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 == '/' || *name == ':' || isspace(*name))
868 EXPORT_SYMBOL(dev_valid_name);
871 * __dev_alloc_name - allocate a name for a device
872 * @net: network namespace to allocate the device name in
873 * @name: name format string
874 * @buf: scratch buffer and result name string
876 * Passed a format string - eg "lt%d" it will try and find a suitable
877 * id. It scans list of devices to build up a free map, then chooses
878 * the first empty slot. The caller must hold the dev_base or rtnl lock
879 * while allocating the name and adding the device in order to avoid
881 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
882 * Returns the number of the unit assigned or a negative errno code.
885 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
889 const int max_netdevices = 8*PAGE_SIZE;
890 unsigned long *inuse;
891 struct net_device *d;
893 p = strnchr(name, IFNAMSIZ-1, '%');
896 * Verify the string as this thing may have come from
897 * the user. There must be either one "%d" and no other "%"
900 if (p[1] != 'd' || strchr(p + 2, '%'))
903 /* Use one page as a bit array of possible slots */
904 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
908 for_each_netdev(net, d) {
909 if (!sscanf(d->name, name, &i))
911 if (i < 0 || i >= max_netdevices)
914 /* avoid cases where sscanf is not exact inverse of printf */
915 snprintf(buf, IFNAMSIZ, name, i);
916 if (!strncmp(buf, d->name, IFNAMSIZ))
920 i = find_first_zero_bit(inuse, max_netdevices);
921 free_page((unsigned long) inuse);
925 snprintf(buf, IFNAMSIZ, name, i);
926 if (!__dev_get_by_name(net, buf))
929 /* It is possible to run out of possible slots
930 * when the name is long and there isn't enough space left
931 * for the digits, or if all bits are used.
937 * dev_alloc_name - allocate a name for a device
939 * @name: name format string
941 * Passed a format string - eg "lt%d" it will try and find a suitable
942 * id. It scans list of devices to build up a free map, then chooses
943 * the first empty slot. The caller must hold the dev_base or rtnl lock
944 * while allocating the name and adding the device in order to avoid
946 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
947 * Returns the number of the unit assigned or a negative errno code.
950 int dev_alloc_name(struct net_device *dev, const char *name)
956 BUG_ON(!dev_net(dev));
958 ret = __dev_alloc_name(net, name, buf);
960 strlcpy(dev->name, buf, IFNAMSIZ);
963 EXPORT_SYMBOL(dev_alloc_name);
965 static int dev_get_valid_name(struct net_device *dev, const char *name)
969 BUG_ON(!dev_net(dev));
972 if (!dev_valid_name(name))
975 if (strchr(name, '%'))
976 return dev_alloc_name(dev, name);
977 else if (__dev_get_by_name(net, name))
979 else if (dev->name != name)
980 strlcpy(dev->name, name, IFNAMSIZ);
986 * dev_change_name - change name of a device
988 * @newname: name (or format string) must be at least IFNAMSIZ
990 * Change name of a device, can pass format strings "eth%d".
993 int dev_change_name(struct net_device *dev, const char *newname)
995 char oldname[IFNAMSIZ];
1001 BUG_ON(!dev_net(dev));
1004 if (dev->flags & IFF_UP)
1007 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1010 memcpy(oldname, dev->name, IFNAMSIZ);
1012 err = dev_get_valid_name(dev, newname);
1017 ret = device_rename(&dev->dev, dev->name);
1019 memcpy(dev->name, oldname, IFNAMSIZ);
1023 write_lock_bh(&dev_base_lock);
1024 hlist_del_rcu(&dev->name_hlist);
1025 write_unlock_bh(&dev_base_lock);
1029 write_lock_bh(&dev_base_lock);
1030 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1031 write_unlock_bh(&dev_base_lock);
1033 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1034 ret = notifier_to_errno(ret);
1037 /* err >= 0 after dev_alloc_name() or stores the first errno */
1040 memcpy(dev->name, oldname, IFNAMSIZ);
1044 "%s: name change rollback failed: %d.\n",
1053 * dev_set_alias - change ifalias of a device
1055 * @alias: name up to IFALIASZ
1056 * @len: limit of bytes to copy from info
1058 * Set ifalias for a device,
1060 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1066 if (len >= IFALIASZ)
1071 kfree(dev->ifalias);
1072 dev->ifalias = NULL;
1077 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1080 dev->ifalias = new_ifalias;
1082 strlcpy(dev->ifalias, alias, len+1);
1088 * netdev_features_change - device changes features
1089 * @dev: device to cause notification
1091 * Called to indicate a device has changed features.
1093 void netdev_features_change(struct net_device *dev)
1095 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1097 EXPORT_SYMBOL(netdev_features_change);
1100 * netdev_state_change - device changes state
1101 * @dev: device to cause notification
1103 * Called to indicate a device has changed state. This function calls
1104 * the notifier chains for netdev_chain and sends a NEWLINK message
1105 * to the routing socket.
1107 void netdev_state_change(struct net_device *dev)
1109 if (dev->flags & IFF_UP) {
1110 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1111 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1114 EXPORT_SYMBOL(netdev_state_change);
1116 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1118 return call_netdevice_notifiers(event, dev);
1120 EXPORT_SYMBOL(netdev_bonding_change);
1123 * dev_load - load a network module
1124 * @net: the applicable net namespace
1125 * @name: name of interface
1127 * If a network interface is not present and the process has suitable
1128 * privileges this function loads the module. If module loading is not
1129 * available in this kernel then it becomes a nop.
1132 void dev_load(struct net *net, const char *name)
1134 struct net_device *dev;
1138 dev = dev_get_by_name_rcu(net, name);
1142 if (no_module && capable(CAP_NET_ADMIN))
1143 no_module = request_module("netdev-%s", name);
1144 if (no_module && capable(CAP_SYS_MODULE)) {
1145 if (!request_module("%s", name))
1146 pr_err("Loading kernel module for a network device "
1147 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1151 EXPORT_SYMBOL(dev_load);
1153 static int __dev_open(struct net_device *dev)
1155 const struct net_device_ops *ops = dev->netdev_ops;
1160 if (!netif_device_present(dev))
1163 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1164 ret = notifier_to_errno(ret);
1168 set_bit(__LINK_STATE_START, &dev->state);
1170 if (ops->ndo_validate_addr)
1171 ret = ops->ndo_validate_addr(dev);
1173 if (!ret && ops->ndo_open)
1174 ret = ops->ndo_open(dev);
1177 clear_bit(__LINK_STATE_START, &dev->state);
1179 dev->flags |= IFF_UP;
1180 net_dmaengine_get();
1181 dev_set_rx_mode(dev);
1183 add_device_randomness(dev->dev_addr, dev->addr_len);
1190 * dev_open - prepare an interface for use.
1191 * @dev: device to open
1193 * Takes a device from down to up state. The device's private open
1194 * function is invoked and then the multicast lists are loaded. Finally
1195 * the device is moved into the up state and a %NETDEV_UP message is
1196 * sent to the netdev notifier chain.
1198 * Calling this function on an active interface is a nop. On a failure
1199 * a negative errno code is returned.
1201 int dev_open(struct net_device *dev)
1205 if (dev->flags & IFF_UP)
1208 ret = __dev_open(dev);
1212 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1213 call_netdevice_notifiers(NETDEV_UP, dev);
1217 EXPORT_SYMBOL(dev_open);
1219 static int __dev_close_many(struct list_head *head)
1221 struct net_device *dev;
1226 list_for_each_entry(dev, head, unreg_list) {
1227 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1229 clear_bit(__LINK_STATE_START, &dev->state);
1231 /* Synchronize to scheduled poll. We cannot touch poll list, it
1232 * can be even on different cpu. So just clear netif_running().
1234 * dev->stop() will invoke napi_disable() on all of it's
1235 * napi_struct instances on this device.
1237 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1240 dev_deactivate_many(head);
1242 list_for_each_entry(dev, head, unreg_list) {
1243 const struct net_device_ops *ops = dev->netdev_ops;
1246 * Call the device specific close. This cannot fail.
1247 * Only if device is UP
1249 * We allow it to be called even after a DETACH hot-plug
1255 dev->flags &= ~IFF_UP;
1256 net_dmaengine_put();
1262 static int __dev_close(struct net_device *dev)
1267 list_add(&dev->unreg_list, &single);
1268 retval = __dev_close_many(&single);
1273 static int dev_close_many(struct list_head *head)
1275 struct net_device *dev, *tmp;
1276 LIST_HEAD(tmp_list);
1278 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1279 if (!(dev->flags & IFF_UP))
1280 list_move(&dev->unreg_list, &tmp_list);
1282 __dev_close_many(head);
1284 list_for_each_entry(dev, head, unreg_list) {
1285 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1286 call_netdevice_notifiers(NETDEV_DOWN, dev);
1289 /* rollback_registered_many needs the complete original list */
1290 list_splice(&tmp_list, head);
1295 * dev_close - shutdown an interface.
1296 * @dev: device to shutdown
1298 * This function moves an active device into down state. A
1299 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1300 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1303 int dev_close(struct net_device *dev)
1305 if (dev->flags & IFF_UP) {
1308 list_add(&dev->unreg_list, &single);
1309 dev_close_many(&single);
1314 EXPORT_SYMBOL(dev_close);
1318 * dev_disable_lro - disable Large Receive Offload on a device
1321 * Disable Large Receive Offload (LRO) on a net device. Must be
1322 * called under RTNL. This is needed if received packets may be
1323 * forwarded to another interface.
1325 void dev_disable_lro(struct net_device *dev)
1330 * If we're trying to disable lro on a vlan device
1331 * use the underlying physical device instead
1333 if (is_vlan_dev(dev))
1334 dev = vlan_dev_real_dev(dev);
1336 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1337 flags = dev->ethtool_ops->get_flags(dev);
1339 flags = ethtool_op_get_flags(dev);
1341 if (!(flags & ETH_FLAG_LRO))
1344 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1345 if (unlikely(dev->features & NETIF_F_LRO))
1346 netdev_WARN(dev, "failed to disable LRO!\n");
1348 EXPORT_SYMBOL(dev_disable_lro);
1351 static int dev_boot_phase = 1;
1354 * register_netdevice_notifier - register a network notifier block
1357 * Register a notifier to be called when network device events occur.
1358 * The notifier passed is linked into the kernel structures and must
1359 * not be reused until it has been unregistered. A negative errno code
1360 * is returned on a failure.
1362 * When registered all registration and up events are replayed
1363 * to the new notifier to allow device to have a race free
1364 * view of the network device list.
1367 int register_netdevice_notifier(struct notifier_block *nb)
1369 struct net_device *dev;
1370 struct net_device *last;
1375 err = raw_notifier_chain_register(&netdev_chain, nb);
1381 for_each_netdev(net, dev) {
1382 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1383 err = notifier_to_errno(err);
1387 if (!(dev->flags & IFF_UP))
1390 nb->notifier_call(nb, NETDEV_UP, dev);
1401 for_each_netdev(net, dev) {
1405 if (dev->flags & IFF_UP) {
1406 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1407 nb->notifier_call(nb, NETDEV_DOWN, dev);
1409 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1410 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1415 raw_notifier_chain_unregister(&netdev_chain, nb);
1418 EXPORT_SYMBOL(register_netdevice_notifier);
1421 * unregister_netdevice_notifier - unregister a network notifier block
1424 * Unregister a notifier previously registered by
1425 * register_netdevice_notifier(). The notifier is unlinked into the
1426 * kernel structures and may then be reused. A negative errno code
1427 * is returned on a failure.
1429 * After unregistering unregister and down device events are synthesized
1430 * for all devices on the device list to the removed notifier to remove
1431 * the need for special case cleanup code.
1434 int unregister_netdevice_notifier(struct notifier_block *nb)
1436 struct net_device *dev;
1441 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1446 for_each_netdev(net, dev) {
1447 if (dev->flags & IFF_UP) {
1448 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1449 nb->notifier_call(nb, NETDEV_DOWN, dev);
1451 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1452 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1459 EXPORT_SYMBOL(unregister_netdevice_notifier);
1462 * call_netdevice_notifiers - call all network notifier blocks
1463 * @val: value passed unmodified to notifier function
1464 * @dev: net_device pointer passed unmodified to notifier function
1466 * Call all network notifier blocks. Parameters and return value
1467 * are as for raw_notifier_call_chain().
1470 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1473 return raw_notifier_call_chain(&netdev_chain, val, dev);
1475 EXPORT_SYMBOL(call_netdevice_notifiers);
1477 /* When > 0 there are consumers of rx skb time stamps */
1478 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1480 void net_enable_timestamp(void)
1482 atomic_inc(&netstamp_needed);
1484 EXPORT_SYMBOL(net_enable_timestamp);
1486 void net_disable_timestamp(void)
1488 atomic_dec(&netstamp_needed);
1490 EXPORT_SYMBOL(net_disable_timestamp);
1492 static inline void net_timestamp_set(struct sk_buff *skb)
1494 if (atomic_read(&netstamp_needed))
1495 __net_timestamp(skb);
1497 skb->tstamp.tv64 = 0;
1500 static inline void net_timestamp_check(struct sk_buff *skb)
1502 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1503 __net_timestamp(skb);
1506 static int net_hwtstamp_validate(struct ifreq *ifr)
1508 struct hwtstamp_config cfg;
1509 enum hwtstamp_tx_types tx_type;
1510 enum hwtstamp_rx_filters rx_filter;
1511 int tx_type_valid = 0;
1512 int rx_filter_valid = 0;
1514 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1517 if (cfg.flags) /* reserved for future extensions */
1520 tx_type = cfg.tx_type;
1521 rx_filter = cfg.rx_filter;
1524 case HWTSTAMP_TX_OFF:
1525 case HWTSTAMP_TX_ON:
1526 case HWTSTAMP_TX_ONESTEP_SYNC:
1531 switch (rx_filter) {
1532 case HWTSTAMP_FILTER_NONE:
1533 case HWTSTAMP_FILTER_ALL:
1534 case HWTSTAMP_FILTER_SOME:
1535 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1536 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1537 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1538 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1539 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1540 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1541 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1542 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1543 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1544 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1545 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1546 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1547 rx_filter_valid = 1;
1551 if (!tx_type_valid || !rx_filter_valid)
1557 static inline bool is_skb_forwardable(struct net_device *dev,
1558 struct sk_buff *skb)
1562 if (!(dev->flags & IFF_UP))
1565 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1566 if (skb->len <= len)
1569 /* if TSO is enabled, we don't care about the length as the packet
1570 * could be forwarded without being segmented before
1572 if (skb_is_gso(skb))
1579 * dev_forward_skb - loopback an skb to another netif
1581 * @dev: destination network device
1582 * @skb: buffer to forward
1585 * NET_RX_SUCCESS (no congestion)
1586 * NET_RX_DROP (packet was dropped, but freed)
1588 * dev_forward_skb can be used for injecting an skb from the
1589 * start_xmit function of one device into the receive queue
1590 * of another device.
1592 * The receiving device may be in another namespace, so
1593 * we have to clear all information in the skb that could
1594 * impact namespace isolation.
