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)
1064 if (len >= IFALIASZ)
1069 kfree(dev->ifalias);
1070 dev->ifalias = NULL;
1075 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1079 strlcpy(dev->ifalias, alias, len+1);
1085 * netdev_features_change - device changes features
1086 * @dev: device to cause notification
1088 * Called to indicate a device has changed features.
1090 void netdev_features_change(struct net_device *dev)
1092 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1094 EXPORT_SYMBOL(netdev_features_change);
1097 * netdev_state_change - device changes state
1098 * @dev: device to cause notification
1100 * Called to indicate a device has changed state. This function calls
1101 * the notifier chains for netdev_chain and sends a NEWLINK message
1102 * to the routing socket.
1104 void netdev_state_change(struct net_device *dev)
1106 if (dev->flags & IFF_UP) {
1107 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1108 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1111 EXPORT_SYMBOL(netdev_state_change);
1113 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1115 return call_netdevice_notifiers(event, dev);
1117 EXPORT_SYMBOL(netdev_bonding_change);
1120 * dev_load - load a network module
1121 * @net: the applicable net namespace
1122 * @name: name of interface
1124 * If a network interface is not present and the process has suitable
1125 * privileges this function loads the module. If module loading is not
1126 * available in this kernel then it becomes a nop.
1129 void dev_load(struct net *net, const char *name)
1131 struct net_device *dev;
1135 dev = dev_get_by_name_rcu(net, name);
1139 if (no_module && capable(CAP_NET_ADMIN))
1140 no_module = request_module("netdev-%s", name);
1141 if (no_module && capable(CAP_SYS_MODULE)) {
1142 if (!request_module("%s", name))
1143 pr_err("Loading kernel module for a network device "
1144 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1148 EXPORT_SYMBOL(dev_load);
1150 static int __dev_open(struct net_device *dev)
1152 const struct net_device_ops *ops = dev->netdev_ops;
1157 if (!netif_device_present(dev))
1160 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1161 ret = notifier_to_errno(ret);
1165 set_bit(__LINK_STATE_START, &dev->state);
1167 if (ops->ndo_validate_addr)
1168 ret = ops->ndo_validate_addr(dev);
1170 if (!ret && ops->ndo_open)
1171 ret = ops->ndo_open(dev);
1174 clear_bit(__LINK_STATE_START, &dev->state);
1176 dev->flags |= IFF_UP;
1177 net_dmaengine_get();
1178 dev_set_rx_mode(dev);
1186 * dev_open - prepare an interface for use.
1187 * @dev: device to open
1189 * Takes a device from down to up state. The device's private open
1190 * function is invoked and then the multicast lists are loaded. Finally
1191 * the device is moved into the up state and a %NETDEV_UP message is
1192 * sent to the netdev notifier chain.
1194 * Calling this function on an active interface is a nop. On a failure
1195 * a negative errno code is returned.
1197 int dev_open(struct net_device *dev)
1201 if (dev->flags & IFF_UP)
1204 ret = __dev_open(dev);
1208 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1209 call_netdevice_notifiers(NETDEV_UP, dev);
1213 EXPORT_SYMBOL(dev_open);
1215 static int __dev_close_many(struct list_head *head)
1217 struct net_device *dev;
1222 list_for_each_entry(dev, head, unreg_list) {
1223 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1225 clear_bit(__LINK_STATE_START, &dev->state);
1227 /* Synchronize to scheduled poll. We cannot touch poll list, it
1228 * can be even on different cpu. So just clear netif_running().
1230 * dev->stop() will invoke napi_disable() on all of it's
1231 * napi_struct instances on this device.
1233 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1236 dev_deactivate_many(head);
1238 list_for_each_entry(dev, head, unreg_list) {
1239 const struct net_device_ops *ops = dev->netdev_ops;
1242 * Call the device specific close. This cannot fail.
1243 * Only if device is UP
1245 * We allow it to be called even after a DETACH hot-plug
1251 dev->flags &= ~IFF_UP;
1252 net_dmaengine_put();
1258 static int __dev_close(struct net_device *dev)
1263 list_add(&dev->unreg_list, &single);
1264 retval = __dev_close_many(&single);
1269 static int dev_close_many(struct list_head *head)
1271 struct net_device *dev, *tmp;
1272 LIST_HEAD(tmp_list);
1274 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1275 if (!(dev->flags & IFF_UP))
1276 list_move(&dev->unreg_list, &tmp_list);
1278 __dev_close_many(head);
1280 list_for_each_entry(dev, head, unreg_list) {
1281 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1282 call_netdevice_notifiers(NETDEV_DOWN, dev);
1285 /* rollback_registered_many needs the complete original list */
1286 list_splice(&tmp_list, head);
1291 * dev_close - shutdown an interface.
1292 * @dev: device to shutdown
1294 * This function moves an active device into down state. A
1295 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1296 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1299 int dev_close(struct net_device *dev)
1301 if (dev->flags & IFF_UP) {
1304 list_add(&dev->unreg_list, &single);
1305 dev_close_many(&single);
1310 EXPORT_SYMBOL(dev_close);
1314 * dev_disable_lro - disable Large Receive Offload on a device
1317 * Disable Large Receive Offload (LRO) on a net device. Must be
1318 * called under RTNL. This is needed if received packets may be
1319 * forwarded to another interface.
1321 void dev_disable_lro(struct net_device *dev)
1326 * If we're trying to disable lro on a vlan device
1327 * use the underlying physical device instead
1329 if (is_vlan_dev(dev))
1330 dev = vlan_dev_real_dev(dev);
1332 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1333 flags = dev->ethtool_ops->get_flags(dev);
1335 flags = ethtool_op_get_flags(dev);
1337 if (!(flags & ETH_FLAG_LRO))
1340 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1341 if (unlikely(dev->features & NETIF_F_LRO))
1342 netdev_WARN(dev, "failed to disable LRO!\n");
1344 EXPORT_SYMBOL(dev_disable_lro);
1347 static int dev_boot_phase = 1;
1350 * register_netdevice_notifier - register a network notifier block
1353 * Register a notifier to be called when network device events occur.
1354 * The notifier passed is linked into the kernel structures and must
1355 * not be reused until it has been unregistered. A negative errno code
1356 * is returned on a failure.
1358 * When registered all registration and up events are replayed
1359 * to the new notifier to allow device to have a race free
1360 * view of the network device list.
1363 int register_netdevice_notifier(struct notifier_block *nb)
1365 struct net_device *dev;
1366 struct net_device *last;
1371 err = raw_notifier_chain_register(&netdev_chain, nb);
1377 for_each_netdev(net, dev) {
1378 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1379 err = notifier_to_errno(err);
1383 if (!(dev->flags & IFF_UP))
1386 nb->notifier_call(nb, NETDEV_UP, dev);
1397 for_each_netdev(net, dev) {
1401 if (dev->flags & IFF_UP) {
1402 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1403 nb->notifier_call(nb, NETDEV_DOWN, dev);
1405 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1406 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1411 raw_notifier_chain_unregister(&netdev_chain, nb);
1414 EXPORT_SYMBOL(register_netdevice_notifier);
1417 * unregister_netdevice_notifier - unregister a network notifier block
1420 * Unregister a notifier previously registered by
1421 * register_netdevice_notifier(). The notifier is unlinked into the
1422 * kernel structures and may then be reused. A negative errno code
1423 * is returned on a failure.
1425 * After unregistering unregister and down device events are synthesized
1426 * for all devices on the device list to the removed notifier to remove
1427 * the need for special case cleanup code.
1430 int unregister_netdevice_notifier(struct notifier_block *nb)
1432 struct net_device *dev;
1437 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1442 for_each_netdev(net, dev) {
1443 if (dev->flags & IFF_UP) {
1444 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1445 nb->notifier_call(nb, NETDEV_DOWN, dev);
1447 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1448 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1455 EXPORT_SYMBOL(unregister_netdevice_notifier);
1458 * call_netdevice_notifiers - call all network notifier blocks
1459 * @val: value passed unmodified to notifier function
1460 * @dev: net_device pointer passed unmodified to notifier function
1462 * Call all network notifier blocks. Parameters and return value
1463 * are as for raw_notifier_call_chain().
1466 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1469 return raw_notifier_call_chain(&netdev_chain, val, dev);
1471 EXPORT_SYMBOL(call_netdevice_notifiers);
1473 /* When > 0 there are consumers of rx skb time stamps */
1474 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1476 void net_enable_timestamp(void)
1478 atomic_inc(&netstamp_needed);
1480 EXPORT_SYMBOL(net_enable_timestamp);
1482 void net_disable_timestamp(void)
1484 atomic_dec(&netstamp_needed);
1486 EXPORT_SYMBOL(net_disable_timestamp);
1488 static inline void net_timestamp_set(struct sk_buff *skb)
1490 if (atomic_read(&netstamp_needed))
1491 __net_timestamp(skb);
1493 skb->tstamp.tv64 = 0;
1496 static inline void net_timestamp_check(struct sk_buff *skb)
1498 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1499 __net_timestamp(skb);
1502 static int net_hwtstamp_validate(struct ifreq *ifr)
1504 struct hwtstamp_config cfg;
1505 enum hwtstamp_tx_types tx_type;
1506 enum hwtstamp_rx_filters rx_filter;
1507 int tx_type_valid = 0;
1508 int rx_filter_valid = 0;
1510 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1513 if (cfg.flags) /* reserved for future extensions */
1516 tx_type = cfg.tx_type;
1517 rx_filter = cfg.rx_filter;
1520 case HWTSTAMP_TX_OFF:
1521 case HWTSTAMP_TX_ON:
1522 case HWTSTAMP_TX_ONESTEP_SYNC:
1527 switch (rx_filter) {
1528 case HWTSTAMP_FILTER_NONE:
1529 case HWTSTAMP_FILTER_ALL:
1530 case HWTSTAMP_FILTER_SOME:
1531 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1532 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1533 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1534 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1535 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1536 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1537 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1538 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1539 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1540 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1541 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1542 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1543 rx_filter_valid = 1;
1547 if (!tx_type_valid || !rx_filter_valid)
1553 static inline bool is_skb_forwardable(struct net_device *dev,
1554 struct sk_buff *skb)
1558 if (!(dev->flags & IFF_UP))
1561 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1562 if (skb->len <= len)
1565 /* if TSO is enabled, we don't care about the length as the packet
1566 * could be forwarded without being segmented before
1568 if (skb_is_gso(skb))
1575 * dev_forward_skb - loopback an skb to another netif
1577 * @dev: destination network device
1578 * @skb: buffer to forward
1581 * NET_RX_SUCCESS (no congestion)
1582 * NET_RX_DROP (packet was dropped, but freed)
1584 * dev_forward_skb can be used for injecting an skb from the
1585 * start_xmit function of one device into the receive queue
1586 * of another device.
1588 * The receiving device may be in another namespace, so
1589 * we have to clear all information in the skb that could
1590 * impact namespace isolation.
1592 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1594 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1595 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1596 atomic_long_inc(&dev->rx_dropped);
1605 if (unlikely(!is_skb_forwardable(dev, skb))) {
1606 atomic_long_inc(&dev->rx_dropped);
1612 skb->tstamp.tv64 = 0;
1613 skb->pkt_type = PACKET_HOST;
1614 skb->protocol = eth_type_trans(skb, dev);
1618 return netif_rx(skb);
1620 EXPORT_SYMBOL_GPL(dev_forward_skb);
1622 static inline int deliver_skb(struct sk_buff *skb,
1623 struct packet_type *pt_prev,
1624 struct net_device *orig_dev)
1626 atomic_inc(&skb->users);
1627 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1631 * Support routine. Sends outgoing frames to any network
1632 * taps currently in use.
1635 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1637 struct packet_type *ptype;
1638 struct sk_buff *skb2 = NULL;
1639 struct packet_type *pt_prev = NULL;
1642 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1643 /* Never send packets back to the socket
1644 * they originated from - MvS (miquels@drinkel.ow.org)
1646 if ((ptype->dev == dev || !ptype->dev) &&
1647 (ptype->af_packet_priv == NULL ||
1648 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1650 deliver_skb(skb2, pt_prev, skb->dev);
1655 skb2 = skb_clone(skb, GFP_ATOMIC);
1659 net_timestamp_set(skb2);
1661 /* skb->nh should be correctly
1662 set by sender, so that the second statement is
1663 just protection against buggy protocols.
1665 skb_reset_mac_header(skb2);
1667 if (skb_network_header(skb2) < skb2->data ||
1668 skb2->network_header > skb2->tail) {
1669 if (net_ratelimit())
1670 printk(KERN_CRIT "protocol %04x is "
1672 ntohs(skb2->protocol),
1674 skb_reset_network_header(skb2);
1677 skb2->transport_header = skb2->network_header;
1678 skb2->pkt_type = PACKET_OUTGOING;
1683 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1687 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1688 * @dev: Network device
1689 * @txq: number of queues available
1691 * If real_num_tx_queues is changed the tc mappings may no longer be
1692 * valid. To resolve this verify the tc mapping remains valid and if
1693 * not NULL the mapping. With no priorities mapping to this
1694 * offset/count pair it will no longer be used. In the worst case TC0
1695 * is invalid nothing can be done so disable priority mappings. If is
1696 * expected that drivers will fix this mapping if they can before
1697 * calling netif_set_real_num_tx_queues.
