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 hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
212 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
215 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
217 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
220 static inline void rps_lock(struct softnet_data *sd)
223 spin_lock(&sd->input_pkt_queue.lock);
227 static inline void rps_unlock(struct softnet_data *sd)
230 spin_unlock(&sd->input_pkt_queue.lock);
234 /* Device list insertion */
235 static int list_netdevice(struct net_device *dev)
237 struct net *net = dev_net(dev);
241 write_lock_bh(&dev_base_lock);
242 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
243 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
244 hlist_add_head_rcu(&dev->index_hlist,
245 dev_index_hash(net, dev->ifindex));
246 write_unlock_bh(&dev_base_lock);
248 dev_base_seq_inc(net);
253 /* Device list removal
254 * caller must respect a RCU grace period before freeing/reusing dev
256 static void unlist_netdevice(struct net_device *dev)
260 /* Unlink dev from the device chain */
261 write_lock_bh(&dev_base_lock);
262 list_del_rcu(&dev->dev_list);
263 hlist_del_rcu(&dev->name_hlist);
264 hlist_del_rcu(&dev->index_hlist);
265 write_unlock_bh(&dev_base_lock);
267 dev_base_seq_inc(dev_net(dev));
274 static RAW_NOTIFIER_HEAD(netdev_chain);
277 * Device drivers call our routines to queue packets here. We empty the
278 * queue in the local softnet handler.
281 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
282 EXPORT_PER_CPU_SYMBOL(softnet_data);
284 #ifdef CONFIG_LOCKDEP
286 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
287 * according to dev->type
289 static const unsigned short netdev_lock_type[] =
290 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
291 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
292 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
293 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
294 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
295 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
296 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
297 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
298 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
299 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
300 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
301 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
302 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
303 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
304 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
305 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_IEEE802_TR", "_xmit_IEEE80211",
321 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
322 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
323 "_xmit_VOID", "_xmit_NONE"};
325 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
326 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
328 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
332 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
333 if (netdev_lock_type[i] == dev_type)
335 /* the last key is used by default */
336 return ARRAY_SIZE(netdev_lock_type) - 1;
339 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
340 unsigned short dev_type)
344 i = netdev_lock_pos(dev_type);
345 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
346 netdev_lock_name[i]);
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
353 i = netdev_lock_pos(dev->type);
354 lockdep_set_class_and_name(&dev->addr_list_lock,
355 &netdev_addr_lock_key[i],
356 netdev_lock_name[i]);
359 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
360 unsigned short dev_type)
363 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
368 /*******************************************************************************
370 Protocol management and registration routines
372 *******************************************************************************/
375 * Add a protocol ID to the list. Now that the input handler is
376 * smarter we can dispense with all the messy stuff that used to be
379 * BEWARE!!! Protocol handlers, mangling input packets,
380 * MUST BE last in hash buckets and checking protocol handlers
381 * MUST start from promiscuous ptype_all chain in net_bh.
382 * It is true now, do not change it.
383 * Explanation follows: if protocol handler, mangling packet, will
384 * be the first on list, it is not able to sense, that packet
385 * is cloned and should be copied-on-write, so that it will
386 * change it and subsequent readers will get broken packet.
390 static inline struct list_head *ptype_head(const struct packet_type *pt)
392 if (pt->type == htons(ETH_P_ALL))
395 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
399 * dev_add_pack - add packet handler
400 * @pt: packet type declaration
402 * Add a protocol handler to the networking stack. The passed &packet_type
403 * is linked into kernel lists and may not be freed until it has been
404 * removed from the kernel lists.
406 * This call does not sleep therefore it can not
407 * guarantee all CPU's that are in middle of receiving packets
408 * will see the new packet type (until the next received packet).
411 void dev_add_pack(struct packet_type *pt)
413 struct list_head *head = ptype_head(pt);
415 spin_lock(&ptype_lock);
416 list_add_rcu(&pt->list, head);
417 spin_unlock(&ptype_lock);
419 EXPORT_SYMBOL(dev_add_pack);
422 * __dev_remove_pack - remove packet handler
423 * @pt: packet type declaration
425 * Remove a protocol handler that was previously added to the kernel
426 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
427 * from the kernel lists and can be freed or reused once this function
430 * The packet type might still be in use by receivers
431 * and must not be freed until after all the CPU's have gone
432 * through a quiescent state.
434 void __dev_remove_pack(struct packet_type *pt)
436 struct list_head *head = ptype_head(pt);
437 struct packet_type *pt1;
439 spin_lock(&ptype_lock);
441 list_for_each_entry(pt1, head, list) {
443 list_del_rcu(&pt->list);
448 pr_warn("dev_remove_pack: %p not found\n", pt);
450 spin_unlock(&ptype_lock);
452 EXPORT_SYMBOL(__dev_remove_pack);
455 * dev_remove_pack - remove packet handler
456 * @pt: packet type declaration
458 * Remove a protocol handler that was previously added to the kernel
459 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
460 * from the kernel lists and can be freed or reused once this function
463 * This call sleeps to guarantee that no CPU is looking at the packet
466 void dev_remove_pack(struct packet_type *pt)
468 __dev_remove_pack(pt);
472 EXPORT_SYMBOL(dev_remove_pack);
474 /******************************************************************************
476 Device Boot-time Settings Routines
478 *******************************************************************************/
480 /* Boot time configuration table */
481 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
484 * netdev_boot_setup_add - add new setup entry
485 * @name: name of the device
486 * @map: configured settings for the device
488 * Adds new setup entry to the dev_boot_setup list. The function
489 * returns 0 on error and 1 on success. This is a generic routine to
492 static int netdev_boot_setup_add(char *name, struct ifmap *map)
494 struct netdev_boot_setup *s;
498 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
499 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
500 memset(s[i].name, 0, sizeof(s[i].name));
501 strlcpy(s[i].name, name, IFNAMSIZ);
502 memcpy(&s[i].map, map, sizeof(s[i].map));
507 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
511 * netdev_boot_setup_check - check boot time settings
512 * @dev: the netdevice
514 * Check boot time settings for the device.
515 * The found settings are set for the device to be used
516 * later in the device probing.
517 * Returns 0 if no settings found, 1 if they are.
519 int netdev_boot_setup_check(struct net_device *dev)
521 struct netdev_boot_setup *s = dev_boot_setup;
524 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
525 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
526 !strcmp(dev->name, s[i].name)) {
527 dev->irq = s[i].map.irq;
528 dev->base_addr = s[i].map.base_addr;
529 dev->mem_start = s[i].map.mem_start;
530 dev->mem_end = s[i].map.mem_end;
536 EXPORT_SYMBOL(netdev_boot_setup_check);
540 * netdev_boot_base - get address from boot time settings
541 * @prefix: prefix for network device
542 * @unit: id for network device
544 * Check boot time settings for the base address of device.
545 * The found settings are set for the device to be used
546 * later in the device probing.
547 * Returns 0 if no settings found.
549 unsigned long netdev_boot_base(const char *prefix, int unit)
551 const struct netdev_boot_setup *s = dev_boot_setup;
555 sprintf(name, "%s%d", prefix, unit);
558 * If device already registered then return base of 1
559 * to indicate not to probe for this interface
561 if (__dev_get_by_name(&init_net, name))
564 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
565 if (!strcmp(name, s[i].name))
566 return s[i].map.base_addr;
571 * Saves at boot time configured settings for any netdevice.
573 int __init netdev_boot_setup(char *str)
578 str = get_options(str, ARRAY_SIZE(ints), ints);
583 memset(&map, 0, sizeof(map));
587 map.base_addr = ints[2];
589 map.mem_start = ints[3];
591 map.mem_end = ints[4];
593 /* Add new entry to the list */
594 return netdev_boot_setup_add(str, &map);
597 __setup("netdev=", netdev_boot_setup);
599 /*******************************************************************************
601 Device Interface Subroutines
603 *******************************************************************************/
606 * __dev_get_by_name - find a device by its name
607 * @net: the applicable net namespace
608 * @name: name to find
610 * Find an interface by name. Must be called under RTNL semaphore
611 * or @dev_base_lock. If the name is found a pointer to the device
612 * is returned. If the name is not found then %NULL is returned. The
613 * reference counters are not incremented so the caller must be
614 * careful with locks.
617 struct net_device *__dev_get_by_name(struct net *net, const char *name)
619 struct hlist_node *p;
620 struct net_device *dev;
621 struct hlist_head *head = dev_name_hash(net, name);
623 hlist_for_each_entry(dev, p, head, name_hlist)
624 if (!strncmp(dev->name, name, IFNAMSIZ))
629 EXPORT_SYMBOL(__dev_get_by_name);
632 * dev_get_by_name_rcu - find a device by its name
633 * @net: the applicable net namespace
634 * @name: name to find
636 * Find an interface by name.
637 * If the name is found a pointer to the device is returned.
638 * If the name is not found then %NULL is returned.
639 * The reference counters are not incremented so the caller must be
640 * careful with locks. The caller must hold RCU lock.
643 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
645 struct hlist_node *p;
646 struct net_device *dev;
647 struct hlist_head *head = dev_name_hash(net, name);
649 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
650 if (!strncmp(dev->name, name, IFNAMSIZ))
655 EXPORT_SYMBOL(dev_get_by_name_rcu);
658 * dev_get_by_name - find a device by its name
659 * @net: the applicable net namespace
660 * @name: name to find
662 * Find an interface by name. This can be called from any
663 * context and does its own locking. The returned handle has
664 * the usage count incremented and the caller must use dev_put() to
665 * release it when it is no longer needed. %NULL is returned if no
666 * matching device is found.
669 struct net_device *dev_get_by_name(struct net *net, const char *name)
671 struct net_device *dev;
674 dev = dev_get_by_name_rcu(net, name);
680 EXPORT_SYMBOL(dev_get_by_name);
683 * __dev_get_by_index - find a device by its ifindex
684 * @net: the applicable net namespace
685 * @ifindex: index of device
687 * Search for an interface by index. Returns %NULL if the device
688 * is not found or a pointer to the device. The device has not
689 * had its reference counter increased so the caller must be careful
690 * about locking. The caller must hold either the RTNL semaphore
694 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
696 struct hlist_node *p;
697 struct net_device *dev;
698 struct hlist_head *head = dev_index_hash(net, ifindex);
700 hlist_for_each_entry(dev, p, head, index_hlist)
701 if (dev->ifindex == ifindex)
706 EXPORT_SYMBOL(__dev_get_by_index);
709 * dev_get_by_index_rcu - find a device by its ifindex
710 * @net: the applicable net namespace
711 * @ifindex: index of device
713 * Search for an interface by index. Returns %NULL if the device
714 * is not found or a pointer to the device. The device has not
715 * had its reference counter increased so the caller must be careful
716 * about locking. The caller must hold RCU lock.
719 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
721 struct hlist_node *p;
722 struct net_device *dev;
723 struct hlist_head *head = dev_index_hash(net, ifindex);
725 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
726 if (dev->ifindex == ifindex)
731 EXPORT_SYMBOL(dev_get_by_index_rcu);
735 * dev_get_by_index - find a device by its ifindex
736 * @net: the applicable net namespace
737 * @ifindex: index of device
739 * Search for an interface by index. Returns NULL if the device
740 * is not found or a pointer to the device. The device returned has
741 * had a reference added and the pointer is safe until the user calls
742 * dev_put to indicate they have finished with it.
745 struct net_device *dev_get_by_index(struct net *net, int ifindex)
747 struct net_device *dev;
750 dev = dev_get_by_index_rcu(net, ifindex);
756 EXPORT_SYMBOL(dev_get_by_index);
759 * dev_getbyhwaddr_rcu - find a device by its hardware address
760 * @net: the applicable net namespace
761 * @type: media type of device
762 * @ha: hardware address
764 * Search for an interface by MAC address. Returns NULL if the device
765 * is not found or a pointer to the device.
766 * The caller must hold RCU or RTNL.
767 * The returned device has not had its ref count increased
768 * and the caller must therefore be careful about locking
772 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
775 struct net_device *dev;
777 for_each_netdev_rcu(net, dev)
778 if (dev->type == type &&
779 !memcmp(dev->dev_addr, ha, dev->addr_len))
784 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
786 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
788 struct net_device *dev;
791 for_each_netdev(net, dev)
792 if (dev->type == type)
797 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
799 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
801 struct net_device *dev, *ret = NULL;
804 for_each_netdev_rcu(net, dev)
805 if (dev->type == type) {
813 EXPORT_SYMBOL(dev_getfirstbyhwtype);
816 * dev_get_by_flags_rcu - find any device with given flags
817 * @net: the applicable net namespace
818 * @if_flags: IFF_* values
819 * @mask: bitmask of bits in if_flags to check
821 * Search for any interface with the given flags. Returns NULL if a device
822 * is not found or a pointer to the device. Must be called inside
823 * rcu_read_lock(), and result refcount is unchanged.
826 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
829 struct net_device *dev, *ret;
832 for_each_netdev_rcu(net, dev) {
833 if (((dev->flags ^ if_flags) & mask) == 0) {
840 EXPORT_SYMBOL(dev_get_by_flags_rcu);
843 * dev_valid_name - check if name is okay for network device
846 * Network device names need to be valid file names to
847 * to allow sysfs to work. We also disallow any kind of
850 bool dev_valid_name(const char *name)
854 if (strlen(name) >= IFNAMSIZ)
856 if (!strcmp(name, ".") || !strcmp(name, ".."))
860 if (*name == '/' || isspace(*name))
866 EXPORT_SYMBOL(dev_valid_name);
869 * __dev_alloc_name - allocate a name for a device
870 * @net: network namespace to allocate the device name in
871 * @name: name format string
872 * @buf: scratch buffer and result name string
874 * Passed a format string - eg "lt%d" it will try and find a suitable
875 * id. It scans list of devices to build up a free map, then chooses
876 * the first empty slot. The caller must hold the dev_base or rtnl lock
877 * while allocating the name and adding the device in order to avoid
879 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
880 * Returns the number of the unit assigned or a negative errno code.
883 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
887 const int max_netdevices = 8*PAGE_SIZE;
888 unsigned long *inuse;
889 struct net_device *d;
891 p = strnchr(name, IFNAMSIZ-1, '%');
894 * Verify the string as this thing may have come from
895 * the user. There must be either one "%d" and no other "%"
898 if (p[1] != 'd' || strchr(p + 2, '%'))
901 /* Use one page as a bit array of possible slots */
902 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
906 for_each_netdev(net, d) {
907 if (!sscanf(d->name, name, &i))
909 if (i < 0 || i >= max_netdevices)
912 /* avoid cases where sscanf is not exact inverse of printf */
913 snprintf(buf, IFNAMSIZ, name, i);
914 if (!strncmp(buf, d->name, IFNAMSIZ))
918 i = find_first_zero_bit(inuse, max_netdevices);
919 free_page((unsigned long) inuse);
923 snprintf(buf, IFNAMSIZ, name, i);
924 if (!__dev_get_by_name(net, buf))
927 /* It is possible to run out of possible slots
928 * when the name is long and there isn't enough space left
929 * for the digits, or if all bits are used.
935 * dev_alloc_name - allocate a name for a device
937 * @name: name format string
939 * Passed a format string - eg "lt%d" it will try and find a suitable
940 * id. It scans list of devices to build up a free map, then chooses
941 * the first empty slot. The caller must hold the dev_base or rtnl lock
942 * while allocating the name and adding the device in order to avoid
944 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
945 * Returns the number of the unit assigned or a negative errno code.
948 int dev_alloc_name(struct net_device *dev, const char *name)
954 BUG_ON(!dev_net(dev));
956 ret = __dev_alloc_name(net, name, buf);
958 strlcpy(dev->name, buf, IFNAMSIZ);
961 EXPORT_SYMBOL(dev_alloc_name);
963 static int dev_get_valid_name(struct net_device *dev, const char *name)
967 BUG_ON(!dev_net(dev));
970 if (!dev_valid_name(name))
973 if (strchr(name, '%'))
974 return dev_alloc_name(dev, name);
975 else if (__dev_get_by_name(net, name))
977 else if (dev->name != name)
978 strlcpy(dev->name, name, IFNAMSIZ);
984 * dev_change_name - change name of a device
986 * @newname: name (or format string) must be at least IFNAMSIZ
988 * Change name of a device, can pass format strings "eth%d".