1596 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1598 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1599 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1600 atomic_long_inc(&dev->rx_dropped);
1609 if (unlikely(!is_skb_forwardable(dev, skb))) {
1610 atomic_long_inc(&dev->rx_dropped);
1616 skb->tstamp.tv64 = 0;
1617 skb->pkt_type = PACKET_HOST;
1618 skb->protocol = eth_type_trans(skb, dev);
1619 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1623 nf_reset_trace(skb);
1624 return netif_rx(skb);
1626 EXPORT_SYMBOL_GPL(dev_forward_skb);
1628 static inline int deliver_skb(struct sk_buff *skb,
1629 struct packet_type *pt_prev,
1630 struct net_device *orig_dev)
1632 atomic_inc(&skb->users);
1633 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1636 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1638 if (!ptype->af_packet_priv || !skb->sk)
1641 if (ptype->id_match)
1642 return ptype->id_match(ptype, skb->sk);
1643 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1650 * Support routine. Sends outgoing frames to any network
1651 * taps currently in use.
1654 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1656 struct packet_type *ptype;
1657 struct sk_buff *skb2 = NULL;
1658 struct packet_type *pt_prev = NULL;
1661 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1662 /* Never send packets back to the socket
1663 * they originated from - MvS (miquels@drinkel.ow.org)
1665 if ((ptype->dev == dev || !ptype->dev) &&
1666 (!skb_loop_sk(ptype, skb))) {
1668 deliver_skb(skb2, pt_prev, skb->dev);
1673 skb2 = skb_clone(skb, GFP_ATOMIC);
1677 net_timestamp_set(skb2);
1679 /* skb->nh should be correctly
1680 set by sender, so that the second statement is
1681 just protection against buggy protocols.
1683 skb_reset_mac_header(skb2);
1685 if (skb_network_header(skb2) < skb2->data ||
1686 skb2->network_header > skb2->tail) {
1687 if (net_ratelimit())
1688 printk(KERN_CRIT "protocol %04x is "
1690 ntohs(skb2->protocol),
1692 skb_reset_network_header(skb2);
1695 skb2->transport_header = skb2->network_header;
1696 skb2->pkt_type = PACKET_OUTGOING;
1701 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1705 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1706 * @dev: Network device
1707 * @txq: number of queues available
1709 * If real_num_tx_queues is changed the tc mappings may no longer be
1710 * valid. To resolve this verify the tc mapping remains valid and if
1711 * not NULL the mapping. With no priorities mapping to this
1712 * offset/count pair it will no longer be used. In the worst case TC0
1713 * is invalid nothing can be done so disable priority mappings. If is
1714 * expected that drivers will fix this mapping if they can before
1715 * calling netif_set_real_num_tx_queues.
1717 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1720 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1722 /* If TC0 is invalidated disable TC mapping */
1723 if (tc->offset + tc->count > txq) {
1724 pr_warning("Number of in use tx queues changed "
1725 "invalidating tc mappings. Priority "
1726 "traffic classification disabled!\n");
1731 /* Invalidated prio to tc mappings set to TC0 */
1732 for (i = 1; i < TC_BITMASK + 1; i++) {
1733 int q = netdev_get_prio_tc_map(dev, i);
1735 tc = &dev->tc_to_txq[q];
1736 if (tc->offset + tc->count > txq) {
1737 pr_warning("Number of in use tx queues "
1738 "changed. Priority %i to tc "
1739 "mapping %i is no longer valid "
1740 "setting map to 0\n",
1742 netdev_set_prio_tc_map(dev, i, 0);
1748 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1749 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1751 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1756 disabling = txq < dev->real_num_tx_queues;
1758 if (txq < 1 || txq > dev->num_tx_queues)
1761 if (dev->reg_state == NETREG_REGISTERED ||
1762 dev->reg_state == NETREG_UNREGISTERING) {
1765 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1771 netif_setup_tc(dev, txq);
1773 dev->real_num_tx_queues = txq;
1777 qdisc_reset_all_tx_gt(dev, txq);
1780 dev->real_num_tx_queues = txq;
1785 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1789 * netif_set_real_num_rx_queues - set actual number of RX queues used
1790 * @dev: Network device
1791 * @rxq: Actual number of RX queues
1793 * This must be called either with the rtnl_lock held or before
1794 * registration of the net device. Returns 0 on success, or a
1795 * negative error code. If called before registration, it always
1798 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1802 if (rxq < 1 || rxq > dev->num_rx_queues)
1805 if (dev->reg_state == NETREG_REGISTERED) {
1808 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1814 dev->real_num_rx_queues = rxq;
1817 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1820 static inline void __netif_reschedule(struct Qdisc *q)
1822 struct softnet_data *sd;
1823 unsigned long flags;
1825 local_irq_save(flags);
1826 sd = &__get_cpu_var(softnet_data);
1827 q->next_sched = NULL;
1828 *sd->output_queue_tailp = q;
1829 sd->output_queue_tailp = &q->next_sched;
1830 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1831 local_irq_restore(flags);
1834 void __netif_schedule(struct Qdisc *q)
1836 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1837 __netif_reschedule(q);
1839 EXPORT_SYMBOL(__netif_schedule);
1841 void dev_kfree_skb_irq(struct sk_buff *skb)
1843 if (atomic_dec_and_test(&skb->users)) {
1844 struct softnet_data *sd;
1845 unsigned long flags;
1847 local_irq_save(flags);
1848 sd = &__get_cpu_var(softnet_data);
1849 skb->next = sd->completion_queue;
1850 sd->completion_queue = skb;
1851 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1852 local_irq_restore(flags);
1855 EXPORT_SYMBOL(dev_kfree_skb_irq);
1857 void dev_kfree_skb_any(struct sk_buff *skb)
1859 if (in_irq() || irqs_disabled())
1860 dev_kfree_skb_irq(skb);
1864 EXPORT_SYMBOL(dev_kfree_skb_any);
1868 * netif_device_detach - mark device as removed
1869 * @dev: network device
1871 * Mark device as removed from system and therefore no longer available.
1873 void netif_device_detach(struct net_device *dev)
1875 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1876 netif_running(dev)) {
1877 netif_tx_stop_all_queues(dev);
1880 EXPORT_SYMBOL(netif_device_detach);
1883 * netif_device_attach - mark device as attached
1884 * @dev: network device
1886 * Mark device as attached from system and restart if needed.
1888 void netif_device_attach(struct net_device *dev)
1890 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1891 netif_running(dev)) {
1892 netif_tx_wake_all_queues(dev);
1893 __netdev_watchdog_up(dev);
1896 EXPORT_SYMBOL(netif_device_attach);
1899 * Invalidate hardware checksum when packet is to be mangled, and
1900 * complete checksum manually on outgoing path.
1902 int skb_checksum_help(struct sk_buff *skb)
1905 int ret = 0, offset;
1907 if (skb->ip_summed == CHECKSUM_COMPLETE)
1908 goto out_set_summed;
1910 if (unlikely(skb_shinfo(skb)->gso_size)) {
1911 /* Let GSO fix up the checksum. */
1912 goto out_set_summed;
1915 offset = skb_checksum_start_offset(skb);
1916 BUG_ON(offset >= skb_headlen(skb));
1917 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1919 offset += skb->csum_offset;
1920 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1922 if (skb_cloned(skb) &&
1923 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1924 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1929 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
1931 skb->ip_summed = CHECKSUM_NONE;
1935 EXPORT_SYMBOL(skb_checksum_help);
1938 * skb_gso_segment - Perform segmentation on skb.
1939 * @skb: buffer to segment
1940 * @features: features for the output path (see dev->features)
1942 * This function segments the given skb and returns a list of segments.
1944 * It may return NULL if the skb requires no segmentation. This is
1945 * only possible when GSO is used for verifying header integrity.
1947 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1949 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1950 struct packet_type *ptype;
1951 __be16 type = skb->protocol;
1952 int vlan_depth = ETH_HLEN;
1955 while (type == htons(ETH_P_8021Q)) {
1956 struct vlan_hdr *vh;
1958 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1959 return ERR_PTR(-EINVAL);
1961 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1962 type = vh->h_vlan_encapsulated_proto;
1963 vlan_depth += VLAN_HLEN;
1966 skb_reset_mac_header(skb);
1967 skb->mac_len = skb->network_header - skb->mac_header;
1968 __skb_pull(skb, skb->mac_len);
1970 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1971 struct net_device *dev = skb->dev;
1972 struct ethtool_drvinfo info = {};
1974 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1975 dev->ethtool_ops->get_drvinfo(dev, &info);
1977 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1978 info.driver, dev ? dev->features : 0L,
1979 skb->sk ? skb->sk->sk_route_caps : 0L,
1980 skb->len, skb->data_len, skb->ip_summed);
1982 if (skb_header_cloned(skb) &&
1983 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1984 return ERR_PTR(err);
1988 list_for_each_entry_rcu(ptype,
1989 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1990 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1991 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1992 err = ptype->gso_send_check(skb);
1993 segs = ERR_PTR(err);
1994 if (err || skb_gso_ok(skb, features))
1996 __skb_push(skb, (skb->data -
1997 skb_network_header(skb)));
1999 segs = ptype->gso_segment(skb, features);
2005 __skb_push(skb, skb->data - skb_mac_header(skb));
2009 EXPORT_SYMBOL(skb_gso_segment);
2011 /* Take action when hardware reception checksum errors are detected. */
2013 void netdev_rx_csum_fault(struct net_device *dev)
2015 if (net_ratelimit()) {
2016 printk(KERN_ERR "%s: hw csum failure.\n",
2017 dev ? dev->name : "<unknown>");
2021 EXPORT_SYMBOL(netdev_rx_csum_fault);
2024 /* Actually, we should eliminate this check as soon as we know, that:
2025 * 1. IOMMU is present and allows to map all the memory.
2026 * 2. No high memory really exists on this machine.
2029 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2031 #ifdef CONFIG_HIGHMEM
2033 if (!(dev->features & NETIF_F_HIGHDMA)) {
2034 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2035 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2036 if (PageHighMem(skb_frag_page(frag)))
2041 if (PCI_DMA_BUS_IS_PHYS) {
2042 struct device *pdev = dev->dev.parent;
2046 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2047 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2048 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2049 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2058 void (*destructor)(struct sk_buff *skb);
2061 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2063 static void dev_gso_skb_destructor(struct sk_buff *skb)
2065 struct dev_gso_cb *cb;
2068 struct sk_buff *nskb = skb->next;
2070 skb->next = nskb->next;
2073 } while (skb->next);
2075 cb = DEV_GSO_CB(skb);
2077 cb->destructor(skb);
2081 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2082 * @skb: buffer to segment
2083 * @features: device features as applicable to this skb
2085 * This function segments the given skb and stores the list of segments
2088 static int dev_gso_segment(struct sk_buff *skb, int features)
2090 struct sk_buff *segs;
2092 segs = skb_gso_segment(skb, features);
2094 /* Verifying header integrity only. */
2099 return PTR_ERR(segs);
2102 DEV_GSO_CB(skb)->destructor = skb->destructor;
2103 skb->destructor = dev_gso_skb_destructor;
2108 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2110 return ((features & NETIF_F_GEN_CSUM) ||
2111 ((features & NETIF_F_V4_CSUM) &&
2112 protocol == htons(ETH_P_IP)) ||
2113 ((features & NETIF_F_V6_CSUM) &&
2114 protocol == htons(ETH_P_IPV6)) ||
2115 ((features & NETIF_F_FCOE_CRC) &&
2116 protocol == htons(ETH_P_FCOE)));
2119 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2121 if (skb->ip_summed != CHECKSUM_NONE &&
2122 !can_checksum_protocol(features, protocol)) {
2123 features &= ~NETIF_F_ALL_CSUM;
2124 features &= ~NETIF_F_SG;
2125 } else if (illegal_highdma(skb->dev, skb)) {
2126 features &= ~NETIF_F_SG;
2132 u32 netif_skb_features(struct sk_buff *skb)
2134 __be16 protocol = skb->protocol;
2135 u32 features = skb->dev->features;
2137 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2138 features &= ~NETIF_F_GSO_MASK;
2140 if (!vlan_tx_tag_present(skb)) {
2141 if (unlikely(protocol == htons(ETH_P_8021Q))) {
2142 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2143 protocol = veh->h_vlan_encapsulated_proto;
2145 return harmonize_features(skb, protocol, features);
2149 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2151 if (protocol != htons(ETH_P_8021Q)) {
2152 return harmonize_features(skb, protocol, features);
2154 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2155 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2156 return harmonize_features(skb, protocol, features);
2159 EXPORT_SYMBOL(netif_skb_features);
2162 * Returns true if either:
2163 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2164 * 2. skb is fragmented and the device does not support SG, or if
2165 * at least one of fragments is in highmem and device does not
2166 * support DMA from it.
2168 static inline int skb_needs_linearize(struct sk_buff *skb,
2171 return skb_is_nonlinear(skb) &&
2172 ((skb_has_frag_list(skb) &&
2173 !(features & NETIF_F_FRAGLIST)) ||
2174 (skb_shinfo(skb)->nr_frags &&
2175 !(features & NETIF_F_SG)));
2178 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2179 struct netdev_queue *txq)
2181 const struct net_device_ops *ops = dev->netdev_ops;
2182 int rc = NETDEV_TX_OK;
2183 unsigned int skb_len;
2185 if (likely(!skb->next)) {
2189 * If device doesn't need skb->dst, release it right now while
2190 * its hot in this cpu cache
2192 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2195 if (!list_empty(&ptype_all))
2196 dev_queue_xmit_nit(skb, dev);
2198 features = netif_skb_features(skb);
2200 if (vlan_tx_tag_present(skb) &&
2201 !(features & NETIF_F_HW_VLAN_TX)) {
2202 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2209 if (netif_needs_gso(skb, features)) {
2210 if (unlikely(dev_gso_segment(skb, features)))
2215 if (skb_needs_linearize(skb, features) &&
2216 __skb_linearize(skb))
2219 /* If packet is not checksummed and device does not
2220 * support checksumming for this protocol, complete
2221 * checksumming here.
2223 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2224 skb_set_transport_header(skb,
2225 skb_checksum_start_offset(skb));
2226 if (!(features & NETIF_F_ALL_CSUM) &&
2227 skb_checksum_help(skb))
2233 rc = ops->ndo_start_xmit(skb, dev);
2234 trace_net_dev_xmit(skb, rc, dev, skb_len);
2235 if (rc == NETDEV_TX_OK)
2236 txq_trans_update(txq);
2242 struct sk_buff *nskb = skb->next;
2244 skb->next = nskb->next;
2248 * If device doesn't need nskb->dst, release it right now while
2249 * its hot in this cpu cache
2251 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2254 skb_len = nskb->len;
2255 rc = ops->ndo_start_xmit(nskb, dev);
2256 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2257 if (unlikely(rc != NETDEV_TX_OK)) {
2258 if (rc & ~NETDEV_TX_MASK)
2259 goto out_kfree_gso_skb;
2260 nskb->next = skb->next;
2264 txq_trans_update(txq);
2265 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2266 return NETDEV_TX_BUSY;
2267 } while (skb->next);
2270 if (likely(skb->next == NULL))
2271 skb->destructor = DEV_GSO_CB(skb)->destructor;
2278 static u32 hashrnd __read_mostly;
2281 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2282 * to be used as a distribution range.
