2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/if_tunnel.h>
137 #include <linux/if_pppox.h>
138 #include <linux/ppp_defs.h>
139 #include <linux/net_tstamp.h>
141 #include "net-sysfs.h"
143 /* Instead of increasing this, you should create a hash table. */
144 #define MAX_GRO_SKBS 8
146 /* This should be increased if a protocol with a bigger head is added. */
147 #define GRO_MAX_HEAD (MAX_HEADER + 128)
150 * The list of packet types we will receive (as opposed to discard)
151 * and the routines to invoke.
153 * Why 16. Because with 16 the only overlap we get on a hash of the
154 * low nibble of the protocol value is RARP/SNAP/X.25.
156 * NOTE: That is no longer true with the addition of VLAN tags. Not
157 * sure which should go first, but I bet it won't make much
158 * difference if we are running VLANs. The good news is that
159 * this protocol won't be in the list unless compiled in, so
160 * the average user (w/out VLANs) will not be adversely affected.
177 #define PTYPE_HASH_SIZE (16)
178 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
180 static DEFINE_SPINLOCK(ptype_lock);
181 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
182 static struct list_head ptype_all __read_mostly; /* Taps */
185 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
188 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
190 * Writers must hold the rtnl semaphore while they loop through the
191 * dev_base_head list, and hold dev_base_lock for writing when they do the
192 * actual updates. This allows pure readers to access the list even
193 * while a writer is preparing to update it.
195 * To put it another way, dev_base_lock is held for writing only to
196 * protect against pure readers; the rtnl semaphore provides the
197 * protection against other writers.
199 * See, for example usages, register_netdevice() and
200 * unregister_netdevice(), which must be called with the rtnl
203 DEFINE_RWLOCK(dev_base_lock);
204 EXPORT_SYMBOL(dev_base_lock);
206 static inline void dev_base_seq_inc(struct net *net)
208 while (++net->dev_base_seq == 0);
211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
213 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
214 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
217 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
219 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
222 static inline void rps_lock(struct softnet_data *sd)
225 spin_lock(&sd->input_pkt_queue.lock);
229 static inline void rps_unlock(struct softnet_data *sd)
232 spin_unlock(&sd->input_pkt_queue.lock);
236 /* Device list insertion */
237 static int list_netdevice(struct net_device *dev)
239 struct net *net = dev_net(dev);
243 write_lock_bh(&dev_base_lock);
244 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
245 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
246 hlist_add_head_rcu(&dev->index_hlist,
247 dev_index_hash(net, dev->ifindex));
248 write_unlock_bh(&dev_base_lock);
250 dev_base_seq_inc(net);
255 /* Device list removal
256 * caller must respect a RCU grace period before freeing/reusing dev
258 static void unlist_netdevice(struct net_device *dev)
262 /* Unlink dev from the device chain */
263 write_lock_bh(&dev_base_lock);
264 list_del_rcu(&dev->dev_list);
265 hlist_del_rcu(&dev->name_hlist);
266 hlist_del_rcu(&dev->index_hlist);
267 write_unlock_bh(&dev_base_lock);
269 dev_base_seq_inc(dev_net(dev));
276 static RAW_NOTIFIER_HEAD(netdev_chain);
279 * Device drivers call our routines to queue packets here. We empty the
280 * queue in the local softnet handler.
283 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
284 EXPORT_PER_CPU_SYMBOL(softnet_data);
286 #ifdef CONFIG_LOCKDEP
288 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
289 * according to dev->type
291 static const unsigned short netdev_lock_type[] =
292 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
293 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
294 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
295 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
296 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
297 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
298 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
299 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
300 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
301 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
302 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
303 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
304 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
305 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
306 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
307 ARPHRD_VOID, ARPHRD_NONE};
309 static const char *const netdev_lock_name[] =
310 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
311 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
312 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
313 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
314 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
315 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
316 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
317 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
318 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
319 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
320 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
321 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
322 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
323 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
324 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
325 "_xmit_VOID", "_xmit_NONE"};
327 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
328 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
330 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
334 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
335 if (netdev_lock_type[i] == dev_type)
337 /* the last key is used by default */
338 return ARRAY_SIZE(netdev_lock_type) - 1;
341 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
342 unsigned short dev_type)
346 i = netdev_lock_pos(dev_type);
347 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
348 netdev_lock_name[i]);
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
355 i = netdev_lock_pos(dev->type);
356 lockdep_set_class_and_name(&dev->addr_list_lock,
357 &netdev_addr_lock_key[i],
358 netdev_lock_name[i]);
361 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
362 unsigned short dev_type)
365 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
370 /*******************************************************************************
372 Protocol management and registration routines
374 *******************************************************************************/
377 * Add a protocol ID to the list. Now that the input handler is
378 * smarter we can dispense with all the messy stuff that used to be
381 * BEWARE!!! Protocol handlers, mangling input packets,
382 * MUST BE last in hash buckets and checking protocol handlers
383 * MUST start from promiscuous ptype_all chain in net_bh.
384 * It is true now, do not change it.
385 * Explanation follows: if protocol handler, mangling packet, will
386 * be the first on list, it is not able to sense, that packet
387 * is cloned and should be copied-on-write, so that it will
388 * change it and subsequent readers will get broken packet.
392 static inline struct list_head *ptype_head(const struct packet_type *pt)
394 if (pt->type == htons(ETH_P_ALL))
397 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
401 * dev_add_pack - add packet handler
402 * @pt: packet type declaration
404 * Add a protocol handler to the networking stack. The passed &packet_type
405 * is linked into kernel lists and may not be freed until it has been
406 * removed from the kernel lists.
408 * This call does not sleep therefore it can not
409 * guarantee all CPU's that are in middle of receiving packets
410 * will see the new packet type (until the next received packet).
413 void dev_add_pack(struct packet_type *pt)
415 struct list_head *head = ptype_head(pt);
417 spin_lock(&ptype_lock);
418 list_add_rcu(&pt->list, head);
419 spin_unlock(&ptype_lock);
421 EXPORT_SYMBOL(dev_add_pack);
424 * __dev_remove_pack - remove packet handler
425 * @pt: packet type declaration
427 * Remove a protocol handler that was previously added to the kernel
428 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
429 * from the kernel lists and can be freed or reused once this function
432 * The packet type might still be in use by receivers
433 * and must not be freed until after all the CPU's have gone
434 * through a quiescent state.
436 void __dev_remove_pack(struct packet_type *pt)
438 struct list_head *head = ptype_head(pt);
439 struct packet_type *pt1;
441 spin_lock(&ptype_lock);
443 list_for_each_entry(pt1, head, list) {
445 list_del_rcu(&pt->list);
450 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
452 spin_unlock(&ptype_lock);
454 EXPORT_SYMBOL(__dev_remove_pack);
457 * dev_remove_pack - remove packet handler
458 * @pt: packet type declaration
460 * Remove a protocol handler that was previously added to the kernel
461 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
462 * from the kernel lists and can be freed or reused once this function
465 * This call sleeps to guarantee that no CPU is looking at the packet
468 void dev_remove_pack(struct packet_type *pt)
470 __dev_remove_pack(pt);
474 EXPORT_SYMBOL(dev_remove_pack);
476 /******************************************************************************
478 Device Boot-time Settings Routines
480 *******************************************************************************/
482 /* Boot time configuration table */
483 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
486 * netdev_boot_setup_add - add new setup entry
487 * @name: name of the device
488 * @map: configured settings for the device
490 * Adds new setup entry to the dev_boot_setup list. The function
491 * returns 0 on error and 1 on success. This is a generic routine to
494 static int netdev_boot_setup_add(char *name, struct ifmap *map)
496 struct netdev_boot_setup *s;
500 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
501 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
502 memset(s[i].name, 0, sizeof(s[i].name));
503 strlcpy(s[i].name, name, IFNAMSIZ);
504 memcpy(&s[i].map, map, sizeof(s[i].map));
509 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
513 * netdev_boot_setup_check - check boot time settings
514 * @dev: the netdevice
516 * Check boot time settings for the device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found, 1 if they are.
521 int netdev_boot_setup_check(struct net_device *dev)
523 struct netdev_boot_setup *s = dev_boot_setup;
526 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
527 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
528 !strcmp(dev->name, s[i].name)) {
529 dev->irq = s[i].map.irq;
530 dev->base_addr = s[i].map.base_addr;
531 dev->mem_start = s[i].map.mem_start;
532 dev->mem_end = s[i].map.mem_end;
538 EXPORT_SYMBOL(netdev_boot_setup_check);
542 * netdev_boot_base - get address from boot time settings
543 * @prefix: prefix for network device
544 * @unit: id for network device
546 * Check boot time settings for the base address of device.
547 * The found settings are set for the device to be used
548 * later in the device probing.
549 * Returns 0 if no settings found.
551 unsigned long netdev_boot_base(const char *prefix, int unit)
553 const struct netdev_boot_setup *s = dev_boot_setup;
557 sprintf(name, "%s%d", prefix, unit);
560 * If device already registered then return base of 1
561 * to indicate not to probe for this interface
563 if (__dev_get_by_name(&init_net, name))
566 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
567 if (!strcmp(name, s[i].name))
568 return s[i].map.base_addr;
573 * Saves at boot time configured settings for any netdevice.
575 int __init netdev_boot_setup(char *str)
580 str = get_options(str, ARRAY_SIZE(ints), ints);
585 memset(&map, 0, sizeof(map));
589 map.base_addr = ints[2];
591 map.mem_start = ints[3];
593 map.mem_end = ints[4];
595 /* Add new entry to the list */
596 return netdev_boot_setup_add(str, &map);
599 __setup("netdev=", netdev_boot_setup);
601 /*******************************************************************************
603 Device Interface Subroutines
605 *******************************************************************************/
608 * __dev_get_by_name - find a device by its name
609 * @net: the applicable net namespace
610 * @name: name to find
612 * Find an interface by name. Must be called under RTNL semaphore
613 * or @dev_base_lock. If the name is found a pointer to the device
614 * is returned. If the name is not found then %NULL is returned. The
615 * reference counters are not incremented so the caller must be
616 * careful with locks.
619 struct net_device *__dev_get_by_name(struct net *net, const char *name)
621 struct hlist_node *p;
622 struct net_device *dev;
623 struct hlist_head *head = dev_name_hash(net, name);
625 hlist_for_each_entry(dev, p, head, name_hlist)
626 if (!strncmp(dev->name, name, IFNAMSIZ))
631 EXPORT_SYMBOL(__dev_get_by_name);
634 * dev_get_by_name_rcu - find a device by its name
635 * @net: the applicable net namespace
636 * @name: name to find
638 * Find an interface by name.
639 * If the name is found a pointer to the device is returned.
640 * If the name is not found then %NULL is returned.
641 * The reference counters are not incremented so the caller must be
642 * careful with locks. The caller must hold RCU lock.
645 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
647 struct hlist_node *p;
648 struct net_device *dev;
649 struct hlist_head *head = dev_name_hash(net, name);
651 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
652 if (!strncmp(dev->name, name, IFNAMSIZ))
657 EXPORT_SYMBOL(dev_get_by_name_rcu);
660 * dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. This can be called from any
665 * context and does its own locking. The returned handle has
666 * the usage count incremented and the caller must use dev_put() to
667 * release it when it is no longer needed. %NULL is returned if no
668 * matching device is found.
671 struct net_device *dev_get_by_name(struct net *net, const char *name)
673 struct net_device *dev;
676 dev = dev_get_by_name_rcu(net, name);
682 EXPORT_SYMBOL(dev_get_by_name);
685 * __dev_get_by_index - find a device by its ifindex
686 * @net: the applicable net namespace
687 * @ifindex: index of device
689 * Search for an interface by index. Returns %NULL if the device
690 * is not found or a pointer to the device. The device has not
691 * had its reference counter increased so the caller must be careful
692 * about locking. The caller must hold either the RTNL semaphore
696 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
698 struct hlist_node *p;
699 struct net_device *dev;
700 struct hlist_head *head = dev_index_hash(net, ifindex);
702 hlist_for_each_entry(dev, p, head, index_hlist)
703 if (dev->ifindex == ifindex)
708 EXPORT_SYMBOL(__dev_get_by_index);
711 * dev_get_by_index_rcu - find a device by its ifindex
712 * @net: the applicable net namespace
713 * @ifindex: index of device
715 * Search for an interface by index. Returns %NULL if the device
716 * is not found or a pointer to the device. The device has not
717 * had its reference counter increased so the caller must be careful
718 * about locking. The caller must hold RCU lock.
721 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
723 struct hlist_node *p;
724 struct net_device *dev;
725 struct hlist_head *head = dev_index_hash(net, ifindex);
727 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
728 if (dev->ifindex == ifindex)
733 EXPORT_SYMBOL(dev_get_by_index_rcu);
737 * dev_get_by_index - find a device by its ifindex
738 * @net: the applicable net namespace
739 * @ifindex: index of device
741 * Search for an interface by index. Returns NULL if the device
742 * is not found or a pointer to the device. The device returned has
743 * had a reference added and the pointer is safe until the user calls
744 * dev_put to indicate they have finished with it.
747 struct net_device *dev_get_by_index(struct net *net, int ifindex)
749 struct net_device *dev;
752 dev = dev_get_by_index_rcu(net, ifindex);
758 EXPORT_SYMBOL(dev_get_by_index);
761 * dev_getbyhwaddr_rcu - find a device by its hardware address
762 * @net: the applicable net namespace
763 * @type: media type of device
764 * @ha: hardware address
766 * Search for an interface by MAC address. Returns NULL if the device
767 * is not found or a pointer to the device.
768 * The caller must hold RCU or RTNL.
769 * The returned device has not had its ref count increased
770 * and the caller must therefore be careful about locking
774 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
777 struct net_device *dev;
779 for_each_netdev_rcu(net, dev)
780 if (dev->type == type &&
781 !memcmp(dev->dev_addr, ha, dev->addr_len))
786 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
788 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
790 struct net_device *dev;
793 for_each_netdev(net, dev)
794 if (dev->type == type)
799 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
801 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
803 struct net_device *dev, *ret = NULL;
806 for_each_netdev_rcu(net, dev)
807 if (dev->type == type) {
815 EXPORT_SYMBOL(dev_getfirstbyhwtype);
818 * dev_get_by_flags_rcu - find any device with given flags
819 * @net: the applicable net namespace
820 * @if_flags: IFF_* values
821 * @mask: bitmask of bits in if_flags to check
823 * Search for any interface with the given flags. Returns NULL if a device
824 * is not found or a pointer to the device. Must be called inside
825 * rcu_read_lock(), and result refcount is unchanged.
828 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
831 struct net_device *dev, *ret;
834 for_each_netdev_rcu(net, dev) {
835 if (((dev->flags ^ if_flags) & mask) == 0) {
842 EXPORT_SYMBOL(dev_get_by_flags_rcu);
845 * dev_valid_name - check if name is okay for network device
848 * Network device names need to be valid file names to
849 * to allow sysfs to work. We also disallow any kind of
852 int dev_valid_name(const char *name)
856 if (strlen(name) >= IFNAMSIZ)
858 if (!strcmp(name, ".") || !strcmp(name, ".."))
862 if (*name == '/' || *name == ':' || isspace(*name))
868 EXPORT_SYMBOL(dev_valid_name);
871 * __dev_alloc_name - allocate a name for a device
872 * @net: network namespace to allocate the device name in
873 * @name: name format string
874 * @buf: scratch buffer and result name string
876 * Passed a format string - eg "lt%d" it will try and find a suitable
877 * id. It scans list of devices to build up a free map, then chooses
878 * the first empty slot. The caller must hold the dev_base or rtnl lock
879 * while allocating the name and adding the device in order to avoid
881 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
882 * Returns the number of the unit assigned or a negative errno code.
885 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
889 const int max_netdevices = 8*PAGE_SIZE;
890 unsigned long *inuse;
891 struct net_device *d;
893 p = strnchr(name, IFNAMSIZ-1, '%');
896 * Verify the string as this thing may have come from
897 * the user. There must be either one "%d" and no other "%"
900 if (p[1] != 'd' || strchr(p + 2, '%'))
903 /* Use one page as a bit array of possible slots */
904 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
908 for_each_netdev(net, d) {
909 if (!sscanf(d->name, name, &i))
911 if (i < 0 || i >= max_netdevices)
914 /* avoid cases where sscanf is not exact inverse of printf */
915 snprintf(buf, IFNAMSIZ, name, i);
916 if (!strncmp(buf, d->name, IFNAMSIZ))
920 i = find_first_zero_bit(inuse, max_netdevices);
921 free_page((unsigned long) inuse);
925 snprintf(buf, IFNAMSIZ, name, i);
926 if (!__dev_get_by_name(net, buf))
929 /* It is possible to run out of possible slots
930 * when the name is long and there isn't enough space left
931 * for the digits, or if all bits are used.
937 * dev_alloc_name - allocate a name for a device
939 * @name: name format string
941 * Passed a format string - eg "lt%d" it will try and find a suitable
942 * id. It scans list of devices to build up a free map, then chooses
943 * the first empty slot. The caller must hold the dev_base or rtnl lock
944 * while allocating the name and adding the device in order to avoid
946 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
947 * Returns the number of the unit assigned or a negative errno code.