1699 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1702 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1704 /* If TC0 is invalidated disable TC mapping */
1705 if (tc->offset + tc->count > txq) {
1706 pr_warning("Number of in use tx queues changed "
1707 "invalidating tc mappings. Priority "
1708 "traffic classification disabled!\n");
1713 /* Invalidated prio to tc mappings set to TC0 */
1714 for (i = 1; i < TC_BITMASK + 1; i++) {
1715 int q = netdev_get_prio_tc_map(dev, i);
1717 tc = &dev->tc_to_txq[q];
1718 if (tc->offset + tc->count > txq) {
1719 pr_warning("Number of in use tx queues "
1720 "changed. Priority %i to tc "
1721 "mapping %i is no longer valid "
1722 "setting map to 0\n",
1724 netdev_set_prio_tc_map(dev, i, 0);
1730 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1731 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1733 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1737 if (txq < 1 || txq > dev->num_tx_queues)
1740 if (dev->reg_state == NETREG_REGISTERED ||
1741 dev->reg_state == NETREG_UNREGISTERING) {
1744 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1750 netif_setup_tc(dev, txq);
1752 if (txq < dev->real_num_tx_queues)
1753 qdisc_reset_all_tx_gt(dev, txq);
1756 dev->real_num_tx_queues = txq;
1759 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1763 * netif_set_real_num_rx_queues - set actual number of RX queues used
1764 * @dev: Network device
1765 * @rxq: Actual number of RX queues
1767 * This must be called either with the rtnl_lock held or before
1768 * registration of the net device. Returns 0 on success, or a
1769 * negative error code. If called before registration, it always
1772 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1776 if (rxq < 1 || rxq > dev->num_rx_queues)
1779 if (dev->reg_state == NETREG_REGISTERED) {
1782 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1788 dev->real_num_rx_queues = rxq;
1791 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1794 static inline void __netif_reschedule(struct Qdisc *q)
1796 struct softnet_data *sd;
1797 unsigned long flags;
1799 local_irq_save(flags);
1800 sd = &__get_cpu_var(softnet_data);
1801 q->next_sched = NULL;
1802 *sd->output_queue_tailp = q;
1803 sd->output_queue_tailp = &q->next_sched;
1804 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1805 local_irq_restore(flags);
1808 void __netif_schedule(struct Qdisc *q)
1810 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1811 __netif_reschedule(q);
1813 EXPORT_SYMBOL(__netif_schedule);
1815 void dev_kfree_skb_irq(struct sk_buff *skb)
1817 if (atomic_dec_and_test(&skb->users)) {
1818 struct softnet_data *sd;
1819 unsigned long flags;
1821 local_irq_save(flags);
1822 sd = &__get_cpu_var(softnet_data);
1823 skb->next = sd->completion_queue;
1824 sd->completion_queue = skb;
1825 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1826 local_irq_restore(flags);
1829 EXPORT_SYMBOL(dev_kfree_skb_irq);
1831 void dev_kfree_skb_any(struct sk_buff *skb)
1833 if (in_irq() || irqs_disabled())
1834 dev_kfree_skb_irq(skb);
1838 EXPORT_SYMBOL(dev_kfree_skb_any);
1842 * netif_device_detach - mark device as removed
1843 * @dev: network device
1845 * Mark device as removed from system and therefore no longer available.
1847 void netif_device_detach(struct net_device *dev)
1849 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1850 netif_running(dev)) {
1851 netif_tx_stop_all_queues(dev);
1854 EXPORT_SYMBOL(netif_device_detach);
1857 * netif_device_attach - mark device as attached
1858 * @dev: network device
1860 * Mark device as attached from system and restart if needed.
1862 void netif_device_attach(struct net_device *dev)
1864 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1865 netif_running(dev)) {
1866 netif_tx_wake_all_queues(dev);
1867 __netdev_watchdog_up(dev);
1870 EXPORT_SYMBOL(netif_device_attach);
1873 * Invalidate hardware checksum when packet is to be mangled, and
1874 * complete checksum manually on outgoing path.
1876 int skb_checksum_help(struct sk_buff *skb)
1879 int ret = 0, offset;
1881 if (skb->ip_summed == CHECKSUM_COMPLETE)
1882 goto out_set_summed;
1884 if (unlikely(skb_shinfo(skb)->gso_size)) {
1885 /* Let GSO fix up the checksum. */
1886 goto out_set_summed;
1889 offset = skb_checksum_start_offset(skb);
1890 BUG_ON(offset >= skb_headlen(skb));
1891 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1893 offset += skb->csum_offset;
1894 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1896 if (skb_cloned(skb) &&
1897 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1898 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1903 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1905 skb->ip_summed = CHECKSUM_NONE;
1909 EXPORT_SYMBOL(skb_checksum_help);
1912 * skb_gso_segment - Perform segmentation on skb.
1913 * @skb: buffer to segment
1914 * @features: features for the output path (see dev->features)
1916 * This function segments the given skb and returns a list of segments.
1918 * It may return NULL if the skb requires no segmentation. This is
1919 * only possible when GSO is used for verifying header integrity.
1921 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1923 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1924 struct packet_type *ptype;
1925 __be16 type = skb->protocol;
1926 int vlan_depth = ETH_HLEN;
1929 while (type == htons(ETH_P_8021Q)) {
1930 struct vlan_hdr *vh;
1932 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1933 return ERR_PTR(-EINVAL);
1935 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1936 type = vh->h_vlan_encapsulated_proto;
1937 vlan_depth += VLAN_HLEN;
1940 skb_reset_mac_header(skb);
1941 skb->mac_len = skb->network_header - skb->mac_header;
1942 __skb_pull(skb, skb->mac_len);
1944 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1945 struct net_device *dev = skb->dev;
1946 struct ethtool_drvinfo info = {};
1948 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1949 dev->ethtool_ops->get_drvinfo(dev, &info);
1951 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1952 info.driver, dev ? dev->features : 0L,
1953 skb->sk ? skb->sk->sk_route_caps : 0L,
1954 skb->len, skb->data_len, skb->ip_summed);
1956 if (skb_header_cloned(skb) &&
1957 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1958 return ERR_PTR(err);
1962 list_for_each_entry_rcu(ptype,
1963 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1964 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1965 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1966 err = ptype->gso_send_check(skb);
1967 segs = ERR_PTR(err);
1968 if (err || skb_gso_ok(skb, features))
1970 __skb_push(skb, (skb->data -
1971 skb_network_header(skb)));
1973 segs = ptype->gso_segment(skb, features);
1979 __skb_push(skb, skb->data - skb_mac_header(skb));
1983 EXPORT_SYMBOL(skb_gso_segment);
1985 /* Take action when hardware reception checksum errors are detected. */
1987 void netdev_rx_csum_fault(struct net_device *dev)
1989 if (net_ratelimit()) {
1990 printk(KERN_ERR "%s: hw csum failure.\n",
1991 dev ? dev->name : "<unknown>");
1995 EXPORT_SYMBOL(netdev_rx_csum_fault);
1998 /* Actually, we should eliminate this check as soon as we know, that:
1999 * 1. IOMMU is present and allows to map all the memory.
2000 * 2. No high memory really exists on this machine.
2003 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2005 #ifdef CONFIG_HIGHMEM
2007 if (!(dev->features & NETIF_F_HIGHDMA)) {
2008 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2009 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2010 if (PageHighMem(skb_frag_page(frag)))
2015 if (PCI_DMA_BUS_IS_PHYS) {
2016 struct device *pdev = dev->dev.parent;
2020 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2021 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2022 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2023 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2032 void (*destructor)(struct sk_buff *skb);
2035 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2037 static void dev_gso_skb_destructor(struct sk_buff *skb)
2039 struct dev_gso_cb *cb;
2042 struct sk_buff *nskb = skb->next;
2044 skb->next = nskb->next;
2047 } while (skb->next);
2049 cb = DEV_GSO_CB(skb);
2051 cb->destructor(skb);
2055 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2056 * @skb: buffer to segment
2057 * @features: device features as applicable to this skb
2059 * This function segments the given skb and stores the list of segments
2062 static int dev_gso_segment(struct sk_buff *skb, int features)
2064 struct sk_buff *segs;
2066 segs = skb_gso_segment(skb, features);
2068 /* Verifying header integrity only. */
2073 return PTR_ERR(segs);
2076 DEV_GSO_CB(skb)->destructor = skb->destructor;
2077 skb->destructor = dev_gso_skb_destructor;
2083 * Try to orphan skb early, right before transmission by the device.
2084 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2085 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2087 static inline void skb_orphan_try(struct sk_buff *skb)
2089 struct sock *sk = skb->sk;
2091 if (sk && !skb_shinfo(skb)->tx_flags) {
2092 /* skb_tx_hash() wont be able to get sk.
2093 * We copy sk_hash into skb->rxhash
2096 skb->rxhash = sk->sk_hash;
2101 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2103 return ((features & NETIF_F_GEN_CSUM) ||
2104 ((features & NETIF_F_V4_CSUM) &&
2105 protocol == htons(ETH_P_IP)) ||
2106 ((features & NETIF_F_V6_CSUM) &&
2107 protocol == htons(ETH_P_IPV6)) ||
2108 ((features & NETIF_F_FCOE_CRC) &&
2109 protocol == htons(ETH_P_FCOE)));
2112 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2114 if (!can_checksum_protocol(features, protocol)) {
2115 features &= ~NETIF_F_ALL_CSUM;
2116 features &= ~NETIF_F_SG;
2117 } else if (illegal_highdma(skb->dev, skb)) {
2118 features &= ~NETIF_F_SG;
2124 u32 netif_skb_features(struct sk_buff *skb)
2126 __be16 protocol = skb->protocol;
2127 u32 features = skb->dev->features;
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 skb_orphan_try(skb);
2187 features = netif_skb_features(skb);
2189 if (vlan_tx_tag_present(skb) &&
2190 !(features & NETIF_F_HW_VLAN_TX)) {
2191 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2198 if (netif_needs_gso(skb, features)) {
2199 if (unlikely(dev_gso_segment(skb, features)))
2204 if (skb_needs_linearize(skb, features) &&
2205 __skb_linearize(skb))
2208 /* If packet is not checksummed and device does not
2209 * support checksumming for this protocol, complete
2210 * checksumming here.
2212 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2213 skb_set_transport_header(skb,
2214 skb_checksum_start_offset(skb));
2215 if (!(features & NETIF_F_ALL_CSUM) &&
2216 skb_checksum_help(skb))
2222 rc = ops->ndo_start_xmit(skb, dev);
2223 trace_net_dev_xmit(skb, rc, dev, skb_len);
2224 if (rc == NETDEV_TX_OK)
2225 txq_trans_update(txq);
2231 struct sk_buff *nskb = skb->next;
2233 skb->next = nskb->next;
2237 * If device doesn't need nskb->dst, release it right now while
2238 * its hot in this cpu cache
2240 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2243 skb_len = nskb->len;
2244 rc = ops->ndo_start_xmit(nskb, dev);
2245 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2246 if (unlikely(rc != NETDEV_TX_OK)) {
2247 if (rc & ~NETDEV_TX_MASK)
2248 goto out_kfree_gso_skb;
2249 nskb->next = skb->next;
2253 txq_trans_update(txq);
2254 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2255 return NETDEV_TX_BUSY;
2256 } while (skb->next);
2259 if (likely(skb->next == NULL))
2260 skb->destructor = DEV_GSO_CB(skb)->destructor;
2267 static u32 hashrnd __read_mostly;
2270 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2271 * to be used as a distribution range.
2273 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2274 unsigned int num_tx_queues)
2278 u16 qcount = num_tx_queues;
2280 if (skb_rx_queue_recorded(skb)) {
2281 hash = skb_get_rx_queue(skb);
2282 while (unlikely(hash >= num_tx_queues))
2283 hash -= num_tx_queues;
2288 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2289 qoffset = dev->tc_to_txq[tc].offset;
2290 qcount = dev->tc_to_txq[tc].count;
2293 if (skb->sk && skb->sk->sk_hash)
2294 hash = skb->sk->sk_hash;
2296 hash = (__force u16) skb->protocol ^ skb->rxhash;
2297 hash = jhash_1word(hash, hashrnd);
2299 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2301 EXPORT_SYMBOL(__skb_tx_hash);
2303 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2305 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2306 if (net_ratelimit()) {
2307 pr_warning("%s selects TX queue %d, but "
2308 "real number of TX queues is %d\n",
2309 dev->name, queue_index, dev->real_num_tx_queues);
2316 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2319 struct xps_dev_maps *dev_maps;
2320 struct xps_map *map;
2321 int queue_index = -1;
2324 dev_maps = rcu_dereference(dev->xps_maps);
2326 map = rcu_dereference(
2327 dev_maps->cpu_map[raw_smp_processor_id()]);
2330 queue_index = map->queues[0];
2333 if (skb->sk && skb->sk->sk_hash)
2334 hash = skb->sk->sk_hash;
2336 hash = (__force u16) skb->protocol ^
2338 hash = jhash_1word(hash, hashrnd);
2339 queue_index = map->queues[
2340 ((u64)hash * map->len) >> 32];
2342 if (unlikely(queue_index >= dev->real_num_tx_queues))
2354 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2355 struct sk_buff *skb)
2358 const struct net_device_ops *ops = dev->netdev_ops;
2360 if (dev->real_num_tx_queues == 1)
2362 else if (ops->ndo_select_queue) {
2363 queue_index = ops->ndo_select_queue(dev, skb);
2364 queue_index = dev_cap_txqueue(dev, queue_index);
2366 struct sock *sk = skb->sk;
2367 queue_index = sk_tx_queue_get(sk);
2369 if (queue_index < 0 || skb->ooo_okay ||
2370 queue_index >= dev->real_num_tx_queues) {
2371 int old_index = queue_index;
2373 queue_index = get_xps_queue(dev, skb);
2374 if (queue_index < 0)
2375 queue_index = skb_tx_hash(dev, skb);
2377 if (queue_index != old_index && sk) {
2378 struct dst_entry *dst =
2379 rcu_dereference_check(sk->sk_dst_cache, 1);
2381 if (dst && skb_dst(skb) == dst)
2382 sk_tx_queue_set(sk, queue_index);
2387 skb_set_queue_mapping(skb, queue_index);
2388 return netdev_get_tx_queue(dev, queue_index);
2391 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2392 struct net_device *dev,
2393 struct netdev_queue *txq)
2395 spinlock_t *root_lock = qdisc_lock(q);
2399 qdisc_skb_cb(skb)->pkt_len = skb->len;
2400 qdisc_calculate_pkt_len(skb, q);
2402 * Heuristic to force contended enqueues to serialize on a
2403 * separate lock before trying to get qdisc main lock.
