2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/if_tunnel.h>
137 #include <linux/if_pppox.h>
138 #include <linux/ppp_defs.h>
139 #include <linux/net_tstamp.h>
141 #include "net-sysfs.h"
143 /* Instead of increasing this, you should create a hash table. */
144 #define MAX_GRO_SKBS 8
146 /* This should be increased if a protocol with a bigger head is added. */
147 #define GRO_MAX_HEAD (MAX_HEADER + 128)
150 * The list of packet types we will receive (as opposed to discard)
151 * and the routines to invoke.
153 * Why 16. Because with 16 the only overlap we get on a hash of the
154 * low nibble of the protocol value is RARP/SNAP/X.25.
156 * NOTE: That is no longer true with the addition of VLAN tags. Not
157 * sure which should go first, but I bet it won't make much
158 * difference if we are running VLANs. The good news is that
159 * this protocol won't be in the list unless compiled in, so
160 * the average user (w/out VLANs) will not be adversely affected.
177 #define PTYPE_HASH_SIZE (16)
178 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
180 static DEFINE_SPINLOCK(ptype_lock);
181 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
182 static struct list_head ptype_all __read_mostly; /* Taps */
185 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
188 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
190 * Writers must hold the rtnl semaphore while they loop through the
191 * dev_base_head list, and hold dev_base_lock for writing when they do the
192 * actual updates. This allows pure readers to access the list even
193 * while a writer is preparing to update it.
195 * To put it another way, dev_base_lock is held for writing only to
196 * protect against pure readers; the rtnl semaphore provides the
197 * protection against other writers.
199 * See, for example usages, register_netdevice() and
200 * unregister_netdevice(), which must be called with the rtnl
203 DEFINE_RWLOCK(dev_base_lock);
204 EXPORT_SYMBOL(dev_base_lock);
206 static inline void dev_base_seq_inc(struct net *net)
208 while (++net->dev_base_seq == 0);
211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
213 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
214 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
217 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
219 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
222 static inline void rps_lock(struct softnet_data *sd)
225 spin_lock(&sd->input_pkt_queue.lock);
229 static inline void rps_unlock(struct softnet_data *sd)
232 spin_unlock(&sd->input_pkt_queue.lock);
236 /* Device list insertion */
237 static int list_netdevice(struct net_device *dev)
239 struct net *net = dev_net(dev);
243 write_lock_bh(&dev_base_lock);
244 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
245 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
246 hlist_add_head_rcu(&dev->index_hlist,
247 dev_index_hash(net, dev->ifindex));
248 write_unlock_bh(&dev_base_lock);
250 dev_base_seq_inc(net);
255 /* Device list removal
256 * caller must respect a RCU grace period before freeing/reusing dev
258 static void unlist_netdevice(struct net_device *dev)
262 /* Unlink dev from the device chain */
263 write_lock_bh(&dev_base_lock);
264 list_del_rcu(&dev->dev_list);
265 hlist_del_rcu(&dev->name_hlist);
266 hlist_del_rcu(&dev->index_hlist);
267 write_unlock_bh(&dev_base_lock);
269 dev_base_seq_inc(dev_net(dev));
276 static RAW_NOTIFIER_HEAD(netdev_chain);
279 * Device drivers call our routines to queue packets here. We empty the
280 * queue in the local softnet handler.
283 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
284 EXPORT_PER_CPU_SYMBOL(softnet_data);
286 #ifdef CONFIG_LOCKDEP
288 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
289 * according to dev->type
291 static const unsigned short netdev_lock_type[] =
292 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
293 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
294 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
295 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
296 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
297 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
298 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
299 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
300 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
301 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
302 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
303 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
304 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
305 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
306 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
307 ARPHRD_VOID, ARPHRD_NONE};
309 static const char *const netdev_lock_name[] =
310 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
311 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
312 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
313 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
314 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
315 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
316 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
317 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
318 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
319 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
320 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
321 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
322 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
323 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
324 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
325 "_xmit_VOID", "_xmit_NONE"};
327 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
328 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
330 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
334 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
335 if (netdev_lock_type[i] == dev_type)
337 /* the last key is used by default */
338 return ARRAY_SIZE(netdev_lock_type) - 1;
341 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
342 unsigned short dev_type)
346 i = netdev_lock_pos(dev_type);
347 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
348 netdev_lock_name[i]);
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
355 i = netdev_lock_pos(dev->type);
356 lockdep_set_class_and_name(&dev->addr_list_lock,
357 &netdev_addr_lock_key[i],
358 netdev_lock_name[i]);
361 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
362 unsigned short dev_type)
365 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
370 /*******************************************************************************
372 Protocol management and registration routines
374 *******************************************************************************/
377 * Add a protocol ID to the list. Now that the input handler is
378 * smarter we can dispense with all the messy stuff that used to be
381 * BEWARE!!! Protocol handlers, mangling input packets,
382 * MUST BE last in hash buckets and checking protocol handlers
383 * MUST start from promiscuous ptype_all chain in net_bh.
384 * It is true now, do not change it.
385 * Explanation follows: if protocol handler, mangling packet, will
386 * be the first on list, it is not able to sense, that packet
387 * is cloned and should be copied-on-write, so that it will
388 * change it and subsequent readers will get broken packet.
392 static inline struct list_head *ptype_head(const struct packet_type *pt)
394 if (pt->type == htons(ETH_P_ALL))
397 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
401 * dev_add_pack - add packet handler
402 * @pt: packet type declaration
404 * Add a protocol handler to the networking stack. The passed &packet_type
405 * is linked into kernel lists and may not be freed until it has been
406 * removed from the kernel lists.
408 * This call does not sleep therefore it can not
409 * guarantee all CPU's that are in middle of receiving packets
410 * will see the new packet type (until the next received packet).
413 void dev_add_pack(struct packet_type *pt)
415 struct list_head *head = ptype_head(pt);
417 spin_lock(&ptype_lock);
418 list_add_rcu(&pt->list, head);
419 spin_unlock(&ptype_lock);
421 EXPORT_SYMBOL(dev_add_pack);
424 * __dev_remove_pack - remove packet handler
425 * @pt: packet type declaration
427 * Remove a protocol handler that was previously added to the kernel
428 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
429 * from the kernel lists and can be freed or reused once this function
432 * The packet type might still be in use by receivers
433 * and must not be freed until after all the CPU's have gone
434 * through a quiescent state.
436 void __dev_remove_pack(struct packet_type *pt)
438 struct list_head *head = ptype_head(pt);
439 struct packet_type *pt1;
441 spin_lock(&ptype_lock);
443 list_for_each_entry(pt1, head, list) {
445 list_del_rcu(&pt->list);
450 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
452 spin_unlock(&ptype_lock);
454 EXPORT_SYMBOL(__dev_remove_pack);
457 * dev_remove_pack - remove packet handler
458 * @pt: packet type declaration
460 * Remove a protocol handler that was previously added to the kernel
461 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
462 * from the kernel lists and can be freed or reused once this function
465 * This call sleeps to guarantee that no CPU is looking at the packet
468 void dev_remove_pack(struct packet_type *pt)
470 __dev_remove_pack(pt);
474 EXPORT_SYMBOL(dev_remove_pack);
476 /******************************************************************************
478 Device Boot-time Settings Routines
480 *******************************************************************************/
482 /* Boot time configuration table */
483 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
486 * netdev_boot_setup_add - add new setup entry
487 * @name: name of the device
488 * @map: configured settings for the device
490 * Adds new setup entry to the dev_boot_setup list. The function
491 * returns 0 on error and 1 on success. This is a generic routine to
494 static int netdev_boot_setup_add(char *name, struct ifmap *map)
496 struct netdev_boot_setup *s;
500 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
501 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
502 memset(s[i].name, 0, sizeof(s[i].name));
503 strlcpy(s[i].name, name, IFNAMSIZ);
504 memcpy(&s[i].map, map, sizeof(s[i].map));
509 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
513 * netdev_boot_setup_check - check boot time settings
514 * @dev: the netdevice
516 * Check boot time settings for the device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found, 1 if they are.
521 int netdev_boot_setup_check(struct net_device *dev)
523 struct netdev_boot_setup *s = dev_boot_setup;
526 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
527 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
528 !strcmp(dev->name, s[i].name)) {
529 dev->irq = s[i].map.irq;
530 dev->base_addr = s[i].map.base_addr;
531 dev->mem_start = s[i].map.mem_start;
532 dev->mem_end = s[i].map.mem_end;
538 EXPORT_SYMBOL(netdev_boot_setup_check);
542 * netdev_boot_base - get address from boot time settings
543 * @prefix: prefix for network device
544 * @unit: id for network device
546 * Check boot time settings for the base address of device.
547 * The found settings are set for the device to be used
548 * later in the device probing.
549 * Returns 0 if no settings found.
551 unsigned long netdev_boot_base(const char *prefix, int unit)
553 const struct netdev_boot_setup *s = dev_boot_setup;
557 sprintf(name, "%s%d", prefix, unit);
560 * If device already registered then return base of 1
561 * to indicate not to probe for this interface
563 if (__dev_get_by_name(&init_net, name))
566 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
567 if (!strcmp(name, s[i].name))
568 return s[i].map.base_addr;
573 * Saves at boot time configured settings for any netdevice.
575 int __init netdev_boot_setup(char *str)
580 str = get_options(str, ARRAY_SIZE(ints), ints);
585 memset(&map, 0, sizeof(map));
589 map.base_addr = ints[2];
591 map.mem_start = ints[3];
593 map.mem_end = ints[4];
595 /* Add new entry to the list */
596 return netdev_boot_setup_add(str, &map);
599 __setup("netdev=", netdev_boot_setup);
601 /*******************************************************************************
603 Device Interface Subroutines
605 *******************************************************************************/
608 * __dev_get_by_name - find a device by its name
609 * @net: the applicable net namespace
610 * @name: name to find
612 * Find an interface by name. Must be called under RTNL semaphore
613 * or @dev_base_lock. If the name is found a pointer to the device
614 * is returned. If the name is not found then %NULL is returned. The
615 * reference counters are not incremented so the caller must be
616 * careful with locks.
619 struct net_device *__dev_get_by_name(struct net *net, const char *name)
621 struct hlist_node *p;
622 struct net_device *dev;
623 struct hlist_head *head = dev_name_hash(net, name);
625 hlist_for_each_entry(dev, p, head, name_hlist)
626 if (!strncmp(dev->name, name, IFNAMSIZ))
631 EXPORT_SYMBOL(__dev_get_by_name);
634 * dev_get_by_name_rcu - find a device by its name
635 * @net: the applicable net namespace
636 * @name: name to find
638 * Find an interface by name.
639 * If the name is found a pointer to the device is returned.
640 * If the name is not found then %NULL is returned.
641 * The reference counters are not incremented so the caller must be
642 * careful with locks. The caller must hold RCU lock.
645 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
647 struct hlist_node *p;
648 struct net_device *dev;
649 struct hlist_head *head = dev_name_hash(net, name);
651 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
652 if (!strncmp(dev->name, name, IFNAMSIZ))
657 EXPORT_SYMBOL(dev_get_by_name_rcu);
660 * dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. This can be called from any
665 * context and does its own locking. The returned handle has
666 * the usage count incremented and the caller must use dev_put() to
667 * release it when it is no longer needed. %NULL is returned if no
668 * matching device is found.
671 struct net_device *dev_get_by_name(struct net *net, const char *name)
673 struct net_device *dev;
676 dev = dev_get_by_name_rcu(net, name);
682 EXPORT_SYMBOL(dev_get_by_name);
685 * __dev_get_by_index - find a device by its ifindex
686 * @net: the applicable net namespace
687 * @ifindex: index of device
689 * Search for an interface by index. Returns %NULL if the device
690 * is not found or a pointer to the device. The device has not
691 * had its reference counter increased so the caller must be careful
692 * about locking. The caller must hold either the RTNL semaphore
696 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
698 struct hlist_node *p;
699 struct net_device *dev;
700 struct hlist_head *head = dev_index_hash(net, ifindex);
702 hlist_for_each_entry(dev, p, head, index_hlist)
703 if (dev->ifindex == ifindex)
708 EXPORT_SYMBOL(__dev_get_by_index);
711 * dev_get_by_index_rcu - find a device by its ifindex
712 * @net: the applicable net namespace
713 * @ifindex: index of device
715 * Search for an interface by index. Returns %NULL if the device
716 * is not found or a pointer to the device. The device has not
717 * had its reference counter increased so the caller must be careful
718 * about locking. The caller must hold RCU lock.
721 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
723 struct hlist_node *p;
724 struct net_device *dev;
725 struct hlist_head *head = dev_index_hash(net, ifindex);
727 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
728 if (dev->ifindex == ifindex)
733 EXPORT_SYMBOL(dev_get_by_index_rcu);
737 * dev_get_by_index - find a device by its ifindex
738 * @net: the applicable net namespace
739 * @ifindex: index of device
741 * Search for an interface by index. Returns NULL if the device
742 * is not found or a pointer to the device. The device returned has
743 * had a reference added and the pointer is safe until the user calls
744 * dev_put to indicate they have finished with it.
747 struct net_device *dev_get_by_index(struct net *net, int ifindex)
749 struct net_device *dev;
752 dev = dev_get_by_index_rcu(net, ifindex);
758 EXPORT_SYMBOL(dev_get_by_index);
761 * dev_getbyhwaddr_rcu - find a device by its hardware address
762 * @net: the applicable net namespace
763 * @type: media type of device
764 * @ha: hardware address
766 * Search for an interface by MAC address. Returns NULL if the device
767 * is not found or a pointer to the device.
768 * The caller must hold RCU or RTNL.
769 * The returned device has not had its ref count increased
770 * and the caller must therefore be careful about locking
774 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
777 struct net_device *dev;
779 for_each_netdev_rcu(net, dev)
780 if (dev->type == type &&
781 !memcmp(dev->dev_addr, ha, dev->addr_len))
786 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
788 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
790 struct net_device *dev;
793 for_each_netdev(net, dev)
794 if (dev->type == type)
799 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
801 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
803 struct net_device *dev, *ret = NULL;
806 for_each_netdev_rcu(net, dev)
807 if (dev->type == type) {
815 EXPORT_SYMBOL(dev_getfirstbyhwtype);
818 * dev_get_by_flags_rcu - find any device with given flags
819 * @net: the applicable net namespace
820 * @if_flags: IFF_* values
821 * @mask: bitmask of bits in if_flags to check
823 * Search for any interface with the given flags. Returns NULL if a device
824 * is not found or a pointer to the device. Must be called inside
825 * rcu_read_lock(), and result refcount is unchanged.
828 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
831 struct net_device *dev, *ret;
834 for_each_netdev_rcu(net, dev) {
835 if (((dev->flags ^ if_flags) & mask) == 0) {
842 EXPORT_SYMBOL(dev_get_by_flags_rcu);
845 * dev_valid_name - check if name is okay for network device
848 * Network device names need to be valid file names to
849 * to allow sysfs to work. We also disallow any kind of
852 int dev_valid_name(const char *name)
856 if (strlen(name) >= IFNAMSIZ)
858 if (!strcmp(name, ".") || !strcmp(name, ".."))
862 if (*name == '/' || isspace(*name))
868 EXPORT_SYMBOL(dev_valid_name);
871 * __dev_alloc_name - allocate a name for a device
872 * @net: network namespace to allocate the device name in
873 * @name: name format string
874 * @buf: scratch buffer and result name string
876 * Passed a format string - eg "lt%d" it will try and find a suitable
877 * id. It scans list of devices to build up a free map, then chooses
878 * the first empty slot. The caller must hold the dev_base or rtnl lock
879 * while allocating the name and adding the device in order to avoid
881 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
882 * Returns the number of the unit assigned or a negative errno code.
885 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
889 const int max_netdevices = 8*PAGE_SIZE;
890 unsigned long *inuse;
891 struct net_device *d;
893 p = strnchr(name, IFNAMSIZ-1, '%');
896 * Verify the string as this thing may have come from
897 * the user. There must be either one "%d" and no other "%"
900 if (p[1] != 'd' || strchr(p + 2, '%'))
903 /* Use one page as a bit array of possible slots */
904 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
908 for_each_netdev(net, d) {
909 if (!sscanf(d->name, name, &i))
911 if (i < 0 || i >= max_netdevices)
914 /* avoid cases where sscanf is not exact inverse of printf */
915 snprintf(buf, IFNAMSIZ, name, i);
916 if (!strncmp(buf, d->name, IFNAMSIZ))
920 i = find_first_zero_bit(inuse, max_netdevices);
921 free_page((unsigned long) inuse);
925 snprintf(buf, IFNAMSIZ, name, i);
926 if (!__dev_get_by_name(net, buf))
929 /* It is possible to run out of possible slots
930 * when the name is long and there isn't enough space left
931 * for the digits, or if all bits are used.
937 * dev_alloc_name - allocate a name for a device
939 * @name: name format string
941 * Passed a format string - eg "lt%d" it will try and find a suitable
942 * id. It scans list of devices to build up a free map, then chooses
943 * the first empty slot. The caller must hold the dev_base or rtnl lock
944 * while allocating the name and adding the device in order to avoid
946 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
947 * Returns the number of the unit assigned or a negative errno code.