991 int dev_change_name(struct net_device *dev, const char *newname)
993 char oldname[IFNAMSIZ];
999 BUG_ON(!dev_net(dev));
1002 if (dev->flags & IFF_UP)
1005 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1008 memcpy(oldname, dev->name, IFNAMSIZ);
1010 err = dev_get_valid_name(dev, newname);
1015 ret = device_rename(&dev->dev, dev->name);
1017 memcpy(dev->name, oldname, IFNAMSIZ);
1021 write_lock_bh(&dev_base_lock);
1022 hlist_del_rcu(&dev->name_hlist);
1023 write_unlock_bh(&dev_base_lock);
1027 write_lock_bh(&dev_base_lock);
1028 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1029 write_unlock_bh(&dev_base_lock);
1031 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1032 ret = notifier_to_errno(ret);
1035 /* err >= 0 after dev_alloc_name() or stores the first errno */
1038 memcpy(dev->name, oldname, IFNAMSIZ);
1041 pr_err("%s: name change rollback failed: %d\n",
1050 * dev_set_alias - change ifalias of a device
1052 * @alias: name up to IFALIASZ
1053 * @len: limit of bytes to copy from info
1055 * Set ifalias for a device,
1057 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1061 if (len >= IFALIASZ)
1066 kfree(dev->ifalias);
1067 dev->ifalias = NULL;
1072 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1076 strlcpy(dev->ifalias, alias, len+1);
1082 * netdev_features_change - device changes features
1083 * @dev: device to cause notification
1085 * Called to indicate a device has changed features.
1087 void netdev_features_change(struct net_device *dev)
1089 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1091 EXPORT_SYMBOL(netdev_features_change);
1094 * netdev_state_change - device changes state
1095 * @dev: device to cause notification
1097 * Called to indicate a device has changed state. This function calls
1098 * the notifier chains for netdev_chain and sends a NEWLINK message
1099 * to the routing socket.
1101 void netdev_state_change(struct net_device *dev)
1103 if (dev->flags & IFF_UP) {
1104 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1105 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1108 EXPORT_SYMBOL(netdev_state_change);
1110 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1112 return call_netdevice_notifiers(event, dev);
1114 EXPORT_SYMBOL(netdev_bonding_change);
1117 * dev_load - load a network module
1118 * @net: the applicable net namespace
1119 * @name: name of interface
1121 * If a network interface is not present and the process has suitable
1122 * privileges this function loads the module. If module loading is not
1123 * available in this kernel then it becomes a nop.
1126 void dev_load(struct net *net, const char *name)
1128 struct net_device *dev;
1132 dev = dev_get_by_name_rcu(net, name);
1136 if (no_module && capable(CAP_NET_ADMIN))
1137 no_module = request_module("netdev-%s", name);
1138 if (no_module && capable(CAP_SYS_MODULE)) {
1139 if (!request_module("%s", name))
1140 pr_err("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1144 EXPORT_SYMBOL(dev_load);
1146 static int __dev_open(struct net_device *dev)
1148 const struct net_device_ops *ops = dev->netdev_ops;
1153 if (!netif_device_present(dev))
1156 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1157 ret = notifier_to_errno(ret);
1161 set_bit(__LINK_STATE_START, &dev->state);
1163 if (ops->ndo_validate_addr)
1164 ret = ops->ndo_validate_addr(dev);
1166 if (!ret && ops->ndo_open)
1167 ret = ops->ndo_open(dev);
1170 clear_bit(__LINK_STATE_START, &dev->state);
1172 dev->flags |= IFF_UP;
1173 net_dmaengine_get();
1174 dev_set_rx_mode(dev);
1176 add_device_randomness(dev->dev_addr, dev->addr_len);
1183 * dev_open - prepare an interface for use.
1184 * @dev: device to open
1186 * Takes a device from down to up state. The device's private open
1187 * function is invoked and then the multicast lists are loaded. Finally
1188 * the device is moved into the up state and a %NETDEV_UP message is
1189 * sent to the netdev notifier chain.
1191 * Calling this function on an active interface is a nop. On a failure
1192 * a negative errno code is returned.
1194 int dev_open(struct net_device *dev)
1198 if (dev->flags & IFF_UP)
1201 ret = __dev_open(dev);
1205 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1206 call_netdevice_notifiers(NETDEV_UP, dev);
1210 EXPORT_SYMBOL(dev_open);
1212 static int __dev_close_many(struct list_head *head)
1214 struct net_device *dev;
1219 list_for_each_entry(dev, head, unreg_list) {
1220 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1222 clear_bit(__LINK_STATE_START, &dev->state);
1224 /* Synchronize to scheduled poll. We cannot touch poll list, it
1225 * can be even on different cpu. So just clear netif_running().
1227 * dev->stop() will invoke napi_disable() on all of it's
1228 * napi_struct instances on this device.
1230 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1233 dev_deactivate_many(head);
1235 list_for_each_entry(dev, head, unreg_list) {
1236 const struct net_device_ops *ops = dev->netdev_ops;
1239 * Call the device specific close. This cannot fail.
1240 * Only if device is UP
1242 * We allow it to be called even after a DETACH hot-plug
1248 dev->flags &= ~IFF_UP;
1249 net_dmaengine_put();
1255 static int __dev_close(struct net_device *dev)
1260 list_add(&dev->unreg_list, &single);
1261 retval = __dev_close_many(&single);
1266 static int dev_close_many(struct list_head *head)
1268 struct net_device *dev, *tmp;
1269 LIST_HEAD(tmp_list);
1271 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1272 if (!(dev->flags & IFF_UP))
1273 list_move(&dev->unreg_list, &tmp_list);
1275 __dev_close_many(head);
1277 list_for_each_entry(dev, head, unreg_list) {
1278 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1279 call_netdevice_notifiers(NETDEV_DOWN, dev);
1282 /* rollback_registered_many needs the complete original list */
1283 list_splice(&tmp_list, head);
1288 * dev_close - shutdown an interface.
1289 * @dev: device to shutdown
1291 * This function moves an active device into down state. A
1292 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1293 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1296 int dev_close(struct net_device *dev)
1298 if (dev->flags & IFF_UP) {
1301 list_add(&dev->unreg_list, &single);
1302 dev_close_many(&single);
1307 EXPORT_SYMBOL(dev_close);
1311 * dev_disable_lro - disable Large Receive Offload on a device
1314 * Disable Large Receive Offload (LRO) on a net device. Must be
1315 * called under RTNL. This is needed if received packets may be
1316 * forwarded to another interface.
1318 void dev_disable_lro(struct net_device *dev)
1321 * If we're trying to disable lro on a vlan device
1322 * use the underlying physical device instead
1324 if (is_vlan_dev(dev))
1325 dev = vlan_dev_real_dev(dev);
1327 dev->wanted_features &= ~NETIF_F_LRO;
1328 netdev_update_features(dev);
1330 if (unlikely(dev->features & NETIF_F_LRO))
1331 netdev_WARN(dev, "failed to disable LRO!\n");
1333 EXPORT_SYMBOL(dev_disable_lro);
1336 static int dev_boot_phase = 1;
1339 * register_netdevice_notifier - register a network notifier block
1342 * Register a notifier to be called when network device events occur.
1343 * The notifier passed is linked into the kernel structures and must
1344 * not be reused until it has been unregistered. A negative errno code
1345 * is returned on a failure.
1347 * When registered all registration and up events are replayed
1348 * to the new notifier to allow device to have a race free
1349 * view of the network device list.
1352 int register_netdevice_notifier(struct notifier_block *nb)
1354 struct net_device *dev;
1355 struct net_device *last;
1360 err = raw_notifier_chain_register(&netdev_chain, nb);
1366 for_each_netdev(net, dev) {
1367 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1368 err = notifier_to_errno(err);
1372 if (!(dev->flags & IFF_UP))
1375 nb->notifier_call(nb, NETDEV_UP, dev);
1386 for_each_netdev(net, dev) {
1390 if (dev->flags & IFF_UP) {
1391 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1392 nb->notifier_call(nb, NETDEV_DOWN, dev);
1394 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1395 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1400 raw_notifier_chain_unregister(&netdev_chain, nb);
1403 EXPORT_SYMBOL(register_netdevice_notifier);
1406 * unregister_netdevice_notifier - unregister a network notifier block
1409 * Unregister a notifier previously registered by
1410 * register_netdevice_notifier(). The notifier is unlinked into the
1411 * kernel structures and may then be reused. A negative errno code
1412 * is returned on a failure.
1414 * After unregistering unregister and down device events are synthesized
1415 * for all devices on the device list to the removed notifier to remove
1416 * the need for special case cleanup code.
1419 int unregister_netdevice_notifier(struct notifier_block *nb)
1421 struct net_device *dev;
1426 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1431 for_each_netdev(net, dev) {
1432 if (dev->flags & IFF_UP) {
1433 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1434 nb->notifier_call(nb, NETDEV_DOWN, dev);
1436 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1437 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1444 EXPORT_SYMBOL(unregister_netdevice_notifier);
1447 * call_netdevice_notifiers - call all network notifier blocks
1448 * @val: value passed unmodified to notifier function
1449 * @dev: net_device pointer passed unmodified to notifier function
1451 * Call all network notifier blocks. Parameters and return value
1452 * are as for raw_notifier_call_chain().
1455 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1458 return raw_notifier_call_chain(&netdev_chain, val, dev);
1460 EXPORT_SYMBOL(call_netdevice_notifiers);
1462 static struct static_key netstamp_needed __read_mostly;
1463 #ifdef HAVE_JUMP_LABEL
1464 /* We are not allowed to call static_key_slow_dec() from irq context
1465 * If net_disable_timestamp() is called from irq context, defer the
1466 * static_key_slow_dec() calls.
1468 static atomic_t netstamp_needed_deferred;
1471 void net_enable_timestamp(void)
1473 #ifdef HAVE_JUMP_LABEL
1474 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1478 static_key_slow_dec(&netstamp_needed);
1482 WARN_ON(in_interrupt());
1483 static_key_slow_inc(&netstamp_needed);
1485 EXPORT_SYMBOL(net_enable_timestamp);
1487 void net_disable_timestamp(void)
1489 #ifdef HAVE_JUMP_LABEL
1490 if (in_interrupt()) {
1491 atomic_inc(&netstamp_needed_deferred);
1495 static_key_slow_dec(&netstamp_needed);
1497 EXPORT_SYMBOL(net_disable_timestamp);
1499 static inline void net_timestamp_set(struct sk_buff *skb)
1501 skb->tstamp.tv64 = 0;
1502 if (static_key_false(&netstamp_needed))
1503 __net_timestamp(skb);
1506 #define net_timestamp_check(COND, SKB) \
1507 if (static_key_false(&netstamp_needed)) { \
1508 if ((COND) && !(SKB)->tstamp.tv64) \
1509 __net_timestamp(SKB); \
1512 static int net_hwtstamp_validate(struct ifreq *ifr)
1514 struct hwtstamp_config cfg;
1515 enum hwtstamp_tx_types tx_type;
1516 enum hwtstamp_rx_filters rx_filter;
1517 int tx_type_valid = 0;
1518 int rx_filter_valid = 0;
1520 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1523 if (cfg.flags) /* reserved for future extensions */
1526 tx_type = cfg.tx_type;
1527 rx_filter = cfg.rx_filter;
1530 case HWTSTAMP_TX_OFF:
1531 case HWTSTAMP_TX_ON:
1532 case HWTSTAMP_TX_ONESTEP_SYNC:
1537 switch (rx_filter) {
1538 case HWTSTAMP_FILTER_NONE:
1539 case HWTSTAMP_FILTER_ALL:
1540 case HWTSTAMP_FILTER_SOME:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1542 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1543 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1545 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1546 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1548 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1549 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1550 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1551 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1552 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1553 rx_filter_valid = 1;
1557 if (!tx_type_valid || !rx_filter_valid)
1563 static inline bool is_skb_forwardable(struct net_device *dev,
1564 struct sk_buff *skb)
1568 if (!(dev->flags & IFF_UP))
1571 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1572 if (skb->len <= len)
1575 /* if TSO is enabled, we don't care about the length as the packet
1576 * could be forwarded without being segmented before
1578 if (skb_is_gso(skb))
1585 * dev_forward_skb - loopback an skb to another netif
1587 * @dev: destination network device
1588 * @skb: buffer to forward
1591 * NET_RX_SUCCESS (no congestion)
1592 * NET_RX_DROP (packet was dropped, but freed)
1594 * dev_forward_skb can be used for injecting an skb from the
1595 * start_xmit function of one device into the receive queue
1596 * of another device.
1598 * The receiving device may be in another namespace, so
1599 * we have to clear all information in the skb that could
1600 * impact namespace isolation.
1602 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1604 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1605 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1606 atomic_long_inc(&dev->rx_dropped);
1615 if (unlikely(!is_skb_forwardable(dev, skb))) {
1616 atomic_long_inc(&dev->rx_dropped);
1623 skb->tstamp.tv64 = 0;
1624 skb->pkt_type = PACKET_HOST;
1625 skb->protocol = eth_type_trans(skb, dev);
1629 return netif_rx(skb);
1631 EXPORT_SYMBOL_GPL(dev_forward_skb);
1633 static inline int deliver_skb(struct sk_buff *skb,
1634 struct packet_type *pt_prev,
1635 struct net_device *orig_dev)
1637 atomic_inc(&skb->users);
1638 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1642 * Support routine. Sends outgoing frames to any network
1643 * taps currently in use.
1646 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1648 struct packet_type *ptype;
1649 struct sk_buff *skb2 = NULL;
1650 struct packet_type *pt_prev = NULL;
1653 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1654 /* Never send packets back to the socket
1655 * they originated from - MvS (miquels@drinkel.ow.org)
1657 if ((ptype->dev == dev || !ptype->dev) &&
1658 (ptype->af_packet_priv == NULL ||
1659 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1661 deliver_skb(skb2, pt_prev, skb->dev);
1666 skb2 = skb_clone(skb, GFP_ATOMIC);
1670 net_timestamp_set(skb2);
1672 /* skb->nh should be correctly
1673 set by sender, so that the second statement is
1674 just protection against buggy protocols.
1676 skb_reset_mac_header(skb2);
1678 if (skb_network_header(skb2) < skb2->data ||
1679 skb2->network_header > skb2->tail) {
1680 if (net_ratelimit())
1681 pr_crit("protocol %04x is buggy, dev %s\n",
1682 ntohs(skb2->protocol),
1684 skb_reset_network_header(skb2);
1687 skb2->transport_header = skb2->network_header;
1688 skb2->pkt_type = PACKET_OUTGOING;
1693 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1697 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1698 * @dev: Network device
1699 * @txq: number of queues available
1701 * If real_num_tx_queues is changed the tc mappings may no longer be
1702 * valid. To resolve this verify the tc mapping remains valid and if
1703 * not NULL the mapping. With no priorities mapping to this
1704 * offset/count pair it will no longer be used. In the worst case TC0
1705 * is invalid nothing can be done so disable priority mappings. If is
1706 * expected that drivers will fix this mapping if they can before
1707 * calling netif_set_real_num_tx_queues.
1709 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1712 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1714 /* If TC0 is invalidated disable TC mapping */
1715 if (tc->offset + tc->count > txq) {
1716 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1721 /* Invalidated prio to tc mappings set to TC0 */
1722 for (i = 1; i < TC_BITMASK + 1; i++) {
1723 int q = netdev_get_prio_tc_map(dev, i);
1725 tc = &dev->tc_to_txq[q];
1726 if (tc->offset + tc->count > txq) {
1727 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1729 netdev_set_prio_tc_map(dev, i, 0);
1735 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1736 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1738 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1742 if (txq < 1 || txq > dev->num_tx_queues)
1745 if (dev->reg_state == NETREG_REGISTERED ||
1746 dev->reg_state == NETREG_UNREGISTERING) {
1749 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1755 netif_setup_tc(dev, txq);
1757 if (txq < dev->real_num_tx_queues)
1758 qdisc_reset_all_tx_gt(dev, txq);
1761 dev->real_num_tx_queues = txq;
1764 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1768 * netif_set_real_num_rx_queues - set actual number of RX queues used
1769 * @dev: Network device
1770 * @rxq: Actual number of RX queues
1772 * This must be called either with the rtnl_lock held or before
1773 * registration of the net device. Returns 0 on success, or a
1774 * negative error code. If called before registration, it always
1777 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1781 if (rxq < 1 || rxq > dev->num_rx_queues)
1784 if (dev->reg_state == NETREG_REGISTERED) {
1787 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1793 dev->real_num_rx_queues = rxq;
1796 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1799 static inline void __netif_reschedule(struct Qdisc *q)
1801 struct softnet_data *sd;
1802 unsigned long flags;
1804 local_irq_save(flags);
1805 sd = &__get_cpu_var(softnet_data);
1806 q->next_sched = NULL;
1807 *sd->output_queue_tailp = q;
1808 sd->output_queue_tailp = &q->next_sched;
1809 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1810 local_irq_restore(flags);
1813 void __netif_schedule(struct Qdisc *q)
1815 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1816 __netif_reschedule(q);
1818 EXPORT_SYMBOL(__netif_schedule);
1820 void dev_kfree_skb_irq(struct sk_buff *skb)
1822 if (atomic_dec_and_test(&skb->users)) {
1823 struct softnet_data *sd;
1824 unsigned long flags;
1826 local_irq_save(flags);
1827 sd = &__get_cpu_var(softnet_data);
1828 skb->next = sd->completion_queue;
1829 sd->completion_queue = skb;
1830 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1831 local_irq_restore(flags);
1834 EXPORT_SYMBOL(dev_kfree_skb_irq);
1836 void dev_kfree_skb_any(struct sk_buff *skb)
1838 if (in_irq() || irqs_disabled())
1839 dev_kfree_skb_irq(skb);
1843 EXPORT_SYMBOL(dev_kfree_skb_any);
1847 * netif_device_detach - mark device as removed
1848 * @dev: network device
1850 * Mark device as removed from system and therefore no longer available.