2404 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2405 * and dequeue packets faster.
2407 contended = qdisc_is_running(q);
2408 if (unlikely(contended))
2409 spin_lock(&q->busylock);
2411 spin_lock(root_lock);
2412 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2415 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2416 qdisc_run_begin(q)) {
2418 * This is a work-conserving queue; there are no old skbs
2419 * waiting to be sent out; and the qdisc is not running -
2420 * xmit the skb directly.
2422 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2425 qdisc_bstats_update(q, skb);
2427 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2428 if (unlikely(contended)) {
2429 spin_unlock(&q->busylock);
2436 rc = NET_XMIT_SUCCESS;
2439 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2440 if (qdisc_run_begin(q)) {
2441 if (unlikely(contended)) {
2442 spin_unlock(&q->busylock);
2448 spin_unlock(root_lock);
2449 if (unlikely(contended))
2450 spin_unlock(&q->busylock);
2454 static DEFINE_PER_CPU(int, xmit_recursion);
2455 #define RECURSION_LIMIT 10
2458 * dev_queue_xmit - transmit a buffer
2459 * @skb: buffer to transmit
2461 * Queue a buffer for transmission to a network device. The caller must
2462 * have set the device and priority and built the buffer before calling
2463 * this function. The function can be called from an interrupt.
2465 * A negative errno code is returned on a failure. A success does not
2466 * guarantee the frame will be transmitted as it may be dropped due
2467 * to congestion or traffic shaping.
2469 * -----------------------------------------------------------------------------------
2470 * I notice this method can also return errors from the queue disciplines,
2471 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2474 * Regardless of the return value, the skb is consumed, so it is currently
2475 * difficult to retry a send to this method. (You can bump the ref count
2476 * before sending to hold a reference for retry if you are careful.)
2478 * When calling this method, interrupts MUST be enabled. This is because
2479 * the BH enable code must have IRQs enabled so that it will not deadlock.
2482 int dev_queue_xmit(struct sk_buff *skb)
2484 struct net_device *dev = skb->dev;
2485 struct netdev_queue *txq;
2489 /* Disable soft irqs for various locks below. Also
2490 * stops preemption for RCU.
2494 txq = dev_pick_tx(dev, skb);
2495 q = rcu_dereference_bh(txq->qdisc);
2497 #ifdef CONFIG_NET_CLS_ACT
2498 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2500 trace_net_dev_queue(skb);
2502 rc = __dev_xmit_skb(skb, q, dev, txq);
2506 /* The device has no queue. Common case for software devices:
2507 loopback, all the sorts of tunnels...
2509 Really, it is unlikely that netif_tx_lock protection is necessary
2510 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2512 However, it is possible, that they rely on protection
2515 Check this and shot the lock. It is not prone from deadlocks.
2516 Either shot noqueue qdisc, it is even simpler 8)
2518 if (dev->flags & IFF_UP) {
2519 int cpu = smp_processor_id(); /* ok because BHs are off */
2521 if (txq->xmit_lock_owner != cpu) {
2523 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2524 goto recursion_alert;
2526 HARD_TX_LOCK(dev, txq, cpu);
2528 if (!netif_tx_queue_stopped(txq)) {
2529 __this_cpu_inc(xmit_recursion);
2530 rc = dev_hard_start_xmit(skb, dev, txq);
2531 __this_cpu_dec(xmit_recursion);
2532 if (dev_xmit_complete(rc)) {
2533 HARD_TX_UNLOCK(dev, txq);
2537 HARD_TX_UNLOCK(dev, txq);
2538 if (net_ratelimit())
2539 printk(KERN_CRIT "Virtual device %s asks to "
2540 "queue packet!\n", dev->name);
2542 /* Recursion is detected! It is possible,
2546 if (net_ratelimit())
2547 printk(KERN_CRIT "Dead loop on virtual device "
2548 "%s, fix it urgently!\n", dev->name);
2553 rcu_read_unlock_bh();
2558 rcu_read_unlock_bh();
2561 EXPORT_SYMBOL(dev_queue_xmit);
2564 /*=======================================================================
2566 =======================================================================*/
2568 int netdev_max_backlog __read_mostly = 1000;
2569 int netdev_tstamp_prequeue __read_mostly = 1;
2570 int netdev_budget __read_mostly = 300;
2571 int weight_p __read_mostly = 64; /* old backlog weight */
2573 /* Called with irq disabled */
2574 static inline void ____napi_schedule(struct softnet_data *sd,
2575 struct napi_struct *napi)
2577 list_add_tail(&napi->poll_list, &sd->poll_list);
2578 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2582 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2583 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2584 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2585 * if hash is a canonical 4-tuple hash over transport ports.
2587 void __skb_get_rxhash(struct sk_buff *skb)
2589 int nhoff, hash = 0, poff;
2590 const struct ipv6hdr *ip6;
2591 const struct iphdr *ip;
2592 const struct vlan_hdr *vlan;
2601 nhoff = skb_network_offset(skb);
2602 proto = skb->protocol;
2606 case __constant_htons(ETH_P_IP):
2608 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2611 ip = (const struct iphdr *) (skb->data + nhoff);
2612 if (ip_is_fragment(ip))
2615 ip_proto = ip->protocol;
2616 addr1 = (__force u32) ip->saddr;
2617 addr2 = (__force u32) ip->daddr;
2618 nhoff += ip->ihl * 4;
2620 case __constant_htons(ETH_P_IPV6):
2622 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2625 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2626 ip_proto = ip6->nexthdr;
2627 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2628 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2631 case __constant_htons(ETH_P_8021Q):
2632 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2634 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2635 proto = vlan->h_vlan_encapsulated_proto;
2636 nhoff += sizeof(*vlan);
2638 case __constant_htons(ETH_P_PPP_SES):
2639 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2641 proto = *((__be16 *) (skb->data + nhoff +
2642 sizeof(struct pppoe_hdr)));
2643 nhoff += PPPOE_SES_HLEN;
2645 case __constant_htons(PPP_IP):
2647 case __constant_htons(PPP_IPV6):
2658 if (pskb_may_pull(skb, nhoff + 16)) {
2659 u8 *h = skb->data + nhoff;
2660 __be16 flags = *(__be16 *)h;
2663 * Only look inside GRE if version zero and no
2666 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2667 proto = *(__be16 *)(h + 2);
2669 if (flags & GRE_CSUM)
2671 if (flags & GRE_KEY)
2673 if (flags & GRE_SEQ)
2686 poff = proto_ports_offset(ip_proto);
2689 if (pskb_may_pull(skb, nhoff + 4)) {
2690 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2691 if (ports.v16[1] < ports.v16[0])
2692 swap(ports.v16[0], ports.v16[1]);
2697 /* get a consistent hash (same value on both flow directions) */
2701 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2708 EXPORT_SYMBOL(__skb_get_rxhash);
2712 /* One global table that all flow-based protocols share. */
2713 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2714 EXPORT_SYMBOL(rps_sock_flow_table);
2716 static struct rps_dev_flow *
2717 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2718 struct rps_dev_flow *rflow, u16 next_cpu)
2720 if (next_cpu != RPS_NO_CPU) {
2721 #ifdef CONFIG_RFS_ACCEL
2722 struct netdev_rx_queue *rxqueue;
2723 struct rps_dev_flow_table *flow_table;
2724 struct rps_dev_flow *old_rflow;
2729 /* Should we steer this flow to a different hardware queue? */
2730 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2731 !(dev->features & NETIF_F_NTUPLE))
2733 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2734 if (rxq_index == skb_get_rx_queue(skb))
2737 rxqueue = dev->_rx + rxq_index;
2738 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2741 flow_id = skb->rxhash & flow_table->mask;
2742 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2743 rxq_index, flow_id);
2747 rflow = &flow_table->flows[flow_id];
2749 if (old_rflow->filter == rflow->filter)
2750 old_rflow->filter = RPS_NO_FILTER;
2754 per_cpu(softnet_data, next_cpu).input_queue_head;
2757 rflow->cpu = next_cpu;
2762 * get_rps_cpu is called from netif_receive_skb and returns the target
2763 * CPU from the RPS map of the receiving queue for a given skb.
2764 * rcu_read_lock must be held on entry.
2766 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2767 struct rps_dev_flow **rflowp)
2769 struct netdev_rx_queue *rxqueue;
2770 struct rps_map *map;
2771 struct rps_dev_flow_table *flow_table;
2772 struct rps_sock_flow_table *sock_flow_table;
2776 if (skb_rx_queue_recorded(skb)) {
2777 u16 index = skb_get_rx_queue(skb);
2778 if (unlikely(index >= dev->real_num_rx_queues)) {
2779 WARN_ONCE(dev->real_num_rx_queues > 1,
2780 "%s received packet on queue %u, but number "
2781 "of RX queues is %u\n",
2782 dev->name, index, dev->real_num_rx_queues);
2785 rxqueue = dev->_rx + index;
2789 map = rcu_dereference(rxqueue->rps_map);
2791 if (map->len == 1 &&
2792 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2793 tcpu = map->cpus[0];
2794 if (cpu_online(tcpu))
2798 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2802 skb_reset_network_header(skb);
2803 if (!skb_get_rxhash(skb))
2806 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2807 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2808 if (flow_table && sock_flow_table) {
2810 struct rps_dev_flow *rflow;
2812 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2815 next_cpu = sock_flow_table->ents[skb->rxhash &
2816 sock_flow_table->mask];
2819 * If the desired CPU (where last recvmsg was done) is
2820 * different from current CPU (one in the rx-queue flow
2821 * table entry), switch if one of the following holds:
2822 * - Current CPU is unset (equal to RPS_NO_CPU).
2823 * - Current CPU is offline.
2824 * - The current CPU's queue tail has advanced beyond the
2825 * last packet that was enqueued using this table entry.
2826 * This guarantees that all previous packets for the flow
2827 * have been dequeued, thus preserving in order delivery.
2829 if (unlikely(tcpu != next_cpu) &&
2830 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2831 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2832 rflow->last_qtail)) >= 0))
2833 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2835 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2843 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2845 if (cpu_online(tcpu)) {
2855 #ifdef CONFIG_RFS_ACCEL
2858 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2859 * @dev: Device on which the filter was set
2860 * @rxq_index: RX queue index
2861 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2862 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2864 * Drivers that implement ndo_rx_flow_steer() should periodically call
2865 * this function for each installed filter and remove the filters for
2866 * which it returns %true.
2868 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2869 u32 flow_id, u16 filter_id)
2871 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2872 struct rps_dev_flow_table *flow_table;
2873 struct rps_dev_flow *rflow;
2878 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2879 if (flow_table && flow_id <= flow_table->mask) {
2880 rflow = &flow_table->flows[flow_id];
2881 cpu = ACCESS_ONCE(rflow->cpu);
2882 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2883 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2884 rflow->last_qtail) <
2885 (int)(10 * flow_table->mask)))
2891 EXPORT_SYMBOL(rps_may_expire_flow);
2893 #endif /* CONFIG_RFS_ACCEL */
2895 /* Called from hardirq (IPI) context */
2896 static void rps_trigger_softirq(void *data)
2898 struct softnet_data *sd = data;
2900 ____napi_schedule(sd, &sd->backlog);
2904 #endif /* CONFIG_RPS */
2907 * Check if this softnet_data structure is another cpu one
2908 * If yes, queue it to our IPI list and return 1
2911 static int rps_ipi_queued(struct softnet_data *sd)
2914 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2917 sd->rps_ipi_next = mysd->rps_ipi_list;
2918 mysd->rps_ipi_list = sd;
2920 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2923 #endif /* CONFIG_RPS */
2928 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2929 * queue (may be a remote CPU queue).