950 int dev_alloc_name(struct net_device *dev, const char *name)
956 BUG_ON(!dev_net(dev));
958 ret = __dev_alloc_name(net, name, buf);
960 strlcpy(dev->name, buf, IFNAMSIZ);
963 EXPORT_SYMBOL(dev_alloc_name);
965 static int dev_get_valid_name(struct net_device *dev, const char *name)
969 BUG_ON(!dev_net(dev));
972 if (!dev_valid_name(name))
975 if (strchr(name, '%'))
976 return dev_alloc_name(dev, name);
977 else if (__dev_get_by_name(net, name))
979 else if (dev->name != name)
980 strlcpy(dev->name, name, IFNAMSIZ);
986 * dev_change_name - change name of a device
988 * @newname: name (or format string) must be at least IFNAMSIZ
990 * Change name of a device, can pass format strings "eth%d".
993 int dev_change_name(struct net_device *dev, const char *newname)
995 char oldname[IFNAMSIZ];
1001 BUG_ON(!dev_net(dev));
1004 if (dev->flags & IFF_UP)
1007 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1010 memcpy(oldname, dev->name, IFNAMSIZ);
1012 err = dev_get_valid_name(dev, newname);
1017 ret = device_rename(&dev->dev, dev->name);
1019 memcpy(dev->name, oldname, IFNAMSIZ);
1023 write_lock_bh(&dev_base_lock);
1024 hlist_del_rcu(&dev->name_hlist);
1025 write_unlock_bh(&dev_base_lock);
1029 write_lock_bh(&dev_base_lock);
1030 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1031 write_unlock_bh(&dev_base_lock);
1033 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1034 ret = notifier_to_errno(ret);
1037 /* err >= 0 after dev_alloc_name() or stores the first errno */
1040 memcpy(dev->name, oldname, IFNAMSIZ);
1044 "%s: name change rollback failed: %d.\n",
1053 * dev_set_alias - change ifalias of a device
1055 * @alias: name up to IFALIASZ
1056 * @len: limit of bytes to copy from info
1058 * Set ifalias for a device,
1060 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1066 if (len >= IFALIASZ)
1071 kfree(dev->ifalias);
1072 dev->ifalias = NULL;
1077 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1080 dev->ifalias = new_ifalias;
1082 strlcpy(dev->ifalias, alias, len+1);
1088 * netdev_features_change - device changes features
1089 * @dev: device to cause notification
1091 * Called to indicate a device has changed features.
1093 void netdev_features_change(struct net_device *dev)
1095 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1097 EXPORT_SYMBOL(netdev_features_change);
1100 * netdev_state_change - device changes state
1101 * @dev: device to cause notification
1103 * Called to indicate a device has changed state. This function calls
1104 * the notifier chains for netdev_chain and sends a NEWLINK message
1105 * to the routing socket.
1107 void netdev_state_change(struct net_device *dev)
1109 if (dev->flags & IFF_UP) {
1110 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1111 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1114 EXPORT_SYMBOL(netdev_state_change);
1116 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1118 return call_netdevice_notifiers(event, dev);
1120 EXPORT_SYMBOL(netdev_bonding_change);
1123 * dev_load - load a network module
1124 * @net: the applicable net namespace
1125 * @name: name of interface
1127 * If a network interface is not present and the process has suitable
1128 * privileges this function loads the module. If module loading is not
1129 * available in this kernel then it becomes a nop.
1132 void dev_load(struct net *net, const char *name)
1134 struct net_device *dev;
1138 dev = dev_get_by_name_rcu(net, name);
1142 if (no_module && capable(CAP_NET_ADMIN))
1143 no_module = request_module("netdev-%s", name);
1144 if (no_module && capable(CAP_SYS_MODULE)) {
1145 if (!request_module("%s", name))
1146 pr_err("Loading kernel module for a network device "
1147 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1151 EXPORT_SYMBOL(dev_load);
1153 static int __dev_open(struct net_device *dev)
1155 const struct net_device_ops *ops = dev->netdev_ops;
1160 if (!netif_device_present(dev))
1163 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1164 ret = notifier_to_errno(ret);
1168 set_bit(__LINK_STATE_START, &dev->state);
1170 if (ops->ndo_validate_addr)
1171 ret = ops->ndo_validate_addr(dev);
1173 if (!ret && ops->ndo_open)
1174 ret = ops->ndo_open(dev);
1177 clear_bit(__LINK_STATE_START, &dev->state);
1179 dev->flags |= IFF_UP;
1180 net_dmaengine_get();
1181 dev_set_rx_mode(dev);
1183 add_device_randomness(dev->dev_addr, dev->addr_len);
1190 * dev_open - prepare an interface for use.
1191 * @dev: device to open
1193 * Takes a device from down to up state. The device's private open
1194 * function is invoked and then the multicast lists are loaded. Finally
1195 * the device is moved into the up state and a %NETDEV_UP message is
1196 * sent to the netdev notifier chain.
1198 * Calling this function on an active interface is a nop. On a failure
1199 * a negative errno code is returned.
1201 int dev_open(struct net_device *dev)
1205 if (dev->flags & IFF_UP)
1208 ret = __dev_open(dev);
1212 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1213 call_netdevice_notifiers(NETDEV_UP, dev);
1217 EXPORT_SYMBOL(dev_open);
1219 static int __dev_close_many(struct list_head *head)
1221 struct net_device *dev;
1226 list_for_each_entry(dev, head, unreg_list) {
1227 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1229 clear_bit(__LINK_STATE_START, &dev->state);
1231 /* Synchronize to scheduled poll. We cannot touch poll list, it
1232 * can be even on different cpu. So just clear netif_running().
1234 * dev->stop() will invoke napi_disable() on all of it's
1235 * napi_struct instances on this device.
1237 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1240 dev_deactivate_many(head);
1242 list_for_each_entry(dev, head, unreg_list) {
1243 const struct net_device_ops *ops = dev->netdev_ops;
1246 * Call the device specific close. This cannot fail.
1247 * Only if device is UP
1249 * We allow it to be called even after a DETACH hot-plug
1255 dev->flags &= ~IFF_UP;
1256 net_dmaengine_put();
1262 static int __dev_close(struct net_device *dev)
1267 list_add(&dev->unreg_list, &single);
1268 retval = __dev_close_many(&single);
1273 static int dev_close_many(struct list_head *head)
1275 struct net_device *dev, *tmp;
1276 LIST_HEAD(tmp_list);
1278 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1279 if (!(dev->flags & IFF_UP))
1280 list_move(&dev->unreg_list, &tmp_list);
1282 __dev_close_many(head);
1284 list_for_each_entry(dev, head, unreg_list) {
1285 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1286 call_netdevice_notifiers(NETDEV_DOWN, dev);
1289 /* rollback_registered_many needs the complete original list */
1290 list_splice(&tmp_list, head);
1295 * dev_close - shutdown an interface.
1296 * @dev: device to shutdown
1298 * This function moves an active device into down state. A
1299 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1300 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1303 int dev_close(struct net_device *dev)
1305 if (dev->flags & IFF_UP) {
1308 list_add(&dev->unreg_list, &single);
1309 dev_close_many(&single);
1314 EXPORT_SYMBOL(dev_close);
1318 * dev_disable_lro - disable Large Receive Offload on a device
1321 * Disable Large Receive Offload (LRO) on a net device. Must be
1322 * called under RTNL. This is needed if received packets may be
1323 * forwarded to another interface.
1325 void dev_disable_lro(struct net_device *dev)
1330 * If we're trying to disable lro on a vlan device
1331 * use the underlying physical device instead
1333 if (is_vlan_dev(dev))
1334 dev = vlan_dev_real_dev(dev);
1336 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1337 flags = dev->ethtool_ops->get_flags(dev);
1339 flags = ethtool_op_get_flags(dev);
1341 if (!(flags & ETH_FLAG_LRO))
1344 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1345 if (unlikely(dev->features & NETIF_F_LRO))
1346 netdev_WARN(dev, "failed to disable LRO!\n");
1348 EXPORT_SYMBOL(dev_disable_lro);
1351 static int dev_boot_phase = 1;
1354 * register_netdevice_notifier - register a network notifier block
1357 * Register a notifier to be called when network device events occur.
1358 * The notifier passed is linked into the kernel structures and must
1359 * not be reused until it has been unregistered. A negative errno code
1360 * is returned on a failure.
1362 * When registered all registration and up events are replayed
1363 * to the new notifier to allow device to have a race free
1364 * view of the network device list.
1367 int register_netdevice_notifier(struct notifier_block *nb)
1369 struct net_device *dev;
1370 struct net_device *last;
1375 err = raw_notifier_chain_register(&netdev_chain, nb);
1381 for_each_netdev(net, dev) {
1382 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1383 err = notifier_to_errno(err);
1387 if (!(dev->flags & IFF_UP))
1390 nb->notifier_call(nb, NETDEV_UP, dev);
1401 for_each_netdev(net, dev) {
1405 if (dev->flags & IFF_UP) {
1406 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1407 nb->notifier_call(nb, NETDEV_DOWN, dev);
1409 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1410 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1415 raw_notifier_chain_unregister(&netdev_chain, nb);
1418 EXPORT_SYMBOL(register_netdevice_notifier);
1421 * unregister_netdevice_notifier - unregister a network notifier block
1424 * Unregister a notifier previously registered by
1425 * register_netdevice_notifier(). The notifier is unlinked into the
1426 * kernel structures and may then be reused. A negative errno code
1427 * is returned on a failure.
1429 * After unregistering unregister and down device events are synthesized
1430 * for all devices on the device list to the removed notifier to remove
1431 * the need for special case cleanup code.
1434 int unregister_netdevice_notifier(struct notifier_block *nb)
1436 struct net_device *dev;
1441 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1446 for_each_netdev(net, dev) {
1447 if (dev->flags & IFF_UP) {
1448 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1449 nb->notifier_call(nb, NETDEV_DOWN, dev);
1451 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1452 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1459 EXPORT_SYMBOL(unregister_netdevice_notifier);
1462 * call_netdevice_notifiers - call all network notifier blocks
1463 * @val: value passed unmodified to notifier function
1464 * @dev: net_device pointer passed unmodified to notifier function
1466 * Call all network notifier blocks. Parameters and return value
1467 * are as for raw_notifier_call_chain().
1470 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1473 return raw_notifier_call_chain(&netdev_chain, val, dev);
1475 EXPORT_SYMBOL(call_netdevice_notifiers);
1477 /* When > 0 there are consumers of rx skb time stamps */
1478 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1480 void net_enable_timestamp(void)
1482 atomic_inc(&netstamp_needed);
1484 EXPORT_SYMBOL(net_enable_timestamp);
1486 void net_disable_timestamp(void)
1488 atomic_dec(&netstamp_needed);
1490 EXPORT_SYMBOL(net_disable_timestamp);
1492 static inline void net_timestamp_set(struct sk_buff *skb)
1494 if (atomic_read(&netstamp_needed))
1495 __net_timestamp(skb);
1497 skb->tstamp.tv64 = 0;
1500 static inline void net_timestamp_check(struct sk_buff *skb)
1502 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1503 __net_timestamp(skb);
1506 static int net_hwtstamp_validate(struct ifreq *ifr)
1508 struct hwtstamp_config cfg;
1509 enum hwtstamp_tx_types tx_type;
1510 enum hwtstamp_rx_filters rx_filter;
1511 int tx_type_valid = 0;
1512 int rx_filter_valid = 0;
1514 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1517 if (cfg.flags) /* reserved for future extensions */
1520 tx_type = cfg.tx_type;
1521 rx_filter = cfg.rx_filter;
1524 case HWTSTAMP_TX_OFF:
1525 case HWTSTAMP_TX_ON:
1526 case HWTSTAMP_TX_ONESTEP_SYNC:
1531 switch (rx_filter) {
1532 case HWTSTAMP_FILTER_NONE:
1533 case HWTSTAMP_FILTER_ALL:
1534 case HWTSTAMP_FILTER_SOME:
1535 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1536 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1537 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1538 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1539 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1540 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1541 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1542 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1543 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1544 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1545 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1546 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1547 rx_filter_valid = 1;
1551 if (!tx_type_valid || !rx_filter_valid)
1557 static inline bool is_skb_forwardable(struct net_device *dev,
1558 struct sk_buff *skb)
1562 if (!(dev->flags & IFF_UP))
1565 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1566 if (skb->len <= len)
1569 /* if TSO is enabled, we don't care about the length as the packet
1570 * could be forwarded without being segmented before
1572 if (skb_is_gso(skb))
1579 * dev_forward_skb - loopback an skb to another netif
1581 * @dev: destination network device
1582 * @skb: buffer to forward
1585 * NET_RX_SUCCESS (no congestion)
1586 * NET_RX_DROP (packet was dropped, but freed)
1588 * dev_forward_skb can be used for injecting an skb from the
1589 * start_xmit function of one device into the receive queue
1590 * of another device.
1592 * The receiving device may be in another namespace, so
1593 * we have to clear all information in the skb that could
1594 * impact namespace isolation.
1596 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1598 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1599 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1600 atomic_long_inc(&dev->rx_dropped);
1609 if (unlikely(!is_skb_forwardable(dev, skb))) {
1610 atomic_long_inc(&dev->rx_dropped);
1616 skb->tstamp.tv64 = 0;
1617 skb->pkt_type = PACKET_HOST;
1618 skb->protocol = eth_type_trans(skb, dev);
1619 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1623 nf_reset_trace(skb);
1624 return netif_rx(skb);
1626 EXPORT_SYMBOL_GPL(dev_forward_skb);
1628 static inline int deliver_skb(struct sk_buff *skb,
1629 struct packet_type *pt_prev,
1630 struct net_device *orig_dev)
1632 atomic_inc(&skb->users);
1633 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1636 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1638 if (!ptype->af_packet_priv || !skb->sk)
1641 if (ptype->id_match)
1642 return ptype->id_match(ptype, skb->sk);
1643 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1650 * Support routine. Sends outgoing frames to any network
1651 * taps currently in use.
1654 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1656 struct packet_type *ptype;
1657 struct sk_buff *skb2 = NULL;
1658 struct packet_type *pt_prev = NULL;
1661 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1662 /* Never send packets back to the socket
1663 * they originated from - MvS (miquels@drinkel.ow.org)
1665 if ((ptype->dev == dev || !ptype->dev) &&
1666 (!skb_loop_sk(ptype, skb))) {
1668 deliver_skb(skb2, pt_prev, skb->dev);
1673 skb2 = skb_clone(skb, GFP_ATOMIC);
1677 net_timestamp_set(skb2);
1679 /* skb->nh should be correctly
1680 set by sender, so that the second statement is
1681 just protection against buggy protocols.
1683 skb_reset_mac_header(skb2);
1685 if (skb_network_header(skb2) < skb2->data ||
1686 skb2->network_header > skb2->tail) {
1687 if (net_ratelimit())
1688 printk(KERN_CRIT "protocol %04x is "
1690 ntohs(skb2->protocol),
1692 skb_reset_network_header(skb2);
1695 skb2->transport_header = skb2->network_header;
1696 skb2->pkt_type = PACKET_OUTGOING;
1701 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1705 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1706 * @dev: Network device
1707 * @txq: number of queues available
1709 * If real_num_tx_queues is changed the tc mappings may no longer be
1710 * valid. To resolve this verify the tc mapping remains valid and if
1711 * not NULL the mapping. With no priorities mapping to this
1712 * offset/count pair it will no longer be used. In the worst case TC0
1713 * is invalid nothing can be done so disable priority mappings. If is
1714 * expected that drivers will fix this mapping if they can before
1715 * calling netif_set_real_num_tx_queues.