1852 void netif_device_detach(struct net_device *dev)
1854 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1855 netif_running(dev)) {
1856 netif_tx_stop_all_queues(dev);
1859 EXPORT_SYMBOL(netif_device_detach);
1862 * netif_device_attach - mark device as attached
1863 * @dev: network device
1865 * Mark device as attached from system and restart if needed.
1867 void netif_device_attach(struct net_device *dev)
1869 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1870 netif_running(dev)) {
1871 netif_tx_wake_all_queues(dev);
1872 __netdev_watchdog_up(dev);
1875 EXPORT_SYMBOL(netif_device_attach);
1877 static void skb_warn_bad_offload(const struct sk_buff *skb)
1879 static const netdev_features_t null_features = 0;
1880 struct net_device *dev = skb->dev;
1881 const char *driver = "";
1883 if (dev && dev->dev.parent)
1884 driver = dev_driver_string(dev->dev.parent);
1886 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1887 "gso_type=%d ip_summed=%d\n",
1888 driver, dev ? &dev->features : &null_features,
1889 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1890 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1891 skb_shinfo(skb)->gso_type, skb->ip_summed);
1895 * Invalidate hardware checksum when packet is to be mangled, and
1896 * complete checksum manually on outgoing path.
1898 int skb_checksum_help(struct sk_buff *skb)
1901 int ret = 0, offset;
1903 if (skb->ip_summed == CHECKSUM_COMPLETE)
1904 goto out_set_summed;
1906 if (unlikely(skb_shinfo(skb)->gso_size)) {
1907 skb_warn_bad_offload(skb);
1911 offset = skb_checksum_start_offset(skb);
1912 BUG_ON(offset >= skb_headlen(skb));
1913 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1915 offset += skb->csum_offset;
1916 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1918 if (skb_cloned(skb) &&
1919 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1920 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1925 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1927 skb->ip_summed = CHECKSUM_NONE;
1931 EXPORT_SYMBOL(skb_checksum_help);
1934 * skb_gso_segment - Perform segmentation on skb.
1935 * @skb: buffer to segment
1936 * @features: features for the output path (see dev->features)
1938 * This function segments the given skb and returns a list of segments.
1940 * It may return NULL if the skb requires no segmentation. This is
1941 * only possible when GSO is used for verifying header integrity.
1943 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1944 netdev_features_t features)
1946 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1947 struct packet_type *ptype;
1948 __be16 type = skb->protocol;
1949 int vlan_depth = ETH_HLEN;
1952 while (type == htons(ETH_P_8021Q)) {
1953 struct vlan_hdr *vh;
1955 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1956 return ERR_PTR(-EINVAL);
1958 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1959 type = vh->h_vlan_encapsulated_proto;
1960 vlan_depth += VLAN_HLEN;
1963 skb_reset_mac_header(skb);
1964 skb->mac_len = skb->network_header - skb->mac_header;
1965 __skb_pull(skb, skb->mac_len);
1967 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1968 skb_warn_bad_offload(skb);
1970 if (skb_header_cloned(skb) &&
1971 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1972 return ERR_PTR(err);
1976 list_for_each_entry_rcu(ptype,
1977 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1978 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1979 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1980 err = ptype->gso_send_check(skb);
1981 segs = ERR_PTR(err);
1982 if (err || skb_gso_ok(skb, features))
1984 __skb_push(skb, (skb->data -
1985 skb_network_header(skb)));
1987 segs = ptype->gso_segment(skb, features);
1993 __skb_push(skb, skb->data - skb_mac_header(skb));
1997 EXPORT_SYMBOL(skb_gso_segment);
1999 /* Take action when hardware reception checksum errors are detected. */
2001 void netdev_rx_csum_fault(struct net_device *dev)
2003 if (net_ratelimit()) {
2004 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2008 EXPORT_SYMBOL(netdev_rx_csum_fault);
2011 /* Actually, we should eliminate this check as soon as we know, that:
2012 * 1. IOMMU is present and allows to map all the memory.
2013 * 2. No high memory really exists on this machine.
2016 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2018 #ifdef CONFIG_HIGHMEM
2020 if (!(dev->features & NETIF_F_HIGHDMA)) {
2021 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2022 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2023 if (PageHighMem(skb_frag_page(frag)))
2028 if (PCI_DMA_BUS_IS_PHYS) {
2029 struct device *pdev = dev->dev.parent;
2033 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2034 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2035 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2036 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2045 void (*destructor)(struct sk_buff *skb);
2048 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2050 static void dev_gso_skb_destructor(struct sk_buff *skb)
2052 struct dev_gso_cb *cb;
2055 struct sk_buff *nskb = skb->next;
2057 skb->next = nskb->next;
2060 } while (skb->next);
2062 cb = DEV_GSO_CB(skb);
2064 cb->destructor(skb);
2068 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2069 * @skb: buffer to segment
2070 * @features: device features as applicable to this skb
2072 * This function segments the given skb and stores the list of segments
2075 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2077 struct sk_buff *segs;
2079 segs = skb_gso_segment(skb, features);
2081 /* Verifying header integrity only. */
2086 return PTR_ERR(segs);
2089 DEV_GSO_CB(skb)->destructor = skb->destructor;
2090 skb->destructor = dev_gso_skb_destructor;
2095 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2097 return ((features & NETIF_F_GEN_CSUM) ||
2098 ((features & NETIF_F_V4_CSUM) &&
2099 protocol == htons(ETH_P_IP)) ||
2100 ((features & NETIF_F_V6_CSUM) &&
2101 protocol == htons(ETH_P_IPV6)) ||
2102 ((features & NETIF_F_FCOE_CRC) &&
2103 protocol == htons(ETH_P_FCOE)));
2106 static netdev_features_t harmonize_features(struct sk_buff *skb,
2107 __be16 protocol, netdev_features_t features)
2109 if (!can_checksum_protocol(features, protocol)) {
2110 features &= ~NETIF_F_ALL_CSUM;
2111 features &= ~NETIF_F_SG;
2112 } else if (illegal_highdma(skb->dev, skb)) {
2113 features &= ~NETIF_F_SG;
2119 netdev_features_t netif_skb_features(struct sk_buff *skb)
2121 __be16 protocol = skb->protocol;
2122 netdev_features_t features = skb->dev->features;
2124 if (protocol == htons(ETH_P_8021Q)) {
2125 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2126 protocol = veh->h_vlan_encapsulated_proto;
2127 } else if (!vlan_tx_tag_present(skb)) {
2128 return harmonize_features(skb, protocol, features);
2131 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2133 if (protocol != htons(ETH_P_8021Q)) {
2134 return harmonize_features(skb, protocol, features);
2136 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2137 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2138 return harmonize_features(skb, protocol, features);
2141 EXPORT_SYMBOL(netif_skb_features);
2144 * Returns true if either:
2145 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2146 * 2. skb is fragmented and the device does not support SG, or if
2147 * at least one of fragments is in highmem and device does not
2148 * support DMA from it.
2150 static inline int skb_needs_linearize(struct sk_buff *skb,
2153 return skb_is_nonlinear(skb) &&
2154 ((skb_has_frag_list(skb) &&
2155 !(features & NETIF_F_FRAGLIST)) ||
2156 (skb_shinfo(skb)->nr_frags &&
2157 !(features & NETIF_F_SG)));
2160 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2161 struct netdev_queue *txq)
2163 const struct net_device_ops *ops = dev->netdev_ops;
2164 int rc = NETDEV_TX_OK;
2165 unsigned int skb_len;
2167 if (likely(!skb->next)) {
2168 netdev_features_t features;
2171 * If device doesn't need skb->dst, release it right now while
2172 * its hot in this cpu cache
2174 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2177 if (!list_empty(&ptype_all))
2178 dev_queue_xmit_nit(skb, dev);
2180 features = netif_skb_features(skb);
2182 if (vlan_tx_tag_present(skb) &&
2183 !(features & NETIF_F_HW_VLAN_TX)) {
2184 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2191 if (netif_needs_gso(skb, features)) {
2192 if (unlikely(dev_gso_segment(skb, features)))
2197 if (skb_needs_linearize(skb, features) &&
2198 __skb_linearize(skb))
2201 /* If packet is not checksummed and device does not
2202 * support checksumming for this protocol, complete
2203 * checksumming here.
2205 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2206 skb_set_transport_header(skb,
2207 skb_checksum_start_offset(skb));
2208 if (!(features & NETIF_F_ALL_CSUM) &&
2209 skb_checksum_help(skb))
2215 rc = ops->ndo_start_xmit(skb, dev);
2216 trace_net_dev_xmit(skb, rc, dev, skb_len);
2217 if (rc == NETDEV_TX_OK)
2218 txq_trans_update(txq);
2224 struct sk_buff *nskb = skb->next;
2226 skb->next = nskb->next;
2230 * If device doesn't need nskb->dst, release it right now while
2231 * its hot in this cpu cache
2233 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2236 skb_len = nskb->len;
2237 rc = ops->ndo_start_xmit(nskb, dev);
2238 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2239 if (unlikely(rc != NETDEV_TX_OK)) {
2240 if (rc & ~NETDEV_TX_MASK)
2241 goto out_kfree_gso_skb;
2242 nskb->next = skb->next;
2246 txq_trans_update(txq);
2247 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2248 return NETDEV_TX_BUSY;
2249 } while (skb->next);
2252 if (likely(skb->next == NULL))
2253 skb->destructor = DEV_GSO_CB(skb)->destructor;
2260 static u32 hashrnd __read_mostly;
2263 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2264 * to be used as a distribution range.
2266 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2267 unsigned int num_tx_queues)
2271 u16 qcount = num_tx_queues;
2273 if (skb_rx_queue_recorded(skb)) {
2274 hash = skb_get_rx_queue(skb);
2275 while (unlikely(hash >= num_tx_queues))
2276 hash -= num_tx_queues;
2281 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2282 qoffset = dev->tc_to_txq[tc].offset;
2283 qcount = dev->tc_to_txq[tc].count;
2286 if (skb->sk && skb->sk->sk_hash)
2287 hash = skb->sk->sk_hash;
2289 hash = (__force u16) skb->protocol;
2290 hash = jhash_1word(hash, hashrnd);
2292 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2294 EXPORT_SYMBOL(__skb_tx_hash);
2296 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2298 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2299 if (net_ratelimit()) {
2300 pr_warn("%s selects TX queue %d, but real number of TX queues is %d\n",
2301 dev->name, queue_index,
2302 dev->real_num_tx_queues);
2309 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2312 struct xps_dev_maps *dev_maps;
2313 struct xps_map *map;
2314 int queue_index = -1;
2317 dev_maps = rcu_dereference(dev->xps_maps);
2319 map = rcu_dereference(
2320 dev_maps->cpu_map[raw_smp_processor_id()]);
2323 queue_index = map->queues[0];
2326 if (skb->sk && skb->sk->sk_hash)
2327 hash = skb->sk->sk_hash;
2329 hash = (__force u16) skb->protocol ^
2331 hash = jhash_1word(hash, hashrnd);
2332 queue_index = map->queues[
2333 ((u64)hash * map->len) >> 32];
2335 if (unlikely(queue_index >= dev->real_num_tx_queues))
2347 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2348 struct sk_buff *skb)
2351 const struct net_device_ops *ops = dev->netdev_ops;
2353 if (dev->real_num_tx_queues == 1)
2355 else if (ops->ndo_select_queue) {
2356 queue_index = ops->ndo_select_queue(dev, skb);
2357 queue_index = dev_cap_txqueue(dev, queue_index);
2359 struct sock *sk = skb->sk;
2360 queue_index = sk_tx_queue_get(sk);
2362 if (queue_index < 0 || skb->ooo_okay ||
2363 queue_index >= dev->real_num_tx_queues) {
2364 int old_index = queue_index;
2366 queue_index = get_xps_queue(dev, skb);
2367 if (queue_index < 0)
2368 queue_index = skb_tx_hash(dev, skb);
2370 if (queue_index != old_index && sk) {
2371 struct dst_entry *dst =
2372 rcu_dereference_check(sk->sk_dst_cache, 1);
2374 if (dst && skb_dst(skb) == dst)
2375 sk_tx_queue_set(sk, queue_index);
2380 skb_set_queue_mapping(skb, queue_index);
2381 return netdev_get_tx_queue(dev, queue_index);
2384 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2385 struct net_device *dev,
2386 struct netdev_queue *txq)
2388 spinlock_t *root_lock = qdisc_lock(q);
2392 qdisc_skb_cb(skb)->pkt_len = skb->len;
2393 qdisc_calculate_pkt_len(skb, q);
2395 * Heuristic to force contended enqueues to serialize on a
2396 * separate lock before trying to get qdisc main lock.
2397 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2398 * and dequeue packets faster.
2400 contended = qdisc_is_running(q);
2401 if (unlikely(contended))
2402 spin_lock(&q->busylock);
2404 spin_lock(root_lock);
2405 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2408 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2409 qdisc_run_begin(q)) {
2411 * This is a work-conserving queue; there are no old skbs
2412 * waiting to be sent out; and the qdisc is not running -
2413 * xmit the skb directly.
2415 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2418 qdisc_bstats_update(q, skb);
2420 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2421 if (unlikely(contended)) {
2422 spin_unlock(&q->busylock);
2429 rc = NET_XMIT_SUCCESS;
2432 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2433 if (qdisc_run_begin(q)) {
2434 if (unlikely(contended)) {
2435 spin_unlock(&q->busylock);
2441 spin_unlock(root_lock);
2442 if (unlikely(contended))
2443 spin_unlock(&q->busylock);
2447 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2448 static void skb_update_prio(struct sk_buff *skb)
2450 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2452 if ((!skb->priority) && (skb->sk) && map)
2453 skb->priority = map->priomap[skb->sk->sk_cgrp_prioidx];
2456 #define skb_update_prio(skb)
2459 static DEFINE_PER_CPU(int, xmit_recursion);
2460 #define RECURSION_LIMIT 10
2463 * dev_queue_xmit - transmit a buffer
2464 * @skb: buffer to transmit
2466 * Queue a buffer for transmission to a network device. The caller must
2467 * have set the device and priority and built the buffer before calling
2468 * this function. The function can be called from an interrupt.
2470 * A negative errno code is returned on a failure. A success does not
2471 * guarantee the frame will be transmitted as it may be dropped due
2472 * to congestion or traffic shaping.
2474 * -----------------------------------------------------------------------------------
2475 * I notice this method can also return errors from the queue disciplines,
2476 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2479 * Regardless of the return value, the skb is consumed, so it is currently
2480 * difficult to retry a send to this method. (You can bump the ref count
2481 * before sending to hold a reference for retry if you are careful.)
2483 * When calling this method, interrupts MUST be enabled. This is because
2484 * the BH enable code must have IRQs enabled so that it will not deadlock.
2487 int dev_queue_xmit(struct sk_buff *skb)
2489 struct net_device *dev = skb->dev;
2490 struct netdev_queue *txq;
2494 /* Disable soft irqs for various locks below. Also
2495 * stops preemption for RCU.
2499 skb_update_prio(skb);
2501 txq = dev_pick_tx(dev, skb);
2502 q = rcu_dereference_bh(txq->qdisc);
2504 #ifdef CONFIG_NET_CLS_ACT
2505 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2507 trace_net_dev_queue(skb);
2509 rc = __dev_xmit_skb(skb, q, dev, txq);
2513 /* The device has no queue. Common case for software devices:
2514 loopback, all the sorts of tunnels...