2931 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2932 unsigned int *qtail)
2934 struct softnet_data *sd;
2935 unsigned long flags;
2937 sd = &per_cpu(softnet_data, cpu);
2939 local_irq_save(flags);
2942 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2943 if (skb_queue_len(&sd->input_pkt_queue)) {
2945 __skb_queue_tail(&sd->input_pkt_queue, skb);
2946 input_queue_tail_incr_save(sd, qtail);
2948 local_irq_restore(flags);
2949 return NET_RX_SUCCESS;
2952 /* Schedule NAPI for backlog device
2953 * We can use non atomic operation since we own the queue lock
2955 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2956 if (!rps_ipi_queued(sd))
2957 ____napi_schedule(sd, &sd->backlog);
2965 local_irq_restore(flags);
2967 atomic_long_inc(&skb->dev->rx_dropped);
2973 * netif_rx - post buffer to the network code
2974 * @skb: buffer to post
2976 * This function receives a packet from a device driver and queues it for
2977 * the upper (protocol) levels to process. It always succeeds. The buffer
2978 * may be dropped during processing for congestion control or by the
2982 * NET_RX_SUCCESS (no congestion)
2983 * NET_RX_DROP (packet was dropped)
2987 int netif_rx(struct sk_buff *skb)
2991 /* if netpoll wants it, pretend we never saw it */
2992 if (netpoll_rx(skb))
2995 if (netdev_tstamp_prequeue)
2996 net_timestamp_check(skb);
2998 trace_netif_rx(skb);
3001 struct rps_dev_flow voidflow, *rflow = &voidflow;
3007 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3009 cpu = smp_processor_id();
3011 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3019 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3025 EXPORT_SYMBOL(netif_rx);
3027 int netif_rx_ni(struct sk_buff *skb)
3032 err = netif_rx(skb);
3033 if (local_softirq_pending())
3039 EXPORT_SYMBOL(netif_rx_ni);
3041 static void net_tx_action(struct softirq_action *h)
3043 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3045 if (sd->completion_queue) {
3046 struct sk_buff *clist;
3048 local_irq_disable();
3049 clist = sd->completion_queue;
3050 sd->completion_queue = NULL;
3054 struct sk_buff *skb = clist;
3055 clist = clist->next;
3057 WARN_ON(atomic_read(&skb->users));
3058 trace_kfree_skb(skb, net_tx_action);
3063 if (sd->output_queue) {
3066 local_irq_disable();
3067 head = sd->output_queue;
3068 sd->output_queue = NULL;
3069 sd->output_queue_tailp = &sd->output_queue;
3073 struct Qdisc *q = head;
3074 spinlock_t *root_lock;
3076 head = head->next_sched;
3078 root_lock = qdisc_lock(q);
3079 if (spin_trylock(root_lock)) {
3080 smp_mb__before_clear_bit();
3081 clear_bit(__QDISC_STATE_SCHED,
3084 spin_unlock(root_lock);
3086 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3088 __netif_reschedule(q);
3090 smp_mb__before_clear_bit();
3091 clear_bit(__QDISC_STATE_SCHED,
3099 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3100 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3101 /* This hook is defined here for ATM LANE */
3102 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3103 unsigned char *addr) __read_mostly;
3104 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3107 #ifdef CONFIG_NET_CLS_ACT
3108 /* TODO: Maybe we should just force sch_ingress to be compiled in
3109 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3110 * a compare and 2 stores extra right now if we dont have it on
3111 * but have CONFIG_NET_CLS_ACT
3112 * NOTE: This doesn't stop any functionality; if you dont have
3113 * the ingress scheduler, you just can't add policies on ingress.
3116 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3118 struct net_device *dev = skb->dev;
3119 u32 ttl = G_TC_RTTL(skb->tc_verd);
3120 int result = TC_ACT_OK;
3123 if (unlikely(MAX_RED_LOOP < ttl++)) {
3124 if (net_ratelimit())
3125 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3126 skb->skb_iif, dev->ifindex);
3130 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3131 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3134 if (q != &noop_qdisc) {
3135 spin_lock(qdisc_lock(q));
3136 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3137 result = qdisc_enqueue_root(skb, q);
3138 spin_unlock(qdisc_lock(q));
3144 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3145 struct packet_type **pt_prev,
3146 int *ret, struct net_device *orig_dev)
3148 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3150 if (!rxq || rxq->qdisc == &noop_qdisc)
3154 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3158 switch (ing_filter(skb, rxq)) {
3172 * netdev_rx_handler_register - register receive handler
3173 * @dev: device to register a handler for
3174 * @rx_handler: receive handler to register
3175 * @rx_handler_data: data pointer that is used by rx handler
3177 * Register a receive hander for a device. This handler will then be
3178 * called from __netif_receive_skb. A negative errno code is returned
3181 * The caller must hold the rtnl_mutex.
3183 * For a general description of rx_handler, see enum rx_handler_result.
3185 int netdev_rx_handler_register(struct net_device *dev,
3186 rx_handler_func_t *rx_handler,
3187 void *rx_handler_data)
3191 if (dev->rx_handler)
3194 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3195 rcu_assign_pointer(dev->rx_handler, rx_handler);
3199 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3202 * netdev_rx_handler_unregister - unregister receive handler
3203 * @dev: device to unregister a handler from
3205 * Unregister a receive hander from a device.
3207 * The caller must hold the rtnl_mutex.
3209 void netdev_rx_handler_unregister(struct net_device *dev)
3213 RCU_INIT_POINTER(dev->rx_handler, NULL);
3214 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3216 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3218 static int __netif_receive_skb(struct sk_buff *skb)
3220 struct packet_type *ptype, *pt_prev;
3221 rx_handler_func_t *rx_handler;
3222 struct net_device *orig_dev;
3223 struct net_device *null_or_dev;
3224 bool deliver_exact = false;
3225 int ret = NET_RX_DROP;
3228 if (!netdev_tstamp_prequeue)
3229 net_timestamp_check(skb);
3231 trace_netif_receive_skb(skb);
3233 /* if we've gotten here through NAPI, check netpoll */
3234 if (netpoll_receive_skb(skb))
3238 skb->skb_iif = skb->dev->ifindex;
3239 orig_dev = skb->dev;
3241 skb_reset_network_header(skb);
3242 skb_reset_transport_header(skb);
3243 skb_reset_mac_len(skb);
3251 __this_cpu_inc(softnet_data.processed);
3253 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3254 skb = vlan_untag(skb);
3259 #ifdef CONFIG_NET_CLS_ACT
3260 if (skb->tc_verd & TC_NCLS) {
3261 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3266 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3267 if (!ptype->dev || ptype->dev == skb->dev) {
3269 ret = deliver_skb(skb, pt_prev, orig_dev);
3274 #ifdef CONFIG_NET_CLS_ACT
3275 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3281 rx_handler = rcu_dereference(skb->dev->rx_handler);
3282 if (vlan_tx_tag_present(skb)) {
3284 ret = deliver_skb(skb, pt_prev, orig_dev);
3287 if (vlan_do_receive(&skb, !rx_handler))
3289 else if (unlikely(!skb))
3295 ret = deliver_skb(skb, pt_prev, orig_dev);
3298 switch (rx_handler(&skb)) {
3299 case RX_HANDLER_CONSUMED:
3301 case RX_HANDLER_ANOTHER:
3303 case RX_HANDLER_EXACT:
3304 deliver_exact = true;
3305 case RX_HANDLER_PASS:
3312 /* deliver only exact match when indicated */
3313 null_or_dev = deliver_exact ? skb->dev : NULL;
3315 type = skb->protocol;
3316 list_for_each_entry_rcu(ptype,
3317 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3318 if (ptype->type == type &&
3319 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3320 ptype->dev == orig_dev)) {
3322 ret = deliver_skb(skb, pt_prev, orig_dev);
3328 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3330 atomic_long_inc(&skb->dev->rx_dropped);
3332 /* Jamal, now you will not able to escape explaining
3333 * me how you were going to use this. :-)
3344 * netif_receive_skb - process receive buffer from network
3345 * @skb: buffer to process
3347 * netif_receive_skb() is the main receive data processing function.
3348 * It always succeeds. The buffer may be dropped during processing
3349 * for congestion control or by the protocol layers.
3351 * This function may only be called from softirq context and interrupts
3352 * should be enabled.
3354 * Return values (usually ignored):
3355 * NET_RX_SUCCESS: no congestion
3356 * NET_RX_DROP: packet was dropped
3358 int netif_receive_skb(struct sk_buff *skb)
3360 if (netdev_tstamp_prequeue)
3361 net_timestamp_check(skb);
3363 if (skb_defer_rx_timestamp(skb))
3364 return NET_RX_SUCCESS;
3368 struct rps_dev_flow voidflow, *rflow = &voidflow;
3373 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3376 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3380 ret = __netif_receive_skb(skb);
3386 return __netif_receive_skb(skb);
3389 EXPORT_SYMBOL(netif_receive_skb);
3391 /* Network device is going away, flush any packets still pending
3392 * Called with irqs disabled.
3394 static void flush_backlog(void *arg)
3396 struct net_device *dev = arg;
3397 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3398 struct sk_buff *skb, *tmp;
3401 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3402 if (skb->dev == dev) {
3403 __skb_unlink(skb, &sd->input_pkt_queue);
3405 input_queue_head_incr(sd);
3410 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3411 if (skb->dev == dev) {
3412 __skb_unlink(skb, &sd->process_queue);
3414 input_queue_head_incr(sd);
3419 static int napi_gro_complete(struct sk_buff *skb)
3421 struct packet_type *ptype;
3422 __be16 type = skb->protocol;
3423 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3426 if (NAPI_GRO_CB(skb)->count == 1) {
3427 skb_shinfo(skb)->gso_size = 0;
3432 list_for_each_entry_rcu(ptype, head, list) {
3433 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3436 err = ptype->gro_complete(skb);
3442 WARN_ON(&ptype->list == head);
3444 return NET_RX_SUCCESS;
3448 return netif_receive_skb(skb);
3451 inline void napi_gro_flush(struct napi_struct *napi)
3453 struct sk_buff *skb, *next;
3455 for (skb = napi->gro_list; skb; skb = next) {
3458 napi_gro_complete(skb);
3461 napi->gro_count = 0;
3462 napi->gro_list = NULL;
3464 EXPORT_SYMBOL(napi_gro_flush);
3466 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3468 struct sk_buff **pp = NULL;
3469 struct packet_type *ptype;
3470 __be16 type = skb->protocol;
3471 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3474 enum gro_result ret;
3476 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3479 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3483 list_for_each_entry_rcu(ptype, head, list) {
3484 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3487 skb_set_network_header(skb, skb_gro_offset(skb));
3488 mac_len = skb->network_header - skb->mac_header;
3489 skb->mac_len = mac_len;
3490 NAPI_GRO_CB(skb)->same_flow = 0;
3491 NAPI_GRO_CB(skb)->flush = 0;
3492 NAPI_GRO_CB(skb)->free = 0;
3494 pp = ptype->gro_receive(&napi->gro_list, skb);
3499 if (&ptype->list == head)
3502 same_flow = NAPI_GRO_CB(skb)->same_flow;
3503 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3506 struct sk_buff *nskb = *pp;
3510 napi_gro_complete(nskb);
3517 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3521 NAPI_GRO_CB(skb)->count = 1;
3522 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3523 skb->next = napi->gro_list;
3524 napi->gro_list = skb;
3528 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3529 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3531 BUG_ON(skb->end - skb->tail < grow);
3533 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3536 skb->data_len -= grow;
3538 skb_shinfo(skb)->frags[0].page_offset += grow;
3539 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3541 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3542 skb_frag_unref(skb, 0);
3543 memmove(skb_shinfo(skb)->frags,
3544 skb_shinfo(skb)->frags + 1,
3545 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3556 EXPORT_SYMBOL(dev_gro_receive);
3558 static inline gro_result_t
3559 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3562 unsigned int maclen = skb->dev->hard_header_len;
3564 for (p = napi->gro_list; p; p = p->next) {
3565 unsigned long diffs;
3567 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3568 diffs |= p->vlan_tci ^ skb->vlan_tci;
3569 if (maclen == ETH_HLEN)
3570 diffs |= compare_ether_header(skb_mac_header(p),
3571 skb_gro_mac_header(skb));
3573 diffs = memcmp(skb_mac_header(p),
3574 skb_gro_mac_header(skb),
3576 NAPI_GRO_CB(p)->same_flow = !diffs;
3577 NAPI_GRO_CB(p)->flush = 0;
3580 return dev_gro_receive(napi, skb);
3583 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3587 if (netif_receive_skb(skb))
3592 case GRO_MERGED_FREE:
3603 EXPORT_SYMBOL(napi_skb_finish);
3605 void skb_gro_reset_offset(struct sk_buff *skb)
3607 NAPI_GRO_CB(skb)->data_offset = 0;
3608 NAPI_GRO_CB(skb)->frag0 = NULL;
3609 NAPI_GRO_CB(skb)->frag0_len = 0;
3611 if (skb->mac_header == skb->tail &&
3612 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3613 NAPI_GRO_CB(skb)->frag0 =
3614 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3615 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3618 EXPORT_SYMBOL(skb_gro_reset_offset);
3620 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3622 skb_gro_reset_offset(skb);
3624 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3626 EXPORT_SYMBOL(napi_gro_receive);
3628 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3630 __skb_pull(skb, skb_headlen(skb));
3631 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3632 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3634 skb->dev = napi->dev;
3640 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3642 struct sk_buff *skb = napi->skb;
3645 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3651 EXPORT_SYMBOL(napi_get_frags);
3653 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3659 skb->protocol = eth_type_trans(skb, skb->dev);
3661 if (ret == GRO_HELD)
3662 skb_gro_pull(skb, -ETH_HLEN);
3663 else if (netif_receive_skb(skb))
3668 case GRO_MERGED_FREE:
3669 napi_reuse_skb(napi, skb);
3678 EXPORT_SYMBOL(napi_frags_finish);
3680 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3682 struct sk_buff *skb = napi->skb;
3689 skb_reset_mac_header(skb);
3690 skb_gro_reset_offset(skb);
3692 off = skb_gro_offset(skb);
3693 hlen = off + sizeof(*eth);
3694 eth = skb_gro_header_fast(skb, off);
3695 if (skb_gro_header_hard(skb, hlen)) {
3696 eth = skb_gro_header_slow(skb, hlen, off);
3697 if (unlikely(!eth)) {
3698 napi_reuse_skb(napi, skb);
3704 skb_gro_pull(skb, sizeof(*eth));
3707 * This works because the only protocols we care about don't require
3708 * special handling. We'll fix it up properly at the end.