1717 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1720 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1722 /* If TC0 is invalidated disable TC mapping */
1723 if (tc->offset + tc->count > txq) {
1724 pr_warning("Number of in use tx queues changed "
1725 "invalidating tc mappings. Priority "
1726 "traffic classification disabled!\n");
1731 /* Invalidated prio to tc mappings set to TC0 */
1732 for (i = 1; i < TC_BITMASK + 1; i++) {
1733 int q = netdev_get_prio_tc_map(dev, i);
1735 tc = &dev->tc_to_txq[q];
1736 if (tc->offset + tc->count > txq) {
1737 pr_warning("Number of in use tx queues "
1738 "changed. Priority %i to tc "
1739 "mapping %i is no longer valid "
1740 "setting map to 0\n",
1742 netdev_set_prio_tc_map(dev, i, 0);
1748 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1749 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1751 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1755 if (txq < 1 || txq > dev->num_tx_queues)
1758 if (dev->reg_state == NETREG_REGISTERED ||
1759 dev->reg_state == NETREG_UNREGISTERING) {
1762 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1768 netif_setup_tc(dev, txq);
1770 if (txq < dev->real_num_tx_queues)
1771 qdisc_reset_all_tx_gt(dev, txq);
1774 dev->real_num_tx_queues = txq;
1777 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1781 * netif_set_real_num_rx_queues - set actual number of RX queues used
1782 * @dev: Network device
1783 * @rxq: Actual number of RX queues
1785 * This must be called either with the rtnl_lock held or before
1786 * registration of the net device. Returns 0 on success, or a
1787 * negative error code. If called before registration, it always
1790 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1794 if (rxq < 1 || rxq > dev->num_rx_queues)
1797 if (dev->reg_state == NETREG_REGISTERED) {
1800 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1806 dev->real_num_rx_queues = rxq;
1809 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1812 static inline void __netif_reschedule(struct Qdisc *q)
1814 struct softnet_data *sd;
1815 unsigned long flags;
1817 local_irq_save(flags);
1818 sd = &__get_cpu_var(softnet_data);
1819 q->next_sched = NULL;
1820 *sd->output_queue_tailp = q;
1821 sd->output_queue_tailp = &q->next_sched;
1822 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1823 local_irq_restore(flags);
1826 void __netif_schedule(struct Qdisc *q)
1828 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1829 __netif_reschedule(q);
1831 EXPORT_SYMBOL(__netif_schedule);
1833 void dev_kfree_skb_irq(struct sk_buff *skb)
1835 if (atomic_dec_and_test(&skb->users)) {
1836 struct softnet_data *sd;
1837 unsigned long flags;
1839 local_irq_save(flags);
1840 sd = &__get_cpu_var(softnet_data);
1841 skb->next = sd->completion_queue;
1842 sd->completion_queue = skb;
1843 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1844 local_irq_restore(flags);
1847 EXPORT_SYMBOL(dev_kfree_skb_irq);
1849 void dev_kfree_skb_any(struct sk_buff *skb)
1851 if (in_irq() || irqs_disabled())
1852 dev_kfree_skb_irq(skb);
1856 EXPORT_SYMBOL(dev_kfree_skb_any);
1860 * netif_device_detach - mark device as removed
1861 * @dev: network device
1863 * Mark device as removed from system and therefore no longer available.
1865 void netif_device_detach(struct net_device *dev)
1867 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1868 netif_running(dev)) {
1869 netif_tx_stop_all_queues(dev);
1872 EXPORT_SYMBOL(netif_device_detach);
1875 * netif_device_attach - mark device as attached
1876 * @dev: network device
1878 * Mark device as attached from system and restart if needed.
1880 void netif_device_attach(struct net_device *dev)
1882 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1883 netif_running(dev)) {
1884 netif_tx_wake_all_queues(dev);
1885 __netdev_watchdog_up(dev);
1888 EXPORT_SYMBOL(netif_device_attach);
1891 * Invalidate hardware checksum when packet is to be mangled, and
1892 * complete checksum manually on outgoing path.
1894 int skb_checksum_help(struct sk_buff *skb)
1897 int ret = 0, offset;
1899 if (skb->ip_summed == CHECKSUM_COMPLETE)
1900 goto out_set_summed;
1902 if (unlikely(skb_shinfo(skb)->gso_size)) {
1903 /* Let GSO fix up the checksum. */
1904 goto out_set_summed;
1907 offset = skb_checksum_start_offset(skb);
1908 BUG_ON(offset >= skb_headlen(skb));
1909 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1911 offset += skb->csum_offset;
1912 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1914 if (skb_cloned(skb) &&
1915 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1916 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1921 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1923 skb->ip_summed = CHECKSUM_NONE;
1927 EXPORT_SYMBOL(skb_checksum_help);
1930 * skb_gso_segment - Perform segmentation on skb.
1931 * @skb: buffer to segment
1932 * @features: features for the output path (see dev->features)
1934 * This function segments the given skb and returns a list of segments.
1936 * It may return NULL if the skb requires no segmentation. This is
1937 * only possible when GSO is used for verifying header integrity.
1939 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1941 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1942 struct packet_type *ptype;
1943 __be16 type = skb->protocol;
1944 int vlan_depth = ETH_HLEN;
1947 while (type == htons(ETH_P_8021Q)) {
1948 struct vlan_hdr *vh;
1950 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1951 return ERR_PTR(-EINVAL);
1953 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1954 type = vh->h_vlan_encapsulated_proto;
1955 vlan_depth += VLAN_HLEN;
1958 skb_reset_mac_header(skb);
1959 skb->mac_len = skb->network_header - skb->mac_header;
1960 __skb_pull(skb, skb->mac_len);
1962 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1963 struct net_device *dev = skb->dev;
1964 struct ethtool_drvinfo info = {};
1966 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1967 dev->ethtool_ops->get_drvinfo(dev, &info);
1969 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1970 info.driver, dev ? dev->features : 0L,
1971 skb->sk ? skb->sk->sk_route_caps : 0L,
1972 skb->len, skb->data_len, skb->ip_summed);
1974 if (skb_header_cloned(skb) &&
1975 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1976 return ERR_PTR(err);
1980 list_for_each_entry_rcu(ptype,
1981 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1982 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1983 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1984 err = ptype->gso_send_check(skb);
1985 segs = ERR_PTR(err);
1986 if (err || skb_gso_ok(skb, features))
1988 __skb_push(skb, (skb->data -
1989 skb_network_header(skb)));
1991 segs = ptype->gso_segment(skb, features);
1997 __skb_push(skb, skb->data - skb_mac_header(skb));
2001 EXPORT_SYMBOL(skb_gso_segment);
2003 /* Take action when hardware reception checksum errors are detected. */
2005 void netdev_rx_csum_fault(struct net_device *dev)
2007 if (net_ratelimit()) {
2008 printk(KERN_ERR "%s: hw csum failure.\n",
2009 dev ? dev->name : "<unknown>");
2013 EXPORT_SYMBOL(netdev_rx_csum_fault);
2016 /* Actually, we should eliminate this check as soon as we know, that:
2017 * 1. IOMMU is present and allows to map all the memory.
2018 * 2. No high memory really exists on this machine.
2021 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2023 #ifdef CONFIG_HIGHMEM
2025 if (!(dev->features & NETIF_F_HIGHDMA)) {
2026 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2027 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2028 if (PageHighMem(skb_frag_page(frag)))
2033 if (PCI_DMA_BUS_IS_PHYS) {
2034 struct device *pdev = dev->dev.parent;
2038 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2039 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2040 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2041 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2050 void (*destructor)(struct sk_buff *skb);
2053 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2055 static void dev_gso_skb_destructor(struct sk_buff *skb)
2057 struct dev_gso_cb *cb;
2060 struct sk_buff *nskb = skb->next;
2062 skb->next = nskb->next;
2065 } while (skb->next);
2067 cb = DEV_GSO_CB(skb);
2069 cb->destructor(skb);
2073 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2074 * @skb: buffer to segment
2075 * @features: device features as applicable to this skb
2077 * This function segments the given skb and stores the list of segments
2080 static int dev_gso_segment(struct sk_buff *skb, int features)
2082 struct sk_buff *segs;
2084 segs = skb_gso_segment(skb, features);
2086 /* Verifying header integrity only. */
2091 return PTR_ERR(segs);
2094 DEV_GSO_CB(skb)->destructor = skb->destructor;
2095 skb->destructor = dev_gso_skb_destructor;
2100 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2102 return ((features & NETIF_F_GEN_CSUM) ||
2103 ((features & NETIF_F_V4_CSUM) &&
2104 protocol == htons(ETH_P_IP)) ||
2105 ((features & NETIF_F_V6_CSUM) &&
2106 protocol == htons(ETH_P_IPV6)) ||
2107 ((features & NETIF_F_FCOE_CRC) &&
2108 protocol == htons(ETH_P_FCOE)));
2111 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2113 if (skb->ip_summed != CHECKSUM_NONE &&
2114 !can_checksum_protocol(features, protocol)) {
2115 features &= ~NETIF_F_ALL_CSUM;
2116 features &= ~NETIF_F_SG;
2117 } else if (illegal_highdma(skb->dev, skb)) {
2118 features &= ~NETIF_F_SG;
2124 u32 netif_skb_features(struct sk_buff *skb)
2126 __be16 protocol = skb->protocol;
2127 u32 features = skb->dev->features;
2129 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2130 features &= ~NETIF_F_GSO_MASK;
2132 if (!vlan_tx_tag_present(skb)) {
2133 if (unlikely(protocol == htons(ETH_P_8021Q))) {
2134 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2135 protocol = veh->h_vlan_encapsulated_proto;
2137 return harmonize_features(skb, protocol, features);
2141 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2143 if (protocol != htons(ETH_P_8021Q)) {
2144 return harmonize_features(skb, protocol, features);
2146 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2147 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2148 return harmonize_features(skb, protocol, features);
2151 EXPORT_SYMBOL(netif_skb_features);
2154 * Returns true if either:
2155 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2156 * 2. skb is fragmented and the device does not support SG, or if
2157 * at least one of fragments is in highmem and device does not
2158 * support DMA from it.
2160 static inline int skb_needs_linearize(struct sk_buff *skb,
2163 return skb_is_nonlinear(skb) &&
2164 ((skb_has_frag_list(skb) &&
2165 !(features & NETIF_F_FRAGLIST)) ||
2166 (skb_shinfo(skb)->nr_frags &&
2167 !(features & NETIF_F_SG)));
2170 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2171 struct netdev_queue *txq)
2173 const struct net_device_ops *ops = dev->netdev_ops;
2174 int rc = NETDEV_TX_OK;
2175 unsigned int skb_len;
2177 if (likely(!skb->next)) {
2181 * If device doesn't need skb->dst, release it right now while
2182 * its hot in this cpu cache
2184 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2187 if (!list_empty(&ptype_all))
2188 dev_queue_xmit_nit(skb, dev);
2190 features = netif_skb_features(skb);
2192 if (vlan_tx_tag_present(skb) &&
2193 !(features & NETIF_F_HW_VLAN_TX)) {
2194 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2201 if (netif_needs_gso(skb, features)) {
2202 if (unlikely(dev_gso_segment(skb, features)))
2207 if (skb_needs_linearize(skb, features) &&
2208 __skb_linearize(skb))
2211 /* If packet is not checksummed and device does not
2212 * support checksumming for this protocol, complete
2213 * checksumming here.
2215 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2216 skb_set_transport_header(skb,
2217 skb_checksum_start_offset(skb));
2218 if (!(features & NETIF_F_ALL_CSUM) &&
2219 skb_checksum_help(skb))
2225 rc = ops->ndo_start_xmit(skb, dev);
2226 trace_net_dev_xmit(skb, rc, dev, skb_len);
2227 if (rc == NETDEV_TX_OK)
2228 txq_trans_update(txq);
2234 struct sk_buff *nskb = skb->next;
2236 skb->next = nskb->next;
2240 * If device doesn't need nskb->dst, release it right now while
2241 * its hot in this cpu cache
2243 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2246 skb_len = nskb->len;
2247 rc = ops->ndo_start_xmit(nskb, dev);
2248 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2249 if (unlikely(rc != NETDEV_TX_OK)) {
2250 if (rc & ~NETDEV_TX_MASK)
2251 goto out_kfree_gso_skb;
2252 nskb->next = skb->next;
2256 txq_trans_update(txq);
2257 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2258 return NETDEV_TX_BUSY;
2259 } while (skb->next);
2262 if (likely(skb->next == NULL))
2263 skb->destructor = DEV_GSO_CB(skb)->destructor;
2270 static u32 hashrnd __read_mostly;
2273 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2274 * to be used as a distribution range.
2276 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2277 unsigned int num_tx_queues)
2281 u16 qcount = num_tx_queues;
2283 if (skb_rx_queue_recorded(skb)) {
2284 hash = skb_get_rx_queue(skb);
2285 while (unlikely(hash >= num_tx_queues))
2286 hash -= num_tx_queues;
2291 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2292 qoffset = dev->tc_to_txq[tc].offset;
2293 qcount = dev->tc_to_txq[tc].count;
2296 if (skb->sk && skb->sk->sk_hash)
2297 hash = skb->sk->sk_hash;
2299 hash = (__force u16) skb->protocol;
2300 hash = jhash_1word(hash, hashrnd);
2302 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2304 EXPORT_SYMBOL(__skb_tx_hash);
2306 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2308 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2309 if (net_ratelimit()) {
2310 pr_warning("%s selects TX queue %d, but "
2311 "real number of TX queues is %d\n",
2312 dev->name, queue_index, dev->real_num_tx_queues);
2319 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2322 struct xps_dev_maps *dev_maps;
2323 struct xps_map *map;
2324 int queue_index = -1;
2327 dev_maps = rcu_dereference(dev->xps_maps);
2329 map = rcu_dereference(
2330 dev_maps->cpu_map[raw_smp_processor_id()]);
2333 queue_index = map->queues[0];
2336 if (skb->sk && skb->sk->sk_hash)
2337 hash = skb->sk->sk_hash;
2339 hash = (__force u16) skb->protocol ^
2341 hash = jhash_1word(hash, hashrnd);
2342 queue_index = map->queues[
2343 ((u64)hash * map->len) >> 32];
2345 if (unlikely(queue_index >= dev->real_num_tx_queues))
2357 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2358 struct sk_buff *skb)
2361 const struct net_device_ops *ops = dev->netdev_ops;
2363 if (dev->real_num_tx_queues == 1)
2365 else if (ops->ndo_select_queue) {
2366 queue_index = ops->ndo_select_queue(dev, skb);
2367 queue_index = dev_cap_txqueue(dev, queue_index);
2369 struct sock *sk = skb->sk;
2370 queue_index = sk_tx_queue_get(sk);
2372 if (queue_index < 0 || skb->ooo_okay ||
2373 queue_index >= dev->real_num_tx_queues) {
2374 int old_index = queue_index;
2376 queue_index = get_xps_queue(dev, skb);
2377 if (queue_index < 0)
2378 queue_index = skb_tx_hash(dev, skb);
2380 if (queue_index != old_index && sk) {
2381 struct dst_entry *dst =
2382 rcu_dereference_check(sk->sk_dst_cache, 1);
2384 if (dst && skb_dst(skb) == dst)
2385 sk_tx_queue_set(sk, queue_index);
2390 skb_set_queue_mapping(skb, queue_index);
2391 return netdev_get_tx_queue(dev, queue_index);
2394 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2395 struct net_device *dev,
2396 struct netdev_queue *txq)
2398 spinlock_t *root_lock = qdisc_lock(q);
2402 qdisc_skb_cb(skb)->pkt_len = skb->len;
2403 qdisc_calculate_pkt_len(skb, q);
2405 * Heuristic to force contended enqueues to serialize on a
2406 * separate lock before trying to get qdisc main lock.
2407 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2408 * and dequeue packets faster.
2410 contended = qdisc_is_running(q);
2411 if (unlikely(contended))
2412 spin_lock(&q->busylock);
2414 spin_lock(root_lock);
2415 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2418 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2419 qdisc_run_begin(q)) {
2421 * This is a work-conserving queue; there are no old skbs
2422 * waiting to be sent out; and the qdisc is not running -
2423 * xmit the skb directly.
2425 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2428 qdisc_bstats_update(q, skb);
2430 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2431 if (unlikely(contended)) {
2432 spin_unlock(&q->busylock);
2439 rc = NET_XMIT_SUCCESS;
2442 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2443 if (qdisc_run_begin(q)) {
2444 if (unlikely(contended)) {
2445 spin_unlock(&q->busylock);
2451 spin_unlock(root_lock);
2452 if (unlikely(contended))
2453 spin_unlock(&q->busylock);
2457 static DEFINE_PER_CPU(int, xmit_recursion);
2458 #define RECURSION_LIMIT 10
2461 * dev_queue_xmit - transmit a buffer
2462 * @skb: buffer to transmit
2464 * Queue a buffer for transmission to a network device. The caller must
2465 * have set the device and priority and built the buffer before calling
2466 * this function. The function can be called from an interrupt.
2468 * A negative errno code is returned on a failure. A success does not
2469 * guarantee the frame will be transmitted as it may be dropped due
2470 * to congestion or traffic shaping.
2472 * -----------------------------------------------------------------------------------
2473 * I notice this method can also return errors from the queue disciplines,
2474 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2477 * Regardless of the return value, the skb is consumed, so it is currently
2478 * difficult to retry a send to this method. (You can bump the ref count
2479 * before sending to hold a reference for retry if you are careful.)
2481 * When calling this method, interrupts MUST be enabled. This is because
2482 * the BH enable code must have IRQs enabled so that it will not deadlock.
2485 int dev_queue_xmit(struct sk_buff *skb)
2487 struct net_device *dev = skb->dev;
2488 struct netdev_queue *txq;
2492 /* Disable soft irqs for various locks below. Also
2493 * stops preemption for RCU.
2497 txq = dev_pick_tx(dev, skb);
2498 q = rcu_dereference_bh(txq->qdisc);
2500 #ifdef CONFIG_NET_CLS_ACT
2501 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2503 trace_net_dev_queue(skb);
2505 rc = __dev_xmit_skb(skb, q, dev, txq);
2509 /* The device has no queue. Common case for software devices:
2510 loopback, all the sorts of tunnels...
