head, unreg_list) {
5294 /* Shutdown queueing discipline. */
5298 /* Notify protocols, that we are about to destroy
5299 this device. They should clean all the things.
5301 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5303 if (!dev->rtnl_link_ops ||
5304 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5305 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5308 * Flush the unicast and multicast chains
5313 if (dev->netdev_ops->ndo_uninit)
5314 dev->netdev_ops->ndo_uninit(dev);
5316 /* Notifier chain MUST detach us from master device. */
5317 WARN_ON(dev->master);
5319 /* Remove entries from kobject tree */
5320 netdev_unregister_kobject(dev);
5323 /* Process any work delayed until the end of the batch */
5324 dev = list_first_entry(head, struct net_device, unreg_list);
5325 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5329 list_for_each_entry(dev, head, unreg_list)
5333 static void rollback_registered(struct net_device *dev)
5337 list_add(&dev->unreg_list, &single);
5338 rollback_registered_many(&single);
5342 static netdev_features_t netdev_fix_features(struct net_device *dev,
5343 netdev_features_t features)
5345 /* Fix illegal checksum combinations */
5346 if ((features & NETIF_F_HW_CSUM) &&
5347 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5348 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5349 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5352 /* Fix illegal SG+CSUM combinations. */
5353 if ((features & NETIF_F_SG) &&
5354 !(features & NETIF_F_ALL_CSUM)) {
5356 "Dropping NETIF_F_SG since no checksum feature.\n");
5357 features &= ~NETIF_F_SG;
5360 /* TSO requires that SG is present as well. */
5361 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5362 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5363 features &= ~NETIF_F_ALL_TSO;
5366 /* TSO ECN requires that TSO is present as well. */
5367 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5368 features &= ~NETIF_F_TSO_ECN;
5370 /* Software GSO depends on SG. */
5371 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5372 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5373 features &= ~NETIF_F_GSO;
5376 /* UFO needs SG and checksumming */
5377 if (features & NETIF_F_UFO) {
5378 /* maybe split UFO into V4 and V6? */
5379 if (!((features & NETIF_F_GEN_CSUM) ||
5380 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5381 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5383 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5384 features &= ~NETIF_F_UFO;
5387 if (!(features & NETIF_F_SG)) {
5389 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5390 features &= ~NETIF_F_UFO;
5397 int __netdev_update_features(struct net_device *dev)
5399 netdev_features_t features;
5404 features = netdev_get_wanted_features(dev);
5406 if (dev->netdev_ops->ndo_fix_features)
5407 features = dev->netdev_ops->ndo_fix_features(dev, features);
5409 /* driver might be less strict about feature dependencies */
5410 features = netdev_fix_features(dev, features);
5412 if (dev->features == features)
5415 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5416 &dev->features, &features);
5418 if (dev->netdev_ops->ndo_set_features)
5419 err = dev->netdev_ops->ndo_set_features(dev, features);
5421 if (unlikely(err < 0)) {
5423 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5424 err, &features, &dev->features);
5429 dev->features = features;
5435 * netdev_update_features - recalculate device features
5436 * @dev: the device to check
5438 * Recalculate dev->features set and send notifications if it
5439 * has changed. Should be called after driver or hardware dependent
5440 * conditions might have changed that influence the features.
5442 void netdev_update_features(struct net_device *dev)
5444 if (__netdev_update_features(dev))
5445 netdev_features_change(dev);
5447 EXPORT_SYMBOL(netdev_update_features);
5450 * netdev_change_features - recalculate device features
5451 * @dev: the device to check
5453 * Recalculate dev->features set and send notifications even
5454 * if they have not changed. Should be called instead of
5455 * netdev_update_features() if also dev->vlan_features might
5456 * have changed to allow the changes to be propagated to stacked
5459 void netdev_change_features(struct net_device *dev)
5461 __netdev_update_features(dev);
5462 netdev_features_change(dev);
5464 EXPORT_SYMBOL(netdev_change_features);
5467 * netif_stacked_transfer_operstate - transfer operstate
5468 * @rootdev: the root or lower level device to transfer state from
5469 * @dev: the device to transfer operstate to
5471 * Transfer operational state from root to device. This is normally
5472 * called when a stacking relationship exists between the root
5473 * device and the device(a leaf device).