2284 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2285 unsigned int num_tx_queues)
2289 u16 qcount = num_tx_queues;
2291 if (skb_rx_queue_recorded(skb)) {
2292 hash = skb_get_rx_queue(skb);
2293 while (unlikely(hash >= num_tx_queues))
2294 hash -= num_tx_queues;
2299 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2300 qoffset = dev->tc_to_txq[tc].offset;
2301 qcount = dev->tc_to_txq[tc].count;
2304 if (skb->sk && skb->sk->sk_hash)
2305 hash = skb->sk->sk_hash;
2307 hash = (__force u16) skb->protocol;
2308 hash = jhash_1word(hash, hashrnd);
2310 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2312 EXPORT_SYMBOL(__skb_tx_hash);
2314 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2316 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2317 if (net_ratelimit()) {
2318 pr_warning("%s selects TX queue %d, but "
2319 "real number of TX queues is %d\n",
2320 dev->name, queue_index, dev->real_num_tx_queues);
2327 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2330 struct xps_dev_maps *dev_maps;
2331 struct xps_map *map;
2332 int queue_index = -1;
2335 dev_maps = rcu_dereference(dev->xps_maps);
2337 map = rcu_dereference(
2338 dev_maps->cpu_map[raw_smp_processor_id()]);
2341 queue_index = map->queues[0];
2344 if (skb->sk && skb->sk->sk_hash)
2345 hash = skb->sk->sk_hash;
2347 hash = (__force u16) skb->protocol ^
2349 hash = jhash_1word(hash, hashrnd);
2350 queue_index = map->queues[
2351 ((u64)hash * map->len) >> 32];
2353 if (unlikely(queue_index >= dev->real_num_tx_queues))
2365 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2366 struct sk_buff *skb)
2369 const struct net_device_ops *ops = dev->netdev_ops;
2371 if (dev->real_num_tx_queues == 1)
2373 else if (ops->ndo_select_queue) {
2374 queue_index = ops->ndo_select_queue(dev, skb);
2375 queue_index = dev_cap_txqueue(dev, queue_index);
2377 struct sock *sk = skb->sk;
2378 queue_index = sk_tx_queue_get(sk);
2380 if (queue_index < 0 || skb->ooo_okay ||
2381 queue_index >= dev->real_num_tx_queues) {
2382 int old_index = queue_index;
2384 queue_index = get_xps_queue(dev, skb);
2385 if (queue_index < 0)
2386 queue_index = skb_tx_hash(dev, skb);
2388 if (queue_index != old_index && sk) {
2389 struct dst_entry *dst =
2390 rcu_dereference_check(sk->sk_dst_cache, 1);
2392 if (dst && skb_dst(skb) == dst)
2393 sk_tx_queue_set(sk, queue_index);
2398 skb_set_queue_mapping(skb, queue_index);
2399 return netdev_get_tx_queue(dev, queue_index);
2402 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2403 struct net_device *dev,
2404 struct netdev_queue *txq)
2406 spinlock_t *root_lock = qdisc_lock(q);
2410 qdisc_skb_cb(skb)->pkt_len = skb->len;
2411 qdisc_calculate_pkt_len(skb, q);
2413 * Heuristic to force contended enqueues to serialize on a
2414 * separate lock before trying to get qdisc main lock.
2415 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2416 * and dequeue packets faster.
2418 contended = qdisc_is_running(q);
2419 if (unlikely(contended))
2420 spin_lock(&q->busylock);
2422 spin_lock(root_lock);
2423 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2426 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2427 qdisc_run_begin(q)) {
2429 * This is a work-conserving queue; there are no old skbs
2430 * waiting to be sent out; and the qdisc is not running -
2431 * xmit the skb directly.
2433 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2436 qdisc_bstats_update(q, skb);
2438 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2439 if (unlikely(contended)) {
2440 spin_unlock(&q->busylock);
2447 rc = NET_XMIT_SUCCESS;
2450 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2451 if (qdisc_run_begin(q)) {
2452 if (unlikely(contended)) {
2453 spin_unlock(&q->busylock);
2459 spin_unlock(root_lock);
2460 if (unlikely(contended))
2461 spin_unlock(&q->busylock);
2465 static DEFINE_PER_CPU(int, xmit_recursion);
2466 #define RECURSION_LIMIT 10
2469 * dev_queue_xmit - transmit a buffer
2470 * @skb: buffer to transmit
2472 * Queue a buffer for transmission to a network device. The caller must
2473 * have set the device and priority and built the buffer before calling
2474 * this function. The function can be called from an interrupt.
2476 * A negative errno code is returned on a failure. A success does not
2477 * guarantee the frame will be transmitted as it may be dropped due
2478 * to congestion or traffic shaping.
2480 * -----------------------------------------------------------------------------------
2481 * I notice this method can also return errors from the queue disciplines,
2482 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2485 * Regardless of the return value, the skb is consumed, so it is currently
2486 * difficult to retry a send to this method. (You can bump the ref count
2487 * before sending to hold a reference for retry if you are careful.)
2489 * When calling this method, interrupts MUST be enabled. This is because
2490 * the BH enable code must have IRQs enabled so that it will not deadlock.
2493 int dev_queue_xmit(struct sk_buff *skb)
2495 struct net_device *dev = skb->dev;
2496 struct netdev_queue *txq;
2500 /* Disable soft irqs for various locks below. Also
2501 * stops preemption for RCU.
2505 txq = dev_pick_tx(dev, skb);
2506 q = rcu_dereference_bh(txq->qdisc);
2508 #ifdef CONFIG_NET_CLS_ACT
2509 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2511 trace_net_dev_queue(skb);
2513 rc = __dev_xmit_skb(skb, q, dev, txq);
2517 /* The device has no queue. Common case for software devices:
2518 loopback, all the sorts of tunnels...
2520 Really, it is unlikely that netif_tx_lock protection is necessary
2521 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2523 However, it is possible, that they rely on protection
2526 Check this and shot the lock. It is not prone from deadlocks.
2527 Either shot noqueue qdisc, it is even simpler 8)
2529 if (dev->flags & IFF_UP) {
2530 int cpu = smp_processor_id(); /* ok because BHs are off */
2532 if (txq->xmit_lock_owner != cpu) {
2534 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2535 goto recursion_alert;
2537 HARD_TX_LOCK(dev, txq, cpu);
2539 if (!netif_tx_queue_stopped(txq)) {
2540 __this_cpu_inc(xmit_recursion);
2541 rc = dev_hard_start_xmit(skb, dev, txq);
2542 __this_cpu_dec(xmit_recursion);
2543 if (dev_xmit_complete(rc)) {
2544 HARD_TX_UNLOCK(dev, txq);
2548 HARD_TX_UNLOCK(dev, txq);
2549 if (net_ratelimit())
2550 printk(KERN_CRIT "Virtual device %s asks to "
2551 "queue packet!\n", dev->name);
2553 /* Recursion is detected! It is possible,
2557 if (net_ratelimit())
2558 printk(KERN_CRIT "Dead loop on virtual device "
2559 "%s, fix it urgently!\n", dev->name);
2564 rcu_read_unlock_bh();
2569 rcu_read_unlock_bh();
2572 EXPORT_SYMBOL(dev_queue_xmit);
2575 /*=======================================================================
2577 =======================================================================*/
2579 int netdev_max_backlog __read_mostly = 1000;
2580 int netdev_tstamp_prequeue __read_mostly = 1;
2581 int netdev_budget __read_mostly = 300;
2582 int weight_p __read_mostly = 64; /* old backlog weight */
2584 /* Called with irq disabled */
2585 static inline void ____napi_schedule(struct softnet_data *sd,
2586 struct napi_struct *napi)
2588 list_add_tail(&napi->poll_list, &sd->poll_list);
2589 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2593 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2594 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2595 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2596 * if hash is a canonical 4-tuple hash over transport ports.
2598 void __skb_get_rxhash(struct sk_buff *skb)
2600 int nhoff, hash = 0, poff;
2601 const struct ipv6hdr *ip6;
2602 const struct iphdr *ip;
2603 const struct vlan_hdr *vlan;
2612 nhoff = skb_network_offset(skb);
2613 proto = skb->protocol;
2617 case __constant_htons(ETH_P_IP):
2619 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2622 ip = (const struct iphdr *) (skb->data + nhoff);
2625 if (ip_is_fragment(ip))
2628 ip_proto = ip->protocol;
2629 addr1 = (__force u32) ip->saddr;
2630 addr2 = (__force u32) ip->daddr;
2631 nhoff += ip->ihl * 4;
2633 case __constant_htons(ETH_P_IPV6):
2635 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2638 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2639 ip_proto = ip6->nexthdr;
2640 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2641 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2644 case __constant_htons(ETH_P_8021Q):
2645 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2647 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2648 proto = vlan->h_vlan_encapsulated_proto;
2649 nhoff += sizeof(*vlan);
2651 case __constant_htons(ETH_P_PPP_SES):
2652 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2654 proto = *((__be16 *) (skb->data + nhoff +
2655 sizeof(struct pppoe_hdr)));
2656 nhoff += PPPOE_SES_HLEN;
2658 case __constant_htons(PPP_IP):
2660 case __constant_htons(PPP_IPV6):
2671 if (pskb_may_pull(skb, nhoff + 16)) {
2672 u8 *h = skb->data + nhoff;
2673 __be16 flags = *(__be16 *)h;
2676 * Only look inside GRE if version zero and no
2679 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2680 proto = *(__be16 *)(h + 2);
2682 if (flags & GRE_CSUM)
2684 if (flags & GRE_KEY)
2686 if (flags & GRE_SEQ)
2699 poff = proto_ports_offset(ip_proto);
2702 if (pskb_may_pull(skb, nhoff + 4)) {
2703 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2708 /* get a consistent hash (same value on both flow directions) */
2709 if (addr2 < addr1 ||
2711 ports.v16[1] < ports.v16[0])) {
2713 swap(ports.v16[0], ports.v16[1]);
2715 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2722 EXPORT_SYMBOL(__skb_get_rxhash);
2726 /* One global table that all flow-based protocols share. */
2727 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2728 EXPORT_SYMBOL(rps_sock_flow_table);
2730 static struct rps_dev_flow *
2731 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2732 struct rps_dev_flow *rflow, u16 next_cpu)
2734 if (next_cpu != RPS_NO_CPU) {
2735 #ifdef CONFIG_RFS_ACCEL
2736 struct netdev_rx_queue *rxqueue;
2737 struct rps_dev_flow_table *flow_table;
2738 struct rps_dev_flow *old_rflow;
2743 /* Should we steer this flow to a different hardware queue? */
2744 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2745 !(dev->features & NETIF_F_NTUPLE))
2747 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2748 if (rxq_index == skb_get_rx_queue(skb))
2751 rxqueue = dev->_rx + rxq_index;
2752 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2755 flow_id = skb->rxhash & flow_table->mask;
2756 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2757 rxq_index, flow_id);
2761 rflow = &flow_table->flows[flow_id];
2763 if (old_rflow->filter == rflow->filter)
2764 old_rflow->filter = RPS_NO_FILTER;
2768 per_cpu(softnet_data, next_cpu).input_queue_head;
2771 rflow->cpu = next_cpu;
2776 * get_rps_cpu is called from netif_receive_skb and returns the target
2777 * CPU from the RPS map of the receiving queue for a given skb.
2778 * rcu_read_lock must be held on entry.
2780 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2781 struct rps_dev_flow **rflowp)
2783 struct netdev_rx_queue *rxqueue;
2784 struct rps_map *map;
2785 struct rps_dev_flow_table *flow_table;
2786 struct rps_sock_flow_table *sock_flow_table;
2790 if (skb_rx_queue_recorded(skb)) {
2791 u16 index = skb_get_rx_queue(skb);
2792 if (unlikely(index >= dev->real_num_rx_queues)) {
2793 WARN_ONCE(dev->real_num_rx_queues > 1,
2794 "%s received packet on queue %u, but number "
2795 "of RX queues is %u\n",
2796 dev->name, index, dev->real_num_rx_queues);
2799 rxqueue = dev->_rx + index;
2803 map = rcu_dereference(rxqueue->rps_map);
2805 if (map->len == 1 &&
2806 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2807 tcpu = map->cpus[0];
2808 if (cpu_online(tcpu))
2812 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2816 skb_reset_network_header(skb);
2817 if (!skb_get_rxhash(skb))
2820 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2821 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2822 if (flow_table && sock_flow_table) {
2824 struct rps_dev_flow *rflow;
2826 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2829 next_cpu = sock_flow_table->ents[skb->rxhash &
2830 sock_flow_table->mask];
2833 * If the desired CPU (where last recvmsg was done) is
2834 * different from current CPU (one in the rx-queue flow
2835 * table entry), switch if one of the following holds:
2836 * - Current CPU is unset (equal to RPS_NO_CPU).
2837 * - Current CPU is offline.
2838 * - The current CPU's queue tail has advanced beyond the
2839 * last packet that was enqueued using this table entry.
2840 * This guarantees that all previous packets for the flow
2841 * have been dequeued, thus preserving in order delivery.
2843 if (unlikely(tcpu != next_cpu) &&
2844 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2845 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2846 rflow->last_qtail)) >= 0)) {
2848 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2851 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2859 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2861 if (cpu_online(tcpu)) {
2871 #ifdef CONFIG_RFS_ACCEL
2874 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2875 * @dev: Device on which the filter was set
2876 * @rxq_index: RX queue index
2877 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2878 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2880 * Drivers that implement ndo_rx_flow_steer() should periodically call
2881 * this function for each installed filter and remove the filters for
2882 * which it returns %true.
2884 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2885 u32 flow_id, u16 filter_id)
2887 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2888 struct rps_dev_flow_table *flow_table;
2889 struct rps_dev_flow *rflow;
2894 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2895 if (flow_table && flow_id <= flow_table->mask) {
2896 rflow = &flow_table->flows[flow_id];
2897 cpu = ACCESS_ONCE(rflow->cpu);
2898 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2899 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2900 rflow->last_qtail) <
2901 (int)(10 * flow_table->mask)))
2907 EXPORT_SYMBOL(rps_may_expire_flow);
2909 #endif /* CONFIG_RFS_ACCEL */
2911 /* Called from hardirq (IPI) context */
2912 static void rps_trigger_softirq(void *data)
2914 struct softnet_data *sd = data;
2916 ____napi_schedule(sd, &sd->backlog);
2920 #endif /* CONFIG_RPS */
2923 * Check if this softnet_data structure is another cpu one
2924 * If yes, queue it to our IPI list and return 1
2927 static int rps_ipi_queued(struct softnet_data *sd)
2930 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2933 sd->rps_ipi_next = mysd->rps_ipi_list;
2934 mysd->rps_ipi_list = sd;
2936 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2939 #endif /* CONFIG_RPS */
2944 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2945 * queue (may be a remote CPU queue).