2512 Really, it is unlikely that netif_tx_lock protection is necessary
2513 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2515 However, it is possible, that they rely on protection
2518 Check this and shot the lock. It is not prone from deadlocks.
2519 Either shot noqueue qdisc, it is even simpler 8)
2521 if (dev->flags & IFF_UP) {
2522 int cpu = smp_processor_id(); /* ok because BHs are off */
2524 if (txq->xmit_lock_owner != cpu) {
2526 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2527 goto recursion_alert;
2529 HARD_TX_LOCK(dev, txq, cpu);
2531 if (!netif_tx_queue_stopped(txq)) {
2532 __this_cpu_inc(xmit_recursion);
2533 rc = dev_hard_start_xmit(skb, dev, txq);
2534 __this_cpu_dec(xmit_recursion);
2535 if (dev_xmit_complete(rc)) {
2536 HARD_TX_UNLOCK(dev, txq);
2540 HARD_TX_UNLOCK(dev, txq);
2541 if (net_ratelimit())
2542 printk(KERN_CRIT "Virtual device %s asks to "
2543 "queue packet!\n", dev->name);
2545 /* Recursion is detected! It is possible,
2549 if (net_ratelimit())
2550 printk(KERN_CRIT "Dead loop on virtual device "
2551 "%s, fix it urgently!\n", dev->name);
2556 rcu_read_unlock_bh();
2561 rcu_read_unlock_bh();
2564 EXPORT_SYMBOL(dev_queue_xmit);
2567 /*=======================================================================
2569 =======================================================================*/
2571 int netdev_max_backlog __read_mostly = 1000;
2572 int netdev_tstamp_prequeue __read_mostly = 1;
2573 int netdev_budget __read_mostly = 300;
2574 int weight_p __read_mostly = 64; /* old backlog weight */
2576 /* Called with irq disabled */
2577 static inline void ____napi_schedule(struct softnet_data *sd,
2578 struct napi_struct *napi)
2580 list_add_tail(&napi->poll_list, &sd->poll_list);
2581 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2585 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2586 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2587 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2588 * if hash is a canonical 4-tuple hash over transport ports.
2590 void __skb_get_rxhash(struct sk_buff *skb)
2592 int nhoff, hash = 0, poff;
2593 const struct ipv6hdr *ip6;
2594 const struct iphdr *ip;
2595 const struct vlan_hdr *vlan;
2604 nhoff = skb_network_offset(skb);
2605 proto = skb->protocol;
2609 case __constant_htons(ETH_P_IP):
2611 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2614 ip = (const struct iphdr *) (skb->data + nhoff);
2617 if (ip_is_fragment(ip))
2620 ip_proto = ip->protocol;
2621 addr1 = (__force u32) ip->saddr;
2622 addr2 = (__force u32) ip->daddr;
2623 nhoff += ip->ihl * 4;
2625 case __constant_htons(ETH_P_IPV6):
2627 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2630 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2631 ip_proto = ip6->nexthdr;
2632 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2633 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2636 case __constant_htons(ETH_P_8021Q):
2637 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2639 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2640 proto = vlan->h_vlan_encapsulated_proto;
2641 nhoff += sizeof(*vlan);
2643 case __constant_htons(ETH_P_PPP_SES):
2644 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2646 proto = *((__be16 *) (skb->data + nhoff +
2647 sizeof(struct pppoe_hdr)));
2648 nhoff += PPPOE_SES_HLEN;
2650 case __constant_htons(PPP_IP):
2652 case __constant_htons(PPP_IPV6):
2663 if (pskb_may_pull(skb, nhoff + 16)) {
2664 u8 *h = skb->data + nhoff;
2665 __be16 flags = *(__be16 *)h;
2668 * Only look inside GRE if version zero and no
2671 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2672 proto = *(__be16 *)(h + 2);
2674 if (flags & GRE_CSUM)
2676 if (flags & GRE_KEY)
2678 if (flags & GRE_SEQ)
2691 poff = proto_ports_offset(ip_proto);
2694 if (pskb_may_pull(skb, nhoff + 4)) {
2695 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2700 /* get a consistent hash (same value on both flow directions) */
2701 if (addr2 < addr1 ||
2703 ports.v16[1] < ports.v16[0])) {
2705 swap(ports.v16[0], ports.v16[1]);
2707 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2714 EXPORT_SYMBOL(__skb_get_rxhash);
2718 /* One global table that all flow-based protocols share. */
2719 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2720 EXPORT_SYMBOL(rps_sock_flow_table);
2722 static struct rps_dev_flow *
2723 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2724 struct rps_dev_flow *rflow, u16 next_cpu)
2726 if (next_cpu != RPS_NO_CPU) {
2727 #ifdef CONFIG_RFS_ACCEL
2728 struct netdev_rx_queue *rxqueue;
2729 struct rps_dev_flow_table *flow_table;
2730 struct rps_dev_flow *old_rflow;
2735 /* Should we steer this flow to a different hardware queue? */
2736 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2737 !(dev->features & NETIF_F_NTUPLE))
2739 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2740 if (rxq_index == skb_get_rx_queue(skb))
2743 rxqueue = dev->_rx + rxq_index;
2744 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2747 flow_id = skb->rxhash & flow_table->mask;
2748 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2749 rxq_index, flow_id);
2753 rflow = &flow_table->flows[flow_id];
2755 if (old_rflow->filter == rflow->filter)
2756 old_rflow->filter = RPS_NO_FILTER;
2760 per_cpu(softnet_data, next_cpu).input_queue_head;
2763 rflow->cpu = next_cpu;
2768 * get_rps_cpu is called from netif_receive_skb and returns the target
2769 * CPU from the RPS map of the receiving queue for a given skb.
2770 * rcu_read_lock must be held on entry.
2772 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2773 struct rps_dev_flow **rflowp)
2775 struct netdev_rx_queue *rxqueue;
2776 struct rps_map *map;
2777 struct rps_dev_flow_table *flow_table;
2778 struct rps_sock_flow_table *sock_flow_table;
2782 if (skb_rx_queue_recorded(skb)) {
2783 u16 index = skb_get_rx_queue(skb);
2784 if (unlikely(index >= dev->real_num_rx_queues)) {
2785 WARN_ONCE(dev->real_num_rx_queues > 1,
2786 "%s received packet on queue %u, but number "
2787 "of RX queues is %u\n",
2788 dev->name, index, dev->real_num_rx_queues);
2791 rxqueue = dev->_rx + index;
2795 map = rcu_dereference(rxqueue->rps_map);
2797 if (map->len == 1 &&
2798 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2799 tcpu = map->cpus[0];
2800 if (cpu_online(tcpu))
2804 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2808 skb_reset_network_header(skb);
2809 if (!skb_get_rxhash(skb))
2812 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2813 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2814 if (flow_table && sock_flow_table) {
2816 struct rps_dev_flow *rflow;
2818 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2821 next_cpu = sock_flow_table->ents[skb->rxhash &
2822 sock_flow_table->mask];
2825 * If the desired CPU (where last recvmsg was done) is
2826 * different from current CPU (one in the rx-queue flow
2827 * table entry), switch if one of the following holds:
2828 * - Current CPU is unset (equal to RPS_NO_CPU).
2829 * - Current CPU is offline.
2830 * - The current CPU's queue tail has advanced beyond the
2831 * last packet that was enqueued using this table entry.
2832 * This guarantees that all previous packets for the flow
2833 * have been dequeued, thus preserving in order delivery.
2835 if (unlikely(tcpu != next_cpu) &&
2836 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2837 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2838 rflow->last_qtail)) >= 0)) {
2840 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2843 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2851 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2853 if (cpu_online(tcpu)) {
2863 #ifdef CONFIG_RFS_ACCEL
2866 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2867 * @dev: Device on which the filter was set
2868 * @rxq_index: RX queue index
2869 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2870 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2872 * Drivers that implement ndo_rx_flow_steer() should periodically call
2873 * this function for each installed filter and remove the filters for
2874 * which it returns %true.
2876 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2877 u32 flow_id, u16 filter_id)
2879 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2880 struct rps_dev_flow_table *flow_table;
2881 struct rps_dev_flow *rflow;
2886 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2887 if (flow_table && flow_id <= flow_table->mask) {
2888 rflow = &flow_table->flows[flow_id];
2889 cpu = ACCESS_ONCE(rflow->cpu);
2890 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2891 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2892 rflow->last_qtail) <
2893 (int)(10 * flow_table->mask)))
2899 EXPORT_SYMBOL(rps_may_expire_flow);
2901 #endif /* CONFIG_RFS_ACCEL */
2903 /* Called from hardirq (IPI) context */
2904 static void rps_trigger_softirq(void *data)
2906 struct softnet_data *sd = data;
2908 ____napi_schedule(sd, &sd->backlog);
2912 #endif /* CONFIG_RPS */
2915 * Check if this softnet_data structure is another cpu one
2916 * If yes, queue it to our IPI list and return 1
2919 static int rps_ipi_queued(struct softnet_data *sd)
2922 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2925 sd->rps_ipi_next = mysd->rps_ipi_list;
2926 mysd->rps_ipi_list = sd;
2928 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2931 #endif /* CONFIG_RPS */
2936 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2937 * queue (may be a remote CPU queue).
2939 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2940 unsigned int *qtail)
2942 struct softnet_data *sd;
2943 unsigned long flags;
2945 sd = &per_cpu(softnet_data, cpu);
2947 local_irq_save(flags);
2950 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2951 if (skb_queue_len(&sd->input_pkt_queue)) {
2953 __skb_queue_tail(&sd->input_pkt_queue, skb);
2954 input_queue_tail_incr_save(sd, qtail);
2956 local_irq_restore(flags);
2957 return NET_RX_SUCCESS;
2960 /* Schedule NAPI for backlog device
2961 * We can use non atomic operation since we own the queue lock
2963 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2964 if (!rps_ipi_queued(sd))
2965 ____napi_schedule(sd, &sd->backlog);
2973 local_irq_restore(flags);
2975 atomic_long_inc(&skb->dev->rx_dropped);
2981 * netif_rx - post buffer to the network code
2982 * @skb: buffer to post
2984 * This function receives a packet from a device driver and queues it for
2985 * the upper (protocol) levels to process. It always succeeds. The buffer
2986 * may be dropped during processing for congestion control or by the
2990 * NET_RX_SUCCESS (no congestion)
2991 * NET_RX_DROP (packet was dropped)
2995 int netif_rx(struct sk_buff *skb)
2999 /* if netpoll wants it, pretend we never saw it */
3000 if (netpoll_rx(skb))
3003 if (netdev_tstamp_prequeue)
3004 net_timestamp_check(skb);
3006 trace_netif_rx(skb);
3009 struct rps_dev_flow voidflow, *rflow = &voidflow;
3015 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3017 cpu = smp_processor_id();
3019 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3027 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3033 EXPORT_SYMBOL(netif_rx);
3035 int netif_rx_ni(struct sk_buff *skb)
3040 err = netif_rx(skb);
3041 if (local_softirq_pending())
3047 EXPORT_SYMBOL(netif_rx_ni);
3049 static void net_tx_action(struct softirq_action *h)
3051 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3053 if (sd->completion_queue) {
3054 struct sk_buff *clist;
3056 local_irq_disable();
3057 clist = sd->completion_queue;
3058 sd->completion_queue = NULL;
3062 struct sk_buff *skb = clist;
3063 clist = clist->next;
3065 WARN_ON(atomic_read(&skb->users));
3066 trace_kfree_skb(skb, net_tx_action);
3071 if (sd->output_queue) {
3074 local_irq_disable();
3075 head = sd->output_queue;
3076 sd->output_queue = NULL;
3077 sd->output_queue_tailp = &sd->output_queue;
3081 struct Qdisc *q = head;
3082 spinlock_t *root_lock;
3084 head = head->next_sched;
3086 root_lock = qdisc_lock(q);
3087 if (spin_trylock(root_lock)) {
3088 smp_mb__before_clear_bit();
3089 clear_bit(__QDISC_STATE_SCHED,
3092 spin_unlock(root_lock);
3094 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3096 __netif_reschedule(q);
3098 smp_mb__before_clear_bit();
3099 clear_bit(__QDISC_STATE_SCHED,
3107 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3108 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3109 /* This hook is defined here for ATM LANE */
3110 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3111 unsigned char *addr) __read_mostly;
3112 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3115 #ifdef CONFIG_NET_CLS_ACT
3116 /* TODO: Maybe we should just force sch_ingress to be compiled in
3117 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3118 * a compare and 2 stores extra right now if we dont have it on
3119 * but have CONFIG_NET_CLS_ACT
3120 * NOTE: This doesn't stop any functionality; if you dont have
3121 * the ingress scheduler, you just can't add policies on ingress.
3124 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3126 struct net_device *dev = skb->dev;
3127 u32 ttl = G_TC_RTTL(skb->tc_verd);
3128 int result = TC_ACT_OK;
3131 if (unlikely(MAX_RED_LOOP < ttl++)) {
3132 if (net_ratelimit())
3133 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3134 skb->skb_iif, dev->ifindex);
3138 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3139 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3142 if (q != &noop_qdisc) {
3143 spin_lock(qdisc_lock(q));
3144 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3145 result = qdisc_enqueue_root(skb, q);
3146 spin_unlock(qdisc_lock(q));
3152 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3153 struct packet_type **pt_prev,
3154 int *ret, struct net_device *orig_dev)
3156 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3158 if (!rxq || rxq->qdisc == &noop_qdisc)
3162 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3166 switch (ing_filter(skb, rxq)) {
3180 * netdev_rx_handler_register - register receive handler
3181 * @dev: device to register a handler for
3182 * @rx_handler: receive handler to register
3183 * @rx_handler_data: data pointer that is used by rx handler
3185 * Register a receive hander for a device. This handler will then be
3186 * called from __netif_receive_skb. A negative errno code is returned
3189 * The caller must hold the rtnl_mutex.
3191 * For a general description of rx_handler, see enum rx_handler_result.
3193 int netdev_rx_handler_register(struct net_device *dev,
3194 rx_handler_func_t *rx_handler,
3195 void *rx_handler_data)
3199 if (dev->rx_handler)
3202 /* Note: rx_handler_data must be set before rx_handler */
3203 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3204 rcu_assign_pointer(dev->rx_handler, rx_handler);
3208 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3211 * netdev_rx_handler_unregister - unregister receive handler
3212 * @dev: device to unregister a handler from
3214 * Unregister a receive hander from a device.
3216 * The caller must hold the rtnl_mutex.
3218 void netdev_rx_handler_unregister(struct net_device *dev)
3222 RCU_INIT_POINTER(dev->rx_handler, NULL);
3223 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3224 * section has a guarantee to see a non NULL rx_handler_data
3228 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3230 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3232 static int __netif_receive_skb(struct sk_buff *skb)
3234 struct packet_type *ptype, *pt_prev;
3235 rx_handler_func_t *rx_handler;
3236 struct net_device *orig_dev;
3237 struct net_device *null_or_dev;
3238 bool deliver_exact = false;
3239 int ret = NET_RX_DROP;
3242 if (!netdev_tstamp_prequeue)
3243 net_timestamp_check(skb);
3245 trace_netif_receive_skb(skb);
3247 /* if we've gotten here through NAPI, check netpoll */
3248 if (netpoll_receive_skb(skb))
3252 skb->skb_iif = skb->dev->ifindex;
3253 orig_dev = skb->dev;
3255 skb_reset_network_header(skb);
3256 skb_reset_transport_header(skb);
3257 skb_reset_mac_len(skb);
3265 __this_cpu_inc(softnet_data.processed);
3267 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3268 skb = vlan_untag(skb);
3273 #ifdef CONFIG_NET_CLS_ACT
3274 if (skb->tc_verd & TC_NCLS) {
3275 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3280 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3281 if (!ptype->dev || ptype->dev == skb->dev) {
3283 ret = deliver_skb(skb, pt_prev, orig_dev);
3288 #ifdef CONFIG_NET_CLS_ACT
3289 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3295 if (vlan_tx_tag_present(skb)) {
3297 ret = deliver_skb(skb, pt_prev, orig_dev);
3300 if (vlan_do_receive(&skb))
3302 else if (unlikely(!skb))
3306 rx_handler = rcu_dereference(skb->dev->rx_handler);
3309 ret = deliver_skb(skb, pt_prev, orig_dev);
3312 switch (rx_handler(&skb)) {
3313 case RX_HANDLER_CONSUMED:
3314 ret = NET_RX_SUCCESS;
3316 case RX_HANDLER_ANOTHER:
3318 case RX_HANDLER_EXACT:
3319 deliver_exact = true;
3320 case RX_HANDLER_PASS:
3327 if (vlan_tx_nonzero_tag_present(skb))
3328 skb->pkt_type = PACKET_OTHERHOST;
3330 /* deliver only exact match when indicated */
3331 null_or_dev = deliver_exact ? skb->dev : NULL;
3333 type = skb->protocol;
3334 list_for_each_entry_rcu(ptype,
3335 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3336 if (ptype->type == type &&
3337 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3338 ptype->dev == orig_dev)) {
3340 ret = deliver_skb(skb, pt_prev, orig_dev);
3346 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3348 atomic_long_inc(&skb->dev->rx_dropped);
3350 /* Jamal, now you will not able to escape explaining
3351 * me how you were going to use this. :-)
3362 * netif_receive_skb - process receive buffer from network
3363 * @skb: buffer to process
3365 * netif_receive_skb() is the main receive data processing function.
3366 * It always succeeds. The buffer may be dropped during processing
3367 * for congestion control or by the protocol layers.
3369 * This function may only be called from softirq context and interrupts
3370 * should be enabled.