950 int dev_alloc_name(struct net_device *dev, const char *name)
956 BUG_ON(!dev_net(dev));
958 ret = __dev_alloc_name(net, name, buf);
960 strlcpy(dev->name, buf, IFNAMSIZ);
963 EXPORT_SYMBOL(dev_alloc_name);
965 static int dev_get_valid_name(struct net_device *dev, const char *name)
969 BUG_ON(!dev_net(dev));
972 if (!dev_valid_name(name))
975 if (strchr(name, '%'))
976 return dev_alloc_name(dev, name);
977 else if (__dev_get_by_name(net, name))
979 else if (dev->name != name)
980 strlcpy(dev->name, name, IFNAMSIZ);
986 * dev_change_name - change name of a device
988 * @newname: name (or format string) must be at least IFNAMSIZ
990 * Change name of a device, can pass format strings "eth%d".
993 int dev_change_name(struct net_device *dev, const char *newname)
995 char oldname[IFNAMSIZ];
1001 BUG_ON(!dev_net(dev));
1004 if (dev->flags & IFF_UP)
1007 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1010 memcpy(oldname, dev->name, IFNAMSIZ);
1012 err = dev_get_valid_name(dev, newname);
1017 ret = device_rename(&dev->dev, dev->name);
1019 memcpy(dev->name, oldname, IFNAMSIZ);
1023 write_lock_bh(&dev_base_lock);
1024 hlist_del_rcu(&dev->name_hlist);
1025 write_unlock_bh(&dev_base_lock);
1029 write_lock_bh(&dev_base_lock);
1030 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1031 write_unlock_bh(&dev_base_lock);
1033 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1034 ret = notifier_to_errno(ret);
1037 /* err >= 0 after dev_alloc_name() or stores the first errno */
1040 memcpy(dev->name, oldname, IFNAMSIZ);
1044 "%s: name change rollback failed: %d.\n",
1053 * dev_set_alias - change ifalias of a device
1055 * @alias: name up to IFALIASZ
1056 * @len: limit of bytes to copy from info
1058 * Set ifalias for a device,
1060 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1064 if (len >= IFALIASZ)
1069 kfree(dev->ifalias);
1070 dev->ifalias = NULL;
1075 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1079 strlcpy(dev->ifalias, alias, len+1);
1085 * netdev_features_change - device changes features
1086 * @dev: device to cause notification
1088 * Called to indicate a device has changed features.
1090 void netdev_features_change(struct net_device *dev)
1092 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1094 EXPORT_SYMBOL(netdev_features_change);
1097 * netdev_state_change - device changes state
1098 * @dev: device to cause notification
1100 * Called to indicate a device has changed state. This function calls
1101 * the notifier chains for netdev_chain and sends a NEWLINK message
1102 * to the routing socket.
1104 void netdev_state_change(struct net_device *dev)
1106 if (dev->flags & IFF_UP) {
1107 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1108 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1111 EXPORT_SYMBOL(netdev_state_change);
1113 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1115 return call_netdevice_notifiers(event, dev);
1117 EXPORT_SYMBOL(netdev_bonding_change);
1120 * dev_load - load a network module
1121 * @net: the applicable net namespace
1122 * @name: name of interface
1124 * If a network interface is not present and the process has suitable
1125 * privileges this function loads the module. If module loading is not
1126 * available in this kernel then it becomes a nop.
1129 void dev_load(struct net *net, const char *name)
1131 struct net_device *dev;
1135 dev = dev_get_by_name_rcu(net, name);
1139 if (no_module && capable(CAP_NET_ADMIN))
1140 no_module = request_module("netdev-%s", name);
1141 if (no_module && capable(CAP_SYS_MODULE)) {
1142 if (!request_module("%s", name))
1143 pr_err("Loading kernel module for a network device "
1144 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1148 EXPORT_SYMBOL(dev_load);
1150 static int __dev_open(struct net_device *dev)
1152 const struct net_device_ops *ops = dev->netdev_ops;
1157 if (!netif_device_present(dev))
1160 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1161 ret = notifier_to_errno(ret);
1165 set_bit(__LINK_STATE_START, &dev->state);
1167 if (ops->ndo_validate_addr)
1168 ret = ops->ndo_validate_addr(dev);
1170 if (!ret && ops->ndo_open)
1171 ret = ops->ndo_open(dev);
1174 clear_bit(__LINK_STATE_START, &dev->state);
1176 dev->flags |= IFF_UP;
1177 net_dmaengine_get();
1178 dev_set_rx_mode(dev);
1180 add_device_randomness(dev->dev_addr, dev->addr_len);
1187 * dev_open - prepare an interface for use.
1188 * @dev: device to open
1190 * Takes a device from down to up state. The device's private open
1191 * function is invoked and then the multicast lists are loaded. Finally
1192 * the device is moved into the up state and a %NETDEV_UP message is
1193 * sent to the netdev notifier chain.
1195 * Calling this function on an active interface is a nop. On a failure
1196 * a negative errno code is returned.
1198 int dev_open(struct net_device *dev)
1202 if (dev->flags & IFF_UP)
1205 ret = __dev_open(dev);
1209 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1210 call_netdevice_notifiers(NETDEV_UP, dev);
1214 EXPORT_SYMBOL(dev_open);
1216 static int __dev_close_many(struct list_head *head)
1218 struct net_device *dev;
1223 list_for_each_entry(dev, head, unreg_list) {
1224 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1226 clear_bit(__LINK_STATE_START, &dev->state);
1228 /* Synchronize to scheduled poll. We cannot touch poll list, it
1229 * can be even on different cpu. So just clear netif_running().
1231 * dev->stop() will invoke napi_disable() on all of it's
1232 * napi_struct instances on this device.
1234 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1237 dev_deactivate_many(head);
1239 list_for_each_entry(dev, head, unreg_list) {
1240 const struct net_device_ops *ops = dev->netdev_ops;
1243 * Call the device specific close. This cannot fail.
1244 * Only if device is UP
1246 * We allow it to be called even after a DETACH hot-plug
1252 dev->flags &= ~IFF_UP;
1253 net_dmaengine_put();
1259 static int __dev_close(struct net_device *dev)
1264 list_add(&dev->unreg_list, &single);
1265 retval = __dev_close_many(&single);
1270 static int dev_close_many(struct list_head *head)
1272 struct net_device *dev, *tmp;
1273 LIST_HEAD(tmp_list);
1275 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1276 if (!(dev->flags & IFF_UP))
1277 list_move(&dev->unreg_list, &tmp_list);
1279 __dev_close_many(head);
1281 list_for_each_entry(dev, head, unreg_list) {
1282 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1283 call_netdevice_notifiers(NETDEV_DOWN, dev);
1286 /* rollback_registered_many needs the complete original list */
1287 list_splice(&tmp_list, head);
1292 * dev_close - shutdown an interface.
1293 * @dev: device to shutdown
1295 * This function moves an active device into down state. A
1296 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1297 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1300 int dev_close(struct net_device *dev)
1302 if (dev->flags & IFF_UP) {
1305 list_add(&dev->unreg_list, &single);
1306 dev_close_many(&single);
1311 EXPORT_SYMBOL(dev_close);
1315 * dev_disable_lro - disable Large Receive Offload on a device
1318 * Disable Large Receive Offload (LRO) on a net device. Must be
1319 * called under RTNL. This is needed if received packets may be
1320 * forwarded to another interface.
1322 void dev_disable_lro(struct net_device *dev)
1327 * If we're trying to disable lro on a vlan device
1328 * use the underlying physical device instead
1330 if (is_vlan_dev(dev))
1331 dev = vlan_dev_real_dev(dev);
1333 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1334 flags = dev->ethtool_ops->get_flags(dev);
1336 flags = ethtool_op_get_flags(dev);
1338 if (!(flags & ETH_FLAG_LRO))
1341 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1342 if (unlikely(dev->features & NETIF_F_LRO))
1343 netdev_WARN(dev, "failed to disable LRO!\n");
1345 EXPORT_SYMBOL(dev_disable_lro);
1348 static int dev_boot_phase = 1;
1351 * register_netdevice_notifier - register a network notifier block
1354 * Register a notifier to be called when network device events occur.
1355 * The notifier passed is linked into the kernel structures and must
1356 * not be reused until it has been unregistered. A negative errno code
1357 * is returned on a failure.
1359 * When registered all registration and up events are replayed
1360 * to the new notifier to allow device to have a race free
1361 * view of the network device list.
1364 int register_netdevice_notifier(struct notifier_block *nb)
1366 struct net_device *dev;
1367 struct net_device *last;
1372 err = raw_notifier_chain_register(&netdev_chain, nb);
1378 for_each_netdev(net, dev) {
1379 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1380 err = notifier_to_errno(err);
1384 if (!(dev->flags & IFF_UP))
1387 nb->notifier_call(nb, NETDEV_UP, dev);
1398 for_each_netdev(net, dev) {
1402 if (dev->flags & IFF_UP) {
1403 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1404 nb->notifier_call(nb, NETDEV_DOWN, dev);
1406 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1407 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1412 raw_notifier_chain_unregister(&netdev_chain, nb);
1415 EXPORT_SYMBOL(register_netdevice_notifier);
1418 * unregister_netdevice_notifier - unregister a network notifier block
1421 * Unregister a notifier previously registered by
1422 * register_netdevice_notifier(). The notifier is unlinked into the
1423 * kernel structures and may then be reused. A negative errno code
1424 * is returned on a failure.
1426 * After unregistering unregister and down device events are synthesized
1427 * for all devices on the device list to the removed notifier to remove
1428 * the need for special case cleanup code.
1431 int unregister_netdevice_notifier(struct notifier_block *nb)
1433 struct net_device *dev;
1438 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1443 for_each_netdev(net, dev) {
1444 if (dev->flags & IFF_UP) {
1445 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1446 nb->notifier_call(nb, NETDEV_DOWN, dev);
1448 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1449 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1456 EXPORT_SYMBOL(unregister_netdevice_notifier);
1459 * call_netdevice_notifiers - call all network notifier blocks
1460 * @val: value passed unmodified to notifier function
1461 * @dev: net_device pointer passed unmodified to notifier function
1463 * Call all network notifier blocks. Parameters and return value
1464 * are as for raw_notifier_call_chain().
1467 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1470 return raw_notifier_call_chain(&netdev_chain, val, dev);
1472 EXPORT_SYMBOL(call_netdevice_notifiers);
1474 /* When > 0 there are consumers of rx skb time stamps */
1475 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1477 void net_enable_timestamp(void)
1479 atomic_inc(&netstamp_needed);
1481 EXPORT_SYMBOL(net_enable_timestamp);
1483 void net_disable_timestamp(void)
1485 atomic_dec(&netstamp_needed);
1487 EXPORT_SYMBOL(net_disable_timestamp);
1489 static inline void net_timestamp_set(struct sk_buff *skb)
1491 if (atomic_read(&netstamp_needed))
1492 __net_timestamp(skb);
1494 skb->tstamp.tv64 = 0;
1497 static inline void net_timestamp_check(struct sk_buff *skb)
1499 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1500 __net_timestamp(skb);
1503 static int net_hwtstamp_validate(struct ifreq *ifr)
1505 struct hwtstamp_config cfg;
1506 enum hwtstamp_tx_types tx_type;
1507 enum hwtstamp_rx_filters rx_filter;
1508 int tx_type_valid = 0;
1509 int rx_filter_valid = 0;
1511 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1514 if (cfg.flags) /* reserved for future extensions */
1517 tx_type = cfg.tx_type;
1518 rx_filter = cfg.rx_filter;
1521 case HWTSTAMP_TX_OFF:
1522 case HWTSTAMP_TX_ON:
1523 case HWTSTAMP_TX_ONESTEP_SYNC:
1528 switch (rx_filter) {
1529 case HWTSTAMP_FILTER_NONE:
1530 case HWTSTAMP_FILTER_ALL:
1531 case HWTSTAMP_FILTER_SOME:
1532 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1533 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1534 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1535 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1536 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1537 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1538 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1539 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1540 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1541 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1542 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1543 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1544 rx_filter_valid = 1;
1548 if (!tx_type_valid || !rx_filter_valid)
1554 static inline bool is_skb_forwardable(struct net_device *dev,
1555 struct sk_buff *skb)
1559 if (!(dev->flags & IFF_UP))
1562 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1563 if (skb->len <= len)
1566 /* if TSO is enabled, we don't care about the length as the packet
1567 * could be forwarded without being segmented before
1569 if (skb_is_gso(skb))
1576 * dev_forward_skb - loopback an skb to another netif
1578 * @dev: destination network device
1579 * @skb: buffer to forward
1582 * NET_RX_SUCCESS (no congestion)
1583 * NET_RX_DROP (packet was dropped, but freed)
1585 * dev_forward_skb can be used for injecting an skb from the
1586 * start_xmit function of one device into the receive queue
1587 * of another device.
1589 * The receiving device may be in another namespace, so
1590 * we have to clear all information in the skb that could
1591 * impact namespace isolation.
1593 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1595 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1596 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1597 atomic_long_inc(&dev->rx_dropped);
1606 if (unlikely(!is_skb_forwardable(dev, skb))) {
1607 atomic_long_inc(&dev->rx_dropped);
1613 skb->tstamp.tv64 = 0;
1614 skb->pkt_type = PACKET_HOST;
1615 skb->protocol = eth_type_trans(skb, dev);
1619 return netif_rx(skb);
1621 EXPORT_SYMBOL_GPL(dev_forward_skb);
1623 static inline int deliver_skb(struct sk_buff *skb,
1624 struct packet_type *pt_prev,
1625 struct net_device *orig_dev)
1627 atomic_inc(&skb->users);
1628 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1632 * Support routine. Sends outgoing frames to any network
1633 * taps currently in use.
1636 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1638 struct packet_type *ptype;
1639 struct sk_buff *skb2 = NULL;
1640 struct packet_type *pt_prev = NULL;
1643 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1644 /* Never send packets back to the socket
1645 * they originated from - MvS (miquels@drinkel.ow.org)
1647 if ((ptype->dev == dev || !ptype->dev) &&
1648 (ptype->af_packet_priv == NULL ||
1649 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1651 deliver_skb(skb2, pt_prev, skb->dev);
1656 skb2 = skb_clone(skb, GFP_ATOMIC);
1660 net_timestamp_set(skb2);
1662 /* skb->nh should be correctly
1663 set by sender, so that the second statement is
1664 just protection against buggy protocols.
1666 skb_reset_mac_header(skb2);
1668 if (skb_network_header(skb2) < skb2->data ||
1669 skb2->network_header > skb2->tail) {
1670 if (net_ratelimit())
1671 printk(KERN_CRIT "protocol %04x is "
1673 ntohs(skb2->protocol),
1675 skb_reset_network_header(skb2);
1678 skb2->transport_header = skb2->network_header;
1679 skb2->pkt_type = PACKET_OUTGOING;
1684 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1688 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1689 * @dev: Network device
1690 * @txq: number of queues available
1692 * If real_num_tx_queues is changed the tc mappings may no longer be
1693 * valid. To resolve this verify the tc mapping remains valid and if
1694 * not NULL the mapping. With no priorities mapping to this
1695 * offset/count pair it will no longer be used. In the worst case TC0
1696 * is invalid nothing can be done so disable priority mappings. If is
1697 * expected that drivers will fix this mapping if they can before
1698 * calling netif_set_real_num_tx_queues.