2947 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2948 unsigned int *qtail)
2950 struct softnet_data *sd;
2951 unsigned long flags;
2953 sd = &per_cpu(softnet_data, cpu);
2955 local_irq_save(flags);
2958 if (!netif_running(skb->dev))
2960 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2961 if (skb_queue_len(&sd->input_pkt_queue)) {
2963 __skb_queue_tail(&sd->input_pkt_queue, skb);
2964 input_queue_tail_incr_save(sd, qtail);
2966 local_irq_restore(flags);
2967 return NET_RX_SUCCESS;
2970 /* Schedule NAPI for backlog device
2971 * We can use non atomic operation since we own the queue lock
2973 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2974 if (!rps_ipi_queued(sd))
2975 ____napi_schedule(sd, &sd->backlog);
2984 local_irq_restore(flags);
2986 atomic_long_inc(&skb->dev->rx_dropped);
2992 * netif_rx - post buffer to the network code
2993 * @skb: buffer to post
2995 * This function receives a packet from a device driver and queues it for
2996 * the upper (protocol) levels to process. It always succeeds. The buffer
2997 * may be dropped during processing for congestion control or by the
3001 * NET_RX_SUCCESS (no congestion)
3002 * NET_RX_DROP (packet was dropped)
3006 int netif_rx(struct sk_buff *skb)
3010 /* if netpoll wants it, pretend we never saw it */
3011 if (netpoll_rx(skb))
3014 if (netdev_tstamp_prequeue)
3015 net_timestamp_check(skb);
3017 trace_netif_rx(skb);
3020 struct rps_dev_flow voidflow, *rflow = &voidflow;
3026 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3028 cpu = smp_processor_id();
3030 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3038 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3044 EXPORT_SYMBOL(netif_rx);
3046 int netif_rx_ni(struct sk_buff *skb)
3051 err = netif_rx(skb);
3052 if (local_softirq_pending())
3058 EXPORT_SYMBOL(netif_rx_ni);
3060 static void net_tx_action(struct softirq_action *h)
3062 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3064 if (sd->completion_queue) {
3065 struct sk_buff *clist;
3067 local_irq_disable();
3068 clist = sd->completion_queue;
3069 sd->completion_queue = NULL;
3073 struct sk_buff *skb = clist;
3074 clist = clist->next;
3076 WARN_ON(atomic_read(&skb->users));
3077 trace_kfree_skb(skb, net_tx_action);
3082 if (sd->output_queue) {
3085 local_irq_disable();
3086 head = sd->output_queue;
3087 sd->output_queue = NULL;
3088 sd->output_queue_tailp = &sd->output_queue;
3092 struct Qdisc *q = head;
3093 spinlock_t *root_lock;
3095 head = head->next_sched;
3097 root_lock = qdisc_lock(q);
3098 if (spin_trylock(root_lock)) {
3099 smp_mb__before_clear_bit();
3100 clear_bit(__QDISC_STATE_SCHED,
3103 spin_unlock(root_lock);
3105 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3107 __netif_reschedule(q);
3109 smp_mb__before_clear_bit();
3110 clear_bit(__QDISC_STATE_SCHED,
3118 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3119 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3120 /* This hook is defined here for ATM LANE */
3121 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3122 unsigned char *addr) __read_mostly;
3123 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3126 #ifdef CONFIG_NET_CLS_ACT
3127 /* TODO: Maybe we should just force sch_ingress to be compiled in
3128 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3129 * a compare and 2 stores extra right now if we dont have it on
3130 * but have CONFIG_NET_CLS_ACT
3131 * NOTE: This doesn't stop any functionality; if you dont have
3132 * the ingress scheduler, you just can't add policies on ingress.
3135 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3137 struct net_device *dev = skb->dev;
3138 u32 ttl = G_TC_RTTL(skb->tc_verd);
3139 int result = TC_ACT_OK;
3142 if (unlikely(MAX_RED_LOOP < ttl++)) {
3143 if (net_ratelimit())
3144 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3145 skb->skb_iif, dev->ifindex);
3149 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3150 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3153 if (q != &noop_qdisc) {
3154 spin_lock(qdisc_lock(q));
3155 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3156 result = qdisc_enqueue_root(skb, q);
3157 spin_unlock(qdisc_lock(q));
3163 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3164 struct packet_type **pt_prev,
3165 int *ret, struct net_device *orig_dev)
3167 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3169 if (!rxq || rxq->qdisc == &noop_qdisc)
3173 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3177 switch (ing_filter(skb, rxq)) {
3191 * netdev_rx_handler_register - register receive handler
3192 * @dev: device to register a handler for
3193 * @rx_handler: receive handler to register
3194 * @rx_handler_data: data pointer that is used by rx handler
3196 * Register a receive hander for a device. This handler will then be
3197 * called from __netif_receive_skb. A negative errno code is returned
3200 * The caller must hold the rtnl_mutex.
3202 * For a general description of rx_handler, see enum rx_handler_result.
3204 int netdev_rx_handler_register(struct net_device *dev,
3205 rx_handler_func_t *rx_handler,
3206 void *rx_handler_data)
3210 if (dev->rx_handler)
3213 /* Note: rx_handler_data must be set before rx_handler */
3214 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3215 rcu_assign_pointer(dev->rx_handler, rx_handler);
3219 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3222 * netdev_rx_handler_unregister - unregister receive handler
3223 * @dev: device to unregister a handler from
3225 * Unregister a receive hander from a device.
3227 * The caller must hold the rtnl_mutex.
3229 void netdev_rx_handler_unregister(struct net_device *dev)
3233 RCU_INIT_POINTER(dev->rx_handler, NULL);
3234 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3235 * section has a guarantee to see a non NULL rx_handler_data
3239 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3241 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3243 static int __netif_receive_skb(struct sk_buff *skb)
3245 struct packet_type *ptype, *pt_prev;
3246 rx_handler_func_t *rx_handler;
3247 struct net_device *orig_dev;
3248 struct net_device *null_or_dev;
3249 bool deliver_exact = false;
3250 int ret = NET_RX_DROP;
3253 if (!netdev_tstamp_prequeue)
3254 net_timestamp_check(skb);
3256 trace_netif_receive_skb(skb);
3258 /* if we've gotten here through NAPI, check netpoll */
3259 if (netpoll_receive_skb(skb))
3263 skb->skb_iif = skb->dev->ifindex;
3264 orig_dev = skb->dev;
3266 skb_reset_network_header(skb);
3267 skb_reset_transport_header(skb);
3268 skb_reset_mac_len(skb);
3274 __this_cpu_inc(softnet_data.processed);
3276 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3277 skb = vlan_untag(skb);
3282 #ifdef CONFIG_NET_CLS_ACT
3283 if (skb->tc_verd & TC_NCLS) {
3284 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3289 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3290 if (!ptype->dev || ptype->dev == skb->dev) {
3292 ret = deliver_skb(skb, pt_prev, orig_dev);
3297 #ifdef CONFIG_NET_CLS_ACT
3298 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3304 if (vlan_tx_tag_present(skb)) {
3306 ret = deliver_skb(skb, pt_prev, orig_dev);
3309 if (vlan_do_receive(&skb))
3311 else if (unlikely(!skb))
3315 rx_handler = rcu_dereference(skb->dev->rx_handler);
3318 ret = deliver_skb(skb, pt_prev, orig_dev);
3321 switch (rx_handler(&skb)) {
3322 case RX_HANDLER_CONSUMED:
3323 ret = NET_RX_SUCCESS;
3325 case RX_HANDLER_ANOTHER:
3327 case RX_HANDLER_EXACT:
3328 deliver_exact = true;
3329 case RX_HANDLER_PASS:
3336 if (vlan_tx_nonzero_tag_present(skb))
3337 skb->pkt_type = PACKET_OTHERHOST;
3339 /* deliver only exact match when indicated */
3340 null_or_dev = deliver_exact ? skb->dev : NULL;
3342 type = skb->protocol;
3343 list_for_each_entry_rcu(ptype,
3344 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3345 if (ptype->type == type &&
3346 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3347 ptype->dev == orig_dev)) {
3349 ret = deliver_skb(skb, pt_prev, orig_dev);
3355 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3357 atomic_long_inc(&skb->dev->rx_dropped);
3359 /* Jamal, now you will not able to escape explaining
3360 * me how you were going to use this. :-)
3370 * netif_receive_skb - process receive buffer from network
3371 * @skb: buffer to process
3373 * netif_receive_skb() is the main receive data processing function.
3374 * It always succeeds. The buffer may be dropped during processing
3375 * for congestion control or by the protocol layers.
3377 * This function may only be called from softirq context and interrupts
3378 * should be enabled.
3380 * Return values (usually ignored):
3381 * NET_RX_SUCCESS: no congestion
3382 * NET_RX_DROP: packet was dropped
3384 int netif_receive_skb(struct sk_buff *skb)
3388 if (netdev_tstamp_prequeue)
3389 net_timestamp_check(skb);
3391 if (skb_defer_rx_timestamp(skb))
3392 return NET_RX_SUCCESS;
3398 struct rps_dev_flow voidflow, *rflow = &voidflow;
3399 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
3402 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3408 ret = __netif_receive_skb(skb);
3412 EXPORT_SYMBOL(netif_receive_skb);
3414 /* Network device is going away, flush any packets still pending
3415 * Called with irqs disabled.
3417 static void flush_backlog(void *arg)
3419 struct net_device *dev = arg;
3420 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3421 struct sk_buff *skb, *tmp;
3424 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3425 if (skb->dev == dev) {
3426 __skb_unlink(skb, &sd->input_pkt_queue);
3428 input_queue_head_incr(sd);
3433 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3434 if (skb->dev == dev) {
3435 __skb_unlink(skb, &sd->process_queue);
3437 input_queue_head_incr(sd);
3442 static int napi_gro_complete(struct sk_buff *skb)
3444 struct packet_type *ptype;
3445 __be16 type = skb->protocol;
3446 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3449 if (NAPI_GRO_CB(skb)->count == 1) {
3450 skb_shinfo(skb)->gso_size = 0;
3455 list_for_each_entry_rcu(ptype, head, list) {
3456 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3459 err = ptype->gro_complete(skb);
3465 WARN_ON(&ptype->list == head);
3467 return NET_RX_SUCCESS;
3471 return netif_receive_skb(skb);
3474 inline void napi_gro_flush(struct napi_struct *napi)
3476 struct sk_buff *skb, *next;
3478 for (skb = napi->gro_list; skb; skb = next) {
3481 napi_gro_complete(skb);
3484 napi->gro_count = 0;
3485 napi->gro_list = NULL;
3487 EXPORT_SYMBOL(napi_gro_flush);
3489 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3491 struct sk_buff **pp = NULL;
3492 struct packet_type *ptype;
3493 __be16 type = skb->protocol;
3494 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3497 enum gro_result ret;
3499 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3502 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3506 list_for_each_entry_rcu(ptype, head, list) {
3507 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3510 skb_set_network_header(skb, skb_gro_offset(skb));
3511 mac_len = skb->network_header - skb->mac_header;
3512 skb->mac_len = mac_len;
3513 NAPI_GRO_CB(skb)->same_flow = 0;
3514 NAPI_GRO_CB(skb)->flush = 0;
3515 NAPI_GRO_CB(skb)->free = 0;
3517 pp = ptype->gro_receive(&napi->gro_list, skb);
3522 if (&ptype->list == head)
3525 same_flow = NAPI_GRO_CB(skb)->same_flow;
3526 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3529 struct sk_buff *nskb = *pp;
3533 napi_gro_complete(nskb);
3540 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3544 NAPI_GRO_CB(skb)->count = 1;
3545 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3546 skb->next = napi->gro_list;
3547 napi->gro_list = skb;
3551 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3552 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3554 BUG_ON(skb->end - skb->tail < grow);
3556 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3559 skb->data_len -= grow;
3561 skb_shinfo(skb)->frags[0].page_offset += grow;
3562 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3564 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3565 skb_frag_unref(skb, 0);
3566 memmove(skb_shinfo(skb)->frags,
3567 skb_shinfo(skb)->frags + 1,
3568 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3579 EXPORT_SYMBOL(dev_gro_receive);
3581 static inline gro_result_t
3582 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3585 unsigned int maclen = skb->dev->hard_header_len;
3587 for (p = napi->gro_list; p; p = p->next) {
3588 unsigned long diffs;
3590 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3591 diffs |= p->vlan_tci ^ skb->vlan_tci;
3592 if (maclen == ETH_HLEN)
3593 diffs |= compare_ether_header(skb_mac_header(p),
3594 skb_gro_mac_header(skb));
3596 diffs = memcmp(skb_mac_header(p),
3597 skb_gro_mac_header(skb),
3599 NAPI_GRO_CB(p)->same_flow = !diffs;
3600 NAPI_GRO_CB(p)->flush = 0;
3603 return dev_gro_receive(napi, skb);
3606 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3610 if (netif_receive_skb(skb))
3615 case GRO_MERGED_FREE:
3626 EXPORT_SYMBOL(napi_skb_finish);
3628 void skb_gro_reset_offset(struct sk_buff *skb)
3630 NAPI_GRO_CB(skb)->data_offset = 0;
3631 NAPI_GRO_CB(skb)->frag0 = NULL;
3632 NAPI_GRO_CB(skb)->frag0_len = 0;
3634 if (skb->mac_header == skb->tail &&
3635 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3636 NAPI_GRO_CB(skb)->frag0 =
3637 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3638 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
3639 skb_frag_size(&skb_shinfo(skb)->frags[0]),
3640 skb->end - skb->tail);
3643 EXPORT_SYMBOL(skb_gro_reset_offset);
3645 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3647 skb_gro_reset_offset(skb);
3649 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3651 EXPORT_SYMBOL(napi_gro_receive);
3653 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3655 __skb_pull(skb, skb_headlen(skb));
3656 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3657 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3659 skb->dev = napi->dev;
3661 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
3666 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3668 struct sk_buff *skb = napi->skb;
3671 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3677 EXPORT_SYMBOL(napi_get_frags);
3679 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3685 skb->protocol = eth_type_trans(skb, skb->dev);
3687 if (ret == GRO_HELD)
3688 skb_gro_pull(skb, -ETH_HLEN);
3689 else if (netif_receive_skb(skb))
3694 case GRO_MERGED_FREE:
3695 napi_reuse_skb(napi, skb);
3704 EXPORT_SYMBOL(napi_frags_finish);
3706 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3708 struct sk_buff *skb = napi->skb;
3715 skb_reset_mac_header(skb);
3716 skb_gro_reset_offset(skb);
3718 off = skb_gro_offset(skb);
3719 hlen = off + sizeof(*eth);
3720 eth = skb_gro_header_fast(skb, off);
3721 if (skb_gro_header_hard(skb, hlen)) {
3722 eth = skb_gro_header_slow(skb, hlen, off);
3723 if (unlikely(!eth)) {
3724 napi_reuse_skb(napi, skb);
3730 skb_gro_pull(skb, sizeof(*eth));
3733 * This works because the only protocols we care about don't require
3734 * special handling. We'll fix it up properly at the end.
3736 skb->protocol = eth->h_proto;
3741 EXPORT_SYMBOL(napi_frags_skb);
3743 gro_result_t napi_gro_frags(struct napi_struct *napi)
3745 struct sk_buff *skb = napi_frags_skb(napi);
3750 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3752 EXPORT_SYMBOL(napi_gro_frags);
3755 * net_rps_action sends any pending IPI's for rps.
3756 * Note: called with local irq disabled, but exits with local irq enabled.
3758 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3761 struct softnet_data *remsd = sd->rps_ipi_list;
3764 sd->rps_ipi_list = NULL;
3768 /* Send pending IPI's to kick RPS processing on remote cpus. */
3770 struct softnet_data *next = remsd->rps_ipi_next;
3772 if (cpu_online(remsd->cpu))
3773 __smp_call_function_single(remsd->cpu,
3782 static int process_backlog(struct napi_struct *napi, int quota)
3785 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3788 /* Check if we have pending ipi, its better to send them now,
3789 * not waiting net_rx_action() end.
