2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/if_tunnel.h>
137 #include <linux/if_pppox.h>
138 #include <linux/ppp_defs.h>
139 #include <linux/net_tstamp.h>
141 #include "net-sysfs.h"
143 /* Instead of increasing this, you should create a hash table. */
144 #define MAX_GRO_SKBS 8
146 /* This should be increased if a protocol with a bigger head is added. */
147 #define GRO_MAX_HEAD (MAX_HEADER + 128)
150 * The list of packet types we will receive (as opposed to discard)
151 * and the routines to invoke.
153 * Why 16. Because with 16 the only overlap we get on a hash of the
154 * low nibble of the protocol value is RARP/SNAP/X.25.
156 * NOTE: That is no longer true with the addition of VLAN tags. Not
157 * sure which should go first, but I bet it won't make much
158 * difference if we are running VLANs. The good news is that
159 * this protocol won't be in the list unless compiled in, so
160 * the average user (w/out VLANs) will not be adversely affected.
177 #define PTYPE_HASH_SIZE (16)
178 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
180 static DEFINE_SPINLOCK(ptype_lock);
181 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
182 static struct list_head ptype_all __read_mostly; /* Taps */
185 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
188 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
190 * Writers must hold the rtnl semaphore while they loop through the
191 * dev_base_head list, and hold dev_base_lock for writing when they do the
192 * actual updates. This allows pure readers to access the list even
193 * while a writer is preparing to update it.
195 * To put it another way, dev_base_lock is held for writing only to
196 * protect against pure readers; the rtnl semaphore provides the
197 * protection against other writers.
199 * See, for example usages, register_netdevice() and
200 * unregister_netdevice(), which must be called with the rtnl
203 DEFINE_RWLOCK(dev_base_lock);
204 EXPORT_SYMBOL(dev_base_lock);
206 static inline void dev_base_seq_inc(struct net *net)
208 while (++net->dev_base_seq == 0);
211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
213 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
214 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
217 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
219 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
222 static inline void rps_lock(struct softnet_data *sd)
225 spin_lock(&sd->input_pkt_queue.lock);
229 static inline void rps_unlock(struct softnet_data *sd)
232 spin_unlock(&sd->input_pkt_queue.lock);
236 /* Device list insertion */
237 static int list_netdevice(struct net_device *dev)
239 struct net *net = dev_net(dev);
243 write_lock_bh(&dev_base_lock);
244 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
245 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
246 hlist_add_head_rcu(&dev->index_hlist,
247 dev_index_hash(net, dev->ifindex));
248 write_unlock_bh(&dev_base_lock);
250 dev_base_seq_inc(net);
255 /* Device list removal
256 * caller must respect a RCU grace period before freeing/reusing dev
258 static void unlist_netdevice(struct net_device *dev)
262 /* Unlink dev from the device chain */
263 write_lock_bh(&dev_base_lock);
264 list_del_rcu(&dev->dev_list);
265 hlist_del_rcu(&dev->name_hlist);
266 hlist_del_rcu(&dev->index_hlist);
267 write_unlock_bh(&dev_base_lock);
269 dev_base_seq_inc(dev_net(dev));
276 static RAW_NOTIFIER_HEAD(netdev_chain);
279 * Device drivers call our routines to queue packets here. We empty the
280 * queue in the local softnet handler.
283 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
284 EXPORT_PER_CPU_SYMBOL(softnet_data);
286 #ifdef CONFIG_LOCKDEP
288 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
289 * according to dev->type
291 static const unsigned short netdev_lock_type[] =
292 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
293 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
294 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
295 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
296 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
297 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
298 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
299 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
300 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
301 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
302 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
303 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
304 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
305 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
306 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
307 ARPHRD_VOID, ARPHRD_NONE};
309 static const char *const netdev_lock_name[] =
310 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
311 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
312 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
313 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
314 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
315 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
316 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
317 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
318 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
319 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
320 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
321 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
322 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
323 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
324 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
325 "_xmit_VOID", "_xmit_NONE"};
327 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
328 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
330 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
334 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
335 if (netdev_lock_type[i] == dev_type)
337 /* the last key is used by default */
338 return ARRAY_SIZE(netdev_lock_type) - 1;
341 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
342 unsigned short dev_type)
346 i = netdev_lock_pos(dev_type);
347 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
348 netdev_lock_name[i]);
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
355 i = netdev_lock_pos(dev->type);
356 lockdep_set_class_and_name(&dev->addr_list_lock,
357 &netdev_addr_lock_key[i],
358 netdev_lock_name[i]);
361 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
362 unsigned short dev_type)
365 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
370 /*******************************************************************************
372 Protocol management and registration routines
374 *******************************************************************************/
377 * Add a protocol ID to the list. Now that the input handler is
378 * smarter we can dispense with all the messy stuff that used to be
381 * BEWARE!!! Protocol handlers, mangling input packets,
382 * MUST BE last in hash buckets and checking protocol handlers
383 * MUST start from promiscuous ptype_all chain in net_bh.
384 * It is true now, do not change it.
385 * Explanation follows: if protocol handler, mangling packet, will
386 * be the first on list, it is not able to sense, that packet
387 * is cloned and should be copied-on-write, so that it will
388 * change it and subsequent readers will get broken packet.
392 static inline struct list_head *ptype_head(const struct packet_type *pt)
394 if (pt->type == htons(ETH_P_ALL))
397 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
401 * dev_add_pack - add packet handler
402 * @pt: packet type declaration
404 * Add a protocol handler to the networking stack. The passed &packet_type
405 * is linked into kernel lists and may not be freed until it has been
406 * removed from the kernel lists.
408 * This call does not sleep therefore it can not
409 * guarantee all CPU's that are in middle of receiving packets
410 * will see the new packet type (until the next received packet).
413 void dev_add_pack(struct packet_type *pt)
415 struct list_head *head = ptype_head(pt);
417 spin_lock(&ptype_lock);
418 list_add_rcu(&pt->list, head);
419 spin_unlock(&ptype_lock);
421 EXPORT_SYMBOL(dev_add_pack);
424 * __dev_remove_pack - remove packet handler
425 * @pt: packet type declaration
427 * Remove a protocol handler that was previously added to the kernel
428 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
429 * from the kernel lists and can be freed or reused once this function
432 * The packet type might still be in use by receivers
433 * and must not be freed until after all the CPU's have gone
434 * through a quiescent state.
436 void __dev_remove_pack(struct packet_type *pt)
438 struct list_head *head = ptype_head(pt);
439 struct packet_type *pt1;
441 spin_lock(&ptype_lock);
443 list_for_each_entry(pt1, head, list) {
445 list_del_rcu(&pt->list);
450 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
452 spin_unlock(&ptype_lock);
454 EXPORT_SYMBOL(__dev_remove_pack);
457 * dev_remove_pack - remove packet handler
458 * @pt: packet type declaration
460 * Remove a protocol handler that was previously added to the kernel
461 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
462 * from the kernel lists and can be freed or reused once this function
465 * This call sleeps to guarantee that no CPU is looking at the packet
468 void dev_remove_pack(struct packet_type *pt)
470 __dev_remove_pack(pt);
474 EXPORT_SYMBOL(dev_remove_pack);
476 /******************************************************************************
478 Device Boot-time Settings Routines
480 *******************************************************************************/
482 /* Boot time configuration table */
483 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
486 * netdev_boot_setup_add - add new setup entry
487 * @name: name of the device
488 * @map: configured settings for the device
490 * Adds new setup entry to the dev_boot_setup list. The function
491 * returns 0 on error and 1 on success. This is a generic routine to
494 static int netdev_boot_setup_add(char *name, struct ifmap *map)
496 struct netdev_boot_setup *s;
500 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
501 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
502 memset(s[i].name, 0, sizeof(s[i].name));
503 strlcpy(s[i].name, name, IFNAMSIZ);
504 memcpy(&s[i].map, map, sizeof(s[i].map));
509 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
513 * netdev_boot_setup_check - check boot time settings
514 * @dev: the netdevice
516 * Check boot time settings for the device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found, 1 if they are.
521 int netdev_boot_setup_check(struct net_device *dev)
523 struct netdev_boot_setup *s = dev_boot_setup;
526 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
527 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
528 !strcmp(dev->name, s[i].name)) {
529 dev->irq = s[i].map.irq;
530 dev->base_addr = s[i].map.base_addr;
531 dev->mem_start = s[i].map.mem_start;
532 dev->mem_end = s[i].map.mem_end;
538 EXPORT_SYMBOL(netdev_boot_setup_check);
542 * netdev_boot_base - get address from boot time settings
543 * @prefix: prefix for network device
544 * @unit: id for network device
546 * Check boot time settings for the base address of device.
547 * The found settings are set for the device to be used
548 * later in the device probing.
549 * Returns 0 if no settings found.
551 unsigned long netdev_boot_base(const char *prefix, int unit)
553 const struct netdev_boot_setup *s = dev_boot_setup;
557 sprintf(name, "%s%d", prefix, unit);
560 * If device already registered then return base of 1
561 * to indicate not to probe for this interface
563 if (__dev_get_by_name(&init_net, name))
566 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
567 if (!strcmp(name, s[i].name))
568 return s[i].map.base_addr;
573 * Saves at boot time configured settings for any netdevice.
575 int __init netdev_boot_setup(char *str)
580 str = get_options(str, ARRAY_SIZE(ints), ints);
585 memset(&map, 0, sizeof(map));
589 map.base_addr = ints[2];
591 map.mem_start = ints[3];
593 map.mem_end = ints[4];
595 /* Add new entry to the list */
596 return netdev_boot_setup_add(str, &map);
599 __setup("netdev=", netdev_boot_setup);
601 /*******************************************************************************
603 Device Interface Subroutines
605 *******************************************************************************/
608 * __dev_get_by_name - find a device by its name
609 * @net: the applicable net namespace
610 * @name: name to find
612 * Find an interface by name. Must be called under RTNL semaphore
613 * or @dev_base_lock. If the name is found a pointer to the device
614 * is returned. If the name is not found then %NULL is returned. The
615 * reference counters are not incremented so the caller must be
616 * careful with locks.
619 struct net_device *__dev_get_by_name(struct net *net, const char *name)
621 struct hlist_node *p;
622 struct net_device *dev;
623 struct hlist_head *head = dev_name_hash(net, name);
625 hlist_for_each_entry(dev, p, head, name_hlist)
626 if (!strncmp(dev->name, name, IFNAMSIZ))
631 EXPORT_SYMBOL(__dev_get_by_name);
634 * dev_get_by_name_rcu - find a device by its name
635 * @net: the applicable net namespace
636 * @name: name to find
638 * Find an interface by name.
639 * If the name is found a pointer to the device is returned.
640 * If the name is not found then %NULL is returned.
641 * The reference counters are not incremented so the caller must be
642 * careful with locks. The caller must hold RCU lock.
645 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
647 struct hlist_node *p;
648 struct net_device *dev;
649 struct hlist_head *head = dev_name_hash(net, name);
651 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
652 if (!strncmp(dev->name, name, IFNAMSIZ))
657 EXPORT_SYMBOL(dev_get_by_name_rcu);
660 * dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. This can be called from any
665 * context and does its own locking. The returned handle has
666 * the usage count incremented and the caller must use dev_put() to
667 * release it when it is no longer needed. %NULL is returned if no
668 * matching device is found.
671 struct net_device *dev_get_by_name(struct net *net, const char *name)
673 struct net_device *dev;
676 dev = dev_get_by_name_rcu(net, name);
682 EXPORT_SYMBOL(dev_get_by_name);
685 * __dev_get_by_index - find a device by its ifindex
686 * @net: the applicable net namespace
687 * @ifindex: index of device
689 * Search for an interface by index. Returns %NULL if the device
690 * is not found or a pointer to the device. The device has not
691 * had its reference counter increased so the caller must be careful
692 * about locking. The caller must hold either the RTNL semaphore
696 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
698 struct hlist_node *p;
699 struct net_device *dev;
700 struct hlist_head *head = dev_index_hash(net, ifindex);
702 hlist_for_each_entry(dev, p, head, index_hlist)
703 if (dev->ifindex == ifindex)
708 EXPORT_SYMBOL(__dev_get_by_index);
711 * dev_get_by_index_rcu - find a device by its ifindex
712 * @net: the applicable net namespace
713 * @ifindex: index of device
715 * Search for an interface by index. Returns %NULL if the device
716 * is not found or a pointer to the device. The device has not
717 * had its reference counter increased so the caller must be careful
718 * about locking. The caller must hold RCU lock.
721 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
723 struct hlist_node *p;
724 struct net_device *dev;
725 struct hlist_head *head = dev_index_hash(net, ifindex);
727 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
728 if (dev->ifindex == ifindex)
733 EXPORT_SYMBOL(dev_get_by_index_rcu);
737 * dev_get_by_index - find a device by its ifindex
738 * @net: the applicable net namespace
739 * @ifindex: index of device
741 * Search for an interface by index. Returns NULL if the device
742 * is not found or a pointer to the device. The device returned has
743 * had a reference added and the pointer is safe until the user calls
744 * dev_put to indicate they have finished with it.