1700 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1703 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1705 /* If TC0 is invalidated disable TC mapping */
1706 if (tc->offset + tc->count > txq) {
1707 pr_warning("Number of in use tx queues changed "
1708 "invalidating tc mappings. Priority "
1709 "traffic classification disabled!\n");
1714 /* Invalidated prio to tc mappings set to TC0 */
1715 for (i = 1; i < TC_BITMASK + 1; i++) {
1716 int q = netdev_get_prio_tc_map(dev, i);
1718 tc = &dev->tc_to_txq[q];
1719 if (tc->offset + tc->count > txq) {
1720 pr_warning("Number of in use tx queues "
1721 "changed. Priority %i to tc "
1722 "mapping %i is no longer valid "
1723 "setting map to 0\n",
1725 netdev_set_prio_tc_map(dev, i, 0);
1731 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1732 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1734 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1738 if (txq < 1 || txq > dev->num_tx_queues)
1741 if (dev->reg_state == NETREG_REGISTERED ||
1742 dev->reg_state == NETREG_UNREGISTERING) {
1745 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1751 netif_setup_tc(dev, txq);
1753 if (txq < dev->real_num_tx_queues)
1754 qdisc_reset_all_tx_gt(dev, txq);
1757 dev->real_num_tx_queues = txq;
1760 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1764 * netif_set_real_num_rx_queues - set actual number of RX queues used
1765 * @dev: Network device
1766 * @rxq: Actual number of RX queues
1768 * This must be called either with the rtnl_lock held or before
1769 * registration of the net device. Returns 0 on success, or a
1770 * negative error code. If called before registration, it always
1773 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1777 if (rxq < 1 || rxq > dev->num_rx_queues)
1780 if (dev->reg_state == NETREG_REGISTERED) {
1783 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1789 dev->real_num_rx_queues = rxq;
1792 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1795 static inline void __netif_reschedule(struct Qdisc *q)
1797 struct softnet_data *sd;
1798 unsigned long flags;
1800 local_irq_save(flags);
1801 sd = &__get_cpu_var(softnet_data);
1802 q->next_sched = NULL;
1803 *sd->output_queue_tailp = q;
1804 sd->output_queue_tailp = &q->next_sched;
1805 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1806 local_irq_restore(flags);
1809 void __netif_schedule(struct Qdisc *q)
1811 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1812 __netif_reschedule(q);
1814 EXPORT_SYMBOL(__netif_schedule);
1816 void dev_kfree_skb_irq(struct sk_buff *skb)
1818 if (atomic_dec_and_test(&skb->users)) {
1819 struct softnet_data *sd;
1820 unsigned long flags;
1822 local_irq_save(flags);
1823 sd = &__get_cpu_var(softnet_data);
1824 skb->next = sd->completion_queue;
1825 sd->completion_queue = skb;
1826 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1827 local_irq_restore(flags);
1830 EXPORT_SYMBOL(dev_kfree_skb_irq);
1832 void dev_kfree_skb_any(struct sk_buff *skb)
1834 if (in_irq() || irqs_disabled())
1835 dev_kfree_skb_irq(skb);
1839 EXPORT_SYMBOL(dev_kfree_skb_any);
1843 * netif_device_detach - mark device as removed
1844 * @dev: network device
1846 * Mark device as removed from system and therefore no longer available.
1848 void netif_device_detach(struct net_device *dev)
1850 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1851 netif_running(dev)) {
1852 netif_tx_stop_all_queues(dev);
1855 EXPORT_SYMBOL(netif_device_detach);
1858 * netif_device_attach - mark device as attached
1859 * @dev: network device
1861 * Mark device as attached from system and restart if needed.
1863 void netif_device_attach(struct net_device *dev)
1865 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1866 netif_running(dev)) {
1867 netif_tx_wake_all_queues(dev);
1868 __netdev_watchdog_up(dev);
1871 EXPORT_SYMBOL(netif_device_attach);
1874 * Invalidate hardware checksum when packet is to be mangled, and
1875 * complete checksum manually on outgoing path.
1877 int skb_checksum_help(struct sk_buff *skb)
1880 int ret = 0, offset;
1882 if (skb->ip_summed == CHECKSUM_COMPLETE)
1883 goto out_set_summed;
1885 if (unlikely(skb_shinfo(skb)->gso_size)) {
1886 /* Let GSO fix up the checksum. */
1887 goto out_set_summed;
1890 offset = skb_checksum_start_offset(skb);
1891 BUG_ON(offset >= skb_headlen(skb));
1892 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1894 offset += skb->csum_offset;
1895 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1897 if (skb_cloned(skb) &&
1898 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1899 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1904 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1906 skb->ip_summed = CHECKSUM_NONE;
1910 EXPORT_SYMBOL(skb_checksum_help);
1913 * skb_gso_segment - Perform segmentation on skb.
1914 * @skb: buffer to segment
1915 * @features: features for the output path (see dev->features)
1917 * This function segments the given skb and returns a list of segments.
1919 * It may return NULL if the skb requires no segmentation. This is
1920 * only possible when GSO is used for verifying header integrity.
1922 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1924 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1925 struct packet_type *ptype;
1926 __be16 type = skb->protocol;
1927 int vlan_depth = ETH_HLEN;
1930 while (type == htons(ETH_P_8021Q)) {
1931 struct vlan_hdr *vh;
1933 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1934 return ERR_PTR(-EINVAL);
1936 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1937 type = vh->h_vlan_encapsulated_proto;
1938 vlan_depth += VLAN_HLEN;
1941 skb_reset_mac_header(skb);
1942 skb->mac_len = skb->network_header - skb->mac_header;
1943 __skb_pull(skb, skb->mac_len);
1945 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1946 struct net_device *dev = skb->dev;
1947 struct ethtool_drvinfo info = {};
1949 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1950 dev->ethtool_ops->get_drvinfo(dev, &info);
1952 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1953 info.driver, dev ? dev->features : 0L,
1954 skb->sk ? skb->sk->sk_route_caps : 0L,
1955 skb->len, skb->data_len, skb->ip_summed);
1957 if (skb_header_cloned(skb) &&
1958 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1959 return ERR_PTR(err);
1963 list_for_each_entry_rcu(ptype,
1964 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1965 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1966 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1967 err = ptype->gso_send_check(skb);
1968 segs = ERR_PTR(err);
1969 if (err || skb_gso_ok(skb, features))
1971 __skb_push(skb, (skb->data -
1972 skb_network_header(skb)));
1974 segs = ptype->gso_segment(skb, features);
1980 __skb_push(skb, skb->data - skb_mac_header(skb));
1984 EXPORT_SYMBOL(skb_gso_segment);
1986 /* Take action when hardware reception checksum errors are detected. */
1988 void netdev_rx_csum_fault(struct net_device *dev)
1990 if (net_ratelimit()) {
1991 printk(KERN_ERR "%s: hw csum failure.\n",
1992 dev ? dev->name : "<unknown>");
1996 EXPORT_SYMBOL(netdev_rx_csum_fault);
1999 /* Actually, we should eliminate this check as soon as we know, that:
2000 * 1. IOMMU is present and allows to map all the memory.
2001 * 2. No high memory really exists on this machine.
2004 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2006 #ifdef CONFIG_HIGHMEM
2008 if (!(dev->features & NETIF_F_HIGHDMA)) {
2009 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2010 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2011 if (PageHighMem(skb_frag_page(frag)))
2016 if (PCI_DMA_BUS_IS_PHYS) {
2017 struct device *pdev = dev->dev.parent;
2021 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2022 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2023 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2024 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2033 void (*destructor)(struct sk_buff *skb);
2036 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2038 static void dev_gso_skb_destructor(struct sk_buff *skb)
2040 struct dev_gso_cb *cb;
2043 struct sk_buff *nskb = skb->next;
2045 skb->next = nskb->next;
2048 } while (skb->next);
2050 cb = DEV_GSO_CB(skb);
2052 cb->destructor(skb);
2056 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2057 * @skb: buffer to segment
2058 * @features: device features as applicable to this skb
2060 * This function segments the given skb and stores the list of segments
2063 static int dev_gso_segment(struct sk_buff *skb, int features)
2065 struct sk_buff *segs;
2067 segs = skb_gso_segment(skb, features);
2069 /* Verifying header integrity only. */
2074 return PTR_ERR(segs);
2077 DEV_GSO_CB(skb)->destructor = skb->destructor;
2078 skb->destructor = dev_gso_skb_destructor;
2083 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2085 return ((features & NETIF_F_GEN_CSUM) ||
2086 ((features & NETIF_F_V4_CSUM) &&
2087 protocol == htons(ETH_P_IP)) ||
2088 ((features & NETIF_F_V6_CSUM) &&
2089 protocol == htons(ETH_P_IPV6)) ||
2090 ((features & NETIF_F_FCOE_CRC) &&
2091 protocol == htons(ETH_P_FCOE)));
2094 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2096 if (!can_checksum_protocol(features, protocol)) {
2097 features &= ~NETIF_F_ALL_CSUM;
2098 features &= ~NETIF_F_SG;
2099 } else if (illegal_highdma(skb->dev, skb)) {
2100 features &= ~NETIF_F_SG;
2106 u32 netif_skb_features(struct sk_buff *skb)
2108 __be16 protocol = skb->protocol;
2109 u32 features = skb->dev->features;
2111 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2112 features &= ~NETIF_F_GSO_MASK;
2114 if (protocol == htons(ETH_P_8021Q)) {
2115 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2116 protocol = veh->h_vlan_encapsulated_proto;
2117 } else if (!vlan_tx_tag_present(skb)) {
2118 return harmonize_features(skb, protocol, features);
2121 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2123 if (protocol != htons(ETH_P_8021Q)) {
2124 return harmonize_features(skb, protocol, features);
2126 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2127 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2128 return harmonize_features(skb, protocol, features);
2131 EXPORT_SYMBOL(netif_skb_features);
2134 * Returns true if either:
2135 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2136 * 2. skb is fragmented and the device does not support SG, or if
2137 * at least one of fragments is in highmem and device does not
2138 * support DMA from it.
2140 static inline int skb_needs_linearize(struct sk_buff *skb,
2143 return skb_is_nonlinear(skb) &&
2144 ((skb_has_frag_list(skb) &&
2145 !(features & NETIF_F_FRAGLIST)) ||
2146 (skb_shinfo(skb)->nr_frags &&
2147 !(features & NETIF_F_SG)));
2150 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2151 struct netdev_queue *txq)
2153 const struct net_device_ops *ops = dev->netdev_ops;
2154 int rc = NETDEV_TX_OK;
2155 unsigned int skb_len;
2157 if (likely(!skb->next)) {
2161 * If device doesn't need skb->dst, release it right now while
2162 * its hot in this cpu cache
2164 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2167 if (!list_empty(&ptype_all))
2168 dev_queue_xmit_nit(skb, dev);
2170 features = netif_skb_features(skb);
2172 if (vlan_tx_tag_present(skb) &&
2173 !(features & NETIF_F_HW_VLAN_TX)) {
2174 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2181 if (netif_needs_gso(skb, features)) {
2182 if (unlikely(dev_gso_segment(skb, features)))
2187 if (skb_needs_linearize(skb, features) &&
2188 __skb_linearize(skb))
2191 /* If packet is not checksummed and device does not
2192 * support checksumming for this protocol, complete
2193 * checksumming here.
2195 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2196 skb_set_transport_header(skb,
2197 skb_checksum_start_offset(skb));
2198 if (!(features & NETIF_F_ALL_CSUM) &&
2199 skb_checksum_help(skb))
2205 rc = ops->ndo_start_xmit(skb, dev);
2206 trace_net_dev_xmit(skb, rc, dev, skb_len);
2207 if (rc == NETDEV_TX_OK)
2208 txq_trans_update(txq);
2214 struct sk_buff *nskb = skb->next;
2216 skb->next = nskb->next;
2220 * If device doesn't need nskb->dst, release it right now while
2221 * its hot in this cpu cache
2223 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2226 skb_len = nskb->len;
2227 rc = ops->ndo_start_xmit(nskb, dev);
2228 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2229 if (unlikely(rc != NETDEV_TX_OK)) {
2230 if (rc & ~NETDEV_TX_MASK)
2231 goto out_kfree_gso_skb;
2232 nskb->next = skb->next;
2236 txq_trans_update(txq);
2237 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2238 return NETDEV_TX_BUSY;
2239 } while (skb->next);
2242 if (likely(skb->next == NULL))
2243 skb->destructor = DEV_GSO_CB(skb)->destructor;
2250 static u32 hashrnd __read_mostly;
2253 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2254 * to be used as a distribution range.
2256 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2257 unsigned int num_tx_queues)
2261 u16 qcount = num_tx_queues;
2263 if (skb_rx_queue_recorded(skb)) {
2264 hash = skb_get_rx_queue(skb);
2265 while (unlikely(hash >= num_tx_queues))
2266 hash -= num_tx_queues;
2271 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2272 qoffset = dev->tc_to_txq[tc].offset;
2273 qcount = dev->tc_to_txq[tc].count;
2276 if (skb->sk && skb->sk->sk_hash)
2277 hash = skb->sk->sk_hash;
2279 hash = (__force u16) skb->protocol;
2280 hash = jhash_1word(hash, hashrnd);
2282 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2284 EXPORT_SYMBOL(__skb_tx_hash);
2286 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2288 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2289 if (net_ratelimit()) {
2290 pr_warning("%s selects TX queue %d, but "
2291 "real number of TX queues is %d\n",
2292 dev->name, queue_index, dev->real_num_tx_queues);
2299 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2302 struct xps_dev_maps *dev_maps;
2303 struct xps_map *map;
2304 int queue_index = -1;
2307 dev_maps = rcu_dereference(dev->xps_maps);
2309 map = rcu_dereference(
2310 dev_maps->cpu_map[raw_smp_processor_id()]);
2313 queue_index = map->queues[0];
2316 if (skb->sk && skb->sk->sk_hash)
2317 hash = skb->sk->sk_hash;
2319 hash = (__force u16) skb->protocol ^
2321 hash = jhash_1word(hash, hashrnd);
2322 queue_index = map->queues[
2323 ((u64)hash * map->len) >> 32];
2325 if (unlikely(queue_index >= dev->real_num_tx_queues))
2337 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2338 struct sk_buff *skb)
2341 const struct net_device_ops *ops = dev->netdev_ops;
2343 if (dev->real_num_tx_queues == 1)
2345 else if (ops->ndo_select_queue) {
2346 queue_index = ops->ndo_select_queue(dev, skb);
2347 queue_index = dev_cap_txqueue(dev, queue_index);
2349 struct sock *sk = skb->sk;
2350 queue_index = sk_tx_queue_get(sk);
2352 if (queue_index < 0 || skb->ooo_okay ||
2353 queue_index >= dev->real_num_tx_queues) {
2354 int old_index = queue_index;
2356 queue_index = get_xps_queue(dev, skb);
2357 if (queue_index < 0)
2358 queue_index = skb_tx_hash(dev, skb);
2360 if (queue_index != old_index && sk) {
2361 struct dst_entry *dst =
2362 rcu_dereference_check(sk->sk_dst_cache, 1);
2364 if (dst && skb_dst(skb) == dst)
2365 sk_tx_queue_set(sk, queue_index);
2370 skb_set_queue_mapping(skb, queue_index);
2371 return netdev_get_tx_queue(dev, queue_index);
2374 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2375 struct net_device *dev,
2376 struct netdev_queue *txq)
2378 spinlock_t *root_lock = qdisc_lock(q);
2382 qdisc_skb_cb(skb)->pkt_len = skb->len;
2383 qdisc_calculate_pkt_len(skb, q);
2385 * Heuristic to force contended enqueues to serialize on a
2386 * separate lock before trying to get qdisc main lock.
2387 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2388 * and dequeue packets faster.
2390 contended = qdisc_is_running(q);
2391 if (unlikely(contended))
2392 spin_lock(&q->busylock);
2394 spin_lock(root_lock);
2395 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2398 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2399 qdisc_run_begin(q)) {
2401 * This is a work-conserving queue; there are no old skbs
2402 * waiting to be sent out; and the qdisc is not running -
2403 * xmit the skb directly.
2405 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2408 qdisc_bstats_update(q, skb);
2410 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2411 if (unlikely(contended)) {
2412 spin_unlock(&q->busylock);
2419 rc = NET_XMIT_SUCCESS;
2422 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2423 if (qdisc_run_begin(q)) {
2424 if (unlikely(contended)) {
2425 spin_unlock(&q->busylock);
2431 spin_unlock(root_lock);
2432 if (unlikely(contended))
2433 spin_unlock(&q->busylock);
2437 static DEFINE_PER_CPU(int, xmit_recursion);
2438 #define RECURSION_LIMIT 10
2441 * dev_queue_xmit - transmit a buffer
2442 * @skb: buffer to transmit
2444 * Queue a buffer for transmission to a network device. The caller must
2445 * have set the device and priority and built the buffer before calling
2446 * this function. The function can be called from an interrupt.
2448 * A negative errno code is returned on a failure. A success does not
2449 * guarantee the frame will be transmitted as it may be dropped due
2450 * to congestion or traffic shaping.
2452 * -----------------------------------------------------------------------------------
2453 * I notice this method can also return errors from the queue disciplines,
2454 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2457 * Regardless of the return value, the skb is consumed, so it is currently
2458 * difficult to retry a send to this method. (You can bump the ref count
2459 * before sending to hold a reference for retry if you are careful.)
2461 * When calling this method, interrupts MUST be enabled. This is because
2462 * the BH enable code must have IRQs enabled so that it will not deadlock.
2465 int dev_queue_xmit(struct sk_buff *skb)
2467 struct net_device *dev = skb->dev;
2468 struct netdev_queue *txq;
2472 /* Disable soft irqs for various locks below. Also
2473 * stops preemption for RCU.
2477 txq = dev_pick_tx(dev, skb);
2478 q = rcu_dereference_bh(txq->qdisc);
2480 #ifdef CONFIG_NET_CLS_ACT
2481 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2483 trace_net_dev_queue(skb);
2485 rc = __dev_xmit_skb(skb, q, dev, txq);
2489 /* The device has no queue. Common case for software devices:
2490 loopback, all the sorts of tunnels...