3791 if (sd->rps_ipi_list) {
3792 local_irq_disable();
3793 net_rps_action_and_irq_enable(sd);
3796 napi->weight = weight_p;
3797 local_irq_disable();
3798 while (work < quota) {
3799 struct sk_buff *skb;
3802 while ((skb = __skb_dequeue(&sd->process_queue))) {
3805 __netif_receive_skb(skb);
3807 local_irq_disable();
3808 input_queue_head_incr(sd);
3809 if (++work >= quota) {
3816 qlen = skb_queue_len(&sd->input_pkt_queue);
3818 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3819 &sd->process_queue);
3821 if (qlen < quota - work) {
3823 * Inline a custom version of __napi_complete().
3824 * only current cpu owns and manipulates this napi,
3825 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3826 * we can use a plain write instead of clear_bit(),
3827 * and we dont need an smp_mb() memory barrier.
3829 list_del(&napi->poll_list);
3832 quota = work + qlen;
3842 * __napi_schedule - schedule for receive
3843 * @n: entry to schedule
3845 * The entry's receive function will be scheduled to run
3847 void __napi_schedule(struct napi_struct *n)
3849 unsigned long flags;
3851 local_irq_save(flags);
3852 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3853 local_irq_restore(flags);
3855 EXPORT_SYMBOL(__napi_schedule);
3857 void __napi_complete(struct napi_struct *n)
3859 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3860 BUG_ON(n->gro_list);
3862 list_del(&n->poll_list);
3863 smp_mb__before_clear_bit();
3864 clear_bit(NAPI_STATE_SCHED, &n->state);
3866 EXPORT_SYMBOL(__napi_complete);
3868 void napi_complete(struct napi_struct *n)
3870 unsigned long flags;
3873 * don't let napi dequeue from the cpu poll list
3874 * just in case its running on a different cpu
3876 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3880 local_irq_save(flags);
3882 local_irq_restore(flags);
3884 EXPORT_SYMBOL(napi_complete);
3886 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3887 int (*poll)(struct napi_struct *, int), int weight)
3889 INIT_LIST_HEAD(&napi->poll_list);
3890 napi->gro_count = 0;
3891 napi->gro_list = NULL;
3894 napi->weight = weight;
3895 list_add(&napi->dev_list, &dev->napi_list);
3897 #ifdef CONFIG_NETPOLL
3898 spin_lock_init(&napi->poll_lock);
3899 napi->poll_owner = -1;
3901 set_bit(NAPI_STATE_SCHED, &napi->state);
3903 EXPORT_SYMBOL(netif_napi_add);
3905 void netif_napi_del(struct napi_struct *napi)
3907 struct sk_buff *skb, *next;
3909 list_del_init(&napi->dev_list);
3910 napi_free_frags(napi);
3912 for (skb = napi->gro_list; skb; skb = next) {
3918 napi->gro_list = NULL;
3919 napi->gro_count = 0;
3921 EXPORT_SYMBOL(netif_napi_del);
3923 static void net_rx_action(struct softirq_action *h)
3925 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3926 unsigned long time_limit = jiffies + 2;
3927 int budget = netdev_budget;
3930 local_irq_disable();
3932 while (!list_empty(&sd->poll_list)) {
3933 struct napi_struct *n;
3936 /* If softirq window is exhuasted then punt.
3937 * Allow this to run for 2 jiffies since which will allow
3938 * an average latency of 1.5/HZ.
3940 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3945 /* Even though interrupts have been re-enabled, this
3946 * access is safe because interrupts can only add new
3947 * entries to the tail of this list, and only ->poll()
3948 * calls can remove this head entry from the list.
3950 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3952 have = netpoll_poll_lock(n);
3956 /* This NAPI_STATE_SCHED test is for avoiding a race
3957 * with netpoll's poll_napi(). Only the entity which
3958 * obtains the lock and sees NAPI_STATE_SCHED set will
3959 * actually make the ->poll() call. Therefore we avoid
3960 * accidentally calling ->poll() when NAPI is not scheduled.
3963 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3964 work = n->poll(n, weight);
3968 WARN_ON_ONCE(work > weight);
3972 local_irq_disable();
3974 /* Drivers must not modify the NAPI state if they
3975 * consume the entire weight. In such cases this code
3976 * still "owns" the NAPI instance and therefore can
3977 * move the instance around on the list at-will.
3979 if (unlikely(work == weight)) {
3980 if (unlikely(napi_disable_pending(n))) {
3983 local_irq_disable();
3985 list_move_tail(&n->poll_list, &sd->poll_list);
3988 netpoll_poll_unlock(have);
3991 net_rps_action_and_irq_enable(sd);
3993 #ifdef CONFIG_NET_DMA
3995 * There may not be any more sk_buffs coming right now, so push
3996 * any pending DMA copies to hardware
3998 dma_issue_pending_all();
4005 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4009 static gifconf_func_t *gifconf_list[NPROTO];
4012 * register_gifconf - register a SIOCGIF handler
4013 * @family: Address family
4014 * @gifconf: Function handler
4016 * Register protocol dependent address dumping routines. The handler
4017 * that is passed must not be freed or reused until it has been replaced
4018 * by another handler.
4020 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4022 if (family >= NPROTO)
4024 gifconf_list[family] = gifconf;
4027 EXPORT_SYMBOL(register_gifconf);
4031 * Map an interface index to its name (SIOCGIFNAME)
4035 * We need this ioctl for efficient implementation of the
4036 * if_indextoname() function required by the IPv6 API. Without
4037 * it, we would have to search all the interfaces to find a
4041 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4043 struct net_device *dev;
4047 * Fetch the caller's info block.
4050 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4054 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4060 strcpy(ifr.ifr_name, dev->name);
4063 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4069 * Perform a SIOCGIFCONF call. This structure will change
4070 * size eventually, and there is nothing I can do about it.
4071 * Thus we will need a 'compatibility mode'.
4074 static int dev_ifconf(struct net *net, char __user *arg)
4077 struct net_device *dev;
4084 * Fetch the caller's info block.
4087 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4094 * Loop over the interfaces, and write an info block for each.
4098 for_each_netdev(net, dev) {
4099 for (i = 0; i < NPROTO; i++) {
4100 if (gifconf_list[i]) {
4103 done = gifconf_list[i](dev, NULL, 0);
4105 done = gifconf_list[i](dev, pos + total,
4115 * All done. Write the updated control block back to the caller.
4117 ifc.ifc_len = total;
4120 * Both BSD and Solaris return 0 here, so we do too.
4122 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4125 #ifdef CONFIG_PROC_FS
4127 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4129 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4130 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4131 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4133 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4135 struct net *net = seq_file_net(seq);
4136 struct net_device *dev;
4137 struct hlist_node *p;
4138 struct hlist_head *h;
4139 unsigned int count = 0, offset = get_offset(*pos);
4141 h = &net->dev_name_head[get_bucket(*pos)];
4142 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4143 if (++count == offset)
4150 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4152 struct net_device *dev;
4153 unsigned int bucket;
4156 dev = dev_from_same_bucket(seq, pos);
4160 bucket = get_bucket(*pos) + 1;
4161 *pos = set_bucket_offset(bucket, 1);
4162 } while (bucket < NETDEV_HASHENTRIES);
4168 * This is invoked by the /proc filesystem handler to display a device
4171 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4176 return SEQ_START_TOKEN;
4178 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4181 return dev_from_bucket(seq, pos);
4184 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4187 return dev_from_bucket(seq, pos);
4190 void dev_seq_stop(struct seq_file *seq, void *v)
4196 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4198 struct rtnl_link_stats64 temp;
4199 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4201 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4202 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4203 dev->name, stats->rx_bytes, stats->rx_packets,
4205 stats->rx_dropped + stats->rx_missed_errors,
4206 stats->rx_fifo_errors,
4207 stats->rx_length_errors + stats->rx_over_errors +
4208 stats->rx_crc_errors + stats->rx_frame_errors,
4209 stats->rx_compressed, stats->multicast,
4210 stats->tx_bytes, stats->tx_packets,
4211 stats->tx_errors, stats->tx_dropped,
4212 stats->tx_fifo_errors, stats->collisions,
4213 stats->tx_carrier_errors +
4214 stats->tx_aborted_errors +
4215 stats->tx_window_errors +
4216 stats->tx_heartbeat_errors,
4217 stats->tx_compressed);
4221 * Called from the PROCfs module. This now uses the new arbitrary sized
4222 * /proc/net interface to create /proc/net/dev
4224 static int dev_seq_show(struct seq_file *seq, void *v)
4226 if (v == SEQ_START_TOKEN)
4227 seq_puts(seq, "Inter-| Receive "
4229 " face |bytes packets errs drop fifo frame "
4230 "compressed multicast|bytes packets errs "
4231 "drop fifo colls carrier compressed\n");
4233 dev_seq_printf_stats(seq, v);
4237 static struct softnet_data *softnet_get_online(loff_t *pos)
4239 struct softnet_data *sd = NULL;
4241 while (*pos < nr_cpu_ids)
4242 if (cpu_online(*pos)) {
4243 sd = &per_cpu(softnet_data, *pos);
4250 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4252 return softnet_get_online(pos);
4255 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4258 return softnet_get_online(pos);
4261 static void softnet_seq_stop(struct seq_file *seq, void *v)
4265 static int softnet_seq_show(struct seq_file *seq, void *v)
4267 struct softnet_data *sd = v;
4269 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4270 sd->processed, sd->dropped, sd->time_squeeze, 0,
4271 0, 0, 0, 0, /* was fastroute */
4272 sd->cpu_collision, sd->received_rps);
4276 static const struct seq_operations dev_seq_ops = {
4277 .start = dev_seq_start,
4278 .next = dev_seq_next,
4279 .stop = dev_seq_stop,
4280 .show = dev_seq_show,
4283 static int dev_seq_open(struct inode *inode, struct file *file)
4285 return seq_open_net(inode, file, &dev_seq_ops,
4286 sizeof(struct seq_net_private));
4289 static const struct file_operations dev_seq_fops = {
4290 .owner = THIS_MODULE,
4291 .open = dev_seq_open,
4293 .llseek = seq_lseek,
4294 .release = seq_release_net,
4297 static const struct seq_operations softnet_seq_ops = {
4298 .start = softnet_seq_start,
4299 .next = softnet_seq_next,
4300 .stop = softnet_seq_stop,
4301 .show = softnet_seq_show,
4304 static int softnet_seq_open(struct inode *inode, struct file *file)
4306 return seq_open(file, &softnet_seq_ops);
4309 static const struct file_operations softnet_seq_fops = {
4310 .owner = THIS_MODULE,
4311 .open = softnet_seq_open,
4313 .llseek = seq_lseek,
4314 .release = seq_release,
4317 static void *ptype_get_idx(loff_t pos)
4319 struct packet_type *pt = NULL;
4323 list_for_each_entry_rcu(pt, &ptype_all, list) {
4329 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4330 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4339 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4343 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4346 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4348 struct packet_type *pt;
4349 struct list_head *nxt;
4353 if (v == SEQ_START_TOKEN)
4354 return ptype_get_idx(0);
4357 nxt = pt->list.next;
4358 if (pt->type == htons(ETH_P_ALL)) {
4359 if (nxt != &ptype_all)
4362 nxt = ptype_base[0].next;
4364 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4366 while (nxt == &ptype_base[hash]) {
4367 if (++hash >= PTYPE_HASH_SIZE)
4369 nxt = ptype_base[hash].next;
4372 return list_entry(nxt, struct packet_type, list);
4375 static void ptype_seq_stop(struct seq_file *seq, void *v)
4381 static int ptype_seq_show(struct seq_file *seq, void *v)
4383 struct packet_type *pt = v;
4385 if (v == SEQ_START_TOKEN)
4386 seq_puts(seq, "Type Device Function\n");
4387 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4388 if (pt->type == htons(ETH_P_ALL))
4389 seq_puts(seq, "ALL ");
4391 seq_printf(seq, "%04x", ntohs(pt->type));
4393 seq_printf(seq, " %-8s %pF\n",
4394 pt->dev ? pt->dev->name : "", pt->func);
4400 static const struct seq_operations ptype_seq_ops = {
4401 .start = ptype_seq_start,
4402 .next = ptype_seq_next,
4403 .stop = ptype_seq_stop,
4404 .show = ptype_seq_show,
4407 static int ptype_seq_open(struct inode *inode, struct file *file)
4409 return seq_open_net(inode, file, &ptype_seq_ops,
4410 sizeof(struct seq_net_private));
4413 static const struct file_operations ptype_seq_fops = {
4414 .owner = THIS_MODULE,
4415 .open = ptype_seq_open,
4417 .llseek = seq_lseek,
4418 .release = seq_release_net,
4422 static int __net_init dev_proc_net_init(struct net *net)
4426 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4428 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4430 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4433 if (wext_proc_init(net))
4439 proc_net_remove(net, "ptype");
4441 proc_net_remove(net, "softnet_stat");
4443 proc_net_remove(net, "dev");
4447 static void __net_exit dev_proc_net_exit(struct net *net)
4449 wext_proc_exit(net);
4451 proc_net_remove(net, "ptype");
4452 proc_net_remove(net, "softnet_stat");
4453 proc_net_remove(net, "dev");
4456 static struct pernet_operations __net_initdata dev_proc_ops = {
4457 .init = dev_proc_net_init,
4458 .exit = dev_proc_net_exit,
4461 static int __init dev_proc_init(void)
4463 return register_pernet_subsys(&dev_proc_ops);
4466 #define dev_proc_init() 0
4467 #endif /* CONFIG_PROC_FS */
4471 * netdev_set_master - set up master pointer
4472 * @slave: slave device
4473 * @master: new master device
4475 * Changes the master device of the slave. Pass %NULL to break the
4476 * bonding. The caller must hold the RTNL semaphore. On a failure
4477 * a negative errno code is returned. On success the reference counts
4478 * are adjusted and the function returns zero.
4480 int netdev_set_master(struct net_device *slave, struct net_device *master)
4482 struct net_device *old = slave->master;
4492 slave->master = master;
4498 EXPORT_SYMBOL(netdev_set_master);
4501 * netdev_set_bond_master - set up bonding master/slave pair
4502 * @slave: slave device
4503 * @master: new master device
4505 * Changes the master device of the slave. Pass %NULL to break the
4506 * bonding. The caller must hold the RTNL semaphore. On a failure
4507 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4508 * to the routing socket and the function returns zero.
4510 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4516 err = netdev_set_master(slave, master);
4520 slave->flags |= IFF_SLAVE;
4522 slave->flags &= ~IFF_SLAVE;
4524 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4527 EXPORT_SYMBOL(netdev_set_bond_master);
4529 static void dev_change_rx_flags(struct net_device *dev, int flags)
4531 const struct net_device_ops *ops = dev->netdev_ops;
4533 if (ops->ndo_change_rx_flags)
4534 ops->ndo_change_rx_flags(dev, flags);
4537 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4539 unsigned short old_flags = dev->flags;
4545 dev->flags |= IFF_PROMISC;
4546 dev->promiscuity += inc;
4547 if (dev->promiscuity == 0) {
4550 * If inc causes overflow, untouch promisc and return error.