3710 skb->protocol = eth->h_proto;
3715 EXPORT_SYMBOL(napi_frags_skb);
3717 gro_result_t napi_gro_frags(struct napi_struct *napi)
3719 struct sk_buff *skb = napi_frags_skb(napi);
3724 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3726 EXPORT_SYMBOL(napi_gro_frags);
3729 * net_rps_action sends any pending IPI's for rps.
3730 * Note: called with local irq disabled, but exits with local irq enabled.
3732 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3735 struct softnet_data *remsd = sd->rps_ipi_list;
3738 sd->rps_ipi_list = NULL;
3742 /* Send pending IPI's to kick RPS processing on remote cpus. */
3744 struct softnet_data *next = remsd->rps_ipi_next;
3746 if (cpu_online(remsd->cpu))
3747 __smp_call_function_single(remsd->cpu,
3756 static int process_backlog(struct napi_struct *napi, int quota)
3759 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3762 /* Check if we have pending ipi, its better to send them now,
3763 * not waiting net_rx_action() end.
3765 if (sd->rps_ipi_list) {
3766 local_irq_disable();
3767 net_rps_action_and_irq_enable(sd);
3770 napi->weight = weight_p;
3771 local_irq_disable();
3772 while (work < quota) {
3773 struct sk_buff *skb;
3776 while ((skb = __skb_dequeue(&sd->process_queue))) {
3778 __netif_receive_skb(skb);
3779 local_irq_disable();
3780 input_queue_head_incr(sd);
3781 if (++work >= quota) {
3788 qlen = skb_queue_len(&sd->input_pkt_queue);
3790 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3791 &sd->process_queue);
3793 if (qlen < quota - work) {
3795 * Inline a custom version of __napi_complete().
3796 * only current cpu owns and manipulates this napi,
3797 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3798 * we can use a plain write instead of clear_bit(),
3799 * and we dont need an smp_mb() memory barrier.
3801 list_del(&napi->poll_list);
3804 quota = work + qlen;
3814 * __napi_schedule - schedule for receive
3815 * @n: entry to schedule
3817 * The entry's receive function will be scheduled to run
3819 void __napi_schedule(struct napi_struct *n)
3821 unsigned long flags;
3823 local_irq_save(flags);
3824 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3825 local_irq_restore(flags);
3827 EXPORT_SYMBOL(__napi_schedule);
3829 void __napi_complete(struct napi_struct *n)
3831 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3832 BUG_ON(n->gro_list);
3834 list_del(&n->poll_list);
3835 smp_mb__before_clear_bit();
3836 clear_bit(NAPI_STATE_SCHED, &n->state);
3838 EXPORT_SYMBOL(__napi_complete);
3840 void napi_complete(struct napi_struct *n)
3842 unsigned long flags;
3845 * don't let napi dequeue from the cpu poll list
3846 * just in case its running on a different cpu
3848 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3852 local_irq_save(flags);
3854 local_irq_restore(flags);
3856 EXPORT_SYMBOL(napi_complete);
3858 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3859 int (*poll)(struct napi_struct *, int), int weight)
3861 INIT_LIST_HEAD(&napi->poll_list);
3862 napi->gro_count = 0;
3863 napi->gro_list = NULL;
3866 napi->weight = weight;
3867 list_add(&napi->dev_list, &dev->napi_list);
3869 #ifdef CONFIG_NETPOLL
3870 spin_lock_init(&napi->poll_lock);
3871 napi->poll_owner = -1;
3873 set_bit(NAPI_STATE_SCHED, &napi->state);
3875 EXPORT_SYMBOL(netif_napi_add);
3877 void netif_napi_del(struct napi_struct *napi)
3879 struct sk_buff *skb, *next;
3881 list_del_init(&napi->dev_list);
3882 napi_free_frags(napi);
3884 for (skb = napi->gro_list; skb; skb = next) {
3890 napi->gro_list = NULL;
3891 napi->gro_count = 0;
3893 EXPORT_SYMBOL(netif_napi_del);
3895 static void net_rx_action(struct softirq_action *h)
3897 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3898 unsigned long time_limit = jiffies + 2;
3899 int budget = netdev_budget;
3902 local_irq_disable();
3904 while (!list_empty(&sd->poll_list)) {
3905 struct napi_struct *n;
3908 /* If softirq window is exhuasted then punt.
3909 * Allow this to run for 2 jiffies since which will allow
3910 * an average latency of 1.5/HZ.
3912 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3917 /* Even though interrupts have been re-enabled, this
3918 * access is safe because interrupts can only add new
3919 * entries to the tail of this list, and only ->poll()
3920 * calls can remove this head entry from the list.
3922 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3924 have = netpoll_poll_lock(n);
3928 /* This NAPI_STATE_SCHED test is for avoiding a race
3929 * with netpoll's poll_napi(). Only the entity which
3930 * obtains the lock and sees NAPI_STATE_SCHED set will
3931 * actually make the ->poll() call. Therefore we avoid
3932 * accidentally calling ->poll() when NAPI is not scheduled.
3935 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3936 work = n->poll(n, weight);
3940 WARN_ON_ONCE(work > weight);
3944 local_irq_disable();
3946 /* Drivers must not modify the NAPI state if they
3947 * consume the entire weight. In such cases this code
3948 * still "owns" the NAPI instance and therefore can
3949 * move the instance around on the list at-will.
3951 if (unlikely(work == weight)) {
3952 if (unlikely(napi_disable_pending(n))) {
3955 local_irq_disable();
3957 list_move_tail(&n->poll_list, &sd->poll_list);
3960 netpoll_poll_unlock(have);
3963 net_rps_action_and_irq_enable(sd);
3965 #ifdef CONFIG_NET_DMA
3967 * There may not be any more sk_buffs coming right now, so push
3968 * any pending DMA copies to hardware
3970 dma_issue_pending_all();
3977 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3981 static gifconf_func_t *gifconf_list[NPROTO];
3984 * register_gifconf - register a SIOCGIF handler
3985 * @family: Address family
3986 * @gifconf: Function handler
3988 * Register protocol dependent address dumping routines. The handler
3989 * that is passed must not be freed or reused until it has been replaced
3990 * by another handler.
3992 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3994 if (family >= NPROTO)
3996 gifconf_list[family] = gifconf;
3999 EXPORT_SYMBOL(register_gifconf);
4003 * Map an interface index to its name (SIOCGIFNAME)
4007 * We need this ioctl for efficient implementation of the
4008 * if_indextoname() function required by the IPv6 API. Without
4009 * it, we would have to search all the interfaces to find a
4013 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4015 struct net_device *dev;
4019 * Fetch the caller's info block.
4022 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4026 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4032 strcpy(ifr.ifr_name, dev->name);
4035 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4041 * Perform a SIOCGIFCONF call. This structure will change
4042 * size eventually, and there is nothing I can do about it.
4043 * Thus we will need a 'compatibility mode'.
4046 static int dev_ifconf(struct net *net, char __user *arg)
4049 struct net_device *dev;
4056 * Fetch the caller's info block.
4059 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4066 * Loop over the interfaces, and write an info block for each.
4070 for_each_netdev(net, dev) {
4071 for (i = 0; i < NPROTO; i++) {
4072 if (gifconf_list[i]) {
4075 done = gifconf_list[i](dev, NULL, 0);
4077 done = gifconf_list[i](dev, pos + total,
4087 * All done. Write the updated control block back to the caller.
4089 ifc.ifc_len = total;
4092 * Both BSD and Solaris return 0 here, so we do too.
4094 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4097 #ifdef CONFIG_PROC_FS
4099 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4101 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4102 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4103 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4105 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4107 struct net *net = seq_file_net(seq);
4108 struct net_device *dev;
4109 struct hlist_node *p;
4110 struct hlist_head *h;
4111 unsigned int count = 0, offset = get_offset(*pos);
4113 h = &net->dev_name_head[get_bucket(*pos)];
4114 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4115 if (++count == offset)
4122 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4124 struct net_device *dev;
4125 unsigned int bucket;
4128 dev = dev_from_same_bucket(seq, pos);
4132 bucket = get_bucket(*pos) + 1;
4133 *pos = set_bucket_offset(bucket, 1);
4134 } while (bucket < NETDEV_HASHENTRIES);
4140 * This is invoked by the /proc filesystem handler to display a device
4143 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4148 return SEQ_START_TOKEN;
4150 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4153 return dev_from_bucket(seq, pos);
4156 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4159 return dev_from_bucket(seq, pos);
4162 void dev_seq_stop(struct seq_file *seq, void *v)
4168 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4170 struct rtnl_link_stats64 temp;
4171 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4173 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4174 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4175 dev->name, stats->rx_bytes, stats->rx_packets,
4177 stats->rx_dropped + stats->rx_missed_errors,
4178 stats->rx_fifo_errors,
4179 stats->rx_length_errors + stats->rx_over_errors +
4180 stats->rx_crc_errors + stats->rx_frame_errors,
4181 stats->rx_compressed, stats->multicast,
4182 stats->tx_bytes, stats->tx_packets,
4183 stats->tx_errors, stats->tx_dropped,
4184 stats->tx_fifo_errors, stats->collisions,
4185 stats->tx_carrier_errors +
4186 stats->tx_aborted_errors +
4187 stats->tx_window_errors +
4188 stats->tx_heartbeat_errors,
4189 stats->tx_compressed);
4193 * Called from the PROCfs module. This now uses the new arbitrary sized
4194 * /proc/net interface to create /proc/net/dev
4196 static int dev_seq_show(struct seq_file *seq, void *v)
4198 if (v == SEQ_START_TOKEN)
4199 seq_puts(seq, "Inter-| Receive "
4201 " face |bytes packets errs drop fifo frame "
4202 "compressed multicast|bytes packets errs "
4203 "drop fifo colls carrier compressed\n");
4205 dev_seq_printf_stats(seq, v);
4209 static struct softnet_data *softnet_get_online(loff_t *pos)
4211 struct softnet_data *sd = NULL;
4213 while (*pos < nr_cpu_ids)
4214 if (cpu_online(*pos)) {
4215 sd = &per_cpu(softnet_data, *pos);
4222 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4224 return softnet_get_online(pos);
4227 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4230 return softnet_get_online(pos);
4233 static void softnet_seq_stop(struct seq_file *seq, void *v)
4237 static int softnet_seq_show(struct seq_file *seq, void *v)
4239 struct softnet_data *sd = v;
4241 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4242 sd->processed, sd->dropped, sd->time_squeeze, 0,
4243 0, 0, 0, 0, /* was fastroute */
4244 sd->cpu_collision, sd->received_rps);
4248 static const struct seq_operations dev_seq_ops = {
4249 .start = dev_seq_start,
4250 .next = dev_seq_next,
4251 .stop = dev_seq_stop,
4252 .show = dev_seq_show,
4255 static int dev_seq_open(struct inode *inode, struct file *file)
4257 return seq_open_net(inode, file, &dev_seq_ops,
4258 sizeof(struct seq_net_private));
4261 static const struct file_operations dev_seq_fops = {
4262 .owner = THIS_MODULE,
4263 .open = dev_seq_open,
4265 .llseek = seq_lseek,
4266 .release = seq_release_net,
4269 static const struct seq_operations softnet_seq_ops = {
4270 .start = softnet_seq_start,
4271 .next = softnet_seq_next,
4272 .stop = softnet_seq_stop,
4273 .show = softnet_seq_show,
4276 static int softnet_seq_open(struct inode *inode, struct file *file)
4278 return seq_open(file, &softnet_seq_ops);
4281 static const struct file_operations softnet_seq_fops = {
4282 .owner = THIS_MODULE,
4283 .open = softnet_seq_open,
4285 .llseek = seq_lseek,
4286 .release = seq_release,
4289 static void *ptype_get_idx(loff_t pos)
4291 struct packet_type *pt = NULL;
4295 list_for_each_entry_rcu(pt, &ptype_all, list) {
4301 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4302 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4311 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4315 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4318 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4320 struct packet_type *pt;
4321 struct list_head *nxt;
4325 if (v == SEQ_START_TOKEN)
4326 return ptype_get_idx(0);
4329 nxt = pt->list.next;
4330 if (pt->type == htons(ETH_P_ALL)) {
4331 if (nxt != &ptype_all)
4334 nxt = ptype_base[0].next;
4336 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4338 while (nxt == &ptype_base[hash]) {
4339 if (++hash >= PTYPE_HASH_SIZE)
4341 nxt = ptype_base[hash].next;
4344 return list_entry(nxt, struct packet_type, list);
4347 static void ptype_seq_stop(struct seq_file *seq, void *v)
4353 static int ptype_seq_show(struct seq_file *seq, void *v)
4355 struct packet_type *pt = v;
4357 if (v == SEQ_START_TOKEN)
4358 seq_puts(seq, "Type Device Function\n");
4359 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4360 if (pt->type == htons(ETH_P_ALL))
4361 seq_puts(seq, "ALL ");
4363 seq_printf(seq, "%04x", ntohs(pt->type));
4365 seq_printf(seq, " %-8s %pF\n",
4366 pt->dev ? pt->dev->name : "", pt->func);
4372 static const struct seq_operations ptype_seq_ops = {
4373 .start = ptype_seq_start,
4374 .next = ptype_seq_next,
4375 .stop = ptype_seq_stop,
4376 .show = ptype_seq_show,
4379 static int ptype_seq_open(struct inode *inode, struct file *file)
4381 return seq_open_net(inode, file, &ptype_seq_ops,
4382 sizeof(struct seq_net_private));
4385 static const struct file_operations ptype_seq_fops = {
4386 .owner = THIS_MODULE,
4387 .open = ptype_seq_open,
4389 .llseek = seq_lseek,
4390 .release = seq_release_net,
4394 static int __net_init dev_proc_net_init(struct net *net)
4398 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4400 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4402 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4405 if (wext_proc_init(net))
4411 proc_net_remove(net, "ptype");
4413 proc_net_remove(net, "softnet_stat");
4415 proc_net_remove(net, "dev");
4419 static void __net_exit dev_proc_net_exit(struct net *net)
4421 wext_proc_exit(net);
4423 proc_net_remove(net, "ptype");
4424 proc_net_remove(net, "softnet_stat");
4425 proc_net_remove(net, "dev");
4428 static struct pernet_operations __net_initdata dev_proc_ops = {
4429 .init = dev_proc_net_init,
4430 .exit = dev_proc_net_exit,
4433 static int __init dev_proc_init(void)
4435 return register_pernet_subsys(&dev_proc_ops);
4438 #define dev_proc_init() 0
4439 #endif /* CONFIG_PROC_FS */
4443 * netdev_set_master - set up master pointer
4444 * @slave: slave device
4445 * @master: new master device
4447 * Changes the master device of the slave. Pass %NULL to break the
4448 * bonding. The caller must hold the RTNL semaphore. On a failure
4449 * a negative errno code is returned. On success the reference counts
4450 * are adjusted and the function returns zero.
