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 <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/if_tunnel.h>
137 #include <linux/if_pppox.h>
138 #include <linux/ppp_defs.h>
139 #include <linux/net_tstamp.h>
141 #include "net-sysfs.h"
143 /* Instead of increasing this, you should create a hash table. */
144 #define MAX_GRO_SKBS 8
146 /* This should be increased if a protocol with a bigger head is added. */
147 #define GRO_MAX_HEAD (MAX_HEADER + 128)
150 * The list of packet types we will receive (as opposed to discard)
151 * and the routines to invoke.
153 * Why 16. Because with 16 the only overlap we get on a hash of the
154 * low nibble of the protocol value is RARP/SNAP/X.25.
156 * NOTE: That is no longer true with the addition of VLAN tags. Not
157 * sure which should go first, but I bet it won't make much
158 * difference if we are running VLANs. The good news is that
159 * this protocol won't be in the list unless compiled in, so
160 * the average user (w/out VLANs) will not be adversely affected.
177 #define PTYPE_HASH_SIZE (16)
178 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
180 static DEFINE_SPINLOCK(ptype_lock);
181 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
182 static struct list_head ptype_all __read_mostly; /* Taps */
185 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
188 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
190 * Writers must hold the rtnl semaphore while they loop through the
191 * dev_base_head list, and hold dev_base_lock for writing when they do the
192 * actual updates. This allows pure readers to access the list even
193 * while a writer is preparing to update it.
195 * To put it another way, dev_base_lock is held for writing only to
196 * protect against pure readers; the rtnl semaphore provides the
197 * protection against other writers.
199 * See, for example usages, register_netdevice() and
200 * unregister_netdevice(), which must be called with the rtnl
203 DEFINE_RWLOCK(dev_base_lock);
204 EXPORT_SYMBOL(dev_base_lock);
206 static inline void dev_base_seq_inc(struct net *net)
208 while (++net->dev_base_seq == 0);
211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
213 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
214 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
217 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
219 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
222 static inline void rps_lock(struct softnet_data *sd)
225 spin_lock(&sd->input_pkt_queue.lock);
229 static inline void rps_unlock(struct softnet_data *sd)
232 spin_unlock(&sd->input_pkt_queue.lock);
236 /* Device list insertion */
237 static int list_netdevice(struct net_device *dev)
239 struct net *net = dev_net(dev);
243 write_lock_bh(&dev_base_lock);
244 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
245 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
246 hlist_add_head_rcu(&dev->index_hlist,
247 dev_index_hash(net, dev->ifindex));
248 write_unlock_bh(&dev_base_lock);
250 dev_base_seq_inc(net);
255 /* Device list removal
256 * caller must respect a RCU grace period before freeing/reusing dev
258 static void unlist_netdevice(struct net_device *dev)
262 /* Unlink dev from the device chain */
263 write_lock_bh(&dev_base_lock);
264 list_del_rcu(&dev->dev_list);
265 hlist_del_rcu(&dev->name_hlist);
266 hlist_del_rcu(&dev->index_hlist);
267 write_unlock_bh(&dev_base_lock);
269 dev_base_seq_inc(dev_net(dev));
276 static RAW_NOTIFIER_HEAD(netdev_chain);
279 * Device drivers call our routines to queue packets here. We empty the
280 * queue in the local softnet handler.
283 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
284 EXPORT_PER_CPU_SYMBOL(softnet_data);
286 #ifdef CONFIG_LOCKDEP
288 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
289 * according to dev->type
291 static const unsigned short netdev_lock_type[] =
292 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
293 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
294 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
295 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
296 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
297 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
298 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
299 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
300 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
301 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
302 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
303 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
304 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
305 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
306 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
307 ARPHRD_VOID, ARPHRD_NONE};
309 static const char *const netdev_lock_name[] =
310 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
311 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
312 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
313 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
314 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
315 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
316 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
317 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
318 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
319 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
320 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
321 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
322 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
323 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
324 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
325 "_xmit_VOID", "_xmit_NONE"};
327 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
328 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
330 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
334 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
335 if (netdev_lock_type[i] == dev_type)
337 /* the last key is used by default */
338 return ARRAY_SIZE(netdev_lock_type) - 1;
341 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
342 unsigned short dev_type)
346 i = netdev_lock_pos(dev_type);
347 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
348 netdev_lock_name[i]);
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
355 i = netdev_lock_pos(dev->type);
356 lockdep_set_class_and_name(&dev->addr_list_lock,
357 &netdev_addr_lock_key[i],
358 netdev_lock_name[i]);
361 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
362 unsigned short dev_type)
365 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
370 /*******************************************************************************
372 Protocol management and registration routines
374 *******************************************************************************/
377 * Add a protocol ID to the list. Now that the input handler is
378 * smarter we can dispense with all the messy stuff that used to be
381 * BEWARE!!! Protocol handlers, mangling input packets,
382 * MUST BE last in hash buckets and checking protocol handlers
383 * MUST start from promiscuous ptype_all chain in net_bh.
384 * It is true now, do not change it.
385 * Explanation follows: if protocol handler, mangling packet, will
386 * be the first on list, it is not able to sense, that packet
387 * is cloned and should be copied-on-write, so that it will
388 * change it and subsequent readers will get broken packet.
392 static inline struct list_head *ptype_head(const struct packet_type *pt)
394 if (pt->type == htons(ETH_P_ALL))
397 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
401 * dev_add_pack - add packet handler
402 * @pt: packet type declaration
404 * Add a protocol handler to the networking stack. The passed &packet_type
405 * is linked into kernel lists and may not be freed until it has been
406 * removed from the kernel lists.
408 * This call does not sleep therefore it can not
409 * guarantee all CPU's that are in middle of receiving packets
410 * will see the new packet type (until the next received packet).
413 void dev_add_pack(struct packet_type *pt)
415 struct list_head *head = ptype_head(pt);
417 spin_lock(&ptype_lock);
418 list_add_rcu(&pt->list, head);
419 spin_unlock(&ptype_lock);
421 EXPORT_SYMBOL(dev_add_pack);
424 * __dev_remove_pack - remove packet handler
425 * @pt: packet type declaration
427 * Remove a protocol handler that was previously added to the kernel
428 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
429 * from the kernel lists and can be freed or reused once this function
432 * The packet type might still be in use by receivers
433 * and must not be freed until after all the CPU's have gone
434 * through a quiescent state.
436 void __dev_remove_pack(struct packet_type *pt)
438 struct list_head *head = ptype_head(pt);
439 struct packet_type *pt1;
441 spin_lock(&ptype_lock);
443 list_for_each_entry(pt1, head, list) {
445 list_del_rcu(&pt->list);
450 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
452 spin_unlock(&ptype_lock);
454 EXPORT_SYMBOL(__dev_remove_pack);
457 * dev_remove_pack - remove packet handler
458 * @pt: packet type declaration
460 * Remove a protocol handler that was previously added to the kernel
461 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
462 * from the kernel lists and can be freed or reused once this function
465 * This call sleeps to guarantee that no CPU is looking at the packet
468 void dev_remove_pack(struct packet_type *pt)
470 __dev_remove_pack(pt);
474 EXPORT_SYMBOL(dev_remove_pack);
476 /******************************************************************************
478 Device Boot-time Settings Routines
480 *******************************************************************************/
482 /* Boot time configuration table */
483 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
486 * netdev_boot_setup_add - add new setup entry
487 * @name: name of the device
488 * @map: configured settings for the device
490 * Adds new setup entry to the dev_boot_setup list. The function
491 * returns 0 on error and 1 on success. This is a generic routine to
494 static int netdev_boot_setup_add(char *name, struct ifmap *map)
496 struct netdev_boot_setup *s;
500 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
501 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
502 memset(s[i].name, 0, sizeof(s[i].name));
503 strlcpy(s[i].name, name, IFNAMSIZ);
504 memcpy(&s[i].map, map, sizeof(s[i].map));
509 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
513 * netdev_boot_setup_check - check boot time settings
514 * @dev: the netdevice
516 * Check boot time settings for the device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found, 1 if they are.
521 int netdev_boot_setup_check(struct net_device *dev)
523 struct netdev_boot_setup *s = dev_boot_setup;
526 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
527 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
528 !strcmp(dev->name, s[i].name)) {
529 dev->irq = s[i].map.irq;
530 dev->base_addr = s[i].map.base_addr;
531 dev->mem_start = s[i].map.mem_start;
532 dev->mem_end = s[i].map.mem_end;
538 EXPORT_SYMBOL(netdev_boot_setup_check);
542 * netdev_boot_base - get address from boot time settings
543 * @prefix: prefix for network device
544 * @unit: id for network device
546 * Check boot time settings for the base address of device.
547 * The found settings are set for the device to be used
548 * later in the device probing.
549 * Returns 0 if no settings found.
551 unsigned long netdev_boot_base(const char *prefix, int unit)
553 const struct netdev_boot_setup *s = dev_boot_setup;
557 sprintf(name, "%s%d", prefix, unit);
560 * If device already registered then return base of 1
561 * to indicate not to probe for this interface
563 if (__dev_get_by_name(&init_net, name))
566 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
567 if (!strcmp(name, s[i].name))
568 return s[i].map.base_addr;
573 * Saves at boot time configured settings for any netdevice.
575 int __init netdev_boot_setup(char *str)
580 str = get_options(str, ARRAY_SIZE(ints), ints);
585 memset(&map, 0, sizeof(map));
589 map.base_addr = ints[2];
591 map.mem_start = ints[3];
593 map.mem_end = ints[4];
595 /* Add new entry to the list */
596 return netdev_boot_setup_add(str, &map);
599 __setup("netdev=", netdev_boot_setup);
601 /*******************************************************************************
603 Device Interface Subroutines
605 *******************************************************************************/
608 * __dev_get_by_name - find a device by its name
609 * @net: the applicable net namespace
610 * @name: name to find
612 * Find an interface by name. Must be called under RTNL semaphore
613 * or @dev_base_lock. If the name is found a pointer to the device
614 * is returned. If the name is not found then %NULL is returned. The
615 * reference counters are not incremented so the caller must be
616 * careful with locks.
619 struct net_device *__dev_get_by_name(struct net *net, const char *name)
621 struct hlist_node *p;
622 struct net_device *dev;
623 struct hlist_head *head = dev_name_hash(net, name);
625 hlist_for_each_entry(dev, p, head, name_hlist)
626 if (!strncmp(dev->name, name, IFNAMSIZ))
631 EXPORT_SYMBOL(__dev_get_by_name);
634 * dev_get_by_name_rcu - find a device by its name
635 * @net: the applicable net namespace
636 * @name: name to find
638 * Find an interface by name.
639 * If the name is found a pointer to the device is returned.
640 * If the name is not found then %NULL is returned.
641 * The reference counters are not incremented so the caller must be
642 * careful with locks. The caller must hold RCU lock.
645 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
647 struct hlist_node *p;
648 struct net_device *dev;
649 struct hlist_head *head = dev_name_hash(net, name);
651 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
652 if (!strncmp(dev->name, name, IFNAMSIZ))
657 EXPORT_SYMBOL(dev_get_by_name_rcu);
660 * dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. This can be called from any
665 * context and does its own locking. The returned handle has
666 * the usage count incremented and the caller must use dev_put() to
667 * release it when it is no longer needed. %NULL is returned if no
668 * matching device is found.
671 struct net_device *dev_get_by_name(struct net *net, const char *name)
673 struct net_device *dev;
676 dev = dev_get_by_name_rcu(net, name);
682 EXPORT_SYMBOL(dev_get_by_name);
685 * __dev_get_by_index - find a device by its ifindex
686 * @net: the applicable net namespace
687 * @ifindex: index of device
689 * Search for an interface by index. Returns %NULL if the device
690 * is not found or a pointer to the device. The device has not
691 * had its reference counter increased so the caller must be careful
692 * about locking. The caller must hold either the RTNL semaphore
696 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
698 struct hlist_node *p;
699 struct net_device *dev;
700 struct hlist_head *head = dev_index_hash(net, ifindex);
702 hlist_for_each_entry(dev, p, head, index_hlist)
703 if (dev->ifindex == ifindex)
708 EXPORT_SYMBOL(__dev_get_by_index);
711 * dev_get_by_index_rcu - find a device by its ifindex
712 * @net: the applicable net namespace
713 * @ifindex: index of device
715 * Search for an interface by index. Returns %NULL if the device
716 * is not found or a pointer to the device. The device has not
717 * had its reference counter increased so the caller must be careful
718 * about locking. The caller must hold RCU lock.
721 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
723 struct hlist_node *p;
724 struct net_device *dev;
725 struct hlist_head *head = dev_index_hash(net, ifindex);
727 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
728 if (dev->ifindex == ifindex)
733 EXPORT_SYMBOL(dev_get_by_index_rcu);
737 * dev_get_by_index - find a device by its ifindex
738 * @net: the applicable net namespace
739 * @ifindex: index of device
741 * Search for an interface by index. Returns NULL if the device
742 * is not found or a pointer to the device. The device returned has
743 * had a reference added and the pointer is safe until the user calls
744 * dev_put to indicate they have finished with it.
747 struct net_device *dev_get_by_index(struct net *net, int ifindex)
749 struct net_device *dev;
752 dev = dev_get_by_index_rcu(net, ifindex);
758 EXPORT_SYMBOL(dev_get_by_index);
761 * dev_getbyhwaddr_rcu - find a device by its hardware address
762 * @net: the applicable net namespace
763 * @type: media type of device
764 * @ha: hardware address
766 * Search for an interface by MAC address. Returns NULL if the device
767 * is not found or a pointer to the device.
768 * The caller must hold RCU or RTNL.
769 * The returned device has not had its ref count increased
770 * and the caller must therefore be careful about locking
774 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
777 struct net_device *dev;
779 for_each_netdev_rcu(net, dev)
780 if (dev->type == type &&
781 !memcmp(dev->dev_addr, ha, dev->addr_len))
786 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
788 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
790 struct net_device *dev;
793 for_each_netdev(net, dev)
794 if (dev->type == type)
799 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
801 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
803 struct net_device *dev, *ret = NULL;
806 for_each_netdev_rcu(net, dev)
807 if (dev->type == type) {
815 EXPORT_SYMBOL(dev_getfirstbyhwtype);
818 * dev_get_by_flags_rcu - find any device with given flags
819 * @net: the applicable net namespace
820 * @if_flags: IFF_* values
821 * @mask: bitmask of bits in if_flags to check
823 * Search for any interface with the given flags. Returns NULL if a device
824 * is not found or a pointer to the device. Must be called inside
825 * rcu_read_lock(), and result refcount is unchanged.
828 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
831 struct net_device *dev, *ret;
834 for_each_netdev_rcu(net, dev) {
835 if (((dev->flags ^ if_flags) & mask) == 0) {
842 EXPORT_SYMBOL(dev_get_by_flags_rcu);
845 * dev_valid_name - check if name is okay for network device
848 * Network device names need to be valid file names to
849 * to allow sysfs to work. We also disallow any kind of
852 int dev_valid_name(const char *name)
856 if (strlen(name) >= IFNAMSIZ)
858 if (!strcmp(name, ".") || !strcmp(name, ".."))
862 if (*name == '/' || isspace(*name))
868 EXPORT_SYMBOL(dev_valid_name);
871 * __dev_alloc_name - allocate a name for a device
872 * @net: network namespace to allocate the device name in
873 * @name: name format string
874 * @buf: scratch buffer and result name string
876 * Passed a format string - eg "lt%d" it will try and find a suitable
877 * id. It scans list of devices to build up a free map, then chooses
878 * the first empty slot. The caller must hold the dev_base or rtnl lock
879 * while allocating the name and adding the device in order to avoid
881 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
882 * Returns the number of the unit assigned or a negative errno code.
885 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
889 const int max_netdevices = 8*PAGE_SIZE;
890 unsigned long *inuse;
891 struct net_device *d;
893 p = strnchr(name, IFNAMSIZ-1, '%');
896 * Verify the string as this thing may have come from
897 * the user. There must be either one "%d" and no other "%"
900 if (p[1] != 'd' || strchr(p + 2, '%'))
903 /* Use one page as a bit array of possible slots */
904 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
908 for_each_netdev(net, d) {
909 if (!sscanf(d->name, name, &i))
911 if (i < 0 || i >= max_netdevices)
914 /* avoid cases where sscanf is not exact inverse of printf */
915 snprintf(buf, IFNAMSIZ, name, i);
916 if (!strncmp(buf, d->name, IFNAMSIZ))
920 i = find_first_zero_bit(inuse, max_netdevices);
921 free_page((unsigned long) inuse);
925 snprintf(buf, IFNAMSIZ, name, i);
926 if (!__dev_get_by_name(net, buf))
929 /* It is possible to run out of possible slots
930 * when the name is long and there isn't enough space left
931 * for the digits, or if all bits are used.
937 * dev_alloc_name - allocate a name for a device
939 * @name: name format string
941 * Passed a format string - eg "lt%d" it will try and find a suitable
942 * id. It scans list of devices to build up a free map, then chooses
943 * the first empty slot. The caller must hold the dev_base or rtnl lock
944 * while allocating the name and adding the device in order to avoid
946 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
947 * Returns the number of the unit assigned or a negative errno code.