747 struct net_device *dev_get_by_index(struct net *net, int ifindex)
749 struct net_device *dev;
752 dev = dev_get_by_index_rcu(net, ifindex);
758 EXPORT_SYMBOL(dev_get_by_index);
761 * dev_getbyhwaddr_rcu - find a device by its hardware address
762 * @net: the applicable net namespace
763 * @type: media type of device
764 * @ha: hardware address
766 * Search for an interface by MAC address. Returns NULL if the device
767 * is not found or a pointer to the device.
768 * The caller must hold RCU or RTNL.
769 * The returned device has not had its ref count increased
770 * and the caller must therefore be careful about locking
774 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
777 struct net_device *dev;
779 for_each_netdev_rcu(net, dev)
780 if (dev->type == type &&
781 !memcmp(dev->dev_addr, ha, dev->addr_len))
786 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
788 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
790 struct net_device *dev;
793 for_each_netdev(net, dev)
794 if (dev->type == type)
799 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
801 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
803 struct net_device *dev, *ret = NULL;
806 for_each_netdev_rcu(net, dev)
807 if (dev->type == type) {
815 EXPORT_SYMBOL(dev_getfirstbyhwtype);
818 * dev_get_by_flags_rcu - find any device with given flags
819 * @net: the applicable net namespace
820 * @if_flags: IFF_* values
821 * @mask: bitmask of bits in if_flags to check
823 * Search for any interface with the given flags. Returns NULL if a device
824 * is not found or a pointer to the device. Must be called inside
825 * rcu_read_lock(), and result refcount is unchanged.
828 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
831 struct net_device *dev, *ret;
834 for_each_netdev_rcu(net, dev) {
835 if (((dev->flags ^ if_flags) & mask) == 0) {
842 EXPORT_SYMBOL(dev_get_by_flags_rcu);
845 * dev_valid_name - check if name is okay for network device
848 * Network device names need to be valid file names to
849 * to allow sysfs to work. We also disallow any kind of
852 int dev_valid_name(const char *name)
856 if (strlen(name) >= IFNAMSIZ)
858 if (!strcmp(name, ".") || !strcmp(name, ".."))
862 if (*name == '/' || isspace(*name))
868 EXPORT_SYMBOL(dev_valid_name);
871 * __dev_alloc_name - allocate a name for a device
872 * @net: network namespace to allocate the device name in
873 * @name: name format string
874 * @buf: scratch buffer and result name string
876 * Passed a format string - eg "lt%d" it will try and find a suitable
877 * id. It scans list of devices to build up a free map, then chooses
878 * the first empty slot. The caller must hold the dev_base or rtnl lock
879 * while allocating the name and adding the device in order to avoid
881 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
882 * Returns the number of the unit assigned or a negative errno code.
885 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
889 const int max_netdevices = 8*PAGE_SIZE;
890 unsigned long *inuse;
891 struct net_device *d;
893 p = strnchr(name, IFNAMSIZ-1, '%');
896 * Verify the string as this thing may have come from
897 * the user. There must be either one "%d" and no other "%"
900 if (p[1] != 'd' || strchr(p + 2, '%'))
903 /* Use one page as a bit array of possible slots */
904 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
908 for_each_netdev(net, d) {
909 if (!sscanf(d->name, name, &i))
911 if (i < 0 || i >= max_netdevices)
914 /* avoid cases where sscanf is not exact inverse of printf */
915 snprintf(buf, IFNAMSIZ, name, i);
916 if (!strncmp(buf, d->name, IFNAMSIZ))
920 i = find_first_zero_bit(inuse, max_netdevices);
921 free_page((unsigned long) inuse);
925 snprintf(buf, IFNAMSIZ, name, i);
926 if (!__dev_get_by_name(net, buf))
929 /* It is possible to run out of possible slots
930 * when the name is long and there isn't enough space left
931 * for the digits, or if all bits are used.
937 * dev_alloc_name - allocate a name for a device
939 * @name: name format string
941 * Passed a format string - eg "lt%d" it will try and find a suitable
942 * id. It scans list of devices to build up a free map, then chooses
943 * the first empty slot. The caller must hold the dev_base or rtnl lock
944 * while allocating the name and adding the device in order to avoid
946 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
947 * Returns the number of the unit assigned or a negative errno code.
950 int dev_alloc_name(struct net_device *dev, const char *name)
956 BUG_ON(!dev_net(dev));
958 ret = __dev_alloc_name(net, name, buf);
960 strlcpy(dev->name, buf, IFNAMSIZ);
963 EXPORT_SYMBOL(dev_alloc_name);
965 static int dev_get_valid_name(struct net_device *dev, const char *name)
969 BUG_ON(!dev_net(dev));
972 if (!dev_valid_name(name))
975 if (strchr(name, '%'))
976 return dev_alloc_name(dev, name);
977 else if (__dev_get_by_name(net, name))
979 else if (dev->name != name)
980 strlcpy(dev->name, name, IFNAMSIZ);
986 * dev_change_name - change name of a device
988 * @newname: name (or format string) must be at least IFNAMSIZ
990 * Change name of a device, can pass format strings "eth%d".
993 int dev_change_name(struct net_device *dev, const char *newname)
995 char oldname[IFNAMSIZ];
1001 BUG_ON(!dev_net(dev));
1004 if (dev->flags & IFF_UP)
1007 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1010 memcpy(oldname, dev->name, IFNAMSIZ);
1012 err = dev_get_valid_name(dev, newname);
1017 ret = device_rename(&dev->dev, dev->name);
1019 memcpy(dev->name, oldname, IFNAMSIZ);
1023 write_lock_bh(&dev_base_lock);
1024 hlist_del_rcu(&dev->name_hlist);
1025 write_unlock_bh(&dev_base_lock);
1029 write_lock_bh(&dev_base_lock);
1030 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1031 write_unlock_bh(&dev_base_lock);
1033 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1034 ret = notifier_to_errno(ret);
1037 /* err >= 0 after dev_alloc_name() or stores the first errno */
1040 memcpy(dev->name, oldname, IFNAMSIZ);
1044 "%s: name change rollback failed: %d.\n",
1053 * dev_set_alias - change ifalias of a device
1055 * @alias: name up to IFALIASZ
1056 * @len: limit of bytes to copy from info
1058 * Set ifalias for a device,
1060 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1066 if (len >= IFALIASZ)
1071 kfree(dev->ifalias);
1072 dev->ifalias = NULL;
1077 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1080 dev->ifalias = new_ifalias;
1082 strlcpy(dev->ifalias, alias, len+1);
1088 * netdev_features_change - device changes features
1089 * @dev: device to cause notification
1091 * Called to indicate a device has changed features.
1093 void netdev_features_change(struct net_device *dev)
1095 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1097 EXPORT_SYMBOL(netdev_features_change);
1100 * netdev_state_change - device changes state
1101 * @dev: device to cause notification
1103 * Called to indicate a device has changed state. This function calls
1104 * the notifier chains for netdev_chain and sends a NEWLINK message
1105 * to the routing socket.
1107 void netdev_state_change(struct net_device *dev)
1109 if (dev->flags & IFF_UP) {
1110 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1111 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1114 EXPORT_SYMBOL(netdev_state_change);
1116 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1118 return call_netdevice_notifiers(event, dev);
1120 EXPORT_SYMBOL(netdev_bonding_change);
1123 * dev_load - load a network module
1124 * @net: the applicable net namespace
1125 * @name: name of interface
1127 * If a network interface is not present and the process has suitable
1128 * privileges this function loads the module. If module loading is not
1129 * available in this kernel then it becomes a nop.
1132 void dev_load(struct net *net, const char *name)
1134 struct net_device *dev;
1138 dev = dev_get_by_name_rcu(net, name);
1142 if (no_module && capable(CAP_NET_ADMIN))
1143 no_module = request_module("netdev-%s", name);
1144 if (no_module && capable(CAP_SYS_MODULE)) {
1145 if (!request_module("%s", name))
1146 pr_err("Loading kernel module for a network device "
1147 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1151 EXPORT_SYMBOL(dev_load);
1153 static int __dev_open(struct net_device *dev)
1155 const struct net_device_ops *ops = dev->netdev_ops;
1160 if (!netif_device_present(dev))
1163 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1164 ret = notifier_to_errno(ret);
1168 set_bit(__LINK_STATE_START, &dev->state);
1170 if (ops->ndo_validate_addr)
1171 ret = ops->ndo_validate_addr(dev);
1173 if (!ret && ops->ndo_open)
1174 ret = ops->ndo_open(dev);
1177 clear_bit(__LINK_STATE_START, &dev->state);
1179 dev->flags |= IFF_UP;
1180 net_dmaengine_get();
1181 dev_set_rx_mode(dev);
1183 add_device_randomness(dev->dev_addr, dev->addr_len);
1190 * dev_open - prepare an interface for use.
1191 * @dev: device to open
1193 * Takes a device from down to up state. The device's private open
1194 * function is invoked and then the multicast lists are loaded. Finally
1195 * the device is moved into the up state and a %NETDEV_UP message is
1196 * sent to the netdev notifier chain.
1198 * Calling this function on an active interface is a nop. On a failure
1199 * a negative errno code is returned.
1201 int dev_open(struct net_device *dev)
1205 if (dev->flags & IFF_UP)
1208 ret = __dev_open(dev);
1212 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1213 call_netdevice_notifiers(NETDEV_UP, dev);
1217 EXPORT_SYMBOL(dev_open);
1219 static int __dev_close_many(struct list_head *head)
1221 struct net_device *dev;
1226 list_for_each_entry(dev, head, unreg_list) {
1227 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1229 clear_bit(__LINK_STATE_START, &dev->state);
1231 /* Synchronize to scheduled poll. We cannot touch poll list, it
1232 * can be even on different cpu. So just clear netif_running().
1234 * dev->stop() will invoke napi_disable() on all of it's
1235 * napi_struct instances on this device.
1237 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1240 dev_deactivate_many(head);
1242 list_for_each_entry(dev, head, unreg_list) {
1243 const struct net_device_ops *ops = dev->netdev_ops;
1246 * Call the device specific close. This cannot fail.
1247 * Only if device is UP
1249 * We allow it to be called even after a DETACH hot-plug
1255 dev->flags &= ~IFF_UP;
1256 net_dmaengine_put();
1262 static int __dev_close(struct net_device *dev)
1267 list_add(&dev->unreg_list, &single);
1268 retval = __dev_close_many(&single);
1273 static int dev_close_many(struct list_head *head)
1275 struct net_device *dev, *tmp;
1276 LIST_HEAD(tmp_list);
1278 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1279 if (!(dev->flags & IFF_UP))
1280 list_move(&dev->unreg_list, &tmp_list);
1282 __dev_close_many(head);
1284 list_for_each_entry(dev, head, unreg_list) {
1285 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1286 call_netdevice_notifiers(NETDEV_DOWN, dev);
1289 /* rollback_registered_many needs the complete original list */
1290 list_splice(&tmp_list, head);
1295 * dev_close - shutdown an interface.
1296 * @dev: device to shutdown
1298 * This function moves an active device into down state. A
1299 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1300 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1303 int dev_close(struct net_device *dev)
1305 if (dev->flags & IFF_UP) {
1308 list_add(&dev->unreg_list, &single);
1309 dev_close_many(&single);
1314 EXPORT_SYMBOL(dev_close);
1318 * dev_disable_lro - disable Large Receive Offload on a device
1321 * Disable Large Receive Offload (LRO) on a net device. Must be
1322 * called under RTNL. This is needed if received packets may be
1323 * forwarded to another interface.
1325 void dev_disable_lro(struct net_device *dev)
1330 * If we're trying to disable lro on a vlan device
1331 * use the underlying physical device instead
1333 if (is_vlan_dev(dev))
1334 dev = vlan_dev_real_dev(dev);
1336 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1337 flags = dev->ethtool_ops->get_flags(dev);
1339 flags = ethtool_op_get_flags(dev);
1341 if (!(flags & ETH_FLAG_LRO))
1344 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1345 if (unlikely(dev->features & NETIF_F_LRO))
1346 netdev_WARN(dev, "failed to disable LRO!\n");
1348 EXPORT_SYMBOL(dev_disable_lro);
1351 static int dev_boot_phase = 1;
1354 * register_netdevice_notifier - register a network notifier block
1357 * Register a notifier to be called when network device events occur.
1358 * The notifier passed is linked into the kernel structures and must
1359 * not be reused until it has been unregistered. A negative errno code
1360 * is returned on a failure.
1362 * When registered all registration and up events are replayed
1363 * to the new notifier to allow device to have a race free
1364 * view of the network device list.
1367 int register_netdevice_notifier(struct notifier_block *nb)
1369 struct net_device *dev;
1370 struct net_device *last;
1375 err = raw_notifier_chain_register(&netdev_chain, nb);
1381 for_each_netdev(net, dev) {
1382 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1383 err = notifier_to_errno(err);
1387 if (!(dev->flags & IFF_UP))
1390 nb->notifier_call(nb, NETDEV_UP, dev);
1401 for_each_netdev(net, dev) {
1405 if (dev->flags & IFF_UP) {
1406 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1407 nb->notifier_call(nb, NETDEV_DOWN, dev);
1409 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1410 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1415 raw_notifier_chain_unregister(&netdev_chain, nb);
1418 EXPORT_SYMBOL(register_netdevice_notifier);
1421 * unregister_netdevice_notifier - unregister a network notifier block
1424 * Unregister a notifier previously registered by
1425 * register_netdevice_notifier(). The notifier is unlinked into the
1426 * kernel structures and may then be reused. A negative errno code
1427 * is returned on a failure.