2516 Really, it is unlikely that netif_tx_lock protection is necessary
2517 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2519 However, it is possible, that they rely on protection
2522 Check this and shot the lock. It is not prone from deadlocks.
2523 Either shot noqueue qdisc, it is even simpler 8)
2525 if (dev->flags & IFF_UP) {
2526 int cpu = smp_processor_id(); /* ok because BHs are off */
2528 if (txq->xmit_lock_owner != cpu) {
2530 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2531 goto recursion_alert;
2533 HARD_TX_LOCK(dev, txq, cpu);
2535 if (!netif_xmit_stopped(txq)) {
2536 __this_cpu_inc(xmit_recursion);
2537 rc = dev_hard_start_xmit(skb, dev, txq);
2538 __this_cpu_dec(xmit_recursion);
2539 if (dev_xmit_complete(rc)) {
2540 HARD_TX_UNLOCK(dev, txq);
2544 HARD_TX_UNLOCK(dev, txq);
2545 if (net_ratelimit())
2546 pr_crit("Virtual device %s asks to queue packet!\n",
2549 /* Recursion is detected! It is possible,
2553 if (net_ratelimit())
2554 pr_crit("Dead loop on virtual device %s, fix it urgently!\n",
2560 rcu_read_unlock_bh();
2565 rcu_read_unlock_bh();
2568 EXPORT_SYMBOL(dev_queue_xmit);
2571 /*=======================================================================
2573 =======================================================================*/
2575 int netdev_max_backlog __read_mostly = 1000;
2576 int netdev_tstamp_prequeue __read_mostly = 1;
2577 int netdev_budget __read_mostly = 300;
2578 int weight_p __read_mostly = 64; /* old backlog weight */
2580 /* Called with irq disabled */
2581 static inline void ____napi_schedule(struct softnet_data *sd,
2582 struct napi_struct *napi)
2584 list_add_tail(&napi->poll_list, &sd->poll_list);
2585 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2589 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2590 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2591 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2592 * if hash is a canonical 4-tuple hash over transport ports.
2594 void __skb_get_rxhash(struct sk_buff *skb)
2596 struct flow_keys keys;
2599 if (!skb_flow_dissect(skb, &keys))
2603 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2604 swap(keys.port16[0], keys.port16[1]);
2608 /* get a consistent hash (same value on both flow directions) */
2609 if ((__force u32)keys.dst < (__force u32)keys.src)
2610 swap(keys.dst, keys.src);
2612 hash = jhash_3words((__force u32)keys.dst,
2613 (__force u32)keys.src,
2614 (__force u32)keys.ports, hashrnd);
2620 EXPORT_SYMBOL(__skb_get_rxhash);
2624 /* One global table that all flow-based protocols share. */
2625 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2626 EXPORT_SYMBOL(rps_sock_flow_table);
2628 struct static_key rps_needed __read_mostly;
2630 static struct rps_dev_flow *
2631 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2632 struct rps_dev_flow *rflow, u16 next_cpu)
2634 if (next_cpu != RPS_NO_CPU) {
2635 #ifdef CONFIG_RFS_ACCEL
2636 struct netdev_rx_queue *rxqueue;
2637 struct rps_dev_flow_table *flow_table;
2638 struct rps_dev_flow *old_rflow;
2643 /* Should we steer this flow to a different hardware queue? */
2644 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2645 !(dev->features & NETIF_F_NTUPLE))
2647 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2648 if (rxq_index == skb_get_rx_queue(skb))
2651 rxqueue = dev->_rx + rxq_index;
2652 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2655 flow_id = skb->rxhash & flow_table->mask;
2656 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2657 rxq_index, flow_id);
2661 rflow = &flow_table->flows[flow_id];
2663 if (old_rflow->filter == rflow->filter)
2664 old_rflow->filter = RPS_NO_FILTER;
2668 per_cpu(softnet_data, next_cpu).input_queue_head;
2671 rflow->cpu = next_cpu;
2676 * get_rps_cpu is called from netif_receive_skb and returns the target
2677 * CPU from the RPS map of the receiving queue for a given skb.
2678 * rcu_read_lock must be held on entry.
2680 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2681 struct rps_dev_flow **rflowp)
2683 struct netdev_rx_queue *rxqueue;
2684 struct rps_map *map;
2685 struct rps_dev_flow_table *flow_table;
2686 struct rps_sock_flow_table *sock_flow_table;
2690 if (skb_rx_queue_recorded(skb)) {
2691 u16 index = skb_get_rx_queue(skb);
2692 if (unlikely(index >= dev->real_num_rx_queues)) {
2693 WARN_ONCE(dev->real_num_rx_queues > 1,
2694 "%s received packet on queue %u, but number "
2695 "of RX queues is %u\n",
2696 dev->name, index, dev->real_num_rx_queues);
2699 rxqueue = dev->_rx + index;
2703 map = rcu_dereference(rxqueue->rps_map);
2705 if (map->len == 1 &&
2706 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2707 tcpu = map->cpus[0];
2708 if (cpu_online(tcpu))
2712 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2716 skb_reset_network_header(skb);
2717 if (!skb_get_rxhash(skb))
2720 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2721 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2722 if (flow_table && sock_flow_table) {
2724 struct rps_dev_flow *rflow;
2726 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2729 next_cpu = sock_flow_table->ents[skb->rxhash &
2730 sock_flow_table->mask];
2733 * If the desired CPU (where last recvmsg was done) is
2734 * different from current CPU (one in the rx-queue flow
2735 * table entry), switch if one of the following holds:
2736 * - Current CPU is unset (equal to RPS_NO_CPU).
2737 * - Current CPU is offline.
2738 * - The current CPU's queue tail has advanced beyond the
2739 * last packet that was enqueued using this table entry.
2740 * This guarantees that all previous packets for the flow
2741 * have been dequeued, thus preserving in order delivery.
2743 if (unlikely(tcpu != next_cpu) &&
2744 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2745 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2746 rflow->last_qtail)) >= 0))
2747 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2749 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2757 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2759 if (cpu_online(tcpu)) {
2769 #ifdef CONFIG_RFS_ACCEL
2772 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2773 * @dev: Device on which the filter was set
2774 * @rxq_index: RX queue index
2775 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2776 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2778 * Drivers that implement ndo_rx_flow_steer() should periodically call
2779 * this function for each installed filter and remove the filters for
2780 * which it returns %true.
2782 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2783 u32 flow_id, u16 filter_id)
2785 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2786 struct rps_dev_flow_table *flow_table;
2787 struct rps_dev_flow *rflow;
2792 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2793 if (flow_table && flow_id <= flow_table->mask) {
2794 rflow = &flow_table->flows[flow_id];
2795 cpu = ACCESS_ONCE(rflow->cpu);
2796 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2797 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2798 rflow->last_qtail) <
2799 (int)(10 * flow_table->mask)))
2805 EXPORT_SYMBOL(rps_may_expire_flow);
2807 #endif /* CONFIG_RFS_ACCEL */
2809 /* Called from hardirq (IPI) context */
2810 static void rps_trigger_softirq(void *data)
2812 struct softnet_data *sd = data;
2814 ____napi_schedule(sd, &sd->backlog);
2818 #endif /* CONFIG_RPS */
2821 * Check if this softnet_data structure is another cpu one
2822 * If yes, queue it to our IPI list and return 1
2825 static int rps_ipi_queued(struct softnet_data *sd)
2828 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2831 sd->rps_ipi_next = mysd->rps_ipi_list;
2832 mysd->rps_ipi_list = sd;
2834 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2837 #endif /* CONFIG_RPS */
2842 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2843 * queue (may be a remote CPU queue).
2845 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2846 unsigned int *qtail)
2848 struct softnet_data *sd;
2849 unsigned long flags;
2851 sd = &per_cpu(softnet_data, cpu);
2853 local_irq_save(flags);
2856 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2857 if (skb_queue_len(&sd->input_pkt_queue)) {
2859 __skb_queue_tail(&sd->input_pkt_queue, skb);
2860 input_queue_tail_incr_save(sd, qtail);
2862 local_irq_restore(flags);
2863 return NET_RX_SUCCESS;
2866 /* Schedule NAPI for backlog device
2867 * We can use non atomic operation since we own the queue lock
2869 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2870 if (!rps_ipi_queued(sd))
2871 ____napi_schedule(sd, &sd->backlog);
2879 local_irq_restore(flags);
2881 atomic_long_inc(&skb->dev->rx_dropped);
2887 * netif_rx - post buffer to the network code
2888 * @skb: buffer to post
2890 * This function receives a packet from a device driver and queues it for
2891 * the upper (protocol) levels to process. It always succeeds. The buffer
2892 * may be dropped during processing for congestion control or by the
2896 * NET_RX_SUCCESS (no congestion)
2897 * NET_RX_DROP (packet was dropped)
2901 int netif_rx(struct sk_buff *skb)
2905 /* if netpoll wants it, pretend we never saw it */
2906 if (netpoll_rx(skb))
2909 net_timestamp_check(netdev_tstamp_prequeue, skb);
2911 trace_netif_rx(skb);
2913 if (static_key_false(&rps_needed)) {
2914 struct rps_dev_flow voidflow, *rflow = &voidflow;
2920 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2922 cpu = smp_processor_id();
2924 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2932 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2937 EXPORT_SYMBOL(netif_rx);
2939 int netif_rx_ni(struct sk_buff *skb)
2944 err = netif_rx(skb);
2945 if (local_softirq_pending())
2951 EXPORT_SYMBOL(netif_rx_ni);
2953 static void net_tx_action(struct softirq_action *h)
2955 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2957 if (sd->completion_queue) {
2958 struct sk_buff *clist;
2960 local_irq_disable();
2961 clist = sd->completion_queue;
2962 sd->completion_queue = NULL;
2966 struct sk_buff *skb = clist;
2967 clist = clist->next;
2969 WARN_ON(atomic_read(&skb->users));
2970 trace_kfree_skb(skb, net_tx_action);
2975 if (sd->output_queue) {
2978 local_irq_disable();
2979 head = sd->output_queue;
2980 sd->output_queue = NULL;
2981 sd->output_queue_tailp = &sd->output_queue;
2985 struct Qdisc *q = head;
2986 spinlock_t *root_lock;
2988 head = head->next_sched;
2990 root_lock = qdisc_lock(q);
2991 if (spin_trylock(root_lock)) {
2992 smp_mb__before_clear_bit();
2993 clear_bit(__QDISC_STATE_SCHED,
2996 spin_unlock(root_lock);
2998 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3000 __netif_reschedule(q);
3002 smp_mb__before_clear_bit();
3003 clear_bit(__QDISC_STATE_SCHED,
3011 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3012 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3013 /* This hook is defined here for ATM LANE */
3014 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3015 unsigned char *addr) __read_mostly;
3016 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3019 #ifdef CONFIG_NET_CLS_ACT
3020 /* TODO: Maybe we should just force sch_ingress to be compiled in
3021 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3022 * a compare and 2 stores extra right now if we dont have it on
3023 * but have CONFIG_NET_CLS_ACT
3024 * NOTE: This doesn't stop any functionality; if you dont have
3025 * the ingress scheduler, you just can't add policies on ingress.
3028 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3030 struct net_device *dev = skb->dev;
3031 u32 ttl = G_TC_RTTL(skb->tc_verd);
3032 int result = TC_ACT_OK;
3035 if (unlikely(MAX_RED_LOOP < ttl++)) {
3036 if (net_ratelimit())
3037 pr_warn("Redir loop detected Dropping packet (%d->%d)\n",
3038 skb->skb_iif, dev->ifindex);
3042 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3043 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3046 if (q != &noop_qdisc) {
3047 spin_lock(qdisc_lock(q));
3048 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3049 result = qdisc_enqueue_root(skb, q);
3050 spin_unlock(qdisc_lock(q));
3056 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3057 struct packet_type **pt_prev,
3058 int *ret, struct net_device *orig_dev)
3060 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3062 if (!rxq || rxq->qdisc == &noop_qdisc)
3066 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3070 switch (ing_filter(skb, rxq)) {
3084 * netdev_rx_handler_register - register receive handler
3085 * @dev: device to register a handler for
3086 * @rx_handler: receive handler to register
3087 * @rx_handler_data: data pointer that is used by rx handler
3089 * Register a receive hander for a device. This handler will then be
3090 * called from __netif_receive_skb. A negative errno code is returned
3093 * The caller must hold the rtnl_mutex.
3095 * For a general description of rx_handler, see enum rx_handler_result.
3097 int netdev_rx_handler_register(struct net_device *dev,
3098 rx_handler_func_t *rx_handler,
3099 void *rx_handler_data)
3103 if (dev->rx_handler)
3106 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3107 rcu_assign_pointer(dev->rx_handler, rx_handler);
3111 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3114 * netdev_rx_handler_unregister - unregister receive handler
3115 * @dev: device to unregister a handler from
3117 * Unregister a receive hander from a device.
3119 * The caller must hold the rtnl_mutex.
3121 void netdev_rx_handler_unregister(struct net_device *dev)
3125 RCU_INIT_POINTER(dev->rx_handler, NULL);
3126 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3128 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3130 static int __netif_receive_skb(struct sk_buff *skb)
3132 struct packet_type *ptype, *pt_prev;
3133 rx_handler_func_t *rx_handler;
3134 struct net_device *orig_dev;
3135 struct net_device *null_or_dev;
3136 bool deliver_exact = false;
3137 int ret = NET_RX_DROP;
3140 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3142 trace_netif_receive_skb(skb);
3144 /* if we've gotten here through NAPI, check netpoll */
3145 if (netpoll_receive_skb(skb))
3149 skb->skb_iif = skb->dev->ifindex;
3150 orig_dev = skb->dev;
3152 skb_reset_network_header(skb);
3153 skb_reset_transport_header(skb);
3154 skb_reset_mac_len(skb);
3162 __this_cpu_inc(softnet_data.processed);
3164 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3165 skb = vlan_untag(skb);
3170 #ifdef CONFIG_NET_CLS_ACT
3171 if (skb->tc_verd & TC_NCLS) {
3172 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3177 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3178 if (!ptype->dev || ptype->dev == skb->dev) {
3180 ret = deliver_skb(skb, pt_prev, orig_dev);
3185 #ifdef CONFIG_NET_CLS_ACT
3186 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3192 rx_handler = rcu_dereference(skb->dev->rx_handler);
3193 if (vlan_tx_tag_present(skb)) {
3195 ret = deliver_skb(skb, pt_prev, orig_dev);
3198 if (vlan_do_receive(&skb, !rx_handler))
3200 else if (unlikely(!skb))
3206 ret = deliver_skb(skb, pt_prev, orig_dev);
3209 switch (rx_handler(&skb)) {
3210 case RX_HANDLER_CONSUMED:
3212 case RX_HANDLER_ANOTHER:
3214 case RX_HANDLER_EXACT:
3215 deliver_exact = true;
3216 case RX_HANDLER_PASS:
3223 /* deliver only exact match when indicated */
3224 null_or_dev = deliver_exact ? skb->dev : NULL;
3226 type = skb->protocol;
3227 list_for_each_entry_rcu(ptype,
3228 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3229 if (ptype->type == type &&
3230 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3231 ptype->dev == orig_dev)) {
3233 ret = deliver_skb(skb, pt_prev, orig_dev);
3239 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3241 atomic_long_inc(&skb->dev->rx_dropped);
3243 /* Jamal, now you will not able to escape explaining
3244 * me how you were going to use this. :-)
3255 * netif_receive_skb - process receive buffer from network
3256 * @skb: buffer to process
3258 * netif_receive_skb() is the main receive data processing function.
3259 * It always succeeds. The buffer may be dropped during processing
3260 * for congestion control or by the protocol layers.
3262 * This function may only be called from softirq context and interrupts
3263 * should be enabled.
3265 * Return values (usually ignored):
3266 * NET_RX_SUCCESS: no congestion
3267 * NET_RX_DROP: packet was dropped
3269 int netif_receive_skb(struct sk_buff *skb)
3271 net_timestamp_check(netdev_tstamp_prequeue, skb);
3273 if (skb_defer_rx_timestamp(skb))
3274 return NET_RX_SUCCESS;
3277 if (static_key_false(&rps_needed)) {
3278 struct rps_dev_flow voidflow, *rflow = &voidflow;
3283 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3286 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3293 return __netif_receive_skb(skb);
3295 EXPORT_SYMBOL(netif_receive_skb);
3297 /* Network device is going away, flush any packets still pending
3298 * Called with irqs disabled.