950 int dev_alloc_name(struct net_device *dev, const char *name)
956 BUG_ON(!dev_net(dev));
958 ret = __dev_alloc_name(net, name, buf);
960 strlcpy(dev->name, buf, IFNAMSIZ);
963 EXPORT_SYMBOL(dev_alloc_name);
965 static int dev_get_valid_name(struct net_device *dev, const char *name)
969 BUG_ON(!dev_net(dev));
972 if (!dev_valid_name(name))
975 if (strchr(name, '%'))
976 return dev_alloc_name(dev, name);
977 else if (__dev_get_by_name(net, name))
979 else if (dev->name != name)
980 strlcpy(dev->name, name, IFNAMSIZ);
986 * dev_change_name - change name of a device
988 * @newname: name (or format string) must be at least IFNAMSIZ
990 * Change name of a device, can pass format strings "eth%d".
993 int dev_change_name(struct net_device *dev, const char *newname)
995 char oldname[IFNAMSIZ];
1001 BUG_ON(!dev_net(dev));
1004 if (dev->flags & IFF_UP)
1007 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1010 memcpy(oldname, dev->name, IFNAMSIZ);
1012 err = dev_get_valid_name(dev, newname);
1017 ret = device_rename(&dev->dev, dev->name);
1019 memcpy(dev->name, oldname, IFNAMSIZ);
1023 write_lock_bh(&dev_base_lock);
1024 hlist_del_rcu(&dev->name_hlist);
1025 write_unlock_bh(&dev_base_lock);
1029 write_lock_bh(&dev_base_lock);
1030 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1031 write_unlock_bh(&dev_base_lock);
1033 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1034 ret = notifier_to_errno(ret);
1037 /* err >= 0 after dev_alloc_name() or stores the first errno */
1040 memcpy(dev->name, oldname, IFNAMSIZ);
1044 "%s: name change rollback failed: %d.\n",
1053 * dev_set_alias - change ifalias of a device
1055 * @alias: name up to IFALIASZ
1056 * @len: limit of bytes to copy from info
1058 * Set ifalias for a device,
1060 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1066 if (len >= IFALIASZ)
1071 kfree(dev->ifalias);
1072 dev->ifalias = NULL;
1077 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1080 dev->ifalias = new_ifalias;
1082 strlcpy(dev->ifalias, alias, len+1);
1088 * netdev_features_change - device changes features
1089 * @dev: device to cause notification
1091 * Called to indicate a device has changed features.
1093 void netdev_features_change(struct net_device *dev)
1095 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1097 EXPORT_SYMBOL(netdev_features_change);
1100 * netdev_state_change - device changes state
1101 * @dev: device to cause notification
1103 * Called to indicate a device has changed state. This function calls
1104 * the notifier chains for netdev_chain and sends a NEWLINK message
1105 * to the routing socket.
1107 void netdev_state_change(struct net_device *dev)
1109 if (dev->flags & IFF_UP) {
1110 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1111 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1114 EXPORT_SYMBOL(netdev_state_change);
1116 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1118 return call_netdevice_notifiers(event, dev);
1120 EXPORT_SYMBOL(netdev_bonding_change);
1123 * dev_load - load a network module
1124 * @net: the applicable net namespace
1125 * @name: name of interface
1127 * If a network interface is not present and the process has suitable
1128 * privileges this function loads the module. If module loading is not
1129 * available in this kernel then it becomes a nop.
1132 void dev_load(struct net *net, const char *name)
1134 struct net_device *dev;
1138 dev = dev_get_by_name_rcu(net, name);
1142 if (no_module && capable(CAP_NET_ADMIN))
1143 no_module = request_module("netdev-%s", name);
1144 if (no_module && capable(CAP_SYS_MODULE)) {
1145 if (!request_module("%s", name))
1146 pr_err("Loading kernel module for a network device "
1147 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1151 EXPORT_SYMBOL(dev_load);
1153 static int __dev_open(struct net_device *dev)
1155 const struct net_device_ops *ops = dev->netdev_ops;
1160 if (!netif_device_present(dev))
1163 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1164 ret = notifier_to_errno(ret);
1168 set_bit(__LINK_STATE_START, &dev->state);
1170 if (ops->ndo_validate_addr)
1171 ret = ops->ndo_validate_addr(dev);
1173 if (!ret && ops->ndo_open)
1174 ret = ops->ndo_open(dev);
1177 clear_bit(__LINK_STATE_START, &dev->state);
1179 dev->flags |= IFF_UP;
1180 net_dmaengine_get();
1181 dev_set_rx_mode(dev);
1183 add_device_randomness(dev->dev_addr, dev->addr_len);
1190 * dev_open - prepare an interface for use.
1191 * @dev: device to open
1193 * Takes a device from down to up state. The device's private open
1194 * function is invoked and then the multicast lists are loaded. Finally
1195 * the device is moved into the up state and a %NETDEV_UP message is
1196 * sent to the netdev notifier chain.
1198 * Calling this function on an active interface is a nop. On a failure
1199 * a negative errno code is returned.
1201 int dev_open(struct net_device *dev)
1205 if (dev->flags & IFF_UP)
1208 ret = __dev_open(dev);
1212 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1213 call_netdevice_notifiers(NETDEV_UP, dev);
1217 EXPORT_SYMBOL(dev_open);
1219 static int __dev_close_many(struct list_head *head)
1221 struct net_device *dev;
1226 list_for_each_entry(dev, head, unreg_list) {
1227 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1229 clear_bit(__LINK_STATE_START, &dev->state);
1231 /* Synchronize to scheduled poll. We cannot touch poll list, it
1232 * can be even on different cpu. So just clear netif_running().
1234 * dev->stop() will invoke napi_disable() on all of it's
1235 * napi_struct instances on this device.
1237 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1240 dev_deactivate_many(head);
1242 list_for_each_entry(dev, head, unreg_list) {
1243 const struct net_device_ops *ops = dev->netdev_ops;
1246 * Call the device specific close. This cannot fail.
1247 * Only if device is UP
1249 * We allow it to be called even after a DETACH hot-plug
1255 dev->flags &= ~IFF_UP;
1256 net_dmaengine_put();
1262 static int __dev_close(struct net_device *dev)
1267 list_add(&dev->unreg_list, &single);
1268 retval = __dev_close_many(&single);
1273 static int dev_close_many(struct list_head *head)
1275 struct net_device *dev, *tmp;
1276 LIST_HEAD(tmp_list);
1278 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1279 if (!(dev->flags & IFF_UP))
1280 list_move(&dev->unreg_list, &tmp_list);
1282 __dev_close_many(head);
1284 list_for_each_entry(dev, head, unreg_list) {
1285 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1286 call_netdevice_notifiers(NETDEV_DOWN, dev);
1289 /* rollback_registered_many needs the complete original list */
1290 list_splice(&tmp_list, head);
1295 * dev_close - shutdown an interface.
1296 * @dev: device to shutdown
1298 * This function moves an active device into down state. A
1299 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1300 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1303 int dev_close(struct net_device *dev)
1305 if (dev->flags & IFF_UP) {
1308 list_add(&dev->unreg_list, &single);
1309 dev_close_many(&single);
1314 EXPORT_SYMBOL(dev_close);
1318 * dev_disable_lro - disable Large Receive Offload on a device
1321 * Disable Large Receive Offload (LRO) on a net device. Must be
1322 * called under RTNL. This is needed if received packets may be
1323 * forwarded to another interface.
1325 void dev_disable_lro(struct net_device *dev)
1330 * If we're trying to disable lro on a vlan device
1331 * use the underlying physical device instead
1333 if (is_vlan_dev(dev))
1334 dev = vlan_dev_real_dev(dev);
1336 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1337 flags = dev->ethtool_ops->get_flags(dev);
1339 flags = ethtool_op_get_flags(dev);
1341 if (!(flags & ETH_FLAG_LRO))
1344 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1345 if (unlikely(dev->features & NETIF_F_LRO))
1346 netdev_WARN(dev, "failed to disable LRO!\n");
1348 EXPORT_SYMBOL(dev_disable_lro);
1351 static int dev_boot_phase = 1;
1354 * register_netdevice_notifier - register a network notifier block
1357 * Register a notifier to be called when network device events occur.
1358 * The notifier passed is linked into the kernel structures and must
1359 * not be reused until it has been unregistered. A negative errno code
1360 * is returned on a failure.
1362 * When registered all registration and up events are replayed
1363 * to the new notifier to allow device to have a race free
1364 * view of the network device list.
1367 int register_netdevice_notifier(struct notifier_block *nb)
1369 struct net_device *dev;
1370 struct net_device *last;
1375 err = raw_notifier_chain_register(&netdev_chain, nb);
1381 for_each_netdev(net, dev) {
1382 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1383 err = notifier_to_errno(err);
1387 if (!(dev->flags & IFF_UP))
1390 nb->notifier_call(nb, NETDEV_UP, dev);
1401 for_each_netdev(net, dev) {
1405 if (dev->flags & IFF_UP) {
1406 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1407 nb->notifier_call(nb, NETDEV_DOWN, dev);
1409 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1410 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1415 raw_notifier_chain_unregister(&netdev_chain, nb);
1418 EXPORT_SYMBOL(register_netdevice_notifier);
1421 * unregister_netdevice_notifier - unregister a network notifier block
1424 * Unregister a notifier previously registered by
1425 * register_netdevice_notifier(). The notifier is unlinked into the
1426 * kernel structures and may then be reused. A negative errno code
1427 * is returned on a failure.
1429 * After unregistering unregister and down device events are synthesized
1430 * for all devices on the device list to the removed notifier to remove
1431 * the need for special case cleanup code.
1434 int unregister_netdevice_notifier(struct notifier_block *nb)
1436 struct net_device *dev;
1441 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1446 for_each_netdev(net, dev) {
1447 if (dev->flags & IFF_UP) {
1448 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1449 nb->notifier_call(nb, NETDEV_DOWN, dev);
1451 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1452 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1459 EXPORT_SYMBOL(unregister_netdevice_notifier);
1462 * call_netdevice_notifiers - call all network notifier blocks
1463 * @val: value passed unmodified to notifier function
1464 * @dev: net_device pointer passed unmodified to notifier function
1466 * Call all network notifier blocks. Parameters and return value
1467 * are as for raw_notifier_call_chain().
1470 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1473 return raw_notifier_call_chain(&netdev_chain, val, dev);
1475 EXPORT_SYMBOL(call_netdevice_notifiers);
1477 /* When > 0 there are consumers of rx skb time stamps */
1478 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1480 void net_enable_timestamp(void)
1482 atomic_inc(&netstamp_needed);
1484 EXPORT_SYMBOL(net_enable_timestamp);
1486 void net_disable_timestamp(void)
1488 atomic_dec(&netstamp_needed);
1490 EXPORT_SYMBOL(net_disable_timestamp);
1492 static inline void net_timestamp_set(struct sk_buff *skb)
1494 if (atomic_read(&netstamp_needed))
1495 __net_timestamp(skb);
1497 skb->tstamp.tv64 = 0;
1500 static inline void net_timestamp_check(struct sk_buff *skb)
1502 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1503 __net_timestamp(skb);
1506 static int net_hwtstamp_validate(struct ifreq *ifr)
1508 struct hwtstamp_config cfg;
1509 enum hwtstamp_tx_types tx_type;
1510 enum hwtstamp_rx_filters rx_filter;
1511 int tx_type_valid = 0;
1512 int rx_filter_valid = 0;
1514 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1517 if (cfg.flags) /* reserved for future extensions */
1520 tx_type = cfg.tx_type;
1521 rx_filter = cfg.rx_filter;
1524 case HWTSTAMP_TX_OFF:
1525 case HWTSTAMP_TX_ON:
1526 case HWTSTAMP_TX_ONESTEP_SYNC:
1531 switch (rx_filter) {
1532 case HWTSTAMP_FILTER_NONE:
1533 case HWTSTAMP_FILTER_ALL:
1534 case HWTSTAMP_FILTER_SOME:
1535 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1536 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1537 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1538 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1539 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1540 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1541 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1542 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1543 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1544 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1545 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1546 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1547 rx_filter_valid = 1;
1551 if (!tx_type_valid || !rx_filter_valid)
1557 static inline bool is_skb_forwardable(struct net_device *dev,
1558 struct sk_buff *skb)
1562 if (!(dev->flags & IFF_UP))
1565 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1566 if (skb->len <= len)
1569 /* if TSO is enabled, we don't care about the length as the packet
1570 * could be forwarded without being segmented before
1572 if (skb_is_gso(skb))
1579 * dev_forward_skb - loopback an skb to another netif
1581 * @dev: destination network device
1582 * @skb: buffer to forward
1585 * NET_RX_SUCCESS (no congestion)
1586 * NET_RX_DROP (packet was dropped, but freed)
1588 * dev_forward_skb can be used for injecting an skb from the
1589 * start_xmit function of one device into the receive queue
1590 * of another device.
1592 * The receiving device may be in another namespace, so
1593 * we have to clear all information in the skb that could
1594 * impact namespace isolation.
1596 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1598 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1599 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1600 atomic_long_inc(&dev->rx_dropped);
1609 if (unlikely(!is_skb_forwardable(dev, skb))) {
1610 atomic_long_inc(&dev->rx_dropped);
1616 skb->tstamp.tv64 = 0;
1617 skb->pkt_type = PACKET_HOST;
1618 skb->protocol = eth_type_trans(skb, dev);
1622 return netif_rx(skb);
1624 EXPORT_SYMBOL_GPL(dev_forward_skb);
1626 static inline int deliver_skb(struct sk_buff *skb,
1627 struct packet_type *pt_prev,
1628 struct net_device *orig_dev)
1630 atomic_inc(&skb->users);
1631 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1634 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1636 if (ptype->af_packet_priv == NULL)
1639 if (ptype->id_match)
1640 return ptype->id_match(ptype, skb->sk);
1641 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1648 * Support routine. Sends outgoing frames to any network
1649 * taps currently in use.
1652 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1654 struct packet_type *ptype;
1655 struct sk_buff *skb2 = NULL;
1656 struct packet_type *pt_prev = NULL;
1659 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1660 /* Never send packets back to the socket
1661 * they originated from - MvS (miquels@drinkel.ow.org)
1663 if ((ptype->dev == dev || !ptype->dev) &&
1664 (!skb_loop_sk(ptype, skb))) {
1666 deliver_skb(skb2, pt_prev, skb->dev);
1671 skb2 = skb_clone(skb, GFP_ATOMIC);
1675 net_timestamp_set(skb2);
1677 /* skb->nh should be correctly
1678 set by sender, so that the second statement is
1679 just protection against buggy protocols.
1681 skb_reset_mac_header(skb2);
1683 if (skb_network_header(skb2) < skb2->data ||
1684 skb2->network_header > skb2->tail) {
1685 if (net_ratelimit())
1686 printk(KERN_CRIT "protocol %04x is "
1688 ntohs(skb2->protocol),
1690 skb_reset_network_header(skb2);
1693 skb2->transport_header = skb2->network_header;
1694 skb2->pkt_type = PACKET_OUTGOING;
1699 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1703 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1704 * @dev: Network device
1705 * @txq: number of queues available
1707 * If real_num_tx_queues is changed the tc mappings may no longer be
1708 * valid. To resolve this verify the tc mapping remains valid and if
1709 * not NULL the mapping. With no priorities mapping to this
1710 * offset/count pair it will no longer be used. In the worst case TC0
1711 * is invalid nothing can be done so disable priority mappings. If is
1712 * expected that drivers will fix this mapping if they can before
1713 * calling netif_set_real_num_tx_queues.