3372 * Return values (usually ignored):
3373 * NET_RX_SUCCESS: no congestion
3374 * NET_RX_DROP: packet was dropped
3376 int netif_receive_skb(struct sk_buff *skb)
3378 if (netdev_tstamp_prequeue)
3379 net_timestamp_check(skb);
3381 if (skb_defer_rx_timestamp(skb))
3382 return NET_RX_SUCCESS;
3386 struct rps_dev_flow voidflow, *rflow = &voidflow;
3391 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3394 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3398 ret = __netif_receive_skb(skb);
3404 return __netif_receive_skb(skb);
3407 EXPORT_SYMBOL(netif_receive_skb);
3409 /* Network device is going away, flush any packets still pending
3410 * Called with irqs disabled.
3412 static void flush_backlog(void *arg)
3414 struct net_device *dev = arg;
3415 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3416 struct sk_buff *skb, *tmp;
3419 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3420 if (skb->dev == dev) {
3421 __skb_unlink(skb, &sd->input_pkt_queue);
3423 input_queue_head_incr(sd);
3428 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3429 if (skb->dev == dev) {
3430 __skb_unlink(skb, &sd->process_queue);
3432 input_queue_head_incr(sd);
3437 static int napi_gro_complete(struct sk_buff *skb)
3439 struct packet_type *ptype;
3440 __be16 type = skb->protocol;
3441 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3444 if (NAPI_GRO_CB(skb)->count == 1) {
3445 skb_shinfo(skb)->gso_size = 0;
3450 list_for_each_entry_rcu(ptype, head, list) {
3451 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3454 err = ptype->gro_complete(skb);
3460 WARN_ON(&ptype->list == head);
3462 return NET_RX_SUCCESS;
3466 return netif_receive_skb(skb);
3469 inline void napi_gro_flush(struct napi_struct *napi)
3471 struct sk_buff *skb, *next;
3473 for (skb = napi->gro_list; skb; skb = next) {
3476 napi_gro_complete(skb);
3479 napi->gro_count = 0;
3480 napi->gro_list = NULL;
3482 EXPORT_SYMBOL(napi_gro_flush);
3484 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3486 struct sk_buff **pp = NULL;
3487 struct packet_type *ptype;
3488 __be16 type = skb->protocol;
3489 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3492 enum gro_result ret;
3494 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3497 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3501 list_for_each_entry_rcu(ptype, head, list) {
3502 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3505 skb_set_network_header(skb, skb_gro_offset(skb));
3506 mac_len = skb->network_header - skb->mac_header;
3507 skb->mac_len = mac_len;
3508 NAPI_GRO_CB(skb)->same_flow = 0;
3509 NAPI_GRO_CB(skb)->flush = 0;
3510 NAPI_GRO_CB(skb)->free = 0;
3512 pp = ptype->gro_receive(&napi->gro_list, skb);
3517 if (&ptype->list == head)
3520 same_flow = NAPI_GRO_CB(skb)->same_flow;
3521 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3524 struct sk_buff *nskb = *pp;
3528 napi_gro_complete(nskb);
3535 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3539 NAPI_GRO_CB(skb)->count = 1;
3540 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3541 skb->next = napi->gro_list;
3542 napi->gro_list = skb;
3546 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3547 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3549 BUG_ON(skb->end - skb->tail < grow);
3551 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3554 skb->data_len -= grow;
3556 skb_shinfo(skb)->frags[0].page_offset += grow;
3557 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3559 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3560 skb_frag_unref(skb, 0);
3561 memmove(skb_shinfo(skb)->frags,
3562 skb_shinfo(skb)->frags + 1,
3563 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3574 EXPORT_SYMBOL(dev_gro_receive);
3576 static inline gro_result_t
3577 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3580 unsigned int maclen = skb->dev->hard_header_len;
3582 for (p = napi->gro_list; p; p = p->next) {
3583 unsigned long diffs;
3585 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3586 diffs |= p->vlan_tci ^ skb->vlan_tci;
3587 if (maclen == ETH_HLEN)
3588 diffs |= compare_ether_header(skb_mac_header(p),
3589 skb_gro_mac_header(skb));
3591 diffs = memcmp(skb_mac_header(p),
3592 skb_gro_mac_header(skb),
3594 NAPI_GRO_CB(p)->same_flow = !diffs;
3595 NAPI_GRO_CB(p)->flush = 0;
3598 return dev_gro_receive(napi, skb);
3601 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3605 if (netif_receive_skb(skb))
3610 case GRO_MERGED_FREE:
3621 EXPORT_SYMBOL(napi_skb_finish);
3623 void skb_gro_reset_offset(struct sk_buff *skb)
3625 NAPI_GRO_CB(skb)->data_offset = 0;
3626 NAPI_GRO_CB(skb)->frag0 = NULL;
3627 NAPI_GRO_CB(skb)->frag0_len = 0;
3629 if (skb->mac_header == skb->tail &&
3630 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3631 NAPI_GRO_CB(skb)->frag0 =
3632 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3633 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3636 EXPORT_SYMBOL(skb_gro_reset_offset);
3638 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3640 skb_gro_reset_offset(skb);
3642 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3644 EXPORT_SYMBOL(napi_gro_receive);
3646 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3648 __skb_pull(skb, skb_headlen(skb));
3649 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3650 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3652 skb->dev = napi->dev;
3654 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
3659 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3661 struct sk_buff *skb = napi->skb;
3664 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3670 EXPORT_SYMBOL(napi_get_frags);
3672 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3678 skb->protocol = eth_type_trans(skb, skb->dev);
3680 if (ret == GRO_HELD)
3681 skb_gro_pull(skb, -ETH_HLEN);
3682 else if (netif_receive_skb(skb))
3687 case GRO_MERGED_FREE:
3688 napi_reuse_skb(napi, skb);
3697 EXPORT_SYMBOL(napi_frags_finish);
3699 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3701 struct sk_buff *skb = napi->skb;
3708 skb_reset_mac_header(skb);
3709 skb_gro_reset_offset(skb);
3711 off = skb_gro_offset(skb);
3712 hlen = off + sizeof(*eth);
3713 eth = skb_gro_header_fast(skb, off);
3714 if (skb_gro_header_hard(skb, hlen)) {
3715 eth = skb_gro_header_slow(skb, hlen, off);
3716 if (unlikely(!eth)) {
3717 napi_reuse_skb(napi, skb);
3723 skb_gro_pull(skb, sizeof(*eth));
3726 * This works because the only protocols we care about don't require
3727 * special handling. We'll fix it up properly at the end.
3729 skb->protocol = eth->h_proto;
3734 EXPORT_SYMBOL(napi_frags_skb);
3736 gro_result_t napi_gro_frags(struct napi_struct *napi)
3738 struct sk_buff *skb = napi_frags_skb(napi);
3743 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3745 EXPORT_SYMBOL(napi_gro_frags);
3748 * net_rps_action sends any pending IPI's for rps.
3749 * Note: called with local irq disabled, but exits with local irq enabled.
3751 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3754 struct softnet_data *remsd = sd->rps_ipi_list;
3757 sd->rps_ipi_list = NULL;
3761 /* Send pending IPI's to kick RPS processing on remote cpus. */
3763 struct softnet_data *next = remsd->rps_ipi_next;
3765 if (cpu_online(remsd->cpu))
3766 __smp_call_function_single(remsd->cpu,
3775 static int process_backlog(struct napi_struct *napi, int quota)
3778 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3781 /* Check if we have pending ipi, its better to send them now,
3782 * not waiting net_rx_action() end.
3784 if (sd->rps_ipi_list) {
3785 local_irq_disable();
3786 net_rps_action_and_irq_enable(sd);
3789 napi->weight = weight_p;
3790 local_irq_disable();
3791 while (work < quota) {
3792 struct sk_buff *skb;
3795 while ((skb = __skb_dequeue(&sd->process_queue))) {
3797 __netif_receive_skb(skb);
3798 local_irq_disable();
3799 input_queue_head_incr(sd);
3800 if (++work >= quota) {
3807 qlen = skb_queue_len(&sd->input_pkt_queue);
3809 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3810 &sd->process_queue);
3812 if (qlen < quota - work) {
3814 * Inline a custom version of __napi_complete().
3815 * only current cpu owns and manipulates this napi,
3816 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3817 * we can use a plain write instead of clear_bit(),
3818 * and we dont need an smp_mb() memory barrier.
3820 list_del(&napi->poll_list);
3823 quota = work + qlen;
3833 * __napi_schedule - schedule for receive
3834 * @n: entry to schedule
3836 * The entry's receive function will be scheduled to run
3838 void __napi_schedule(struct napi_struct *n)
3840 unsigned long flags;
3842 local_irq_save(flags);
3843 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3844 local_irq_restore(flags);
3846 EXPORT_SYMBOL(__napi_schedule);
3848 void __napi_complete(struct napi_struct *n)
3850 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3851 BUG_ON(n->gro_list);
3853 list_del(&n->poll_list);
3854 smp_mb__before_clear_bit();
3855 clear_bit(NAPI_STATE_SCHED, &n->state);
3857 EXPORT_SYMBOL(__napi_complete);
3859 void napi_complete(struct napi_struct *n)
3861 unsigned long flags;
3864 * don't let napi dequeue from the cpu poll list
3865 * just in case its running on a different cpu
3867 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3871 local_irq_save(flags);
3873 local_irq_restore(flags);
3875 EXPORT_SYMBOL(napi_complete);
3877 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3878 int (*poll)(struct napi_struct *, int), int weight)
3880 INIT_LIST_HEAD(&napi->poll_list);
3881 napi->gro_count = 0;
3882 napi->gro_list = NULL;
3885 napi->weight = weight;
3886 list_add(&napi->dev_list, &dev->napi_list);
3888 #ifdef CONFIG_NETPOLL
3889 spin_lock_init(&napi->poll_lock);
3890 napi->poll_owner = -1;
3892 set_bit(NAPI_STATE_SCHED, &napi->state);
3894 EXPORT_SYMBOL(netif_napi_add);
3896 void netif_napi_del(struct napi_struct *napi)
3898 struct sk_buff *skb, *next;
3900 list_del_init(&napi->dev_list);
3901 napi_free_frags(napi);
3903 for (skb = napi->gro_list; skb; skb = next) {
3909 napi->gro_list = NULL;
3910 napi->gro_count = 0;
3912 EXPORT_SYMBOL(netif_napi_del);
3914 static void net_rx_action(struct softirq_action *h)
3916 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3917 unsigned long time_limit = jiffies + 2;
3918 int budget = netdev_budget;
3921 local_irq_disable();
3923 while (!list_empty(&sd->poll_list)) {
3924 struct napi_struct *n;
3927 /* If softirq window is exhuasted then punt.
3928 * Allow this to run for 2 jiffies since which will allow
3929 * an average latency of 1.5/HZ.
3931 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3936 /* Even though interrupts have been re-enabled, this
3937 * access is safe because interrupts can only add new
3938 * entries to the tail of this list, and only ->poll()
3939 * calls can remove this head entry from the list.
3941 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3943 have = netpoll_poll_lock(n);
3947 /* This NAPI_STATE_SCHED test is for avoiding a race
3948 * with netpoll's poll_napi(). Only the entity which
3949 * obtains the lock and sees NAPI_STATE_SCHED set will
3950 * actually make the ->poll() call. Therefore we avoid
3951 * accidentally calling ->poll() when NAPI is not scheduled.
3954 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3955 work = n->poll(n, weight);
3959 WARN_ON_ONCE(work > weight);
3963 local_irq_disable();
3965 /* Drivers must not modify the NAPI state if they
3966 * consume the entire weight. In such cases this code
3967 * still "owns" the NAPI instance and therefore can
3968 * move the instance around on the list at-will.
3970 if (unlikely(work == weight)) {
3971 if (unlikely(napi_disable_pending(n))) {
3974 local_irq_disable();
3976 list_move_tail(&n->poll_list, &sd->poll_list);
3979 netpoll_poll_unlock(have);
3982 net_rps_action_and_irq_enable(sd);
3984 #ifdef CONFIG_NET_DMA
3986 * There may not be any more sk_buffs coming right now, so push
3987 * any pending DMA copies to hardware
3989 dma_issue_pending_all();
3996 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4000 static gifconf_func_t *gifconf_list[NPROTO];
4003 * register_gifconf - register a SIOCGIF handler
4004 * @family: Address family
4005 * @gifconf: Function handler
4007 * Register protocol dependent address dumping routines. The handler
4008 * that is passed must not be freed or reused until it has been replaced
4009 * by another handler.
4011 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
4013 if (family >= NPROTO)
4015 gifconf_list[family] = gifconf;
4018 EXPORT_SYMBOL(register_gifconf);
4022 * Map an interface index to its name (SIOCGIFNAME)
4026 * We need this ioctl for efficient implementation of the
4027 * if_indextoname() function required by the IPv6 API. Without
4028 * it, we would have to search all the interfaces to find a
4032 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4034 struct net_device *dev;
4038 * Fetch the caller's info block.
4041 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4045 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4051 strcpy(ifr.ifr_name, dev->name);
4054 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4060 * Perform a SIOCGIFCONF call. This structure will change
4061 * size eventually, and there is nothing I can do about it.
4062 * Thus we will need a 'compatibility mode'.
4065 static int dev_ifconf(struct net *net, char __user *arg)
4068 struct net_device *dev;
4075 * Fetch the caller's info block.
4078 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4085 * Loop over the interfaces, and write an info block for each.
4089 for_each_netdev(net, dev) {
4090 for (i = 0; i < NPROTO; i++) {
4091 if (gifconf_list[i]) {
4094 done = gifconf_list[i](dev, NULL, 0);
4096 done = gifconf_list[i](dev, pos + total,
4106 * All done. Write the updated control block back to the caller.
4108 ifc.ifc_len = total;
4111 * Both BSD and Solaris return 0 here, so we do too.