3300 static void flush_backlog(void *arg)
3302 struct net_device *dev = arg;
3303 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3304 struct sk_buff *skb, *tmp;
3307 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3308 if (skb->dev == dev) {
3309 __skb_unlink(skb, &sd->input_pkt_queue);
3311 input_queue_head_incr(sd);
3316 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3317 if (skb->dev == dev) {
3318 __skb_unlink(skb, &sd->process_queue);
3320 input_queue_head_incr(sd);
3325 static int napi_gro_complete(struct sk_buff *skb)
3327 struct packet_type *ptype;
3328 __be16 type = skb->protocol;
3329 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3332 if (NAPI_GRO_CB(skb)->count == 1) {
3333 skb_shinfo(skb)->gso_size = 0;
3338 list_for_each_entry_rcu(ptype, head, list) {
3339 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3342 err = ptype->gro_complete(skb);
3348 WARN_ON(&ptype->list == head);
3350 return NET_RX_SUCCESS;
3354 return netif_receive_skb(skb);
3357 inline void napi_gro_flush(struct napi_struct *napi)
3359 struct sk_buff *skb, *next;
3361 for (skb = napi->gro_list; skb; skb = next) {
3364 napi_gro_complete(skb);
3367 napi->gro_count = 0;
3368 napi->gro_list = NULL;
3370 EXPORT_SYMBOL(napi_gro_flush);
3372 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3374 struct sk_buff **pp = NULL;
3375 struct packet_type *ptype;
3376 __be16 type = skb->protocol;
3377 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3380 enum gro_result ret;
3382 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3385 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3389 list_for_each_entry_rcu(ptype, head, list) {
3390 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3393 skb_set_network_header(skb, skb_gro_offset(skb));
3394 mac_len = skb->network_header - skb->mac_header;
3395 skb->mac_len = mac_len;
3396 NAPI_GRO_CB(skb)->same_flow = 0;
3397 NAPI_GRO_CB(skb)->flush = 0;
3398 NAPI_GRO_CB(skb)->free = 0;
3400 pp = ptype->gro_receive(&napi->gro_list, skb);
3405 if (&ptype->list == head)
3408 same_flow = NAPI_GRO_CB(skb)->same_flow;
3409 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3412 struct sk_buff *nskb = *pp;
3416 napi_gro_complete(nskb);
3423 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3427 NAPI_GRO_CB(skb)->count = 1;
3428 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3429 skb->next = napi->gro_list;
3430 napi->gro_list = skb;
3434 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3435 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3437 BUG_ON(skb->end - skb->tail < grow);
3439 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3442 skb->data_len -= grow;
3444 skb_shinfo(skb)->frags[0].page_offset += grow;
3445 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3447 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3448 skb_frag_unref(skb, 0);
3449 memmove(skb_shinfo(skb)->frags,
3450 skb_shinfo(skb)->frags + 1,
3451 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3462 EXPORT_SYMBOL(dev_gro_receive);
3464 static inline gro_result_t
3465 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3468 unsigned int maclen = skb->dev->hard_header_len;
3470 for (p = napi->gro_list; p; p = p->next) {
3471 unsigned long diffs;
3473 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3474 diffs |= p->vlan_tci ^ skb->vlan_tci;
3475 if (maclen == ETH_HLEN)
3476 diffs |= compare_ether_header(skb_mac_header(p),
3477 skb_gro_mac_header(skb));
3479 diffs = memcmp(skb_mac_header(p),
3480 skb_gro_mac_header(skb),
3482 NAPI_GRO_CB(p)->same_flow = !diffs;
3483 NAPI_GRO_CB(p)->flush = 0;
3486 return dev_gro_receive(napi, skb);
3489 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3493 if (netif_receive_skb(skb))
3498 case GRO_MERGED_FREE:
3509 EXPORT_SYMBOL(napi_skb_finish);
3511 void skb_gro_reset_offset(struct sk_buff *skb)
3513 NAPI_GRO_CB(skb)->data_offset = 0;
3514 NAPI_GRO_CB(skb)->frag0 = NULL;
3515 NAPI_GRO_CB(skb)->frag0_len = 0;
3517 if (skb->mac_header == skb->tail &&
3518 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3519 NAPI_GRO_CB(skb)->frag0 =
3520 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3521 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3524 EXPORT_SYMBOL(skb_gro_reset_offset);
3526 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3528 skb_gro_reset_offset(skb);
3530 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3532 EXPORT_SYMBOL(napi_gro_receive);
3534 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3536 __skb_pull(skb, skb_headlen(skb));
3537 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3538 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3540 skb->dev = napi->dev;
3546 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3548 struct sk_buff *skb = napi->skb;
3551 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3557 EXPORT_SYMBOL(napi_get_frags);
3559 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3565 skb->protocol = eth_type_trans(skb, skb->dev);
3567 if (ret == GRO_HELD)
3568 skb_gro_pull(skb, -ETH_HLEN);
3569 else if (netif_receive_skb(skb))
3574 case GRO_MERGED_FREE:
3575 napi_reuse_skb(napi, skb);
3584 EXPORT_SYMBOL(napi_frags_finish);
3586 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3588 struct sk_buff *skb = napi->skb;
3595 skb_reset_mac_header(skb);
3596 skb_gro_reset_offset(skb);
3598 off = skb_gro_offset(skb);
3599 hlen = off + sizeof(*eth);
3600 eth = skb_gro_header_fast(skb, off);
3601 if (skb_gro_header_hard(skb, hlen)) {
3602 eth = skb_gro_header_slow(skb, hlen, off);
3603 if (unlikely(!eth)) {
3604 napi_reuse_skb(napi, skb);
3610 skb_gro_pull(skb, sizeof(*eth));
3613 * This works because the only protocols we care about don't require
3614 * special handling. We'll fix it up properly at the end.
3616 skb->protocol = eth->h_proto;
3621 EXPORT_SYMBOL(napi_frags_skb);
3623 gro_result_t napi_gro_frags(struct napi_struct *napi)
3625 struct sk_buff *skb = napi_frags_skb(napi);
3630 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3632 EXPORT_SYMBOL(napi_gro_frags);
3635 * net_rps_action sends any pending IPI's for rps.
3636 * Note: called with local irq disabled, but exits with local irq enabled.
3638 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3641 struct softnet_data *remsd = sd->rps_ipi_list;
3644 sd->rps_ipi_list = NULL;
3648 /* Send pending IPI's to kick RPS processing on remote cpus. */
3650 struct softnet_data *next = remsd->rps_ipi_next;
3652 if (cpu_online(remsd->cpu))
3653 __smp_call_function_single(remsd->cpu,
3662 static int process_backlog(struct napi_struct *napi, int quota)
3665 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3668 /* Check if we have pending ipi, its better to send them now,
3669 * not waiting net_rx_action() end.
3671 if (sd->rps_ipi_list) {
3672 local_irq_disable();
3673 net_rps_action_and_irq_enable(sd);
3676 napi->weight = weight_p;
3677 local_irq_disable();
3678 while (work < quota) {
3679 struct sk_buff *skb;
3682 while ((skb = __skb_dequeue(&sd->process_queue))) {
3684 __netif_receive_skb(skb);
3685 local_irq_disable();
3686 input_queue_head_incr(sd);
3687 if (++work >= quota) {
3694 qlen = skb_queue_len(&sd->input_pkt_queue);
3696 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3697 &sd->process_queue);
3699 if (qlen < quota - work) {
3701 * Inline a custom version of __napi_complete().
3702 * only current cpu owns and manipulates this napi,
3703 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3704 * we can use a plain write instead of clear_bit(),
3705 * and we dont need an smp_mb() memory barrier.
3707 list_del(&napi->poll_list);
3710 quota = work + qlen;
3720 * __napi_schedule - schedule for receive
3721 * @n: entry to schedule
3723 * The entry's receive function will be scheduled to run
3725 void __napi_schedule(struct napi_struct *n)
3727 unsigned long flags;
3729 local_irq_save(flags);
3730 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3731 local_irq_restore(flags);
3733 EXPORT_SYMBOL(__napi_schedule);
3735 void __napi_complete(struct napi_struct *n)
3737 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3738 BUG_ON(n->gro_list);
3740 list_del(&n->poll_list);
3741 smp_mb__before_clear_bit();
3742 clear_bit(NAPI_STATE_SCHED, &n->state);
3744 EXPORT_SYMBOL(__napi_complete);
3746 void napi_complete(struct napi_struct *n)
3748 unsigned long flags;
3751 * don't let napi dequeue from the cpu poll list
3752 * just in case its running on a different cpu
3754 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3758 local_irq_save(flags);
3760 local_irq_restore(flags);
3762 EXPORT_SYMBOL(napi_complete);
3764 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3765 int (*poll)(struct napi_struct *, int), int weight)
3767 INIT_LIST_HEAD(&napi->poll_list);
3768 napi->gro_count = 0;
3769 napi->gro_list = NULL;
3772 napi->weight = weight;
3773 list_add(&napi->dev_list, &dev->napi_list);
3775 #ifdef CONFIG_NETPOLL
3776 spin_lock_init(&napi->poll_lock);
3777 napi->poll_owner = -1;
3779 set_bit(NAPI_STATE_SCHED, &napi->state);
3781 EXPORT_SYMBOL(netif_napi_add);
3783 void netif_napi_del(struct napi_struct *napi)
3785 struct sk_buff *skb, *next;
3787 list_del_init(&napi->dev_list);
3788 napi_free_frags(napi);
3790 for (skb = napi->gro_list; skb; skb = next) {
3796 napi->gro_list = NULL;
3797 napi->gro_count = 0;
3799 EXPORT_SYMBOL(netif_napi_del);
3801 static void net_rx_action(struct softirq_action *h)
3803 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3804 unsigned long time_limit = jiffies + 2;
3805 int budget = netdev_budget;
3808 local_irq_disable();
3810 while (!list_empty(&sd->poll_list)) {
3811 struct napi_struct *n;
3814 /* If softirq window is exhuasted then punt.
3815 * Allow this to run for 2 jiffies since which will allow
3816 * an average latency of 1.5/HZ.
3818 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3823 /* Even though interrupts have been re-enabled, this
3824 * access is safe because interrupts can only add new
3825 * entries to the tail of this list, and only ->poll()
3826 * calls can remove this head entry from the list.
3828 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3830 have = netpoll_poll_lock(n);
3834 /* This NAPI_STATE_SCHED test is for avoiding a race
3835 * with netpoll's poll_napi(). Only the entity which
3836 * obtains the lock and sees NAPI_STATE_SCHED set will
3837 * actually make the ->poll() call. Therefore we avoid
3838 * accidentally calling ->poll() when NAPI is not scheduled.
3841 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3842 work = n->poll(n, weight);
3846 WARN_ON_ONCE(work > weight);
3850 local_irq_disable();
3852 /* Drivers must not modify the NAPI state if they
3853 * consume the entire weight. In such cases this code
3854 * still "owns" the NAPI instance and therefore can
3855 * move the instance around on the list at-will.
3857 if (unlikely(work == weight)) {
3858 if (unlikely(napi_disable_pending(n))) {
3861 local_irq_disable();
3863 list_move_tail(&n->poll_list, &sd->poll_list);
3866 netpoll_poll_unlock(have);
3869 net_rps_action_and_irq_enable(sd);
3871 #ifdef CONFIG_NET_DMA
3873 * There may not be any more sk_buffs coming right now, so push
3874 * any pending DMA copies to hardware
3876 dma_issue_pending_all();
3883 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3887 static gifconf_func_t *gifconf_list[NPROTO];
3890 * register_gifconf - register a SIOCGIF handler
3891 * @family: Address family
3892 * @gifconf: Function handler
3894 * Register protocol dependent address dumping routines. The handler
3895 * that is passed must not be freed or reused until it has been replaced
3896 * by another handler.
3898 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3900 if (family >= NPROTO)
3902 gifconf_list[family] = gifconf;
3905 EXPORT_SYMBOL(register_gifconf);
3909 * Map an interface index to its name (SIOCGIFNAME)
3913 * We need this ioctl for efficient implementation of the
3914 * if_indextoname() function required by the IPv6 API. Without
3915 * it, we would have to search all the interfaces to find a
3919 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3921 struct net_device *dev;
3925 * Fetch the caller's info block.
3928 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3932 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3938 strcpy(ifr.ifr_name, dev->name);
3941 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3947 * Perform a SIOCGIFCONF call. This structure will change
3948 * size eventually, and there is nothing I can do about it.
3949 * Thus we will need a 'compatibility mode'.
3952 static int dev_ifconf(struct net *net, char __user *arg)
3955 struct net_device *dev;
3962 * Fetch the caller's info block.
3965 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3972 * Loop over the interfaces, and write an info block for each.
3976 for_each_netdev(net, dev) {
3977 for (i = 0; i < NPROTO; i++) {
3978 if (gifconf_list[i]) {
3981 done = gifconf_list[i](dev, NULL, 0);
3983 done = gifconf_list[i](dev, pos + total,
3993 * All done. Write the updated control block back to the caller.
3995 ifc.ifc_len = total;
3998 * Both BSD and Solaris return 0 here, so we do too.
4000 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4003 #ifdef CONFIG_PROC_FS
4005 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4007 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4008 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4009 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4011 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4013 struct net *net = seq_file_net(seq);
4014 struct net_device *dev;
4015 struct hlist_node *p;
4016 struct hlist_head *h;
4017 unsigned int count = 0, offset = get_offset(*pos);
4019 h = &net->dev_name_head[get_bucket(*pos)];
4020 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4021 if (++count == offset)
4028 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4030 struct net_device *dev;
4031 unsigned int bucket;
4034 dev = dev_from_same_bucket(seq, pos);
4038 bucket = get_bucket(*pos) + 1;
4039 *pos = set_bucket_offset(bucket, 1);
4040 } while (bucket < NETDEV_HASHENTRIES);
4046 * This is invoked by the /proc filesystem handler to display a device
4049 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4054 return SEQ_START_TOKEN;
4056 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4059 return dev_from_bucket(seq, pos);
4062 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4065 return dev_from_bucket(seq, pos);
4068 void dev_seq_stop(struct seq_file *seq, void *v)
4074 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4076 struct rtnl_link_stats64 temp;
4077 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4079 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4080 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4081 dev->name, stats->rx_bytes, stats->rx_packets,
4083 stats->rx_dropped + stats->rx_missed_errors,
4084 stats->rx_fifo_errors,
4085 stats->rx_length_errors + stats->rx_over_errors +
4086 stats->rx_crc_errors + stats->rx_frame_errors,
4087 stats->rx_compressed, stats->multicast,
4088 stats->tx_bytes, stats->tx_packets,
4089 stats->tx_errors, stats->tx_dropped,
4090 stats->tx_fifo_errors, stats->collisions,
4091 stats->tx_carrier_errors +
4092 stats->tx_aborted_errors +
4093 stats->tx_window_errors +
4094 stats->tx_heartbeat_errors,
4095 stats->tx_compressed);
4099 * Called from the PROCfs module. This now uses the new arbitrary sized
4100 * /proc/net interface to create /proc/net/dev
4102 static int dev_seq_show(struct seq_file *seq, void *v)
4104 if (v == SEQ_START_TOKEN)
4105 seq_puts(seq, "Inter-| Receive "
4107 " face |bytes packets errs drop fifo frame "
4108 "compressed multicast|bytes packets errs "
4109 "drop fifo colls carrier compressed\n");
4111 dev_seq_printf_stats(seq, v);
4115 static struct softnet_data *softnet_get_online(loff_t *pos)
4117 struct softnet_data *sd = NULL;
4119 while (*pos < nr_cpu_ids)
4120 if (cpu_online(*pos)) {
4121 sd = &per_cpu(softnet_data, *pos);
4128 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4130 return softnet_get_online(pos);
4133 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4136 return softnet_get_online(pos);
4139 static void softnet_seq_stop(struct seq_file *seq, void *v)
4143 static int softnet_seq_show(struct seq_file *seq, void *v)
4145 struct softnet_data *sd = v;
4147 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4148 sd->processed, sd->dropped, sd->time_squeeze, 0,
4149 0, 0, 0, 0, /* was fastroute */
4150 sd->cpu_collision, sd->received_rps);
4154 static const struct seq_operations dev_seq_ops = {
4155 .start = dev_seq_start,
4156 .next = dev_seq_next,
4157 .stop = dev_seq_stop,
4158 .show = dev_seq_show,
4161 static int dev_seq_open(struct inode *inode, struct file *file)
4163 return seq_open_net(inode, file, &dev_seq_ops,
4164 sizeof(struct seq_net_private));
4167 static const struct file_operations dev_seq_fops = {
4168 .owner = THIS_MODULE,
4169 .open = dev_seq_open,
4171 .llseek = seq_lseek,
4172 .release = seq_release_net,
4175 static const struct seq_operations softnet_seq_ops = {
4176 .start = softnet_seq_start,
4177 .next = softnet_seq_next,
4178 .stop = softnet_seq_stop,
4179 .show = softnet_seq_show,
4182 static int softnet_seq_open(struct inode *inode, struct file *file)
4184 return seq_open(file, &softnet_seq_ops);
4187 static const struct file_operations softnet_seq_fops = {
4188 .owner = THIS_MODULE,
4189 .open = softnet_seq_open,
4191 .llseek = seq_lseek,
4192 .release = seq_release,
4195 static void *ptype_get_idx(loff_t pos)
4197 struct packet_type *pt = NULL;
4201 list_for_each_entry_rcu(pt, &ptype_all, list) {
4207 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4208 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4217 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4221 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4224 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4226 struct packet_type *pt;
4227 struct list_head *nxt;
4231 if (v == SEQ_START_TOKEN)
4232 return ptype_get_idx(0);
4235 nxt = pt->list.next;
4236 if (pt->type == htons(ETH_P_ALL)) {
4237 if (nxt != &ptype_all)
4240 nxt = ptype_base[0].next;
4242 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4244 while (nxt == &ptype_base[hash]) {
4245 if (++hash >= PTYPE_HASH_SIZE)
4247 nxt = ptype_base[hash].next;
4250 return list_entry(nxt, struct packet_type, list);
4253 static void ptype_seq_stop(struct seq_file *seq, void *v)
4259 static int ptype_seq_show(struct seq_file *seq, void *v)
4261 struct packet_type *pt = v;
4263 if (v == SEQ_START_TOKEN)
4264 seq_puts(seq, "Type Device Function\n");
4265 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4266 if (pt->type == htons(ETH_P_ALL))
4267 seq_puts(seq, "ALL ");
4269 seq_printf(seq, "%04x", ntohs(pt->type));
4271 seq_printf(seq, " %-8s %pF\n",
4272 pt->dev ? pt->dev->name : "", pt->func);
4278 static const struct seq_operations ptype_seq_ops = {
4279 .start = ptype_seq_start,
4280 .next = ptype_seq_next,
4281 .stop = ptype_seq_stop,
4282 .show = ptype_seq_show,
4285 static int ptype_seq_open(struct inode *inode, struct file *file)
4287 return seq_open_net(inode, file, &ptype_seq_ops,
4288 sizeof(struct seq_net_private));
4291 static const struct file_operations ptype_seq_fops = {
4292 .owner = THIS_MODULE,
4293 .open = ptype_seq_open,
4295 .llseek = seq_lseek,
4296 .release = seq_release_net,
4300 static int __net_init dev_proc_net_init(struct net *net)
4304 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4306 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4308 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4311 if (wext_proc_init(net))
4317 proc_net_remove(net, "ptype");
4319 proc_net_remove(net, "softnet_stat");
4321 proc_net_remove(net, "dev");
4325 static void __net_exit dev_proc_net_exit(struct net *net)
4327 wext_proc_exit(net);
4329 proc_net_remove(net, "ptype");
4330 proc_net_remove(net, "softnet_stat");
4331 proc_net_remove(net, "dev");
4334 static struct pernet_operations __net_initdata dev_proc_ops = {
4335 .init = dev_proc_net_init,
4336 .exit = dev_proc_net_exit,
4339 static int __init dev_proc_init(void)
4341 return register_pernet_subsys(&dev_proc_ops);
4344 #define dev_proc_init() 0
4345 #endif /* CONFIG_PROC_FS */
4349 * netdev_set_master - set up master pointer
4350 * @slave: slave device
4351 * @master: new master device
4353 * Changes the master device of the slave. Pass %NULL to break the
4354 * bonding. The caller must hold the RTNL semaphore. On a failure
4355 * a negative errno code is returned. On success the reference counts
4356 * are adjusted and the function returns zero.