1715 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1718 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1720 /* If TC0 is invalidated disable TC mapping */
1721 if (tc->offset + tc->count > txq) {
1722 pr_warning("Number of in use tx queues changed "
1723 "invalidating tc mappings. Priority "
1724 "traffic classification disabled!\n");
1729 /* Invalidated prio to tc mappings set to TC0 */
1730 for (i = 1; i < TC_BITMASK + 1; i++) {
1731 int q = netdev_get_prio_tc_map(dev, i);
1733 tc = &dev->tc_to_txq[q];
1734 if (tc->offset + tc->count > txq) {
1735 pr_warning("Number of in use tx queues "
1736 "changed. Priority %i to tc "
1737 "mapping %i is no longer valid "
1738 "setting map to 0\n",
1740 netdev_set_prio_tc_map(dev, i, 0);
1746 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1747 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1749 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1753 if (txq < 1 || txq > dev->num_tx_queues)
1756 if (dev->reg_state == NETREG_REGISTERED ||
1757 dev->reg_state == NETREG_UNREGISTERING) {
1760 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1766 netif_setup_tc(dev, txq);
1768 if (txq < dev->real_num_tx_queues)
1769 qdisc_reset_all_tx_gt(dev, txq);
1772 dev->real_num_tx_queues = txq;
1775 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1779 * netif_set_real_num_rx_queues - set actual number of RX queues used
1780 * @dev: Network device
1781 * @rxq: Actual number of RX queues
1783 * This must be called either with the rtnl_lock held or before
1784 * registration of the net device. Returns 0 on success, or a
1785 * negative error code. If called before registration, it always
1788 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1792 if (rxq < 1 || rxq > dev->num_rx_queues)
1795 if (dev->reg_state == NETREG_REGISTERED) {
1798 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1804 dev->real_num_rx_queues = rxq;
1807 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1810 static inline void __netif_reschedule(struct Qdisc *q)
1812 struct softnet_data *sd;
1813 unsigned long flags;
1815 local_irq_save(flags);
1816 sd = &__get_cpu_var(softnet_data);
1817 q->next_sched = NULL;
1818 *sd->output_queue_tailp = q;
1819 sd->output_queue_tailp = &q->next_sched;
1820 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1821 local_irq_restore(flags);
1824 void __netif_schedule(struct Qdisc *q)
1826 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1827 __netif_reschedule(q);
1829 EXPORT_SYMBOL(__netif_schedule);
1831 void dev_kfree_skb_irq(struct sk_buff *skb)
1833 if (atomic_dec_and_test(&skb->users)) {
1834 struct softnet_data *sd;
1835 unsigned long flags;
1837 local_irq_save(flags);
1838 sd = &__get_cpu_var(softnet_data);
1839 skb->next = sd->completion_queue;
1840 sd->completion_queue = skb;
1841 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1842 local_irq_restore(flags);
1845 EXPORT_SYMBOL(dev_kfree_skb_irq);
1847 void dev_kfree_skb_any(struct sk_buff *skb)
1849 if (in_irq() || irqs_disabled())
1850 dev_kfree_skb_irq(skb);
1854 EXPORT_SYMBOL(dev_kfree_skb_any);
1858 * netif_device_detach - mark device as removed
1859 * @dev: network device
1861 * Mark device as removed from system and therefore no longer available.
1863 void netif_device_detach(struct net_device *dev)
1865 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1866 netif_running(dev)) {
1867 netif_tx_stop_all_queues(dev);
1870 EXPORT_SYMBOL(netif_device_detach);
1873 * netif_device_attach - mark device as attached
1874 * @dev: network device
1876 * Mark device as attached from system and restart if needed.
1878 void netif_device_attach(struct net_device *dev)
1880 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1881 netif_running(dev)) {
1882 netif_tx_wake_all_queues(dev);
1883 __netdev_watchdog_up(dev);
1886 EXPORT_SYMBOL(netif_device_attach);
1889 * Invalidate hardware checksum when packet is to be mangled, and
1890 * complete checksum manually on outgoing path.
1892 int skb_checksum_help(struct sk_buff *skb)
1895 int ret = 0, offset;
1897 if (skb->ip_summed == CHECKSUM_COMPLETE)
1898 goto out_set_summed;
1900 if (unlikely(skb_shinfo(skb)->gso_size)) {
1901 /* Let GSO fix up the checksum. */
1902 goto out_set_summed;
1905 offset = skb_checksum_start_offset(skb);
1906 BUG_ON(offset >= skb_headlen(skb));
1907 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1909 offset += skb->csum_offset;
1910 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1912 if (skb_cloned(skb) &&
1913 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1914 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1919 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1921 skb->ip_summed = CHECKSUM_NONE;
1925 EXPORT_SYMBOL(skb_checksum_help);
1928 * skb_gso_segment - Perform segmentation on skb.
1929 * @skb: buffer to segment
1930 * @features: features for the output path (see dev->features)
1932 * This function segments the given skb and returns a list of segments.
1934 * It may return NULL if the skb requires no segmentation. This is
1935 * only possible when GSO is used for verifying header integrity.
1937 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1939 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1940 struct packet_type *ptype;
1941 __be16 type = skb->protocol;
1942 int vlan_depth = ETH_HLEN;
1945 while (type == htons(ETH_P_8021Q)) {
1946 struct vlan_hdr *vh;
1948 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1949 return ERR_PTR(-EINVAL);
1951 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1952 type = vh->h_vlan_encapsulated_proto;
1953 vlan_depth += VLAN_HLEN;
1956 skb_reset_mac_header(skb);
1957 skb->mac_len = skb->network_header - skb->mac_header;
1958 __skb_pull(skb, skb->mac_len);
1960 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1961 struct net_device *dev = skb->dev;
1962 struct ethtool_drvinfo info = {};
1964 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1965 dev->ethtool_ops->get_drvinfo(dev, &info);
1967 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1968 info.driver, dev ? dev->features : 0L,
1969 skb->sk ? skb->sk->sk_route_caps : 0L,
1970 skb->len, skb->data_len, skb->ip_summed);
1972 if (skb_header_cloned(skb) &&
1973 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1974 return ERR_PTR(err);
1978 list_for_each_entry_rcu(ptype,
1979 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1980 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1981 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1982 err = ptype->gso_send_check(skb);
1983 segs = ERR_PTR(err);
1984 if (err || skb_gso_ok(skb, features))
1986 __skb_push(skb, (skb->data -
1987 skb_network_header(skb)));
1989 segs = ptype->gso_segment(skb, features);
1995 __skb_push(skb, skb->data - skb_mac_header(skb));
1999 EXPORT_SYMBOL(skb_gso_segment);
2001 /* Take action when hardware reception checksum errors are detected. */
2003 void netdev_rx_csum_fault(struct net_device *dev)
2005 if (net_ratelimit()) {
2006 printk(KERN_ERR "%s: hw csum failure.\n",
2007 dev ? dev->name : "<unknown>");
2011 EXPORT_SYMBOL(netdev_rx_csum_fault);
2014 /* Actually, we should eliminate this check as soon as we know, that:
2015 * 1. IOMMU is present and allows to map all the memory.
2016 * 2. No high memory really exists on this machine.
2019 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2021 #ifdef CONFIG_HIGHMEM
2023 if (!(dev->features & NETIF_F_HIGHDMA)) {
2024 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2025 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2026 if (PageHighMem(skb_frag_page(frag)))
2031 if (PCI_DMA_BUS_IS_PHYS) {
2032 struct device *pdev = dev->dev.parent;
2036 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2037 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2038 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2039 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2048 void (*destructor)(struct sk_buff *skb);
2051 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2053 static void dev_gso_skb_destructor(struct sk_buff *skb)
2055 struct dev_gso_cb *cb;
2058 struct sk_buff *nskb = skb->next;
2060 skb->next = nskb->next;
2063 } while (skb->next);
2065 cb = DEV_GSO_CB(skb);
2067 cb->destructor(skb);
2071 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2072 * @skb: buffer to segment
2073 * @features: device features as applicable to this skb
2075 * This function segments the given skb and stores the list of segments
2078 static int dev_gso_segment(struct sk_buff *skb, int features)
2080 struct sk_buff *segs;
2082 segs = skb_gso_segment(skb, features);
2084 /* Verifying header integrity only. */
2089 return PTR_ERR(segs);
2092 DEV_GSO_CB(skb)->destructor = skb->destructor;
2093 skb->destructor = dev_gso_skb_destructor;
2098 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2100 return ((features & NETIF_F_GEN_CSUM) ||
2101 ((features & NETIF_F_V4_CSUM) &&
2102 protocol == htons(ETH_P_IP)) ||
2103 ((features & NETIF_F_V6_CSUM) &&
2104 protocol == htons(ETH_P_IPV6)) ||
2105 ((features & NETIF_F_FCOE_CRC) &&
2106 protocol == htons(ETH_P_FCOE)));
2109 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2111 if (!can_checksum_protocol(features, protocol)) {
2112 features &= ~NETIF_F_ALL_CSUM;
2113 features &= ~NETIF_F_SG;
2114 } else if (illegal_highdma(skb->dev, skb)) {
2115 features &= ~NETIF_F_SG;
2121 u32 netif_skb_features(struct sk_buff *skb)
2123 __be16 protocol = skb->protocol;
2124 u32 features = skb->dev->features;
2126 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2127 features &= ~NETIF_F_GSO_MASK;
2129 if (protocol == htons(ETH_P_8021Q)) {
2130 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2131 protocol = veh->h_vlan_encapsulated_proto;
2132 } else if (!vlan_tx_tag_present(skb)) {
2133 return harmonize_features(skb, protocol, features);
2136 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2138 if (protocol != htons(ETH_P_8021Q)) {
2139 return harmonize_features(skb, protocol, features);
2141 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2142 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2143 return harmonize_features(skb, protocol, features);
2146 EXPORT_SYMBOL(netif_skb_features);
2149 * Returns true if either:
2150 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2151 * 2. skb is fragmented and the device does not support SG, or if
2152 * at least one of fragments is in highmem and device does not
2153 * support DMA from it.
2155 static inline int skb_needs_linearize(struct sk_buff *skb,
2158 return skb_is_nonlinear(skb) &&
2159 ((skb_has_frag_list(skb) &&
2160 !(features & NETIF_F_FRAGLIST)) ||
2161 (skb_shinfo(skb)->nr_frags &&
2162 !(features & NETIF_F_SG)));
2165 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2166 struct netdev_queue *txq)
2168 const struct net_device_ops *ops = dev->netdev_ops;
2169 int rc = NETDEV_TX_OK;
2170 unsigned int skb_len;
2172 if (likely(!skb->next)) {
2176 * If device doesn't need skb->dst, release it right now while
2177 * its hot in this cpu cache
2179 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2182 if (!list_empty(&ptype_all))
2183 dev_queue_xmit_nit(skb, dev);
2185 features = netif_skb_features(skb);
2187 if (vlan_tx_tag_present(skb) &&
2188 !(features & NETIF_F_HW_VLAN_TX)) {
2189 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2196 if (netif_needs_gso(skb, features)) {
2197 if (unlikely(dev_gso_segment(skb, features)))
2202 if (skb_needs_linearize(skb, features) &&
2203 __skb_linearize(skb))
2206 /* If packet is not checksummed and device does not
2207 * support checksumming for this protocol, complete
2208 * checksumming here.
2210 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2211 skb_set_transport_header(skb,
2212 skb_checksum_start_offset(skb));
2213 if (!(features & NETIF_F_ALL_CSUM) &&
2214 skb_checksum_help(skb))
2220 rc = ops->ndo_start_xmit(skb, dev);
2221 trace_net_dev_xmit(skb, rc, dev, skb_len);
2222 if (rc == NETDEV_TX_OK)
2223 txq_trans_update(txq);
2229 struct sk_buff *nskb = skb->next;
2231 skb->next = nskb->next;
2235 * If device doesn't need nskb->dst, release it right now while
2236 * its hot in this cpu cache
2238 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2241 skb_len = nskb->len;
2242 rc = ops->ndo_start_xmit(nskb, dev);
2243 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2244 if (unlikely(rc != NETDEV_TX_OK)) {
2245 if (rc & ~NETDEV_TX_MASK)
2246 goto out_kfree_gso_skb;
2247 nskb->next = skb->next;
2251 txq_trans_update(txq);
2252 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2253 return NETDEV_TX_BUSY;
2254 } while (skb->next);
2257 if (likely(skb->next == NULL))
2258 skb->destructor = DEV_GSO_CB(skb)->destructor;
2265 static u32 hashrnd __read_mostly;
2268 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2269 * to be used as a distribution range.
2271 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2272 unsigned int num_tx_queues)
2276 u16 qcount = num_tx_queues;
2278 if (skb_rx_queue_recorded(skb)) {
2279 hash = skb_get_rx_queue(skb);
2280 while (unlikely(hash >= num_tx_queues))
2281 hash -= num_tx_queues;
2286 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2287 qoffset = dev->tc_to_txq[tc].offset;
2288 qcount = dev->tc_to_txq[tc].count;
2291 if (skb->sk && skb->sk->sk_hash)
2292 hash = skb->sk->sk_hash;
2294 hash = (__force u16) skb->protocol;
2295 hash = jhash_1word(hash, hashrnd);
2297 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2299 EXPORT_SYMBOL(__skb_tx_hash);
2301 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2303 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2304 if (net_ratelimit()) {
2305 pr_warning("%s selects TX queue %d, but "
2306 "real number of TX queues is %d\n",
2307 dev->name, queue_index, dev->real_num_tx_queues);
2314 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2317 struct xps_dev_maps *dev_maps;
2318 struct xps_map *map;
2319 int queue_index = -1;
2322 dev_maps = rcu_dereference(dev->xps_maps);
2324 map = rcu_dereference(
2325 dev_maps->cpu_map[raw_smp_processor_id()]);
2328 queue_index = map->queues[0];
2331 if (skb->sk && skb->sk->sk_hash)
2332 hash = skb->sk->sk_hash;
2334 hash = (__force u16) skb->protocol ^
2336 hash = jhash_1word(hash, hashrnd);
2337 queue_index = map->queues[
2338 ((u64)hash * map->len) >> 32];
2340 if (unlikely(queue_index >= dev->real_num_tx_queues))
2352 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2353 struct sk_buff *skb)
2356 const struct net_device_ops *ops = dev->netdev_ops;
2358 if (dev->real_num_tx_queues == 1)
2360 else if (ops->ndo_select_queue) {
2361 queue_index = ops->ndo_select_queue(dev, skb);
2362 queue_index = dev_cap_txqueue(dev, queue_index);
2364 struct sock *sk = skb->sk;
2365 queue_index = sk_tx_queue_get(sk);
2367 if (queue_index < 0 || skb->ooo_okay ||
2368 queue_index >= dev->real_num_tx_queues) {
2369 int old_index = queue_index;
2371 queue_index = get_xps_queue(dev, skb);
2372 if (queue_index < 0)
2373 queue_index = skb_tx_hash(dev, skb);
2375 if (queue_index != old_index && sk) {
2376 struct dst_entry *dst =
2377 rcu_dereference_check(sk->sk_dst_cache, 1);
2379 if (dst && skb_dst(skb) == dst)
2380 sk_tx_queue_set(sk, queue_index);
2385 skb_set_queue_mapping(skb, queue_index);
2386 return netdev_get_tx_queue(dev, queue_index);
2389 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2390 struct net_device *dev,
2391 struct netdev_queue *txq)
2393 spinlock_t *root_lock = qdisc_lock(q);
2397 qdisc_skb_cb(skb)->pkt_len = skb->len;
2398 qdisc_calculate_pkt_len(skb, q);
2400 * Heuristic to force contended enqueues to serialize on a
2401 * separate lock before trying to get qdisc main lock.
2402 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2403 * and dequeue packets faster.
2405 contended = qdisc_is_running(q);
2406 if (unlikely(contended))
2407 spin_lock(&q->busylock);
2409 spin_lock(root_lock);
2410 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2413 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2414 qdisc_run_begin(q)) {
2416 * This is a work-conserving queue; there are no old skbs
2417 * waiting to be sent out; and the qdisc is not running -
2418 * xmit the skb directly.
2420 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2423 qdisc_bstats_update(q, skb);
2425 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2426 if (unlikely(contended)) {
2427 spin_unlock(&q->busylock);
2434 rc = NET_XMIT_SUCCESS;
2437 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2438 if (qdisc_run_begin(q)) {
2439 if (unlikely(contended)) {
2440 spin_unlock(&q->busylock);
2446 spin_unlock(root_lock);
2447 if (unlikely(contended))
2448 spin_unlock(&q->busylock);
2452 static DEFINE_PER_CPU(int, xmit_recursion);
2453 #define RECURSION_LIMIT 10
2456 * dev_queue_xmit - transmit a buffer
2457 * @skb: buffer to transmit
2459 * Queue a buffer for transmission to a network device. The caller must
2460 * have set the device and priority and built the buffer before calling
2461 * this function. The function can be called from an interrupt.
2463 * A negative errno code is returned on a failure. A success does not
2464 * guarantee the frame will be transmitted as it may be dropped due
2465 * to congestion or traffic shaping.
2467 * -----------------------------------------------------------------------------------
2468 * I notice this method can also return errors from the queue disciplines,
2469 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2472 * Regardless of the return value, the skb is consumed, so it is currently
2473 * difficult to retry a send to this method. (You can bump the ref count
2474 * before sending to hold a reference for retry if you are careful.)
2476 * When calling this method, interrupts MUST be enabled. This is because
2477 * the BH enable code must have IRQs enabled so that it will not deadlock.
2480 int dev_queue_xmit(struct sk_buff *skb)
2482 struct net_device *dev = skb->dev;
2483 struct netdev_queue *txq;
2487 /* Disable soft irqs for various locks below. Also
2488 * stops preemption for RCU.
2492 txq = dev_pick_tx(dev, skb);
2493 q = rcu_dereference_bh(txq->qdisc);
2495 #ifdef CONFIG_NET_CLS_ACT
2496 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2498 trace_net_dev_queue(skb);
2500 rc = __dev_xmit_skb(skb, q, dev, txq);
2504 /* The device has no queue. Common case for software devices:
2505 loopback, all the sorts of tunnels...