1429 * After unregistering unregister and down device events are synthesized
1430 * for all devices on the device list to the removed notifier to remove
1431 * the need for special case cleanup code.
1434 int unregister_netdevice_notifier(struct notifier_block *nb)
1436 struct net_device *dev;
1441 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1446 for_each_netdev(net, dev) {
1447 if (dev->flags & IFF_UP) {
1448 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1449 nb->notifier_call(nb, NETDEV_DOWN, dev);
1451 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1452 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1459 EXPORT_SYMBOL(unregister_netdevice_notifier);
1462 * call_netdevice_notifiers - call all network notifier blocks
1463 * @val: value passed unmodified to notifier function
1464 * @dev: net_device pointer passed unmodified to notifier function
1466 * Call all network notifier blocks. Parameters and return value
1467 * are as for raw_notifier_call_chain().
1470 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1473 return raw_notifier_call_chain(&netdev_chain, val, dev);
1475 EXPORT_SYMBOL(call_netdevice_notifiers);
1477 /* When > 0 there are consumers of rx skb time stamps */
1478 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1480 void net_enable_timestamp(void)
1482 atomic_inc(&netstamp_needed);
1484 EXPORT_SYMBOL(net_enable_timestamp);
1486 void net_disable_timestamp(void)
1488 atomic_dec(&netstamp_needed);
1490 EXPORT_SYMBOL(net_disable_timestamp);
1492 static inline void net_timestamp_set(struct sk_buff *skb)
1494 if (atomic_read(&netstamp_needed))
1495 __net_timestamp(skb);
1497 skb->tstamp.tv64 = 0;
1500 static inline void net_timestamp_check(struct sk_buff *skb)
1502 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1503 __net_timestamp(skb);
1506 static int net_hwtstamp_validate(struct ifreq *ifr)
1508 struct hwtstamp_config cfg;
1509 enum hwtstamp_tx_types tx_type;
1510 enum hwtstamp_rx_filters rx_filter;
1511 int tx_type_valid = 0;
1512 int rx_filter_valid = 0;
1514 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1517 if (cfg.flags) /* reserved for future extensions */
1520 tx_type = cfg.tx_type;
1521 rx_filter = cfg.rx_filter;
1524 case HWTSTAMP_TX_OFF:
1525 case HWTSTAMP_TX_ON:
1526 case HWTSTAMP_TX_ONESTEP_SYNC:
1531 switch (rx_filter) {
1532 case HWTSTAMP_FILTER_NONE:
1533 case HWTSTAMP_FILTER_ALL:
1534 case HWTSTAMP_FILTER_SOME:
1535 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1536 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1537 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1538 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1539 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1540 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1541 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1542 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1543 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1544 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1545 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1546 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1547 rx_filter_valid = 1;
1551 if (!tx_type_valid || !rx_filter_valid)
1557 static inline bool is_skb_forwardable(struct net_device *dev,
1558 struct sk_buff *skb)
1562 if (!(dev->flags & IFF_UP))
1565 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1566 if (skb->len <= len)
1569 /* if TSO is enabled, we don't care about the length as the packet
1570 * could be forwarded without being segmented before
1572 if (skb_is_gso(skb))
1579 * dev_forward_skb - loopback an skb to another netif
1581 * @dev: destination network device
1582 * @skb: buffer to forward
1585 * NET_RX_SUCCESS (no congestion)
1586 * NET_RX_DROP (packet was dropped, but freed)
1588 * dev_forward_skb can be used for injecting an skb from the
1589 * start_xmit function of one device into the receive queue
1590 * of another device.
1592 * The receiving device may be in another namespace, so
1593 * we have to clear all information in the skb that could
1594 * impact namespace isolation.
1596 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1598 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1599 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1600 atomic_long_inc(&dev->rx_dropped);
1609 if (unlikely(!is_skb_forwardable(dev, skb))) {
1610 atomic_long_inc(&dev->rx_dropped);
1616 skb->tstamp.tv64 = 0;
1617 skb->pkt_type = PACKET_HOST;
1618 skb->protocol = eth_type_trans(skb, dev);
1622 return netif_rx(skb);
1624 EXPORT_SYMBOL_GPL(dev_forward_skb);
1626 static inline int deliver_skb(struct sk_buff *skb,
1627 struct packet_type *pt_prev,
1628 struct net_device *orig_dev)
1630 atomic_inc(&skb->users);
1631 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1635 * Support routine. Sends outgoing frames to any network
1636 * taps currently in use.
1639 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1641 struct packet_type *ptype;
1642 struct sk_buff *skb2 = NULL;
1643 struct packet_type *pt_prev = NULL;
1646 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1647 /* Never send packets back to the socket
1648 * they originated from - MvS (miquels@drinkel.ow.org)
1650 if ((ptype->dev == dev || !ptype->dev) &&
1651 (ptype->af_packet_priv == NULL ||
1652 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1654 deliver_skb(skb2, pt_prev, skb->dev);
1659 skb2 = skb_clone(skb, GFP_ATOMIC);
1663 net_timestamp_set(skb2);
1665 /* skb->nh should be correctly
1666 set by sender, so that the second statement is
1667 just protection against buggy protocols.
1669 skb_reset_mac_header(skb2);
1671 if (skb_network_header(skb2) < skb2->data ||
1672 skb2->network_header > skb2->tail) {
1673 if (net_ratelimit())
1674 printk(KERN_CRIT "protocol %04x is "
1676 ntohs(skb2->protocol),
1678 skb_reset_network_header(skb2);
1681 skb2->transport_header = skb2->network_header;
1682 skb2->pkt_type = PACKET_OUTGOING;
1687 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1691 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1692 * @dev: Network device
1693 * @txq: number of queues available
1695 * If real_num_tx_queues is changed the tc mappings may no longer be
1696 * valid. To resolve this verify the tc mapping remains valid and if
1697 * not NULL the mapping. With no priorities mapping to this
1698 * offset/count pair it will no longer be used. In the worst case TC0
1699 * is invalid nothing can be done so disable priority mappings. If is
1700 * expected that drivers will fix this mapping if they can before
1701 * calling netif_set_real_num_tx_queues.
1703 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1706 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1708 /* If TC0 is invalidated disable TC mapping */
1709 if (tc->offset + tc->count > txq) {
1710 pr_warning("Number of in use tx queues changed "
1711 "invalidating tc mappings. Priority "
1712 "traffic classification disabled!\n");
1717 /* Invalidated prio to tc mappings set to TC0 */
1718 for (i = 1; i < TC_BITMASK + 1; i++) {
1719 int q = netdev_get_prio_tc_map(dev, i);
1721 tc = &dev->tc_to_txq[q];
1722 if (tc->offset + tc->count > txq) {
1723 pr_warning("Number of in use tx queues "
1724 "changed. Priority %i to tc "
1725 "mapping %i is no longer valid "
1726 "setting map to 0\n",
1728 netdev_set_prio_tc_map(dev, i, 0);
1734 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1735 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1737 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1741 if (txq < 1 || txq > dev->num_tx_queues)
1744 if (dev->reg_state == NETREG_REGISTERED ||
1745 dev->reg_state == NETREG_UNREGISTERING) {
1748 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1754 netif_setup_tc(dev, txq);
1756 if (txq < dev->real_num_tx_queues)
1757 qdisc_reset_all_tx_gt(dev, txq);
1760 dev->real_num_tx_queues = txq;
1763 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1767 * netif_set_real_num_rx_queues - set actual number of RX queues used
1768 * @dev: Network device
1769 * @rxq: Actual number of RX queues
1771 * This must be called either with the rtnl_lock held or before
1772 * registration of the net device. Returns 0 on success, or a
1773 * negative error code. If called before registration, it always
1776 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1780 if (rxq < 1 || rxq > dev->num_rx_queues)
1783 if (dev->reg_state == NETREG_REGISTERED) {
1786 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1792 dev->real_num_rx_queues = rxq;
1795 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1798 static inline void __netif_reschedule(struct Qdisc *q)
1800 struct softnet_data *sd;
1801 unsigned long flags;
1803 local_irq_save(flags);
1804 sd = &__get_cpu_var(softnet_data);
1805 q->next_sched = NULL;
1806 *sd->output_queue_tailp = q;
1807 sd->output_queue_tailp = &q->next_sched;
1808 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1809 local_irq_restore(flags);
1812 void __netif_schedule(struct Qdisc *q)
1814 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1815 __netif_reschedule(q);
1817 EXPORT_SYMBOL(__netif_schedule);
1819 void dev_kfree_skb_irq(struct sk_buff *skb)
1821 if (atomic_dec_and_test(&skb->users)) {
1822 struct softnet_data *sd;
1823 unsigned long flags;
1825 local_irq_save(flags);
1826 sd = &__get_cpu_var(softnet_data);
1827 skb->next = sd->completion_queue;
1828 sd->completion_queue = skb;
1829 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1830 local_irq_restore(flags);
1833 EXPORT_SYMBOL(dev_kfree_skb_irq);
1835 void dev_kfree_skb_any(struct sk_buff *skb)
1837 if (in_irq() || irqs_disabled())
1838 dev_kfree_skb_irq(skb);
1842 EXPORT_SYMBOL(dev_kfree_skb_any);
1846 * netif_device_detach - mark device as removed
1847 * @dev: network device
1849 * Mark device as removed from system and therefore no longer available.
1851 void netif_device_detach(struct net_device *dev)
1853 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1854 netif_running(dev)) {
1855 netif_tx_stop_all_queues(dev);
1858 EXPORT_SYMBOL(netif_device_detach);
1861 * netif_device_attach - mark device as attached
1862 * @dev: network device
1864 * Mark device as attached from system and restart if needed.
1866 void netif_device_attach(struct net_device *dev)
1868 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1869 netif_running(dev)) {
1870 netif_tx_wake_all_queues(dev);
1871 __netdev_watchdog_up(dev);
1874 EXPORT_SYMBOL(netif_device_attach);
1877 * Invalidate hardware checksum when packet is to be mangled, and
1878 * complete checksum manually on outgoing path.
1880 int skb_checksum_help(struct sk_buff *skb)
1883 int ret = 0, offset;
1885 if (skb->ip_summed == CHECKSUM_COMPLETE)
1886 goto out_set_summed;
1888 if (unlikely(skb_shinfo(skb)->gso_size)) {
1889 /* Let GSO fix up the checksum. */
1890 goto out_set_summed;
1893 offset = skb_checksum_start_offset(skb);
1894 BUG_ON(offset >= skb_headlen(skb));
1895 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1897 offset += skb->csum_offset;
1898 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1900 if (skb_cloned(skb) &&
1901 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1902 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1907 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1909 skb->ip_summed = CHECKSUM_NONE;
1913 EXPORT_SYMBOL(skb_checksum_help);
1916 * skb_gso_segment - Perform segmentation on skb.
1917 * @skb: buffer to segment
1918 * @features: features for the output path (see dev->features)
1920 * This function segments the given skb and returns a list of segments.
1922 * It may return NULL if the skb requires no segmentation. This is
1923 * only possible when GSO is used for verifying header integrity.
1925 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1927 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1928 struct packet_type *ptype;
1929 __be16 type = skb->protocol;
1930 int vlan_depth = ETH_HLEN;
1933 while (type == htons(ETH_P_8021Q)) {
1934 struct vlan_hdr *vh;
1936 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1937 return ERR_PTR(-EINVAL);
1939 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1940 type = vh->h_vlan_encapsulated_proto;
1941 vlan_depth += VLAN_HLEN;
1944 skb_reset_mac_header(skb);
1945 skb->mac_len = skb->network_header - skb->mac_header;
1946 __skb_pull(skb, skb->mac_len);
1948 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1949 struct net_device *dev = skb->dev;
1950 struct ethtool_drvinfo info = {};
1952 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1953 dev->ethtool_ops->get_drvinfo(dev, &info);
1955 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1956 info.driver, dev ? dev->features : 0L,
1957 skb->sk ? skb->sk->sk_route_caps : 0L,
1958 skb->len, skb->data_len, skb->ip_summed);
1960 if (skb_header_cloned(skb) &&
1961 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1962 return ERR_PTR(err);
1966 list_for_each_entry_rcu(ptype,
1967 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1968 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1969 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1970 err = ptype->gso_send_check(skb);
1971 segs = ERR_PTR(err);
1972 if (err || skb_gso_ok(skb, features))
1974 __skb_push(skb, (skb->data -
1975 skb_network_header(skb)));
1977 segs = ptype->gso_segment(skb, features);
1983 __skb_push(skb, skb->data - skb_mac_header(skb));
1987 EXPORT_SYMBOL(skb_gso_segment);
1989 /* Take action when hardware reception checksum errors are detected. */
1991 void netdev_rx_csum_fault(struct net_device *dev)
1993 if (net_ratelimit()) {
1994 printk(KERN_ERR "%s: hw csum failure.\n",
1995 dev ? dev->name : "<unknown>");
1999 EXPORT_SYMBOL(netdev_rx_csum_fault);
2002 /* Actually, we should eliminate this check as soon as we know, that:
2003 * 1. IOMMU is present and allows to map all the memory.