5475 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5476 struct net_device *dev)
5478 if (rootdev->operstate == IF_OPER_DORMANT)
5479 netif_dormant_on(dev);
5481 netif_dormant_off(dev);
5483 if (netif_carrier_ok(rootdev)) {
5484 if (!netif_carrier_ok(dev))
5485 netif_carrier_on(dev);
5487 if (netif_carrier_ok(dev))
5488 netif_carrier_off(dev);
5491 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5494 static int netif_alloc_rx_queues(struct net_device *dev)
5496 unsigned int i, count = dev->num_rx_queues;
5497 struct netdev_rx_queue *rx;
5501 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5503 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5508 for (i = 0; i < count; i++)
5514 static void netdev_init_one_queue(struct net_device *dev,
5515 struct netdev_queue *queue, void *_unused)
5517 /* Initialize queue lock */
5518 spin_lock_init(&queue->_xmit_lock);
5519 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5520 queue->xmit_lock_owner = -1;
5521 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5524 dql_init(&queue->dql, HZ);
5528 static int netif_alloc_netdev_queues(struct net_device *dev)
5530 unsigned int count = dev->num_tx_queues;
5531 struct netdev_queue *tx;
5535 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5537 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5542 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5543 spin_lock_init(&dev->tx_global_lock);
5549 * register_netdevice - register a network device
5550 * @dev: device to register
5552 * Take a completed network device structure and add it to the kernel
5553 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5554 * chain. 0 is returned on success. A negative errno code is returned
5555 * on a failure to set up the device, or if the name is a duplicate.
5557 * Callers must hold the rtnl semaphore. You may want
5558 * register_netdev() instead of this.
5561 * The locking appears insufficient to guarantee two parallel registers
5562 * will not get the same name.
5565 int register_netdevice(struct net_device *dev)
5568 struct net *net = dev_net(dev);
5570 BUG_ON(dev_boot_phase);
5575 /* When net_device's are persistent, this will be fatal. */
5576 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5579 spin_lock_init(&dev->addr_list_lock);
5580 netdev_set_addr_lockdep_class(dev);
5584 ret = dev_get_valid_name(dev, dev->name);
5588 /* Init, if this function is available */
5589 if (dev->netdev_ops->ndo_init) {
5590 ret = dev->netdev_ops->ndo_init(dev);
5598 dev->ifindex = dev_new_index(net);
5599 if (dev->iflink == -1)
5600 dev->iflink = dev->ifindex;
5602 /* Transfer changeable features to wanted_features and enable
5603 * software offloads (GSO and GRO).
5605 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5606 dev->features |= NETIF_F_SOFT_FEATURES;
5607 dev->wanted_features = dev->features & dev->hw_features;
5609 /* Turn on no cache copy if HW is doing checksum */
5610 if (!(dev->flags & IFF_LOOPBACK)) {
5611 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5612 if (dev->features & NETIF_F_ALL_CSUM) {
5613 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5614 dev->features |= NETIF_F_NOCACHE_COPY;
5618 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5620 dev->vlan_features |= NETIF_F_HIGHDMA;
5622 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5623 ret = notifier_to_errno(ret);
5627 ret = netdev_register_kobject(dev);
5630 dev->reg_state = NETREG_REGISTERED;
5632 __netdev_update_features(dev);
5635 * Default initial state at registry is that the
5636 * device is present.