4358 int netdev_set_master(struct net_device *slave, struct net_device *master)
4360 struct net_device *old = slave->master;
4370 slave->master = master;
4376 EXPORT_SYMBOL(netdev_set_master);
4379 * netdev_set_bond_master - set up bonding master/slave pair
4380 * @slave: slave device
4381 * @master: new master device
4383 * Changes the master device of the slave. Pass %NULL to break the
4384 * bonding. The caller must hold the RTNL semaphore. On a failure
4385 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4386 * to the routing socket and the function returns zero.
4388 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4394 err = netdev_set_master(slave, master);
4398 slave->flags |= IFF_SLAVE;
4400 slave->flags &= ~IFF_SLAVE;
4402 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4405 EXPORT_SYMBOL(netdev_set_bond_master);
4407 static void dev_change_rx_flags(struct net_device *dev, int flags)
4409 const struct net_device_ops *ops = dev->netdev_ops;
4411 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4412 ops->ndo_change_rx_flags(dev, flags);
4415 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4417 unsigned int old_flags = dev->flags;
4423 dev->flags |= IFF_PROMISC;
4424 dev->promiscuity += inc;
4425 if (dev->promiscuity == 0) {
4428 * If inc causes overflow, untouch promisc and return error.
4431 dev->flags &= ~IFF_PROMISC;
4433 dev->promiscuity -= inc;
4434 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4439 if (dev->flags != old_flags) {
4440 pr_info("device %s %s promiscuous mode\n",
4442 dev->flags & IFF_PROMISC ? "entered" : "left");
4443 if (audit_enabled) {
4444 current_uid_gid(&uid, &gid);
4445 audit_log(current->audit_context, GFP_ATOMIC,
4446 AUDIT_ANOM_PROMISCUOUS,
4447 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4448 dev->name, (dev->flags & IFF_PROMISC),
4449 (old_flags & IFF_PROMISC),
4450 audit_get_loginuid(current),
4452 audit_get_sessionid(current));
4455 dev_change_rx_flags(dev, IFF_PROMISC);
4461 * dev_set_promiscuity - update promiscuity count on a device
4465 * Add or remove promiscuity from a device. While the count in the device
4466 * remains above zero the interface remains promiscuous. Once it hits zero
4467 * the device reverts back to normal filtering operation. A negative inc
4468 * value is used to drop promiscuity on the device.
4469 * Return 0 if successful or a negative errno code on error.
4471 int dev_set_promiscuity(struct net_device *dev, int inc)
4473 unsigned int old_flags = dev->flags;
4476 err = __dev_set_promiscuity(dev, inc);
4479 if (dev->flags != old_flags)
4480 dev_set_rx_mode(dev);
4483 EXPORT_SYMBOL(dev_set_promiscuity);
4486 * dev_set_allmulti - update allmulti count on a device
4490 * Add or remove reception of all multicast frames to a device. While the
4491 * count in the device remains above zero the interface remains listening
4492 * to all interfaces. Once it hits zero the device reverts back to normal
4493 * filtering operation. A negative @inc value is used to drop the counter
4494 * when releasing a resource needing all multicasts.
4495 * Return 0 if successful or a negative errno code on error.
4498 int dev_set_allmulti(struct net_device *dev, int inc)
4500 unsigned int old_flags = dev->flags;
4504 dev->flags |= IFF_ALLMULTI;
4505 dev->allmulti += inc;
4506 if (dev->allmulti == 0) {
4509 * If inc causes overflow, untouch allmulti and return error.
4512 dev->flags &= ~IFF_ALLMULTI;
4514 dev->allmulti -= inc;
4515 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4520 if (dev->flags ^ old_flags) {
4521 dev_change_rx_flags(dev, IFF_ALLMULTI);
4522 dev_set_rx_mode(dev);
4526 EXPORT_SYMBOL(dev_set_allmulti);
4529 * Upload unicast and multicast address lists to device and
4530 * configure RX filtering. When the device doesn't support unicast
4531 * filtering it is put in promiscuous mode while unicast addresses
4534 void __dev_set_rx_mode(struct net_device *dev)
4536 const struct net_device_ops *ops = dev->netdev_ops;
4538 /* dev_open will call this function so the list will stay sane. */
4539 if (!(dev->flags&IFF_UP))
4542 if (!netif_device_present(dev))
4545 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4546 /* Unicast addresses changes may only happen under the rtnl,
4547 * therefore calling __dev_set_promiscuity here is safe.
4549 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4550 __dev_set_promiscuity(dev, 1);
4551 dev->uc_promisc = true;
4552 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4553 __dev_set_promiscuity(dev, -1);
4554 dev->uc_promisc = false;
4558 if (ops->ndo_set_rx_mode)
4559 ops->ndo_set_rx_mode(dev);
4562 void dev_set_rx_mode(struct net_device *dev)
4564 netif_addr_lock_bh(dev);
4565 __dev_set_rx_mode(dev);
4566 netif_addr_unlock_bh(dev);
4570 * dev_get_flags - get flags reported to userspace
4573 * Get the combination of flag bits exported through APIs to userspace.
4575 unsigned dev_get_flags(const struct net_device *dev)
4579 flags = (dev->flags & ~(IFF_PROMISC |
4584 (dev->gflags & (IFF_PROMISC |
4587 if (netif_running(dev)) {
4588 if (netif_oper_up(dev))
4589 flags |= IFF_RUNNING;
4590 if (netif_carrier_ok(dev))
4591 flags |= IFF_LOWER_UP;
4592 if (netif_dormant(dev))
4593 flags |= IFF_DORMANT;
4598 EXPORT_SYMBOL(dev_get_flags);
4600 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4602 unsigned int old_flags = dev->flags;
4608 * Set the flags on our device.
4611 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4612 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4614 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4618 * Load in the correct multicast list now the flags have changed.
4621 if ((old_flags ^ flags) & IFF_MULTICAST)
4622 dev_change_rx_flags(dev, IFF_MULTICAST);
4624 dev_set_rx_mode(dev);
4627 * Have we downed the interface. We handle IFF_UP ourselves
4628 * according to user attempts to set it, rather than blindly
4633 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4634 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4637 dev_set_rx_mode(dev);
4640 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4641 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4643 dev->gflags ^= IFF_PROMISC;
4644 dev_set_promiscuity(dev, inc);
4647 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4648 is important. Some (broken) drivers set IFF_PROMISC, when
4649 IFF_ALLMULTI is requested not asking us and not reporting.
4651 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4652 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4654 dev->gflags ^= IFF_ALLMULTI;
4655 dev_set_allmulti(dev, inc);
4661 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4663 unsigned int changes = dev->flags ^ old_flags;
4665 if (changes & IFF_UP) {
4666 if (dev->flags & IFF_UP)
4667 call_netdevice_notifiers(NETDEV_UP, dev);
4669 call_netdevice_notifiers(NETDEV_DOWN, dev);
4672 if (dev->flags & IFF_UP &&
4673 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4674 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4678 * dev_change_flags - change device settings
4680 * @flags: device state flags
4682 * Change settings on device based state flags. The flags are
4683 * in the userspace exported format.
4685 int dev_change_flags(struct net_device *dev, unsigned int flags)
4688 unsigned int changes, old_flags = dev->flags;
4690 ret = __dev_change_flags(dev, flags);
4694 changes = old_flags ^ dev->flags;
4696 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4698 __dev_notify_flags(dev, old_flags);
4701 EXPORT_SYMBOL(dev_change_flags);
4704 * dev_set_mtu - Change maximum transfer unit
4706 * @new_mtu: new transfer unit
4708 * Change the maximum transfer size of the network device.
4710 int dev_set_mtu(struct net_device *dev, int new_mtu)
4712 const struct net_device_ops *ops = dev->netdev_ops;
4715 if (new_mtu == dev->mtu)
4718 /* MTU must be positive. */
4722 if (!netif_device_present(dev))
4726 if (ops->ndo_change_mtu)
4727 err = ops->ndo_change_mtu(dev, new_mtu);
4731 if (!err && dev->flags & IFF_UP)
4732 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4735 EXPORT_SYMBOL(dev_set_mtu);
4738 * dev_set_group - Change group this device belongs to
4740 * @new_group: group this device should belong to
4742 void dev_set_group(struct net_device *dev, int new_group)
4744 dev->group = new_group;
4746 EXPORT_SYMBOL(dev_set_group);
4749 * dev_set_mac_address - Change Media Access Control Address
4753 * Change the hardware (MAC) address of the device
4755 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4757 const struct net_device_ops *ops = dev->netdev_ops;
4760 if (!ops->ndo_set_mac_address)
4762 if (sa->sa_family != dev->type)
4764 if (!netif_device_present(dev))
4766 err = ops->ndo_set_mac_address(dev, sa);
4768 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4769 add_device_randomness(dev->dev_addr, dev->addr_len);
4772 EXPORT_SYMBOL(dev_set_mac_address);
4775 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4777 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4780 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4786 case SIOCGIFFLAGS: /* Get interface flags */
4787 ifr->ifr_flags = (short) dev_get_flags(dev);
4790 case SIOCGIFMETRIC: /* Get the metric on the interface
4791 (currently unused) */
4792 ifr->ifr_metric = 0;
4795 case SIOCGIFMTU: /* Get the MTU of a device */
4796 ifr->ifr_mtu = dev->mtu;
4801 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4803 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4804 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4805 ifr->ifr_hwaddr.sa_family = dev->type;
4813 ifr->ifr_map.mem_start = dev->mem_start;
4814 ifr->ifr_map.mem_end = dev->mem_end;
4815 ifr->ifr_map.base_addr = dev->base_addr;
4816 ifr->ifr_map.irq = dev->irq;
4817 ifr->ifr_map.dma = dev->dma;
4818 ifr->ifr_map.port = dev->if_port;
4822 ifr->ifr_ifindex = dev->ifindex;
4826 ifr->ifr_qlen = dev->tx_queue_len;
4830 /* dev_ioctl() should ensure this case
4842 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4844 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4847 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4848 const struct net_device_ops *ops;
4853 ops = dev->netdev_ops;
4856 case SIOCSIFFLAGS: /* Set interface flags */
4857 return dev_change_flags(dev, ifr->ifr_flags);
4859 case SIOCSIFMETRIC: /* Set the metric on the interface
4860 (currently unused) */
4863 case SIOCSIFMTU: /* Set the MTU of a device */
4864 return dev_set_mtu(dev, ifr->ifr_mtu);
4867 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4869 case SIOCSIFHWBROADCAST:
4870 if (ifr->ifr_hwaddr.sa_family != dev->type)
4872 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4873 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4874 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4878 if (ops->ndo_set_config) {
4879 if (!netif_device_present(dev))
4881 return ops->ndo_set_config(dev, &ifr->ifr_map);
4886 if (!ops->ndo_set_rx_mode ||
4887 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4889 if (!netif_device_present(dev))
4891 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4894 if (!ops->ndo_set_rx_mode ||
4895 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4897 if (!netif_device_present(dev))
4899 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4902 if (ifr->ifr_qlen < 0)
4904 dev->tx_queue_len = ifr->ifr_qlen;
4908 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4909 return dev_change_name(dev, ifr->ifr_newname);
4912 err = net_hwtstamp_validate(ifr);
4918 * Unknown or private ioctl
4921 if ((cmd >= SIOCDEVPRIVATE &&
4922 cmd <= SIOCDEVPRIVATE + 15) ||
4923 cmd == SIOCBONDENSLAVE ||
4924 cmd == SIOCBONDRELEASE ||
4925 cmd == SIOCBONDSETHWADDR ||
4926 cmd == SIOCBONDSLAVEINFOQUERY ||
4927 cmd == SIOCBONDINFOQUERY ||
4928 cmd == SIOCBONDCHANGEACTIVE ||
4929 cmd == SIOCGMIIPHY ||
4930 cmd == SIOCGMIIREG ||
4931 cmd == SIOCSMIIREG ||
4932 cmd == SIOCBRADDIF ||
4933 cmd == SIOCBRDELIF ||
4934 cmd == SIOCSHWTSTAMP ||
4935 cmd == SIOCWANDEV) {
4937 if (ops->ndo_do_ioctl) {
4938 if (netif_device_present(dev))
4939 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4951 * This function handles all "interface"-type I/O control requests. The actual
4952 * 'doing' part of this is dev_ifsioc above.
4956 * dev_ioctl - network device ioctl
4957 * @net: the applicable net namespace
4958 * @cmd: command to issue
4959 * @arg: pointer to a struct ifreq in user space
4961 * Issue ioctl functions to devices. This is normally called by the
4962 * user space syscall interfaces but can sometimes be useful for
4963 * other purposes. The return value is the return from the syscall if
4964 * positive or a negative errno code on error.
4967 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4973 /* One special case: SIOCGIFCONF takes ifconf argument
4974 and requires shared lock, because it sleeps writing
4978 if (cmd == SIOCGIFCONF) {
4980 ret = dev_ifconf(net, (char __user *) arg);
4984 if (cmd == SIOCGIFNAME)
4985 return dev_ifname(net, (struct ifreq __user *)arg);
4987 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4990 ifr.ifr_name[IFNAMSIZ-1] = 0;
4992 colon = strchr(ifr.ifr_name, ':');
4997 * See which interface the caller is talking about.
5002 * These ioctl calls:
5003 * - can be done by all.
5004 * - atomic and do not require locking.