2004 * 2. No high memory really exists on this machine.
2007 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2009 #ifdef CONFIG_HIGHMEM
2011 if (!(dev->features & NETIF_F_HIGHDMA)) {
2012 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2013 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2014 if (PageHighMem(skb_frag_page(frag)))
2019 if (PCI_DMA_BUS_IS_PHYS) {
2020 struct device *pdev = dev->dev.parent;
2024 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2025 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2026 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2027 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2036 void (*destructor)(struct sk_buff *skb);
2039 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2041 static void dev_gso_skb_destructor(struct sk_buff *skb)
2043 struct dev_gso_cb *cb;
2046 struct sk_buff *nskb = skb->next;
2048 skb->next = nskb->next;
2051 } while (skb->next);
2053 cb = DEV_GSO_CB(skb);
2055 cb->destructor(skb);
2059 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2060 * @skb: buffer to segment
2061 * @features: device features as applicable to this skb
2063 * This function segments the given skb and stores the list of segments
2066 static int dev_gso_segment(struct sk_buff *skb, int features)
2068 struct sk_buff *segs;
2070 segs = skb_gso_segment(skb, features);
2072 /* Verifying header integrity only. */
2077 return PTR_ERR(segs);
2080 DEV_GSO_CB(skb)->destructor = skb->destructor;
2081 skb->destructor = dev_gso_skb_destructor;
2086 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2088 return ((features & NETIF_F_GEN_CSUM) ||
2089 ((features & NETIF_F_V4_CSUM) &&
2090 protocol == htons(ETH_P_IP)) ||
2091 ((features & NETIF_F_V6_CSUM) &&
2092 protocol == htons(ETH_P_IPV6)) ||
2093 ((features & NETIF_F_FCOE_CRC) &&
2094 protocol == htons(ETH_P_FCOE)));
2097 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2099 if (!can_checksum_protocol(features, protocol)) {
2100 features &= ~NETIF_F_ALL_CSUM;
2101 features &= ~NETIF_F_SG;
2102 } else if (illegal_highdma(skb->dev, skb)) {
2103 features &= ~NETIF_F_SG;
2109 u32 netif_skb_features(struct sk_buff *skb)
2111 __be16 protocol = skb->protocol;
2112 u32 features = skb->dev->features;
2114 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2115 features &= ~NETIF_F_GSO_MASK;
2117 if (protocol == htons(ETH_P_8021Q)) {
2118 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2119 protocol = veh->h_vlan_encapsulated_proto;
2120 } else if (!vlan_tx_tag_present(skb)) {
2121 return harmonize_features(skb, protocol, features);
2124 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2126 if (protocol != htons(ETH_P_8021Q)) {
2127 return harmonize_features(skb, protocol, features);
2129 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2130 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2131 return harmonize_features(skb, protocol, features);
2134 EXPORT_SYMBOL(netif_skb_features);
2137 * Returns true if either:
2138 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2139 * 2. skb is fragmented and the device does not support SG, or if
2140 * at least one of fragments is in highmem and device does not
2141 * support DMA from it.
2143 static inline int skb_needs_linearize(struct sk_buff *skb,
2146 return skb_is_nonlinear(skb) &&
2147 ((skb_has_frag_list(skb) &&
2148 !(features & NETIF_F_FRAGLIST)) ||
2149 (skb_shinfo(skb)->nr_frags &&
2150 !(features & NETIF_F_SG)));
2153 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2154 struct netdev_queue *txq)
2156 const struct net_device_ops *ops = dev->netdev_ops;
2157 int rc = NETDEV_TX_OK;
2158 unsigned int skb_len;
2160 if (likely(!skb->next)) {
2164 * If device doesn't need skb->dst, release it right now while
2165 * its hot in this cpu cache
2167 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2170 if (!list_empty(&ptype_all))
2171 dev_queue_xmit_nit(skb, dev);
2173 features = netif_skb_features(skb);
2175 if (vlan_tx_tag_present(skb) &&
2176 !(features & NETIF_F_HW_VLAN_TX)) {
2177 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2184 if (netif_needs_gso(skb, features)) {
2185 if (unlikely(dev_gso_segment(skb, features)))
2190 if (skb_needs_linearize(skb, features) &&
2191 __skb_linearize(skb))
2194 /* If packet is not checksummed and device does not
2195 * support checksumming for this protocol, complete
2196 * checksumming here.
2198 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2199 skb_set_transport_header(skb,
2200 skb_checksum_start_offset(skb));
2201 if (!(features & NETIF_F_ALL_CSUM) &&
2202 skb_checksum_help(skb))
2208 rc = ops->ndo_start_xmit(skb, dev);
2209 trace_net_dev_xmit(skb, rc, dev, skb_len);
2210 if (rc == NETDEV_TX_OK)
2211 txq_trans_update(txq);
2217 struct sk_buff *nskb = skb->next;
2219 skb->next = nskb->next;
2223 * If device doesn't need nskb->dst, release it right now while
2224 * its hot in this cpu cache
2226 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2229 skb_len = nskb->len;
2230 rc = ops->ndo_start_xmit(nskb, dev);
2231 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2232 if (unlikely(rc != NETDEV_TX_OK)) {
2233 if (rc & ~NETDEV_TX_MASK)
2234 goto out_kfree_gso_skb;
2235 nskb->next = skb->next;
2239 txq_trans_update(txq);
2240 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2241 return NETDEV_TX_BUSY;
2242 } while (skb->next);
2245 if (likely(skb->next == NULL))
2246 skb->destructor = DEV_GSO_CB(skb)->destructor;
2253 static u32 hashrnd __read_mostly;
2256 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2257 * to be used as a distribution range.
2259 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2260 unsigned int num_tx_queues)
2264 u16 qcount = num_tx_queues;
2266 if (skb_rx_queue_recorded(skb)) {
2267 hash = skb_get_rx_queue(skb);
2268 while (unlikely(hash >= num_tx_queues))
2269 hash -= num_tx_queues;
2274 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2275 qoffset = dev->tc_to_txq[tc].offset;
2276 qcount = dev->tc_to_txq[tc].count;
2279 if (skb->sk && skb->sk->sk_hash)
2280 hash = skb->sk->sk_hash;
2282 hash = (__force u16) skb->protocol;
2283 hash = jhash_1word(hash, hashrnd);
2285 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2287 EXPORT_SYMBOL(__skb_tx_hash);
2289 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2291 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2292 if (net_ratelimit()) {
2293 pr_warning("%s selects TX queue %d, but "
2294 "real number of TX queues is %d\n",
2295 dev->name, queue_index, dev->real_num_tx_queues);
2302 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2305 struct xps_dev_maps *dev_maps;
2306 struct xps_map *map;
2307 int queue_index = -1;
2310 dev_maps = rcu_dereference(dev->xps_maps);
2312 map = rcu_dereference(
2313 dev_maps->cpu_map[raw_smp_processor_id()]);
2316 queue_index = map->queues[0];
2319 if (skb->sk && skb->sk->sk_hash)
2320 hash = skb->sk->sk_hash;
2322 hash = (__force u16) skb->protocol ^
2324 hash = jhash_1word(hash, hashrnd);
2325 queue_index = map->queues[
2326 ((u64)hash * map->len) >> 32];
2328 if (unlikely(queue_index >= dev->real_num_tx_queues))
2340 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2341 struct sk_buff *skb)
2344 const struct net_device_ops *ops = dev->netdev_ops;
2346 if (dev->real_num_tx_queues == 1)
2348 else if (ops->ndo_select_queue) {
2349 queue_index = ops->ndo_select_queue(dev, skb);
2350 queue_index = dev_cap_txqueue(dev, queue_index);
2352 struct sock *sk = skb->sk;
2353 queue_index = sk_tx_queue_get(sk);
2355 if (queue_index < 0 || skb->ooo_okay ||
2356 queue_index >= dev->real_num_tx_queues) {
2357 int old_index = queue_index;
2359 queue_index = get_xps_queue(dev, skb);
2360 if (queue_index < 0)
2361 queue_index = skb_tx_hash(dev, skb);
2363 if (queue_index != old_index && sk) {
2364 struct dst_entry *dst =
2365 rcu_dereference_check(sk->sk_dst_cache, 1);
2367 if (dst && skb_dst(skb) == dst)
2368 sk_tx_queue_set(sk, queue_index);
2373 skb_set_queue_mapping(skb, queue_index);
2374 return netdev_get_tx_queue(dev, queue_index);
2377 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2378 struct net_device *dev,
2379 struct netdev_queue *txq)
2381 spinlock_t *root_lock = qdisc_lock(q);
2385 qdisc_skb_cb(skb)->pkt_len = skb->len;
2386 qdisc_calculate_pkt_len(skb, q);
2388 * Heuristic to force contended enqueues to serialize on a
2389 * separate lock before trying to get qdisc main lock.
2390 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2391 * and dequeue packets faster.
2393 contended = qdisc_is_running(q);
2394 if (unlikely(contended))
2395 spin_lock(&q->busylock);
2397 spin_lock(root_lock);
2398 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2401 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2402 qdisc_run_begin(q)) {
2404 * This is a work-conserving queue; there are no old skbs
2405 * waiting to be sent out; and the qdisc is not running -
2406 * xmit the skb directly.
2408 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2411 qdisc_bstats_update(q, skb);
2413 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2414 if (unlikely(contended)) {
2415 spin_unlock(&q->busylock);
2422 rc = NET_XMIT_SUCCESS;
2425 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2426 if (qdisc_run_begin(q)) {
2427 if (unlikely(contended)) {
2428 spin_unlock(&q->busylock);
2434 spin_unlock(root_lock);
2435 if (unlikely(contended))
2436 spin_unlock(&q->busylock);
2440 static DEFINE_PER_CPU(int, xmit_recursion);
2441 #define RECURSION_LIMIT 10
2444 * dev_queue_xmit - transmit a buffer
2445 * @skb: buffer to transmit
2447 * Queue a buffer for transmission to a network device. The caller must
2448 * have set the device and priority and built the buffer before calling
2449 * this function. The function can be called from an interrupt.
2451 * A negative errno code is returned on a failure. A success does not
2452 * guarantee the frame will be transmitted as it may be dropped due
2453 * to congestion or traffic shaping.
2455 * -----------------------------------------------------------------------------------
2456 * I notice this method can also return errors from the queue disciplines,
2457 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2460 * Regardless of the return value, the skb is consumed, so it is currently
2461 * difficult to retry a send to this method. (You can bump the ref count
2462 * before sending to hold a reference for retry if you are careful.)
2464 * When calling this method, interrupts MUST be enabled. This is because
2465 * the BH enable code must have IRQs enabled so that it will not deadlock.
2468 int dev_queue_xmit(struct sk_buff *skb)
2470 struct net_device *dev = skb->dev;
2471 struct netdev_queue *txq;
2475 /* Disable soft irqs for various locks below. Also
2476 * stops preemption for RCU.
2480 txq = dev_pick_tx(dev, skb);
2481 q = rcu_dereference_bh(txq->qdisc);
2483 #ifdef CONFIG_NET_CLS_ACT
2484 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2486 trace_net_dev_queue(skb);
2488 rc = __dev_xmit_skb(skb, q, dev, txq);
2492 /* The device has no queue. Common case for software devices:
2493 loopback, all the sorts of tunnels...
2495 Really, it is unlikely that netif_tx_lock protection is necessary
2496 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2498 However, it is possible, that they rely on protection
2501 Check this and shot the lock. It is not prone from deadlocks.
2502 Either shot noqueue qdisc, it is even simpler 8)
2504 if (dev->flags & IFF_UP) {
2505 int cpu = smp_processor_id(); /* ok because BHs are off */
2507 if (txq->xmit_lock_owner != cpu) {
2509 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2510 goto recursion_alert;
2512 HARD_TX_LOCK(dev, txq, cpu);
2514 if (!netif_tx_queue_stopped(txq)) {
2515 __this_cpu_inc(xmit_recursion);
2516 rc = dev_hard_start_xmit(skb, dev, txq);
2517 __this_cpu_dec(xmit_recursion);
2518 if (dev_xmit_complete(rc)) {
2519 HARD_TX_UNLOCK(dev, txq);
2523 HARD_TX_UNLOCK(dev, txq);
2524 if (net_ratelimit())
2525 printk(KERN_CRIT "Virtual device %s asks to "
2526 "queue packet!\n", dev->name);
2528 /* Recursion is detected! It is possible,
2532 if (net_ratelimit())
2533 printk(KERN_CRIT "Dead loop on virtual device "
2534 "%s, fix it urgently!\n", dev->name);
2539 rcu_read_unlock_bh();
2544 rcu_read_unlock_bh();
2547 EXPORT_SYMBOL(dev_queue_xmit);
2550 /*=======================================================================
2552 =======================================================================*/
2554 int netdev_max_backlog __read_mostly = 1000;
2555 int netdev_tstamp_prequeue __read_mostly = 1;
2556 int netdev_budget __read_mostly = 300;
2557 int weight_p __read_mostly = 64; /* old backlog weight */
2559 /* Called with irq disabled */
2560 static inline void ____napi_schedule(struct softnet_data *sd,
2561 struct napi_struct *napi)
2563 list_add_tail(&napi->poll_list, &sd->poll_list);
2564 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2568 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2569 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2570 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2571 * if hash is a canonical 4-tuple hash over transport ports.