2507 Really, it is unlikely that netif_tx_lock protection is necessary
2508 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2510 However, it is possible, that they rely on protection
2513 Check this and shot the lock. It is not prone from deadlocks.
2514 Either shot noqueue qdisc, it is even simpler 8)
2516 if (dev->flags & IFF_UP) {
2517 int cpu = smp_processor_id(); /* ok because BHs are off */
2519 if (txq->xmit_lock_owner != cpu) {
2521 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2522 goto recursion_alert;
2524 HARD_TX_LOCK(dev, txq, cpu);
2526 if (!netif_tx_queue_stopped(txq)) {
2527 __this_cpu_inc(xmit_recursion);
2528 rc = dev_hard_start_xmit(skb, dev, txq);
2529 __this_cpu_dec(xmit_recursion);
2530 if (dev_xmit_complete(rc)) {
2531 HARD_TX_UNLOCK(dev, txq);
2535 HARD_TX_UNLOCK(dev, txq);
2536 if (net_ratelimit())
2537 printk(KERN_CRIT "Virtual device %s asks to "
2538 "queue packet!\n", dev->name);
2540 /* Recursion is detected! It is possible,
2544 if (net_ratelimit())
2545 printk(KERN_CRIT "Dead loop on virtual device "
2546 "%s, fix it urgently!\n", dev->name);
2551 rcu_read_unlock_bh();
2556 rcu_read_unlock_bh();
2559 EXPORT_SYMBOL(dev_queue_xmit);
2562 /*=======================================================================
2564 =======================================================================*/
2566 int netdev_max_backlog __read_mostly = 1000;
2567 int netdev_tstamp_prequeue __read_mostly = 1;
2568 int netdev_budget __read_mostly = 300;
2569 int weight_p __read_mostly = 64; /* old backlog weight */
2571 /* Called with irq disabled */
2572 static inline void ____napi_schedule(struct softnet_data *sd,
2573 struct napi_struct *napi)
2575 list_add_tail(&napi->poll_list, &sd->poll_list);
2576 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2580 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2581 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2582 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2583 * if hash is a canonical 4-tuple hash over transport ports.
2585 void __skb_get_rxhash(struct sk_buff *skb)
2587 int nhoff, hash = 0, poff;
2588 const struct ipv6hdr *ip6;
2589 const struct iphdr *ip;
2590 const struct vlan_hdr *vlan;
2599 nhoff = skb_network_offset(skb);
2600 proto = skb->protocol;
2604 case __constant_htons(ETH_P_IP):
2606 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2609 ip = (const struct iphdr *) (skb->data + nhoff);
2610 if (ip_is_fragment(ip))
2613 ip_proto = ip->protocol;
2614 addr1 = (__force u32) ip->saddr;
2615 addr2 = (__force u32) ip->daddr;
2616 nhoff += ip->ihl * 4;
2618 case __constant_htons(ETH_P_IPV6):
2620 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2623 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2624 ip_proto = ip6->nexthdr;
2625 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2626 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2629 case __constant_htons(ETH_P_8021Q):
2630 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2632 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2633 proto = vlan->h_vlan_encapsulated_proto;
2634 nhoff += sizeof(*vlan);
2636 case __constant_htons(ETH_P_PPP_SES):
2637 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2639 proto = *((__be16 *) (skb->data + nhoff +
2640 sizeof(struct pppoe_hdr)));
2641 nhoff += PPPOE_SES_HLEN;
2643 case __constant_htons(PPP_IP):
2645 case __constant_htons(PPP_IPV6):
2656 if (pskb_may_pull(skb, nhoff + 16)) {
2657 u8 *h = skb->data + nhoff;
2658 __be16 flags = *(__be16 *)h;
2661 * Only look inside GRE if version zero and no
2664 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2665 proto = *(__be16 *)(h + 2);
2667 if (flags & GRE_CSUM)
2669 if (flags & GRE_KEY)
2671 if (flags & GRE_SEQ)
2684 poff = proto_ports_offset(ip_proto);
2687 if (pskb_may_pull(skb, nhoff + 4)) {
2688 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2689 if (ports.v16[1] < ports.v16[0])
2690 swap(ports.v16[0], ports.v16[1]);
2695 /* get a consistent hash (same value on both flow directions) */
2699 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2706 EXPORT_SYMBOL(__skb_get_rxhash);
2710 /* One global table that all flow-based protocols share. */
2711 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2712 EXPORT_SYMBOL(rps_sock_flow_table);
2714 static struct rps_dev_flow *
2715 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2716 struct rps_dev_flow *rflow, u16 next_cpu)
2718 if (next_cpu != RPS_NO_CPU) {
2719 #ifdef CONFIG_RFS_ACCEL
2720 struct netdev_rx_queue *rxqueue;
2721 struct rps_dev_flow_table *flow_table;
2722 struct rps_dev_flow *old_rflow;
2727 /* Should we steer this flow to a different hardware queue? */
2728 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2729 !(dev->features & NETIF_F_NTUPLE))
2731 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2732 if (rxq_index == skb_get_rx_queue(skb))
2735 rxqueue = dev->_rx + rxq_index;
2736 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2739 flow_id = skb->rxhash & flow_table->mask;
2740 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2741 rxq_index, flow_id);
2745 rflow = &flow_table->flows[flow_id];
2747 if (old_rflow->filter == rflow->filter)
2748 old_rflow->filter = RPS_NO_FILTER;
2752 per_cpu(softnet_data, next_cpu).input_queue_head;
2755 rflow->cpu = next_cpu;
2760 * get_rps_cpu is called from netif_receive_skb and returns the target
2761 * CPU from the RPS map of the receiving queue for a given skb.
2762 * rcu_read_lock must be held on entry.
2764 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2765 struct rps_dev_flow **rflowp)
2767 struct netdev_rx_queue *rxqueue;
2768 struct rps_map *map;
2769 struct rps_dev_flow_table *flow_table;
2770 struct rps_sock_flow_table *sock_flow_table;
2774 if (skb_rx_queue_recorded(skb)) {
2775 u16 index = skb_get_rx_queue(skb);
2776 if (unlikely(index >= dev->real_num_rx_queues)) {
2777 WARN_ONCE(dev->real_num_rx_queues > 1,
2778 "%s received packet on queue %u, but number "
2779 "of RX queues is %u\n",
2780 dev->name, index, dev->real_num_rx_queues);
2783 rxqueue = dev->_rx + index;
2787 map = rcu_dereference(rxqueue->rps_map);
2789 if (map->len == 1 &&
2790 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2791 tcpu = map->cpus[0];
2792 if (cpu_online(tcpu))
2796 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2800 skb_reset_network_header(skb);
2801 if (!skb_get_rxhash(skb))
2804 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2805 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2806 if (flow_table && sock_flow_table) {
2808 struct rps_dev_flow *rflow;
2810 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2813 next_cpu = sock_flow_table->ents[skb->rxhash &
2814 sock_flow_table->mask];
2817 * If the desired CPU (where last recvmsg was done) is
2818 * different from current CPU (one in the rx-queue flow
2819 * table entry), switch if one of the following holds:
2820 * - Current CPU is unset (equal to RPS_NO_CPU).
2821 * - Current CPU is offline.
2822 * - The current CPU's queue tail has advanced beyond the
2823 * last packet that was enqueued using this table entry.
2824 * This guarantees that all previous packets for the flow
2825 * have been dequeued, thus preserving in order delivery.
2827 if (unlikely(tcpu != next_cpu) &&
2828 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2829 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2830 rflow->last_qtail)) >= 0))
2831 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2833 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2841 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2843 if (cpu_online(tcpu)) {
2853 #ifdef CONFIG_RFS_ACCEL
2856 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2857 * @dev: Device on which the filter was set
2858 * @rxq_index: RX queue index
2859 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2860 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2862 * Drivers that implement ndo_rx_flow_steer() should periodically call
2863 * this function for each installed filter and remove the filters for
2864 * which it returns %true.
2866 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2867 u32 flow_id, u16 filter_id)
2869 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2870 struct rps_dev_flow_table *flow_table;
2871 struct rps_dev_flow *rflow;
2876 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2877 if (flow_table && flow_id <= flow_table->mask) {
2878 rflow = &flow_table->flows[flow_id];
2879 cpu = ACCESS_ONCE(rflow->cpu);
2880 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2881 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2882 rflow->last_qtail) <
2883 (int)(10 * flow_table->mask)))
2889 EXPORT_SYMBOL(rps_may_expire_flow);
2891 #endif /* CONFIG_RFS_ACCEL */
2893 /* Called from hardirq (IPI) context */
2894 static void rps_trigger_softirq(void *data)
2896 struct softnet_data *sd = data;
2898 ____napi_schedule(sd, &sd->backlog);
2902 #endif /* CONFIG_RPS */
2905 * Check if this softnet_data structure is another cpu one
2906 * If yes, queue it to our IPI list and return 1
2909 static int rps_ipi_queued(struct softnet_data *sd)
2912 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2915 sd->rps_ipi_next = mysd->rps_ipi_list;
2916 mysd->rps_ipi_list = sd;
2918 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2921 #endif /* CONFIG_RPS */
2926 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2927 * queue (may be a remote CPU queue).
2929 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2930 unsigned int *qtail)
2932 struct softnet_data *sd;
2933 unsigned long flags;
2935 sd = &per_cpu(softnet_data, cpu);
2937 local_irq_save(flags);
2940 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2941 if (skb_queue_len(&sd->input_pkt_queue)) {
2943 __skb_queue_tail(&sd->input_pkt_queue, skb);
2944 input_queue_tail_incr_save(sd, qtail);
2946 local_irq_restore(flags);
2947 return NET_RX_SUCCESS;
2950 /* Schedule NAPI for backlog device
2951 * We can use non atomic operation since we own the queue lock
2953 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2954 if (!rps_ipi_queued(sd))
2955 ____napi_schedule(sd, &sd->backlog);
2963 local_irq_restore(flags);
2965 atomic_long_inc(&skb->dev->rx_dropped);
2971 * netif_rx - post buffer to the network code
2972 * @skb: buffer to post
2974 * This function receives a packet from a device driver and queues it for
2975 * the upper (protocol) levels to process. It always succeeds. The buffer
2976 * may be dropped during processing for congestion control or by the
2980 * NET_RX_SUCCESS (no congestion)
2981 * NET_RX_DROP (packet was dropped)
2985 int netif_rx(struct sk_buff *skb)
2989 /* if netpoll wants it, pretend we never saw it */
2990 if (netpoll_rx(skb))
2993 if (netdev_tstamp_prequeue)
2994 net_timestamp_check(skb);
2996 trace_netif_rx(skb);
2999 struct rps_dev_flow voidflow, *rflow = &voidflow;
3005 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3007 cpu = smp_processor_id();
3009 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3017 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3023 EXPORT_SYMBOL(netif_rx);
3025 int netif_rx_ni(struct sk_buff *skb)
3030 err = netif_rx(skb);
3031 if (local_softirq_pending())
3037 EXPORT_SYMBOL(netif_rx_ni);
3039 static void net_tx_action(struct softirq_action *h)
3041 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3043 if (sd->completion_queue) {
3044 struct sk_buff *clist;
3046 local_irq_disable();
3047 clist = sd->completion_queue;
3048 sd->completion_queue = NULL;
3052 struct sk_buff *skb = clist;
3053 clist = clist->next;
3055 WARN_ON(atomic_read(&skb->users));
3056 trace_kfree_skb(skb, net_tx_action);
3061 if (sd->output_queue) {
3064 local_irq_disable();
3065 head = sd->output_queue;
3066 sd->output_queue = NULL;
3067 sd->output_queue_tailp = &sd->output_queue;
3071 struct Qdisc *q = head;
3072 spinlock_t *root_lock;
3074 head = head->next_sched;
3076 root_lock = qdisc_lock(q);
3077 if (spin_trylock(root_lock)) {
3078 smp_mb__before_clear_bit();
3079 clear_bit(__QDISC_STATE_SCHED,
3082 spin_unlock(root_lock);
3084 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3086 __netif_reschedule(q);
3088 smp_mb__before_clear_bit();
3089 clear_bit(__QDISC_STATE_SCHED,
3097 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3098 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3099 /* This hook is defined here for ATM LANE */
3100 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3101 unsigned char *addr) __read_mostly;
3102 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3105 #ifdef CONFIG_NET_CLS_ACT
3106 /* TODO: Maybe we should just force sch_ingress to be compiled in
3107 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3108 * a compare and 2 stores extra right now if we dont have it on
3109 * but have CONFIG_NET_CLS_ACT
3110 * NOTE: This doesn't stop any functionality; if you dont have
3111 * the ingress scheduler, you just can't add policies on ingress.
3114 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3116 struct net_device *dev = skb->dev;
3117 u32 ttl = G_TC_RTTL(skb->tc_verd);
3118 int result = TC_ACT_OK;
3121 if (unlikely(MAX_RED_LOOP < ttl++)) {
3122 if (net_ratelimit())
3123 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3124 skb->skb_iif, dev->ifindex);
3128 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3129 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3132 if (q != &noop_qdisc) {
3133 spin_lock(qdisc_lock(q));
3134 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3135 result = qdisc_enqueue_root(skb, q);
3136 spin_unlock(qdisc_lock(q));
3142 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3143 struct packet_type **pt_prev,
3144 int *ret, struct net_device *orig_dev)
3146 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3148 if (!rxq || rxq->qdisc == &noop_qdisc)
3152 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3156 switch (ing_filter(skb, rxq)) {
3170 * netdev_rx_handler_register - register receive handler
3171 * @dev: device to register a handler for
3172 * @rx_handler: receive handler to register
3173 * @rx_handler_data: data pointer that is used by rx handler
3175 * Register a receive hander for a device. This handler will then be
3176 * called from __netif_receive_skb. A negative errno code is returned
3179 * The caller must hold the rtnl_mutex.
3181 * For a general description of rx_handler, see enum rx_handler_result.
3183 int netdev_rx_handler_register(struct net_device *dev,
3184 rx_handler_func_t *rx_handler,
3185 void *rx_handler_data)
3189 if (dev->rx_handler)
3192 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3193 rcu_assign_pointer(dev->rx_handler, rx_handler);
3197 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3200 * netdev_rx_handler_unregister - unregister receive handler
3201 * @dev: device to unregister a handler from
3203 * Unregister a receive hander from a device.
3205 * The caller must hold the rtnl_mutex.
3207 void netdev_rx_handler_unregister(struct net_device *dev)
3211 RCU_INIT_POINTER(dev->rx_handler, NULL);
3212 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3214 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3216 static int __netif_receive_skb(struct sk_buff *skb)
3218 struct packet_type *ptype, *pt_prev;
3219 rx_handler_func_t *rx_handler;
3220 struct net_device *orig_dev;
3221 struct net_device *null_or_dev;
3222 bool deliver_exact = false;
3223 int ret = NET_RX_DROP;
3226 if (!netdev_tstamp_prequeue)
3227 net_timestamp_check(skb);
3229 trace_netif_receive_skb(skb);
3231 /* if we've gotten here through NAPI, check netpoll */
3232 if (netpoll_receive_skb(skb))
3236 skb->skb_iif = skb->dev->ifindex;
3237 orig_dev = skb->dev;
3239 skb_reset_network_header(skb);
3240 skb_reset_transport_header(skb);
3241 skb_reset_mac_len(skb);
3249 __this_cpu_inc(softnet_data.processed);
3251 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3252 skb = vlan_untag(skb);
3257 #ifdef CONFIG_NET_CLS_ACT
3258 if (skb->tc_verd & TC_NCLS) {
3259 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3264 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3265 if (!ptype->dev || ptype->dev == skb->dev) {
3267 ret = deliver_skb(skb, pt_prev, orig_dev);
3272 #ifdef CONFIG_NET_CLS_ACT
3273 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3279 rx_handler = rcu_dereference(skb->dev->rx_handler);
3280 if (vlan_tx_tag_present(skb)) {
3282 ret = deliver_skb(skb, pt_prev, orig_dev);
3285 if (vlan_do_receive(&skb, !rx_handler))
3287 else if (unlikely(!skb))
3293 ret = deliver_skb(skb, pt_prev, orig_dev);
3296 switch (rx_handler(&skb)) {
3297 case RX_HANDLER_CONSUMED:
3299 case RX_HANDLER_ANOTHER:
3301 case RX_HANDLER_EXACT:
3302 deliver_exact = true;
3303 case RX_HANDLER_PASS:
3310 /* deliver only exact match when indicated */
3311 null_or_dev = deliver_exact ? skb->dev : NULL;
3313 type = skb->protocol;
3314 list_for_each_entry_rcu(ptype,
3315 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3316 if (ptype->type == type &&
3317 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3318 ptype->dev == orig_dev)) {
3320 ret = deliver_skb(skb, pt_prev, orig_dev);
3326 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3328 atomic_long_inc(&skb->dev->rx_dropped);
3330 /* Jamal, now you will not able to escape explaining
3331 * me how you were going to use this. :-)
3342 * netif_receive_skb - process receive buffer from network
3343 * @skb: buffer to process
3345 * netif_receive_skb() is the main receive data processing function.
3346 * It always succeeds. The buffer may be dropped during processing
3347 * for congestion control or by the protocol layers.
3349 * This function may only be called from softirq context and interrupts
3350 * should be enabled.