5015 dev_load(net, ifr.ifr_name);
5017 ret = dev_ifsioc_locked(net, &ifr, cmd);
5022 if (copy_to_user(arg, &ifr,
5023 sizeof(struct ifreq)))
5029 dev_load(net, ifr.ifr_name);
5031 ret = dev_ethtool(net, &ifr);
5036 if (copy_to_user(arg, &ifr,
5037 sizeof(struct ifreq)))
5043 * These ioctl calls:
5044 * - require superuser power.
5045 * - require strict serialization.
5051 if (!capable(CAP_NET_ADMIN))
5053 dev_load(net, ifr.ifr_name);
5055 ret = dev_ifsioc(net, &ifr, cmd);
5060 if (copy_to_user(arg, &ifr,
5061 sizeof(struct ifreq)))
5067 * These ioctl calls:
5068 * - require superuser power.
5069 * - require strict serialization.
5070 * - do not return a value
5080 case SIOCSIFHWBROADCAST:
5083 case SIOCBONDENSLAVE:
5084 case SIOCBONDRELEASE:
5085 case SIOCBONDSETHWADDR:
5086 case SIOCBONDCHANGEACTIVE:
5090 if (!capable(CAP_NET_ADMIN))
5093 case SIOCBONDSLAVEINFOQUERY:
5094 case SIOCBONDINFOQUERY:
5095 dev_load(net, ifr.ifr_name);
5097 ret = dev_ifsioc(net, &ifr, cmd);
5102 /* Get the per device memory space. We can add this but
5103 * currently do not support it */
5105 /* Set the per device memory buffer space.
5106 * Not applicable in our case */
5111 * Unknown or private ioctl.
5114 if (cmd == SIOCWANDEV ||
5115 (cmd >= SIOCDEVPRIVATE &&
5116 cmd <= SIOCDEVPRIVATE + 15)) {
5117 dev_load(net, ifr.ifr_name);
5119 ret = dev_ifsioc(net, &ifr, cmd);
5121 if (!ret && copy_to_user(arg, &ifr,
5122 sizeof(struct ifreq)))
5126 /* Take care of Wireless Extensions */
5127 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5128 return wext_handle_ioctl(net, &ifr, cmd, arg);
5135 * dev_new_index - allocate an ifindex
5136 * @net: the applicable net namespace
5138 * Returns a suitable unique value for a new device interface
5139 * number. The caller must hold the rtnl semaphore or the
5140 * dev_base_lock to be sure it remains unique.
5142 static int dev_new_index(struct net *net)
5148 if (!__dev_get_by_index(net, ifindex))
5153 /* Delayed registration/unregisteration */
5154 static LIST_HEAD(net_todo_list);
5156 static void net_set_todo(struct net_device *dev)
5158 list_add_tail(&dev->todo_list, &net_todo_list);
5161 static void rollback_registered_many(struct list_head *head)
5163 struct net_device *dev, *tmp;
5165 BUG_ON(dev_boot_phase);
5168 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5169 /* Some devices call without registering
5170 * for initialization unwind. Remove those
5171 * devices and proceed with the remaining.
5173 if (dev->reg_state == NETREG_UNINITIALIZED) {
5174 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5178 list_del(&dev->unreg_list);
5181 dev->dismantle = true;
5182 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5185 /* If device is running, close it first. */
5186 dev_close_many(head);
5188 list_for_each_entry(dev, head, unreg_list) {
5189 /* And unlink it from device chain. */
5190 unlist_netdevice(dev);
5192 dev->reg_state = NETREG_UNREGISTERING;
5197 list_for_each_entry(dev, head, unreg_list) {
5198 /* Shutdown queueing discipline. */
5202 /* Notify protocols, that we are about to destroy
5203 this device. They should clean all the things.
5205 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5207 if (!dev->rtnl_link_ops ||
5208 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5209 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5212 * Flush the unicast and multicast chains
5217 if (dev->netdev_ops->ndo_uninit)
5218 dev->netdev_ops->ndo_uninit(dev);
5220 /* Notifier chain MUST detach us from master device. */
5221 WARN_ON(dev->master);
5223 /* Remove entries from kobject tree */
5224 netdev_unregister_kobject(dev);
5227 /* Process any work delayed until the end of the batch */
5228 dev = list_first_entry(head, struct net_device, unreg_list);
5229 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5233 list_for_each_entry(dev, head, unreg_list)
5237 static void rollback_registered(struct net_device *dev)
5241 list_add(&dev->unreg_list, &single);
5242 rollback_registered_many(&single);
5246 static netdev_features_t netdev_fix_features(struct net_device *dev,
5247 netdev_features_t features)
5249 /* Fix illegal checksum combinations */
5250 if ((features & NETIF_F_HW_CSUM) &&
5251 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5252 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5253 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5256 /* Fix illegal SG+CSUM combinations. */
5257 if ((features & NETIF_F_SG) &&
5258 !(features & NETIF_F_ALL_CSUM)) {
5260 "Dropping NETIF_F_SG since no checksum feature.\n");
5261 features &= ~NETIF_F_SG;
5264 /* TSO requires that SG is present as well. */
5265 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5266 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5267 features &= ~NETIF_F_ALL_TSO;
5270 /* TSO ECN requires that TSO is present as well. */
5271 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5272 features &= ~NETIF_F_TSO_ECN;
5274 /* Software GSO depends on SG. */
5275 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5276 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5277 features &= ~NETIF_F_GSO;
5280 /* UFO needs SG and checksumming */
5281 if (features & NETIF_F_UFO) {
5282 /* maybe split UFO into V4 and V6? */
5283 if (!((features & NETIF_F_GEN_CSUM) ||
5284 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5285 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5287 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5288 features &= ~NETIF_F_UFO;
5291 if (!(features & NETIF_F_SG)) {
5293 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5294 features &= ~NETIF_F_UFO;
5301 int __netdev_update_features(struct net_device *dev)
5303 netdev_features_t features;
5308 features = netdev_get_wanted_features(dev);
5310 if (dev->netdev_ops->ndo_fix_features)
5311 features = dev->netdev_ops->ndo_fix_features(dev, features);
5313 /* driver might be less strict about feature dependencies */
5314 features = netdev_fix_features(dev, features);
5316 if (dev->features == features)
5319 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5320 &dev->features, &features);
5322 if (dev->netdev_ops->ndo_set_features)
5323 err = dev->netdev_ops->ndo_set_features(dev, features);
5325 if (unlikely(err < 0)) {
5327 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5328 err, &features, &dev->features);
5333 dev->features = features;
5339 * netdev_update_features - recalculate device features
5340 * @dev: the device to check
5342 * Recalculate dev->features set and send notifications if it
5343 * has changed. Should be called after driver or hardware dependent
5344 * conditions might have changed that influence the features.
5346 void netdev_update_features(struct net_device *dev)
5348 if (__netdev_update_features(dev))
5349 netdev_features_change(dev);
5351 EXPORT_SYMBOL(netdev_update_features);
5354 * netdev_change_features - recalculate device features
5355 * @dev: the device to check
5357 * Recalculate dev->features set and send notifications even
5358 * if they have not changed. Should be called instead of
5359 * netdev_update_features() if also dev->vlan_features might
5360 * have changed to allow the changes to be propagated to stacked
5363 void netdev_change_features(struct net_device *dev)
5365 __netdev_update_features(dev);
5366 netdev_features_change(dev);
5368 EXPORT_SYMBOL(netdev_change_features);
5371 * netif_stacked_transfer_operstate - transfer operstate
5372 * @rootdev: the root or lower level device to transfer state from
5373 * @dev: the device to transfer operstate to
5375 * Transfer operational state from root to device. This is normally
5376 * called when a stacking relationship exists between the root
5377 * device and the device(a leaf device).
5379 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5380 struct net_device *dev)
5382 if (rootdev->operstate == IF_OPER_DORMANT)
5383 netif_dormant_on(dev);
5385 netif_dormant_off(dev);
5387 if (netif_carrier_ok(rootdev)) {
5388 if (!netif_carrier_ok(dev))
5389 netif_carrier_on(dev);
5391 if (netif_carrier_ok(dev))
5392 netif_carrier_off(dev);
5395 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5398 static int netif_alloc_rx_queues(struct net_device *dev)
5400 unsigned int i, count = dev->num_rx_queues;
5401 struct netdev_rx_queue *rx;
5405 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5407 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5412 for (i = 0; i < count; i++)
5418 static void netdev_init_one_queue(struct net_device *dev,
5419 struct netdev_queue *queue, void *_unused)
5421 /* Initialize queue lock */
5422 spin_lock_init(&queue->_xmit_lock);
5423 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5424 queue->xmit_lock_owner = -1;
5425 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5428 dql_init(&queue->dql, HZ);
5432 static int netif_alloc_netdev_queues(struct net_device *dev)
5434 unsigned int count = dev->num_tx_queues;
5435 struct netdev_queue *tx;
5439 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5441 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5446 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5447 spin_lock_init(&dev->tx_global_lock);
5453 * register_netdevice - register a network device
5454 * @dev: device to register
5456 * Take a completed network device structure and add it to the kernel
5457 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5458 * chain. 0 is returned on success. A negative errno code is returned
5459 * on a failure to set up the device, or if the name is a duplicate.
5461 * Callers must hold the rtnl semaphore. You may want
5462 * register_netdev() instead of this.
5465 * The locking appears insufficient to guarantee two parallel registers
5466 * will not get the same name.
5469 int register_netdevice(struct net_device *dev)
5472 struct net *net = dev_net(dev);
5474 BUG_ON(dev_boot_phase);
5479 /* When net_device's are persistent, this will be fatal. */
5480 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5483 spin_lock_init(&dev->addr_list_lock);
5484 netdev_set_addr_lockdep_class(dev);
5488 ret = dev_get_valid_name(dev, dev->name);
5492 /* Init, if this function is available */
5493 if (dev->netdev_ops->ndo_init) {
5494 ret = dev->netdev_ops->ndo_init(dev);
5502 dev->ifindex = dev_new_index(net);
5503 if (dev->iflink == -1)
5504 dev->iflink = dev->ifindex;
5506 /* Transfer changeable features to wanted_features and enable
5507 * software offloads (GSO and GRO).
5509 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5510 dev->features |= NETIF_F_SOFT_FEATURES;
5511 dev->wanted_features = dev->features & dev->hw_features;
5513 /* Turn on no cache copy if HW is doing checksum */
5514 if (!(dev->flags & IFF_LOOPBACK)) {
5515 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5516 if (dev->features & NETIF_F_ALL_CSUM) {
5517 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5518 dev->features |= NETIF_F_NOCACHE_COPY;
5522 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5524 dev->vlan_features |= NETIF_F_HIGHDMA;
5526 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5527 ret = notifier_to_errno(ret);
5531 ret = netdev_register_kobject(dev);
5534 dev->reg_state = NETREG_REGISTERED;
5536 __netdev_update_features(dev);
5539 * Default initial state at registry is that the
5540 * device is present.
5543 set_bit(__LINK_STATE_PRESENT, &dev->state);
5545 dev_init_scheduler(dev);
5547 list_netdevice(dev);
5548 add_device_randomness(dev->dev_addr, dev->addr_len);
5550 /* Notify protocols, that a new device appeared. */
5551 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5552 ret = notifier_to_errno(ret);
5554 rollback_registered(dev);
5555 dev->reg_state = NETREG_UNREGISTERED;
5558 * Prevent userspace races by waiting until the network
5559 * device is fully setup before sending notifications.
5561 if (!dev->rtnl_link_ops ||
5562 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5563 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5569 if (dev->netdev_ops->ndo_uninit)
5570 dev->netdev_ops->ndo_uninit(dev);
5573 EXPORT_SYMBOL(register_netdevice);
5576 * init_dummy_netdev - init a dummy network device for NAPI
5577 * @dev: device to init
5579 * This takes a network device structure and initialize the minimum
5580 * amount of fields so it can be used to schedule NAPI polls without
5581 * registering a full blown interface. This is to be used by drivers
5582 * that need to tie several hardware interfaces to a single NAPI
5583 * poll scheduler due to HW limitations.
5585 int init_dummy_netdev(struct net_device *dev)
5587 /* Clear everything. Note we don't initialize spinlocks
5588 * are they aren't supposed to be taken by any of the
5589 * NAPI code and this dummy netdev is supposed to be
5590 * only ever used for NAPI polls
5592 memset(dev, 0, sizeof(struct net_device));
5594 /* make sure we BUG if trying to hit standard
5595 * register/unregister code path
5597 dev->reg_state = NETREG_DUMMY;
5599 /* NAPI wants this */
5600 INIT_LIST_HEAD(&dev->napi_list);
5602 /* a dummy interface is started by default */
5603 set_bit(__LINK_STATE_PRESENT, &dev->state);
5604 set_bit(__LINK_STATE_START, &dev->state);
5606 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5607 * because users of this 'device' dont need to change
5613 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5617 * register_netdev - register a network device
5618 * @dev: device to register
5620 * Take a completed network device structure and add it to the kernel
5621 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5622 * chain. 0 is returned on success. A negative errno code is returned
5623 * on a failure to set up the device, or if the name is a duplicate.
5625 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5626 * and expands the device name if you passed a format string to
5629 int register_netdev(struct net_device *dev)
5634 err = register_netdevice(dev);
5638 EXPORT_SYMBOL(register_netdev);
5640 int netdev_refcnt_read(const struct net_device *dev)
5644 for_each_possible_cpu(i)
5645 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5648 EXPORT_SYMBOL(netdev_refcnt_read);
5651 * netdev_wait_allrefs - wait until all references are gone.
5653 * This is called when unregistering network devices.
5655 * Any protocol or device that holds a reference should register
5656 * for netdevice notification, and cleanup and put back the
5657 * reference if they receive an UNREGISTER event.
5658 * We can get stuck here if buggy protocols don't correctly
5661 static void netdev_wait_allrefs(struct net_device *dev)
5663 unsigned long rebroadcast_time, warning_time;
5666 linkwatch_forget_dev(dev);
5668 rebroadcast_time = warning_time = jiffies;
5669 refcnt = netdev_refcnt_read(dev);
5671 while (refcnt != 0) {
5672 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5675 /* Rebroadcast unregister notification */
5676 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5677 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5678 * should have already handle it the first time */
5680 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5682 /* We must not have linkwatch events
5683 * pending on unregister. If this
5684 * happens, we simply run the queue
5685 * unscheduled, resulting in a noop
5688 linkwatch_run_queue();
5693 rebroadcast_time = jiffies;
5698 refcnt = netdev_refcnt_read(dev);
5700 if (time_after(jiffies, warning_time + 10 * HZ)) {
5701 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5703 warning_time = jiffies;
5712 * register_netdevice(x1);
5713 * register_netdevice(x2);
5715 * unregister_netdevice(y1);
5716 * unregister_netdevice(y2);
5722 * We are invoked by rtnl_unlock().
5723 * This allows us to deal with problems:
5724 * 1) We can delete sysfs objects which invoke hotplug
5725 * without deadlocking with linkwatch via keventd.
5726 * 2) Since we run with the RTNL semaphore not held, we can sleep
5727 * safely in order to wait for the netdev refcnt to drop to zero.
5729 * We must not return until all unregister events added during
5730 * the interval the lock was held have been completed.
5732 void netdev_run_todo(void)
5734 struct list_head list;
5736 /* Snapshot list, allow later requests */
5737 list_replace_init(&net_todo_list, &list);
5741 /* Wait for rcu callbacks to finish before attempting to drain
5742 * the device list. This usually avoids a 250ms wait.
5744 if (!list_empty(&list))
5747 while (!list_empty(&list)) {
5748 struct net_device *dev
5749 = list_first_entry(&list, struct net_device, todo_list);
5750 list_del(&dev->todo_list);
5752 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5753 pr_err("network todo '%s' but state %d\n",
5754 dev->name, dev->reg_state);
5759 dev->reg_state = NETREG_UNREGISTERED;
5761 on_each_cpu(flush_backlog, dev, 1);
5763 netdev_wait_allrefs(dev);
5766 BUG_ON(netdev_refcnt_read(dev));
5767 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5768 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5769 WARN_ON(dev->dn_ptr);
5771 if (dev->destructor)
5772 dev->destructor(dev);
5774 /* Free network device */
5775 kobject_put(&dev->dev.kobj);
5779 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5780 * fields in the same order, with only the type differing.
5782 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5783 const struct net_device_stats *netdev_stats)
5785 #if BITS_PER_LONG == 64
5786 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5787 memcpy(stats64, netdev_stats, sizeof(*stats64));
5789 size_t i, n = sizeof(*stats64) / sizeof(u64);
5790 const unsigned long *src = (const unsigned long *)netdev_stats;
5791 u64 *dst = (u64 *)stats64;
5793 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5794 sizeof(*stats64) / sizeof(u64));
5795 for (i = 0; i < n; i++)
5799 EXPORT_SYMBOL(netdev_stats_to_stats64);
5802 * dev_get_stats - get network device statistics
5803 * @dev: device to get statistics from
5804 * @storage: place to store stats
5806 * Get network statistics from device. Return @storage.