2492 Really, it is unlikely that netif_tx_lock protection is necessary
2493 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2495 However, it is possible, that they rely on protection
2498 Check this and shot the lock. It is not prone from deadlocks.
2499 Either shot noqueue qdisc, it is even simpler 8)
2501 if (dev->flags & IFF_UP) {
2502 int cpu = smp_processor_id(); /* ok because BHs are off */
2504 if (txq->xmit_lock_owner != cpu) {
2506 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2507 goto recursion_alert;
2509 HARD_TX_LOCK(dev, txq, cpu);
2511 if (!netif_tx_queue_stopped(txq)) {
2512 __this_cpu_inc(xmit_recursion);
2513 rc = dev_hard_start_xmit(skb, dev, txq);
2514 __this_cpu_dec(xmit_recursion);
2515 if (dev_xmit_complete(rc)) {
2516 HARD_TX_UNLOCK(dev, txq);
2520 HARD_TX_UNLOCK(dev, txq);
2521 if (net_ratelimit())
2522 printk(KERN_CRIT "Virtual device %s asks to "
2523 "queue packet!\n", dev->name);
2525 /* Recursion is detected! It is possible,
2529 if (net_ratelimit())
2530 printk(KERN_CRIT "Dead loop on virtual device "
2531 "%s, fix it urgently!\n", dev->name);
2536 rcu_read_unlock_bh();
2541 rcu_read_unlock_bh();
2544 EXPORT_SYMBOL(dev_queue_xmit);
2547 /*=======================================================================
2549 =======================================================================*/
2551 int netdev_max_backlog __read_mostly = 1000;
2552 int netdev_tstamp_prequeue __read_mostly = 1;
2553 int netdev_budget __read_mostly = 300;
2554 int weight_p __read_mostly = 64; /* old backlog weight */
2556 /* Called with irq disabled */
2557 static inline void ____napi_schedule(struct softnet_data *sd,
2558 struct napi_struct *napi)
2560 list_add_tail(&napi->poll_list, &sd->poll_list);
2561 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2565 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2566 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2567 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2568 * if hash is a canonical 4-tuple hash over transport ports.
2570 void __skb_get_rxhash(struct sk_buff *skb)
2572 int nhoff, hash = 0, poff;
2573 const struct ipv6hdr *ip6;
2574 const struct iphdr *ip;
2575 const struct vlan_hdr *vlan;
2584 nhoff = skb_network_offset(skb);
2585 proto = skb->protocol;
2589 case __constant_htons(ETH_P_IP):
2591 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2594 ip = (const struct iphdr *) (skb->data + nhoff);
2595 if (ip_is_fragment(ip))
2598 ip_proto = ip->protocol;
2599 addr1 = (__force u32) ip->saddr;
2600 addr2 = (__force u32) ip->daddr;
2601 nhoff += ip->ihl * 4;
2603 case __constant_htons(ETH_P_IPV6):
2605 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2608 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2609 ip_proto = ip6->nexthdr;
2610 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2611 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2614 case __constant_htons(ETH_P_8021Q):
2615 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2617 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2618 proto = vlan->h_vlan_encapsulated_proto;
2619 nhoff += sizeof(*vlan);
2621 case __constant_htons(ETH_P_PPP_SES):
2622 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2624 proto = *((__be16 *) (skb->data + nhoff +
2625 sizeof(struct pppoe_hdr)));
2626 nhoff += PPPOE_SES_HLEN;
2628 case __constant_htons(PPP_IP):
2630 case __constant_htons(PPP_IPV6):
2641 if (pskb_may_pull(skb, nhoff + 16)) {
2642 u8 *h = skb->data + nhoff;
2643 __be16 flags = *(__be16 *)h;
2646 * Only look inside GRE if version zero and no
2649 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2650 proto = *(__be16 *)(h + 2);
2652 if (flags & GRE_CSUM)
2654 if (flags & GRE_KEY)
2656 if (flags & GRE_SEQ)
2669 poff = proto_ports_offset(ip_proto);
2672 if (pskb_may_pull(skb, nhoff + 4)) {
2673 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2674 if (ports.v16[1] < ports.v16[0])
2675 swap(ports.v16[0], ports.v16[1]);
2680 /* get a consistent hash (same value on both flow directions) */
2684 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2691 EXPORT_SYMBOL(__skb_get_rxhash);
2695 /* One global table that all flow-based protocols share. */
2696 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2697 EXPORT_SYMBOL(rps_sock_flow_table);
2699 static struct rps_dev_flow *
2700 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2701 struct rps_dev_flow *rflow, u16 next_cpu)
2703 if (next_cpu != RPS_NO_CPU) {
2704 #ifdef CONFIG_RFS_ACCEL
2705 struct netdev_rx_queue *rxqueue;
2706 struct rps_dev_flow_table *flow_table;
2707 struct rps_dev_flow *old_rflow;
2712 /* Should we steer this flow to a different hardware queue? */
2713 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2714 !(dev->features & NETIF_F_NTUPLE))
2716 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2717 if (rxq_index == skb_get_rx_queue(skb))
2720 rxqueue = dev->_rx + rxq_index;
2721 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2724 flow_id = skb->rxhash & flow_table->mask;
2725 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2726 rxq_index, flow_id);
2730 rflow = &flow_table->flows[flow_id];
2732 if (old_rflow->filter == rflow->filter)
2733 old_rflow->filter = RPS_NO_FILTER;
2737 per_cpu(softnet_data, next_cpu).input_queue_head;
2740 rflow->cpu = next_cpu;
2745 * get_rps_cpu is called from netif_receive_skb and returns the target
2746 * CPU from the RPS map of the receiving queue for a given skb.
2747 * rcu_read_lock must be held on entry.
2749 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2750 struct rps_dev_flow **rflowp)
2752 struct netdev_rx_queue *rxqueue;
2753 struct rps_map *map;
2754 struct rps_dev_flow_table *flow_table;
2755 struct rps_sock_flow_table *sock_flow_table;
2759 if (skb_rx_queue_recorded(skb)) {
2760 u16 index = skb_get_rx_queue(skb);
2761 if (unlikely(index >= dev->real_num_rx_queues)) {
2762 WARN_ONCE(dev->real_num_rx_queues > 1,
2763 "%s received packet on queue %u, but number "
2764 "of RX queues is %u\n",
2765 dev->name, index, dev->real_num_rx_queues);
2768 rxqueue = dev->_rx + index;
2772 map = rcu_dereference(rxqueue->rps_map);
2774 if (map->len == 1 &&
2775 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2776 tcpu = map->cpus[0];
2777 if (cpu_online(tcpu))
2781 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2785 skb_reset_network_header(skb);
2786 if (!skb_get_rxhash(skb))
2789 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2790 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2791 if (flow_table && sock_flow_table) {
2793 struct rps_dev_flow *rflow;
2795 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2798 next_cpu = sock_flow_table->ents[skb->rxhash &
2799 sock_flow_table->mask];
2802 * If the desired CPU (where last recvmsg was done) is
2803 * different from current CPU (one in the rx-queue flow
2804 * table entry), switch if one of the following holds:
2805 * - Current CPU is unset (equal to RPS_NO_CPU).
2806 * - Current CPU is offline.
2807 * - The current CPU's queue tail has advanced beyond the
2808 * last packet that was enqueued using this table entry.
2809 * This guarantees that all previous packets for the flow
2810 * have been dequeued, thus preserving in order delivery.
2812 if (unlikely(tcpu != next_cpu) &&
2813 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2814 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2815 rflow->last_qtail)) >= 0))
2816 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2818 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2826 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2828 if (cpu_online(tcpu)) {
2838 #ifdef CONFIG_RFS_ACCEL
2841 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2842 * @dev: Device on which the filter was set
2843 * @rxq_index: RX queue index
2844 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2845 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2847 * Drivers that implement ndo_rx_flow_steer() should periodically call
2848 * this function for each installed filter and remove the filters for
2849 * which it returns %true.
2851 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2852 u32 flow_id, u16 filter_id)
2854 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2855 struct rps_dev_flow_table *flow_table;
2856 struct rps_dev_flow *rflow;
2861 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2862 if (flow_table && flow_id <= flow_table->mask) {
2863 rflow = &flow_table->flows[flow_id];
2864 cpu = ACCESS_ONCE(rflow->cpu);
2865 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2866 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2867 rflow->last_qtail) <
2868 (int)(10 * flow_table->mask)))
2874 EXPORT_SYMBOL(rps_may_expire_flow);
2876 #endif /* CONFIG_RFS_ACCEL */
2878 /* Called from hardirq (IPI) context */
2879 static void rps_trigger_softirq(void *data)
2881 struct softnet_data *sd = data;
2883 ____napi_schedule(sd, &sd->backlog);
2887 #endif /* CONFIG_RPS */
2890 * Check if this softnet_data structure is another cpu one
2891 * If yes, queue it to our IPI list and return 1
2894 static int rps_ipi_queued(struct softnet_data *sd)
2897 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2900 sd->rps_ipi_next = mysd->rps_ipi_list;
2901 mysd->rps_ipi_list = sd;
2903 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2906 #endif /* CONFIG_RPS */
2911 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2912 * queue (may be a remote CPU queue).
2914 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2915 unsigned int *qtail)
2917 struct softnet_data *sd;
2918 unsigned long flags;
2920 sd = &per_cpu(softnet_data, cpu);
2922 local_irq_save(flags);
2925 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2926 if (skb_queue_len(&sd->input_pkt_queue)) {
2928 __skb_queue_tail(&sd->input_pkt_queue, skb);
2929 input_queue_tail_incr_save(sd, qtail);
2931 local_irq_restore(flags);
2932 return NET_RX_SUCCESS;
2935 /* Schedule NAPI for backlog device
2936 * We can use non atomic operation since we own the queue lock
2938 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2939 if (!rps_ipi_queued(sd))
2940 ____napi_schedule(sd, &sd->backlog);
2948 local_irq_restore(flags);
2950 atomic_long_inc(&skb->dev->rx_dropped);
2956 * netif_rx - post buffer to the network code
2957 * @skb: buffer to post
2959 * This function receives a packet from a device driver and queues it for
2960 * the upper (protocol) levels to process. It always succeeds. The buffer
2961 * may be dropped during processing for congestion control or by the
2965 * NET_RX_SUCCESS (no congestion)
2966 * NET_RX_DROP (packet was dropped)
2970 int netif_rx(struct sk_buff *skb)
2974 /* if netpoll wants it, pretend we never saw it */
2975 if (netpoll_rx(skb))
2978 if (netdev_tstamp_prequeue)
2979 net_timestamp_check(skb);
2981 trace_netif_rx(skb);
2984 struct rps_dev_flow voidflow, *rflow = &voidflow;
2990 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2992 cpu = smp_processor_id();
2994 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3002 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3008 EXPORT_SYMBOL(netif_rx);
3010 int netif_rx_ni(struct sk_buff *skb)
3015 err = netif_rx(skb);
3016 if (local_softirq_pending())
3022 EXPORT_SYMBOL(netif_rx_ni);
3024 static void net_tx_action(struct softirq_action *h)
3026 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3028 if (sd->completion_queue) {
3029 struct sk_buff *clist;
3031 local_irq_disable();
3032 clist = sd->completion_queue;
3033 sd->completion_queue = NULL;
3037 struct sk_buff *skb = clist;
3038 clist = clist->next;
3040 WARN_ON(atomic_read(&skb->users));
3041 trace_kfree_skb(skb, net_tx_action);
3046 if (sd->output_queue) {
3049 local_irq_disable();
3050 head = sd->output_queue;
3051 sd->output_queue = NULL;
3052 sd->output_queue_tailp = &sd->output_queue;
3056 struct Qdisc *q = head;
3057 spinlock_t *root_lock;
3059 head = head->next_sched;
3061 root_lock = qdisc_lock(q);
3062 if (spin_trylock(root_lock)) {
3063 smp_mb__before_clear_bit();
3064 clear_bit(__QDISC_STATE_SCHED,
3067 spin_unlock(root_lock);
3069 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3071 __netif_reschedule(q);
3073 smp_mb__before_clear_bit();
3074 clear_bit(__QDISC_STATE_SCHED,
3082 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3083 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3084 /* This hook is defined here for ATM LANE */
3085 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3086 unsigned char *addr) __read_mostly;
3087 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3090 #ifdef CONFIG_NET_CLS_ACT
3091 /* TODO: Maybe we should just force sch_ingress to be compiled in
3092 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3093 * a compare and 2 stores extra right now if we dont have it on
3094 * but have CONFIG_NET_CLS_ACT
3095 * NOTE: This doesn't stop any functionality; if you dont have
3096 * the ingress scheduler, you just can't add policies on ingress.
3099 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3101 struct net_device *dev = skb->dev;
3102 u32 ttl = G_TC_RTTL(skb->tc_verd);
3103 int result = TC_ACT_OK;
3106 if (unlikely(MAX_RED_LOOP < ttl++)) {
3107 if (net_ratelimit())
3108 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3109 skb->skb_iif, dev->ifindex);
3113 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3114 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3117 if (q != &noop_qdisc) {
3118 spin_lock(qdisc_lock(q));
3119 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3120 result = qdisc_enqueue_root(skb, q);
3121 spin_unlock(qdisc_lock(q));
3127 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3128 struct packet_type **pt_prev,
3129 int *ret, struct net_device *orig_dev)
3131 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3133 if (!rxq || rxq->qdisc == &noop_qdisc)
3137 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3141 switch (ing_filter(skb, rxq)) {
3155 * netdev_rx_handler_register - register receive handler
3156 * @dev: device to register a handler for
3157 * @rx_handler: receive handler to register
3158 * @rx_handler_data: data pointer that is used by rx handler
3160 * Register a receive hander for a device. This handler will then be
3161 * called from __netif_receive_skb. A negative errno code is returned
3164 * The caller must hold the rtnl_mutex.
3166 * For a general description of rx_handler, see enum rx_handler_result.
3168 int netdev_rx_handler_register(struct net_device *dev,
3169 rx_handler_func_t *rx_handler,
3170 void *rx_handler_data)
3174 if (dev->rx_handler)
3177 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3178 rcu_assign_pointer(dev->rx_handler, rx_handler);
3182 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3185 * netdev_rx_handler_unregister - unregister receive handler
3186 * @dev: device to unregister a handler from
3188 * Unregister a receive hander from a device.
3190 * The caller must hold the rtnl_mutex.
3192 void netdev_rx_handler_unregister(struct net_device *dev)
3196 RCU_INIT_POINTER(dev->rx_handler, NULL);
3197 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3199 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3201 static int __netif_receive_skb(struct sk_buff *skb)
3203 struct packet_type *ptype, *pt_prev;
3204 rx_handler_func_t *rx_handler;
3205 struct net_device *orig_dev;
3206 struct net_device *null_or_dev;
3207 bool deliver_exact = false;
3208 int ret = NET_RX_DROP;
3211 if (!netdev_tstamp_prequeue)
3212 net_timestamp_check(skb);
3214 trace_netif_receive_skb(skb);
3216 /* if we've gotten here through NAPI, check netpoll */
3217 if (netpoll_receive_skb(skb))
3221 skb->skb_iif = skb->dev->ifindex;
3222 orig_dev = skb->dev;
3224 skb_reset_network_header(skb);
3225 skb_reset_transport_header(skb);
3226 skb_reset_mac_len(skb);
3234 __this_cpu_inc(softnet_data.processed);
3236 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3237 skb = vlan_untag(skb);
3242 #ifdef CONFIG_NET_CLS_ACT
3243 if (skb->tc_verd & TC_NCLS) {
3244 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3249 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3250 if (!ptype->dev || ptype->dev == skb->dev) {
3252 ret = deliver_skb(skb, pt_prev, orig_dev);
3257 #ifdef CONFIG_NET_CLS_ACT
3258 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3264 rx_handler = rcu_dereference(skb->dev->rx_handler);
3265 if (vlan_tx_tag_present(skb)) {
3267 ret = deliver_skb(skb, pt_prev, orig_dev);
3270 if (vlan_do_receive(&skb, !rx_handler))
3272 else if (unlikely(!skb))
3278 ret = deliver_skb(skb, pt_prev, orig_dev);
3281 switch (rx_handler(&skb)) {
3282 case RX_HANDLER_CONSUMED:
3284 case RX_HANDLER_ANOTHER:
3286 case RX_HANDLER_EXACT:
3287 deliver_exact = true;
3288 case RX_HANDLER_PASS:
3295 /* deliver only exact match when indicated */
3296 null_or_dev = deliver_exact ? skb->dev : NULL;
3298 type = skb->protocol;
3299 list_for_each_entry_rcu(ptype,
3300 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3301 if (ptype->type == type &&
3302 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3303 ptype->dev == orig_dev)) {
3305 ret = deliver_skb(skb, pt_prev, orig_dev);
3311 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3313 atomic_long_inc(&skb->dev->rx_dropped);
3315 /* Jamal, now you will not able to escape explaining
3316 * me how you were going to use this. :-)
3327 * netif_receive_skb - process receive buffer from network
3328 * @skb: buffer to process
3330 * netif_receive_skb() is the main receive data processing function.
3331 * It always succeeds. The buffer may be dropped during processing
3332 * for congestion control or by the protocol layers.
3334 * This function may only be called from softirq context and interrupts
3335 * should be enabled.