5639 set_bit(__LINK_STATE_PRESENT, &dev->state);
5641 dev_init_scheduler(dev);
5643 list_netdevice(dev);
5644 add_device_randomness(dev->dev_addr, dev->addr_len);
5646 /* Notify protocols, that a new device appeared. */
5647 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5648 ret = notifier_to_errno(ret);
5650 rollback_registered(dev);
5651 dev->reg_state = NETREG_UNREGISTERED;
5654 * Prevent userspace races by waiting until the network
5655 * device is fully setup before sending notifications.
5657 if (!dev->rtnl_link_ops ||
5658 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5659 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5665 if (dev->netdev_ops->ndo_uninit)
5666 dev->netdev_ops->ndo_uninit(dev);
5669 EXPORT_SYMBOL(register_netdevice);
5672 * init_dummy_netdev - init a dummy network device for NAPI
5673 * @dev: device to init
5675 * This takes a network device structure and initialize the minimum
5676 * amount of fields so it can be used to schedule NAPI polls without
5677 * registering a full blown interface. This is to be used by drivers
5678 * that need to tie several hardware interfaces to a single NAPI
5679 * poll scheduler due to HW limitations.
5681 int init_dummy_netdev(struct net_device *dev)
5683 /* Clear everything. Note we don't initialize spinlocks
5684 * are they aren't supposed to be taken by any of the
5685 * NAPI code and this dummy netdev is supposed to be
5686 * only ever used for NAPI polls
5688 memset(dev, 0, sizeof(struct net_device));
5690 /* make sure we BUG if trying to hit standard
5691 * register/unregister code path
5693 dev->reg_state = NETREG_DUMMY;
5695 /* NAPI wants this */
5696 INIT_LIST_HEAD(&dev->napi_list);
5698 /* a dummy interface is started by default */
5699 set_bit(__LINK_STATE_PRESENT, &dev->state);
5700 set_bit(__LINK_STATE_START, &dev->state);
5702 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5703 * because users of this 'device' dont need to change
5709 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5713 * register_netdev - register a network device
5714 * @dev: device to register
5716 * Take a completed network device structure and add it to the kernel
5717 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5718 * chain. 0 is returned on success. A negative errno code is returned
5719 * on a failure to set up the device, or if the name is a duplicate.
5721 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5722 * and expands the device name if you passed a format string to
5725 int register_netdev(struct net_device *dev)
5730 err = register_netdevice(dev);
5734 EXPORT_SYMBOL(register_netdev);
5736 int netdev_refcnt_read(const struct net_device *dev)
5740 for_each_possible_cpu(i)
5741 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5744 EXPORT_SYMBOL(netdev_refcnt_read);
5747 * netdev_wait_allrefs - wait until all references are gone.
5748 * @dev: target net_device
5750 * This is called when unregistering network devices.
5752 * Any protocol or device that holds a reference should register
5753 * for netdevice notification, and cleanup and put back the
5754 * reference if they receive an UNREGISTER event.
5755 * We can get stuck here if buggy protocols don't correctly
5758 static void netdev_wait_allrefs(struct net_device *dev)
5760 unsigned long rebroadcast_time, warning_time;
5763 linkwatch_forget_dev(dev);
5765 rebroadcast_time = warning_time = jiffies;
5766 refcnt = netdev_refcnt_read(dev);
5768 while (refcnt != 0) {
5769 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5772 /* Rebroadcast unregister notification */
5773 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5774 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5775 * should have already handle it the first time */
5777 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5779 /* We must not have linkwatch events
5780 * pending on unregister. If this
5781 * happens, we simply run the queue
5782 * unscheduled, resulting in a noop
5785 linkwatch_run_queue();
5790 rebroadcast_time = jiffies;
5795 refcnt = netdev_refcnt_read(dev);
5797 if (time_after(jiffies, warning_time + 10 * HZ)) {
5798 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5800 warning_time = jiffies;
5809 * register_netdevice(x1);
5810 * register_netdevice(x2);
5812 * unregister_netdevice(y1);
5813 * unregister_netdevice(y2);
5819 * We are invoked by rtnl_unlock().