4452 int netdev_set_master(struct net_device *slave, struct net_device *master)
4454 struct net_device *old = slave->master;
4464 slave->master = master;
4470 EXPORT_SYMBOL(netdev_set_master);
4473 * netdev_set_bond_master - set up bonding master/slave pair
4474 * @slave: slave device
4475 * @master: new master device
4477 * Changes the master device of the slave. Pass %NULL to break the
4478 * bonding. The caller must hold the RTNL semaphore. On a failure
4479 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4480 * to the routing socket and the function returns zero.
4482 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4488 err = netdev_set_master(slave, master);
4492 slave->flags |= IFF_SLAVE;
4494 slave->flags &= ~IFF_SLAVE;
4496 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4499 EXPORT_SYMBOL(netdev_set_bond_master);
4501 static void dev_change_rx_flags(struct net_device *dev, int flags)
4503 const struct net_device_ops *ops = dev->netdev_ops;
4505 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4506 ops->ndo_change_rx_flags(dev, flags);
4509 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4511 unsigned short old_flags = dev->flags;
4517 dev->flags |= IFF_PROMISC;
4518 dev->promiscuity += inc;
4519 if (dev->promiscuity == 0) {
4522 * If inc causes overflow, untouch promisc and return error.
4525 dev->flags &= ~IFF_PROMISC;
4527 dev->promiscuity -= inc;
4528 printk(KERN_WARNING "%s: promiscuity touches roof, "
4529 "set promiscuity failed, promiscuity feature "
4530 "of device might be broken.\n", dev->name);
4534 if (dev->flags != old_flags) {
4535 printk(KERN_INFO "device %s %s promiscuous mode\n",
4536 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4538 if (audit_enabled) {
4539 current_uid_gid(&uid, &gid);
4540 audit_log(current->audit_context, GFP_ATOMIC,
4541 AUDIT_ANOM_PROMISCUOUS,
4542 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4543 dev->name, (dev->flags & IFF_PROMISC),
4544 (old_flags & IFF_PROMISC),
4545 audit_get_loginuid(current),
4547 audit_get_sessionid(current));
4550 dev_change_rx_flags(dev, IFF_PROMISC);
4556 * dev_set_promiscuity - update promiscuity count on a device
4560 * Add or remove promiscuity from a device. While the count in the device
4561 * remains above zero the interface remains promiscuous. Once it hits zero
4562 * the device reverts back to normal filtering operation. A negative inc
4563 * value is used to drop promiscuity on the device.
4564 * Return 0 if successful or a negative errno code on error.
4566 int dev_set_promiscuity(struct net_device *dev, int inc)
4568 unsigned short old_flags = dev->flags;
4571 err = __dev_set_promiscuity(dev, inc);
4574 if (dev->flags != old_flags)
4575 dev_set_rx_mode(dev);
4578 EXPORT_SYMBOL(dev_set_promiscuity);
4581 * dev_set_allmulti - update allmulti count on a device
4585 * Add or remove reception of all multicast frames to a device. While the
4586 * count in the device remains above zero the interface remains listening
4587 * to all interfaces. Once it hits zero the device reverts back to normal
4588 * filtering operation. A negative @inc value is used to drop the counter
4589 * when releasing a resource needing all multicasts.
4590 * Return 0 if successful or a negative errno code on error.
4593 int dev_set_allmulti(struct net_device *dev, int inc)
4595 unsigned short old_flags = dev->flags;
4599 dev->flags |= IFF_ALLMULTI;
4600 dev->allmulti += inc;
4601 if (dev->allmulti == 0) {
4604 * If inc causes overflow, untouch allmulti and return error.
4607 dev->flags &= ~IFF_ALLMULTI;
4609 dev->allmulti -= inc;
4610 printk(KERN_WARNING "%s: allmulti touches roof, "
4611 "set allmulti failed, allmulti feature of "
4612 "device might be broken.\n", dev->name);
4616 if (dev->flags ^ old_flags) {
4617 dev_change_rx_flags(dev, IFF_ALLMULTI);
4618 dev_set_rx_mode(dev);
4622 EXPORT_SYMBOL(dev_set_allmulti);
4625 * Upload unicast and multicast address lists to device and
4626 * configure RX filtering. When the device doesn't support unicast
4627 * filtering it is put in promiscuous mode while unicast addresses
4630 void __dev_set_rx_mode(struct net_device *dev)
4632 const struct net_device_ops *ops = dev->netdev_ops;
4634 /* dev_open will call this function so the list will stay sane. */
4635 if (!(dev->flags&IFF_UP))
4638 if (!netif_device_present(dev))
4641 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4642 /* Unicast addresses changes may only happen under the rtnl,
4643 * therefore calling __dev_set_promiscuity here is safe.
4645 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4646 __dev_set_promiscuity(dev, 1);
4647 dev->uc_promisc = true;
4648 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4649 __dev_set_promiscuity(dev, -1);
4650 dev->uc_promisc = false;
4654 if (ops->ndo_set_rx_mode)
4655 ops->ndo_set_rx_mode(dev);
4658 void dev_set_rx_mode(struct net_device *dev)
4660 netif_addr_lock_bh(dev);
4661 __dev_set_rx_mode(dev);
4662 netif_addr_unlock_bh(dev);
4666 * dev_get_flags - get flags reported to userspace
4669 * Get the combination of flag bits exported through APIs to userspace.
4671 unsigned dev_get_flags(const struct net_device *dev)
4675 flags = (dev->flags & ~(IFF_PROMISC |
4680 (dev->gflags & (IFF_PROMISC |
4683 if (netif_running(dev)) {
4684 if (netif_oper_up(dev))
4685 flags |= IFF_RUNNING;
4686 if (netif_carrier_ok(dev))
4687 flags |= IFF_LOWER_UP;
4688 if (netif_dormant(dev))
4689 flags |= IFF_DORMANT;
4694 EXPORT_SYMBOL(dev_get_flags);
4696 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4698 int old_flags = dev->flags;
4704 * Set the flags on our device.
4707 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4708 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4710 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4714 * Load in the correct multicast list now the flags have changed.
4717 if ((old_flags ^ flags) & IFF_MULTICAST)
4718 dev_change_rx_flags(dev, IFF_MULTICAST);
4720 dev_set_rx_mode(dev);
4723 * Have we downed the interface. We handle IFF_UP ourselves
4724 * according to user attempts to set it, rather than blindly
4729 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4730 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4733 dev_set_rx_mode(dev);
4736 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4737 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4739 dev->gflags ^= IFF_PROMISC;
4740 dev_set_promiscuity(dev, inc);
4743 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4744 is important. Some (broken) drivers set IFF_PROMISC, when
4745 IFF_ALLMULTI is requested not asking us and not reporting.
4747 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4748 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4750 dev->gflags ^= IFF_ALLMULTI;
4751 dev_set_allmulti(dev, inc);
4757 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4759 unsigned int changes = dev->flags ^ old_flags;
4761 if (changes & IFF_UP) {
4762 if (dev->flags & IFF_UP)
4763 call_netdevice_notifiers(NETDEV_UP, dev);
4765 call_netdevice_notifiers(NETDEV_DOWN, dev);
4768 if (dev->flags & IFF_UP &&
4769 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4770 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4774 * dev_change_flags - change device settings
4776 * @flags: device state flags
4778 * Change settings on device based state flags. The flags are
4779 * in the userspace exported format.
4781 int dev_change_flags(struct net_device *dev, unsigned flags)
4784 int old_flags = dev->flags;
4786 ret = __dev_change_flags(dev, flags);
4790 changes = old_flags ^ dev->flags;
4792 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4794 __dev_notify_flags(dev, old_flags);
4797 EXPORT_SYMBOL(dev_change_flags);
4800 * dev_set_mtu - Change maximum transfer unit
4802 * @new_mtu: new transfer unit
4804 * Change the maximum transfer size of the network device.
4806 int dev_set_mtu(struct net_device *dev, int new_mtu)
4808 const struct net_device_ops *ops = dev->netdev_ops;
4811 if (new_mtu == dev->mtu)
4814 /* MTU must be positive. */
4818 if (!netif_device_present(dev))
4822 if (ops->ndo_change_mtu)
4823 err = ops->ndo_change_mtu(dev, new_mtu);
4827 if (!err && dev->flags & IFF_UP)
4828 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4831 EXPORT_SYMBOL(dev_set_mtu);
4834 * dev_set_group - Change group this device belongs to
4836 * @new_group: group this device should belong to
4838 void dev_set_group(struct net_device *dev, int new_group)
4840 dev->group = new_group;
4842 EXPORT_SYMBOL(dev_set_group);
4845 * dev_set_mac_address - Change Media Access Control Address
4849 * Change the hardware (MAC) address of the device
4851 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4853 const struct net_device_ops *ops = dev->netdev_ops;
4856 if (!ops->ndo_set_mac_address)
4858 if (sa->sa_family != dev->type)
4860 if (!netif_device_present(dev))
4862 err = ops->ndo_set_mac_address(dev, sa);
4864 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4867 EXPORT_SYMBOL(dev_set_mac_address);
4870 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4872 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4875 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4881 case SIOCGIFFLAGS: /* Get interface flags */
4882 ifr->ifr_flags = (short) dev_get_flags(dev);
4885 case SIOCGIFMETRIC: /* Get the metric on the interface
4886 (currently unused) */
4887 ifr->ifr_metric = 0;
4890 case SIOCGIFMTU: /* Get the MTU of a device */
4891 ifr->ifr_mtu = dev->mtu;
4896 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4898 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4899 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4900 ifr->ifr_hwaddr.sa_family = dev->type;
4908 ifr->ifr_map.mem_start = dev->mem_start;
4909 ifr->ifr_map.mem_end = dev->mem_end;
4910 ifr->ifr_map.base_addr = dev->base_addr;
4911 ifr->ifr_map.irq = dev->irq;
4912 ifr->ifr_map.dma = dev->dma;
4913 ifr->ifr_map.port = dev->if_port;
4917 ifr->ifr_ifindex = dev->ifindex;
4921 ifr->ifr_qlen = dev->tx_queue_len;
4925 /* dev_ioctl() should ensure this case
4937 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4939 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4942 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4943 const struct net_device_ops *ops;
4948 ops = dev->netdev_ops;
4951 case SIOCSIFFLAGS: /* Set interface flags */
4952 return dev_change_flags(dev, ifr->ifr_flags);
4954 case SIOCSIFMETRIC: /* Set the metric on the interface
4955 (currently unused) */
4958 case SIOCSIFMTU: /* Set the MTU of a device */
4959 return dev_set_mtu(dev, ifr->ifr_mtu);
4962 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4964 case SIOCSIFHWBROADCAST:
4965 if (ifr->ifr_hwaddr.sa_family != dev->type)
4967 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4968 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4969 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4973 if (ops->ndo_set_config) {
4974 if (!netif_device_present(dev))
4976 return ops->ndo_set_config(dev, &ifr->ifr_map);
4981 if (!ops->ndo_set_rx_mode ||
4982 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4984 if (!netif_device_present(dev))
4986 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4989 if (!ops->ndo_set_rx_mode ||
4990 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4992 if (!netif_device_present(dev))
4994 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4997 if (ifr->ifr_qlen < 0)
4999 dev->tx_queue_len = ifr->ifr_qlen;
5003 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5004 return dev_change_name(dev, ifr->ifr_newname);
5007 err = net_hwtstamp_validate(ifr);
5013 * Unknown or private ioctl
5016 if ((cmd >= SIOCDEVPRIVATE &&
5017 cmd <= SIOCDEVPRIVATE + 15) ||
5018 cmd == SIOCBONDENSLAVE ||
5019 cmd == SIOCBONDRELEASE ||
5020 cmd == SIOCBONDSETHWADDR ||
5021 cmd == SIOCBONDSLAVEINFOQUERY ||
5022 cmd == SIOCBONDINFOQUERY ||
5023 cmd == SIOCBONDCHANGEACTIVE ||
5024 cmd == SIOCGMIIPHY ||
5025 cmd == SIOCGMIIREG ||
5026 cmd == SIOCSMIIREG ||
5027 cmd == SIOCBRADDIF ||
5028 cmd == SIOCBRDELIF ||
5029 cmd == SIOCSHWTSTAMP ||
5030 cmd == SIOCWANDEV) {
5032 if (ops->ndo_do_ioctl) {
5033 if (netif_device_present(dev))
5034 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5046 * This function handles all "interface"-type I/O control requests. The actual
5047 * 'doing' part of this is dev_ifsioc above.