2573 void __skb_get_rxhash(struct sk_buff *skb)
2575 int nhoff, hash = 0, poff;
2576 const struct ipv6hdr *ip6;
2577 const struct iphdr *ip;
2578 const struct vlan_hdr *vlan;
2587 nhoff = skb_network_offset(skb);
2588 proto = skb->protocol;
2592 case __constant_htons(ETH_P_IP):
2594 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2597 ip = (const struct iphdr *) (skb->data + nhoff);
2598 if (ip_is_fragment(ip))
2601 ip_proto = ip->protocol;
2602 addr1 = (__force u32) ip->saddr;
2603 addr2 = (__force u32) ip->daddr;
2604 nhoff += ip->ihl * 4;
2606 case __constant_htons(ETH_P_IPV6):
2608 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2611 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2612 ip_proto = ip6->nexthdr;
2613 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2614 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2617 case __constant_htons(ETH_P_8021Q):
2618 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2620 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2621 proto = vlan->h_vlan_encapsulated_proto;
2622 nhoff += sizeof(*vlan);
2624 case __constant_htons(ETH_P_PPP_SES):
2625 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2627 proto = *((__be16 *) (skb->data + nhoff +
2628 sizeof(struct pppoe_hdr)));
2629 nhoff += PPPOE_SES_HLEN;
2631 case __constant_htons(PPP_IP):
2633 case __constant_htons(PPP_IPV6):
2644 if (pskb_may_pull(skb, nhoff + 16)) {
2645 u8 *h = skb->data + nhoff;
2646 __be16 flags = *(__be16 *)h;
2649 * Only look inside GRE if version zero and no
2652 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2653 proto = *(__be16 *)(h + 2);
2655 if (flags & GRE_CSUM)
2657 if (flags & GRE_KEY)
2659 if (flags & GRE_SEQ)
2672 poff = proto_ports_offset(ip_proto);
2675 if (pskb_may_pull(skb, nhoff + 4)) {
2676 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2677 if (ports.v16[1] < ports.v16[0])
2678 swap(ports.v16[0], ports.v16[1]);
2683 /* get a consistent hash (same value on both flow directions) */
2687 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2694 EXPORT_SYMBOL(__skb_get_rxhash);
2698 /* One global table that all flow-based protocols share. */
2699 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2700 EXPORT_SYMBOL(rps_sock_flow_table);
2702 static struct rps_dev_flow *
2703 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2704 struct rps_dev_flow *rflow, u16 next_cpu)
2706 if (next_cpu != RPS_NO_CPU) {
2707 #ifdef CONFIG_RFS_ACCEL
2708 struct netdev_rx_queue *rxqueue;
2709 struct rps_dev_flow_table *flow_table;
2710 struct rps_dev_flow *old_rflow;
2715 /* Should we steer this flow to a different hardware queue? */
2716 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2717 !(dev->features & NETIF_F_NTUPLE))
2719 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2720 if (rxq_index == skb_get_rx_queue(skb))
2723 rxqueue = dev->_rx + rxq_index;
2724 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2727 flow_id = skb->rxhash & flow_table->mask;
2728 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2729 rxq_index, flow_id);
2733 rflow = &flow_table->flows[flow_id];
2735 if (old_rflow->filter == rflow->filter)
2736 old_rflow->filter = RPS_NO_FILTER;
2740 per_cpu(softnet_data, next_cpu).input_queue_head;
2743 rflow->cpu = next_cpu;
2748 * get_rps_cpu is called from netif_receive_skb and returns the target
2749 * CPU from the RPS map of the receiving queue for a given skb.
2750 * rcu_read_lock must be held on entry.
2752 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2753 struct rps_dev_flow **rflowp)
2755 struct netdev_rx_queue *rxqueue;
2756 struct rps_map *map;
2757 struct rps_dev_flow_table *flow_table;
2758 struct rps_sock_flow_table *sock_flow_table;
2762 if (skb_rx_queue_recorded(skb)) {
2763 u16 index = skb_get_rx_queue(skb);
2764 if (unlikely(index >= dev->real_num_rx_queues)) {
2765 WARN_ONCE(dev->real_num_rx_queues > 1,
2766 "%s received packet on queue %u, but number "
2767 "of RX queues is %u\n",
2768 dev->name, index, dev->real_num_rx_queues);
2771 rxqueue = dev->_rx + index;
2775 map = rcu_dereference(rxqueue->rps_map);
2777 if (map->len == 1 &&
2778 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2779 tcpu = map->cpus[0];
2780 if (cpu_online(tcpu))
2784 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2788 skb_reset_network_header(skb);
2789 if (!skb_get_rxhash(skb))
2792 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2793 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2794 if (flow_table && sock_flow_table) {
2796 struct rps_dev_flow *rflow;
2798 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2801 next_cpu = sock_flow_table->ents[skb->rxhash &
2802 sock_flow_table->mask];
2805 * If the desired CPU (where last recvmsg was done) is
2806 * different from current CPU (one in the rx-queue flow
2807 * table entry), switch if one of the following holds:
2808 * - Current CPU is unset (equal to RPS_NO_CPU).
2809 * - Current CPU is offline.
2810 * - The current CPU's queue tail has advanced beyond the
2811 * last packet that was enqueued using this table entry.
2812 * This guarantees that all previous packets for the flow
2813 * have been dequeued, thus preserving in order delivery.
2815 if (unlikely(tcpu != next_cpu) &&
2816 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2817 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2818 rflow->last_qtail)) >= 0))
2819 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2821 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2829 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2831 if (cpu_online(tcpu)) {
2841 #ifdef CONFIG_RFS_ACCEL
2844 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2845 * @dev: Device on which the filter was set
2846 * @rxq_index: RX queue index
2847 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2848 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2850 * Drivers that implement ndo_rx_flow_steer() should periodically call
2851 * this function for each installed filter and remove the filters for
2852 * which it returns %true.
2854 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2855 u32 flow_id, u16 filter_id)
2857 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2858 struct rps_dev_flow_table *flow_table;
2859 struct rps_dev_flow *rflow;
2864 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2865 if (flow_table && flow_id <= flow_table->mask) {
2866 rflow = &flow_table->flows[flow_id];
2867 cpu = ACCESS_ONCE(rflow->cpu);
2868 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2869 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2870 rflow->last_qtail) <
2871 (int)(10 * flow_table->mask)))
2877 EXPORT_SYMBOL(rps_may_expire_flow);
2879 #endif /* CONFIG_RFS_ACCEL */
2881 /* Called from hardirq (IPI) context */
2882 static void rps_trigger_softirq(void *data)
2884 struct softnet_data *sd = data;
2886 ____napi_schedule(sd, &sd->backlog);
2890 #endif /* CONFIG_RPS */
2893 * Check if this softnet_data structure is another cpu one
2894 * If yes, queue it to our IPI list and return 1
2897 static int rps_ipi_queued(struct softnet_data *sd)
2900 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2903 sd->rps_ipi_next = mysd->rps_ipi_list;
2904 mysd->rps_ipi_list = sd;
2906 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2909 #endif /* CONFIG_RPS */
2914 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2915 * queue (may be a remote CPU queue).
2917 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2918 unsigned int *qtail)
2920 struct softnet_data *sd;
2921 unsigned long flags;
2923 sd = &per_cpu(softnet_data, cpu);
2925 local_irq_save(flags);
2928 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2929 if (skb_queue_len(&sd->input_pkt_queue)) {
2931 __skb_queue_tail(&sd->input_pkt_queue, skb);
2932 input_queue_tail_incr_save(sd, qtail);
2934 local_irq_restore(flags);
2935 return NET_RX_SUCCESS;
2938 /* Schedule NAPI for backlog device
2939 * We can use non atomic operation since we own the queue lock
2941 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2942 if (!rps_ipi_queued(sd))
2943 ____napi_schedule(sd, &sd->backlog);
2951 local_irq_restore(flags);
2953 atomic_long_inc(&skb->dev->rx_dropped);
2959 * netif_rx - post buffer to the network code
2960 * @skb: buffer to post
2962 * This function receives a packet from a device driver and queues it for
2963 * the upper (protocol) levels to process. It always succeeds. The buffer
2964 * may be dropped during processing for congestion control or by the
2968 * NET_RX_SUCCESS (no congestion)
2969 * NET_RX_DROP (packet was dropped)
2973 int netif_rx(struct sk_buff *skb)
2977 /* if netpoll wants it, pretend we never saw it */
2978 if (netpoll_rx(skb))
2981 if (netdev_tstamp_prequeue)
2982 net_timestamp_check(skb);
2984 trace_netif_rx(skb);
2987 struct rps_dev_flow voidflow, *rflow = &voidflow;
2993 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2995 cpu = smp_processor_id();
2997 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3005 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3011 EXPORT_SYMBOL(netif_rx);
3013 int netif_rx_ni(struct sk_buff *skb)
3018 err = netif_rx(skb);
3019 if (local_softirq_pending())
3025 EXPORT_SYMBOL(netif_rx_ni);
3027 static void net_tx_action(struct softirq_action *h)
3029 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3031 if (sd->completion_queue) {
3032 struct sk_buff *clist;
3034 local_irq_disable();
3035 clist = sd->completion_queue;
3036 sd->completion_queue = NULL;
3040 struct sk_buff *skb = clist;
3041 clist = clist->next;
3043 WARN_ON(atomic_read(&skb->users));
3044 trace_kfree_skb(skb, net_tx_action);
3049 if (sd->output_queue) {
3052 local_irq_disable();
3053 head = sd->output_queue;
3054 sd->output_queue = NULL;
3055 sd->output_queue_tailp = &sd->output_queue;
3059 struct Qdisc *q = head;
3060 spinlock_t *root_lock;
3062 head = head->next_sched;
3064 root_lock = qdisc_lock(q);
3065 if (spin_trylock(root_lock)) {
3066 smp_mb__before_clear_bit();
3067 clear_bit(__QDISC_STATE_SCHED,
3070 spin_unlock(root_lock);
3072 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3074 __netif_reschedule(q);
3076 smp_mb__before_clear_bit();
3077 clear_bit(__QDISC_STATE_SCHED,
3085 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3086 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3087 /* This hook is defined here for ATM LANE */
3088 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3089 unsigned char *addr) __read_mostly;
3090 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3093 #ifdef CONFIG_NET_CLS_ACT
3094 /* TODO: Maybe we should just force sch_ingress to be compiled in
3095 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3096 * a compare and 2 stores extra right now if we dont have it on
3097 * but have CONFIG_NET_CLS_ACT
3098 * NOTE: This doesn't stop any functionality; if you dont have
3099 * the ingress scheduler, you just can't add policies on ingress.
3102 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3104 struct net_device *dev = skb->dev;
3105 u32 ttl = G_TC_RTTL(skb->tc_verd);
3106 int result = TC_ACT_OK;
3109 if (unlikely(MAX_RED_LOOP < ttl++)) {
3110 if (net_ratelimit())
3111 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3112 skb->skb_iif, dev->ifindex);
3116 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3117 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3120 if (q != &noop_qdisc) {
3121 spin_lock(qdisc_lock(q));
3122 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3123 result = qdisc_enqueue_root(skb, q);
3124 spin_unlock(qdisc_lock(q));
3130 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3131 struct packet_type **pt_prev,
3132 int *ret, struct net_device *orig_dev)
3134 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3136 if (!rxq || rxq->qdisc == &noop_qdisc)
3140 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3144 switch (ing_filter(skb, rxq)) {
3158 * netdev_rx_handler_register - register receive handler
3159 * @dev: device to register a handler for
3160 * @rx_handler: receive handler to register
3161 * @rx_handler_data: data pointer that is used by rx handler
3163 * Register a receive hander for a device. This handler will then be
3164 * called from __netif_receive_skb. A negative errno code is returned
3167 * The caller must hold the rtnl_mutex.
3169 * For a general description of rx_handler, see enum rx_handler_result.
3171 int netdev_rx_handler_register(struct net_device *dev,
3172 rx_handler_func_t *rx_handler,
3173 void *rx_handler_data)
3177 if (dev->rx_handler)
3180 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3181 rcu_assign_pointer(dev->rx_handler, rx_handler);
3185 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3188 * netdev_rx_handler_unregister - unregister receive handler
3189 * @dev: device to unregister a handler from
3191 * Unregister a receive hander from a device.
3193 * The caller must hold the rtnl_mutex.