5820 * This allows us to deal with problems:
5821 * 1) We can delete sysfs objects which invoke hotplug
5822 * without deadlocking with linkwatch via keventd.
5823 * 2) Since we run with the RTNL semaphore not held, we can sleep
5824 * safely in order to wait for the netdev refcnt to drop to zero.
5826 * We must not return until all unregister events added during
5827 * the interval the lock was held have been completed.
5829 void netdev_run_todo(void)
5831 struct list_head list;
5833 /* Snapshot list, allow later requests */
5834 list_replace_init(&net_todo_list, &list);
5838 /* Wait for rcu callbacks to finish before attempting to drain
5839 * the device list. This usually avoids a 250ms wait.
5841 if (!list_empty(&list))
5844 while (!list_empty(&list)) {
5845 struct net_device *dev
5846 = list_first_entry(&list, struct net_device, todo_list);
5847 list_del(&dev->todo_list);
5849 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5850 pr_err("network todo '%s' but state %d\n",
5851 dev->name, dev->reg_state);
5856 dev->reg_state = NETREG_UNREGISTERED;
5858 on_each_cpu(flush_backlog, dev, 1);
5860 netdev_wait_allrefs(dev);
5863 BUG_ON(netdev_refcnt_read(dev));
5864 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5865 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5866 WARN_ON(dev->dn_ptr);
5868 if (dev->destructor)
5869 dev->destructor(dev);
5871 /* Free network device */
5872 kobject_put(&dev->dev.kobj);
5876 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5877 * fields in the same order, with only the type differing.
5879 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5880 const struct net_device_stats *netdev_stats)
5882 #if BITS_PER_LONG == 64
5883 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5884 memcpy(stats64, netdev_stats, sizeof(*stats64));
5886 size_t i, n = sizeof(*stats64) / sizeof(u64);
5887 const unsigned long *src = (const unsigned long *)netdev_stats;
5888 u64 *dst = (u64 *)stats64;
5890 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5891 sizeof(*stats64) / sizeof(u64));
5892 for (i = 0; i < n; i++)
5896 EXPORT_SYMBOL(netdev_stats_to_stats64);
5899 * dev_get_stats - get network device statistics
5900 * @dev: device to get statistics from
5901 * @storage: place to store stats
5903 * Get network statistics from device. Return @storage.
5904 * The device driver may provide its own method by setting
5905 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5906 * otherwise the internal statistics structure is used.
5908 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5909 struct rtnl_link_stats64 *storage)
5911 const struct net_device_ops *ops = dev->netdev_ops;
5913 if (ops->ndo_get_stats64) {
5914 memset(storage, 0, sizeof(*storage));
5915 ops->ndo_get_stats64(dev, storage);
5916 } else if (ops->ndo_get_stats) {
5917 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5919 netdev_stats_to_stats64(storage, &dev->stats);
5921 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5924 EXPORT_SYMBOL(dev_get_stats);
5926 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5928 struct netdev_queue *queue = dev_ingress_queue(dev);
5930 #ifdef CONFIG_NET_CLS_ACT
5933 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5936 netdev_init_one_queue(dev, queue, NULL);
5937 queue->qdisc = &noop_qdisc;
5938 queue->qdisc_sleeping = &noop_qdisc;
5939 rcu_assign_pointer(dev->ingress_queue, queue);
5945 * alloc_netdev_mqs - allocate network device
5946 * @sizeof_priv: size of private data to allocate space for
5947 * @name: device name format string
5948 * @setup: callback to initialize device
5949 * @txqs: the number of TX subqueues to allocate
5950 * @rxqs: the number of RX subqueues to allocate
5952 * Allocates a struct net_device with private data area for driver use
5953 * and performs basic initialization. Also allocates subquue structs
5954 * for each queue on the device.