3337 * Return values (usually ignored):
3338 * NET_RX_SUCCESS: no congestion
3339 * NET_RX_DROP: packet was dropped
3341 int netif_receive_skb(struct sk_buff *skb)
3343 if (netdev_tstamp_prequeue)
3344 net_timestamp_check(skb);
3346 if (skb_defer_rx_timestamp(skb))
3347 return NET_RX_SUCCESS;
3351 struct rps_dev_flow voidflow, *rflow = &voidflow;
3356 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3359 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3363 ret = __netif_receive_skb(skb);
3369 return __netif_receive_skb(skb);
3372 EXPORT_SYMBOL(netif_receive_skb);
3374 /* Network device is going away, flush any packets still pending
3375 * Called with irqs disabled.
3377 static void flush_backlog(void *arg)
3379 struct net_device *dev = arg;
3380 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3381 struct sk_buff *skb, *tmp;
3384 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3385 if (skb->dev == dev) {
3386 __skb_unlink(skb, &sd->input_pkt_queue);
3388 input_queue_head_incr(sd);
3393 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3394 if (skb->dev == dev) {
3395 __skb_unlink(skb, &sd->process_queue);
3397 input_queue_head_incr(sd);
3402 static int napi_gro_complete(struct sk_buff *skb)
3404 struct packet_type *ptype;
3405 __be16 type = skb->protocol;
3406 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3409 if (NAPI_GRO_CB(skb)->count == 1) {
3410 skb_shinfo(skb)->gso_size = 0;
3415 list_for_each_entry_rcu(ptype, head, list) {
3416 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3419 err = ptype->gro_complete(skb);
3425 WARN_ON(&ptype->list == head);
3427 return NET_RX_SUCCESS;
3431 return netif_receive_skb(skb);
3434 inline void napi_gro_flush(struct napi_struct *napi)
3436 struct sk_buff *skb, *next;
3438 for (skb = napi->gro_list; skb; skb = next) {
3441 napi_gro_complete(skb);
3444 napi->gro_count = 0;
3445 napi->gro_list = NULL;
3447 EXPORT_SYMBOL(napi_gro_flush);
3449 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3451 struct sk_buff **pp = NULL;
3452 struct packet_type *ptype;
3453 __be16 type = skb->protocol;
3454 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3457 enum gro_result ret;
3459 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3462 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3466 list_for_each_entry_rcu(ptype, head, list) {
3467 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3470 skb_set_network_header(skb, skb_gro_offset(skb));
3471 mac_len = skb->network_header - skb->mac_header;
3472 skb->mac_len = mac_len;
3473 NAPI_GRO_CB(skb)->same_flow = 0;
3474 NAPI_GRO_CB(skb)->flush = 0;
3475 NAPI_GRO_CB(skb)->free = 0;
3477 pp = ptype->gro_receive(&napi->gro_list, skb);
3482 if (&ptype->list == head)
3485 same_flow = NAPI_GRO_CB(skb)->same_flow;
3486 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3489 struct sk_buff *nskb = *pp;
3493 napi_gro_complete(nskb);
3500 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3504 NAPI_GRO_CB(skb)->count = 1;
3505 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3506 skb->next = napi->gro_list;
3507 napi->gro_list = skb;
3511 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3512 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3514 BUG_ON(skb->end - skb->tail < grow);
3516 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3519 skb->data_len -= grow;
3521 skb_shinfo(skb)->frags[0].page_offset += grow;
3522 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3524 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3525 skb_frag_unref(skb, 0);
3526 memmove(skb_shinfo(skb)->frags,
3527 skb_shinfo(skb)->frags + 1,
3528 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3539 EXPORT_SYMBOL(dev_gro_receive);
3541 static inline gro_result_t
3542 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3545 unsigned int maclen = skb->dev->hard_header_len;
3547 for (p = napi->gro_list; p; p = p->next) {
3548 unsigned long diffs;
3550 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3551 diffs |= p->vlan_tci ^ skb->vlan_tci;
3552 if (maclen == ETH_HLEN)
3553 diffs |= compare_ether_header(skb_mac_header(p),
3554 skb_gro_mac_header(skb));
3556 diffs = memcmp(skb_mac_header(p),
3557 skb_gro_mac_header(skb),
3559 NAPI_GRO_CB(p)->same_flow = !diffs;
3560 NAPI_GRO_CB(p)->flush = 0;
3563 return dev_gro_receive(napi, skb);
3566 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3570 if (netif_receive_skb(skb))
3575 case GRO_MERGED_FREE:
3586 EXPORT_SYMBOL(napi_skb_finish);
3588 void skb_gro_reset_offset(struct sk_buff *skb)
3590 NAPI_GRO_CB(skb)->data_offset = 0;
3591 NAPI_GRO_CB(skb)->frag0 = NULL;
3592 NAPI_GRO_CB(skb)->frag0_len = 0;
3594 if (skb->mac_header == skb->tail &&
3595 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3596 NAPI_GRO_CB(skb)->frag0 =
3597 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3598 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3601 EXPORT_SYMBOL(skb_gro_reset_offset);
3603 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3605 skb_gro_reset_offset(skb);
3607 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3609 EXPORT_SYMBOL(napi_gro_receive);
3611 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3613 __skb_pull(skb, skb_headlen(skb));
3614 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3615 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3617 skb->dev = napi->dev;
3623 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3625 struct sk_buff *skb = napi->skb;
3628 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3634 EXPORT_SYMBOL(napi_get_frags);
3636 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3642 skb->protocol = eth_type_trans(skb, skb->dev);
3644 if (ret == GRO_HELD)
3645 skb_gro_pull(skb, -ETH_HLEN);
3646 else if (netif_receive_skb(skb))
3651 case GRO_MERGED_FREE:
3652 napi_reuse_skb(napi, skb);
3661 EXPORT_SYMBOL(napi_frags_finish);
3663 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3665 struct sk_buff *skb = napi->skb;
3672 skb_reset_mac_header(skb);
3673 skb_gro_reset_offset(skb);
3675 off = skb_gro_offset(skb);
3676 hlen = off + sizeof(*eth);
3677 eth = skb_gro_header_fast(skb, off);
3678 if (skb_gro_header_hard(skb, hlen)) {
3679 eth = skb_gro_header_slow(skb, hlen, off);
3680 if (unlikely(!eth)) {
3681 napi_reuse_skb(napi, skb);
3687 skb_gro_pull(skb, sizeof(*eth));
3690 * This works because the only protocols we care about don't require
3691 * special handling. We'll fix it up properly at the end.
3693 skb->protocol = eth->h_proto;
3698 EXPORT_SYMBOL(napi_frags_skb);
3700 gro_result_t napi_gro_frags(struct napi_struct *napi)
3702 struct sk_buff *skb = napi_frags_skb(napi);
3707 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3709 EXPORT_SYMBOL(napi_gro_frags);
3712 * net_rps_action sends any pending IPI's for rps.
3713 * Note: called with local irq disabled, but exits with local irq enabled.
3715 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3718 struct softnet_data *remsd = sd->rps_ipi_list;
3721 sd->rps_ipi_list = NULL;
3725 /* Send pending IPI's to kick RPS processing on remote cpus. */
3727 struct softnet_data *next = remsd->rps_ipi_next;
3729 if (cpu_online(remsd->cpu))
3730 __smp_call_function_single(remsd->cpu,
3739 static int process_backlog(struct napi_struct *napi, int quota)
3742 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3745 /* Check if we have pending ipi, its better to send them now,
3746 * not waiting net_rx_action() end.
3748 if (sd->rps_ipi_list) {
3749 local_irq_disable();
3750 net_rps_action_and_irq_enable(sd);
3753 napi->weight = weight_p;
3754 local_irq_disable();
3755 while (work < quota) {
3756 struct sk_buff *skb;
3759 while ((skb = __skb_dequeue(&sd->process_queue))) {
3761 __netif_receive_skb(skb);
3762 local_irq_disable();
3763 input_queue_head_incr(sd);
3764 if (++work >= quota) {
3771 qlen = skb_queue_len(&sd->input_pkt_queue);
3773 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3774 &sd->process_queue);
3776 if (qlen < quota - work) {
3778 * Inline a custom version of __napi_complete().
3779 * only current cpu owns and manipulates this napi,
3780 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3781 * we can use a plain write instead of clear_bit(),
3782 * and we dont need an smp_mb() memory barrier.
3784 list_del(&napi->poll_list);
3787 quota = work + qlen;
3797 * __napi_schedule - schedule for receive
3798 * @n: entry to schedule
3800 * The entry's receive function will be scheduled to run
3802 void __napi_schedule(struct napi_struct *n)
3804 unsigned long flags;
3806 local_irq_save(flags);
3807 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3808 local_irq_restore(flags);
3810 EXPORT_SYMBOL(__napi_schedule);
3812 void __napi_complete(struct napi_struct *n)
3814 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3815 BUG_ON(n->gro_list);
3817 list_del(&n->poll_list);
3818 smp_mb__before_clear_bit();
3819 clear_bit(NAPI_STATE_SCHED, &n->state);
3821 EXPORT_SYMBOL(__napi_complete);
3823 void napi_complete(struct napi_struct *n)
3825 unsigned long flags;
3828 * don't let napi dequeue from the cpu poll list
3829 * just in case its running on a different cpu
3831 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3835 local_irq_save(flags);
3837 local_irq_restore(flags);
3839 EXPORT_SYMBOL(napi_complete);
3841 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3842 int (*poll)(struct napi_struct *, int), int weight)
3844 INIT_LIST_HEAD(&napi->poll_list);
3845 napi->gro_count = 0;
3846 napi->gro_list = NULL;
3849 napi->weight = weight;
3850 list_add(&napi->dev_list, &dev->napi_list);
3852 #ifdef CONFIG_NETPOLL
3853 spin_lock_init(&napi->poll_lock);
3854 napi->poll_owner = -1;
3856 set_bit(NAPI_STATE_SCHED, &napi->state);
3858 EXPORT_SYMBOL(netif_napi_add);
3860 void netif_napi_del(struct napi_struct *napi)
3862 struct sk_buff *skb, *next;
3864 list_del_init(&napi->dev_list);
3865 napi_free_frags(napi);
3867 for (skb = napi->gro_list; skb; skb = next) {
3873 napi->gro_list = NULL;
3874 napi->gro_count = 0;
3876 EXPORT_SYMBOL(netif_napi_del);
3878 static void net_rx_action(struct softirq_action *h)
3880 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3881 unsigned long time_limit = jiffies + 2;
3882 int budget = netdev_budget;
3885 local_irq_disable();
3887 while (!list_empty(&sd->poll_list)) {
3888 struct napi_struct *n;
3891 /* If softirq window is exhuasted then punt.
3892 * Allow this to run for 2 jiffies since which will allow
3893 * an average latency of 1.5/HZ.
3895 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3900 /* Even though interrupts have been re-enabled, this
3901 * access is safe because interrupts can only add new
3902 * entries to the tail of this list, and only ->poll()
3903 * calls can remove this head entry from the list.
3905 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3907 have = netpoll_poll_lock(n);
3911 /* This NAPI_STATE_SCHED test is for avoiding a race
3912 * with netpoll's poll_napi(). Only the entity which
3913 * obtains the lock and sees NAPI_STATE_SCHED set will
3914 * actually make the ->poll() call. Therefore we avoid
3915 * accidentally calling ->poll() when NAPI is not scheduled.
3918 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3919 work = n->poll(n, weight);
3923 WARN_ON_ONCE(work > weight);
3927 local_irq_disable();
3929 /* Drivers must not modify the NAPI state if they
3930 * consume the entire weight. In such cases this code
3931 * still "owns" the NAPI instance and therefore can
3932 * move the instance around on the list at-will.
3934 if (unlikely(work == weight)) {
3935 if (unlikely(napi_disable_pending(n))) {
3938 local_irq_disable();
3940 list_move_tail(&n->poll_list, &sd->poll_list);
3943 netpoll_poll_unlock(have);
3946 net_rps_action_and_irq_enable(sd);
3948 #ifdef CONFIG_NET_DMA
3950 * There may not be any more sk_buffs coming right now, so push
3951 * any pending DMA copies to hardware
3953 dma_issue_pending_all();
3960 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3964 static gifconf_func_t *gifconf_list[NPROTO];
3967 * register_gifconf - register a SIOCGIF handler
3968 * @family: Address family
3969 * @gifconf: Function handler
3971 * Register protocol dependent address dumping routines. The handler
3972 * that is passed must not be freed or reused until it has been replaced
3973 * by another handler.
3975 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3977 if (family >= NPROTO)
3979 gifconf_list[family] = gifconf;
3982 EXPORT_SYMBOL(register_gifconf);
3986 * Map an interface index to its name (SIOCGIFNAME)
3990 * We need this ioctl for efficient implementation of the
3991 * if_indextoname() function required by the IPv6 API. Without
3992 * it, we would have to search all the interfaces to find a
3996 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3998 struct net_device *dev;
4002 * Fetch the caller's info block.
4005 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4009 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4015 strcpy(ifr.ifr_name, dev->name);
4018 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4024 * Perform a SIOCGIFCONF call. This structure will change
4025 * size eventually, and there is nothing I can do about it.
4026 * Thus we will need a 'compatibility mode'.
4029 static int dev_ifconf(struct net *net, char __user *arg)
4032 struct net_device *dev;
4039 * Fetch the caller's info block.
4042 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4049 * Loop over the interfaces, and write an info block for each.
4053 for_each_netdev(net, dev) {
4054 for (i = 0; i < NPROTO; i++) {
4055 if (gifconf_list[i]) {
4058 done = gifconf_list[i](dev, NULL, 0);
4060 done = gifconf_list[i](dev, pos + total,
4070 * All done. Write the updated control block back to the caller.
4072 ifc.ifc_len = total;
4075 * Both BSD and Solaris return 0 here, so we do too.