3195 void netdev_rx_handler_unregister(struct net_device *dev)
3199 RCU_INIT_POINTER(dev->rx_handler, NULL);
3200 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3202 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3204 static int __netif_receive_skb(struct sk_buff *skb)
3206 struct packet_type *ptype, *pt_prev;
3207 rx_handler_func_t *rx_handler;
3208 struct net_device *orig_dev;
3209 struct net_device *null_or_dev;
3210 bool deliver_exact = false;
3211 int ret = NET_RX_DROP;
3214 if (!netdev_tstamp_prequeue)
3215 net_timestamp_check(skb);
3217 trace_netif_receive_skb(skb);
3219 /* if we've gotten here through NAPI, check netpoll */
3220 if (netpoll_receive_skb(skb))
3224 skb->skb_iif = skb->dev->ifindex;
3225 orig_dev = skb->dev;
3227 skb_reset_network_header(skb);
3228 skb_reset_transport_header(skb);
3229 skb_reset_mac_len(skb);
3237 __this_cpu_inc(softnet_data.processed);
3239 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3240 skb = vlan_untag(skb);
3245 #ifdef CONFIG_NET_CLS_ACT
3246 if (skb->tc_verd & TC_NCLS) {
3247 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3252 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3253 if (!ptype->dev || ptype->dev == skb->dev) {
3255 ret = deliver_skb(skb, pt_prev, orig_dev);
3260 #ifdef CONFIG_NET_CLS_ACT
3261 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3267 rx_handler = rcu_dereference(skb->dev->rx_handler);
3268 if (vlan_tx_tag_present(skb)) {
3270 ret = deliver_skb(skb, pt_prev, orig_dev);
3273 if (vlan_do_receive(&skb, !rx_handler))
3275 else if (unlikely(!skb))
3281 ret = deliver_skb(skb, pt_prev, orig_dev);
3284 switch (rx_handler(&skb)) {
3285 case RX_HANDLER_CONSUMED:
3287 case RX_HANDLER_ANOTHER:
3289 case RX_HANDLER_EXACT:
3290 deliver_exact = true;
3291 case RX_HANDLER_PASS:
3298 /* deliver only exact match when indicated */
3299 null_or_dev = deliver_exact ? skb->dev : NULL;
3301 type = skb->protocol;
3302 list_for_each_entry_rcu(ptype,
3303 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3304 if (ptype->type == type &&
3305 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3306 ptype->dev == orig_dev)) {
3308 ret = deliver_skb(skb, pt_prev, orig_dev);
3314 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3316 atomic_long_inc(&skb->dev->rx_dropped);
3318 /* Jamal, now you will not able to escape explaining
3319 * me how you were going to use this. :-)
3330 * netif_receive_skb - process receive buffer from network
3331 * @skb: buffer to process
3333 * netif_receive_skb() is the main receive data processing function.
3334 * It always succeeds. The buffer may be dropped during processing
3335 * for congestion control or by the protocol layers.
3337 * This function may only be called from softirq context and interrupts
3338 * should be enabled.
3340 * Return values (usually ignored):
3341 * NET_RX_SUCCESS: no congestion
3342 * NET_RX_DROP: packet was dropped
3344 int netif_receive_skb(struct sk_buff *skb)
3346 if (netdev_tstamp_prequeue)
3347 net_timestamp_check(skb);
3349 if (skb_defer_rx_timestamp(skb))
3350 return NET_RX_SUCCESS;
3354 struct rps_dev_flow voidflow, *rflow = &voidflow;
3359 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3362 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3366 ret = __netif_receive_skb(skb);
3372 return __netif_receive_skb(skb);
3375 EXPORT_SYMBOL(netif_receive_skb);
3377 /* Network device is going away, flush any packets still pending
3378 * Called with irqs disabled.
3380 static void flush_backlog(void *arg)
3382 struct net_device *dev = arg;
3383 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3384 struct sk_buff *skb, *tmp;
3387 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3388 if (skb->dev == dev) {
3389 __skb_unlink(skb, &sd->input_pkt_queue);
3391 input_queue_head_incr(sd);
3396 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3397 if (skb->dev == dev) {
3398 __skb_unlink(skb, &sd->process_queue);
3400 input_queue_head_incr(sd);
3405 static int napi_gro_complete(struct sk_buff *skb)
3407 struct packet_type *ptype;
3408 __be16 type = skb->protocol;
3409 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3412 if (NAPI_GRO_CB(skb)->count == 1) {
3413 skb_shinfo(skb)->gso_size = 0;
3418 list_for_each_entry_rcu(ptype, head, list) {
3419 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3422 err = ptype->gro_complete(skb);
3428 WARN_ON(&ptype->list == head);
3430 return NET_RX_SUCCESS;
3434 return netif_receive_skb(skb);
3437 inline void napi_gro_flush(struct napi_struct *napi)
3439 struct sk_buff *skb, *next;
3441 for (skb = napi->gro_list; skb; skb = next) {
3444 napi_gro_complete(skb);
3447 napi->gro_count = 0;
3448 napi->gro_list = NULL;
3450 EXPORT_SYMBOL(napi_gro_flush);
3452 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3454 struct sk_buff **pp = NULL;
3455 struct packet_type *ptype;
3456 __be16 type = skb->protocol;
3457 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3460 enum gro_result ret;
3462 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3465 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3469 list_for_each_entry_rcu(ptype, head, list) {
3470 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3473 skb_set_network_header(skb, skb_gro_offset(skb));
3474 mac_len = skb->network_header - skb->mac_header;
3475 skb->mac_len = mac_len;
3476 NAPI_GRO_CB(skb)->same_flow = 0;
3477 NAPI_GRO_CB(skb)->flush = 0;
3478 NAPI_GRO_CB(skb)->free = 0;
3480 pp = ptype->gro_receive(&napi->gro_list, skb);
3485 if (&ptype->list == head)
3488 same_flow = NAPI_GRO_CB(skb)->same_flow;
3489 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3492 struct sk_buff *nskb = *pp;
3496 napi_gro_complete(nskb);
3503 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3507 NAPI_GRO_CB(skb)->count = 1;
3508 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3509 skb->next = napi->gro_list;
3510 napi->gro_list = skb;
3514 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3515 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3517 BUG_ON(skb->end - skb->tail < grow);
3519 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3522 skb->data_len -= grow;
3524 skb_shinfo(skb)->frags[0].page_offset += grow;
3525 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3527 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3528 skb_frag_unref(skb, 0);
3529 memmove(skb_shinfo(skb)->frags,
3530 skb_shinfo(skb)->frags + 1,
3531 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3542 EXPORT_SYMBOL(dev_gro_receive);
3544 static inline gro_result_t
3545 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3548 unsigned int maclen = skb->dev->hard_header_len;
3550 for (p = napi->gro_list; p; p = p->next) {
3551 unsigned long diffs;
3553 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3554 diffs |= p->vlan_tci ^ skb->vlan_tci;
3555 if (maclen == ETH_HLEN)
3556 diffs |= compare_ether_header(skb_mac_header(p),
3557 skb_gro_mac_header(skb));
3559 diffs = memcmp(skb_mac_header(p),
3560 skb_gro_mac_header(skb),
3562 NAPI_GRO_CB(p)->same_flow = !diffs;
3563 NAPI_GRO_CB(p)->flush = 0;
3566 return dev_gro_receive(napi, skb);
3569 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3573 if (netif_receive_skb(skb))
3578 case GRO_MERGED_FREE:
3589 EXPORT_SYMBOL(napi_skb_finish);
3591 void skb_gro_reset_offset(struct sk_buff *skb)
3593 NAPI_GRO_CB(skb)->data_offset = 0;
3594 NAPI_GRO_CB(skb)->frag0 = NULL;
3595 NAPI_GRO_CB(skb)->frag0_len = 0;
3597 if (skb->mac_header == skb->tail &&
3598 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3599 NAPI_GRO_CB(skb)->frag0 =
3600 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3601 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3604 EXPORT_SYMBOL(skb_gro_reset_offset);
3606 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3608 skb_gro_reset_offset(skb);
3610 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3612 EXPORT_SYMBOL(napi_gro_receive);
3614 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3616 __skb_pull(skb, skb_headlen(skb));
3617 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3618 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3620 skb->dev = napi->dev;
3626 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3628 struct sk_buff *skb = napi->skb;
3631 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3637 EXPORT_SYMBOL(napi_get_frags);
3639 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3645 skb->protocol = eth_type_trans(skb, skb->dev);
3647 if (ret == GRO_HELD)
3648 skb_gro_pull(skb, -ETH_HLEN);
3649 else if (netif_receive_skb(skb))
3654 case GRO_MERGED_FREE:
3655 napi_reuse_skb(napi, skb);
3664 EXPORT_SYMBOL(napi_frags_finish);
3666 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3668 struct sk_buff *skb = napi->skb;
3675 skb_reset_mac_header(skb);
3676 skb_gro_reset_offset(skb);
3678 off = skb_gro_offset(skb);
3679 hlen = off + sizeof(*eth);
3680 eth = skb_gro_header_fast(skb, off);
3681 if (skb_gro_header_hard(skb, hlen)) {
3682 eth = skb_gro_header_slow(skb, hlen, off);
3683 if (unlikely(!eth)) {
3684 napi_reuse_skb(napi, skb);
3690 skb_gro_pull(skb, sizeof(*eth));
3693 * This works because the only protocols we care about don't require
3694 * special handling. We'll fix it up properly at the end.
3696 skb->protocol = eth->h_proto;
3701 EXPORT_SYMBOL(napi_frags_skb);
3703 gro_result_t napi_gro_frags(struct napi_struct *napi)
3705 struct sk_buff *skb = napi_frags_skb(napi);
3710 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3712 EXPORT_SYMBOL(napi_gro_frags);
3715 * net_rps_action sends any pending IPI's for rps.
3716 * Note: called with local irq disabled, but exits with local irq enabled.
3718 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3721 struct softnet_data *remsd = sd->rps_ipi_list;
3724 sd->rps_ipi_list = NULL;
3728 /* Send pending IPI's to kick RPS processing on remote cpus. */
3730 struct softnet_data *next = remsd->rps_ipi_next;
3732 if (cpu_online(remsd->cpu))
3733 __smp_call_function_single(remsd->cpu,
3742 static int process_backlog(struct napi_struct *napi, int quota)
3745 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3748 /* Check if we have pending ipi, its better to send them now,
3749 * not waiting net_rx_action() end.
3751 if (sd->rps_ipi_list) {
3752 local_irq_disable();
3753 net_rps_action_and_irq_enable(sd);
3756 napi->weight = weight_p;
3757 local_irq_disable();
3758 while (work < quota) {
3759 struct sk_buff *skb;
3762 while ((skb = __skb_dequeue(&sd->process_queue))) {
3764 __netif_receive_skb(skb);
3765 local_irq_disable();
3766 input_queue_head_incr(sd);
3767 if (++work >= quota) {
3774 qlen = skb_queue_len(&sd->input_pkt_queue);
3776 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3777 &sd->process_queue);
3779 if (qlen < quota - work) {
3781 * Inline a custom version of __napi_complete().
3782 * only current cpu owns and manipulates this napi,
3783 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3784 * we can use a plain write instead of clear_bit(),
3785 * and we dont need an smp_mb() memory barrier.
3787 list_del(&napi->poll_list);
3790 quota = work + qlen;
3800 * __napi_schedule - schedule for receive
3801 * @n: entry to schedule
3803 * The entry's receive function will be scheduled to run
3805 void __napi_schedule(struct napi_struct *n)
3807 unsigned long flags;
3809 local_irq_save(flags);
3810 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3811 local_irq_restore(flags);
3813 EXPORT_SYMBOL(__napi_schedule);
3815 void __napi_complete(struct napi_struct *n)
3817 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3818 BUG_ON(n->gro_list);
3820 list_del(&n->poll_list);
3821 smp_mb__before_clear_bit();
3822 clear_bit(NAPI_STATE_SCHED, &n->state);
3824 EXPORT_SYMBOL(__napi_complete);
3826 void napi_complete(struct napi_struct *n)
3828 unsigned long flags;
3831 * don't let napi dequeue from the cpu poll list
3832 * just in case its running on a different cpu
3834 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3838 local_irq_save(flags);
3840 local_irq_restore(flags);
3842 EXPORT_SYMBOL(napi_complete);
3844 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3845 int (*poll)(struct napi_struct *, int), int weight)
3847 INIT_LIST_HEAD(&napi->poll_list);
3848 napi->gro_count = 0;
3849 napi->gro_list = NULL;
3852 napi->weight = weight;
3853 list_add(&napi->dev_list, &dev->napi_list);
3855 #ifdef CONFIG_NETPOLL
3856 spin_lock_init(&napi->poll_lock);
3857 napi->poll_owner = -1;
3859 set_bit(NAPI_STATE_SCHED, &napi->state);
3861 EXPORT_SYMBOL(netif_napi_add);
3863 void netif_napi_del(struct napi_struct *napi)
3865 struct sk_buff *skb, *next;
3867 list_del_init(&napi->dev_list);
3868 napi_free_frags(napi);
3870 for (skb = napi->gro_list; skb; skb = next) {
3876 napi->gro_list = NULL;
3877 napi->gro_count = 0;
3879 EXPORT_SYMBOL(netif_napi_del);
3881 static void net_rx_action(struct softirq_action *h)
3883 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3884 unsigned long time_limit = jiffies + 2;
3885 int budget = netdev_budget;
3888 local_irq_disable();
3890 while (!list_empty(&sd->poll_list)) {
3891 struct napi_struct *n;
3894 /* If softirq window is exhuasted then punt.