5956 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5957 void (*setup)(struct net_device *),
5958 unsigned int txqs, unsigned int rxqs)
5960 struct net_device *dev;
5962 struct net_device *p;
5964 BUG_ON(strlen(name) >= sizeof(dev->name));
5967 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5973 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5978 alloc_size = sizeof(struct net_device);
5980 /* ensure 32-byte alignment of private area */
5981 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5982 alloc_size += sizeof_priv;
5984 /* ensure 32-byte alignment of whole construct */
5985 alloc_size += NETDEV_ALIGN - 1;
5987 p = kzalloc(alloc_size, GFP_KERNEL);
5989 pr_err("alloc_netdev: Unable to allocate device\n");
5993 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5994 dev->padded = (char *)dev - (char *)p;
5996 dev->pcpu_refcnt = alloc_percpu(int);
5997 if (!dev->pcpu_refcnt)
6000 if (dev_addr_init(dev))
6006 dev_net_set(dev, &init_net);
6008 dev->gso_max_size = GSO_MAX_SIZE;
6009 dev->gso_max_segs = GSO_MAX_SEGS;
6011 INIT_LIST_HEAD(&dev->napi_list);
6012 INIT_LIST_HEAD(&dev->unreg_list);
6013 INIT_LIST_HEAD(&dev->link_watch_list);
6014 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6017 dev->num_tx_queues = txqs;
6018 dev->real_num_tx_queues = txqs;
6019 if (netif_alloc_netdev_queues(dev))
6023 dev->num_rx_queues = rxqs;
6024 dev->real_num_rx_queues = rxqs;
6025 if (netif_alloc_rx_queues(dev))
6029 strcpy(dev->name, name);
6030 dev->group = INIT_NETDEV_GROUP;
6038 free_percpu(dev->pcpu_refcnt);
6048 EXPORT_SYMBOL(alloc_netdev_mqs);
6051 * free_netdev - free network device
6054 * This function does the last stage of destroying an allocated device
6055 * interface. The reference to the device object is released.
6056 * If this is the last reference then it will be freed.
6058 void free_netdev(struct net_device *dev)
6060 struct napi_struct *p, *n;
6062 release_net(dev_net(dev));
6069 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6071 /* Flush device addresses */
6072 dev_addr_flush(dev);
6074 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6077 free_percpu(dev->pcpu_refcnt);
6078 dev->pcpu_refcnt = NULL;
6080 /* Compatibility with error handling in drivers */
6081 if (dev->reg_state == NETREG_UNINITIALIZED) {
6082 kfree((char *)dev - dev->padded);
6086 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6087 dev->reg_state = NETREG_RELEASED;
6089 /* will free via device release */
6090 put_device(&dev->dev);
6092 EXPORT_SYMBOL(free_netdev);
6095 * synchronize_net - Synchronize with packet receive processing
6097 * Wait for packets currently being received to be done.
6098 * Does not block later packets from starting.
6100 void synchronize_net(void)
6103 if (rtnl_is_locked())
6104 synchronize_rcu_expedited();
6108 EXPORT_SYMBOL(synchronize_net);
6111 * unregister_netdevice_queue - remove device from the kernel
6115 * This function shuts down a device interface and removes it
6116 * from the kernel tables.
6117 * If head not NULL, device is queued to be unregistered later.
6119 * Callers must hold the rtnl semaphore. You may want
6120 * unregister_netdev() instead of this.
6123 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6128 list_move_tail(&dev->unreg_list, head);
6130 rollback_registered(dev);
6131 /* Finish processing unregister after unlock */
6135 EXPORT_SYMBOL(unregister_netdevice_queue);
6138 * unregister_netdevice_many - unregister many devices
6139 * @head: list of devices
6141 void unregister_netdevice_many(struct list_head *head)
6143 struct net_device *dev;
6145 if (!list_empty(head)) {
6146 rollback_registered_many(head);
6147 list_for_each_entry(dev, head, unreg_list)
6151 EXPORT_SYMBOL(unregister_netdevice_many);
6154 * unregister_netdev - remove device from the kernel
6157 * This function shuts down a device interface and removes it
6158 * from the kernel tables.