4553 dev->flags &= ~IFF_PROMISC;
4555 dev->promiscuity -= inc;
4556 printk(KERN_WARNING "%s: promiscuity touches roof, "
4557 "set promiscuity failed, promiscuity feature "
4558 "of device might be broken.\n", dev->name);
4562 if (dev->flags != old_flags) {
4563 printk(KERN_INFO "device %s %s promiscuous mode\n",
4564 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4566 if (audit_enabled) {
4567 current_uid_gid(&uid, &gid);
4568 audit_log(current->audit_context, GFP_ATOMIC,
4569 AUDIT_ANOM_PROMISCUOUS,
4570 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4571 dev->name, (dev->flags & IFF_PROMISC),
4572 (old_flags & IFF_PROMISC),
4573 audit_get_loginuid(current),
4575 audit_get_sessionid(current));
4578 dev_change_rx_flags(dev, IFF_PROMISC);
4584 * dev_set_promiscuity - update promiscuity count on a device
4588 * Add or remove promiscuity from a device. While the count in the device
4589 * remains above zero the interface remains promiscuous. Once it hits zero
4590 * the device reverts back to normal filtering operation. A negative inc
4591 * value is used to drop promiscuity on the device.
4592 * Return 0 if successful or a negative errno code on error.
4594 int dev_set_promiscuity(struct net_device *dev, int inc)
4596 unsigned short old_flags = dev->flags;
4599 err = __dev_set_promiscuity(dev, inc);
4602 if (dev->flags != old_flags)
4603 dev_set_rx_mode(dev);
4606 EXPORT_SYMBOL(dev_set_promiscuity);
4609 * dev_set_allmulti - update allmulti count on a device
4613 * Add or remove reception of all multicast frames to a device. While the
4614 * count in the device remains above zero the interface remains listening
4615 * to all interfaces. Once it hits zero the device reverts back to normal
4616 * filtering operation. A negative @inc value is used to drop the counter
4617 * when releasing a resource needing all multicasts.
4618 * Return 0 if successful or a negative errno code on error.
4621 int dev_set_allmulti(struct net_device *dev, int inc)
4623 unsigned short old_flags = dev->flags;
4627 dev->flags |= IFF_ALLMULTI;
4628 dev->allmulti += inc;
4629 if (dev->allmulti == 0) {
4632 * If inc causes overflow, untouch allmulti and return error.
4635 dev->flags &= ~IFF_ALLMULTI;
4637 dev->allmulti -= inc;
4638 printk(KERN_WARNING "%s: allmulti touches roof, "
4639 "set allmulti failed, allmulti feature of "
4640 "device might be broken.\n", dev->name);
4644 if (dev->flags ^ old_flags) {
4645 dev_change_rx_flags(dev, IFF_ALLMULTI);
4646 dev_set_rx_mode(dev);
4650 EXPORT_SYMBOL(dev_set_allmulti);
4653 * Upload unicast and multicast address lists to device and
4654 * configure RX filtering. When the device doesn't support unicast
4655 * filtering it is put in promiscuous mode while unicast addresses
4658 void __dev_set_rx_mode(struct net_device *dev)
4660 const struct net_device_ops *ops = dev->netdev_ops;
4662 /* dev_open will call this function so the list will stay sane. */
4663 if (!(dev->flags&IFF_UP))
4666 if (!netif_device_present(dev))
4669 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4670 /* Unicast addresses changes may only happen under the rtnl,
4671 * therefore calling __dev_set_promiscuity here is safe.
4673 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4674 __dev_set_promiscuity(dev, 1);
4675 dev->uc_promisc = true;
4676 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4677 __dev_set_promiscuity(dev, -1);
4678 dev->uc_promisc = false;
4682 if (ops->ndo_set_rx_mode)
4683 ops->ndo_set_rx_mode(dev);
4686 void dev_set_rx_mode(struct net_device *dev)
4688 netif_addr_lock_bh(dev);
4689 __dev_set_rx_mode(dev);
4690 netif_addr_unlock_bh(dev);
4694 * dev_get_flags - get flags reported to userspace
4697 * Get the combination of flag bits exported through APIs to userspace.
4699 unsigned dev_get_flags(const struct net_device *dev)
4703 flags = (dev->flags & ~(IFF_PROMISC |
4708 (dev->gflags & (IFF_PROMISC |
4711 if (netif_running(dev)) {
4712 if (netif_oper_up(dev))
4713 flags |= IFF_RUNNING;
4714 if (netif_carrier_ok(dev))
4715 flags |= IFF_LOWER_UP;
4716 if (netif_dormant(dev))
4717 flags |= IFF_DORMANT;
4722 EXPORT_SYMBOL(dev_get_flags);
4724 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4726 int old_flags = dev->flags;
4732 * Set the flags on our device.
4735 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4736 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4738 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4742 * Load in the correct multicast list now the flags have changed.
4745 if ((old_flags ^ flags) & IFF_MULTICAST)
4746 dev_change_rx_flags(dev, IFF_MULTICAST);
4748 dev_set_rx_mode(dev);
4751 * Have we downed the interface. We handle IFF_UP ourselves
4752 * according to user attempts to set it, rather than blindly
4757 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4758 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4761 dev_set_rx_mode(dev);
4764 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4765 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4767 dev->gflags ^= IFF_PROMISC;
4768 dev_set_promiscuity(dev, inc);
4771 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4772 is important. Some (broken) drivers set IFF_PROMISC, when
4773 IFF_ALLMULTI is requested not asking us and not reporting.
4775 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4776 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4778 dev->gflags ^= IFF_ALLMULTI;
4779 dev_set_allmulti(dev, inc);
4785 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4787 unsigned int changes = dev->flags ^ old_flags;
4789 if (changes & IFF_UP) {
4790 if (dev->flags & IFF_UP)
4791 call_netdevice_notifiers(NETDEV_UP, dev);
4793 call_netdevice_notifiers(NETDEV_DOWN, dev);
4796 if (dev->flags & IFF_UP &&
4797 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4798 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4802 * dev_change_flags - change device settings
4804 * @flags: device state flags
4806 * Change settings on device based state flags. The flags are
4807 * in the userspace exported format.
4809 int dev_change_flags(struct net_device *dev, unsigned flags)
4812 int old_flags = dev->flags;
4814 ret = __dev_change_flags(dev, flags);
4818 changes = old_flags ^ dev->flags;
4820 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4822 __dev_notify_flags(dev, old_flags);
4825 EXPORT_SYMBOL(dev_change_flags);
4828 * dev_set_mtu - Change maximum transfer unit
4830 * @new_mtu: new transfer unit
4832 * Change the maximum transfer size of the network device.
4834 int dev_set_mtu(struct net_device *dev, int new_mtu)
4836 const struct net_device_ops *ops = dev->netdev_ops;
4839 if (new_mtu == dev->mtu)
4842 /* MTU must be positive. */
4846 if (!netif_device_present(dev))
4850 if (ops->ndo_change_mtu)
4851 err = ops->ndo_change_mtu(dev, new_mtu);
4855 if (!err && dev->flags & IFF_UP)
4856 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4859 EXPORT_SYMBOL(dev_set_mtu);
4862 * dev_set_group - Change group this device belongs to
4864 * @new_group: group this device should belong to
4866 void dev_set_group(struct net_device *dev, int new_group)
4868 dev->group = new_group;
4870 EXPORT_SYMBOL(dev_set_group);
4873 * dev_set_mac_address - Change Media Access Control Address
4877 * Change the hardware (MAC) address of the device
4879 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4881 const struct net_device_ops *ops = dev->netdev_ops;
4884 if (!ops->ndo_set_mac_address)
4886 if (sa->sa_family != dev->type)
4888 if (!netif_device_present(dev))
4890 err = ops->ndo_set_mac_address(dev, sa);
4892 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4893 add_device_randomness(dev->dev_addr, dev->addr_len);
4896 EXPORT_SYMBOL(dev_set_mac_address);
4899 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4901 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4904 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4910 case SIOCGIFFLAGS: /* Get interface flags */
4911 ifr->ifr_flags = (short) dev_get_flags(dev);
4914 case SIOCGIFMETRIC: /* Get the metric on the interface
4915 (currently unused) */
4916 ifr->ifr_metric = 0;
4919 case SIOCGIFMTU: /* Get the MTU of a device */
4920 ifr->ifr_mtu = dev->mtu;
4925 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4927 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4928 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4929 ifr->ifr_hwaddr.sa_family = dev->type;
4937 ifr->ifr_map.mem_start = dev->mem_start;
4938 ifr->ifr_map.mem_end = dev->mem_end;
4939 ifr->ifr_map.base_addr = dev->base_addr;
4940 ifr->ifr_map.irq = dev->irq;
4941 ifr->ifr_map.dma = dev->dma;
4942 ifr->ifr_map.port = dev->if_port;
4946 ifr->ifr_ifindex = dev->ifindex;
4950 ifr->ifr_qlen = dev->tx_queue_len;
4954 /* dev_ioctl() should ensure this case
4966 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4968 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4971 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4972 const struct net_device_ops *ops;
4977 ops = dev->netdev_ops;
4980 case SIOCSIFFLAGS: /* Set interface flags */
4981 return dev_change_flags(dev, ifr->ifr_flags);
4983 case SIOCSIFMETRIC: /* Set the metric on the interface
4984 (currently unused) */
4987 case SIOCSIFMTU: /* Set the MTU of a device */
4988 return dev_set_mtu(dev, ifr->ifr_mtu);
4991 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4993 case SIOCSIFHWBROADCAST:
4994 if (ifr->ifr_hwaddr.sa_family != dev->type)
4996 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4997 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4998 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5002 if (ops->ndo_set_config) {
5003 if (!netif_device_present(dev))
5005 return ops->ndo_set_config(dev, &ifr->ifr_map);
5010 if (!ops->ndo_set_rx_mode ||
5011 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5013 if (!netif_device_present(dev))
5015 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5018 if (!ops->ndo_set_rx_mode ||
5019 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5021 if (!netif_device_present(dev))
5023 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5026 if (ifr->ifr_qlen < 0)
5028 dev->tx_queue_len = ifr->ifr_qlen;
5032 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5033 return dev_change_name(dev, ifr->ifr_newname);
5036 err = net_hwtstamp_validate(ifr);
5042 * Unknown or private ioctl
5045 if ((cmd >= SIOCDEVPRIVATE &&
5046 cmd <= SIOCDEVPRIVATE + 15) ||
5047 cmd == SIOCBONDENSLAVE ||
5048 cmd == SIOCBONDRELEASE ||
5049 cmd == SIOCBONDSETHWADDR ||
5050 cmd == SIOCBONDSLAVEINFOQUERY ||
5051 cmd == SIOCBONDINFOQUERY ||
5052 cmd == SIOCBONDCHANGEACTIVE ||
5053 cmd == SIOCGMIIPHY ||
5054 cmd == SIOCGMIIREG ||
5055 cmd == SIOCSMIIREG ||
5056 cmd == SIOCBRADDIF ||
5057 cmd == SIOCBRDELIF ||
5058 cmd == SIOCSHWTSTAMP ||
5059 cmd == SIOCWANDEV) {
5061 if (ops->ndo_do_ioctl) {
5062 if (netif_device_present(dev))
5063 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5075 * This function handles all "interface"-type I/O control requests. The actual
5076 * 'doing' part of this is dev_ifsioc above.
5080 * dev_ioctl - network device ioctl
5081 * @net: the applicable net namespace
5082 * @cmd: command to issue
5083 * @arg: pointer to a struct ifreq in user space
5085 * Issue ioctl functions to devices. This is normally called by the
5086 * user space syscall interfaces but can sometimes be useful for
5087 * other purposes. The return value is the return from the syscall if
5088 * positive or a negative errno code on error.
5091 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5097 /* One special case: SIOCGIFCONF takes ifconf argument
5098 and requires shared lock, because it sleeps writing
5102 if (cmd == SIOCGIFCONF) {
5104 ret = dev_ifconf(net, (char __user *) arg);
5108 if (cmd == SIOCGIFNAME)
5109 return dev_ifname(net, (struct ifreq __user *)arg);
5111 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5114 ifr.ifr_name[IFNAMSIZ-1] = 0;
5116 colon = strchr(ifr.ifr_name, ':');
5121 * See which interface the caller is talking about.
5126 * These ioctl calls:
5127 * - can be done by all.
5128 * - atomic and do not require locking.
5139 dev_load(net, ifr.ifr_name);
5141 ret = dev_ifsioc_locked(net, &ifr, cmd);
5146 if (copy_to_user(arg, &ifr,
5147 sizeof(struct ifreq)))
5153 dev_load(net, ifr.ifr_name);
5155 ret = dev_ethtool(net, &ifr);
5160 if (copy_to_user(arg, &ifr,
5161 sizeof(struct ifreq)))
5167 * These ioctl calls:
5168 * - require superuser power.
5169 * - require strict serialization.
5175 if (!capable(CAP_NET_ADMIN))
5177 dev_load(net, ifr.ifr_name);
5179 ret = dev_ifsioc(net, &ifr, cmd);
5184 if (copy_to_user(arg, &ifr,
5185 sizeof(struct ifreq)))
5191 * These ioctl calls:
5192 * - require superuser power.
5193 * - require strict serialization.
5194 * - do not return a value
5204 case SIOCSIFHWBROADCAST:
5207 case SIOCBONDENSLAVE:
5208 case SIOCBONDRELEASE:
5209 case SIOCBONDSETHWADDR:
5210 case SIOCBONDCHANGEACTIVE:
5214 if (!capable(CAP_NET_ADMIN))
5217 case SIOCBONDSLAVEINFOQUERY:
5218 case SIOCBONDINFOQUERY:
5219 dev_load(net, ifr.ifr_name);
5221 ret = dev_ifsioc(net, &ifr, cmd);
5226 /* Get the per device memory space. We can add this but
5227 * currently do not support it */
5229 /* Set the per device memory buffer space.
5230 * Not applicable in our case */
5235 * Unknown or private ioctl.
5238 if (cmd == SIOCWANDEV ||
5239 (cmd >= SIOCDEVPRIVATE &&
5240 cmd <= SIOCDEVPRIVATE + 15)) {
5241 dev_load(net, ifr.ifr_name);
5243 ret = dev_ifsioc(net, &ifr, cmd);
5245 if (!ret && copy_to_user(arg, &ifr,
5246 sizeof(struct ifreq)))
5250 /* Take care of Wireless Extensions */
5251 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5252 return wext_handle_ioctl(net, &ifr, cmd, arg);
5259 * dev_new_index - allocate an ifindex
5260 * @net: the applicable net namespace
5262 * Returns a suitable unique value for a new device interface
5263 * number. The caller must hold the rtnl semaphore or the
5264 * dev_base_lock to be sure it remains unique.
5266 static int dev_new_index(struct net *net)
5272 if (!__dev_get_by_index(net, ifindex))
5277 /* Delayed registration/unregisteration */
5278 static LIST_HEAD(net_todo_list);
5280 static void net_set_todo(struct net_device *dev)
5282 list_add_tail(&dev->todo_list, &net_todo_list);
5285 static void rollback_registered_many(struct list_head *head)
5287 struct net_device *dev, *tmp;
5289 BUG_ON(dev_boot_phase);
5292 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5293 /* Some devices call without registering
5294 * for initialization unwind. Remove those
5295 * devices and proceed with the remaining.
5297 if (dev->reg_state == NETREG_UNINITIALIZED) {
5298 pr_debug("unregister_netdevice: device %s/%p never "
5299 "was registered\n", dev->name, dev);
5302 list_del(&dev->unreg_list);
5305 dev->dismantle = true;
5306 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5309 /* If device is running, close it first. */
5310 dev_close_many(head);
5312 list_for_each_entry(dev, head, unreg_list) {
5313 /* And unlink it from device chain. */
5314 unlist_netdevice(dev);
5316 dev->reg_state = NETREG_UNREGISTERING;
5317 on_each_cpu(flush_backlog, dev, 1);
5322 list_for_each_entry(dev, head, unreg_list) {
5323 /* Shutdown queueing discipline. */
5327 /* Notify protocols, that we are about to destroy
5328 this device. They should clean all the things.