3895 * Allow this to run for 2 jiffies since which will allow
3896 * an average latency of 1.5/HZ.
3898 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3903 /* Even though interrupts have been re-enabled, this
3904 * access is safe because interrupts can only add new
3905 * entries to the tail of this list, and only ->poll()
3906 * calls can remove this head entry from the list.
3908 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3910 have = netpoll_poll_lock(n);
3914 /* This NAPI_STATE_SCHED test is for avoiding a race
3915 * with netpoll's poll_napi(). Only the entity which
3916 * obtains the lock and sees NAPI_STATE_SCHED set will
3917 * actually make the ->poll() call. Therefore we avoid
3918 * accidentally calling ->poll() when NAPI is not scheduled.
3921 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3922 work = n->poll(n, weight);
3926 WARN_ON_ONCE(work > weight);
3930 local_irq_disable();
3932 /* Drivers must not modify the NAPI state if they
3933 * consume the entire weight. In such cases this code
3934 * still "owns" the NAPI instance and therefore can
3935 * move the instance around on the list at-will.
3937 if (unlikely(work == weight)) {
3938 if (unlikely(napi_disable_pending(n))) {
3941 local_irq_disable();
3943 list_move_tail(&n->poll_list, &sd->poll_list);
3946 netpoll_poll_unlock(have);
3949 net_rps_action_and_irq_enable(sd);
3951 #ifdef CONFIG_NET_DMA
3953 * There may not be any more sk_buffs coming right now, so push
3954 * any pending DMA copies to hardware
3956 dma_issue_pending_all();
3963 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3967 static gifconf_func_t *gifconf_list[NPROTO];
3970 * register_gifconf - register a SIOCGIF handler
3971 * @family: Address family
3972 * @gifconf: Function handler
3974 * Register protocol dependent address dumping routines. The handler
3975 * that is passed must not be freed or reused until it has been replaced
3976 * by another handler.
3978 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3980 if (family >= NPROTO)
3982 gifconf_list[family] = gifconf;
3985 EXPORT_SYMBOL(register_gifconf);
3989 * Map an interface index to its name (SIOCGIFNAME)
3993 * We need this ioctl for efficient implementation of the
3994 * if_indextoname() function required by the IPv6 API. Without
3995 * it, we would have to search all the interfaces to find a
3999 static int dev_ifname(struct net *net, struct ifreq __user *arg)
4001 struct net_device *dev;
4005 * Fetch the caller's info block.
4008 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4012 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
4018 strcpy(ifr.ifr_name, dev->name);
4021 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
4027 * Perform a SIOCGIFCONF call. This structure will change
4028 * size eventually, and there is nothing I can do about it.
4029 * Thus we will need a 'compatibility mode'.
4032 static int dev_ifconf(struct net *net, char __user *arg)
4035 struct net_device *dev;
4042 * Fetch the caller's info block.
4045 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
4052 * Loop over the interfaces, and write an info block for each.
4056 for_each_netdev(net, dev) {
4057 for (i = 0; i < NPROTO; i++) {
4058 if (gifconf_list[i]) {
4061 done = gifconf_list[i](dev, NULL, 0);
4063 done = gifconf_list[i](dev, pos + total,
4073 * All done. Write the updated control block back to the caller.
4075 ifc.ifc_len = total;
4078 * Both BSD and Solaris return 0 here, so we do too.
4080 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4083 #ifdef CONFIG_PROC_FS
4085 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4087 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4088 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4089 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4091 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4093 struct net *net = seq_file_net(seq);
4094 struct net_device *dev;
4095 struct hlist_node *p;
4096 struct hlist_head *h;
4097 unsigned int count = 0, offset = get_offset(*pos);
4099 h = &net->dev_name_head[get_bucket(*pos)];
4100 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4101 if (++count == offset)
4108 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4110 struct net_device *dev;
4111 unsigned int bucket;
4114 dev = dev_from_same_bucket(seq, pos);
4118 bucket = get_bucket(*pos) + 1;
4119 *pos = set_bucket_offset(bucket, 1);
4120 } while (bucket < NETDEV_HASHENTRIES);
4126 * This is invoked by the /proc filesystem handler to display a device
4129 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4134 return SEQ_START_TOKEN;
4136 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4139 return dev_from_bucket(seq, pos);
4142 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4145 return dev_from_bucket(seq, pos);
4148 void dev_seq_stop(struct seq_file *seq, void *v)
4154 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4156 struct rtnl_link_stats64 temp;
4157 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4159 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4160 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4161 dev->name, stats->rx_bytes, stats->rx_packets,
4163 stats->rx_dropped + stats->rx_missed_errors,
4164 stats->rx_fifo_errors,
4165 stats->rx_length_errors + stats->rx_over_errors +
4166 stats->rx_crc_errors + stats->rx_frame_errors,
4167 stats->rx_compressed, stats->multicast,
4168 stats->tx_bytes, stats->tx_packets,
4169 stats->tx_errors, stats->tx_dropped,
4170 stats->tx_fifo_errors, stats->collisions,
4171 stats->tx_carrier_errors +
4172 stats->tx_aborted_errors +
4173 stats->tx_window_errors +
4174 stats->tx_heartbeat_errors,
4175 stats->tx_compressed);
4179 * Called from the PROCfs module. This now uses the new arbitrary sized
4180 * /proc/net interface to create /proc/net/dev
4182 static int dev_seq_show(struct seq_file *seq, void *v)
4184 if (v == SEQ_START_TOKEN)
4185 seq_puts(seq, "Inter-| Receive "
4187 " face |bytes packets errs drop fifo frame "
4188 "compressed multicast|bytes packets errs "
4189 "drop fifo colls carrier compressed\n");
4191 dev_seq_printf_stats(seq, v);
4195 static struct softnet_data *softnet_get_online(loff_t *pos)
4197 struct softnet_data *sd = NULL;
4199 while (*pos < nr_cpu_ids)
4200 if (cpu_online(*pos)) {
4201 sd = &per_cpu(softnet_data, *pos);
4208 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4210 return softnet_get_online(pos);
4213 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4216 return softnet_get_online(pos);
4219 static void softnet_seq_stop(struct seq_file *seq, void *v)
4223 static int softnet_seq_show(struct seq_file *seq, void *v)
4225 struct softnet_data *sd = v;
4227 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4228 sd->processed, sd->dropped, sd->time_squeeze, 0,
4229 0, 0, 0, 0, /* was fastroute */
4230 sd->cpu_collision, sd->received_rps);
4234 static const struct seq_operations dev_seq_ops = {
4235 .start = dev_seq_start,
4236 .next = dev_seq_next,
4237 .stop = dev_seq_stop,
4238 .show = dev_seq_show,
4241 static int dev_seq_open(struct inode *inode, struct file *file)
4243 return seq_open_net(inode, file, &dev_seq_ops,
4244 sizeof(struct seq_net_private));
4247 static const struct file_operations dev_seq_fops = {
4248 .owner = THIS_MODULE,
4249 .open = dev_seq_open,
4251 .llseek = seq_lseek,
4252 .release = seq_release_net,
4255 static const struct seq_operations softnet_seq_ops = {
4256 .start = softnet_seq_start,
4257 .next = softnet_seq_next,
4258 .stop = softnet_seq_stop,
4259 .show = softnet_seq_show,
4262 static int softnet_seq_open(struct inode *inode, struct file *file)
4264 return seq_open(file, &softnet_seq_ops);
4267 static const struct file_operations softnet_seq_fops = {
4268 .owner = THIS_MODULE,
4269 .open = softnet_seq_open,
4271 .llseek = seq_lseek,
4272 .release = seq_release,
4275 static void *ptype_get_idx(loff_t pos)
4277 struct packet_type *pt = NULL;
4281 list_for_each_entry_rcu(pt, &ptype_all, list) {
4287 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4288 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4297 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4301 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4304 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4306 struct packet_type *pt;
4307 struct list_head *nxt;
4311 if (v == SEQ_START_TOKEN)
4312 return ptype_get_idx(0);
4315 nxt = pt->list.next;
4316 if (pt->type == htons(ETH_P_ALL)) {
4317 if (nxt != &ptype_all)
4320 nxt = ptype_base[0].next;
4322 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4324 while (nxt == &ptype_base[hash]) {
4325 if (++hash >= PTYPE_HASH_SIZE)
4327 nxt = ptype_base[hash].next;
4330 return list_entry(nxt, struct packet_type, list);
4333 static void ptype_seq_stop(struct seq_file *seq, void *v)
4339 static int ptype_seq_show(struct seq_file *seq, void *v)
4341 struct packet_type *pt = v;
4343 if (v == SEQ_START_TOKEN)
4344 seq_puts(seq, "Type Device Function\n");
4345 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4346 if (pt->type == htons(ETH_P_ALL))
4347 seq_puts(seq, "ALL ");
4349 seq_printf(seq, "%04x", ntohs(pt->type));
4351 seq_printf(seq, " %-8s %pF\n",
4352 pt->dev ? pt->dev->name : "", pt->func);
4358 static const struct seq_operations ptype_seq_ops = {
4359 .start = ptype_seq_start,
4360 .next = ptype_seq_next,
4361 .stop = ptype_seq_stop,
4362 .show = ptype_seq_show,
4365 static int ptype_seq_open(struct inode *inode, struct file *file)
4367 return seq_open_net(inode, file, &ptype_seq_ops,
4368 sizeof(struct seq_net_private));
4371 static const struct file_operations ptype_seq_fops = {
4372 .owner = THIS_MODULE,
4373 .open = ptype_seq_open,
4375 .llseek = seq_lseek,
4376 .release = seq_release_net,
4380 static int __net_init dev_proc_net_init(struct net *net)
4384 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4386 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4388 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4391 if (wext_proc_init(net))
4397 proc_net_remove(net, "ptype");
4399 proc_net_remove(net, "softnet_stat");
4401 proc_net_remove(net, "dev");
4405 static void __net_exit dev_proc_net_exit(struct net *net)
4407 wext_proc_exit(net);
4409 proc_net_remove(net, "ptype");
4410 proc_net_remove(net, "softnet_stat");
4411 proc_net_remove(net, "dev");
4414 static struct pernet_operations __net_initdata dev_proc_ops = {
4415 .init = dev_proc_net_init,
4416 .exit = dev_proc_net_exit,
4419 static int __init dev_proc_init(void)
4421 return register_pernet_subsys(&dev_proc_ops);
4424 #define dev_proc_init() 0
4425 #endif /* CONFIG_PROC_FS */
4429 * netdev_set_master - set up master pointer
4430 * @slave: slave device
4431 * @master: new master device
4433 * Changes the master device of the slave. Pass %NULL to break the
4434 * bonding. The caller must hold the RTNL semaphore. On a failure
4435 * a negative errno code is returned. On success the reference counts
4436 * are adjusted and the function returns zero.
4438 int netdev_set_master(struct net_device *slave, struct net_device *master)
4440 struct net_device *old = slave->master;
4450 slave->master = master;
4456 EXPORT_SYMBOL(netdev_set_master);
4459 * netdev_set_bond_master - set up bonding master/slave pair
4460 * @slave: slave device
4461 * @master: new master device
4463 * Changes the master device of the slave. Pass %NULL to break the
4464 * bonding. The caller must hold the RTNL semaphore. On a failure
4465 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4466 * to the routing socket and the function returns zero.
4468 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4474 err = netdev_set_master(slave, master);
4478 slave->flags |= IFF_SLAVE;
4480 slave->flags &= ~IFF_SLAVE;
4482 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4485 EXPORT_SYMBOL(netdev_set_bond_master);
4487 static void dev_change_rx_flags(struct net_device *dev, int flags)
4489 const struct net_device_ops *ops = dev->netdev_ops;
4491 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4492 ops->ndo_change_rx_flags(dev, flags);
4495 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4497 unsigned short old_flags = dev->flags;
4503 dev->flags |= IFF_PROMISC;
4504 dev->promiscuity += inc;
4505 if (dev->promiscuity == 0) {
4508 * If inc causes overflow, untouch promisc and return error.
4511 dev->flags &= ~IFF_PROMISC;
4513 dev->promiscuity -= inc;
4514 printk(KERN_WARNING "%s: promiscuity touches roof, "
4515 "set promiscuity failed, promiscuity feature "
4516 "of device might be broken.\n", dev->name);
4520 if (dev->flags != old_flags) {
4521 printk(KERN_INFO "device %s %s promiscuous mode\n",
4522 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4524 if (audit_enabled) {
4525 current_uid_gid(&uid, &gid);
4526 audit_log(current->audit_context, GFP_ATOMIC,
4527 AUDIT_ANOM_PROMISCUOUS,
4528 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4529 dev->name, (dev->flags & IFF_PROMISC),
4530 (old_flags & IFF_PROMISC),
4531 audit_get_loginuid(current),
4533 audit_get_sessionid(current));
4536 dev_change_rx_flags(dev, IFF_PROMISC);
4542 * dev_set_promiscuity - update promiscuity count on a device
4546 * Add or remove promiscuity from a device. While the count in the device
4547 * remains above zero the interface remains promiscuous. Once it hits zero
4548 * the device reverts back to normal filtering operation. A negative inc
4549 * value is used to drop promiscuity on the device.