6160 * This is just a wrapper for unregister_netdevice that takes
6161 * the rtnl semaphore. In general you want to use this and not
6162 * unregister_netdevice.
6164 void unregister_netdev(struct net_device *dev)
6167 unregister_netdevice(dev);
6170 EXPORT_SYMBOL(unregister_netdev);
6173 * dev_change_net_namespace - move device to different nethost namespace
6175 * @net: network namespace
6176 * @pat: If not NULL name pattern to try if the current device name
6177 * is already taken in the destination network namespace.
6179 * This function shuts down a device interface and moves it
6180 * to a new network namespace. On success 0 is returned, on
6181 * a failure a netagive errno code is returned.
6183 * Callers must hold the rtnl semaphore.
6186 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6192 /* Don't allow namespace local devices to be moved. */
6194 if (dev->features & NETIF_F_NETNS_LOCAL)
6197 /* Ensure the device has been registrered */
6199 if (dev->reg_state != NETREG_REGISTERED)
6202 /* Get out if there is nothing todo */
6204 if (net_eq(dev_net(dev), net))
6207 /* Pick the destination device name, and ensure
6208 * we can use it in the destination network namespace.
6211 if (__dev_get_by_name(net, dev->name)) {
6212 /* We get here if we can't use the current device name */
6215 if (dev_get_valid_name(dev, pat) < 0)
6220 * And now a mini version of register_netdevice unregister_netdevice.
6223 /* If device is running close it first. */
6226 /* And unlink it from device chain */
6228 unlist_netdevice(dev);
6232 /* Shutdown queueing discipline. */
6235 /* Notify protocols, that we are about to destroy
6236 this device. They should clean all the things.
6238 Note that dev->reg_state stays at NETREG_REGISTERED.
6239 This is wanted because this way 8021q and macvlan know
6240 the device is just moving and can keep their slaves up.
6242 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6243 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6244 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6247 * Flush the unicast and multicast chains
6252 /* Actually switch the network namespace */
6253 dev_net_set(dev, net);
6255 /* If there is an ifindex conflict assign a new one */
6256 if (__dev_get_by_index(net, dev->ifindex)) {
6257 int iflink = (dev->iflink == dev->ifindex);
6258 dev->ifindex = dev_new_index(net);
6260 dev->iflink = dev->ifindex;
6263 /* Fixup kobjects */
6264 err = device_rename(&dev->dev, dev->name);
6267 /* Add the device back in the hashes */
6268 list_netdevice(dev);
6270 /* Notify protocols, that a new device appeared. */
6271 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6274 * Prevent userspace races by waiting until the network
6275 * device is fully setup before sending notifications.
6277 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6284 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6286 static int dev_cpu_callback(struct notifier_block *nfb,
6287 unsigned long action,
6290 struct sk_buff **list_skb;
6291 struct sk_buff *skb;
6292 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6293 struct softnet_data *sd, *oldsd;
6295 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6298 local_irq_disable();
6299 cpu = smp_processor_id();
6300 sd = &per_cpu(softnet_data, cpu);
6301 oldsd = &per_cpu(softnet_data, oldcpu);
6303 /* Find end of our completion_queue. */
6304 list_skb = &sd->completion_queue;
6306 list_skb = &(*list_skb)->next;
6307 /* Append completion queue from offline CPU. */
6308 *list_skb = oldsd->completion_queue;
6309 oldsd->completion_queue = NULL;
6311 /* Append output queue from offline CPU. */
6312 if (oldsd->output_queue) {
6313 *sd->output_queue_tailp = oldsd->output_queue;
6314 sd->output_queue_tailp = oldsd->output_queue_tailp;
6315 oldsd->output_queue = NULL;
6316 oldsd->output_queue_tailp = &oldsd->output_queue;
6318 /* Append NAPI poll list from offline CPU. */
6319 if (!list_empty(&oldsd->poll_list)) {
6320 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6321 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6324 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6327 /* Process offline CPU's input_pkt_queue */
6328 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6330 input_queue_head_incr(oldsd);
6332 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6334 input_queue_head_incr(oldsd);
6342 * netdev_increment_features - increment feature set by one
6343 * @all: current feature set
6344 * @one: new feature set
6345 * @mask: mask feature set
6347 * Computes a new feature set after adding a device with feature set
6348 * @one to the master device with current feature set @all. Will not
6349 * enable anything that is off in @mask. Returns the new feature set.