4077 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4080 #ifdef CONFIG_PROC_FS
4082 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4084 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4085 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4086 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4088 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4090 struct net *net = seq_file_net(seq);
4091 struct net_device *dev;
4092 struct hlist_node *p;
4093 struct hlist_head *h;
4094 unsigned int count = 0, offset = get_offset(*pos);
4096 h = &net->dev_name_head[get_bucket(*pos)];
4097 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4098 if (++count == offset)
4105 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4107 struct net_device *dev;
4108 unsigned int bucket;
4111 dev = dev_from_same_bucket(seq, pos);
4115 bucket = get_bucket(*pos) + 1;
4116 *pos = set_bucket_offset(bucket, 1);
4117 } while (bucket < NETDEV_HASHENTRIES);
4123 * This is invoked by the /proc filesystem handler to display a device
4126 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4131 return SEQ_START_TOKEN;
4133 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4136 return dev_from_bucket(seq, pos);
4139 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4142 return dev_from_bucket(seq, pos);
4145 void dev_seq_stop(struct seq_file *seq, void *v)
4151 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4153 struct rtnl_link_stats64 temp;
4154 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4156 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4157 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4158 dev->name, stats->rx_bytes, stats->rx_packets,
4160 stats->rx_dropped + stats->rx_missed_errors,
4161 stats->rx_fifo_errors,
4162 stats->rx_length_errors + stats->rx_over_errors +
4163 stats->rx_crc_errors + stats->rx_frame_errors,
4164 stats->rx_compressed, stats->multicast,
4165 stats->tx_bytes, stats->tx_packets,
4166 stats->tx_errors, stats->tx_dropped,
4167 stats->tx_fifo_errors, stats->collisions,
4168 stats->tx_carrier_errors +
4169 stats->tx_aborted_errors +
4170 stats->tx_window_errors +
4171 stats->tx_heartbeat_errors,
4172 stats->tx_compressed);
4176 * Called from the PROCfs module. This now uses the new arbitrary sized
4177 * /proc/net interface to create /proc/net/dev
4179 static int dev_seq_show(struct seq_file *seq, void *v)
4181 if (v == SEQ_START_TOKEN)
4182 seq_puts(seq, "Inter-| Receive "
4184 " face |bytes packets errs drop fifo frame "
4185 "compressed multicast|bytes packets errs "
4186 "drop fifo colls carrier compressed\n");
4188 dev_seq_printf_stats(seq, v);
4192 static struct softnet_data *softnet_get_online(loff_t *pos)
4194 struct softnet_data *sd = NULL;
4196 while (*pos < nr_cpu_ids)
4197 if (cpu_online(*pos)) {
4198 sd = &per_cpu(softnet_data, *pos);
4205 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4207 return softnet_get_online(pos);
4210 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4213 return softnet_get_online(pos);
4216 static void softnet_seq_stop(struct seq_file *seq, void *v)
4220 static int softnet_seq_show(struct seq_file *seq, void *v)
4222 struct softnet_data *sd = v;
4224 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4225 sd->processed, sd->dropped, sd->time_squeeze, 0,
4226 0, 0, 0, 0, /* was fastroute */
4227 sd->cpu_collision, sd->received_rps);
4231 static const struct seq_operations dev_seq_ops = {
4232 .start = dev_seq_start,
4233 .next = dev_seq_next,
4234 .stop = dev_seq_stop,
4235 .show = dev_seq_show,
4238 static int dev_seq_open(struct inode *inode, struct file *file)
4240 return seq_open_net(inode, file, &dev_seq_ops,
4241 sizeof(struct seq_net_private));
4244 static const struct file_operations dev_seq_fops = {
4245 .owner = THIS_MODULE,
4246 .open = dev_seq_open,
4248 .llseek = seq_lseek,
4249 .release = seq_release_net,
4252 static const struct seq_operations softnet_seq_ops = {
4253 .start = softnet_seq_start,
4254 .next = softnet_seq_next,
4255 .stop = softnet_seq_stop,
4256 .show = softnet_seq_show,
4259 static int softnet_seq_open(struct inode *inode, struct file *file)
4261 return seq_open(file, &softnet_seq_ops);
4264 static const struct file_operations softnet_seq_fops = {
4265 .owner = THIS_MODULE,
4266 .open = softnet_seq_open,
4268 .llseek = seq_lseek,
4269 .release = seq_release,
4272 static void *ptype_get_idx(loff_t pos)
4274 struct packet_type *pt = NULL;
4278 list_for_each_entry_rcu(pt, &ptype_all, list) {
4284 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4285 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4294 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4298 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4301 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4303 struct packet_type *pt;
4304 struct list_head *nxt;
4308 if (v == SEQ_START_TOKEN)
4309 return ptype_get_idx(0);
4312 nxt = pt->list.next;
4313 if (pt->type == htons(ETH_P_ALL)) {
4314 if (nxt != &ptype_all)
4317 nxt = ptype_base[0].next;
4319 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4321 while (nxt == &ptype_base[hash]) {
4322 if (++hash >= PTYPE_HASH_SIZE)
4324 nxt = ptype_base[hash].next;
4327 return list_entry(nxt, struct packet_type, list);
4330 static void ptype_seq_stop(struct seq_file *seq, void *v)
4336 static int ptype_seq_show(struct seq_file *seq, void *v)
4338 struct packet_type *pt = v;
4340 if (v == SEQ_START_TOKEN)
4341 seq_puts(seq, "Type Device Function\n");
4342 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4343 if (pt->type == htons(ETH_P_ALL))
4344 seq_puts(seq, "ALL ");
4346 seq_printf(seq, "%04x", ntohs(pt->type));
4348 seq_printf(seq, " %-8s %pF\n",
4349 pt->dev ? pt->dev->name : "", pt->func);
4355 static const struct seq_operations ptype_seq_ops = {
4356 .start = ptype_seq_start,
4357 .next = ptype_seq_next,
4358 .stop = ptype_seq_stop,
4359 .show = ptype_seq_show,
4362 static int ptype_seq_open(struct inode *inode, struct file *file)
4364 return seq_open_net(inode, file, &ptype_seq_ops,
4365 sizeof(struct seq_net_private));
4368 static const struct file_operations ptype_seq_fops = {
4369 .owner = THIS_MODULE,
4370 .open = ptype_seq_open,
4372 .llseek = seq_lseek,
4373 .release = seq_release_net,
4377 static int __net_init dev_proc_net_init(struct net *net)
4381 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4383 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4385 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4388 if (wext_proc_init(net))
4394 proc_net_remove(net, "ptype");
4396 proc_net_remove(net, "softnet_stat");
4398 proc_net_remove(net, "dev");
4402 static void __net_exit dev_proc_net_exit(struct net *net)
4404 wext_proc_exit(net);
4406 proc_net_remove(net, "ptype");
4407 proc_net_remove(net, "softnet_stat");
4408 proc_net_remove(net, "dev");
4411 static struct pernet_operations __net_initdata dev_proc_ops = {
4412 .init = dev_proc_net_init,
4413 .exit = dev_proc_net_exit,
4416 static int __init dev_proc_init(void)
4418 return register_pernet_subsys(&dev_proc_ops);
4421 #define dev_proc_init() 0
4422 #endif /* CONFIG_PROC_FS */
4426 * netdev_set_master - set up master pointer
4427 * @slave: slave device
4428 * @master: new master device
4430 * Changes the master device of the slave. Pass %NULL to break the
4431 * bonding. The caller must hold the RTNL semaphore. On a failure
4432 * a negative errno code is returned. On success the reference counts
4433 * are adjusted and the function returns zero.
4435 int netdev_set_master(struct net_device *slave, struct net_device *master)
4437 struct net_device *old = slave->master;
4447 slave->master = master;
4453 EXPORT_SYMBOL(netdev_set_master);
4456 * netdev_set_bond_master - set up bonding master/slave pair
4457 * @slave: slave device
4458 * @master: new master device
4460 * Changes the master device of the slave. Pass %NULL to break the
4461 * bonding. The caller must hold the RTNL semaphore. On a failure
4462 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4463 * to the routing socket and the function returns zero.
4465 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4471 err = netdev_set_master(slave, master);
4475 slave->flags |= IFF_SLAVE;
4477 slave->flags &= ~IFF_SLAVE;
4479 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4482 EXPORT_SYMBOL(netdev_set_bond_master);
4484 static void dev_change_rx_flags(struct net_device *dev, int flags)
4486 const struct net_device_ops *ops = dev->netdev_ops;
4488 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4489 ops->ndo_change_rx_flags(dev, flags);
4492 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4494 unsigned short old_flags = dev->flags;
4500 dev->flags |= IFF_PROMISC;
4501 dev->promiscuity += inc;
4502 if (dev->promiscuity == 0) {
4505 * If inc causes overflow, untouch promisc and return error.
4508 dev->flags &= ~IFF_PROMISC;
4510 dev->promiscuity -= inc;
4511 printk(KERN_WARNING "%s: promiscuity touches roof, "
4512 "set promiscuity failed, promiscuity feature "
4513 "of device might be broken.\n", dev->name);
4517 if (dev->flags != old_flags) {
4518 printk(KERN_INFO "device %s %s promiscuous mode\n",
4519 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4521 if (audit_enabled) {
4522 current_uid_gid(&uid, &gid);
4523 audit_log(current->audit_context, GFP_ATOMIC,
4524 AUDIT_ANOM_PROMISCUOUS,
4525 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4526 dev->name, (dev->flags & IFF_PROMISC),
4527 (old_flags & IFF_PROMISC),
4528 audit_get_loginuid(current),
4530 audit_get_sessionid(current));
4533 dev_change_rx_flags(dev, IFF_PROMISC);
4539 * dev_set_promiscuity - update promiscuity count on a device
4543 * Add or remove promiscuity from a device. While the count in the device
4544 * remains above zero the interface remains promiscuous. Once it hits zero
4545 * the device reverts back to normal filtering operation. A negative inc
4546 * value is used to drop promiscuity on the device.
4547 * Return 0 if successful or a negative errno code on error.
4549 int dev_set_promiscuity(struct net_device *dev, int inc)
4551 unsigned short old_flags = dev->flags;
4554 err = __dev_set_promiscuity(dev, inc);
4557 if (dev->flags != old_flags)
4558 dev_set_rx_mode(dev);
4561 EXPORT_SYMBOL(dev_set_promiscuity);
4564 * dev_set_allmulti - update allmulti count on a device
4568 * Add or remove reception of all multicast frames to a device. While the
4569 * count in the device remains above zero the interface remains listening
4570 * to all interfaces. Once it hits zero the device reverts back to normal
4571 * filtering operation. A negative @inc value is used to drop the counter
4572 * when releasing a resource needing all multicasts.
4573 * Return 0 if successful or a negative errno code on error.
4576 int dev_set_allmulti(struct net_device *dev, int inc)
4578 unsigned short old_flags = dev->flags;
4582 dev->flags |= IFF_ALLMULTI;
4583 dev->allmulti += inc;
4584 if (dev->allmulti == 0) {
4587 * If inc causes overflow, untouch allmulti and return error.
4590 dev->flags &= ~IFF_ALLMULTI;
4592 dev->allmulti -= inc;
4593 printk(KERN_WARNING "%s: allmulti touches roof, "
4594 "set allmulti failed, allmulti feature of "
4595 "device might be broken.\n", dev->name);
4599 if (dev->flags ^ old_flags) {
4600 dev_change_rx_flags(dev, IFF_ALLMULTI);
4601 dev_set_rx_mode(dev);
4605 EXPORT_SYMBOL(dev_set_allmulti);
4608 * Upload unicast and multicast address lists to device and
4609 * configure RX filtering. When the device doesn't support unicast
4610 * filtering it is put in promiscuous mode while unicast addresses
4613 void __dev_set_rx_mode(struct net_device *dev)
4615 const struct net_device_ops *ops = dev->netdev_ops;
4617 /* dev_open will call this function so the list will stay sane. */
4618 if (!(dev->flags&IFF_UP))
4621 if (!netif_device_present(dev))
4624 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4625 /* Unicast addresses changes may only happen under the rtnl,
4626 * therefore calling __dev_set_promiscuity here is safe.
4628 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4629 __dev_set_promiscuity(dev, 1);
4630 dev->uc_promisc = true;
4631 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4632 __dev_set_promiscuity(dev, -1);
4633 dev->uc_promisc = false;
4637 if (ops->ndo_set_rx_mode)
4638 ops->ndo_set_rx_mode(dev);
4641 void dev_set_rx_mode(struct net_device *dev)
4643 netif_addr_lock_bh(dev);
4644 __dev_set_rx_mode(dev);
4645 netif_addr_unlock_bh(dev);
4649 * dev_get_flags - get flags reported to userspace
4652 * Get the combination of flag bits exported through APIs to userspace.
4654 unsigned dev_get_flags(const struct net_device *dev)
4658 flags = (dev->flags & ~(IFF_PROMISC |
4663 (dev->gflags & (IFF_PROMISC |
4666 if (netif_running(dev)) {
4667 if (netif_oper_up(dev))
4668 flags |= IFF_RUNNING;
4669 if (netif_carrier_ok(dev))
4670 flags |= IFF_LOWER_UP;
4671 if (netif_dormant(dev))
4672 flags |= IFF_DORMANT;
4677 EXPORT_SYMBOL(dev_get_flags);
4679 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4681 int old_flags = dev->flags;
4687 * Set the flags on our device.
4690 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4691 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4693 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4697 * Load in the correct multicast list now the flags have changed.
4700 if ((old_flags ^ flags) & IFF_MULTICAST)
4701 dev_change_rx_flags(dev, IFF_MULTICAST);
4703 dev_set_rx_mode(dev);
4706 * Have we downed the interface. We handle IFF_UP ourselves
4707 * according to user attempts to set it, rather than blindly
4712 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4713 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4716 dev_set_rx_mode(dev);
4719 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4720 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4722 dev->gflags ^= IFF_PROMISC;
4723 dev_set_promiscuity(dev, inc);
4726 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4727 is important. Some (broken) drivers set IFF_PROMISC, when
4728 IFF_ALLMULTI is requested not asking us and not reporting.
4730 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4731 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4733 dev->gflags ^= IFF_ALLMULTI;
4734 dev_set_allmulti(dev, inc);
4740 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4742 unsigned int changes = dev->flags ^ old_flags;
4744 if (changes & IFF_UP) {
4745 if (dev->flags & IFF_UP)
4746 call_netdevice_notifiers(NETDEV_UP, dev);
4748 call_netdevice_notifiers(NETDEV_DOWN, dev);
4751 if (dev->flags & IFF_UP &&
4752 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4753 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4757 * dev_change_flags - change device settings
4759 * @flags: device state flags
4761 * Change settings on device based state flags. The flags are
4762 * in the userspace exported format.
4764 int dev_change_flags(struct net_device *dev, unsigned flags)
4767 int old_flags = dev->flags;
4769 ret = __dev_change_flags(dev, flags);
4773 changes = old_flags ^ dev->flags;
4775 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4777 __dev_notify_flags(dev, old_flags);
4780 EXPORT_SYMBOL(dev_change_flags);
4783 * dev_set_mtu - Change maximum transfer unit
4785 * @new_mtu: new transfer unit
4787 * Change the maximum transfer size of the network device.
4789 int dev_set_mtu(struct net_device *dev, int new_mtu)
4791 const struct net_device_ops *ops = dev->netdev_ops;
4794 if (new_mtu == dev->mtu)
4797 /* MTU must be positive. */
4801 if (!netif_device_present(dev))
4805 if (ops->ndo_change_mtu)
4806 err = ops->ndo_change_mtu(dev, new_mtu);
4810 if (!err && dev->flags & IFF_UP)
4811 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4814 EXPORT_SYMBOL(dev_set_mtu);
4817 * dev_set_group - Change group this device belongs to
4819 * @new_group: group this device should belong to
4821 void dev_set_group(struct net_device *dev, int new_group)
4823 dev->group = new_group;
4825 EXPORT_SYMBOL(dev_set_group);
4828 * dev_set_mac_address - Change Media Access Control Address
4832 * Change the hardware (MAC) address of the device
4834 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4836 const struct net_device_ops *ops = dev->netdev_ops;
4839 if (!ops->ndo_set_mac_address)
4841 if (sa->sa_family != dev->type)
4843 if (!netif_device_present(dev))
4845 err = ops->ndo_set_mac_address(dev, sa);
4847 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4848 add_device_randomness(dev->dev_addr, dev->addr_len);
4851 EXPORT_SYMBOL(dev_set_mac_address);
4854 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4856 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4859 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4865 case SIOCGIFFLAGS: /* Get interface flags */
4866 ifr->ifr_flags = (short) dev_get_flags(dev);
4869 case SIOCGIFMETRIC: /* Get the metric on the interface
4870 (currently unused) */
4871 ifr->ifr_metric = 0;
4874 case SIOCGIFMTU: /* Get the MTU of a device */
4875 ifr->ifr_mtu = dev->mtu;
4880 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4882 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4883 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4884 ifr->ifr_hwaddr.sa_family = dev->type;
4892 ifr->ifr_map.mem_start = dev->mem_start;
4893 ifr->ifr_map.mem_end = dev->mem_end;
4894 ifr->ifr_map.base_addr = dev->base_addr;
4895 ifr->ifr_map.irq = dev->irq;
4896 ifr->ifr_map.dma = dev->dma;
4897 ifr->ifr_map.port = dev->if_port;
4901 ifr->ifr_ifindex = dev->ifindex;
4905 ifr->ifr_qlen = dev->tx_queue_len;
4909 /* dev_ioctl() should ensure this case
4921 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4923 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4926 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4927 const struct net_device_ops *ops;
4932 ops = dev->netdev_ops;
4935 case SIOCSIFFLAGS: /* Set interface flags */
4936 return dev_change_flags(dev, ifr->ifr_flags);
4938 case SIOCSIFMETRIC: /* Set the metric on the interface
4939 (currently unused) */
4942 case SIOCSIFMTU: /* Set the MTU of a device */
4943 return dev_set_mtu(dev, ifr->ifr_mtu);
4946 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4948 case SIOCSIFHWBROADCAST:
4949 if (ifr->ifr_hwaddr.sa_family != dev->type)
4951 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4952 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4953 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4957 if (ops->ndo_set_config) {
4958 if (!netif_device_present(dev))
4960 return ops->ndo_set_config(dev, &ifr->ifr_map);
4965 if (!ops->ndo_set_rx_mode ||
4966 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4968 if (!netif_device_present(dev))
4970 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4973 if (!ops->ndo_set_rx_mode ||
4974 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4976 if (!netif_device_present(dev))
4978 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4981 if (ifr->ifr_qlen < 0)
4983 dev->tx_queue_len = ifr->ifr_qlen;
4987 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4988 return dev_change_name(dev, ifr->ifr_newname);
4991 err = net_hwtstamp_validate(ifr);
4997 * Unknown or private ioctl
5000 if ((cmd >= SIOCDEVPRIVATE &&
5001 cmd <= SIOCDEVPRIVATE + 15) ||
5002 cmd == SIOCBONDENSLAVE ||
5003 cmd == SIOCBONDRELEASE ||
5004 cmd == SIOCBONDSETHWADDR ||
5005 cmd == SIOCBONDSLAVEINFOQUERY ||
5006 cmd == SIOCBONDINFOQUERY ||
5007 cmd == SIOCBONDCHANGEACTIVE ||
5008 cmd == SIOCGMIIPHY ||
5009 cmd == SIOCGMIIREG ||
5010 cmd == SIOCSMIIREG ||
5011 cmd == SIOCBRADDIF ||
5012 cmd == SIOCBRDELIF ||
5013 cmd == SIOCSHWTSTAMP ||
5014 cmd == SIOCWANDEV) {
5016 if (ops->ndo_do_ioctl) {
5017 if (netif_device_present(dev))
5018 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5030 * This function handles all "interface"-type I/O control requests. The actual
5031 * 'doing' part of this is dev_ifsioc above.
5035 * dev_ioctl - network device ioctl
5036 * @net: the applicable net namespace
5037 * @cmd: command to issue
5038 * @arg: pointer to a struct ifreq in user space
5040 * Issue ioctl functions to devices. This is normally called by the
5041 * user space syscall interfaces but can sometimes be useful for
5042 * other purposes. The return value is the return from the syscall if
5043 * positive or a negative errno code on error.