3352 * Return values (usually ignored):
3353 * NET_RX_SUCCESS: no congestion
3354 * NET_RX_DROP: packet was dropped
3356 int netif_receive_skb(struct sk_buff *skb)
3358 if (netdev_tstamp_prequeue)
3359 net_timestamp_check(skb);
3361 if (skb_defer_rx_timestamp(skb))
3362 return NET_RX_SUCCESS;
3366 struct rps_dev_flow voidflow, *rflow = &voidflow;
3371 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3374 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3378 ret = __netif_receive_skb(skb);
3384 return __netif_receive_skb(skb);
3387 EXPORT_SYMBOL(netif_receive_skb);
3389 /* Network device is going away, flush any packets still pending
3390 * Called with irqs disabled.
3392 static void flush_backlog(void *arg)
3394 struct net_device *dev = arg;
3395 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3396 struct sk_buff *skb, *tmp;
3399 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3400 if (skb->dev == dev) {
3401 __skb_unlink(skb, &sd->input_pkt_queue);
3403 input_queue_head_incr(sd);
3408 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3409 if (skb->dev == dev) {
3410 __skb_unlink(skb, &sd->process_queue);
3412 input_queue_head_incr(sd);
3417 static int napi_gro_complete(struct sk_buff *skb)
3419 struct packet_type *ptype;
3420 __be16 type = skb->protocol;
3421 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3424 if (NAPI_GRO_CB(skb)->count == 1) {
3425 skb_shinfo(skb)->gso_size = 0;
3430 list_for_each_entry_rcu(ptype, head, list) {
3431 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3434 err = ptype->gro_complete(skb);
3440 WARN_ON(&ptype->list == head);
3442 return NET_RX_SUCCESS;
3446 return netif_receive_skb(skb);
3449 inline void napi_gro_flush(struct napi_struct *napi)
3451 struct sk_buff *skb, *next;
3453 for (skb = napi->gro_list; skb; skb = next) {
3456 napi_gro_complete(skb);
3459 napi->gro_count = 0;
3460 napi->gro_list = NULL;
3462 EXPORT_SYMBOL(napi_gro_flush);
3464 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3466 struct sk_buff **pp = NULL;
3467 struct packet_type *ptype;
3468 __be16 type = skb->protocol;
3469 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3472 enum gro_result ret;
3474 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3477 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3481 list_for_each_entry_rcu(ptype, head, list) {
3482 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3485 skb_set_network_header(skb, skb_gro_offset(skb));
3486 mac_len = skb->network_header - skb->mac_header;
3487 skb->mac_len = mac_len;
3488 NAPI_GRO_CB(skb)->same_flow = 0;
3489 NAPI_GRO_CB(skb)->flush = 0;
3490 NAPI_GRO_CB(skb)->free = 0;
3492 pp = ptype->gro_receive(&napi->gro_list, skb);
3497 if (&ptype->list == head)
3500 same_flow = NAPI_GRO_CB(skb)->same_flow;
3501 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3504 struct sk_buff *nskb = *pp;
3508 napi_gro_complete(nskb);
3515 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3519 NAPI_GRO_CB(skb)->count = 1;
3520 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3521 skb->next = napi->gro_list;
3522 napi->gro_list = skb;
3526 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3527 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3529 BUG_ON(skb->end - skb->tail < grow);
3531 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3534 skb->data_len -= grow;
3536 skb_shinfo(skb)->frags[0].page_offset += grow;
3537 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3539 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3540 skb_frag_unref(skb, 0);
3541 memmove(skb_shinfo(skb)->frags,
3542 skb_shinfo(skb)->frags + 1,
3543 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3554 EXPORT_SYMBOL(dev_gro_receive);
3556 static inline gro_result_t
3557 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3560 unsigned int maclen = skb->dev->hard_header_len;
3562 for (p = napi->gro_list; p; p = p->next) {
3563 unsigned long diffs;
3565 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3566 diffs |= p->vlan_tci ^ skb->vlan_tci;
3567 if (maclen == ETH_HLEN)
3568 diffs |= compare_ether_header(skb_mac_header(p),
3569 skb_gro_mac_header(skb));
3571 diffs = memcmp(skb_mac_header(p),
3572 skb_gro_mac_header(skb),
3574 NAPI_GRO_CB(p)->same_flow = !diffs;
3575 NAPI_GRO_CB(p)->flush = 0;
3578 return dev_gro_receive(napi, skb);
3581 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3585 if (netif_receive_skb(skb))
3590 case GRO_MERGED_FREE:
3601 EXPORT_SYMBOL(napi_skb_finish);
3603 void skb_gro_reset_offset(struct sk_buff *skb)
3605 NAPI_GRO_CB(skb)->data_offset = 0;
3606 NAPI_GRO_CB(skb)->frag0 = NULL;
3607 NAPI_GRO_CB(skb)->frag0_len = 0;
3609 if (skb->mac_header == skb->tail &&
3610 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3611 NAPI_GRO_CB(skb)->frag0 =
3612 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3613 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3616 EXPORT_SYMBOL(skb_gro_reset_offset);
3618 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3620 skb_gro_reset_offset(skb);
3622 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3624 EXPORT_SYMBOL(napi_gro_receive);
3626 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3628 __skb_pull(skb, skb_headlen(skb));
3629 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3630 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3632 skb->dev = napi->dev;
3638 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3640 struct sk_buff *skb = napi->skb;
3643 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3649 EXPORT_SYMBOL(napi_get_frags);
3651 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3657 skb->protocol = eth_type_trans(skb, skb->dev);
3659 if (ret == GRO_HELD)
3660 skb_gro_pull(skb, -ETH_HLEN);
3661 else if (netif_receive_skb(skb))
3666 case GRO_MERGED_FREE:
3667 napi_reuse_skb(napi, skb);
3676 EXPORT_SYMBOL(napi_frags_finish);
3678 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3680 struct sk_buff *skb = napi->skb;
3687 skb_reset_mac_header(skb);
3688 skb_gro_reset_offset(skb);
3690 off = skb_gro_offset(skb);
3691 hlen = off + sizeof(*eth);
3692 eth = skb_gro_header_fast(skb, off);
3693 if (skb_gro_header_hard(skb, hlen)) {
3694 eth = skb_gro_header_slow(skb, hlen, off);
3695 if (unlikely(!eth)) {
3696 napi_reuse_skb(napi, skb);
3702 skb_gro_pull(skb, sizeof(*eth));
3705 * This works because the only protocols we care about don't require
3706 * special handling. We'll fix it up properly at the end.
3708 skb->protocol = eth->h_proto;
3713 EXPORT_SYMBOL(napi_frags_skb);
3715 gro_result_t napi_gro_frags(struct napi_struct *napi)
3717 struct sk_buff *skb = napi_frags_skb(napi);
3722 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3724 EXPORT_SYMBOL(napi_gro_frags);
3727 * net_rps_action sends any pending IPI's for rps.
3728 * Note: called with local irq disabled, but exits with local irq enabled.
3730 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3733 struct softnet_data *remsd = sd->rps_ipi_list;
3736 sd->rps_ipi_list = NULL;
3740 /* Send pending IPI's to kick RPS processing on remote cpus. */
3742 struct softnet_data *next = remsd->rps_ipi_next;
3744 if (cpu_online(remsd->cpu))
3745 __smp_call_function_single(remsd->cpu,
3754 static int process_backlog(struct napi_struct *napi, int quota)
3757 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3760 /* Check if we have pending ipi, its better to send them now,
3761 * not waiting net_rx_action() end.
3763 if (sd->rps_ipi_list) {
3764 local_irq_disable();
3765 net_rps_action_and_irq_enable(sd);
3768 napi->weight = weight_p;
3769 local_irq_disable();
3770 while (work < quota) {
3771 struct sk_buff *skb;
3774 while ((skb = __skb_dequeue(&sd->process_queue))) {
3776 __netif_receive_skb(skb);
3777 local_irq_disable();
3778 input_queue_head_incr(sd);
3779 if (++work >= quota) {
3786 qlen = skb_queue_len(&sd->input_pkt_queue);
3788 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3789 &sd->process_queue);
3791 if (qlen < quota - work) {
3793 * Inline a custom version of __napi_complete().
3794 * only current cpu owns and manipulates this napi,
3795 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3796 * we can use a plain write instead of clear_bit(),
3797 * and we dont need an smp_mb() memory barrier.
3799 list_del(&napi->poll_list);
3802 quota = work + qlen;
3812 * __napi_schedule - schedule for receive
3813 * @n: entry to schedule
3815 * The entry's receive function will be scheduled to run
3817 void __napi_schedule(struct napi_struct *n)
3819 unsigned long flags;
3821 local_irq_save(flags);
3822 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3823 local_irq_restore(flags);
3825 EXPORT_SYMBOL(__napi_schedule);
3827 void __napi_complete(struct napi_struct *n)
3829 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3830 BUG_ON(n->gro_list);
3832 list_del(&n->poll_list);
3833 smp_mb__before_clear_bit();
3834 clear_bit(NAPI_STATE_SCHED, &n->state);
3836 EXPORT_SYMBOL(__napi_complete);
3838 void napi_complete(struct napi_struct *n)
3840 unsigned long flags;
3843 * don't let napi dequeue from the cpu poll list
3844 * just in case its running on a different cpu
3846 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3850 local_irq_save(flags);
3852 local_irq_restore(flags);
3854 EXPORT_SYMBOL(napi_complete);
3856 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3857 int (*poll)(struct napi_struct *, int), int weight)
3859 INIT_LIST_HEAD(&napi->poll_list);
3860 napi->gro_count = 0;
3861 napi->gro_list = NULL;
3864 napi->weight = weight;
3865 list_add(&napi->dev_list, &dev->napi_list);
3867 #ifdef CONFIG_NETPOLL
3868 spin_lock_init(&napi->poll_lock);
3869 napi->poll_owner = -1;
3871 set_bit(NAPI_STATE_SCHED, &napi->state);
3873 EXPORT_SYMBOL(netif_napi_add);
3875 void netif_napi_del(struct napi_struct *napi)
3877 struct sk_buff *skb, *next;
3879 list_del_init(&napi->dev_list);
3880 napi_free_frags(napi);
3882 for (skb = napi->gro_list; skb; skb = next) {
3888 napi->gro_list = NULL;
3889 napi->gro_count = 0;
3891 EXPORT_SYMBOL(netif_napi_del);
3893 static void net_rx_action(struct softirq_action *h)
3895 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3896 unsigned long time_limit = jiffies + 2;
3897 int budget = netdev_budget;
3900 local_irq_disable();
3902 while (!list_empty(&sd->poll_list)) {
3903 struct napi_struct *n;
3906 /* If softirq window is exhuasted then punt.
3907 * Allow this to run for 2 jiffies since which will allow
3908 * an average latency of 1.5/HZ.
3910 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3915 /* Even though interrupts have been re-enabled, this
3916 * access is safe because interrupts can only add new
3917 * entries to the tail of this list, and only ->poll()
3918 * calls can remove this head entry from the list.
3920 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3922 have = netpoll_poll_lock(n);
3926 /* This NAPI_STATE_SCHED test is for avoiding a race
3927 * with netpoll's poll_napi(). Only the entity which
3928 * obtains the lock and sees NAPI_STATE_SCHED set will
3929 * actually make the ->poll() call. Therefore we avoid
3930 * accidentally calling ->poll() when NAPI is not scheduled.
3933 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3934 work = n->poll(n, weight);
3938 WARN_ON_ONCE(work > weight);
3942 local_irq_disable();
3944 /* Drivers must not modify the NAPI state if they
3945 * consume the entire weight. In such cases this code
3946 * still "owns" the NAPI instance and therefore can
3947 * move the instance around on the list at-will.
3949 if (unlikely(work == weight)) {
3950 if (unlikely(napi_disable_pending(n))) {
3953 local_irq_disable();
3955 list_move_tail(&n->poll_list, &sd->poll_list);
3958 netpoll_poll_unlock(have);
3961 net_rps_action_and_irq_enable(sd);
3963 #ifdef CONFIG_NET_DMA
3965 * There may not be any more sk_buffs coming right now, so push
3966 * any pending DMA copies to hardware
3968 dma_issue_pending_all();
3975 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3979 static gifconf_func_t *gifconf_list[NPROTO];
3982 * register_gifconf - register a SIOCGIF handler
3983 * @family: Address family
3984 * @gifconf: Function handler
3986 * Register protocol dependent address dumping routines. The handler
3987 * that is passed must not be freed or reused until it has been replaced
3988 * by another handler.
3990 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3992 if (family >= NPROTO)
3994 gifconf_list[family] = gifconf;
3997 EXPORT_SYMBOL(register_gifconf);
4001 * Map an interface index to its name (SIOCGIFNAME)
4005 * We need this ioctl for efficient implementation of the
4006 * if_indextoname() function required by the IPv6 API. Without
4007 * it, we would have to search all the interfaces to find a
4011 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4013 struct net_device *dev;
4017 * Fetch the caller's info block.
4020 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4024 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4030 strcpy(ifr.ifr_name, dev->name);
4033 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4039 * Perform a SIOCGIFCONF call. This structure will change
4040 * size eventually, and there is nothing I can do about it.
4041 * Thus we will need a 'compatibility mode'.
4044 static int dev_ifconf(struct net *net, char __user *arg)
4047 struct net_device *dev;
4054 * Fetch the caller's info block.
4057 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4064 * Loop over the interfaces, and write an info block for each.
4068 for_each_netdev(net, dev) {
4069 for (i = 0; i < NPROTO; i++) {
4070 if (gifconf_list[i]) {
4073 done = gifconf_list[i](dev, NULL, 0);
4075 done = gifconf_list[i](dev, pos + total,
4085 * All done. Write the updated control block back to the caller.
4087 ifc.ifc_len = total;
4090 * Both BSD and Solaris return 0 here, so we do too.