6351 netdev_features_t netdev_increment_features(netdev_features_t all,
6352 netdev_features_t one, netdev_features_t mask)
6354 if (mask & NETIF_F_GEN_CSUM)
6355 mask |= NETIF_F_ALL_CSUM;
6356 mask |= NETIF_F_VLAN_CHALLENGED;
6358 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6359 all &= one | ~NETIF_F_ALL_FOR_ALL;
6361 /* If one device supports hw checksumming, set for all. */
6362 if (all & NETIF_F_GEN_CSUM)
6363 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6367 EXPORT_SYMBOL(netdev_increment_features);
6369 static struct hlist_head *netdev_create_hash(void)
6372 struct hlist_head *hash;
6374 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6376 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6377 INIT_HLIST_HEAD(&hash[i]);
6382 /* Initialize per network namespace state */
6383 static int __net_init netdev_init(struct net *net)
6385 if (net != &init_net)
6386 INIT_LIST_HEAD(&net->dev_base_head);
6388 net->dev_name_head = netdev_create_hash();
6389 if (net->dev_name_head == NULL)
6392 net->dev_index_head = netdev_create_hash();
6393 if (net->dev_index_head == NULL)
6399 kfree(net->dev_name_head);
6405 * netdev_drivername - network driver for the device
6406 * @dev: network device
6408 * Determine network driver for device.
6410 const char *netdev_drivername(const struct net_device *dev)
6412 const struct device_driver *driver;
6413 const struct device *parent;
6414 const char *empty = "";
6416 parent = dev->dev.parent;
6420 driver = parent->driver;
6421 if (driver && driver->name)
6422 return driver->name;
6426 int __netdev_printk(const char *level, const struct net_device *dev,
6427 struct va_format *vaf)
6431 if (dev && dev->dev.parent)
6432 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6433 netdev_name(dev), vaf);
6435 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6437 r = printk("%s(NULL net_device): %pV", level, vaf);
6441 EXPORT_SYMBOL(__netdev_printk);
6443 int netdev_printk(const char *level, const struct net_device *dev,
6444 const char *format, ...)
6446 struct va_format vaf;
6450 va_start(args, format);
6455 r = __netdev_printk(level, dev, &vaf);
6460 EXPORT_SYMBOL(netdev_printk);
6462 #define define_netdev_printk_level(func, level) \
6463 int func(const struct net_device *dev, const char *fmt, ...) \
6466 struct va_format vaf; \
6469 va_start(args, fmt); \
6474 r = __netdev_printk(level, dev, &vaf); \
6479 EXPORT_SYMBOL(func);
6481 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6482 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6483 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6484 define_netdev_printk_level(netdev_err, KERN_ERR);
6485 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6486 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6487 define_netdev_printk_level(netdev_info, KERN_INFO);
6489 static void __net_exit netdev_exit(struct net *net)
6491 kfree(net->dev_name_head);
6492 kfree(net->dev_index_head);
6495 static struct pernet_operations __net_initdata netdev_net_ops = {
6496 .init = netdev_init,
6497 .exit = netdev_exit,
6500 static void __net_exit default_device_exit(struct net *net)
6502 struct net_device *dev, *aux;
6504 * Push all migratable network devices back to the
6505 * initial network namespace
6508 for_each_netdev_safe(net, dev, aux) {
6510 char fb_name[IFNAMSIZ];
6512 /* Ignore unmoveable devices (i.e. loopback) */
6513 if (dev->features & NETIF_F_NETNS_LOCAL)
6516 /* Leave virtual devices for the generic cleanup */
6517 if (dev->rtnl_link_ops)
6520 /* Push remaining network devices to init_net */
6521 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6522 err = dev_change_net_namespace(dev, &init_net, fb_name);
6524 pr_emerg("%s: failed to move %s to init_net: %d\n",
6525 __func__, dev->name, err);
6532 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6534 /* At exit all network devices most be removed from a network
6535 * namespace. Do this in the reverse order of registration.