5330 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5332 if (!dev->rtnl_link_ops ||
5333 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5334 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5337 * Flush the unicast and multicast chains
5342 if (dev->netdev_ops->ndo_uninit)
5343 dev->netdev_ops->ndo_uninit(dev);
5345 /* Notifier chain MUST detach us from master device. */
5346 WARN_ON(dev->master);
5348 /* Remove entries from kobject tree */
5349 netdev_unregister_kobject(dev);
5352 /* Process any work delayed until the end of the batch */
5353 dev = list_first_entry(head, struct net_device, unreg_list);
5354 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5358 list_for_each_entry(dev, head, unreg_list)
5362 static void rollback_registered(struct net_device *dev)
5366 list_add(&dev->unreg_list, &single);
5367 rollback_registered_many(&single);
5371 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5373 /* Fix illegal checksum combinations */
5374 if ((features & NETIF_F_HW_CSUM) &&
5375 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5376 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5377 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5380 if ((features & NETIF_F_NO_CSUM) &&
5381 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5382 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5383 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5386 /* Fix illegal SG+CSUM combinations. */
5387 if ((features & NETIF_F_SG) &&
5388 !(features & NETIF_F_ALL_CSUM)) {
5390 "Dropping NETIF_F_SG since no checksum feature.\n");
5391 features &= ~NETIF_F_SG;
5394 /* TSO requires that SG is present as well. */
5395 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5396 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5397 features &= ~NETIF_F_ALL_TSO;
5400 /* TSO ECN requires that TSO is present as well. */
5401 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5402 features &= ~NETIF_F_TSO_ECN;
5404 /* Software GSO depends on SG. */
5405 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5406 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5407 features &= ~NETIF_F_GSO;
5410 /* UFO needs SG and checksumming */
5411 if (features & NETIF_F_UFO) {
5412 /* maybe split UFO into V4 and V6? */
5413 if (!((features & NETIF_F_GEN_CSUM) ||
5414 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5415 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5417 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5418 features &= ~NETIF_F_UFO;
5421 if (!(features & NETIF_F_SG)) {
5423 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5424 features &= ~NETIF_F_UFO;
5431 int __netdev_update_features(struct net_device *dev)
5438 features = netdev_get_wanted_features(dev);
5440 if (dev->netdev_ops->ndo_fix_features)
5441 features = dev->netdev_ops->ndo_fix_features(dev, features);
5443 /* driver might be less strict about feature dependencies */
5444 features = netdev_fix_features(dev, features);
5446 if (dev->features == features)
5449 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5450 dev->features, features);
5452 if (dev->netdev_ops->ndo_set_features)
5453 err = dev->netdev_ops->ndo_set_features(dev, features);
5455 if (unlikely(err < 0)) {
5457 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5458 err, features, dev->features);
5459 /* return non-0 since some features might have changed and
5460 * it's better to fire a spurious notification than miss it
5466 dev->features = features;
5472 * netdev_update_features - recalculate device features
5473 * @dev: the device to check
5475 * Recalculate dev->features set and send notifications if it
5476 * has changed. Should be called after driver or hardware dependent
5477 * conditions might have changed that influence the features.
5479 void netdev_update_features(struct net_device *dev)
5481 if (__netdev_update_features(dev))
5482 netdev_features_change(dev);
5484 EXPORT_SYMBOL(netdev_update_features);
5487 * netdev_change_features - recalculate device features
5488 * @dev: the device to check
5490 * Recalculate dev->features set and send notifications even
5491 * if they have not changed. Should be called instead of
5492 * netdev_update_features() if also dev->vlan_features might
5493 * have changed to allow the changes to be propagated to stacked
5496 void netdev_change_features(struct net_device *dev)
5498 __netdev_update_features(dev);
5499 netdev_features_change(dev);
5501 EXPORT_SYMBOL(netdev_change_features);
5504 * netif_stacked_transfer_operstate - transfer operstate
5505 * @rootdev: the root or lower level device to transfer state from
5506 * @dev: the device to transfer operstate to
5508 * Transfer operational state from root to device. This is normally
5509 * called when a stacking relationship exists between the root
5510 * device and the device(a leaf device).
5512 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5513 struct net_device *dev)
5515 if (rootdev->operstate == IF_OPER_DORMANT)
5516 netif_dormant_on(dev);
5518 netif_dormant_off(dev);
5520 if (netif_carrier_ok(rootdev)) {
5521 if (!netif_carrier_ok(dev))
5522 netif_carrier_on(dev);
5524 if (netif_carrier_ok(dev))
5525 netif_carrier_off(dev);
5528 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5531 static int netif_alloc_rx_queues(struct net_device *dev)
5533 unsigned int i, count = dev->num_rx_queues;
5534 struct netdev_rx_queue *rx;
5538 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5540 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5545 for (i = 0; i < count; i++)
5551 static void netdev_init_one_queue(struct net_device *dev,
5552 struct netdev_queue *queue, void *_unused)
5554 /* Initialize queue lock */
5555 spin_lock_init(&queue->_xmit_lock);
5556 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5557 queue->xmit_lock_owner = -1;
5558 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5562 static int netif_alloc_netdev_queues(struct net_device *dev)
5564 unsigned int count = dev->num_tx_queues;
5565 struct netdev_queue *tx;
5569 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5571 pr_err("netdev: Unable to allocate %u tx queues.\n",
5577 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5578 spin_lock_init(&dev->tx_global_lock);
5584 * register_netdevice - register a network device
5585 * @dev: device to register
5587 * Take a completed network device structure and add it to the kernel
5588 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5589 * chain. 0 is returned on success. A negative errno code is returned
5590 * on a failure to set up the device, or if the name is a duplicate.
5592 * Callers must hold the rtnl semaphore. You may want
5593 * register_netdev() instead of this.
5596 * The locking appears insufficient to guarantee two parallel registers
5597 * will not get the same name.
5600 int register_netdevice(struct net_device *dev)
5603 struct net *net = dev_net(dev);
5605 BUG_ON(dev_boot_phase);
5610 /* When net_device's are persistent, this will be fatal. */
5611 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5614 spin_lock_init(&dev->addr_list_lock);
5615 netdev_set_addr_lockdep_class(dev);
5619 ret = dev_get_valid_name(dev, dev->name);
5623 /* Init, if this function is available */
5624 if (dev->netdev_ops->ndo_init) {
5625 ret = dev->netdev_ops->ndo_init(dev);
5633 dev->ifindex = dev_new_index(net);
5634 if (dev->iflink == -1)
5635 dev->iflink = dev->ifindex;
5637 /* Transfer changeable features to wanted_features and enable
5638 * software offloads (GSO and GRO).
5640 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5641 dev->features |= NETIF_F_SOFT_FEATURES;
5642 dev->wanted_features = dev->features & dev->hw_features;
5644 /* Turn on no cache copy if HW is doing checksum */
5645 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5646 if ((dev->features & NETIF_F_ALL_CSUM) &&
5647 !(dev->features & NETIF_F_NO_CSUM)) {
5648 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5649 dev->features |= NETIF_F_NOCACHE_COPY;
5652 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5654 dev->vlan_features |= NETIF_F_HIGHDMA;
5656 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5657 ret = notifier_to_errno(ret);
5661 ret = netdev_register_kobject(dev);
5664 dev->reg_state = NETREG_REGISTERED;
5666 __netdev_update_features(dev);
5669 * Default initial state at registry is that the
5670 * device is present.
5673 set_bit(__LINK_STATE_PRESENT, &dev->state);
5675 dev_init_scheduler(dev);
5677 list_netdevice(dev);
5678 add_device_randomness(dev->dev_addr, dev->addr_len);
5680 /* Notify protocols, that a new device appeared. */
5681 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5682 ret = notifier_to_errno(ret);
5684 rollback_registered(dev);
5685 dev->reg_state = NETREG_UNREGISTERED;
5688 * Prevent userspace races by waiting until the network
5689 * device is fully setup before sending notifications.
5691 if (!dev->rtnl_link_ops ||
5692 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5693 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5699 if (dev->netdev_ops->ndo_uninit)
5700 dev->netdev_ops->ndo_uninit(dev);
5703 EXPORT_SYMBOL(register_netdevice);
5706 * init_dummy_netdev - init a dummy network device for NAPI
5707 * @dev: device to init
5709 * This takes a network device structure and initialize the minimum
5710 * amount of fields so it can be used to schedule NAPI polls without
5711 * registering a full blown interface. This is to be used by drivers
5712 * that need to tie several hardware interfaces to a single NAPI
5713 * poll scheduler due to HW limitations.
5715 int init_dummy_netdev(struct net_device *dev)
5717 /* Clear everything. Note we don't initialize spinlocks
5718 * are they aren't supposed to be taken by any of the
5719 * NAPI code and this dummy netdev is supposed to be
5720 * only ever used for NAPI polls
5722 memset(dev, 0, sizeof(struct net_device));
5724 /* make sure we BUG if trying to hit standard
5725 * register/unregister code path
5727 dev->reg_state = NETREG_DUMMY;
5729 /* NAPI wants this */
5730 INIT_LIST_HEAD(&dev->napi_list);
5732 /* a dummy interface is started by default */
5733 set_bit(__LINK_STATE_PRESENT, &dev->state);
5734 set_bit(__LINK_STATE_START, &dev->state);
5736 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5737 * because users of this 'device' dont need to change
5743 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5747 * register_netdev - register a network device
5748 * @dev: device to register
5750 * Take a completed network device structure and add it to the kernel
5751 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5752 * chain. 0 is returned on success. A negative errno code is returned
5753 * on a failure to set up the device, or if the name is a duplicate.
5755 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5756 * and expands the device name if you passed a format string to
5759 int register_netdev(struct net_device *dev)
5764 err = register_netdevice(dev);
5768 EXPORT_SYMBOL(register_netdev);
5770 int netdev_refcnt_read(const struct net_device *dev)
5774 for_each_possible_cpu(i)
5775 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5778 EXPORT_SYMBOL(netdev_refcnt_read);
5781 * netdev_wait_allrefs - wait until all references are gone.
5783 * This is called when unregistering network devices.
5785 * Any protocol or device that holds a reference should register
5786 * for netdevice notification, and cleanup and put back the
5787 * reference if they receive an UNREGISTER event.
5788 * We can get stuck here if buggy protocols don't correctly
5791 static void netdev_wait_allrefs(struct net_device *dev)
5793 unsigned long rebroadcast_time, warning_time;
5796 linkwatch_forget_dev(dev);
5798 rebroadcast_time = warning_time = jiffies;
5799 refcnt = netdev_refcnt_read(dev);
5801 while (refcnt != 0) {
5802 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5805 /* Rebroadcast unregister notification */
5806 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5807 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5808 * should have already handle it the first time */
5810 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5812 /* We must not have linkwatch events
5813 * pending on unregister. If this
5814 * happens, we simply run the queue
5815 * unscheduled, resulting in a noop
5818 linkwatch_run_queue();
5823 rebroadcast_time = jiffies;
5828 refcnt = netdev_refcnt_read(dev);
5830 if (time_after(jiffies, warning_time + 10 * HZ)) {
5831 printk(KERN_EMERG "unregister_netdevice: "
5832 "waiting for %s to become free. Usage "
5835 warning_time = jiffies;
5844 * register_netdevice(x1);
5845 * register_netdevice(x2);
5847 * unregister_netdevice(y1);
5848 * unregister_netdevice(y2);
5854 * We are invoked by rtnl_unlock().
5855 * This allows us to deal with problems:
5856 * 1) We can delete sysfs objects which invoke hotplug
5857 * without deadlocking with linkwatch via keventd.
5858 * 2) Since we run with the RTNL semaphore not held, we can sleep
5859 * safely in order to wait for the netdev refcnt to drop to zero.
5861 * We must not return until all unregister events added during
5862 * the interval the lock was held have been completed.
5864 void netdev_run_todo(void)
5866 struct list_head list;
5868 /* Snapshot list, allow later requests */
5869 list_replace_init(&net_todo_list, &list);
5873 /* Wait for rcu callbacks to finish before attempting to drain
5874 * the device list. This usually avoids a 250ms wait.
5876 if (!list_empty(&list))
5879 while (!list_empty(&list)) {
5880 struct net_device *dev
5881 = list_first_entry(&list, struct net_device, todo_list);
5882 list_del(&dev->todo_list);
5884 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5885 printk(KERN_ERR "network todo '%s' but state %d\n",
5886 dev->name, dev->reg_state);
5891 dev->reg_state = NETREG_UNREGISTERED;
5893 netdev_wait_allrefs(dev);
5896 BUG_ON(netdev_refcnt_read(dev));
5897 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5898 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5899 WARN_ON(dev->dn_ptr);
5901 if (dev->destructor)
5902 dev->destructor(dev);
5904 /* Free network device */
5905 kobject_put(&dev->dev.kobj);
5909 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5910 * fields in the same order, with only the type differing.
5912 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5913 const struct net_device_stats *netdev_stats)
5915 #if BITS_PER_LONG == 64
5916 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5917 memcpy(stats64, netdev_stats, sizeof(*stats64));
5919 size_t i, n = sizeof(*stats64) / sizeof(u64);
5920 const unsigned long *src = (const unsigned long *)netdev_stats;
5921 u64 *dst = (u64 *)stats64;
5923 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5924 sizeof(*stats64) / sizeof(u64));
5925 for (i = 0; i < n; i++)
5931 * dev_get_stats - get network device statistics
5932 * @dev: device to get statistics from
5933 * @storage: place to store stats
5935 * Get network statistics from device. Return @storage.
5936 * The device driver may provide its own method by setting
5937 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5938 * otherwise the internal statistics structure is used.
5940 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5941 struct rtnl_link_stats64 *storage)
5943 const struct net_device_ops *ops = dev->netdev_ops;
5945 if (ops->ndo_get_stats64) {
5946 memset(storage, 0, sizeof(*storage));
5947 ops->ndo_get_stats64(dev, storage);
5948 } else if (ops->ndo_get_stats) {
5949 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5951 netdev_stats_to_stats64(storage, &dev->stats);
5953 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
5956 EXPORT_SYMBOL(dev_get_stats);
5958 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5960 struct netdev_queue *queue = dev_ingress_queue(dev);
5962 #ifdef CONFIG_NET_CLS_ACT
5965 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5968 netdev_init_one_queue(dev, queue, NULL);
5969 queue->qdisc = &noop_qdisc;
5970 queue->qdisc_sleeping = &noop_qdisc;
5971 rcu_assign_pointer(dev->ingress_queue, queue);
5977 * alloc_netdev_mqs - allocate network device
5978 * @sizeof_priv: size of private data to allocate space for
5979 * @name: device name format string
5980 * @setup: callback to initialize device
5981 * @txqs: the number of TX subqueues to allocate
5982 * @rxqs: the number of RX subqueues to allocate
5984 * Allocates a struct net_device with private data area for driver use
5985 * and performs basic initialization. Also allocates subquue structs
5986 * for each queue on the device.