4092 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4095 #ifdef CONFIG_PROC_FS
4097 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4099 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4100 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4101 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4103 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4105 struct net *net = seq_file_net(seq);
4106 struct net_device *dev;
4107 struct hlist_node *p;
4108 struct hlist_head *h;
4109 unsigned int count = 0, offset = get_offset(*pos);
4111 h = &net->dev_name_head[get_bucket(*pos)];
4112 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4113 if (++count == offset)
4120 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4122 struct net_device *dev;
4123 unsigned int bucket;
4126 dev = dev_from_same_bucket(seq, pos);
4130 bucket = get_bucket(*pos) + 1;
4131 *pos = set_bucket_offset(bucket, 1);
4132 } while (bucket < NETDEV_HASHENTRIES);
4138 * This is invoked by the /proc filesystem handler to display a device
4141 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4146 return SEQ_START_TOKEN;
4148 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4151 return dev_from_bucket(seq, pos);
4154 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4157 return dev_from_bucket(seq, pos);
4160 void dev_seq_stop(struct seq_file *seq, void *v)
4166 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4168 struct rtnl_link_stats64 temp;
4169 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4171 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4172 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4173 dev->name, stats->rx_bytes, stats->rx_packets,
4175 stats->rx_dropped + stats->rx_missed_errors,
4176 stats->rx_fifo_errors,
4177 stats->rx_length_errors + stats->rx_over_errors +
4178 stats->rx_crc_errors + stats->rx_frame_errors,
4179 stats->rx_compressed, stats->multicast,
4180 stats->tx_bytes, stats->tx_packets,
4181 stats->tx_errors, stats->tx_dropped,
4182 stats->tx_fifo_errors, stats->collisions,
4183 stats->tx_carrier_errors +
4184 stats->tx_aborted_errors +
4185 stats->tx_window_errors +
4186 stats->tx_heartbeat_errors,
4187 stats->tx_compressed);
4191 * Called from the PROCfs module. This now uses the new arbitrary sized
4192 * /proc/net interface to create /proc/net/dev
4194 static int dev_seq_show(struct seq_file *seq, void *v)
4196 if (v == SEQ_START_TOKEN)
4197 seq_puts(seq, "Inter-| Receive "
4199 " face |bytes packets errs drop fifo frame "
4200 "compressed multicast|bytes packets errs "
4201 "drop fifo colls carrier compressed\n");
4203 dev_seq_printf_stats(seq, v);
4207 static struct softnet_data *softnet_get_online(loff_t *pos)
4209 struct softnet_data *sd = NULL;
4211 while (*pos < nr_cpu_ids)
4212 if (cpu_online(*pos)) {
4213 sd = &per_cpu(softnet_data, *pos);
4220 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4222 return softnet_get_online(pos);
4225 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4228 return softnet_get_online(pos);
4231 static void softnet_seq_stop(struct seq_file *seq, void *v)
4235 static int softnet_seq_show(struct seq_file *seq, void *v)
4237 struct softnet_data *sd = v;
4239 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4240 sd->processed, sd->dropped, sd->time_squeeze, 0,
4241 0, 0, 0, 0, /* was fastroute */
4242 sd->cpu_collision, sd->received_rps);
4246 static const struct seq_operations dev_seq_ops = {
4247 .start = dev_seq_start,
4248 .next = dev_seq_next,
4249 .stop = dev_seq_stop,
4250 .show = dev_seq_show,
4253 static int dev_seq_open(struct inode *inode, struct file *file)
4255 return seq_open_net(inode, file, &dev_seq_ops,
4256 sizeof(struct seq_net_private));
4259 static const struct file_operations dev_seq_fops = {
4260 .owner = THIS_MODULE,
4261 .open = dev_seq_open,
4263 .llseek = seq_lseek,
4264 .release = seq_release_net,
4267 static const struct seq_operations softnet_seq_ops = {
4268 .start = softnet_seq_start,
4269 .next = softnet_seq_next,
4270 .stop = softnet_seq_stop,
4271 .show = softnet_seq_show,
4274 static int softnet_seq_open(struct inode *inode, struct file *file)
4276 return seq_open(file, &softnet_seq_ops);
4279 static const struct file_operations softnet_seq_fops = {
4280 .owner = THIS_MODULE,
4281 .open = softnet_seq_open,
4283 .llseek = seq_lseek,
4284 .release = seq_release,
4287 static void *ptype_get_idx(loff_t pos)
4289 struct packet_type *pt = NULL;
4293 list_for_each_entry_rcu(pt, &ptype_all, list) {
4299 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4300 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4309 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4313 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4316 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4318 struct packet_type *pt;
4319 struct list_head *nxt;
4323 if (v == SEQ_START_TOKEN)
4324 return ptype_get_idx(0);
4327 nxt = pt->list.next;
4328 if (pt->type == htons(ETH_P_ALL)) {
4329 if (nxt != &ptype_all)
4332 nxt = ptype_base[0].next;
4334 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4336 while (nxt == &ptype_base[hash]) {
4337 if (++hash >= PTYPE_HASH_SIZE)
4339 nxt = ptype_base[hash].next;
4342 return list_entry(nxt, struct packet_type, list);
4345 static void ptype_seq_stop(struct seq_file *seq, void *v)
4351 static int ptype_seq_show(struct seq_file *seq, void *v)
4353 struct packet_type *pt = v;
4355 if (v == SEQ_START_TOKEN)
4356 seq_puts(seq, "Type Device Function\n");
4357 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4358 if (pt->type == htons(ETH_P_ALL))
4359 seq_puts(seq, "ALL ");
4361 seq_printf(seq, "%04x", ntohs(pt->type));
4363 seq_printf(seq, " %-8s %pF\n",
4364 pt->dev ? pt->dev->name : "", pt->func);
4370 static const struct seq_operations ptype_seq_ops = {
4371 .start = ptype_seq_start,
4372 .next = ptype_seq_next,
4373 .stop = ptype_seq_stop,
4374 .show = ptype_seq_show,
4377 static int ptype_seq_open(struct inode *inode, struct file *file)
4379 return seq_open_net(inode, file, &ptype_seq_ops,
4380 sizeof(struct seq_net_private));
4383 static const struct file_operations ptype_seq_fops = {
4384 .owner = THIS_MODULE,
4385 .open = ptype_seq_open,
4387 .llseek = seq_lseek,
4388 .release = seq_release_net,
4392 static int __net_init dev_proc_net_init(struct net *net)
4396 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4398 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4400 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4403 if (wext_proc_init(net))
4409 proc_net_remove(net, "ptype");
4411 proc_net_remove(net, "softnet_stat");
4413 proc_net_remove(net, "dev");
4417 static void __net_exit dev_proc_net_exit(struct net *net)
4419 wext_proc_exit(net);
4421 proc_net_remove(net, "ptype");
4422 proc_net_remove(net, "softnet_stat");
4423 proc_net_remove(net, "dev");
4426 static struct pernet_operations __net_initdata dev_proc_ops = {
4427 .init = dev_proc_net_init,
4428 .exit = dev_proc_net_exit,
4431 static int __init dev_proc_init(void)
4433 return register_pernet_subsys(&dev_proc_ops);
4436 #define dev_proc_init() 0
4437 #endif /* CONFIG_PROC_FS */
4441 * netdev_set_master - set up master pointer
4442 * @slave: slave device
4443 * @master: new master device
4445 * Changes the master device of the slave. Pass %NULL to break the
4446 * bonding. The caller must hold the RTNL semaphore. On a failure
4447 * a negative errno code is returned. On success the reference counts
4448 * are adjusted and the function returns zero.
4450 int netdev_set_master(struct net_device *slave, struct net_device *master)
4452 struct net_device *old = slave->master;
4462 slave->master = master;
4468 EXPORT_SYMBOL(netdev_set_master);
4471 * netdev_set_bond_master - set up bonding master/slave pair
4472 * @slave: slave device
4473 * @master: new master device
4475 * Changes the master device of the slave. Pass %NULL to break the
4476 * bonding. The caller must hold the RTNL semaphore. On a failure
4477 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4478 * to the routing socket and the function returns zero.
4480 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4486 err = netdev_set_master(slave, master);
4490 slave->flags |= IFF_SLAVE;
4492 slave->flags &= ~IFF_SLAVE;
4494 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4497 EXPORT_SYMBOL(netdev_set_bond_master);
4499 static void dev_change_rx_flags(struct net_device *dev, int flags)
4501 const struct net_device_ops *ops = dev->netdev_ops;
4503 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4504 ops->ndo_change_rx_flags(dev, flags);
4507 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4509 unsigned short old_flags = dev->flags;
4515 dev->flags |= IFF_PROMISC;
4516 dev->promiscuity += inc;
4517 if (dev->promiscuity == 0) {
4520 * If inc causes overflow, untouch promisc and return error.
4523 dev->flags &= ~IFF_PROMISC;
4525 dev->promiscuity -= inc;
4526 printk(KERN_WARNING "%s: promiscuity touches roof, "
4527 "set promiscuity failed, promiscuity feature "
4528 "of device might be broken.\n", dev->name);
4532 if (dev->flags != old_flags) {
4533 printk(KERN_INFO "device %s %s promiscuous mode\n",
4534 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4536 if (audit_enabled) {
4537 current_uid_gid(&uid, &gid);
4538 audit_log(current->audit_context, GFP_ATOMIC,
4539 AUDIT_ANOM_PROMISCUOUS,
4540 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4541 dev->name, (dev->flags & IFF_PROMISC),
4542 (old_flags & IFF_PROMISC),
4543 audit_get_loginuid(current),
4545 audit_get_sessionid(current));
4548 dev_change_rx_flags(dev, IFF_PROMISC);
4554 * dev_set_promiscuity - update promiscuity count on a device
4558 * Add or remove promiscuity from a device. While the count in the device
4559 * remains above zero the interface remains promiscuous. Once it hits zero
4560 * the device reverts back to normal filtering operation. A negative inc
4561 * value is used to drop promiscuity on the device.
4562 * Return 0 if successful or a negative errno code on error.
4564 int dev_set_promiscuity(struct net_device *dev, int inc)
4566 unsigned short old_flags = dev->flags;
4569 err = __dev_set_promiscuity(dev, inc);
4572 if (dev->flags != old_flags)
4573 dev_set_rx_mode(dev);
4576 EXPORT_SYMBOL(dev_set_promiscuity);
4579 * dev_set_allmulti - update allmulti count on a device
4583 * Add or remove reception of all multicast frames to a device. While the
4584 * count in the device remains above zero the interface remains listening
4585 * to all interfaces. Once it hits zero the device reverts back to normal
4586 * filtering operation. A negative @inc value is used to drop the counter
4587 * when releasing a resource needing all multicasts.
4588 * Return 0 if successful or a negative errno code on error.
4591 int dev_set_allmulti(struct net_device *dev, int inc)
4593 unsigned short old_flags = dev->flags;
4597 dev->flags |= IFF_ALLMULTI;
4598 dev->allmulti += inc;
4599 if (dev->allmulti == 0) {
4602 * If inc causes overflow, untouch allmulti and return error.
4605 dev->flags &= ~IFF_ALLMULTI;
4607 dev->allmulti -= inc;
4608 printk(KERN_WARNING "%s: allmulti touches roof, "
4609 "set allmulti failed, allmulti feature of "
4610 "device might be broken.\n", dev->name);
4614 if (dev->flags ^ old_flags) {
4615 dev_change_rx_flags(dev, IFF_ALLMULTI);
4616 dev_set_rx_mode(dev);
4620 EXPORT_SYMBOL(dev_set_allmulti);
4623 * Upload unicast and multicast address lists to device and
4624 * configure RX filtering. When the device doesn't support unicast
4625 * filtering it is put in promiscuous mode while unicast addresses
4628 void __dev_set_rx_mode(struct net_device *dev)
4630 const struct net_device_ops *ops = dev->netdev_ops;
4632 /* dev_open will call this function so the list will stay sane. */
4633 if (!(dev->flags&IFF_UP))
4636 if (!netif_device_present(dev))
4639 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4640 /* Unicast addresses changes may only happen under the rtnl,
4641 * therefore calling __dev_set_promiscuity here is safe.
4643 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4644 __dev_set_promiscuity(dev, 1);
4645 dev->uc_promisc = true;
4646 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4647 __dev_set_promiscuity(dev, -1);
4648 dev->uc_promisc = false;
4652 if (ops->ndo_set_rx_mode)
4653 ops->ndo_set_rx_mode(dev);
4656 void dev_set_rx_mode(struct net_device *dev)
4658 netif_addr_lock_bh(dev);
4659 __dev_set_rx_mode(dev);
4660 netif_addr_unlock_bh(dev);
4664 * dev_get_flags - get flags reported to userspace
4667 * Get the combination of flag bits exported through APIs to userspace.
4669 unsigned dev_get_flags(const struct net_device *dev)
4673 flags = (dev->flags & ~(IFF_PROMISC |
4678 (dev->gflags & (IFF_PROMISC |
4681 if (netif_running(dev)) {
4682 if (netif_oper_up(dev))
4683 flags |= IFF_RUNNING;
4684 if (netif_carrier_ok(dev))
4685 flags |= IFF_LOWER_UP;
4686 if (netif_dormant(dev))
4687 flags |= IFF_DORMANT;
4692 EXPORT_SYMBOL(dev_get_flags);
4694 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4696 int old_flags = dev->flags;
4702 * Set the flags on our device.
4705 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4706 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4708 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4712 * Load in the correct multicast list now the flags have changed.
4715 if ((old_flags ^ flags) & IFF_MULTICAST)
4716 dev_change_rx_flags(dev, IFF_MULTICAST);
4718 dev_set_rx_mode(dev);
4721 * Have we downed the interface. We handle IFF_UP ourselves
4722 * according to user attempts to set it, rather than blindly
4727 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4728 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4731 dev_set_rx_mode(dev);
4734 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4735 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4737 dev->gflags ^= IFF_PROMISC;
4738 dev_set_promiscuity(dev, inc);
4741 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4742 is important. Some (broken) drivers set IFF_PROMISC, when
4743 IFF_ALLMULTI is requested not asking us and not reporting.
4745 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4746 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4748 dev->gflags ^= IFF_ALLMULTI;
4749 dev_set_allmulti(dev, inc);
4755 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4757 unsigned int changes = dev->flags ^ old_flags;
4759 if (changes & IFF_UP) {
4760 if (dev->flags & IFF_UP)
4761 call_netdevice_notifiers(NETDEV_UP, dev);
4763 call_netdevice_notifiers(NETDEV_DOWN, dev);
4766 if (dev->flags & IFF_UP &&
4767 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4768 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4772 * dev_change_flags - change device settings
4774 * @flags: device state flags
4776 * Change settings on device based state flags. The flags are
4777 * in the userspace exported format.
4779 int dev_change_flags(struct net_device *dev, unsigned flags)
4782 int old_flags = dev->flags;
4784 ret = __dev_change_flags(dev, flags);
4788 changes = old_flags ^ dev->flags;
4790 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4792 __dev_notify_flags(dev, old_flags);
4795 EXPORT_SYMBOL(dev_change_flags);
4798 * dev_set_mtu - Change maximum transfer unit
4800 * @new_mtu: new transfer unit
4802 * Change the maximum transfer size of the network device.
4804 int dev_set_mtu(struct net_device *dev, int new_mtu)
4806 const struct net_device_ops *ops = dev->netdev_ops;
4809 if (new_mtu == dev->mtu)
4812 /* MTU must be positive. */
4816 if (!netif_device_present(dev))
4820 if (ops->ndo_change_mtu)
4821 err = ops->ndo_change_mtu(dev, new_mtu);
4825 if (!err && dev->flags & IFF_UP)
4826 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4829 EXPORT_SYMBOL(dev_set_mtu);
4832 * dev_set_group - Change group this device belongs to
4834 * @new_group: group this device should belong to
4836 void dev_set_group(struct net_device *dev, int new_group)
4838 dev->group = new_group;
4840 EXPORT_SYMBOL(dev_set_group);
4843 * dev_set_mac_address - Change Media Access Control Address
4847 * Change the hardware (MAC) address of the device
4849 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4851 const struct net_device_ops *ops = dev->netdev_ops;
4854 if (!ops->ndo_set_mac_address)
4856 if (sa->sa_family != dev->type)
4858 if (!netif_device_present(dev))
4860 err = ops->ndo_set_mac_address(dev, sa);
4862 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4863 add_device_randomness(dev->dev_addr, dev->addr_len);
4866 EXPORT_SYMBOL(dev_set_mac_address);
4869 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4871 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4874 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4880 case SIOCGIFFLAGS: /* Get interface flags */
4881 ifr->ifr_flags = (short) dev_get_flags(dev);
4884 case SIOCGIFMETRIC: /* Get the metric on the interface
4885 (currently unused) */
4886 ifr->ifr_metric = 0;
4889 case SIOCGIFMTU: /* Get the MTU of a device */
4890 ifr->ifr_mtu = dev->mtu;
4895 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4897 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4898 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4899 ifr->ifr_hwaddr.sa_family = dev->type;
4907 ifr->ifr_map.mem_start = dev->mem_start;
4908 ifr->ifr_map.mem_end = dev->mem_end;
4909 ifr->ifr_map.base_addr = dev->base_addr;
4910 ifr->ifr_map.irq = dev->irq;
4911 ifr->ifr_map.dma = dev->dma;
4912 ifr->ifr_map.port = dev->if_port;
4916 ifr->ifr_ifindex = dev->ifindex;
4920 ifr->ifr_qlen = dev->tx_queue_len;
4924 /* dev_ioctl() should ensure this case
4936 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4938 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4941 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4942 const struct net_device_ops *ops;
4947 ops = dev->netdev_ops;
4950 case SIOCSIFFLAGS: /* Set interface flags */
4951 return dev_change_flags(dev, ifr->ifr_flags);
4953 case SIOCSIFMETRIC: /* Set the metric on the interface
4954 (currently unused) */
4957 case SIOCSIFMTU: /* Set the MTU of a device */
4958 return dev_set_mtu(dev, ifr->ifr_mtu);
4961 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4963 case SIOCSIFHWBROADCAST:
4964 if (ifr->ifr_hwaddr.sa_family != dev->type)
4966 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4967 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4968 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4972 if (ops->ndo_set_config) {
4973 if (!netif_device_present(dev))
4975 return ops->ndo_set_config(dev, &ifr->ifr_map);
4980 if (!ops->ndo_set_rx_mode ||
4981 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4983 if (!netif_device_present(dev))
4985 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4988 if (!ops->ndo_set_rx_mode ||
4989 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4991 if (!netif_device_present(dev))
4993 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4996 if (ifr->ifr_qlen < 0)
4998 dev->tx_queue_len = ifr->ifr_qlen;
5002 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5003 return dev_change_name(dev, ifr->ifr_newname);
5006 err = net_hwtstamp_validate(ifr);
5012 * Unknown or private ioctl
5015 if ((cmd >= SIOCDEVPRIVATE &&
5016 cmd <= SIOCDEVPRIVATE + 15) ||
5017 cmd == SIOCBONDENSLAVE ||
5018 cmd == SIOCBONDRELEASE ||
5019 cmd == SIOCBONDSETHWADDR ||
5020 cmd == SIOCBONDSLAVEINFOQUERY ||
5021 cmd == SIOCBONDINFOQUERY ||
5022 cmd == SIOCBONDCHANGEACTIVE ||
5023 cmd == SIOCGMIIPHY ||
5024 cmd == SIOCGMIIREG ||
5025 cmd == SIOCSMIIREG ||
5026 cmd == SIOCBRADDIF ||
5027 cmd == SIOCBRDELIF ||
5028 cmd == SIOCSHWTSTAMP ||
5029 cmd == SIOCWANDEV) {
5031 if (ops->ndo_do_ioctl) {
5032 if (netif_device_present(dev))
5033 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5045 * This function handles all "interface"-type I/O control requests. The actual
5046 * 'doing' part of this is dev_ifsioc above.
5050 * dev_ioctl - network device ioctl
5051 * @net: the applicable net namespace
5052 * @cmd: command to issue
5053 * @arg: pointer to a struct ifreq in user space
5055 * Issue ioctl functions to devices. This is normally called by the
5056 * user space syscall interfaces but can sometimes be useful for
5057 * other purposes. The return value is the return from the syscall if
5058 * positive or a negative errno code on error.