4113 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4116 #ifdef CONFIG_PROC_FS
4118 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4120 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4121 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4122 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4124 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4126 struct net *net = seq_file_net(seq);
4127 struct net_device *dev;
4128 struct hlist_node *p;
4129 struct hlist_head *h;
4130 unsigned int count = 0, offset = get_offset(*pos);
4132 h = &net->dev_name_head[get_bucket(*pos)];
4133 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4134 if (++count == offset)
4141 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4143 struct net_device *dev;
4144 unsigned int bucket;
4147 dev = dev_from_same_bucket(seq, pos);
4151 bucket = get_bucket(*pos) + 1;
4152 *pos = set_bucket_offset(bucket, 1);
4153 } while (bucket < NETDEV_HASHENTRIES);
4159 * This is invoked by the /proc filesystem handler to display a device
4162 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4167 return SEQ_START_TOKEN;
4169 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4172 return dev_from_bucket(seq, pos);
4175 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4178 return dev_from_bucket(seq, pos);
4181 void dev_seq_stop(struct seq_file *seq, void *v)
4187 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4189 struct rtnl_link_stats64 temp;
4190 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4192 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4193 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4194 dev->name, stats->rx_bytes, stats->rx_packets,
4196 stats->rx_dropped + stats->rx_missed_errors,
4197 stats->rx_fifo_errors,
4198 stats->rx_length_errors + stats->rx_over_errors +
4199 stats->rx_crc_errors + stats->rx_frame_errors,
4200 stats->rx_compressed, stats->multicast,
4201 stats->tx_bytes, stats->tx_packets,
4202 stats->tx_errors, stats->tx_dropped,
4203 stats->tx_fifo_errors, stats->collisions,
4204 stats->tx_carrier_errors +
4205 stats->tx_aborted_errors +
4206 stats->tx_window_errors +
4207 stats->tx_heartbeat_errors,
4208 stats->tx_compressed);
4212 * Called from the PROCfs module. This now uses the new arbitrary sized
4213 * /proc/net interface to create /proc/net/dev
4215 static int dev_seq_show(struct seq_file *seq, void *v)
4217 if (v == SEQ_START_TOKEN)
4218 seq_puts(seq, "Inter-| Receive "
4220 " face |bytes packets errs drop fifo frame "
4221 "compressed multicast|bytes packets errs "
4222 "drop fifo colls carrier compressed\n");
4224 dev_seq_printf_stats(seq, v);
4228 static struct softnet_data *softnet_get_online(loff_t *pos)
4230 struct softnet_data *sd = NULL;
4232 while (*pos < nr_cpu_ids)
4233 if (cpu_online(*pos)) {
4234 sd = &per_cpu(softnet_data, *pos);
4241 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4243 return softnet_get_online(pos);
4246 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4249 return softnet_get_online(pos);
4252 static void softnet_seq_stop(struct seq_file *seq, void *v)
4256 static int softnet_seq_show(struct seq_file *seq, void *v)
4258 struct softnet_data *sd = v;
4260 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4261 sd->processed, sd->dropped, sd->time_squeeze, 0,
4262 0, 0, 0, 0, /* was fastroute */
4263 sd->cpu_collision, sd->received_rps);
4267 static const struct seq_operations dev_seq_ops = {
4268 .start = dev_seq_start,
4269 .next = dev_seq_next,
4270 .stop = dev_seq_stop,
4271 .show = dev_seq_show,
4274 static int dev_seq_open(struct inode *inode, struct file *file)
4276 return seq_open_net(inode, file, &dev_seq_ops,
4277 sizeof(struct seq_net_private));
4280 static const struct file_operations dev_seq_fops = {
4281 .owner = THIS_MODULE,
4282 .open = dev_seq_open,
4284 .llseek = seq_lseek,
4285 .release = seq_release_net,
4288 static const struct seq_operations softnet_seq_ops = {
4289 .start = softnet_seq_start,
4290 .next = softnet_seq_next,
4291 .stop = softnet_seq_stop,
4292 .show = softnet_seq_show,
4295 static int softnet_seq_open(struct inode *inode, struct file *file)
4297 return seq_open(file, &softnet_seq_ops);
4300 static const struct file_operations softnet_seq_fops = {
4301 .owner = THIS_MODULE,
4302 .open = softnet_seq_open,
4304 .llseek = seq_lseek,
4305 .release = seq_release,
4308 static void *ptype_get_idx(loff_t pos)
4310 struct packet_type *pt = NULL;
4314 list_for_each_entry_rcu(pt, &ptype_all, list) {
4320 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4321 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4330 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4334 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4337 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4339 struct packet_type *pt;
4340 struct list_head *nxt;
4344 if (v == SEQ_START_TOKEN)
4345 return ptype_get_idx(0);
4348 nxt = pt->list.next;
4349 if (pt->type == htons(ETH_P_ALL)) {
4350 if (nxt != &ptype_all)
4353 nxt = ptype_base[0].next;
4355 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4357 while (nxt == &ptype_base[hash]) {
4358 if (++hash >= PTYPE_HASH_SIZE)
4360 nxt = ptype_base[hash].next;
4363 return list_entry(nxt, struct packet_type, list);
4366 static void ptype_seq_stop(struct seq_file *seq, void *v)
4372 static int ptype_seq_show(struct seq_file *seq, void *v)
4374 struct packet_type *pt = v;
4376 if (v == SEQ_START_TOKEN)
4377 seq_puts(seq, "Type Device Function\n");
4378 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4379 if (pt->type == htons(ETH_P_ALL))
4380 seq_puts(seq, "ALL ");
4382 seq_printf(seq, "%04x", ntohs(pt->type));
4384 seq_printf(seq, " %-8s %pF\n",
4385 pt->dev ? pt->dev->name : "", pt->func);
4391 static const struct seq_operations ptype_seq_ops = {
4392 .start = ptype_seq_start,
4393 .next = ptype_seq_next,
4394 .stop = ptype_seq_stop,
4395 .show = ptype_seq_show,
4398 static int ptype_seq_open(struct inode *inode, struct file *file)
4400 return seq_open_net(inode, file, &ptype_seq_ops,
4401 sizeof(struct seq_net_private));
4404 static const struct file_operations ptype_seq_fops = {
4405 .owner = THIS_MODULE,
4406 .open = ptype_seq_open,
4408 .llseek = seq_lseek,
4409 .release = seq_release_net,
4413 static int __net_init dev_proc_net_init(struct net *net)
4417 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4419 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4421 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4424 if (wext_proc_init(net))
4430 proc_net_remove(net, "ptype");
4432 proc_net_remove(net, "softnet_stat");
4434 proc_net_remove(net, "dev");
4438 static void __net_exit dev_proc_net_exit(struct net *net)
4440 wext_proc_exit(net);
4442 proc_net_remove(net, "ptype");
4443 proc_net_remove(net, "softnet_stat");
4444 proc_net_remove(net, "dev");
4447 static struct pernet_operations __net_initdata dev_proc_ops = {
4448 .init = dev_proc_net_init,
4449 .exit = dev_proc_net_exit,
4452 static int __init dev_proc_init(void)
4454 return register_pernet_subsys(&dev_proc_ops);
4457 #define dev_proc_init() 0
4458 #endif /* CONFIG_PROC_FS */
4462 * netdev_set_master - set up master pointer
4463 * @slave: slave device
4464 * @master: new master device
4466 * Changes the master device of the slave. Pass %NULL to break the
4467 * bonding. The caller must hold the RTNL semaphore. On a failure
4468 * a negative errno code is returned. On success the reference counts
4469 * are adjusted and the function returns zero.
4471 int netdev_set_master(struct net_device *slave, struct net_device *master)
4473 struct net_device *old = slave->master;
4483 slave->master = master;
4489 EXPORT_SYMBOL(netdev_set_master);
4492 * netdev_set_bond_master - set up bonding master/slave pair
4493 * @slave: slave device
4494 * @master: new master device
4496 * Changes the master device of the slave. Pass %NULL to break the
4497 * bonding. The caller must hold the RTNL semaphore. On a failure
4498 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4499 * to the routing socket and the function returns zero.
4501 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4507 err = netdev_set_master(slave, master);
4511 slave->flags |= IFF_SLAVE;
4513 slave->flags &= ~IFF_SLAVE;
4515 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4518 EXPORT_SYMBOL(netdev_set_bond_master);
4520 static void dev_change_rx_flags(struct net_device *dev, int flags)
4522 const struct net_device_ops *ops = dev->netdev_ops;
4524 if (ops->ndo_change_rx_flags)
4525 ops->ndo_change_rx_flags(dev, flags);
4528 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4530 unsigned short old_flags = dev->flags;
4536 dev->flags |= IFF_PROMISC;
4537 dev->promiscuity += inc;
4538 if (dev->promiscuity == 0) {
4541 * If inc causes overflow, untouch promisc and return error.
4544 dev->flags &= ~IFF_PROMISC;
4546 dev->promiscuity -= inc;
4547 printk(KERN_WARNING "%s: promiscuity touches roof, "
4548 "set promiscuity failed, promiscuity feature "
4549 "of device might be broken.\n", dev->name);
4553 if (dev->flags != old_flags) {
4554 printk(KERN_INFO "device %s %s promiscuous mode\n",
4555 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4557 if (audit_enabled) {
4558 current_uid_gid(&uid, &gid);
4559 audit_log(current->audit_context, GFP_ATOMIC,
4560 AUDIT_ANOM_PROMISCUOUS,
4561 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4562 dev->name, (dev->flags & IFF_PROMISC),
4563 (old_flags & IFF_PROMISC),
4564 audit_get_loginuid(current),
4566 audit_get_sessionid(current));
4569 dev_change_rx_flags(dev, IFF_PROMISC);
4575 * dev_set_promiscuity - update promiscuity count on a device
4579 * Add or remove promiscuity from a device. While the count in the device
4580 * remains above zero the interface remains promiscuous. Once it hits zero
4581 * the device reverts back to normal filtering operation. A negative inc
4582 * value is used to drop promiscuity on the device.
4583 * Return 0 if successful or a negative errno code on error.
4585 int dev_set_promiscuity(struct net_device *dev, int inc)
4587 unsigned short old_flags = dev->flags;
4590 err = __dev_set_promiscuity(dev, inc);
4593 if (dev->flags != old_flags)
4594 dev_set_rx_mode(dev);
4597 EXPORT_SYMBOL(dev_set_promiscuity);
4600 * dev_set_allmulti - update allmulti count on a device
4604 * Add or remove reception of all multicast frames to a device. While the
4605 * count in the device remains above zero the interface remains listening
4606 * to all interfaces. Once it hits zero the device reverts back to normal
4607 * filtering operation. A negative @inc value is used to drop the counter
4608 * when releasing a resource needing all multicasts.
4609 * Return 0 if successful or a negative errno code on error.
4612 int dev_set_allmulti(struct net_device *dev, int inc)
4614 unsigned short old_flags = dev->flags;
4618 dev->flags |= IFF_ALLMULTI;
4619 dev->allmulti += inc;
4620 if (dev->allmulti == 0) {
4623 * If inc causes overflow, untouch allmulti and return error.
4626 dev->flags &= ~IFF_ALLMULTI;
4628 dev->allmulti -= inc;
4629 printk(KERN_WARNING "%s: allmulti touches roof, "
4630 "set allmulti failed, allmulti feature of "
4631 "device might be broken.\n", dev->name);
4635 if (dev->flags ^ old_flags) {
4636 dev_change_rx_flags(dev, IFF_ALLMULTI);
4637 dev_set_rx_mode(dev);
4641 EXPORT_SYMBOL(dev_set_allmulti);
4644 * Upload unicast and multicast address lists to device and
4645 * configure RX filtering. When the device doesn't support unicast
4646 * filtering it is put in promiscuous mode while unicast addresses
4649 void __dev_set_rx_mode(struct net_device *dev)
4651 const struct net_device_ops *ops = dev->netdev_ops;
4653 /* dev_open will call this function so the list will stay sane. */
4654 if (!(dev->flags&IFF_UP))
4657 if (!netif_device_present(dev))
4660 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4661 /* Unicast addresses changes may only happen under the rtnl,
4662 * therefore calling __dev_set_promiscuity here is safe.
4664 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4665 __dev_set_promiscuity(dev, 1);
4666 dev->uc_promisc = true;
4667 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4668 __dev_set_promiscuity(dev, -1);
4669 dev->uc_promisc = false;
4673 if (ops->ndo_set_rx_mode)
4674 ops->ndo_set_rx_mode(dev);
4677 void dev_set_rx_mode(struct net_device *dev)
4679 netif_addr_lock_bh(dev);
4680 __dev_set_rx_mode(dev);
4681 netif_addr_unlock_bh(dev);
4685 * dev_get_flags - get flags reported to userspace
4688 * Get the combination of flag bits exported through APIs to userspace.
4690 unsigned dev_get_flags(const struct net_device *dev)
4694 flags = (dev->flags & ~(IFF_PROMISC |
4699 (dev->gflags & (IFF_PROMISC |
4702 if (netif_running(dev)) {
4703 if (netif_oper_up(dev))
4704 flags |= IFF_RUNNING;
4705 if (netif_carrier_ok(dev))
4706 flags |= IFF_LOWER_UP;
4707 if (netif_dormant(dev))
4708 flags |= IFF_DORMANT;
4713 EXPORT_SYMBOL(dev_get_flags);
4715 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4717 int old_flags = dev->flags;
4723 * Set the flags on our device.
4726 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4727 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4729 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4733 * Load in the correct multicast list now the flags have changed.
4736 if ((old_flags ^ flags) & IFF_MULTICAST)
4737 dev_change_rx_flags(dev, IFF_MULTICAST);
4739 dev_set_rx_mode(dev);
4742 * Have we downed the interface. We handle IFF_UP ourselves
4743 * according to user attempts to set it, rather than blindly
4748 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4749 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4752 dev_set_rx_mode(dev);
4755 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4756 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4758 dev->gflags ^= IFF_PROMISC;
4759 dev_set_promiscuity(dev, inc);
4762 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4763 is important. Some (broken) drivers set IFF_PROMISC, when
4764 IFF_ALLMULTI is requested not asking us and not reporting.
4766 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4767 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4769 dev->gflags ^= IFF_ALLMULTI;
4770 dev_set_allmulti(dev, inc);
4776 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4778 unsigned int changes = dev->flags ^ old_flags;
4780 if (changes & IFF_UP) {
4781 if (dev->flags & IFF_UP)
4782 call_netdevice_notifiers(NETDEV_UP, dev);
4784 call_netdevice_notifiers(NETDEV_DOWN, dev);
4787 if (dev->flags & IFF_UP &&
4788 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4789 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4793 * dev_change_flags - change device settings
4795 * @flags: device state flags
4797 * Change settings on device based state flags. The flags are
4798 * in the userspace exported format.
4800 int dev_change_flags(struct net_device *dev, unsigned flags)
4803 int old_flags = dev->flags;
4805 ret = __dev_change_flags(dev, flags);
4809 changes = old_flags ^ dev->flags;
4811 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4813 __dev_notify_flags(dev, old_flags);
4816 EXPORT_SYMBOL(dev_change_flags);
4819 * dev_set_mtu - Change maximum transfer unit
4821 * @new_mtu: new transfer unit
4823 * Change the maximum transfer size of the network device.
4825 int dev_set_mtu(struct net_device *dev, int new_mtu)
4827 const struct net_device_ops *ops = dev->netdev_ops;
4830 if (new_mtu == dev->mtu)
4833 /* MTU must be positive. */
4837 if (!netif_device_present(dev))
4841 if (ops->ndo_change_mtu)
4842 err = ops->ndo_change_mtu(dev, new_mtu);
4846 if (!err && dev->flags & IFF_UP)
4847 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4850 EXPORT_SYMBOL(dev_set_mtu);
4853 * dev_set_group - Change group this device belongs to
4855 * @new_group: group this device should belong to
4857 void dev_set_group(struct net_device *dev, int new_group)
4859 dev->group = new_group;
4861 EXPORT_SYMBOL(dev_set_group);
4864 * dev_set_mac_address - Change Media Access Control Address
4868 * Change the hardware (MAC) address of the device
4870 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4872 const struct net_device_ops *ops = dev->netdev_ops;
4875 if (!ops->ndo_set_mac_address)
4877 if (sa->sa_family != dev->type)
4879 if (!netif_device_present(dev))
4881 err = ops->ndo_set_mac_address(dev, sa);
4883 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4884 add_device_randomness(dev->dev_addr, dev->addr_len);
4887 EXPORT_SYMBOL(dev_set_mac_address);
4890 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4892 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4895 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4901 case SIOCGIFFLAGS: /* Get interface flags */
4902 ifr->ifr_flags = (short) dev_get_flags(dev);
4905 case SIOCGIFMETRIC: /* Get the metric on the interface
4906 (currently unused) */
4907 ifr->ifr_metric = 0;
4910 case SIOCGIFMTU: /* Get the MTU of a device */
4911 ifr->ifr_mtu = dev->mtu;
4916 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4918 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4919 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4920 ifr->ifr_hwaddr.sa_family = dev->type;
4928 ifr->ifr_map.mem_start = dev->mem_start;
4929 ifr->ifr_map.mem_end = dev->mem_end;
4930 ifr->ifr_map.base_addr = dev->base_addr;
4931 ifr->ifr_map.irq = dev->irq;
4932 ifr->ifr_map.dma = dev->dma;
4933 ifr->ifr_map.port = dev->if_port;
4937 ifr->ifr_ifindex = dev->ifindex;
4941 ifr->ifr_qlen = dev->tx_queue_len;
4945 /* dev_ioctl() should ensure this case
4957 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4959 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4962 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4963 const struct net_device_ops *ops;
4968 ops = dev->netdev_ops;
4971 case SIOCSIFFLAGS: /* Set interface flags */
4972 return dev_change_flags(dev, ifr->ifr_flags);
4974 case SIOCSIFMETRIC: /* Set the metric on the interface
4975 (currently unused) */
4978 case SIOCSIFMTU: /* Set the MTU of a device */
4979 return dev_set_mtu(dev, ifr->ifr_mtu);
4982 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4984 case SIOCSIFHWBROADCAST:
4985 if (ifr->ifr_hwaddr.sa_family != dev->type)
4987 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4988 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4989 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4993 if (ops->ndo_set_config) {
4994 if (!netif_device_present(dev))
4996 return ops->ndo_set_config(dev, &ifr->ifr_map);
5001 if (!ops->ndo_set_rx_mode ||
5002 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5004 if (!netif_device_present(dev))
5006 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
5009 if (!ops->ndo_set_rx_mode ||
5010 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
5012 if (!netif_device_present(dev))
5014 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
5017 if (ifr->ifr_qlen < 0)
5019 dev->tx_queue_len = ifr->ifr_qlen;
5023 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
5024 return dev_change_name(dev, ifr->ifr_newname);
5027 err = net_hwtstamp_validate(ifr);
5033 * Unknown or private ioctl
5036 if ((cmd >= SIOCDEVPRIVATE &&
5037 cmd <= SIOCDEVPRIVATE + 15) ||
5038 cmd == SIOCBONDENSLAVE ||
5039 cmd == SIOCBONDRELEASE ||
5040 cmd == SIOCBONDSETHWADDR ||
5041 cmd == SIOCBONDSLAVEINFOQUERY ||
5042 cmd == SIOCBONDINFOQUERY ||
5043 cmd == SIOCBONDCHANGEACTIVE ||
5044 cmd == SIOCGMIIPHY ||
5045 cmd == SIOCGMIIREG ||
5046 cmd == SIOCSMIIREG ||
5047 cmd == SIOCBRADDIF ||
5048 cmd == SIOCBRDELIF ||
5049 cmd == SIOCSHWTSTAMP ||
5050 cmd == SIOCWANDEV) {
5052 if (ops->ndo_do_ioctl) {
5053 if (netif_device_present(dev))
5054 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5066 * This function handles all "interface"-type I/O control requests. The actual
5067 * 'doing' part of this is dev_ifsioc above.
5071 * dev_ioctl - network device ioctl
5072 * @net: the applicable net namespace
5073 * @cmd: command to issue
5074 * @arg: pointer to a struct ifreq in user space
5076 * Issue ioctl functions to devices. This is normally called by the
5077 * user space syscall interfaces but can sometimes be useful for
5078 * other purposes. The return value is the return from the syscall if
5079 * positive or a negative errno code on error.