5046 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5052 /* One special case: SIOCGIFCONF takes ifconf argument
5053 and requires shared lock, because it sleeps writing
5057 if (cmd == SIOCGIFCONF) {
5059 ret = dev_ifconf(net, (char __user *) arg);
5063 if (cmd == SIOCGIFNAME)
5064 return dev_ifname(net, (struct ifreq __user *)arg);
5066 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5069 ifr.ifr_name[IFNAMSIZ-1] = 0;
5071 colon = strchr(ifr.ifr_name, ':');
5076 * See which interface the caller is talking about.
5081 * These ioctl calls:
5082 * - can be done by all.
5083 * - atomic and do not require locking.
5094 dev_load(net, ifr.ifr_name);
5096 ret = dev_ifsioc_locked(net, &ifr, cmd);
5101 if (copy_to_user(arg, &ifr,
5102 sizeof(struct ifreq)))
5108 dev_load(net, ifr.ifr_name);
5110 ret = dev_ethtool(net, &ifr);
5115 if (copy_to_user(arg, &ifr,
5116 sizeof(struct ifreq)))
5122 * These ioctl calls:
5123 * - require superuser power.
5124 * - require strict serialization.
5130 if (!capable(CAP_NET_ADMIN))
5132 dev_load(net, ifr.ifr_name);
5134 ret = dev_ifsioc(net, &ifr, cmd);
5139 if (copy_to_user(arg, &ifr,
5140 sizeof(struct ifreq)))
5146 * These ioctl calls:
5147 * - require superuser power.
5148 * - require strict serialization.
5149 * - do not return a value
5159 case SIOCSIFHWBROADCAST:
5162 case SIOCBONDENSLAVE:
5163 case SIOCBONDRELEASE:
5164 case SIOCBONDSETHWADDR:
5165 case SIOCBONDCHANGEACTIVE:
5169 if (!capable(CAP_NET_ADMIN))
5172 case SIOCBONDSLAVEINFOQUERY:
5173 case SIOCBONDINFOQUERY:
5174 dev_load(net, ifr.ifr_name);
5176 ret = dev_ifsioc(net, &ifr, cmd);
5181 /* Get the per device memory space. We can add this but
5182 * currently do not support it */
5184 /* Set the per device memory buffer space.
5185 * Not applicable in our case */
5190 * Unknown or private ioctl.
5193 if (cmd == SIOCWANDEV ||
5194 (cmd >= SIOCDEVPRIVATE &&
5195 cmd <= SIOCDEVPRIVATE + 15)) {
5196 dev_load(net, ifr.ifr_name);
5198 ret = dev_ifsioc(net, &ifr, cmd);
5200 if (!ret && copy_to_user(arg, &ifr,
5201 sizeof(struct ifreq)))
5205 /* Take care of Wireless Extensions */
5206 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5207 return wext_handle_ioctl(net, &ifr, cmd, arg);
5214 * dev_new_index - allocate an ifindex
5215 * @net: the applicable net namespace
5217 * Returns a suitable unique value for a new device interface
5218 * number. The caller must hold the rtnl semaphore or the
5219 * dev_base_lock to be sure it remains unique.
5221 static int dev_new_index(struct net *net)
5227 if (!__dev_get_by_index(net, ifindex))
5232 /* Delayed registration/unregisteration */
5233 static LIST_HEAD(net_todo_list);
5235 static void net_set_todo(struct net_device *dev)
5237 list_add_tail(&dev->todo_list, &net_todo_list);
5240 static void rollback_registered_many(struct list_head *head)
5242 struct net_device *dev, *tmp;
5244 BUG_ON(dev_boot_phase);
5247 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5248 /* Some devices call without registering
5249 * for initialization unwind. Remove those
5250 * devices and proceed with the remaining.
5252 if (dev->reg_state == NETREG_UNINITIALIZED) {
5253 pr_debug("unregister_netdevice: device %s/%p never "
5254 "was registered\n", dev->name, dev);
5257 list_del(&dev->unreg_list);
5260 dev->dismantle = true;
5261 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5264 /* If device is running, close it first. */
5265 dev_close_many(head);
5267 list_for_each_entry(dev, head, unreg_list) {
5268 /* And unlink it from device chain. */
5269 unlist_netdevice(dev);
5271 dev->reg_state = NETREG_UNREGISTERING;
5276 list_for_each_entry(dev, head, unreg_list) {
5277 /* Shutdown queueing discipline. */
5281 /* Notify protocols, that we are about to destroy
5282 this device. They should clean all the things.
5284 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5286 if (!dev->rtnl_link_ops ||
5287 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5288 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5291 * Flush the unicast and multicast chains
5296 if (dev->netdev_ops->ndo_uninit)
5297 dev->netdev_ops->ndo_uninit(dev);
5299 /* Notifier chain MUST detach us from master device. */
5300 WARN_ON(dev->master);
5302 /* Remove entries from kobject tree */
5303 netdev_unregister_kobject(dev);
5306 /* Process any work delayed until the end of the batch */
5307 dev = list_first_entry(head, struct net_device, unreg_list);
5308 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5312 list_for_each_entry(dev, head, unreg_list)
5316 static void rollback_registered(struct net_device *dev)
5320 list_add(&dev->unreg_list, &single);
5321 rollback_registered_many(&single);
5325 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5327 /* Fix illegal checksum combinations */
5328 if ((features & NETIF_F_HW_CSUM) &&
5329 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5330 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5331 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5334 if ((features & NETIF_F_NO_CSUM) &&
5335 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5336 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5337 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5340 /* Fix illegal SG+CSUM combinations. */
5341 if ((features & NETIF_F_SG) &&
5342 !(features & NETIF_F_ALL_CSUM)) {
5344 "Dropping NETIF_F_SG since no checksum feature.\n");
5345 features &= ~NETIF_F_SG;
5348 /* TSO requires that SG is present as well. */
5349 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5350 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5351 features &= ~NETIF_F_ALL_TSO;
5354 /* TSO ECN requires that TSO is present as well. */
5355 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5356 features &= ~NETIF_F_TSO_ECN;
5358 /* Software GSO depends on SG. */
5359 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5360 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5361 features &= ~NETIF_F_GSO;
5364 /* UFO needs SG and checksumming */
5365 if (features & NETIF_F_UFO) {
5366 /* maybe split UFO into V4 and V6? */
5367 if (!((features & NETIF_F_GEN_CSUM) ||
5368 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5369 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5371 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5372 features &= ~NETIF_F_UFO;
5375 if (!(features & NETIF_F_SG)) {
5377 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5378 features &= ~NETIF_F_UFO;
5385 int __netdev_update_features(struct net_device *dev)
5392 features = netdev_get_wanted_features(dev);
5394 if (dev->netdev_ops->ndo_fix_features)
5395 features = dev->netdev_ops->ndo_fix_features(dev, features);
5397 /* driver might be less strict about feature dependencies */
5398 features = netdev_fix_features(dev, features);
5400 if (dev->features == features)
5403 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5404 dev->features, features);
5406 if (dev->netdev_ops->ndo_set_features)
5407 err = dev->netdev_ops->ndo_set_features(dev, features);
5409 if (unlikely(err < 0)) {
5411 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5412 err, features, dev->features);
5417 dev->features = features;
5423 * netdev_update_features - recalculate device features
5424 * @dev: the device to check
5426 * Recalculate dev->features set and send notifications if it
5427 * has changed. Should be called after driver or hardware dependent
5428 * conditions might have changed that influence the features.
5430 void netdev_update_features(struct net_device *dev)
5432 if (__netdev_update_features(dev))
5433 netdev_features_change(dev);
5435 EXPORT_SYMBOL(netdev_update_features);
5438 * netdev_change_features - recalculate device features
5439 * @dev: the device to check
5441 * Recalculate dev->features set and send notifications even
5442 * if they have not changed. Should be called instead of
5443 * netdev_update_features() if also dev->vlan_features might
5444 * have changed to allow the changes to be propagated to stacked
5447 void netdev_change_features(struct net_device *dev)
5449 __netdev_update_features(dev);
5450 netdev_features_change(dev);
5452 EXPORT_SYMBOL(netdev_change_features);
5455 * netif_stacked_transfer_operstate - transfer operstate
5456 * @rootdev: the root or lower level device to transfer state from
5457 * @dev: the device to transfer operstate to
5459 * Transfer operational state from root to device. This is normally
5460 * called when a stacking relationship exists between the root
5461 * device and the device(a leaf device).
5463 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5464 struct net_device *dev)
5466 if (rootdev->operstate == IF_OPER_DORMANT)
5467 netif_dormant_on(dev);
5469 netif_dormant_off(dev);
5471 if (netif_carrier_ok(rootdev)) {
5472 if (!netif_carrier_ok(dev))
5473 netif_carrier_on(dev);
5475 if (netif_carrier_ok(dev))
5476 netif_carrier_off(dev);
5479 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5482 static int netif_alloc_rx_queues(struct net_device *dev)
5484 unsigned int i, count = dev->num_rx_queues;
5485 struct netdev_rx_queue *rx;
5489 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5491 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5496 for (i = 0; i < count; i++)
5502 static void netdev_init_one_queue(struct net_device *dev,
5503 struct netdev_queue *queue, void *_unused)
5505 /* Initialize queue lock */
5506 spin_lock_init(&queue->_xmit_lock);
5507 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5508 queue->xmit_lock_owner = -1;
5509 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5513 static int netif_alloc_netdev_queues(struct net_device *dev)
5515 unsigned int count = dev->num_tx_queues;
5516 struct netdev_queue *tx;
5520 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5522 pr_err("netdev: Unable to allocate %u tx queues.\n",
5528 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5529 spin_lock_init(&dev->tx_global_lock);
5535 * register_netdevice - register a network device
5536 * @dev: device to register
5538 * Take a completed network device structure and add it to the kernel
5539 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5540 * chain. 0 is returned on success. A negative errno code is returned
5541 * on a failure to set up the device, or if the name is a duplicate.
5543 * Callers must hold the rtnl semaphore. You may want
5544 * register_netdev() instead of this.
5547 * The locking appears insufficient to guarantee two parallel registers
5548 * will not get the same name.
5551 int register_netdevice(struct net_device *dev)
5554 struct net *net = dev_net(dev);
5556 BUG_ON(dev_boot_phase);
5561 /* When net_device's are persistent, this will be fatal. */
5562 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5565 spin_lock_init(&dev->addr_list_lock);
5566 netdev_set_addr_lockdep_class(dev);
5570 ret = dev_get_valid_name(dev, dev->name);
5574 /* Init, if this function is available */
5575 if (dev->netdev_ops->ndo_init) {
5576 ret = dev->netdev_ops->ndo_init(dev);
5584 dev->ifindex = dev_new_index(net);
5585 if (dev->iflink == -1)
5586 dev->iflink = dev->ifindex;
5588 /* Transfer changeable features to wanted_features and enable
5589 * software offloads (GSO and GRO).
5591 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5592 dev->features |= NETIF_F_SOFT_FEATURES;
5593 dev->wanted_features = dev->features & dev->hw_features;
5595 /* Turn on no cache copy if HW is doing checksum */
5596 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5597 if ((dev->features & NETIF_F_ALL_CSUM) &&
5598 !(dev->features & NETIF_F_NO_CSUM)) {
5599 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5600 dev->features |= NETIF_F_NOCACHE_COPY;
5603 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5605 dev->vlan_features |= NETIF_F_HIGHDMA;
5607 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5608 ret = notifier_to_errno(ret);
5612 ret = netdev_register_kobject(dev);
5615 dev->reg_state = NETREG_REGISTERED;
5617 __netdev_update_features(dev);
5620 * Default initial state at registry is that the
5621 * device is present.
5624 set_bit(__LINK_STATE_PRESENT, &dev->state);
5626 dev_init_scheduler(dev);
5628 list_netdevice(dev);
5629 add_device_randomness(dev->dev_addr, dev->addr_len);
5631 /* Notify protocols, that a new device appeared. */
5632 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5633 ret = notifier_to_errno(ret);
5635 rollback_registered(dev);
5636 dev->reg_state = NETREG_UNREGISTERED;
5639 * Prevent userspace races by waiting until the network
5640 * device is fully setup before sending notifications.
5642 if (!dev->rtnl_link_ops ||
5643 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5644 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5650 if (dev->netdev_ops->ndo_uninit)
5651 dev->netdev_ops->ndo_uninit(dev);
5654 EXPORT_SYMBOL(register_netdevice);
5657 * init_dummy_netdev - init a dummy network device for NAPI
5658 * @dev: device to init
5660 * This takes a network device structure and initialize the minimum
5661 * amount of fields so it can be used to schedule NAPI polls without
5662 * registering a full blown interface. This is to be used by drivers
5663 * that need to tie several hardware interfaces to a single NAPI
5664 * poll scheduler due to HW limitations.
5666 int init_dummy_netdev(struct net_device *dev)
5668 /* Clear everything. Note we don't initialize spinlocks
5669 * are they aren't supposed to be taken by any of the
5670 * NAPI code and this dummy netdev is supposed to be
5671 * only ever used for NAPI polls
5673 memset(dev, 0, sizeof(struct net_device));
5675 /* make sure we BUG if trying to hit standard
5676 * register/unregister code path
5678 dev->reg_state = NETREG_DUMMY;
5680 /* NAPI wants this */
5681 INIT_LIST_HEAD(&dev->napi_list);
5683 /* a dummy interface is started by default */
5684 set_bit(__LINK_STATE_PRESENT, &dev->state);
5685 set_bit(__LINK_STATE_START, &dev->state);
5687 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5688 * because users of this 'device' dont need to change
5694 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5698 * register_netdev - register a network device
5699 * @dev: device to register
5701 * Take a completed network device structure and add it to the kernel
5702 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5703 * chain. 0 is returned on success. A negative errno code is returned
5704 * on a failure to set up the device, or if the name is a duplicate.
5706 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5707 * and expands the device name if you passed a format string to
5710 int register_netdev(struct net_device *dev)
5715 err = register_netdevice(dev);
5719 EXPORT_SYMBOL(register_netdev);
5721 int netdev_refcnt_read(const struct net_device *dev)
5725 for_each_possible_cpu(i)
5726 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5729 EXPORT_SYMBOL(netdev_refcnt_read);
5732 * netdev_wait_allrefs - wait until all references are gone.
5734 * This is called when unregistering network devices.
5736 * Any protocol or device that holds a reference should register
5737 * for netdevice notification, and cleanup and put back the
5738 * reference if they receive an UNREGISTER event.
5739 * We can get stuck here if buggy protocols don't correctly
5742 static void netdev_wait_allrefs(struct net_device *dev)
5744 unsigned long rebroadcast_time, warning_time;
5747 linkwatch_forget_dev(dev);
5749 rebroadcast_time = warning_time = jiffies;
5750 refcnt = netdev_refcnt_read(dev);
5752 while (refcnt != 0) {
5753 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5756 /* Rebroadcast unregister notification */
5757 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5758 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5759 * should have already handle it the first time */
5761 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5763 /* We must not have linkwatch events
5764 * pending on unregister. If this
5765 * happens, we simply run the queue
5766 * unscheduled, resulting in a noop
5769 linkwatch_run_queue();
5774 rebroadcast_time = jiffies;
5779 refcnt = netdev_refcnt_read(dev);
5781 if (time_after(jiffies, warning_time + 10 * HZ)) {
5782 printk(KERN_EMERG "unregister_netdevice: "
5783 "waiting for %s to become free. Usage "
5786 warning_time = jiffies;
5795 * register_netdevice(x1);
5796 * register_netdevice(x2);
5798 * unregister_netdevice(y1);
5799 * unregister_netdevice(y2);
5805 * We are invoked by rtnl_unlock().
5806 * This allows us to deal with problems:
5807 * 1) We can delete sysfs objects which invoke hotplug
5808 * without deadlocking with linkwatch via keventd.
5809 * 2) Since we run with the RTNL semaphore not held, we can sleep
5810 * safely in order to wait for the netdev refcnt to drop to zero.
5812 * We must not return until all unregister events added during
5813 * the interval the lock was held have been completed.
5815 void netdev_run_todo(void)
5817 struct list_head list;
5819 /* Snapshot list, allow later requests */
5820 list_replace_init(&net_todo_list, &list);
5824 /* Wait for rcu callbacks to finish before attempting to drain
5825 * the device list. This usually avoids a 250ms wait.
5827 if (!list_empty(&list))
5830 while (!list_empty(&list)) {
5831 struct net_device *dev
5832 = list_first_entry(&list, struct net_device, todo_list);
5833 list_del(&dev->todo_list);
5835 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5836 printk(KERN_ERR "network todo '%s' but state %d\n",
5837 dev->name, dev->reg_state);
5842 dev->reg_state = NETREG_UNREGISTERED;
5844 on_each_cpu(flush_backlog, dev, 1);
5846 netdev_wait_allrefs(dev);
5849 BUG_ON(netdev_refcnt_read(dev));
5850 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5851 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5852 WARN_ON(dev->dn_ptr);
5854 if (dev->destructor)
5855 dev->destructor(dev);
5857 /* Free network device */
5858 kobject_put(&dev->dev.kobj);
5862 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5863 * fields in the same order, with only the type differing.