5061 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5067 /* One special case: SIOCGIFCONF takes ifconf argument
5068 and requires shared lock, because it sleeps writing
5072 if (cmd == SIOCGIFCONF) {
5074 ret = dev_ifconf(net, (char __user *) arg);
5078 if (cmd == SIOCGIFNAME)
5079 return dev_ifname(net, (struct ifreq __user *)arg);
5081 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5084 ifr.ifr_name[IFNAMSIZ-1] = 0;
5086 colon = strchr(ifr.ifr_name, ':');
5091 * See which interface the caller is talking about.
5096 * These ioctl calls:
5097 * - can be done by all.
5098 * - atomic and do not require locking.
5109 dev_load(net, ifr.ifr_name);
5111 ret = dev_ifsioc_locked(net, &ifr, cmd);
5116 if (copy_to_user(arg, &ifr,
5117 sizeof(struct ifreq)))
5123 dev_load(net, ifr.ifr_name);
5125 ret = dev_ethtool(net, &ifr);
5130 if (copy_to_user(arg, &ifr,
5131 sizeof(struct ifreq)))
5137 * These ioctl calls:
5138 * - require superuser power.
5139 * - require strict serialization.
5145 if (!capable(CAP_NET_ADMIN))
5147 dev_load(net, ifr.ifr_name);
5149 ret = dev_ifsioc(net, &ifr, cmd);
5154 if (copy_to_user(arg, &ifr,
5155 sizeof(struct ifreq)))
5161 * These ioctl calls:
5162 * - require superuser power.
5163 * - require strict serialization.
5164 * - do not return a value
5174 case SIOCSIFHWBROADCAST:
5177 case SIOCBONDENSLAVE:
5178 case SIOCBONDRELEASE:
5179 case SIOCBONDSETHWADDR:
5180 case SIOCBONDCHANGEACTIVE:
5184 if (!capable(CAP_NET_ADMIN))
5187 case SIOCBONDSLAVEINFOQUERY:
5188 case SIOCBONDINFOQUERY:
5189 dev_load(net, ifr.ifr_name);
5191 ret = dev_ifsioc(net, &ifr, cmd);
5196 /* Get the per device memory space. We can add this but
5197 * currently do not support it */
5199 /* Set the per device memory buffer space.
5200 * Not applicable in our case */
5205 * Unknown or private ioctl.
5208 if (cmd == SIOCWANDEV ||
5209 (cmd >= SIOCDEVPRIVATE &&
5210 cmd <= SIOCDEVPRIVATE + 15)) {
5211 dev_load(net, ifr.ifr_name);
5213 ret = dev_ifsioc(net, &ifr, cmd);
5215 if (!ret && copy_to_user(arg, &ifr,
5216 sizeof(struct ifreq)))
5220 /* Take care of Wireless Extensions */
5221 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5222 return wext_handle_ioctl(net, &ifr, cmd, arg);
5229 * dev_new_index - allocate an ifindex
5230 * @net: the applicable net namespace
5232 * Returns a suitable unique value for a new device interface
5233 * number. The caller must hold the rtnl semaphore or the
5234 * dev_base_lock to be sure it remains unique.
5236 static int dev_new_index(struct net *net)
5242 if (!__dev_get_by_index(net, ifindex))
5247 /* Delayed registration/unregisteration */
5248 static LIST_HEAD(net_todo_list);
5250 static void net_set_todo(struct net_device *dev)
5252 list_add_tail(&dev->todo_list, &net_todo_list);
5255 static void rollback_registered_many(struct list_head *head)
5257 struct net_device *dev, *tmp;
5259 BUG_ON(dev_boot_phase);
5262 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5263 /* Some devices call without registering
5264 * for initialization unwind. Remove those
5265 * devices and proceed with the remaining.
5267 if (dev->reg_state == NETREG_UNINITIALIZED) {
5268 pr_debug("unregister_netdevice: device %s/%p never "
5269 "was registered\n", dev->name, dev);
5272 list_del(&dev->unreg_list);
5275 dev->dismantle = true;
5276 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5279 /* If device is running, close it first. */
5280 dev_close_many(head);
5282 list_for_each_entry(dev, head, unreg_list) {
5283 /* And unlink it from device chain. */
5284 unlist_netdevice(dev);
5286 dev->reg_state = NETREG_UNREGISTERING;
5291 list_for_each_entry(dev, head, unreg_list) {
5292 /* Shutdown queueing discipline. */
5296 /* Notify protocols, that we are about to destroy
5297 this device. They should clean all the things.
5299 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5301 if (!dev->rtnl_link_ops ||
5302 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5303 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5306 * Flush the unicast and multicast chains
5311 if (dev->netdev_ops->ndo_uninit)
5312 dev->netdev_ops->ndo_uninit(dev);
5314 /* Notifier chain MUST detach us from master device. */
5315 WARN_ON(dev->master);
5317 /* Remove entries from kobject tree */
5318 netdev_unregister_kobject(dev);
5321 /* Process any work delayed until the end of the batch */
5322 dev = list_first_entry(head, struct net_device, unreg_list);
5323 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5327 list_for_each_entry(dev, head, unreg_list)
5331 static void rollback_registered(struct net_device *dev)
5335 list_add(&dev->unreg_list, &single);
5336 rollback_registered_many(&single);
5340 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5342 /* Fix illegal checksum combinations */
5343 if ((features & NETIF_F_HW_CSUM) &&
5344 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5345 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5346 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5349 if ((features & NETIF_F_NO_CSUM) &&
5350 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5351 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5352 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5355 /* Fix illegal SG+CSUM combinations. */
5356 if ((features & NETIF_F_SG) &&
5357 !(features & NETIF_F_ALL_CSUM)) {
5359 "Dropping NETIF_F_SG since no checksum feature.\n");
5360 features &= ~NETIF_F_SG;
5363 /* TSO requires that SG is present as well. */
5364 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5365 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5366 features &= ~NETIF_F_ALL_TSO;
5369 /* TSO ECN requires that TSO is present as well. */
5370 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5371 features &= ~NETIF_F_TSO_ECN;
5373 /* Software GSO depends on SG. */
5374 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5375 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5376 features &= ~NETIF_F_GSO;
5379 /* UFO needs SG and checksumming */
5380 if (features & NETIF_F_UFO) {
5381 /* maybe split UFO into V4 and V6? */
5382 if (!((features & NETIF_F_GEN_CSUM) ||
5383 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5384 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5386 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5387 features &= ~NETIF_F_UFO;
5390 if (!(features & NETIF_F_SG)) {
5392 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5393 features &= ~NETIF_F_UFO;
5400 int __netdev_update_features(struct net_device *dev)
5407 features = netdev_get_wanted_features(dev);
5409 if (dev->netdev_ops->ndo_fix_features)
5410 features = dev->netdev_ops->ndo_fix_features(dev, features);
5412 /* driver might be less strict about feature dependencies */
5413 features = netdev_fix_features(dev, features);
5415 if (dev->features == features)
5418 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5419 dev->features, features);
5421 if (dev->netdev_ops->ndo_set_features)
5422 err = dev->netdev_ops->ndo_set_features(dev, features);
5424 if (unlikely(err < 0)) {
5426 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5427 err, features, dev->features);
5432 dev->features = features;
5438 * netdev_update_features - recalculate device features
5439 * @dev: the device to check
5441 * Recalculate dev->features set and send notifications if it
5442 * has changed. Should be called after driver or hardware dependent
5443 * conditions might have changed that influence the features.
5445 void netdev_update_features(struct net_device *dev)
5447 if (__netdev_update_features(dev))
5448 netdev_features_change(dev);
5450 EXPORT_SYMBOL(netdev_update_features);
5453 * netdev_change_features - recalculate device features
5454 * @dev: the device to check
5456 * Recalculate dev->features set and send notifications even
5457 * if they have not changed. Should be called instead of
5458 * netdev_update_features() if also dev->vlan_features might
5459 * have changed to allow the changes to be propagated to stacked
5462 void netdev_change_features(struct net_device *dev)
5464 __netdev_update_features(dev);
5465 netdev_features_change(dev);
5467 EXPORT_SYMBOL(netdev_change_features);
5470 * netif_stacked_transfer_operstate - transfer operstate
5471 * @rootdev: the root or lower level device to transfer state from
5472 * @dev: the device to transfer operstate to
5474 * Transfer operational state from root to device. This is normally
5475 * called when a stacking relationship exists between the root
5476 * device and the device(a leaf device).
5478 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5479 struct net_device *dev)
5481 if (rootdev->operstate == IF_OPER_DORMANT)
5482 netif_dormant_on(dev);
5484 netif_dormant_off(dev);
5486 if (netif_carrier_ok(rootdev)) {
5487 if (!netif_carrier_ok(dev))
5488 netif_carrier_on(dev);
5490 if (netif_carrier_ok(dev))
5491 netif_carrier_off(dev);
5494 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5497 static int netif_alloc_rx_queues(struct net_device *dev)
5499 unsigned int i, count = dev->num_rx_queues;
5500 struct netdev_rx_queue *rx;
5504 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5506 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5511 for (i = 0; i < count; i++)
5517 static void netdev_init_one_queue(struct net_device *dev,
5518 struct netdev_queue *queue, void *_unused)
5520 /* Initialize queue lock */
5521 spin_lock_init(&queue->_xmit_lock);
5522 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5523 queue->xmit_lock_owner = -1;
5524 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5528 static int netif_alloc_netdev_queues(struct net_device *dev)
5530 unsigned int count = dev->num_tx_queues;
5531 struct netdev_queue *tx;
5535 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5537 pr_err("netdev: Unable to allocate %u tx queues.\n",
5543 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5544 spin_lock_init(&dev->tx_global_lock);
5550 * register_netdevice - register a network device
5551 * @dev: device to register
5553 * Take a completed network device structure and add it to the kernel
5554 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5555 * chain. 0 is returned on success. A negative errno code is returned
5556 * on a failure to set up the device, or if the name is a duplicate.
5558 * Callers must hold the rtnl semaphore. You may want
5559 * register_netdev() instead of this.
5562 * The locking appears insufficient to guarantee two parallel registers
5563 * will not get the same name.
5566 int register_netdevice(struct net_device *dev)
5569 struct net *net = dev_net(dev);
5571 BUG_ON(dev_boot_phase);
5576 /* When net_device's are persistent, this will be fatal. */
5577 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5580 spin_lock_init(&dev->addr_list_lock);
5581 netdev_set_addr_lockdep_class(dev);
5585 ret = dev_get_valid_name(dev, dev->name);
5589 /* Init, if this function is available */
5590 if (dev->netdev_ops->ndo_init) {
5591 ret = dev->netdev_ops->ndo_init(dev);
5599 dev->ifindex = dev_new_index(net);
5600 if (dev->iflink == -1)
5601 dev->iflink = dev->ifindex;
5603 /* Transfer changeable features to wanted_features and enable
5604 * software offloads (GSO and GRO).
5606 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5607 dev->features |= NETIF_F_SOFT_FEATURES;
5608 dev->wanted_features = dev->features & dev->hw_features;
5610 /* Turn on no cache copy if HW is doing checksum */
5611 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5612 if ((dev->features & NETIF_F_ALL_CSUM) &&
5613 !(dev->features & NETIF_F_NO_CSUM)) {
5614 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5615 dev->features |= NETIF_F_NOCACHE_COPY;
5618 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5620 dev->vlan_features |= NETIF_F_HIGHDMA;
5622 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5623 ret = notifier_to_errno(ret);
5627 ret = netdev_register_kobject(dev);
5630 dev->reg_state = NETREG_REGISTERED;
5632 __netdev_update_features(dev);
5635 * Default initial state at registry is that the
5636 * device is present.
5639 set_bit(__LINK_STATE_PRESENT, &dev->state);
5641 dev_init_scheduler(dev);
5643 list_netdevice(dev);
5644 add_device_randomness(dev->dev_addr, dev->addr_len);
5646 /* Notify protocols, that a new device appeared. */
5647 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5648 ret = notifier_to_errno(ret);
5650 rollback_registered(dev);
5651 dev->reg_state = NETREG_UNREGISTERED;
5654 * Prevent userspace races by waiting until the network
5655 * device is fully setup before sending notifications.
5657 if (!dev->rtnl_link_ops ||
5658 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5659 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5665 if (dev->netdev_ops->ndo_uninit)
5666 dev->netdev_ops->ndo_uninit(dev);
5669 EXPORT_SYMBOL(register_netdevice);
5672 * init_dummy_netdev - init a dummy network device for NAPI
5673 * @dev: device to init
5675 * This takes a network device structure and initialize the minimum
5676 * amount of fields so it can be used to schedule NAPI polls without
5677 * registering a full blown interface. This is to be used by drivers
5678 * that need to tie several hardware interfaces to a single NAPI
5679 * poll scheduler due to HW limitations.
5681 int init_dummy_netdev(struct net_device *dev)
5683 /* Clear everything. Note we don't initialize spinlocks
5684 * are they aren't supposed to be taken by any of the
5685 * NAPI code and this dummy netdev is supposed to be
5686 * only ever used for NAPI polls
5688 memset(dev, 0, sizeof(struct net_device));
5690 /* make sure we BUG if trying to hit standard
5691 * register/unregister code path
5693 dev->reg_state = NETREG_DUMMY;
5695 /* NAPI wants this */
5696 INIT_LIST_HEAD(&dev->napi_list);
5698 /* a dummy interface is started by default */
5699 set_bit(__LINK_STATE_PRESENT, &dev->state);
5700 set_bit(__LINK_STATE_START, &dev->state);
5702 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5703 * because users of this 'device' dont need to change
5709 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5713 * register_netdev - register a network device
5714 * @dev: device to register
5716 * Take a completed network device structure and add it to the kernel
5717 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5718 * chain. 0 is returned on success. A negative errno code is returned
5719 * on a failure to set up the device, or if the name is a duplicate.
5721 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5722 * and expands the device name if you passed a format string to
5725 int register_netdev(struct net_device *dev)
5730 err = register_netdevice(dev);
5734 EXPORT_SYMBOL(register_netdev);
5736 int netdev_refcnt_read(const struct net_device *dev)
5740 for_each_possible_cpu(i)
5741 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5744 EXPORT_SYMBOL(netdev_refcnt_read);
5747 * netdev_wait_allrefs - wait until all references are gone.
5749 * This is called when unregistering network devices.
5751 * Any protocol or device that holds a reference should register
5752 * for netdevice notification, and cleanup and put back the
5753 * reference if they receive an UNREGISTER event.
5754 * We can get stuck here if buggy protocols don't correctly
5757 static void netdev_wait_allrefs(struct net_device *dev)
5759 unsigned long rebroadcast_time, warning_time;
5762 linkwatch_forget_dev(dev);
5764 rebroadcast_time = warning_time = jiffies;
5765 refcnt = netdev_refcnt_read(dev);
5767 while (refcnt != 0) {
5768 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5771 /* Rebroadcast unregister notification */
5772 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5773 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5774 * should have already handle it the first time */
5776 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5778 /* We must not have linkwatch events
5779 * pending on unregister. If this
5780 * happens, we simply run the queue
5781 * unscheduled, resulting in a noop
5784 linkwatch_run_queue();
5789 rebroadcast_time = jiffies;
5794 refcnt = netdev_refcnt_read(dev);
5796 if (time_after(jiffies, warning_time + 10 * HZ)) {
5797 printk(KERN_EMERG "unregister_netdevice: "
5798 "waiting for %s to become free. Usage "
5801 warning_time = jiffies;
5810 * register_netdevice(x1);
5811 * register_netdevice(x2);
5813 * unregister_netdevice(y1);
5814 * unregister_netdevice(y2);
5820 * We are invoked by rtnl_unlock().
5821 * This allows us to deal with problems:
5822 * 1) We can delete sysfs objects which invoke hotplug
5823 * without deadlocking with linkwatch via keventd.
5824 * 2) Since we run with the RTNL semaphore not held, we can sleep
5825 * safely in order to wait for the netdev refcnt to drop to zero.
5827 * We must not return until all unregister events added during
5828 * the interval the lock was held have been completed.
5830 void netdev_run_todo(void)
5832 struct list_head list;
5834 /* Snapshot list, allow later requests */
5835 list_replace_init(&net_todo_list, &list);
5839 /* Wait for rcu callbacks to finish before attempting to drain
5840 * the device list. This usually avoids a 250ms wait.
5842 if (!list_empty(&list))
5845 while (!list_empty(&list)) {
5846 struct net_device *dev
5847 = list_first_entry(&list, struct net_device, todo_list);
5848 list_del(&dev->todo_list);
5850 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5851 printk(KERN_ERR "network todo '%s' but state %d\n",
5852 dev->name, dev->reg_state);
5857 dev->reg_state = NETREG_UNREGISTERED;
5859 on_each_cpu(flush_backlog, dev, 1);
5861 netdev_wait_allrefs(dev);
5864 BUG_ON(netdev_refcnt_read(dev));
5865 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5866 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5867 WARN_ON(dev->dn_ptr);
5869 if (dev->destructor)
5870 dev->destructor(dev);
5872 /* Free network device */
5873 kobject_put(&dev->dev.kobj);
5877 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5878 * fields in the same order, with only the type differing.