5082 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5088 /* One special case: SIOCGIFCONF takes ifconf argument
5089 and requires shared lock, because it sleeps writing
5093 if (cmd == SIOCGIFCONF) {
5095 ret = dev_ifconf(net, (char __user *) arg);
5099 if (cmd == SIOCGIFNAME)
5100 return dev_ifname(net, (struct ifreq __user *)arg);
5102 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5105 ifr.ifr_name[IFNAMSIZ-1] = 0;
5107 colon = strchr(ifr.ifr_name, ':');
5112 * See which interface the caller is talking about.
5117 * These ioctl calls:
5118 * - can be done by all.
5119 * - atomic and do not require locking.
5130 dev_load(net, ifr.ifr_name);
5132 ret = dev_ifsioc_locked(net, &ifr, cmd);
5137 if (copy_to_user(arg, &ifr,
5138 sizeof(struct ifreq)))
5144 dev_load(net, ifr.ifr_name);
5146 ret = dev_ethtool(net, &ifr);
5151 if (copy_to_user(arg, &ifr,
5152 sizeof(struct ifreq)))
5158 * These ioctl calls:
5159 * - require superuser power.
5160 * - require strict serialization.
5166 if (!capable(CAP_NET_ADMIN))
5168 dev_load(net, ifr.ifr_name);
5170 ret = dev_ifsioc(net, &ifr, cmd);
5175 if (copy_to_user(arg, &ifr,
5176 sizeof(struct ifreq)))
5182 * These ioctl calls:
5183 * - require superuser power.
5184 * - require strict serialization.
5185 * - do not return a value
5195 case SIOCSIFHWBROADCAST:
5198 case SIOCBONDENSLAVE:
5199 case SIOCBONDRELEASE:
5200 case SIOCBONDSETHWADDR:
5201 case SIOCBONDCHANGEACTIVE:
5205 if (!capable(CAP_NET_ADMIN))
5208 case SIOCBONDSLAVEINFOQUERY:
5209 case SIOCBONDINFOQUERY:
5210 dev_load(net, ifr.ifr_name);
5212 ret = dev_ifsioc(net, &ifr, cmd);
5217 /* Get the per device memory space. We can add this but
5218 * currently do not support it */
5220 /* Set the per device memory buffer space.
5221 * Not applicable in our case */
5226 * Unknown or private ioctl.
5229 if (cmd == SIOCWANDEV ||
5230 (cmd >= SIOCDEVPRIVATE &&
5231 cmd <= SIOCDEVPRIVATE + 15)) {
5232 dev_load(net, ifr.ifr_name);
5234 ret = dev_ifsioc(net, &ifr, cmd);
5236 if (!ret && copy_to_user(arg, &ifr,
5237 sizeof(struct ifreq)))
5241 /* Take care of Wireless Extensions */
5242 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5243 return wext_handle_ioctl(net, &ifr, cmd, arg);
5250 * dev_new_index - allocate an ifindex
5251 * @net: the applicable net namespace
5253 * Returns a suitable unique value for a new device interface
5254 * number. The caller must hold the rtnl semaphore or the
5255 * dev_base_lock to be sure it remains unique.
5257 static int dev_new_index(struct net *net)
5263 if (!__dev_get_by_index(net, ifindex))
5268 /* Delayed registration/unregisteration */
5269 static LIST_HEAD(net_todo_list);
5271 static void net_set_todo(struct net_device *dev)
5273 list_add_tail(&dev->todo_list, &net_todo_list);
5276 static void rollback_registered_many(struct list_head *head)
5278 struct net_device *dev, *tmp;
5280 BUG_ON(dev_boot_phase);
5283 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5284 /* Some devices call without registering
5285 * for initialization unwind. Remove those
5286 * devices and proceed with the remaining.
5288 if (dev->reg_state == NETREG_UNINITIALIZED) {
5289 pr_debug("unregister_netdevice: device %s/%p never "
5290 "was registered\n", dev->name, dev);
5293 list_del(&dev->unreg_list);
5296 dev->dismantle = true;
5297 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5300 /* If device is running, close it first. */
5301 dev_close_many(head);
5303 list_for_each_entry(dev, head, unreg_list) {
5304 /* And unlink it from device chain. */
5305 unlist_netdevice(dev);
5307 dev->reg_state = NETREG_UNREGISTERING;
5312 list_for_each_entry(dev, head, unreg_list) {
5313 /* Shutdown queueing discipline. */
5317 /* Notify protocols, that we are about to destroy
5318 this device. They should clean all the things.
5320 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5322 if (!dev->rtnl_link_ops ||
5323 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5324 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5327 * Flush the unicast and multicast chains
5332 if (dev->netdev_ops->ndo_uninit)
5333 dev->netdev_ops->ndo_uninit(dev);
5335 /* Notifier chain MUST detach us from master device. */
5336 WARN_ON(dev->master);
5338 /* Remove entries from kobject tree */
5339 netdev_unregister_kobject(dev);
5342 /* Process any work delayed until the end of the batch */
5343 dev = list_first_entry(head, struct net_device, unreg_list);
5344 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5348 list_for_each_entry(dev, head, unreg_list)
5352 static void rollback_registered(struct net_device *dev)
5356 list_add(&dev->unreg_list, &single);
5357 rollback_registered_many(&single);
5361 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5363 /* Fix illegal checksum combinations */
5364 if ((features & NETIF_F_HW_CSUM) &&
5365 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5366 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5367 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5370 if ((features & NETIF_F_NO_CSUM) &&
5371 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5372 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5373 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5376 /* Fix illegal SG+CSUM combinations. */
5377 if ((features & NETIF_F_SG) &&
5378 !(features & NETIF_F_ALL_CSUM)) {
5380 "Dropping NETIF_F_SG since no checksum feature.\n");
5381 features &= ~NETIF_F_SG;
5384 /* TSO requires that SG is present as well. */
5385 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5386 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5387 features &= ~NETIF_F_ALL_TSO;
5390 /* TSO ECN requires that TSO is present as well. */
5391 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5392 features &= ~NETIF_F_TSO_ECN;
5394 /* Software GSO depends on SG. */
5395 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5396 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5397 features &= ~NETIF_F_GSO;
5400 /* UFO needs SG and checksumming */
5401 if (features & NETIF_F_UFO) {
5402 /* maybe split UFO into V4 and V6? */
5403 if (!((features & NETIF_F_GEN_CSUM) ||
5404 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5405 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5407 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5408 features &= ~NETIF_F_UFO;
5411 if (!(features & NETIF_F_SG)) {
5413 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5414 features &= ~NETIF_F_UFO;
5421 int __netdev_update_features(struct net_device *dev)
5428 features = netdev_get_wanted_features(dev);
5430 if (dev->netdev_ops->ndo_fix_features)
5431 features = dev->netdev_ops->ndo_fix_features(dev, features);
5433 /* driver might be less strict about feature dependencies */
5434 features = netdev_fix_features(dev, features);
5436 if (dev->features == features)
5439 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5440 dev->features, features);
5442 if (dev->netdev_ops->ndo_set_features)
5443 err = dev->netdev_ops->ndo_set_features(dev, features);
5445 if (unlikely(err < 0)) {
5447 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5448 err, features, dev->features);
5453 dev->features = features;
5459 * netdev_update_features - recalculate device features
5460 * @dev: the device to check
5462 * Recalculate dev->features set and send notifications if it
5463 * has changed. Should be called after driver or hardware dependent
5464 * conditions might have changed that influence the features.
5466 void netdev_update_features(struct net_device *dev)
5468 if (__netdev_update_features(dev))
5469 netdev_features_change(dev);
5471 EXPORT_SYMBOL(netdev_update_features);
5474 * netdev_change_features - recalculate device features
5475 * @dev: the device to check
5477 * Recalculate dev->features set and send notifications even
5478 * if they have not changed. Should be called instead of
5479 * netdev_update_features() if also dev->vlan_features might
5480 * have changed to allow the changes to be propagated to stacked
5483 void netdev_change_features(struct net_device *dev)
5485 __netdev_update_features(dev);
5486 netdev_features_change(dev);
5488 EXPORT_SYMBOL(netdev_change_features);
5491 * netif_stacked_transfer_operstate - transfer operstate
5492 * @rootdev: the root or lower level device to transfer state from
5493 * @dev: the device to transfer operstate to
5495 * Transfer operational state from root to device. This is normally
5496 * called when a stacking relationship exists between the root
5497 * device and the device(a leaf device).
5499 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5500 struct net_device *dev)
5502 if (rootdev->operstate == IF_OPER_DORMANT)
5503 netif_dormant_on(dev);
5505 netif_dormant_off(dev);
5507 if (netif_carrier_ok(rootdev)) {
5508 if (!netif_carrier_ok(dev))
5509 netif_carrier_on(dev);
5511 if (netif_carrier_ok(dev))
5512 netif_carrier_off(dev);
5515 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5518 static int netif_alloc_rx_queues(struct net_device *dev)
5520 unsigned int i, count = dev->num_rx_queues;
5521 struct netdev_rx_queue *rx;
5525 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5527 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5532 for (i = 0; i < count; i++)
5538 static void netdev_init_one_queue(struct net_device *dev,
5539 struct netdev_queue *queue, void *_unused)
5541 /* Initialize queue lock */
5542 spin_lock_init(&queue->_xmit_lock);
5543 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5544 queue->xmit_lock_owner = -1;
5545 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5549 static int netif_alloc_netdev_queues(struct net_device *dev)
5551 unsigned int count = dev->num_tx_queues;
5552 struct netdev_queue *tx;
5556 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5558 pr_err("netdev: Unable to allocate %u tx queues.\n",
5564 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5565 spin_lock_init(&dev->tx_global_lock);
5571 * register_netdevice - register a network device
5572 * @dev: device to register
5574 * Take a completed network device structure and add it to the kernel
5575 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5576 * chain. 0 is returned on success. A negative errno code is returned
5577 * on a failure to set up the device, or if the name is a duplicate.
5579 * Callers must hold the rtnl semaphore. You may want
5580 * register_netdev() instead of this.
5583 * The locking appears insufficient to guarantee two parallel registers
5584 * will not get the same name.
5587 int register_netdevice(struct net_device *dev)
5590 struct net *net = dev_net(dev);
5592 BUG_ON(dev_boot_phase);
5597 /* When net_device's are persistent, this will be fatal. */
5598 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5601 spin_lock_init(&dev->addr_list_lock);
5602 netdev_set_addr_lockdep_class(dev);
5606 ret = dev_get_valid_name(dev, dev->name);
5610 /* Init, if this function is available */
5611 if (dev->netdev_ops->ndo_init) {
5612 ret = dev->netdev_ops->ndo_init(dev);
5620 dev->ifindex = dev_new_index(net);
5621 if (dev->iflink == -1)
5622 dev->iflink = dev->ifindex;
5624 /* Transfer changeable features to wanted_features and enable
5625 * software offloads (GSO and GRO).
5627 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5628 dev->features |= NETIF_F_SOFT_FEATURES;
5629 dev->wanted_features = dev->features & dev->hw_features;
5631 /* Turn on no cache copy if HW is doing checksum */
5632 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5633 if ((dev->features & NETIF_F_ALL_CSUM) &&
5634 !(dev->features & NETIF_F_NO_CSUM)) {
5635 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5636 dev->features |= NETIF_F_NOCACHE_COPY;
5639 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5641 dev->vlan_features |= NETIF_F_HIGHDMA;
5643 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5644 ret = notifier_to_errno(ret);
5648 ret = netdev_register_kobject(dev);
5651 dev->reg_state = NETREG_REGISTERED;
5653 __netdev_update_features(dev);
5656 * Default initial state at registry is that the
5657 * device is present.
5660 set_bit(__LINK_STATE_PRESENT, &dev->state);
5662 dev_init_scheduler(dev);
5664 list_netdevice(dev);
5665 add_device_randomness(dev->dev_addr, dev->addr_len);
5667 /* Notify protocols, that a new device appeared. */
5668 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5669 ret = notifier_to_errno(ret);
5671 rollback_registered(dev);
5672 dev->reg_state = NETREG_UNREGISTERED;
5675 * Prevent userspace races by waiting until the network
5676 * device is fully setup before sending notifications.
5678 if (!dev->rtnl_link_ops ||
5679 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5680 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5686 if (dev->netdev_ops->ndo_uninit)
5687 dev->netdev_ops->ndo_uninit(dev);
5690 EXPORT_SYMBOL(register_netdevice);
5693 * init_dummy_netdev - init a dummy network device for NAPI
5694 * @dev: device to init
5696 * This takes a network device structure and initialize the minimum
5697 * amount of fields so it can be used to schedule NAPI polls without
5698 * registering a full blown interface. This is to be used by drivers
5699 * that need to tie several hardware interfaces to a single NAPI
5700 * poll scheduler due to HW limitations.
5702 int init_dummy_netdev(struct net_device *dev)
5704 /* Clear everything. Note we don't initialize spinlocks
5705 * are they aren't supposed to be taken by any of the
5706 * NAPI code and this dummy netdev is supposed to be
5707 * only ever used for NAPI polls
5709 memset(dev, 0, sizeof(struct net_device));
5711 /* make sure we BUG if trying to hit standard
5712 * register/unregister code path
5714 dev->reg_state = NETREG_DUMMY;
5716 /* NAPI wants this */
5717 INIT_LIST_HEAD(&dev->napi_list);
5719 /* a dummy interface is started by default */
5720 set_bit(__LINK_STATE_PRESENT, &dev->state);
5721 set_bit(__LINK_STATE_START, &dev->state);
5723 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5724 * because users of this 'device' dont need to change
5730 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5734 * register_netdev - register a network device
5735 * @dev: device to register
5737 * Take a completed network device structure and add it to the kernel
5738 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5739 * chain. 0 is returned on success. A negative errno code is returned
5740 * on a failure to set up the device, or if the name is a duplicate.
5742 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5743 * and expands the device name if you passed a format string to
5746 int register_netdev(struct net_device *dev)
5751 err = register_netdevice(dev);
5755 EXPORT_SYMBOL(register_netdev);
5757 int netdev_refcnt_read(const struct net_device *dev)
5761 for_each_possible_cpu(i)
5762 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5765 EXPORT_SYMBOL(netdev_refcnt_read);
5768 * netdev_wait_allrefs - wait until all references are gone.
5770 * This is called when unregistering network devices.
5772 * Any protocol or device that holds a reference should register
5773 * for netdevice notification, and cleanup and put back the
5774 * reference if they receive an UNREGISTER event.
5775 * We can get stuck here if buggy protocols don't correctly
5778 static void netdev_wait_allrefs(struct net_device *dev)
5780 unsigned long rebroadcast_time, warning_time;
5783 linkwatch_forget_dev(dev);
5785 rebroadcast_time = warning_time = jiffies;
5786 refcnt = netdev_refcnt_read(dev);
5788 while (refcnt != 0) {
5789 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5792 /* Rebroadcast unregister notification */
5793 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5794 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5795 * should have already handle it the first time */
5797 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5799 /* We must not have linkwatch events
5800 * pending on unregister. If this
5801 * happens, we simply run the queue
5802 * unscheduled, resulting in a noop
5805 linkwatch_run_queue();
5810 rebroadcast_time = jiffies;
5815 refcnt = netdev_refcnt_read(dev);
5817 if (time_after(jiffies, warning_time + 10 * HZ)) {
5818 printk(KERN_EMERG "unregister_netdevice: "
5819 "waiting for %s to become free. Usage "
5822 warning_time = jiffies;
5831 * register_netdevice(x1);
5832 * register_netdevice(x2);
5834 * unregister_netdevice(y1);
5835 * unregister_netdevice(y2);
5841 * We are invoked by rtnl_unlock().
5842 * This allows us to deal with problems:
5843 * 1) We can delete sysfs objects which invoke hotplug
5844 * without deadlocking with linkwatch via keventd.
5845 * 2) Since we run with the RTNL semaphore not held, we can sleep
5846 * safely in order to wait for the netdev refcnt to drop to zero.
5848 * We must not return until all unregister events added during
5849 * the interval the lock was held have been completed.
5851 void netdev_run_todo(void)
5853 struct list_head list;
5855 /* Snapshot list, allow later requests */
5856 list_replace_init(&net_todo_list, &list);
5860 /* Wait for rcu callbacks to finish before attempting to drain
5861 * the device list. This usually avoids a 250ms wait.
5863 if (!list_empty(&list))
5866 while (!list_empty(&list)) {
5867 struct net_device *dev
5868 = list_first_entry(&list, struct net_device, todo_list);
5869 list_del(&dev->todo_list);
5871 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5872 printk(KERN_ERR "network todo '%s' but state %d\n",
5873 dev->name, dev->reg_state);
5878 dev->reg_state = NETREG_UNREGISTERED;
5880 on_each_cpu(flush_backlog, dev, 1);
5882 netdev_wait_allrefs(dev);
5885 BUG_ON(netdev_refcnt_read(dev));
5886 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5887 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5888 WARN_ON(dev->dn_ptr);
5890 if (dev->destructor)
5891 dev->destructor(dev);
5893 /* Free network device */
5894 kobject_put(&dev->dev.kobj);
5898 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5899 * fields in the same order, with only the type differing.