5865 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5866 const struct net_device_stats *netdev_stats)
5868 #if BITS_PER_LONG == 64
5869 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5870 memcpy(stats64, netdev_stats, sizeof(*stats64));
5872 size_t i, n = sizeof(*stats64) / sizeof(u64);
5873 const unsigned long *src = (const unsigned long *)netdev_stats;
5874 u64 *dst = (u64 *)stats64;
5876 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5877 sizeof(*stats64) / sizeof(u64));
5878 for (i = 0; i < n; i++)
5884 * dev_get_stats - get network device statistics
5885 * @dev: device to get statistics from
5886 * @storage: place to store stats
5888 * Get network statistics from device. Return @storage.
5889 * The device driver may provide its own method by setting
5890 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5891 * otherwise the internal statistics structure is used.
5893 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5894 struct rtnl_link_stats64 *storage)
5896 const struct net_device_ops *ops = dev->netdev_ops;
5898 if (ops->ndo_get_stats64) {
5899 memset(storage, 0, sizeof(*storage));
5900 ops->ndo_get_stats64(dev, storage);
5901 } else if (ops->ndo_get_stats) {
5902 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5904 netdev_stats_to_stats64(storage, &dev->stats);
5906 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5909 EXPORT_SYMBOL(dev_get_stats);
5911 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5913 struct netdev_queue *queue = dev_ingress_queue(dev);
5915 #ifdef CONFIG_NET_CLS_ACT
5918 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5921 netdev_init_one_queue(dev, queue, NULL);
5922 queue->qdisc = &noop_qdisc;
5923 queue->qdisc_sleeping = &noop_qdisc;
5924 rcu_assign_pointer(dev->ingress_queue, queue);
5930 * alloc_netdev_mqs - allocate network device
5931 * @sizeof_priv: size of private data to allocate space for
5932 * @name: device name format string
5933 * @setup: callback to initialize device
5934 * @txqs: the number of TX subqueues to allocate
5935 * @rxqs: the number of RX subqueues to allocate
5937 * Allocates a struct net_device with private data area for driver use
5938 * and performs basic initialization. Also allocates subquue structs
5939 * for each queue on the device.
5941 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5942 void (*setup)(struct net_device *),
5943 unsigned int txqs, unsigned int rxqs)
5945 struct net_device *dev;
5947 struct net_device *p;
5949 BUG_ON(strlen(name) >= sizeof(dev->name));
5952 pr_err("alloc_netdev: Unable to allocate device "
5953 "with zero queues.\n");
5959 pr_err("alloc_netdev: Unable to allocate device "
5960 "with zero RX queues.\n");
5965 alloc_size = sizeof(struct net_device);
5967 /* ensure 32-byte alignment of private area */
5968 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5969 alloc_size += sizeof_priv;
5971 /* ensure 32-byte alignment of whole construct */
5972 alloc_size += NETDEV_ALIGN - 1;
5974 p = kzalloc(alloc_size, GFP_KERNEL);
5976 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5980 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5981 dev->padded = (char *)dev - (char *)p;
5983 dev->pcpu_refcnt = alloc_percpu(int);
5984 if (!dev->pcpu_refcnt)
5987 if (dev_addr_init(dev))
5993 dev_net_set(dev, &init_net);
5995 dev->gso_max_size = GSO_MAX_SIZE;
5996 dev->gso_max_segs = GSO_MAX_SEGS;
5998 INIT_LIST_HEAD(&dev->napi_list);
5999 INIT_LIST_HEAD(&dev->unreg_list);
6000 INIT_LIST_HEAD(&dev->link_watch_list);
6001 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6004 dev->num_tx_queues = txqs;
6005 dev->real_num_tx_queues = txqs;
6006 if (netif_alloc_netdev_queues(dev))
6010 dev->num_rx_queues = rxqs;
6011 dev->real_num_rx_queues = rxqs;
6012 if (netif_alloc_rx_queues(dev))
6016 strcpy(dev->name, name);
6017 dev->group = INIT_NETDEV_GROUP;
6025 free_percpu(dev->pcpu_refcnt);
6035 EXPORT_SYMBOL(alloc_netdev_mqs);
6038 * free_netdev - free network device
6041 * This function does the last stage of destroying an allocated device
6042 * interface. The reference to the device object is released.
6043 * If this is the last reference then it will be freed.
6045 void free_netdev(struct net_device *dev)
6047 struct napi_struct *p, *n;
6049 release_net(dev_net(dev));
6056 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6058 /* Flush device addresses */
6059 dev_addr_flush(dev);
6061 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6064 free_percpu(dev->pcpu_refcnt);
6065 dev->pcpu_refcnt = NULL;
6067 /* Compatibility with error handling in drivers */
6068 if (dev->reg_state == NETREG_UNINITIALIZED) {
6069 kfree((char *)dev - dev->padded);
6073 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6074 dev->reg_state = NETREG_RELEASED;
6076 /* will free via device release */
6077 put_device(&dev->dev);
6079 EXPORT_SYMBOL(free_netdev);
6082 * synchronize_net - Synchronize with packet receive processing
6084 * Wait for packets currently being received to be done.
6085 * Does not block later packets from starting.
6087 void synchronize_net(void)
6090 if (rtnl_is_locked())
6091 synchronize_rcu_expedited();
6095 EXPORT_SYMBOL(synchronize_net);
6098 * unregister_netdevice_queue - remove device from the kernel
6102 * This function shuts down a device interface and removes it
6103 * from the kernel tables.
6104 * If head not NULL, device is queued to be unregistered later.
6106 * Callers must hold the rtnl semaphore. You may want
6107 * unregister_netdev() instead of this.
6110 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6115 list_move_tail(&dev->unreg_list, head);
6117 rollback_registered(dev);
6118 /* Finish processing unregister after unlock */
6122 EXPORT_SYMBOL(unregister_netdevice_queue);
6125 * unregister_netdevice_many - unregister many devices
6126 * @head: list of devices
6128 void unregister_netdevice_many(struct list_head *head)
6130 struct net_device *dev;
6132 if (!list_empty(head)) {
6133 rollback_registered_many(head);
6134 list_for_each_entry(dev, head, unreg_list)
6138 EXPORT_SYMBOL(unregister_netdevice_many);
6141 * unregister_netdev - remove device from the kernel
6144 * This function shuts down a device interface and removes it
6145 * from the kernel tables.
6147 * This is just a wrapper for unregister_netdevice that takes
6148 * the rtnl semaphore. In general you want to use this and not
6149 * unregister_netdevice.
6151 void unregister_netdev(struct net_device *dev)
6154 unregister_netdevice(dev);
6157 EXPORT_SYMBOL(unregister_netdev);
6160 * dev_change_net_namespace - move device to different nethost namespace
6162 * @net: network namespace
6163 * @pat: If not NULL name pattern to try if the current device name
6164 * is already taken in the destination network namespace.
6166 * This function shuts down a device interface and moves it
6167 * to a new network namespace. On success 0 is returned, on
6168 * a failure a netagive errno code is returned.
6170 * Callers must hold the rtnl semaphore.
6173 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6179 /* Don't allow namespace local devices to be moved. */
6181 if (dev->features & NETIF_F_NETNS_LOCAL)
6184 /* Ensure the device has been registrered */
6186 if (dev->reg_state != NETREG_REGISTERED)
6189 /* Get out if there is nothing todo */
6191 if (net_eq(dev_net(dev), net))
6194 /* Pick the destination device name, and ensure
6195 * we can use it in the destination network namespace.
6198 if (__dev_get_by_name(net, dev->name)) {
6199 /* We get here if we can't use the current device name */
6202 if (dev_get_valid_name(dev, pat) < 0)
6207 * And now a mini version of register_netdevice unregister_netdevice.
6210 /* If device is running close it first. */
6213 /* And unlink it from device chain */
6215 unlist_netdevice(dev);
6219 /* Shutdown queueing discipline. */
6222 /* Notify protocols, that we are about to destroy
6223 this device. They should clean all the things.
6225 Note that dev->reg_state stays at NETREG_REGISTERED.
6226 This is wanted because this way 8021q and macvlan know
6227 the device is just moving and can keep their slaves up.
6229 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6230 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6231 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6234 * Flush the unicast and multicast chains
6239 /* Actually switch the network namespace */
6240 dev_net_set(dev, net);
6242 /* If there is an ifindex conflict assign a new one */
6243 if (__dev_get_by_index(net, dev->ifindex)) {
6244 int iflink = (dev->iflink == dev->ifindex);
6245 dev->ifindex = dev_new_index(net);
6247 dev->iflink = dev->ifindex;
6250 /* Fixup kobjects */
6251 err = device_rename(&dev->dev, dev->name);
6254 /* Add the device back in the hashes */
6255 list_netdevice(dev);
6257 /* Notify protocols, that a new device appeared. */
6258 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6261 * Prevent userspace races by waiting until the network
6262 * device is fully setup before sending notifications.
6264 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6271 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6273 static int dev_cpu_callback(struct notifier_block *nfb,
6274 unsigned long action,
6277 struct sk_buff **list_skb;
6278 struct sk_buff *skb;
6279 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6280 struct softnet_data *sd, *oldsd;
6282 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6285 local_irq_disable();
6286 cpu = smp_processor_id();
6287 sd = &per_cpu(softnet_data, cpu);
6288 oldsd = &per_cpu(softnet_data, oldcpu);
6290 /* Find end of our completion_queue. */
6291 list_skb = &sd->completion_queue;
6293 list_skb = &(*list_skb)->next;
6294 /* Append completion queue from offline CPU. */
6295 *list_skb = oldsd->completion_queue;
6296 oldsd->completion_queue = NULL;
6298 /* Append output queue from offline CPU. */
6299 if (oldsd->output_queue) {
6300 *sd->output_queue_tailp = oldsd->output_queue;
6301 sd->output_queue_tailp = oldsd->output_queue_tailp;
6302 oldsd->output_queue = NULL;
6303 oldsd->output_queue_tailp = &oldsd->output_queue;
6305 /* Append NAPI poll list from offline CPU. */
6306 if (!list_empty(&oldsd->poll_list)) {
6307 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6308 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6311 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6314 /* Process offline CPU's input_pkt_queue */
6315 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6317 input_queue_head_incr(oldsd);
6319 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6321 input_queue_head_incr(oldsd);
6329 * netdev_increment_features - increment feature set by one
6330 * @all: current feature set
6331 * @one: new feature set
6332 * @mask: mask feature set
6334 * Computes a new feature set after adding a device with feature set
6335 * @one to the master device with current feature set @all. Will not
6336 * enable anything that is off in @mask. Returns the new feature set.
6338 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6340 if (mask & NETIF_F_GEN_CSUM)
6341 mask |= NETIF_F_ALL_CSUM;
6342 mask |= NETIF_F_VLAN_CHALLENGED;
6344 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6345 all &= one | ~NETIF_F_ALL_FOR_ALL;
6347 /* If device needs checksumming, downgrade to it. */
6348 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6349 all &= ~NETIF_F_NO_CSUM;
6351 /* If one device supports hw checksumming, set for all. */
6352 if (all & NETIF_F_GEN_CSUM)
6353 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6357 EXPORT_SYMBOL(netdev_increment_features);
6359 static struct hlist_head *netdev_create_hash(void)
6362 struct hlist_head *hash;
6364 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6366 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6367 INIT_HLIST_HEAD(&hash[i]);
6372 /* Initialize per network namespace state */
6373 static int __net_init netdev_init(struct net *net)
6375 INIT_LIST_HEAD(&net->dev_base_head);
6377 net->dev_name_head = netdev_create_hash();
6378 if (net->dev_name_head == NULL)
6381 net->dev_index_head = netdev_create_hash();
6382 if (net->dev_index_head == NULL)
6388 kfree(net->dev_name_head);
6394 * netdev_drivername - network driver for the device
6395 * @dev: network device
6397 * Determine network driver for device.
6399 const char *netdev_drivername(const struct net_device *dev)
6401 const struct device_driver *driver;
6402 const struct device *parent;
6403 const char *empty = "";
6405 parent = dev->dev.parent;
6409 driver = parent->driver;
6410 if (driver && driver->name)
6411 return driver->name;
6415 int __netdev_printk(const char *level, const struct net_device *dev,
6416 struct va_format *vaf)
6420 if (dev && dev->dev.parent)
6421 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6422 netdev_name(dev), vaf);
6424 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6426 r = printk("%s(NULL net_device): %pV", level, vaf);
6430 EXPORT_SYMBOL(__netdev_printk);
6432 int netdev_printk(const char *level, const struct net_device *dev,
6433 const char *format, ...)
6435 struct va_format vaf;
6439 va_start(args, format);
6444 r = __netdev_printk(level, dev, &vaf);
6449 EXPORT_SYMBOL(netdev_printk);
6451 #define define_netdev_printk_level(func, level) \
6452 int func(const struct net_device *dev, const char *fmt, ...) \
6455 struct va_format vaf; \
6458 va_start(args, fmt); \
6463 r = __netdev_printk(level, dev, &vaf); \
6468 EXPORT_SYMBOL(func);
6470 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6471 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6472 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6473 define_netdev_printk_level(netdev_err, KERN_ERR);
6474 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6475 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6476 define_netdev_printk_level(netdev_info, KERN_INFO);
6478 static void __net_exit netdev_exit(struct net *net)
6480 kfree(net->dev_name_head);
6481 kfree(net->dev_index_head);
6484 static struct pernet_operations __net_initdata netdev_net_ops = {
6485 .init = netdev_init,
6486 .exit = netdev_exit,
6489 static void __net_exit default_device_exit(struct net *net)
6491 struct net_device *dev, *aux;
6493 * Push all migratable network devices back to the
6494 * initial network namespace
6497 for_each_netdev_safe(net, dev, aux) {
6499 char fb_name[IFNAMSIZ];
6501 /* Ignore unmoveable devices (i.e. loopback) */
6502 if (dev->features & NETIF_F_NETNS_LOCAL)
6505 /* Leave virtual devices for the generic cleanup */
6506 if (dev->rtnl_link_ops)
6509 /* Push remaining network devices to init_net */
6510 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6511 err = dev_change_net_namespace(dev, &init_net, fb_name);
6513 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6514 __func__, dev->name, err);
6521 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6523 /* At exit all network devices most be removed from a network
6524 * namespace. Do this in the reverse order of registration.
6525 * Do this across as many network namespaces as possible to
6526 * improve batching efficiency.
6528 struct net_device *dev;
6530 LIST_HEAD(dev_kill_list);
6533 list_for_each_entry(net, net_list, exit_list) {
6534 for_each_netdev_reverse(net, dev) {
6535 if (dev->rtnl_link_ops)
6536 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6538 unregister_netdevice_queue(dev, &dev_kill_list);
6541 unregister_netdevice_many(&dev_kill_list);
6542 list_del(&dev_kill_list);
6546 static struct pernet_operations __net_initdata default_device_ops = {
6547 .exit = default_device_exit,
6548 .exit_batch = default_device_exit_batch,
6552 * Initialize the DEV module. At boot time this walks the device list and
6553 * unhooks any devices that fail to initialise (normally hardware not
6554 * present) and leaves us with a valid list of present and active devices.
6559 * This is called single threaded during boot, so no need
6560 * to take the rtnl semaphore.
6562 static int __init net_dev_init(void)
6564 int i, rc = -ENOMEM;
6566 BUG_ON(!dev_boot_phase);
6568 if (dev_proc_init())
6571 if (netdev_kobject_init())
6574 INIT_LIST_HEAD(&ptype_all);
6575 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6576 INIT_LIST_HEAD(&ptype_base[i]);
6578 if (register_pernet_subsys(&netdev_net_ops))
6582 * Initialise the packet receive queues.
6585 for_each_possible_cpu(i) {
6586 struct softnet_data *sd = &per_cpu(softnet_data, i);
6588 memset(sd, 0, sizeof(*sd));
6589 skb_queue_head_init(&sd->input_pkt_queue);
6590 skb_queue_head_init(&sd->process_queue);
6591 sd->completion_queue = NULL;
6592 INIT_LIST_HEAD(&sd->poll_list);
6593 sd->output_queue = NULL;
6594 sd->output_queue_tailp = &sd->output_queue;
6596 sd->csd.func = rps_trigger_softirq;
6602 sd->backlog.poll = process_backlog;
6603 sd->backlog.weight = weight_p;
6604 sd->backlog.gro_list = NULL;
6605 sd->backlog.gro_count = 0;
6610 /* The loopback device is special if any other network devices
6611 * is present in a network namespace the loopback device must
6612 * be present. Since we now dynamically allocate and free the
6613 * loopback device ensure this invariant is maintained by
6614 * keeping the loopback device as the first device on the
6615 * list of network devices. Ensuring the loopback devices
6616 * is the first device that appears and the last network device
6619 if (register_pernet_device(&loopback_net_ops))
6622 if (register_pernet_device(&default_device_ops))
6625 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6626 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6628 hotcpu_notifier(dev_cpu_callback, 0);
6636 subsys_initcall(net_dev_init);
6638 static int __init initialize_hashrnd(void)
6640 get_random_bytes(&hashrnd, sizeof(hashrnd));
6644 late_initcall_sync(initialize_hashrnd);