5880 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5881 const struct net_device_stats *netdev_stats)
5883 #if BITS_PER_LONG == 64
5884 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5885 memcpy(stats64, netdev_stats, sizeof(*stats64));
5887 size_t i, n = sizeof(*stats64) / sizeof(u64);
5888 const unsigned long *src = (const unsigned long *)netdev_stats;
5889 u64 *dst = (u64 *)stats64;
5891 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5892 sizeof(*stats64) / sizeof(u64));
5893 for (i = 0; i < n; i++)
5899 * dev_get_stats - get network device statistics
5900 * @dev: device to get statistics from
5901 * @storage: place to store stats
5903 * Get network statistics from device. Return @storage.
5904 * The device driver may provide its own method by setting
5905 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5906 * otherwise the internal statistics structure is used.
5908 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5909 struct rtnl_link_stats64 *storage)
5911 const struct net_device_ops *ops = dev->netdev_ops;
5913 if (ops->ndo_get_stats64) {
5914 memset(storage, 0, sizeof(*storage));
5915 ops->ndo_get_stats64(dev, storage);
5916 } else if (ops->ndo_get_stats) {
5917 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5919 netdev_stats_to_stats64(storage, &dev->stats);
5921 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5924 EXPORT_SYMBOL(dev_get_stats);
5926 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5928 struct netdev_queue *queue = dev_ingress_queue(dev);
5930 #ifdef CONFIG_NET_CLS_ACT
5933 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5936 netdev_init_one_queue(dev, queue, NULL);
5937 queue->qdisc = &noop_qdisc;
5938 queue->qdisc_sleeping = &noop_qdisc;
5939 rcu_assign_pointer(dev->ingress_queue, queue);
5945 * alloc_netdev_mqs - allocate network device
5946 * @sizeof_priv: size of private data to allocate space for
5947 * @name: device name format string
5948 * @setup: callback to initialize device
5949 * @txqs: the number of TX subqueues to allocate
5950 * @rxqs: the number of RX subqueues to allocate
5952 * Allocates a struct net_device with private data area for driver use
5953 * and performs basic initialization. Also allocates subquue structs
5954 * for each queue on the device.
5956 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5957 void (*setup)(struct net_device *),
5958 unsigned int txqs, unsigned int rxqs)
5960 struct net_device *dev;
5962 struct net_device *p;
5964 BUG_ON(strlen(name) >= sizeof(dev->name));
5967 pr_err("alloc_netdev: Unable to allocate device "
5968 "with zero queues.\n");
5974 pr_err("alloc_netdev: Unable to allocate device "
5975 "with zero RX queues.\n");
5980 alloc_size = sizeof(struct net_device);
5982 /* ensure 32-byte alignment of private area */
5983 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5984 alloc_size += sizeof_priv;
5986 /* ensure 32-byte alignment of whole construct */
5987 alloc_size += NETDEV_ALIGN - 1;
5989 p = kzalloc(alloc_size, GFP_KERNEL);
5991 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5995 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5996 dev->padded = (char *)dev - (char *)p;
5998 dev->pcpu_refcnt = alloc_percpu(int);
5999 if (!dev->pcpu_refcnt)
6002 if (dev_addr_init(dev))
6008 dev_net_set(dev, &init_net);
6010 dev->gso_max_size = GSO_MAX_SIZE;
6011 dev->gso_max_segs = GSO_MAX_SEGS;
6013 INIT_LIST_HEAD(&dev->napi_list);
6014 INIT_LIST_HEAD(&dev->unreg_list);
6015 INIT_LIST_HEAD(&dev->link_watch_list);
6016 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6019 dev->num_tx_queues = txqs;
6020 dev->real_num_tx_queues = txqs;
6021 if (netif_alloc_netdev_queues(dev))
6025 dev->num_rx_queues = rxqs;
6026 dev->real_num_rx_queues = rxqs;
6027 if (netif_alloc_rx_queues(dev))
6031 strcpy(dev->name, name);
6032 dev->group = INIT_NETDEV_GROUP;
6040 free_percpu(dev->pcpu_refcnt);
6050 EXPORT_SYMBOL(alloc_netdev_mqs);
6053 * free_netdev - free network device
6056 * This function does the last stage of destroying an allocated device
6057 * interface. The reference to the device object is released.
6058 * If this is the last reference then it will be freed.
6060 void free_netdev(struct net_device *dev)
6062 struct napi_struct *p, *n;
6064 release_net(dev_net(dev));
6071 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6073 /* Flush device addresses */
6074 dev_addr_flush(dev);
6076 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6079 free_percpu(dev->pcpu_refcnt);
6080 dev->pcpu_refcnt = NULL;
6082 /* Compatibility with error handling in drivers */
6083 if (dev->reg_state == NETREG_UNINITIALIZED) {
6084 kfree((char *)dev - dev->padded);
6088 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6089 dev->reg_state = NETREG_RELEASED;
6091 /* will free via device release */
6092 put_device(&dev->dev);
6094 EXPORT_SYMBOL(free_netdev);
6097 * synchronize_net - Synchronize with packet receive processing
6099 * Wait for packets currently being received to be done.
6100 * Does not block later packets from starting.
6102 void synchronize_net(void)
6105 if (rtnl_is_locked())
6106 synchronize_rcu_expedited();
6110 EXPORT_SYMBOL(synchronize_net);
6113 * unregister_netdevice_queue - remove device from the kernel
6117 * This function shuts down a device interface and removes it
6118 * from the kernel tables.
6119 * If head not NULL, device is queued to be unregistered later.
6121 * Callers must hold the rtnl semaphore. You may want
6122 * unregister_netdev() instead of this.
6125 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6130 list_move_tail(&dev->unreg_list, head);
6132 rollback_registered(dev);
6133 /* Finish processing unregister after unlock */
6137 EXPORT_SYMBOL(unregister_netdevice_queue);
6140 * unregister_netdevice_many - unregister many devices
6141 * @head: list of devices
6143 void unregister_netdevice_many(struct list_head *head)
6145 struct net_device *dev;
6147 if (!list_empty(head)) {
6148 rollback_registered_many(head);
6149 list_for_each_entry(dev, head, unreg_list)
6153 EXPORT_SYMBOL(unregister_netdevice_many);
6156 * unregister_netdev - remove device from the kernel
6159 * This function shuts down a device interface and removes it
6160 * from the kernel tables.
6162 * This is just a wrapper for unregister_netdevice that takes
6163 * the rtnl semaphore. In general you want to use this and not
6164 * unregister_netdevice.
6166 void unregister_netdev(struct net_device *dev)
6169 unregister_netdevice(dev);
6172 EXPORT_SYMBOL(unregister_netdev);
6175 * dev_change_net_namespace - move device to different nethost namespace
6177 * @net: network namespace
6178 * @pat: If not NULL name pattern to try if the current device name
6179 * is already taken in the destination network namespace.
6181 * This function shuts down a device interface and moves it
6182 * to a new network namespace. On success 0 is returned, on
6183 * a failure a netagive errno code is returned.
6185 * Callers must hold the rtnl semaphore.
6188 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6194 /* Don't allow namespace local devices to be moved. */
6196 if (dev->features & NETIF_F_NETNS_LOCAL)
6199 /* Ensure the device has been registrered */
6201 if (dev->reg_state != NETREG_REGISTERED)
6204 /* Get out if there is nothing todo */
6206 if (net_eq(dev_net(dev), net))
6209 /* Pick the destination device name, and ensure
6210 * we can use it in the destination network namespace.
6213 if (__dev_get_by_name(net, dev->name)) {
6214 /* We get here if we can't use the current device name */
6217 if (dev_get_valid_name(dev, pat) < 0)
6222 * And now a mini version of register_netdevice unregister_netdevice.
6225 /* If device is running close it first. */
6228 /* And unlink it from device chain */
6230 unlist_netdevice(dev);
6234 /* Shutdown queueing discipline. */
6237 /* Notify protocols, that we are about to destroy
6238 this device. They should clean all the things.
6240 Note that dev->reg_state stays at NETREG_REGISTERED.
6241 This is wanted because this way 8021q and macvlan know
6242 the device is just moving and can keep their slaves up.
6244 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6245 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6246 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6249 * Flush the unicast and multicast chains
6254 /* Actually switch the network namespace */
6255 dev_net_set(dev, net);
6257 /* If there is an ifindex conflict assign a new one */
6258 if (__dev_get_by_index(net, dev->ifindex)) {
6259 int iflink = (dev->iflink == dev->ifindex);
6260 dev->ifindex = dev_new_index(net);
6262 dev->iflink = dev->ifindex;
6265 /* Fixup kobjects */
6266 err = device_rename(&dev->dev, dev->name);
6269 /* Add the device back in the hashes */
6270 list_netdevice(dev);
6272 /* Notify protocols, that a new device appeared. */
6273 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6276 * Prevent userspace races by waiting until the network
6277 * device is fully setup before sending notifications.
6279 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6286 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6288 static int dev_cpu_callback(struct notifier_block *nfb,
6289 unsigned long action,
6292 struct sk_buff **list_skb;
6293 struct sk_buff *skb;
6294 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6295 struct softnet_data *sd, *oldsd;
6297 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6300 local_irq_disable();
6301 cpu = smp_processor_id();
6302 sd = &per_cpu(softnet_data, cpu);
6303 oldsd = &per_cpu(softnet_data, oldcpu);
6305 /* Find end of our completion_queue. */
6306 list_skb = &sd->completion_queue;
6308 list_skb = &(*list_skb)->next;
6309 /* Append completion queue from offline CPU. */
6310 *list_skb = oldsd->completion_queue;
6311 oldsd->completion_queue = NULL;
6313 /* Append output queue from offline CPU. */
6314 if (oldsd->output_queue) {
6315 *sd->output_queue_tailp = oldsd->output_queue;
6316 sd->output_queue_tailp = oldsd->output_queue_tailp;
6317 oldsd->output_queue = NULL;
6318 oldsd->output_queue_tailp = &oldsd->output_queue;
6320 /* Append NAPI poll list from offline CPU. */
6321 if (!list_empty(&oldsd->poll_list)) {
6322 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6323 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6326 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6329 /* Process offline CPU's input_pkt_queue */
6330 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6332 input_queue_head_incr(oldsd);
6334 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6336 input_queue_head_incr(oldsd);
6344 * netdev_increment_features - increment feature set by one
6345 * @all: current feature set
6346 * @one: new feature set
6347 * @mask: mask feature set
6349 * Computes a new feature set after adding a device with feature set
6350 * @one to the master device with current feature set @all. Will not
6351 * enable anything that is off in @mask. Returns the new feature set.
6353 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6355 if (mask & NETIF_F_GEN_CSUM)
6356 mask |= NETIF_F_ALL_CSUM;
6357 mask |= NETIF_F_VLAN_CHALLENGED;
6359 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6360 all &= one | ~NETIF_F_ALL_FOR_ALL;
6362 /* If device needs checksumming, downgrade to it. */
6363 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6364 all &= ~NETIF_F_NO_CSUM;
6366 /* If one device supports hw checksumming, set for all. */
6367 if (all & NETIF_F_GEN_CSUM)
6368 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6372 EXPORT_SYMBOL(netdev_increment_features);
6374 static struct hlist_head *netdev_create_hash(void)
6377 struct hlist_head *hash;
6379 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6381 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6382 INIT_HLIST_HEAD(&hash[i]);
6387 /* Initialize per network namespace state */
6388 static int __net_init netdev_init(struct net *net)
6390 INIT_LIST_HEAD(&net->dev_base_head);
6392 net->dev_name_head = netdev_create_hash();
6393 if (net->dev_name_head == NULL)
6396 net->dev_index_head = netdev_create_hash();
6397 if (net->dev_index_head == NULL)
6403 kfree(net->dev_name_head);
6409 * netdev_drivername - network driver for the device
6410 * @dev: network device
6412 * Determine network driver for device.
6414 const char *netdev_drivername(const struct net_device *dev)
6416 const struct device_driver *driver;
6417 const struct device *parent;
6418 const char *empty = "";
6420 parent = dev->dev.parent;
6424 driver = parent->driver;
6425 if (driver && driver->name)
6426 return driver->name;
6430 int __netdev_printk(const char *level, const struct net_device *dev,
6431 struct va_format *vaf)
6435 if (dev && dev->dev.parent)
6436 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6437 netdev_name(dev), vaf);
6439 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6441 r = printk("%s(NULL net_device): %pV", level, vaf);
6445 EXPORT_SYMBOL(__netdev_printk);
6447 int netdev_printk(const char *level, const struct net_device *dev,
6448 const char *format, ...)
6450 struct va_format vaf;
6454 va_start(args, format);
6459 r = __netdev_printk(level, dev, &vaf);
6464 EXPORT_SYMBOL(netdev_printk);
6466 #define define_netdev_printk_level(func, level) \
6467 int func(const struct net_device *dev, const char *fmt, ...) \
6470 struct va_format vaf; \
6473 va_start(args, fmt); \
6478 r = __netdev_printk(level, dev, &vaf); \
6483 EXPORT_SYMBOL(func);
6485 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6486 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6487 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6488 define_netdev_printk_level(netdev_err, KERN_ERR);
6489 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6490 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6491 define_netdev_printk_level(netdev_info, KERN_INFO);
6493 static void __net_exit netdev_exit(struct net *net)
6495 kfree(net->dev_name_head);
6496 kfree(net->dev_index_head);
6499 static struct pernet_operations __net_initdata netdev_net_ops = {
6500 .init = netdev_init,
6501 .exit = netdev_exit,
6504 static void __net_exit default_device_exit(struct net *net)
6506 struct net_device *dev, *aux;
6508 * Push all migratable network devices back to the
6509 * initial network namespace
6512 for_each_netdev_safe(net, dev, aux) {
6514 char fb_name[IFNAMSIZ];
6516 /* Ignore unmoveable devices (i.e. loopback) */
6517 if (dev->features & NETIF_F_NETNS_LOCAL)
6520 /* Leave virtual devices for the generic cleanup */
6521 if (dev->rtnl_link_ops)
6524 /* Push remaining network devices to init_net */
6525 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6526 err = dev_change_net_namespace(dev, &init_net, fb_name);
6528 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6529 __func__, dev->name, err);
6536 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6538 /* At exit all network devices most be removed from a network
6539 * namespace. Do this in the reverse order of registration.
6540 * Do this across as many network namespaces as possible to
6541 * improve batching efficiency.
6543 struct net_device *dev;
6545 LIST_HEAD(dev_kill_list);
6548 list_for_each_entry(net, net_list, exit_list) {
6549 for_each_netdev_reverse(net, dev) {
6550 if (dev->rtnl_link_ops)
6551 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6553 unregister_netdevice_queue(dev, &dev_kill_list);
6556 unregister_netdevice_many(&dev_kill_list);
6557 list_del(&dev_kill_list);
6561 static struct pernet_operations __net_initdata default_device_ops = {
6562 .exit = default_device_exit,
6563 .exit_batch = default_device_exit_batch,
6567 * Initialize the DEV module. At boot time this walks the device list and
6568 * unhooks any devices that fail to initialise (normally hardware not
6569 * present) and leaves us with a valid list of present and active devices.
6574 * This is called single threaded during boot, so no need
6575 * to take the rtnl semaphore.
6577 static int __init net_dev_init(void)
6579 int i, rc = -ENOMEM;
6581 BUG_ON(!dev_boot_phase);
6583 if (dev_proc_init())
6586 if (netdev_kobject_init())
6589 INIT_LIST_HEAD(&ptype_all);
6590 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6591 INIT_LIST_HEAD(&ptype_base[i]);
6593 if (register_pernet_subsys(&netdev_net_ops))
6597 * Initialise the packet receive queues.
6600 for_each_possible_cpu(i) {
6601 struct softnet_data *sd = &per_cpu(softnet_data, i);
6603 memset(sd, 0, sizeof(*sd));
6604 skb_queue_head_init(&sd->input_pkt_queue);
6605 skb_queue_head_init(&sd->process_queue);
6606 sd->completion_queue = NULL;
6607 INIT_LIST_HEAD(&sd->poll_list);
6608 sd->output_queue = NULL;
6609 sd->output_queue_tailp = &sd->output_queue;
6611 sd->csd.func = rps_trigger_softirq;
6617 sd->backlog.poll = process_backlog;
6618 sd->backlog.weight = weight_p;
6619 sd->backlog.gro_list = NULL;
6620 sd->backlog.gro_count = 0;
6625 /* The loopback device is special if any other network devices
6626 * is present in a network namespace the loopback device must
6627 * be present. Since we now dynamically allocate and free the
6628 * loopback device ensure this invariant is maintained by
6629 * keeping the loopback device as the first device on the
6630 * list of network devices. Ensuring the loopback devices
6631 * is the first device that appears and the last network device
6634 if (register_pernet_device(&loopback_net_ops))
6637 if (register_pernet_device(&default_device_ops))
6640 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6641 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6643 hotcpu_notifier(dev_cpu_callback, 0);
6651 subsys_initcall(net_dev_init);
6653 static int __init initialize_hashrnd(void)
6655 get_random_bytes(&hashrnd, sizeof(hashrnd));
6659 late_initcall_sync(initialize_hashrnd);