5051 * dev_ioctl - network device ioctl
5052 * @net: the applicable net namespace
5053 * @cmd: command to issue
5054 * @arg: pointer to a struct ifreq in user space
5056 * Issue ioctl functions to devices. This is normally called by the
5057 * user space syscall interfaces but can sometimes be useful for
5058 * other purposes. The return value is the return from the syscall if
5059 * positive or a negative errno code on error.
5062 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5068 /* One special case: SIOCGIFCONF takes ifconf argument
5069 and requires shared lock, because it sleeps writing
5073 if (cmd == SIOCGIFCONF) {
5075 ret = dev_ifconf(net, (char __user *) arg);
5079 if (cmd == SIOCGIFNAME)
5080 return dev_ifname(net, (struct ifreq __user *)arg);
5082 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5085 ifr.ifr_name[IFNAMSIZ-1] = 0;
5087 colon = strchr(ifr.ifr_name, ':');
5092 * See which interface the caller is talking about.
5097 * These ioctl calls:
5098 * - can be done by all.
5099 * - atomic and do not require locking.
5110 dev_load(net, ifr.ifr_name);
5112 ret = dev_ifsioc_locked(net, &ifr, cmd);
5117 if (copy_to_user(arg, &ifr,
5118 sizeof(struct ifreq)))
5124 dev_load(net, ifr.ifr_name);
5126 ret = dev_ethtool(net, &ifr);
5131 if (copy_to_user(arg, &ifr,
5132 sizeof(struct ifreq)))
5138 * These ioctl calls:
5139 * - require superuser power.
5140 * - require strict serialization.
5146 if (!capable(CAP_NET_ADMIN))
5148 dev_load(net, ifr.ifr_name);
5150 ret = dev_ifsioc(net, &ifr, cmd);
5155 if (copy_to_user(arg, &ifr,
5156 sizeof(struct ifreq)))
5162 * These ioctl calls:
5163 * - require superuser power.
5164 * - require strict serialization.
5165 * - do not return a value
5175 case SIOCSIFHWBROADCAST:
5178 case SIOCBONDENSLAVE:
5179 case SIOCBONDRELEASE:
5180 case SIOCBONDSETHWADDR:
5181 case SIOCBONDCHANGEACTIVE:
5185 if (!capable(CAP_NET_ADMIN))
5188 case SIOCBONDSLAVEINFOQUERY:
5189 case SIOCBONDINFOQUERY:
5190 dev_load(net, ifr.ifr_name);
5192 ret = dev_ifsioc(net, &ifr, cmd);
5197 /* Get the per device memory space. We can add this but
5198 * currently do not support it */
5200 /* Set the per device memory buffer space.
5201 * Not applicable in our case */
5206 * Unknown or private ioctl.
5209 if (cmd == SIOCWANDEV ||
5210 (cmd >= SIOCDEVPRIVATE &&
5211 cmd <= SIOCDEVPRIVATE + 15)) {
5212 dev_load(net, ifr.ifr_name);
5214 ret = dev_ifsioc(net, &ifr, cmd);
5216 if (!ret && copy_to_user(arg, &ifr,
5217 sizeof(struct ifreq)))
5221 /* Take care of Wireless Extensions */
5222 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5223 return wext_handle_ioctl(net, &ifr, cmd, arg);
5230 * dev_new_index - allocate an ifindex
5231 * @net: the applicable net namespace
5233 * Returns a suitable unique value for a new device interface
5234 * number. The caller must hold the rtnl semaphore or the
5235 * dev_base_lock to be sure it remains unique.
5237 static int dev_new_index(struct net *net)
5243 if (!__dev_get_by_index(net, ifindex))
5248 /* Delayed registration/unregisteration */
5249 static LIST_HEAD(net_todo_list);
5251 static void net_set_todo(struct net_device *dev)
5253 list_add_tail(&dev->todo_list, &net_todo_list);
5256 static void rollback_registered_many(struct list_head *head)
5258 struct net_device *dev, *tmp;
5260 BUG_ON(dev_boot_phase);
5263 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5264 /* Some devices call without registering
5265 * for initialization unwind. Remove those
5266 * devices and proceed with the remaining.
5268 if (dev->reg_state == NETREG_UNINITIALIZED) {
5269 pr_debug("unregister_netdevice: device %s/%p never "
5270 "was registered\n", dev->name, dev);
5273 list_del(&dev->unreg_list);
5276 dev->dismantle = true;
5277 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5280 /* If device is running, close it first. */
5281 dev_close_many(head);
5283 list_for_each_entry(dev, head, unreg_list) {
5284 /* And unlink it from device chain. */
5285 unlist_netdevice(dev);
5287 dev->reg_state = NETREG_UNREGISTERING;
5292 list_for_each_entry(dev, head, unreg_list) {
5293 /* Shutdown queueing discipline. */
5297 /* Notify protocols, that we are about to destroy
5298 this device. They should clean all the things.
5300 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5302 if (!dev->rtnl_link_ops ||
5303 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5304 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5307 * Flush the unicast and multicast chains
5312 if (dev->netdev_ops->ndo_uninit)
5313 dev->netdev_ops->ndo_uninit(dev);
5315 /* Notifier chain MUST detach us from master device. */
5316 WARN_ON(dev->master);
5318 /* Remove entries from kobject tree */
5319 netdev_unregister_kobject(dev);
5322 /* Process any work delayed until the end of the batch */
5323 dev = list_first_entry(head, struct net_device, unreg_list);
5324 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5328 list_for_each_entry(dev, head, unreg_list)
5332 static void rollback_registered(struct net_device *dev)
5336 list_add(&dev->unreg_list, &single);
5337 rollback_registered_many(&single);
5341 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5343 /* Fix illegal checksum combinations */
5344 if ((features & NETIF_F_HW_CSUM) &&
5345 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5346 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5347 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5350 if ((features & NETIF_F_NO_CSUM) &&
5351 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5352 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5353 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5356 /* Fix illegal SG+CSUM combinations. */
5357 if ((features & NETIF_F_SG) &&
5358 !(features & NETIF_F_ALL_CSUM)) {
5360 "Dropping NETIF_F_SG since no checksum feature.\n");
5361 features &= ~NETIF_F_SG;
5364 /* TSO requires that SG is present as well. */
5365 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5366 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5367 features &= ~NETIF_F_ALL_TSO;
5370 /* TSO ECN requires that TSO is present as well. */
5371 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5372 features &= ~NETIF_F_TSO_ECN;
5374 /* Software GSO depends on SG. */
5375 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5376 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5377 features &= ~NETIF_F_GSO;
5380 /* UFO needs SG and checksumming */
5381 if (features & NETIF_F_UFO) {
5382 /* maybe split UFO into V4 and V6? */
5383 if (!((features & NETIF_F_GEN_CSUM) ||
5384 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5385 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5387 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5388 features &= ~NETIF_F_UFO;
5391 if (!(features & NETIF_F_SG)) {
5393 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5394 features &= ~NETIF_F_UFO;
5401 int __netdev_update_features(struct net_device *dev)
5408 features = netdev_get_wanted_features(dev);
5410 if (dev->netdev_ops->ndo_fix_features)
5411 features = dev->netdev_ops->ndo_fix_features(dev, features);
5413 /* driver might be less strict about feature dependencies */
5414 features = netdev_fix_features(dev, features);
5416 if (dev->features == features)
5419 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5420 dev->features, features);
5422 if (dev->netdev_ops->ndo_set_features)
5423 err = dev->netdev_ops->ndo_set_features(dev, features);
5425 if (unlikely(err < 0)) {
5427 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5428 err, features, dev->features);
5433 dev->features = features;
5439 * netdev_update_features - recalculate device features
5440 * @dev: the device to check
5442 * Recalculate dev->features set and send notifications if it
5443 * has changed. Should be called after driver or hardware dependent
5444 * conditions might have changed that influence the features.
5446 void netdev_update_features(struct net_device *dev)
5448 if (__netdev_update_features(dev))
5449 netdev_features_change(dev);
5451 EXPORT_SYMBOL(netdev_update_features);
5454 * netdev_change_features - recalculate device features
5455 * @dev: the device to check
5457 * Recalculate dev->features set and send notifications even
5458 * if they have not changed. Should be called instead of
5459 * netdev_update_features() if also dev->vlan_features might
5460 * have changed to allow the changes to be propagated to stacked
5463 void netdev_change_features(struct net_device *dev)
5465 __netdev_update_features(dev);
5466 netdev_features_change(dev);
5468 EXPORT_SYMBOL(netdev_change_features);
5471 * netif_stacked_transfer_operstate - transfer operstate
5472 * @rootdev: the root or lower level device to transfer state from
5473 * @dev: the device to transfer operstate to
5475 * Transfer operational state from root to device. This is normally
5476 * called when a stacking relationship exists between the root
5477 * device and the device(a leaf device).
5479 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5480 struct net_device *dev)
5482 if (rootdev->operstate == IF_OPER_DORMANT)
5483 netif_dormant_on(dev);
5485 netif_dormant_off(dev);
5487 if (netif_carrier_ok(rootdev)) {
5488 if (!netif_carrier_ok(dev))
5489 netif_carrier_on(dev);
5491 if (netif_carrier_ok(dev))
5492 netif_carrier_off(dev);
5495 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5498 static int netif_alloc_rx_queues(struct net_device *dev)
5500 unsigned int i, count = dev->num_rx_queues;
5501 struct netdev_rx_queue *rx;
5505 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5507 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5512 for (i = 0; i < count; i++)
5518 static void netdev_init_one_queue(struct net_device *dev,
5519 struct netdev_queue *queue, void *_unused)
5521 /* Initialize queue lock */
5522 spin_lock_init(&queue->_xmit_lock);
5523 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5524 queue->xmit_lock_owner = -1;
5525 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5529 static int netif_alloc_netdev_queues(struct net_device *dev)
5531 unsigned int count = dev->num_tx_queues;
5532 struct netdev_queue *tx;
5536 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5538 pr_err("netdev: Unable to allocate %u tx queues.\n",
5544 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5545 spin_lock_init(&dev->tx_global_lock);
5551 * register_netdevice - register a network device
5552 * @dev: device to register
5554 * Take a completed network device structure and add it to the kernel
5555 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5556 * chain. 0 is returned on success. A negative errno code is returned
5557 * on a failure to set up the device, or if the name is a duplicate.
5559 * Callers must hold the rtnl semaphore. You may want
5560 * register_netdev() instead of this.
5563 * The locking appears insufficient to guarantee two parallel registers
5564 * will not get the same name.
5567 int register_netdevice(struct net_device *dev)
5570 struct net *net = dev_net(dev);
5572 BUG_ON(dev_boot_phase);
5577 /* When net_device's are persistent, this will be fatal. */
5578 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5581 spin_lock_init(&dev->addr_list_lock);
5582 netdev_set_addr_lockdep_class(dev);
5586 ret = dev_get_valid_name(dev, dev->name);
5590 /* Init, if this function is available */
5591 if (dev->netdev_ops->ndo_init) {
5592 ret = dev->netdev_ops->ndo_init(dev);
5600 dev->ifindex = dev_new_index(net);
5601 if (dev->iflink == -1)
5602 dev->iflink = dev->ifindex;
5604 /* Transfer changeable features to wanted_features and enable
5605 * software offloads (GSO and GRO).
5607 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5608 dev->features |= NETIF_F_SOFT_FEATURES;
5609 dev->wanted_features = dev->features & dev->hw_features;
5611 /* Turn on no cache copy if HW is doing checksum */
5612 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5613 if ((dev->features & NETIF_F_ALL_CSUM) &&
5614 !(dev->features & NETIF_F_NO_CSUM)) {
5615 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5616 dev->features |= NETIF_F_NOCACHE_COPY;
5619 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5621 dev->vlan_features |= NETIF_F_HIGHDMA;
5623 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5624 ret = notifier_to_errno(ret);
5628 ret = netdev_register_kobject(dev);
5631 dev->reg_state = NETREG_REGISTERED;
5633 __netdev_update_features(dev);
5636 * Default initial state at registry is that the
5637 * device is present.
5640 set_bit(__LINK_STATE_PRESENT, &dev->state);
5642 dev_init_scheduler(dev);
5644 list_netdevice(dev);
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;
6012 INIT_LIST_HEAD(&dev->napi_list);
6013 INIT_LIST_HEAD(&dev->unreg_list);
6014 INIT_LIST_HEAD(&dev->link_watch_list);
6015 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6018 dev->num_tx_queues = txqs;
6019 dev->real_num_tx_queues = txqs;
6020 if (netif_alloc_netdev_queues(dev))
6024 dev->num_rx_queues = rxqs;
6025 dev->real_num_rx_queues = rxqs;
6026 if (netif_alloc_rx_queues(dev))
6030 strcpy(dev->name, name);
6031 dev->group = INIT_NETDEV_GROUP;
6039 free_percpu(dev->pcpu_refcnt);
6049 EXPORT_SYMBOL(alloc_netdev_mqs);
6052 * free_netdev - free network device
6055 * This function does the last stage of destroying an allocated device
6056 * interface. The reference to the device object is released.