5988 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5989 void (*setup)(struct net_device *),
5990 unsigned int txqs, unsigned int rxqs)
5992 struct net_device *dev;
5994 struct net_device *p;
5996 BUG_ON(strlen(name) >= sizeof(dev->name));
5999 pr_err("alloc_netdev: Unable to allocate device "
6000 "with zero queues.\n");
6006 pr_err("alloc_netdev: Unable to allocate device "
6007 "with zero RX queues.\n");
6012 alloc_size = sizeof(struct net_device);
6014 /* ensure 32-byte alignment of private area */
6015 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6016 alloc_size += sizeof_priv;
6018 /* ensure 32-byte alignment of whole construct */
6019 alloc_size += NETDEV_ALIGN - 1;
6021 p = kzalloc(alloc_size, GFP_KERNEL);
6023 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
6027 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6028 dev->padded = (char *)dev - (char *)p;
6030 dev->pcpu_refcnt = alloc_percpu(int);
6031 if (!dev->pcpu_refcnt)
6034 if (dev_addr_init(dev))
6040 dev_net_set(dev, &init_net);
6042 dev->gso_max_size = GSO_MAX_SIZE;
6043 dev->gso_max_segs = GSO_MAX_SEGS;
6045 INIT_LIST_HEAD(&dev->napi_list);
6046 INIT_LIST_HEAD(&dev->unreg_list);
6047 INIT_LIST_HEAD(&dev->link_watch_list);
6048 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6051 dev->num_tx_queues = txqs;
6052 dev->real_num_tx_queues = txqs;
6053 if (netif_alloc_netdev_queues(dev))
6057 dev->num_rx_queues = rxqs;
6058 dev->real_num_rx_queues = rxqs;
6059 if (netif_alloc_rx_queues(dev))
6063 strcpy(dev->name, name);
6064 dev->group = INIT_NETDEV_GROUP;
6072 free_percpu(dev->pcpu_refcnt);
6082 EXPORT_SYMBOL(alloc_netdev_mqs);
6085 * free_netdev - free network device
6088 * This function does the last stage of destroying an allocated device
6089 * interface. The reference to the device object is released.
6090 * If this is the last reference then it will be freed.
6092 void free_netdev(struct net_device *dev)
6094 struct napi_struct *p, *n;
6096 release_net(dev_net(dev));
6103 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6105 /* Flush device addresses */
6106 dev_addr_flush(dev);
6108 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6111 free_percpu(dev->pcpu_refcnt);
6112 dev->pcpu_refcnt = NULL;
6114 /* Compatibility with error handling in drivers */
6115 if (dev->reg_state == NETREG_UNINITIALIZED) {
6116 kfree((char *)dev - dev->padded);
6120 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6121 dev->reg_state = NETREG_RELEASED;
6123 /* will free via device release */
6124 put_device(&dev->dev);
6126 EXPORT_SYMBOL(free_netdev);
6129 * synchronize_net - Synchronize with packet receive processing
6131 * Wait for packets currently being received to be done.
6132 * Does not block later packets from starting.
6134 void synchronize_net(void)
6137 if (rtnl_is_locked())
6138 synchronize_rcu_expedited();
6142 EXPORT_SYMBOL(synchronize_net);
6145 * unregister_netdevice_queue - remove device from the kernel
6149 * This function shuts down a device interface and removes it
6150 * from the kernel tables.
6151 * If head not NULL, device is queued to be unregistered later.
6153 * Callers must hold the rtnl semaphore. You may want
6154 * unregister_netdev() instead of this.
6157 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6162 list_move_tail(&dev->unreg_list, head);
6164 rollback_registered(dev);
6165 /* Finish processing unregister after unlock */
6169 EXPORT_SYMBOL(unregister_netdevice_queue);
6172 * unregister_netdevice_many - unregister many devices
6173 * @head: list of devices
6175 void unregister_netdevice_many(struct list_head *head)
6177 struct net_device *dev;
6179 if (!list_empty(head)) {
6180 rollback_registered_many(head);
6181 list_for_each_entry(dev, head, unreg_list)
6185 EXPORT_SYMBOL(unregister_netdevice_many);
6188 * unregister_netdev - remove device from the kernel
6191 * This function shuts down a device interface and removes it
6192 * from the kernel tables.
6194 * This is just a wrapper for unregister_netdevice that takes
6195 * the rtnl semaphore. In general you want to use this and not
6196 * unregister_netdevice.
6198 void unregister_netdev(struct net_device *dev)
6201 unregister_netdevice(dev);
6204 EXPORT_SYMBOL(unregister_netdev);
6207 * dev_change_net_namespace - move device to different nethost namespace
6209 * @net: network namespace
6210 * @pat: If not NULL name pattern to try if the current device name
6211 * is already taken in the destination network namespace.
6213 * This function shuts down a device interface and moves it
6214 * to a new network namespace. On success 0 is returned, on
6215 * a failure a netagive errno code is returned.
6217 * Callers must hold the rtnl semaphore.
6220 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6226 /* Don't allow namespace local devices to be moved. */
6228 if (dev->features & NETIF_F_NETNS_LOCAL)
6231 /* Ensure the device has been registrered */
6233 if (dev->reg_state != NETREG_REGISTERED)
6236 /* Get out if there is nothing todo */
6238 if (net_eq(dev_net(dev), net))
6241 /* Pick the destination device name, and ensure
6242 * we can use it in the destination network namespace.
6245 if (__dev_get_by_name(net, dev->name)) {
6246 /* We get here if we can't use the current device name */
6249 if (dev_get_valid_name(dev, pat) < 0)
6254 * And now a mini version of register_netdevice unregister_netdevice.
6257 /* If device is running close it first. */
6260 /* And unlink it from device chain */
6262 unlist_netdevice(dev);
6266 /* Shutdown queueing discipline. */
6269 /* Notify protocols, that we are about to destroy
6270 this device. They should clean all the things.
6272 Note that dev->reg_state stays at NETREG_REGISTERED.
6273 This is wanted because this way 8021q and macvlan know
6274 the device is just moving and can keep their slaves up.
6276 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6277 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6278 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6281 * Flush the unicast and multicast chains
6286 /* Actually switch the network namespace */
6287 dev_net_set(dev, net);
6289 /* If there is an ifindex conflict assign a new one */
6290 if (__dev_get_by_index(net, dev->ifindex)) {
6291 int iflink = (dev->iflink == dev->ifindex);
6292 dev->ifindex = dev_new_index(net);
6294 dev->iflink = dev->ifindex;
6297 /* Fixup kobjects */
6298 err = device_rename(&dev->dev, dev->name);
6301 /* Add the device back in the hashes */
6302 list_netdevice(dev);
6304 /* Notify protocols, that a new device appeared. */
6305 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6308 * Prevent userspace races by waiting until the network
6309 * device is fully setup before sending notifications.
6311 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6318 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6320 static int dev_cpu_callback(struct notifier_block *nfb,
6321 unsigned long action,
6324 struct sk_buff **list_skb;
6325 struct sk_buff *skb;
6326 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6327 struct softnet_data *sd, *oldsd;
6329 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6332 local_irq_disable();
6333 cpu = smp_processor_id();
6334 sd = &per_cpu(softnet_data, cpu);
6335 oldsd = &per_cpu(softnet_data, oldcpu);
6337 /* Find end of our completion_queue. */
6338 list_skb = &sd->completion_queue;
6340 list_skb = &(*list_skb)->next;
6341 /* Append completion queue from offline CPU. */
6342 *list_skb = oldsd->completion_queue;
6343 oldsd->completion_queue = NULL;
6345 /* Append output queue from offline CPU. */
6346 if (oldsd->output_queue) {
6347 *sd->output_queue_tailp = oldsd->output_queue;
6348 sd->output_queue_tailp = oldsd->output_queue_tailp;
6349 oldsd->output_queue = NULL;
6350 oldsd->output_queue_tailp = &oldsd->output_queue;
6352 /* Append NAPI poll list from offline CPU, with one exception :
6353 * process_backlog() must be called by cpu owning percpu backlog.
6354 * We properly handle process_queue & input_pkt_queue later.
6356 while (!list_empty(&oldsd->poll_list)) {
6357 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
6361 list_del_init(&napi->poll_list);
6362 if (napi->poll == process_backlog)
6365 ____napi_schedule(sd, napi);
6368 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6371 /* Process offline CPU's input_pkt_queue */
6372 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6374 input_queue_head_incr(oldsd);
6376 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
6378 input_queue_head_incr(oldsd);
6386 * netdev_increment_features - increment feature set by one
6387 * @all: current feature set
6388 * @one: new feature set
6389 * @mask: mask feature set
6391 * Computes a new feature set after adding a device with feature set
6392 * @one to the master device with current feature set @all. Will not
6393 * enable anything that is off in @mask. Returns the new feature set.
6395 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6397 if (mask & NETIF_F_GEN_CSUM)
6398 mask |= NETIF_F_ALL_CSUM;
6399 mask |= NETIF_F_VLAN_CHALLENGED;
6401 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6402 all &= one | ~NETIF_F_ALL_FOR_ALL;
6404 /* If device needs checksumming, downgrade to it. */
6405 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6406 all &= ~NETIF_F_NO_CSUM;
6408 /* If one device supports hw checksumming, set for all. */
6409 if (all & NETIF_F_GEN_CSUM)
6410 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6414 EXPORT_SYMBOL(netdev_increment_features);
6416 static struct hlist_head *netdev_create_hash(void)
6419 struct hlist_head *hash;
6421 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6423 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6424 INIT_HLIST_HEAD(&hash[i]);
6429 /* Initialize per network namespace state */
6430 static int __net_init netdev_init(struct net *net)
6432 if (net != &init_net)
6433 INIT_LIST_HEAD(&net->dev_base_head);
6435 net->dev_name_head = netdev_create_hash();
6436 if (net->dev_name_head == NULL)
6439 net->dev_index_head = netdev_create_hash();
6440 if (net->dev_index_head == NULL)
6446 kfree(net->dev_name_head);
6452 * netdev_drivername - network driver for the device
6453 * @dev: network device
6455 * Determine network driver for device.
6457 const char *netdev_drivername(const struct net_device *dev)
6459 const struct device_driver *driver;
6460 const struct device *parent;
6461 const char *empty = "";
6463 parent = dev->dev.parent;
6467 driver = parent->driver;
6468 if (driver && driver->name)
6469 return driver->name;
6473 int __netdev_printk(const char *level, const struct net_device *dev,
6474 struct va_format *vaf)
6478 if (dev && dev->dev.parent)
6479 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6480 netdev_name(dev), vaf);
6482 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6484 r = printk("%s(NULL net_device): %pV", level, vaf);
6488 EXPORT_SYMBOL(__netdev_printk);
6490 int netdev_printk(const char *level, const struct net_device *dev,
6491 const char *format, ...)
6493 struct va_format vaf;
6497 va_start(args, format);
6502 r = __netdev_printk(level, dev, &vaf);
6507 EXPORT_SYMBOL(netdev_printk);
6509 #define define_netdev_printk_level(func, level) \
6510 int func(const struct net_device *dev, const char *fmt, ...) \
6513 struct va_format vaf; \
6516 va_start(args, fmt); \
6521 r = __netdev_printk(level, dev, &vaf); \
6526 EXPORT_SYMBOL(func);
6528 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6529 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6530 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6531 define_netdev_printk_level(netdev_err, KERN_ERR);
6532 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6533 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6534 define_netdev_printk_level(netdev_info, KERN_INFO);
6536 static void __net_exit netdev_exit(struct net *net)
6538 kfree(net->dev_name_head);
6539 kfree(net->dev_index_head);
6542 static struct pernet_operations __net_initdata netdev_net_ops = {
6543 .init = netdev_init,
6544 .exit = netdev_exit,
6547 static void __net_exit default_device_exit(struct net *net)
6549 struct net_device *dev, *aux;
6551 * Push all migratable network devices back to the
6552 * initial network namespace
6555 for_each_netdev_safe(net, dev, aux) {
6557 char fb_name[IFNAMSIZ];
6559 /* Ignore unmoveable devices (i.e. loopback) */
6560 if (dev->features & NETIF_F_NETNS_LOCAL)
6563 /* Leave virtual devices for the generic cleanup */
6564 if (dev->rtnl_link_ops)
6567 /* Push remaining network devices to init_net */
6568 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6569 err = dev_change_net_namespace(dev, &init_net, fb_name);
6571 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6572 __func__, dev->name, err);
6579 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6581 /* At exit all network devices most be removed from a network
6582 * namespace. Do this in the reverse order of registration.
6583 * Do this across as many network namespaces as possible to
6584 * improve batching efficiency.
6586 struct net_device *dev;
6588 LIST_HEAD(dev_kill_list);
6591 list_for_each_entry(net, net_list, exit_list) {
6592 for_each_netdev_reverse(net, dev) {
6593 if (dev->rtnl_link_ops)
6594 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6596 unregister_netdevice_queue(dev, &dev_kill_list);
6599 unregister_netdevice_many(&dev_kill_list);
6600 list_del(&dev_kill_list);
6604 static struct pernet_operations __net_initdata default_device_ops = {
6605 .exit = default_device_exit,
6606 .exit_batch = default_device_exit_batch,
6610 * Initialize the DEV module. At boot time this walks the device list and
6611 * unhooks any devices that fail to initialise (normally hardware not
6612 * present) and leaves us with a valid list of present and active devices.
6617 * This is called single threaded during boot, so no need
6618 * to take the rtnl semaphore.
6620 static int __init net_dev_init(void)
6622 int i, rc = -ENOMEM;
6624 BUG_ON(!dev_boot_phase);
6626 if (dev_proc_init())
6629 if (netdev_kobject_init())
6632 INIT_LIST_HEAD(&ptype_all);
6633 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6634 INIT_LIST_HEAD(&ptype_base[i]);
6636 if (register_pernet_subsys(&netdev_net_ops))
6640 * Initialise the packet receive queues.
6643 for_each_possible_cpu(i) {
6644 struct softnet_data *sd = &per_cpu(softnet_data, i);
6646 memset(sd, 0, sizeof(*sd));
6647 skb_queue_head_init(&sd->input_pkt_queue);
6648 skb_queue_head_init(&sd->process_queue);
6649 sd->completion_queue = NULL;
6650 INIT_LIST_HEAD(&sd->poll_list);
6651 sd->output_queue = NULL;
6652 sd->output_queue_tailp = &sd->output_queue;
6654 sd->csd.func = rps_trigger_softirq;
6660 sd->backlog.poll = process_backlog;
6661 sd->backlog.weight = weight_p;
6662 sd->backlog.gro_list = NULL;
6663 sd->backlog.gro_count = 0;
6668 /* The loopback device is special if any other network devices
6669 * is present in a network namespace the loopback device must
6670 * be present. Since we now dynamically allocate and free the
6671 * loopback device ensure this invariant is maintained by
6672 * keeping the loopback device as the first device on the
6673 * list of network devices. Ensuring the loopback devices
6674 * is the first device that appears and the last network device
6677 if (register_pernet_device(&loopback_net_ops))
6680 if (register_pernet_device(&default_device_ops))
6683 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6684 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6686 hotcpu_notifier(dev_cpu_callback, 0);
6694 subsys_initcall(net_dev_init);
6696 static int __init initialize_hashrnd(void)
6698 get_random_bytes(&hashrnd, sizeof(hashrnd));
6702 late_initcall_sync(initialize_hashrnd);