5807 * The device driver may provide its own method by setting
5808 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5809 * otherwise the internal statistics structure is used.
5811 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5812 struct rtnl_link_stats64 *storage)
5814 const struct net_device_ops *ops = dev->netdev_ops;
5816 if (ops->ndo_get_stats64) {
5817 memset(storage, 0, sizeof(*storage));
5818 ops->ndo_get_stats64(dev, storage);
5819 } else if (ops->ndo_get_stats) {
5820 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5822 netdev_stats_to_stats64(storage, &dev->stats);
5824 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5827 EXPORT_SYMBOL(dev_get_stats);
5829 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5831 struct netdev_queue *queue = dev_ingress_queue(dev);
5833 #ifdef CONFIG_NET_CLS_ACT
5836 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5839 netdev_init_one_queue(dev, queue, NULL);
5840 queue->qdisc = &noop_qdisc;
5841 queue->qdisc_sleeping = &noop_qdisc;
5842 rcu_assign_pointer(dev->ingress_queue, queue);
5848 * alloc_netdev_mqs - allocate network device
5849 * @sizeof_priv: size of private data to allocate space for
5850 * @name: device name format string
5851 * @setup: callback to initialize device
5852 * @txqs: the number of TX subqueues to allocate
5853 * @rxqs: the number of RX subqueues to allocate
5855 * Allocates a struct net_device with private data area for driver use
5856 * and performs basic initialization. Also allocates subquue structs
5857 * for each queue on the device.
5859 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5860 void (*setup)(struct net_device *),
5861 unsigned int txqs, unsigned int rxqs)
5863 struct net_device *dev;
5865 struct net_device *p;
5867 BUG_ON(strlen(name) >= sizeof(dev->name));
5870 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5876 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5881 alloc_size = sizeof(struct net_device);
5883 /* ensure 32-byte alignment of private area */
5884 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5885 alloc_size += sizeof_priv;
5887 /* ensure 32-byte alignment of whole construct */
5888 alloc_size += NETDEV_ALIGN - 1;
5890 p = kzalloc(alloc_size, GFP_KERNEL);
5892 pr_err("alloc_netdev: Unable to allocate device\n");
5896 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5897 dev->padded = (char *)dev - (char *)p;
5899 dev->pcpu_refcnt = alloc_percpu(int);
5900 if (!dev->pcpu_refcnt)
5903 if (dev_addr_init(dev))
5909 dev_net_set(dev, &init_net);
5911 dev->gso_max_size = GSO_MAX_SIZE;
5913 INIT_LIST_HEAD(&dev->napi_list);
5914 INIT_LIST_HEAD(&dev->unreg_list);
5915 INIT_LIST_HEAD(&dev->link_watch_list);
5916 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5919 dev->num_tx_queues = txqs;
5920 dev->real_num_tx_queues = txqs;
5921 if (netif_alloc_netdev_queues(dev))
5925 dev->num_rx_queues = rxqs;
5926 dev->real_num_rx_queues = rxqs;
5927 if (netif_alloc_rx_queues(dev))
5931 strcpy(dev->name, name);
5932 dev->group = INIT_NETDEV_GROUP;
5940 free_percpu(dev->pcpu_refcnt);
5950 EXPORT_SYMBOL(alloc_netdev_mqs);
5953 * free_netdev - free network device
5956 * This function does the last stage of destroying an allocated device
5957 * interface. The reference to the device object is released.
5958 * If this is the last reference then it will be freed.
5960 void free_netdev(struct net_device *dev)
5962 struct napi_struct *p, *n;
5964 release_net(dev_net(dev));
5971 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5973 /* Flush device addresses */
5974 dev_addr_flush(dev);
5976 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5979 free_percpu(dev->pcpu_refcnt);
5980 dev->pcpu_refcnt = NULL;
5982 /* Compatibility with error handling in drivers */
5983 if (dev->reg_state == NETREG_UNINITIALIZED) {
5984 kfree((char *)dev - dev->padded);
5988 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5989 dev->reg_state = NETREG_RELEASED;
5991 /* will free via device release */
5992 put_device(&dev->dev);
5994 EXPORT_SYMBOL(free_netdev);
5997 * synchronize_net - Synchronize with packet receive processing
5999 * Wait for packets currently being received to be done.
6000 * Does not block later packets from starting.
6002 void synchronize_net(void)
6005 if (rtnl_is_locked())
6006 synchronize_rcu_expedited();
6010 EXPORT_SYMBOL(synchronize_net);
6013 * unregister_netdevice_queue - remove device from the kernel
6017 * This function shuts down a device interface and removes it
6018 * from the kernel tables.
6019 * If head not NULL, device is queued to be unregistered later.
6021 * Callers must hold the rtnl semaphore. You may want
6022 * unregister_netdev() instead of this.
6025 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6030 list_move_tail(&dev->unreg_list, head);
6032 rollback_registered(dev);
6033 /* Finish processing unregister after unlock */
6037 EXPORT_SYMBOL(unregister_netdevice_queue);
6040 * unregister_netdevice_many - unregister many devices
6041 * @head: list of devices
6043 void unregister_netdevice_many(struct list_head *head)
6045 struct net_device *dev;
6047 if (!list_empty(head)) {
6048 rollback_registered_many(head);
6049 list_for_each_entry(dev, head, unreg_list)
6053 EXPORT_SYMBOL(unregister_netdevice_many);
6056 * unregister_netdev - remove device from the kernel
6059 * This function shuts down a device interface and removes it
6060 * from the kernel tables.
6062 * This is just a wrapper for unregister_netdevice that takes
6063 * the rtnl semaphore. In general you want to use this and not
6064 * unregister_netdevice.
6066 void unregister_netdev(struct net_device *dev)
6069 unregister_netdevice(dev);
6072 EXPORT_SYMBOL(unregister_netdev);
6075 * dev_change_net_namespace - move device to different nethost namespace
6077 * @net: network namespace
6078 * @pat: If not NULL name pattern to try if the current device name
6079 * is already taken in the destination network namespace.
6081 * This function shuts down a device interface and moves it
6082 * to a new network namespace. On success 0 is returned, on
6083 * a failure a netagive errno code is returned.
6085 * Callers must hold the rtnl semaphore.
6088 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6094 /* Don't allow namespace local devices to be moved. */
6096 if (dev->features & NETIF_F_NETNS_LOCAL)
6099 /* Ensure the device has been registrered */
6101 if (dev->reg_state != NETREG_REGISTERED)
6104 /* Get out if there is nothing todo */
6106 if (net_eq(dev_net(dev), net))
6109 /* Pick the destination device name, and ensure
6110 * we can use it in the destination network namespace.
6113 if (__dev_get_by_name(net, dev->name)) {
6114 /* We get here if we can't use the current device name */
6117 if (dev_get_valid_name(dev, pat) < 0)
6122 * And now a mini version of register_netdevice unregister_netdevice.
6125 /* If device is running close it first. */
6128 /* And unlink it from device chain */
6130 unlist_netdevice(dev);
6134 /* Shutdown queueing discipline. */
6137 /* Notify protocols, that we are about to destroy
6138 this device. They should clean all the things.
6140 Note that dev->reg_state stays at NETREG_REGISTERED.
6141 This is wanted because this way 8021q and macvlan know
6142 the device is just moving and can keep their slaves up.
6144 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6145 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6146 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6149 * Flush the unicast and multicast chains
6154 /* Actually switch the network namespace */
6155 dev_net_set(dev, net);
6157 /* If there is an ifindex conflict assign a new one */
6158 if (__dev_get_by_index(net, dev->ifindex)) {
6159 int iflink = (dev->iflink == dev->ifindex);
6160 dev->ifindex = dev_new_index(net);
6162 dev->iflink = dev->ifindex;
6165 /* Fixup kobjects */
6166 err = device_rename(&dev->dev, dev->name);
6169 /* Add the device back in the hashes */
6170 list_netdevice(dev);
6172 /* Notify protocols, that a new device appeared. */
6173 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6176 * Prevent userspace races by waiting until the network
6177 * device is fully setup before sending notifications.
6179 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6186 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6188 static int dev_cpu_callback(struct notifier_block *nfb,
6189 unsigned long action,
6192 struct sk_buff **list_skb;
6193 struct sk_buff *skb;
6194 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6195 struct softnet_data *sd, *oldsd;
6197 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6200 local_irq_disable();
6201 cpu = smp_processor_id();
6202 sd = &per_cpu(softnet_data, cpu);
6203 oldsd = &per_cpu(softnet_data, oldcpu);
6205 /* Find end of our completion_queue. */
6206 list_skb = &sd->completion_queue;
6208 list_skb = &(*list_skb)->next;
6209 /* Append completion queue from offline CPU. */
6210 *list_skb = oldsd->completion_queue;
6211 oldsd->completion_queue = NULL;
6213 /* Append output queue from offline CPU. */
6214 if (oldsd->output_queue) {
6215 *sd->output_queue_tailp = oldsd->output_queue;
6216 sd->output_queue_tailp = oldsd->output_queue_tailp;
6217 oldsd->output_queue = NULL;
6218 oldsd->output_queue_tailp = &oldsd->output_queue;
6220 /* Append NAPI poll list from offline CPU. */
6221 if (!list_empty(&oldsd->poll_list)) {
6222 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6223 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6226 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6229 /* Process offline CPU's input_pkt_queue */
6230 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6232 input_queue_head_incr(oldsd);
6234 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6236 input_queue_head_incr(oldsd);
6244 * netdev_increment_features - increment feature set by one
6245 * @all: current feature set
6246 * @one: new feature set
6247 * @mask: mask feature set
6249 * Computes a new feature set after adding a device with feature set
6250 * @one to the master device with current feature set @all. Will not
6251 * enable anything that is off in @mask. Returns the new feature set.
6253 netdev_features_t netdev_increment_features(netdev_features_t all,
6254 netdev_features_t one, netdev_features_t mask)
6256 if (mask & NETIF_F_GEN_CSUM)
6257 mask |= NETIF_F_ALL_CSUM;
6258 mask |= NETIF_F_VLAN_CHALLENGED;
6260 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6261 all &= one | ~NETIF_F_ALL_FOR_ALL;
6263 /* If one device supports hw checksumming, set for all. */
6264 if (all & NETIF_F_GEN_CSUM)
6265 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6269 EXPORT_SYMBOL(netdev_increment_features);
6271 static struct hlist_head *netdev_create_hash(void)
6274 struct hlist_head *hash;
6276 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6278 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6279 INIT_HLIST_HEAD(&hash[i]);
6284 /* Initialize per network namespace state */
6285 static int __net_init netdev_init(struct net *net)
6287 INIT_LIST_HEAD(&net->dev_base_head);
6289 net->dev_name_head = netdev_create_hash();
6290 if (net->dev_name_head == NULL)
6293 net->dev_index_head = netdev_create_hash();
6294 if (net->dev_index_head == NULL)
6300 kfree(net->dev_name_head);
6306 * netdev_drivername - network driver for the device
6307 * @dev: network device
6309 * Determine network driver for device.
6311 const char *netdev_drivername(const struct net_device *dev)
6313 const struct device_driver *driver;
6314 const struct device *parent;
6315 const char *empty = "";
6317 parent = dev->dev.parent;
6321 driver = parent->driver;
6322 if (driver && driver->name)
6323 return driver->name;
6327 int __netdev_printk(const char *level, const struct net_device *dev,
6328 struct va_format *vaf)
6332 if (dev && dev->dev.parent)
6333 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6334 netdev_name(dev), vaf);
6336 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6338 r = printk("%s(NULL net_device): %pV", level, vaf);
6342 EXPORT_SYMBOL(__netdev_printk);
6344 int netdev_printk(const char *level, const struct net_device *dev,
6345 const char *format, ...)
6347 struct va_format vaf;
6351 va_start(args, format);
6356 r = __netdev_printk(level, dev, &vaf);
6361 EXPORT_SYMBOL(netdev_printk);
6363 #define define_netdev_printk_level(func, level) \
6364 int func(const struct net_device *dev, const char *fmt, ...) \
6367 struct va_format vaf; \
6370 va_start(args, fmt); \
6375 r = __netdev_printk(level, dev, &vaf); \
6380 EXPORT_SYMBOL(func);
6382 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6383 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6384 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6385 define_netdev_printk_level(netdev_err, KERN_ERR);
6386 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6387 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6388 define_netdev_printk_level(netdev_info, KERN_INFO);
6390 static void __net_exit netdev_exit(struct net *net)
6392 kfree(net->dev_name_head);
6393 kfree(net->dev_index_head);
6396 static struct pernet_operations __net_initdata netdev_net_ops = {
6397 .init = netdev_init,
6398 .exit = netdev_exit,
6401 static void __net_exit default_device_exit(struct net *net)
6403 struct net_device *dev, *aux;
6405 * Push all migratable network devices back to the
6406 * initial network namespace
6409 for_each_netdev_safe(net, dev, aux) {
6411 char fb_name[IFNAMSIZ];
6413 /* Ignore unmoveable devices (i.e. loopback) */
6414 if (dev->features & NETIF_F_NETNS_LOCAL)
6417 /* Leave virtual devices for the generic cleanup */
6418 if (dev->rtnl_link_ops)
6421 /* Push remaining network devices to init_net */
6422 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6423 err = dev_change_net_namespace(dev, &init_net, fb_name);
6425 pr_emerg("%s: failed to move %s to init_net: %d\n",
6426 __func__, dev->name, err);
6433 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6435 /* At exit all network devices most be removed from a network
6436 * namespace. Do this in the reverse order of registration.
6437 * Do this across as many network namespaces as possible to
6438 * improve batching efficiency.
6440 struct net_device *dev;
6442 LIST_HEAD(dev_kill_list);
6445 list_for_each_entry(net, net_list, exit_list) {
6446 for_each_netdev_reverse(net, dev) {
6447 if (dev->rtnl_link_ops)
6448 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6450 unregister_netdevice_queue(dev, &dev_kill_list);
6453 unregister_netdevice_many(&dev_kill_list);
6454 list_del(&dev_kill_list);
6458 static struct pernet_operations __net_initdata default_device_ops = {
6459 .exit = default_device_exit,
6460 .exit_batch = default_device_exit_batch,
6464 * Initialize the DEV module. At boot time this walks the device list and
6465 * unhooks any devices that fail to initialise (normally hardware not
6466 * present) and leaves us with a valid list of present and active devices.
6471 * This is called single threaded during boot, so no need
6472 * to take the rtnl semaphore.
6474 static int __init net_dev_init(void)
6476 int i, rc = -ENOMEM;
6478 BUG_ON(!dev_boot_phase);
6480 if (dev_proc_init())
6483 if (netdev_kobject_init())
6486 INIT_LIST_HEAD(&ptype_all);
6487 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6488 INIT_LIST_HEAD(&ptype_base[i]);
6490 if (register_pernet_subsys(&netdev_net_ops))
6494 * Initialise the packet receive queues.
6497 for_each_possible_cpu(i) {
6498 struct softnet_data *sd = &per_cpu(softnet_data, i);
6500 memset(sd, 0, sizeof(*sd));
6501 skb_queue_head_init(&sd->input_pkt_queue);
6502 skb_queue_head_init(&sd->process_queue);
6503 sd->completion_queue = NULL;
6504 INIT_LIST_HEAD(&sd->poll_list);
6505 sd->output_queue = NULL;
6506 sd->output_queue_tailp = &sd->output_queue;
6508 sd->csd.func = rps_trigger_softirq;
6514 sd->backlog.poll = process_backlog;
6515 sd->backlog.weight = weight_p;
6516 sd->backlog.gro_list = NULL;
6517 sd->backlog.gro_count = 0;
6522 /* The loopback device is special if any other network devices
6523 * is present in a network namespace the loopback device must
6524 * be present. Since we now dynamically allocate and free the
6525 * loopback device ensure this invariant is maintained by
6526 * keeping the loopback device as the first device on the
6527 * list of network devices. Ensuring the loopback devices
6528 * is the first device that appears and the last network device
6531 if (register_pernet_device(&loopback_net_ops))
6534 if (register_pernet_device(&default_device_ops))
6537 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6538 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6540 hotcpu_notifier(dev_cpu_callback, 0);
6548 subsys_initcall(net_dev_init);
6550 static int __init initialize_hashrnd(void)
6552 get_random_bytes(&hashrnd, sizeof(hashrnd));
6556 late_initcall_sync(initialize_hashrnd);