6057 * If this is the last reference then it will be freed.
6059 void free_netdev(struct net_device *dev)
6061 struct napi_struct *p, *n;
6063 release_net(dev_net(dev));
6070 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6072 /* Flush device addresses */
6073 dev_addr_flush(dev);
6075 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6078 free_percpu(dev->pcpu_refcnt);
6079 dev->pcpu_refcnt = NULL;
6081 /* Compatibility with error handling in drivers */
6082 if (dev->reg_state == NETREG_UNINITIALIZED) {
6083 kfree((char *)dev - dev->padded);
6087 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6088 dev->reg_state = NETREG_RELEASED;
6090 /* will free via device release */
6091 put_device(&dev->dev);
6093 EXPORT_SYMBOL(free_netdev);
6096 * synchronize_net - Synchronize with packet receive processing
6098 * Wait for packets currently being received to be done.
6099 * Does not block later packets from starting.
6101 void synchronize_net(void)
6104 if (rtnl_is_locked())
6105 synchronize_rcu_expedited();
6109 EXPORT_SYMBOL(synchronize_net);
6112 * unregister_netdevice_queue - remove device from the kernel
6116 * This function shuts down a device interface and removes it
6117 * from the kernel tables.
6118 * If head not NULL, device is queued to be unregistered later.
6120 * Callers must hold the rtnl semaphore. You may want
6121 * unregister_netdev() instead of this.
6124 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6129 list_move_tail(&dev->unreg_list, head);
6131 rollback_registered(dev);
6132 /* Finish processing unregister after unlock */
6136 EXPORT_SYMBOL(unregister_netdevice_queue);
6139 * unregister_netdevice_many - unregister many devices
6140 * @head: list of devices
6142 void unregister_netdevice_many(struct list_head *head)
6144 struct net_device *dev;
6146 if (!list_empty(head)) {
6147 rollback_registered_many(head);
6148 list_for_each_entry(dev, head, unreg_list)
6152 EXPORT_SYMBOL(unregister_netdevice_many);
6155 * unregister_netdev - remove device from the kernel
6158 * This function shuts down a device interface and removes it
6159 * from the kernel tables.
6161 * This is just a wrapper for unregister_netdevice that takes
6162 * the rtnl semaphore. In general you want to use this and not
6163 * unregister_netdevice.
6165 void unregister_netdev(struct net_device *dev)
6168 unregister_netdevice(dev);
6171 EXPORT_SYMBOL(unregister_netdev);
6174 * dev_change_net_namespace - move device to different nethost namespace
6176 * @net: network namespace
6177 * @pat: If not NULL name pattern to try if the current device name
6178 * is already taken in the destination network namespace.
6180 * This function shuts down a device interface and moves it
6181 * to a new network namespace. On success 0 is returned, on
6182 * a failure a netagive errno code is returned.
6184 * Callers must hold the rtnl semaphore.
6187 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6193 /* Don't allow namespace local devices to be moved. */
6195 if (dev->features & NETIF_F_NETNS_LOCAL)
6198 /* Ensure the device has been registrered */
6200 if (dev->reg_state != NETREG_REGISTERED)
6203 /* Get out if there is nothing todo */
6205 if (net_eq(dev_net(dev), net))
6208 /* Pick the destination device name, and ensure
6209 * we can use it in the destination network namespace.
6212 if (__dev_get_by_name(net, dev->name)) {
6213 /* We get here if we can't use the current device name */
6216 if (dev_get_valid_name(dev, pat) < 0)
6221 * And now a mini version of register_netdevice unregister_netdevice.
6224 /* If device is running close it first. */
6227 /* And unlink it from device chain */
6229 unlist_netdevice(dev);
6233 /* Shutdown queueing discipline. */
6236 /* Notify protocols, that we are about to destroy
6237 this device. They should clean all the things.
6239 Note that dev->reg_state stays at NETREG_REGISTERED.
6240 This is wanted because this way 8021q and macvlan know
6241 the device is just moving and can keep their slaves up.
6243 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6244 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6245 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6248 * Flush the unicast and multicast chains
6253 /* Actually switch the network namespace */
6254 dev_net_set(dev, net);
6256 /* If there is an ifindex conflict assign a new one */
6257 if (__dev_get_by_index(net, dev->ifindex)) {
6258 int iflink = (dev->iflink == dev->ifindex);
6259 dev->ifindex = dev_new_index(net);
6261 dev->iflink = dev->ifindex;
6264 /* Fixup kobjects */
6265 err = device_rename(&dev->dev, dev->name);
6268 /* Add the device back in the hashes */
6269 list_netdevice(dev);
6271 /* Notify protocols, that a new device appeared. */
6272 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6275 * Prevent userspace races by waiting until the network
6276 * device is fully setup before sending notifications.
6278 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6285 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6287 static int dev_cpu_callback(struct notifier_block *nfb,
6288 unsigned long action,
6291 struct sk_buff **list_skb;
6292 struct sk_buff *skb;
6293 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6294 struct softnet_data *sd, *oldsd;
6296 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6299 local_irq_disable();
6300 cpu = smp_processor_id();
6301 sd = &per_cpu(softnet_data, cpu);
6302 oldsd = &per_cpu(softnet_data, oldcpu);
6304 /* Find end of our completion_queue. */
6305 list_skb = &sd->completion_queue;
6307 list_skb = &(*list_skb)->next;
6308 /* Append completion queue from offline CPU. */
6309 *list_skb = oldsd->completion_queue;
6310 oldsd->completion_queue = NULL;
6312 /* Append output queue from offline CPU. */
6313 if (oldsd->output_queue) {
6314 *sd->output_queue_tailp = oldsd->output_queue;
6315 sd->output_queue_tailp = oldsd->output_queue_tailp;
6316 oldsd->output_queue = NULL;
6317 oldsd->output_queue_tailp = &oldsd->output_queue;
6319 /* Append NAPI poll list from offline CPU. */
6320 if (!list_empty(&oldsd->poll_list)) {
6321 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6322 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6325 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6328 /* Process offline CPU's input_pkt_queue */
6329 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6331 input_queue_head_incr(oldsd);
6333 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6335 input_queue_head_incr(oldsd);
6343 * netdev_increment_features - increment feature set by one
6344 * @all: current feature set
6345 * @one: new feature set
6346 * @mask: mask feature set
6348 * Computes a new feature set after adding a device with feature set
6349 * @one to the master device with current feature set @all. Will not
6350 * enable anything that is off in @mask. Returns the new feature set.
6352 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6354 if (mask & NETIF_F_GEN_CSUM)
6355 mask |= NETIF_F_ALL_CSUM;
6356 mask |= NETIF_F_VLAN_CHALLENGED;
6358 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6359 all &= one | ~NETIF_F_ALL_FOR_ALL;
6361 /* If device needs checksumming, downgrade to it. */
6362 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6363 all &= ~NETIF_F_NO_CSUM;
6365 /* If one device supports hw checksumming, set for all. */
6366 if (all & NETIF_F_GEN_CSUM)
6367 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6371 EXPORT_SYMBOL(netdev_increment_features);
6373 static struct hlist_head *netdev_create_hash(void)
6376 struct hlist_head *hash;
6378 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6380 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6381 INIT_HLIST_HEAD(&hash[i]);
6386 /* Initialize per network namespace state */
6387 static int __net_init netdev_init(struct net *net)
6389 INIT_LIST_HEAD(&net->dev_base_head);
6391 net->dev_name_head = netdev_create_hash();
6392 if (net->dev_name_head == NULL)
6395 net->dev_index_head = netdev_create_hash();
6396 if (net->dev_index_head == NULL)
6402 kfree(net->dev_name_head);
6408 * netdev_drivername - network driver for the device
6409 * @dev: network device
6411 * Determine network driver for device.
6413 const char *netdev_drivername(const struct net_device *dev)
6415 const struct device_driver *driver;
6416 const struct device *parent;
6417 const char *empty = "";
6419 parent = dev->dev.parent;
6423 driver = parent->driver;
6424 if (driver && driver->name)
6425 return driver->name;
6429 int __netdev_printk(const char *level, const struct net_device *dev,
6430 struct va_format *vaf)
6434 if (dev && dev->dev.parent)
6435 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6436 netdev_name(dev), vaf);
6438 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6440 r = printk("%s(NULL net_device): %pV", level, vaf);
6444 EXPORT_SYMBOL(__netdev_printk);
6446 int netdev_printk(const char *level, const struct net_device *dev,
6447 const char *format, ...)
6449 struct va_format vaf;
6453 va_start(args, format);
6458 r = __netdev_printk(level, dev, &vaf);
6463 EXPORT_SYMBOL(netdev_printk);
6465 #define define_netdev_printk_level(func, level) \
6466 int func(const struct net_device *dev, const char *fmt, ...) \
6469 struct va_format vaf; \
6472 va_start(args, fmt); \
6477 r = __netdev_printk(level, dev, &vaf); \
6482 EXPORT_SYMBOL(func);
6484 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6485 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6486 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6487 define_netdev_printk_level(netdev_err, KERN_ERR);
6488 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6489 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6490 define_netdev_printk_level(netdev_info, KERN_INFO);
6492 static void __net_exit netdev_exit(struct net *net)
6494 kfree(net->dev_name_head);
6495 kfree(net->dev_index_head);
6498 static struct pernet_operations __net_initdata netdev_net_ops = {
6499 .init = netdev_init,
6500 .exit = netdev_exit,
6503 static void __net_exit default_device_exit(struct net *net)
6505 struct net_device *dev, *aux;
6507 * Push all migratable network devices back to the
6508 * initial network namespace
6511 for_each_netdev_safe(net, dev, aux) {
6513 char fb_name[IFNAMSIZ];
6515 /* Ignore unmoveable devices (i.e. loopback) */
6516 if (dev->features & NETIF_F_NETNS_LOCAL)
6519 /* Leave virtual devices for the generic cleanup */
6520 if (dev->rtnl_link_ops)
6523 /* Push remaining network devices to init_net */
6524 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6525 err = dev_change_net_namespace(dev, &init_net, fb_name);
6527 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6528 __func__, dev->name, err);
6535 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6537 /* At exit all network devices most be removed from a network
6538 * namespace. Do this in the reverse order of registration.
6539 * Do this across as many network namespaces as possible to
6540 * improve batching efficiency.
6542 struct net_device *dev;
6544 LIST_HEAD(dev_kill_list);
6547 list_for_each_entry(net, net_list, exit_list) {
6548 for_each_netdev_reverse(net, dev) {
6549 if (dev->rtnl_link_ops)
6550 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6552 unregister_netdevice_queue(dev, &dev_kill_list);
6555 unregister_netdevice_many(&dev_kill_list);
6556 list_del(&dev_kill_list);
6560 static struct pernet_operations __net_initdata default_device_ops = {
6561 .exit = default_device_exit,
6562 .exit_batch = default_device_exit_batch,
6566 * Initialize the DEV module. At boot time this walks the device list and
6567 * unhooks any devices that fail to initialise (normally hardware not
6568 * present) and leaves us with a valid list of present and active devices.
6573 * This is called single threaded during boot, so no need
6574 * to take the rtnl semaphore.
6576 static int __init net_dev_init(void)
6578 int i, rc = -ENOMEM;
6580 BUG_ON(!dev_boot_phase);
6582 if (dev_proc_init())
6585 if (netdev_kobject_init())
6588 INIT_LIST_HEAD(&ptype_all);
6589 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6590 INIT_LIST_HEAD(&ptype_base[i]);
6592 if (register_pernet_subsys(&netdev_net_ops))
6596 * Initialise the packet receive queues.
6599 for_each_possible_cpu(i) {
6600 struct softnet_data *sd = &per_cpu(softnet_data, i);
6602 memset(sd, 0, sizeof(*sd));
6603 skb_queue_head_init(&sd->input_pkt_queue);
6604 skb_queue_head_init(&sd->process_queue);
6605 sd->completion_queue = NULL;
6606 INIT_LIST_HEAD(&sd->poll_list);
6607 sd->output_queue = NULL;
6608 sd->output_queue_tailp = &sd->output_queue;
6610 sd->csd.func = rps_trigger_softirq;
6616 sd->backlog.poll = process_backlog;
6617 sd->backlog.weight = weight_p;
6618 sd->backlog.gro_list = NULL;
6619 sd->backlog.gro_count = 0;
6624 /* The loopback device is special if any other network devices
6625 * is present in a network namespace the loopback device must
6626 * be present. Since we now dynamically allocate and free the
6627 * loopback device ensure this invariant is maintained by
6628 * keeping the loopback device as the first device on the
6629 * list of network devices. Ensuring the loopback devices
6630 * is the first device that appears and the last network device
6633 if (register_pernet_device(&loopback_net_ops))
6636 if (register_pernet_device(&default_device_ops))
6639 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6640 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6642 hotcpu_notifier(dev_cpu_callback, 0);
6650 subsys_initcall(net_dev_init);
6652 static int __init initialize_hashrnd(void)
6654 get_random_bytes(&hashrnd, sizeof(hashrnd));
6658 late_initcall_sync(initialize_hashrnd);