6536 * Do this across as many network namespaces as possible to
6537 * improve batching efficiency.
6539 struct net_device *dev;
6541 LIST_HEAD(dev_kill_list);
6544 list_for_each_entry(net, net_list, exit_list) {
6545 for_each_netdev_reverse(net, dev) {
6546 if (dev->rtnl_link_ops)
6547 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6549 unregister_netdevice_queue(dev, &dev_kill_list);
6552 unregister_netdevice_many(&dev_kill_list);
6553 list_del(&dev_kill_list);
6557 static struct pernet_operations __net_initdata default_device_ops = {
6558 .exit = default_device_exit,
6559 .exit_batch = default_device_exit_batch,
6563 * Initialize the DEV module. At boot time this walks the device list and
6564 * unhooks any devices that fail to initialise (normally hardware not
6565 * present) and leaves us with a valid list of present and active devices.
6570 * This is called single threaded during boot, so no need
6571 * to take the rtnl semaphore.
6573 static int __init net_dev_init(void)
6575 int i, rc = -ENOMEM;
6577 BUG_ON(!dev_boot_phase);
6579 if (dev_proc_init())
6582 if (netdev_kobject_init())
6585 INIT_LIST_HEAD(&ptype_all);
6586 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6587 INIT_LIST_HEAD(&ptype_base[i]);
6589 if (register_pernet_subsys(&netdev_net_ops))
6593 * Initialise the packet receive queues.
6596 for_each_possible_cpu(i) {
6597 struct softnet_data *sd = &per_cpu(softnet_data, i);
6599 memset(sd, 0, sizeof(*sd));
6600 skb_queue_head_init(&sd->input_pkt_queue);
6601 skb_queue_head_init(&sd->process_queue);
6602 sd->completion_queue = NULL;
6603 INIT_LIST_HEAD(&sd->poll_list);
6604 sd->output_queue = NULL;
6605 sd->output_queue_tailp = &sd->output_queue;
6607 sd->csd.func = rps_trigger_softirq;
6613 sd->backlog.poll = process_backlog;
6614 sd->backlog.weight = weight_p;
6615 sd->backlog.gro_list = NULL;
6616 sd->backlog.gro_count = 0;
6621 /* The loopback device is special if any other network devices
6622 * is present in a network namespace the loopback device must
6623 * be present. Since we now dynamically allocate and free the
6624 * loopback device ensure this invariant is maintained by
6625 * keeping the loopback device as the first device on the
6626 * list of network devices. Ensuring the loopback devices
6627 * is the first device that appears and the last network device
6630 if (register_pernet_device(&loopback_net_ops))
6633 if (register_pernet_device(&default_device_ops))
6636 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6637 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6639 hotcpu_notifier(dev_cpu_callback, 0);
6647 subsys_initcall(net_dev_init);
6649 static int __init initialize_hashrnd(void)
6651 get_random_bytes(&hashrnd, sizeof(hashrnd));
6655 late_initcall_sync(initialize_hashrnd);