5901 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5902 const struct net_device_stats *netdev_stats)
5904 #if BITS_PER_LONG == 64
5905 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5906 memcpy(stats64, netdev_stats, sizeof(*stats64));
5908 size_t i, n = sizeof(*stats64) / sizeof(u64);
5909 const unsigned long *src = (const unsigned long *)netdev_stats;
5910 u64 *dst = (u64 *)stats64;
5912 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5913 sizeof(*stats64) / sizeof(u64));
5914 for (i = 0; i < n; i++)
5920 * dev_get_stats - get network device statistics
5921 * @dev: device to get statistics from
5922 * @storage: place to store stats
5924 * Get network statistics from device. Return @storage.
5925 * The device driver may provide its own method by setting
5926 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5927 * otherwise the internal statistics structure is used.
5929 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5930 struct rtnl_link_stats64 *storage)
5932 const struct net_device_ops *ops = dev->netdev_ops;
5934 if (ops->ndo_get_stats64) {
5935 memset(storage, 0, sizeof(*storage));
5936 ops->ndo_get_stats64(dev, storage);
5937 } else if (ops->ndo_get_stats) {
5938 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5940 netdev_stats_to_stats64(storage, &dev->stats);
5942 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5945 EXPORT_SYMBOL(dev_get_stats);
5947 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5949 struct netdev_queue *queue = dev_ingress_queue(dev);
5951 #ifdef CONFIG_NET_CLS_ACT
5954 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5957 netdev_init_one_queue(dev, queue, NULL);
5958 queue->qdisc = &noop_qdisc;
5959 queue->qdisc_sleeping = &noop_qdisc;
5960 rcu_assign_pointer(dev->ingress_queue, queue);
5966 * alloc_netdev_mqs - allocate network device
5967 * @sizeof_priv: size of private data to allocate space for
5968 * @name: device name format string
5969 * @setup: callback to initialize device
5970 * @txqs: the number of TX subqueues to allocate
5971 * @rxqs: the number of RX subqueues to allocate
5973 * Allocates a struct net_device with private data area for driver use
5974 * and performs basic initialization. Also allocates subquue structs
5975 * for each queue on the device.
5977 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5978 void (*setup)(struct net_device *),
5979 unsigned int txqs, unsigned int rxqs)
5981 struct net_device *dev;
5983 struct net_device *p;
5985 BUG_ON(strlen(name) >= sizeof(dev->name));
5988 pr_err("alloc_netdev: Unable to allocate device "
5989 "with zero queues.\n");
5995 pr_err("alloc_netdev: Unable to allocate device "
5996 "with zero RX queues.\n");
6001 alloc_size = sizeof(struct net_device);
6003 /* ensure 32-byte alignment of private area */
6004 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6005 alloc_size += sizeof_priv;
6007 /* ensure 32-byte alignment of whole construct */
6008 alloc_size += NETDEV_ALIGN - 1;
6010 p = kzalloc(alloc_size, GFP_KERNEL);
6012 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
6016 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6017 dev->padded = (char *)dev - (char *)p;
6019 dev->pcpu_refcnt = alloc_percpu(int);
6020 if (!dev->pcpu_refcnt)
6023 if (dev_addr_init(dev))
6029 dev_net_set(dev, &init_net);
6031 dev->gso_max_size = GSO_MAX_SIZE;
6032 dev->gso_max_segs = GSO_MAX_SEGS;
6034 INIT_LIST_HEAD(&dev->napi_list);
6035 INIT_LIST_HEAD(&dev->unreg_list);
6036 INIT_LIST_HEAD(&dev->link_watch_list);
6037 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6040 dev->num_tx_queues = txqs;
6041 dev->real_num_tx_queues = txqs;
6042 if (netif_alloc_netdev_queues(dev))
6046 dev->num_rx_queues = rxqs;
6047 dev->real_num_rx_queues = rxqs;
6048 if (netif_alloc_rx_queues(dev))
6052 strcpy(dev->name, name);
6053 dev->group = INIT_NETDEV_GROUP;
6061 free_percpu(dev->pcpu_refcnt);
6071 EXPORT_SYMBOL(alloc_netdev_mqs);
6074 * free_netdev - free network device
6077 * This function does the last stage of destroying an allocated device
6078 * interface. The reference to the device object is released.
6079 * If this is the last reference then it will be freed.
6081 void free_netdev(struct net_device *dev)
6083 struct napi_struct *p, *n;
6085 release_net(dev_net(dev));
6092 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6094 /* Flush device addresses */
6095 dev_addr_flush(dev);
6097 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6100 free_percpu(dev->pcpu_refcnt);
6101 dev->pcpu_refcnt = NULL;
6103 /* Compatibility with error handling in drivers */
6104 if (dev->reg_state == NETREG_UNINITIALIZED) {
6105 kfree((char *)dev - dev->padded);
6109 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6110 dev->reg_state = NETREG_RELEASED;
6112 /* will free via device release */
6113 put_device(&dev->dev);
6115 EXPORT_SYMBOL(free_netdev);
6118 * synchronize_net - Synchronize with packet receive processing
6120 * Wait for packets currently being received to be done.
6121 * Does not block later packets from starting.
6123 void synchronize_net(void)
6126 if (rtnl_is_locked())
6127 synchronize_rcu_expedited();
6131 EXPORT_SYMBOL(synchronize_net);
6134 * unregister_netdevice_queue - remove device from the kernel
6138 * This function shuts down a device interface and removes it
6139 * from the kernel tables.
6140 * If head not NULL, device is queued to be unregistered later.
6142 * Callers must hold the rtnl semaphore. You may want
6143 * unregister_netdev() instead of this.
6146 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6151 list_move_tail(&dev->unreg_list, head);
6153 rollback_registered(dev);
6154 /* Finish processing unregister after unlock */
6158 EXPORT_SYMBOL(unregister_netdevice_queue);
6161 * unregister_netdevice_many - unregister many devices
6162 * @head: list of devices
6164 void unregister_netdevice_many(struct list_head *head)
6166 struct net_device *dev;
6168 if (!list_empty(head)) {
6169 rollback_registered_many(head);
6170 list_for_each_entry(dev, head, unreg_list)
6174 EXPORT_SYMBOL(unregister_netdevice_many);
6177 * unregister_netdev - remove device from the kernel
6180 * This function shuts down a device interface and removes it
6181 * from the kernel tables.
6183 * This is just a wrapper for unregister_netdevice that takes
6184 * the rtnl semaphore. In general you want to use this and not
6185 * unregister_netdevice.
6187 void unregister_netdev(struct net_device *dev)
6190 unregister_netdevice(dev);
6193 EXPORT_SYMBOL(unregister_netdev);
6196 * dev_change_net_namespace - move device to different nethost namespace
6198 * @net: network namespace
6199 * @pat: If not NULL name pattern to try if the current device name
6200 * is already taken in the destination network namespace.
6202 * This function shuts down a device interface and moves it
6203 * to a new network namespace. On success 0 is returned, on
6204 * a failure a netagive errno code is returned.
6206 * Callers must hold the rtnl semaphore.
6209 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6215 /* Don't allow namespace local devices to be moved. */
6217 if (dev->features & NETIF_F_NETNS_LOCAL)
6220 /* Ensure the device has been registrered */
6222 if (dev->reg_state != NETREG_REGISTERED)
6225 /* Get out if there is nothing todo */
6227 if (net_eq(dev_net(dev), net))
6230 /* Pick the destination device name, and ensure
6231 * we can use it in the destination network namespace.
6234 if (__dev_get_by_name(net, dev->name)) {
6235 /* We get here if we can't use the current device name */
6238 if (dev_get_valid_name(dev, pat) < 0)
6243 * And now a mini version of register_netdevice unregister_netdevice.
6246 /* If device is running close it first. */
6249 /* And unlink it from device chain */
6251 unlist_netdevice(dev);
6255 /* Shutdown queueing discipline. */
6258 /* Notify protocols, that we are about to destroy
6259 this device. They should clean all the things.
6261 Note that dev->reg_state stays at NETREG_REGISTERED.
6262 This is wanted because this way 8021q and macvlan know
6263 the device is just moving and can keep their slaves up.
6265 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6266 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6267 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6270 * Flush the unicast and multicast chains
6275 /* Actually switch the network namespace */
6276 dev_net_set(dev, net);
6278 /* If there is an ifindex conflict assign a new one */
6279 if (__dev_get_by_index(net, dev->ifindex)) {
6280 int iflink = (dev->iflink == dev->ifindex);
6281 dev->ifindex = dev_new_index(net);
6283 dev->iflink = dev->ifindex;
6286 /* Fixup kobjects */
6287 err = device_rename(&dev->dev, dev->name);
6290 /* Add the device back in the hashes */
6291 list_netdevice(dev);
6293 /* Notify protocols, that a new device appeared. */
6294 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6297 * Prevent userspace races by waiting until the network
6298 * device is fully setup before sending notifications.
6300 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6307 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6309 static int dev_cpu_callback(struct notifier_block *nfb,
6310 unsigned long action,
6313 struct sk_buff **list_skb;
6314 struct sk_buff *skb;
6315 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6316 struct softnet_data *sd, *oldsd;
6318 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6321 local_irq_disable();
6322 cpu = smp_processor_id();
6323 sd = &per_cpu(softnet_data, cpu);
6324 oldsd = &per_cpu(softnet_data, oldcpu);
6326 /* Find end of our completion_queue. */
6327 list_skb = &sd->completion_queue;
6329 list_skb = &(*list_skb)->next;
6330 /* Append completion queue from offline CPU. */
6331 *list_skb = oldsd->completion_queue;
6332 oldsd->completion_queue = NULL;
6334 /* Append output queue from offline CPU. */
6335 if (oldsd->output_queue) {
6336 *sd->output_queue_tailp = oldsd->output_queue;
6337 sd->output_queue_tailp = oldsd->output_queue_tailp;
6338 oldsd->output_queue = NULL;
6339 oldsd->output_queue_tailp = &oldsd->output_queue;
6341 /* Append NAPI poll list from offline CPU, with one exception :
6342 * process_backlog() must be called by cpu owning percpu backlog.
6343 * We properly handle process_queue & input_pkt_queue later.
6345 while (!list_empty(&oldsd->poll_list)) {
6346 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
6350 list_del_init(&napi->poll_list);
6351 if (napi->poll == process_backlog)
6354 ____napi_schedule(sd, napi);
6357 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6360 /* Process offline CPU's input_pkt_queue */
6361 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6363 input_queue_head_incr(oldsd);
6365 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
6367 input_queue_head_incr(oldsd);
6375 * netdev_increment_features - increment feature set by one
6376 * @all: current feature set
6377 * @one: new feature set
6378 * @mask: mask feature set
6380 * Computes a new feature set after adding a device with feature set
6381 * @one to the master device with current feature set @all. Will not
6382 * enable anything that is off in @mask. Returns the new feature set.
6384 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6386 if (mask & NETIF_F_GEN_CSUM)
6387 mask |= NETIF_F_ALL_CSUM;
6388 mask |= NETIF_F_VLAN_CHALLENGED;
6390 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6391 all &= one | ~NETIF_F_ALL_FOR_ALL;
6393 /* If device needs checksumming, downgrade to it. */
6394 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6395 all &= ~NETIF_F_NO_CSUM;
6397 /* If one device supports hw checksumming, set for all. */
6398 if (all & NETIF_F_GEN_CSUM)
6399 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6403 EXPORT_SYMBOL(netdev_increment_features);
6405 static struct hlist_head *netdev_create_hash(void)
6408 struct hlist_head *hash;
6410 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6412 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6413 INIT_HLIST_HEAD(&hash[i]);
6418 /* Initialize per network namespace state */
6419 static int __net_init netdev_init(struct net *net)
6421 if (net != &init_net)
6422 INIT_LIST_HEAD(&net->dev_base_head);
6424 net->dev_name_head = netdev_create_hash();
6425 if (net->dev_name_head == NULL)
6428 net->dev_index_head = netdev_create_hash();
6429 if (net->dev_index_head == NULL)
6435 kfree(net->dev_name_head);
6441 * netdev_drivername - network driver for the device
6442 * @dev: network device
6444 * Determine network driver for device.
6446 const char *netdev_drivername(const struct net_device *dev)
6448 const struct device_driver *driver;
6449 const struct device *parent;
6450 const char *empty = "";
6452 parent = dev->dev.parent;
6456 driver = parent->driver;
6457 if (driver && driver->name)
6458 return driver->name;
6462 int __netdev_printk(const char *level, const struct net_device *dev,
6463 struct va_format *vaf)
6467 if (dev && dev->dev.parent)
6468 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6469 netdev_name(dev), vaf);
6471 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6473 r = printk("%s(NULL net_device): %pV", level, vaf);
6477 EXPORT_SYMBOL(__netdev_printk);
6479 int netdev_printk(const char *level, const struct net_device *dev,
6480 const char *format, ...)
6482 struct va_format vaf;
6486 va_start(args, format);
6491 r = __netdev_printk(level, dev, &vaf);
6496 EXPORT_SYMBOL(netdev_printk);
6498 #define define_netdev_printk_level(func, level) \
6499 int func(const struct net_device *dev, const char *fmt, ...) \
6502 struct va_format vaf; \
6505 va_start(args, fmt); \
6510 r = __netdev_printk(level, dev, &vaf); \
6515 EXPORT_SYMBOL(func);
6517 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6518 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6519 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6520 define_netdev_printk_level(netdev_err, KERN_ERR);
6521 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6522 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6523 define_netdev_printk_level(netdev_info, KERN_INFO);
6525 static void __net_exit netdev_exit(struct net *net)
6527 kfree(net->dev_name_head);
6528 kfree(net->dev_index_head);
6531 static struct pernet_operations __net_initdata netdev_net_ops = {
6532 .init = netdev_init,
6533 .exit = netdev_exit,
6536 static void __net_exit default_device_exit(struct net *net)
6538 struct net_device *dev, *aux;
6540 * Push all migratable network devices back to the
6541 * initial network namespace
6544 for_each_netdev_safe(net, dev, aux) {
6546 char fb_name[IFNAMSIZ];
6548 /* Ignore unmoveable devices (i.e. loopback) */
6549 if (dev->features & NETIF_F_NETNS_LOCAL)
6552 /* Leave virtual devices for the generic cleanup */
6553 if (dev->rtnl_link_ops)
6556 /* Push remaining network devices to init_net */
6557 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6558 err = dev_change_net_namespace(dev, &init_net, fb_name);
6560 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6561 __func__, dev->name, err);
6568 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6570 /* At exit all network devices most be removed from a network
6571 * namespace. Do this in the reverse order of registration.
6572 * Do this across as many network namespaces as possible to
6573 * improve batching efficiency.
6575 struct net_device *dev;
6577 LIST_HEAD(dev_kill_list);
6580 list_for_each_entry(net, net_list, exit_list) {
6581 for_each_netdev_reverse(net, dev) {
6582 if (dev->rtnl_link_ops)
6583 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6585 unregister_netdevice_queue(dev, &dev_kill_list);
6588 unregister_netdevice_many(&dev_kill_list);
6589 list_del(&dev_kill_list);
6593 static struct pernet_operations __net_initdata default_device_ops = {
6594 .exit = default_device_exit,
6595 .exit_batch = default_device_exit_batch,
6599 * Initialize the DEV module. At boot time this walks the device list and
6600 * unhooks any devices that fail to initialise (normally hardware not
6601 * present) and leaves us with a valid list of present and active devices.
6606 * This is called single threaded during boot, so no need
6607 * to take the rtnl semaphore.
6609 static int __init net_dev_init(void)
6611 int i, rc = -ENOMEM;
6613 BUG_ON(!dev_boot_phase);
6615 if (dev_proc_init())
6618 if (netdev_kobject_init())
6621 INIT_LIST_HEAD(&ptype_all);
6622 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6623 INIT_LIST_HEAD(&ptype_base[i]);
6625 if (register_pernet_subsys(&netdev_net_ops))
6629 * Initialise the packet receive queues.
6632 for_each_possible_cpu(i) {
6633 struct softnet_data *sd = &per_cpu(softnet_data, i);
6635 memset(sd, 0, sizeof(*sd));
6636 skb_queue_head_init(&sd->input_pkt_queue);
6637 skb_queue_head_init(&sd->process_queue);
6638 sd->completion_queue = NULL;
6639 INIT_LIST_HEAD(&sd->poll_list);
6640 sd->output_queue = NULL;
6641 sd->output_queue_tailp = &sd->output_queue;
6643 sd->csd.func = rps_trigger_softirq;
6649 sd->backlog.poll = process_backlog;
6650 sd->backlog.weight = weight_p;
6651 sd->backlog.gro_list = NULL;
6652 sd->backlog.gro_count = 0;
6657 /* The loopback device is special if any other network devices
6658 * is present in a network namespace the loopback device must
6659 * be present. Since we now dynamically allocate and free the
6660 * loopback device ensure this invariant is maintained by
6661 * keeping the loopback device as the first device on the
6662 * list of network devices. Ensuring the loopback devices
6663 * is the first device that appears and the last network device
6666 if (register_pernet_device(&loopback_net_ops))
6669 if (register_pernet_device(&default_device_ops))
6672 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6673 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6675 hotcpu_notifier(dev_cpu_callback, 0);
6683 subsys_initcall(net_dev_init);
6685 static int __init initialize_hashrnd(void)
6687 get_random_bytes(&hashrnd, sizeof(hashrnd));
6691 late_initcall_sync(initialize_hashrnd);