4550 * Return 0 if successful or a negative errno code on error.
4552 int dev_set_promiscuity(struct net_device *dev, int inc)
4554 unsigned short old_flags = dev->flags;
4557 err = __dev_set_promiscuity(dev, inc);
4560 if (dev->flags != old_flags)
4561 dev_set_rx_mode(dev);
4564 EXPORT_SYMBOL(dev_set_promiscuity);
4567 * dev_set_allmulti - update allmulti count on a device
4571 * Add or remove reception of all multicast frames to a device. While the
4572 * count in the device remains above zero the interface remains listening
4573 * to all interfaces. Once it hits zero the device reverts back to normal
4574 * filtering operation. A negative @inc value is used to drop the counter
4575 * when releasing a resource needing all multicasts.
4576 * Return 0 if successful or a negative errno code on error.
4579 int dev_set_allmulti(struct net_device *dev, int inc)
4581 unsigned short old_flags = dev->flags;
4585 dev->flags |= IFF_ALLMULTI;
4586 dev->allmulti += inc;
4587 if (dev->allmulti == 0) {
4590 * If inc causes overflow, untouch allmulti and return error.
4593 dev->flags &= ~IFF_ALLMULTI;
4595 dev->allmulti -= inc;
4596 printk(KERN_WARNING "%s: allmulti touches roof, "
4597 "set allmulti failed, allmulti feature of "
4598 "device might be broken.\n", dev->name);
4602 if (dev->flags ^ old_flags) {
4603 dev_change_rx_flags(dev, IFF_ALLMULTI);
4604 dev_set_rx_mode(dev);
4608 EXPORT_SYMBOL(dev_set_allmulti);
4611 * Upload unicast and multicast address lists to device and
4612 * configure RX filtering. When the device doesn't support unicast
4613 * filtering it is put in promiscuous mode while unicast addresses
4616 void __dev_set_rx_mode(struct net_device *dev)
4618 const struct net_device_ops *ops = dev->netdev_ops;
4620 /* dev_open will call this function so the list will stay sane. */
4621 if (!(dev->flags&IFF_UP))
4624 if (!netif_device_present(dev))
4627 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4628 /* Unicast addresses changes may only happen under the rtnl,
4629 * therefore calling __dev_set_promiscuity here is safe.
4631 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4632 __dev_set_promiscuity(dev, 1);
4633 dev->uc_promisc = true;
4634 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4635 __dev_set_promiscuity(dev, -1);
4636 dev->uc_promisc = false;
4640 if (ops->ndo_set_rx_mode)
4641 ops->ndo_set_rx_mode(dev);
4644 void dev_set_rx_mode(struct net_device *dev)
4646 netif_addr_lock_bh(dev);
4647 __dev_set_rx_mode(dev);
4648 netif_addr_unlock_bh(dev);
4652 * dev_get_flags - get flags reported to userspace
4655 * Get the combination of flag bits exported through APIs to userspace.
4657 unsigned dev_get_flags(const struct net_device *dev)
4661 flags = (dev->flags & ~(IFF_PROMISC |
4666 (dev->gflags & (IFF_PROMISC |
4669 if (netif_running(dev)) {
4670 if (netif_oper_up(dev))
4671 flags |= IFF_RUNNING;
4672 if (netif_carrier_ok(dev))
4673 flags |= IFF_LOWER_UP;
4674 if (netif_dormant(dev))
4675 flags |= IFF_DORMANT;
4680 EXPORT_SYMBOL(dev_get_flags);
4682 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4684 int old_flags = dev->flags;
4690 * Set the flags on our device.
4693 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4694 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4696 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4700 * Load in the correct multicast list now the flags have changed.
4703 if ((old_flags ^ flags) & IFF_MULTICAST)
4704 dev_change_rx_flags(dev, IFF_MULTICAST);
4706 dev_set_rx_mode(dev);
4709 * Have we downed the interface. We handle IFF_UP ourselves
4710 * according to user attempts to set it, rather than blindly
4715 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4716 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4719 dev_set_rx_mode(dev);
4722 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4723 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4725 dev->gflags ^= IFF_PROMISC;
4726 dev_set_promiscuity(dev, inc);
4729 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4730 is important. Some (broken) drivers set IFF_PROMISC, when
4731 IFF_ALLMULTI is requested not asking us and not reporting.
4733 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4734 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4736 dev->gflags ^= IFF_ALLMULTI;
4737 dev_set_allmulti(dev, inc);
4743 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4745 unsigned int changes = dev->flags ^ old_flags;
4747 if (changes & IFF_UP) {
4748 if (dev->flags & IFF_UP)
4749 call_netdevice_notifiers(NETDEV_UP, dev);
4751 call_netdevice_notifiers(NETDEV_DOWN, dev);
4754 if (dev->flags & IFF_UP &&
4755 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4756 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4760 * dev_change_flags - change device settings
4762 * @flags: device state flags
4764 * Change settings on device based state flags. The flags are
4765 * in the userspace exported format.
4767 int dev_change_flags(struct net_device *dev, unsigned flags)
4770 int old_flags = dev->flags;
4772 ret = __dev_change_flags(dev, flags);
4776 changes = old_flags ^ dev->flags;
4778 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4780 __dev_notify_flags(dev, old_flags);
4783 EXPORT_SYMBOL(dev_change_flags);
4786 * dev_set_mtu - Change maximum transfer unit
4788 * @new_mtu: new transfer unit
4790 * Change the maximum transfer size of the network device.
4792 int dev_set_mtu(struct net_device *dev, int new_mtu)
4794 const struct net_device_ops *ops = dev->netdev_ops;
4797 if (new_mtu == dev->mtu)
4800 /* MTU must be positive. */
4804 if (!netif_device_present(dev))
4808 if (ops->ndo_change_mtu)
4809 err = ops->ndo_change_mtu(dev, new_mtu);
4813 if (!err && dev->flags & IFF_UP)
4814 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4817 EXPORT_SYMBOL(dev_set_mtu);
4820 * dev_set_group - Change group this device belongs to
4822 * @new_group: group this device should belong to
4824 void dev_set_group(struct net_device *dev, int new_group)
4826 dev->group = new_group;
4828 EXPORT_SYMBOL(dev_set_group);
4831 * dev_set_mac_address - Change Media Access Control Address
4835 * Change the hardware (MAC) address of the device
4837 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4839 const struct net_device_ops *ops = dev->netdev_ops;
4842 if (!ops->ndo_set_mac_address)
4844 if (sa->sa_family != dev->type)
4846 if (!netif_device_present(dev))
4848 err = ops->ndo_set_mac_address(dev, sa);
4850 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4851 add_device_randomness(dev->dev_addr, dev->addr_len);
4854 EXPORT_SYMBOL(dev_set_mac_address);
4857 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4859 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4862 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4868 case SIOCGIFFLAGS: /* Get interface flags */
4869 ifr->ifr_flags = (short) dev_get_flags(dev);
4872 case SIOCGIFMETRIC: /* Get the metric on the interface
4873 (currently unused) */
4874 ifr->ifr_metric = 0;
4877 case SIOCGIFMTU: /* Get the MTU of a device */
4878 ifr->ifr_mtu = dev->mtu;
4883 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4885 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4886 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4887 ifr->ifr_hwaddr.sa_family = dev->type;
4895 ifr->ifr_map.mem_start = dev->mem_start;
4896 ifr->ifr_map.mem_end = dev->mem_end;
4897 ifr->ifr_map.base_addr = dev->base_addr;
4898 ifr->ifr_map.irq = dev->irq;
4899 ifr->ifr_map.dma = dev->dma;
4900 ifr->ifr_map.port = dev->if_port;
4904 ifr->ifr_ifindex = dev->ifindex;
4908 ifr->ifr_qlen = dev->tx_queue_len;
4912 /* dev_ioctl() should ensure this case
4924 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4926 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4929 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4930 const struct net_device_ops *ops;
4935 ops = dev->netdev_ops;
4938 case SIOCSIFFLAGS: /* Set interface flags */
4939 return dev_change_flags(dev, ifr->ifr_flags);
4941 case SIOCSIFMETRIC: /* Set the metric on the interface
4942 (currently unused) */
4945 case SIOCSIFMTU: /* Set the MTU of a device */
4946 return dev_set_mtu(dev, ifr->ifr_mtu);
4949 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4951 case SIOCSIFHWBROADCAST:
4952 if (ifr->ifr_hwaddr.sa_family != dev->type)
4954 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4955 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4956 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4960 if (ops->ndo_set_config) {
4961 if (!netif_device_present(dev))
4963 return ops->ndo_set_config(dev, &ifr->ifr_map);
4968 if (!ops->ndo_set_rx_mode ||
4969 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4971 if (!netif_device_present(dev))
4973 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4976 if (!ops->ndo_set_rx_mode ||
4977 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4979 if (!netif_device_present(dev))
4981 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4984 if (ifr->ifr_qlen < 0)
4986 dev->tx_queue_len = ifr->ifr_qlen;
4990 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4991 return dev_change_name(dev, ifr->ifr_newname);
4994 err = net_hwtstamp_validate(ifr);
5000 * Unknown or private ioctl
5003 if ((cmd >= SIOCDEVPRIVATE &&
5004 cmd <= SIOCDEVPRIVATE + 15) ||
5005 cmd == SIOCBONDENSLAVE ||
5006 cmd == SIOCBONDRELEASE ||
5007 cmd == SIOCBONDSETHWADDR ||
5008 cmd == SIOCBONDSLAVEINFOQUERY ||
5009 cmd == SIOCBONDINFOQUERY ||
5010 cmd == SIOCBONDCHANGEACTIVE ||
5011 cmd == SIOCGMIIPHY ||
5012 cmd == SIOCGMIIREG ||
5013 cmd == SIOCSMIIREG ||
5014 cmd == SIOCBRADDIF ||
5015 cmd == SIOCBRDELIF ||
5016 cmd == SIOCSHWTSTAMP ||
5017 cmd == SIOCWANDEV) {
5019 if (ops->ndo_do_ioctl) {
5020 if (netif_device_present(dev))
5021 err = ops->ndo_do_ioctl(dev, ifr, cmd);
5033 * This function handles all "interface"-type I/O control requests. The actual
5034 * 'doing' part of this is dev_ifsioc above.
5038 * dev_ioctl - network device ioctl
5039 * @net: the applicable net namespace
5040 * @cmd: command to issue
5041 * @arg: pointer to a struct ifreq in user space
5043 * Issue ioctl functions to devices. This is normally called by the
5044 * user space syscall interfaces but can sometimes be useful for
5045 * other purposes. The return value is the return from the syscall if
5046 * positive or a negative errno code on error.
5049 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
5055 /* One special case: SIOCGIFCONF takes ifconf argument
5056 and requires shared lock, because it sleeps writing
5060 if (cmd == SIOCGIFCONF) {
5062 ret = dev_ifconf(net, (char __user *) arg);
5066 if (cmd == SIOCGIFNAME)
5067 return dev_ifname(net, (struct ifreq __user *)arg);
5069 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5072 ifr.ifr_name[IFNAMSIZ-1] = 0;
5074 colon = strchr(ifr.ifr_name, ':');
5079 * See which interface the caller is talking about.
5084 * These ioctl calls:
5085 * - can be done by all.
5086 * - atomic and do not require locking.
5097 dev_load(net, ifr.ifr_name);
5099 ret = dev_ifsioc_locked(net, &ifr, cmd);
5104 if (copy_to_user(arg, &ifr,
5105 sizeof(struct ifreq)))
5111 dev_load(net, ifr.ifr_name);
5113 ret = dev_ethtool(net, &ifr);
5118 if (copy_to_user(arg, &ifr,
5119 sizeof(struct ifreq)))
5125 * These ioctl calls:
5126 * - require superuser power.
5127 * - require strict serialization.
5133 if (!capable(CAP_NET_ADMIN))
5135 dev_load(net, ifr.ifr_name);
5137 ret = dev_ifsioc(net, &ifr, cmd);
5142 if (copy_to_user(arg, &ifr,
5143 sizeof(struct ifreq)))
5149 * These ioctl calls:
5150 * - require superuser power.
5151 * - require strict serialization.
5152 * - do not return a value
5162 case SIOCSIFHWBROADCAST:
5165 case SIOCBONDENSLAVE:
5166 case SIOCBONDRELEASE:
5167 case SIOCBONDSETHWADDR:
5168 case SIOCBONDCHANGEACTIVE:
5172 if (!capable(CAP_NET_ADMIN))
5175 case SIOCBONDSLAVEINFOQUERY:
5176 case SIOCBONDINFOQUERY:
5177 dev_load(net, ifr.ifr_name);
5179 ret = dev_ifsioc(net, &ifr, cmd);
5184 /* Get the per device memory space. We can add this but
5185 * currently do not support it */
5187 /* Set the per device memory buffer space.
5188 * Not applicable in our case */
5193 * Unknown or private ioctl.
5196 if (cmd == SIOCWANDEV ||
5197 (cmd >= SIOCDEVPRIVATE &&
5198 cmd <= SIOCDEVPRIVATE + 15)) {
5199 dev_load(net, ifr.ifr_name);
5201 ret = dev_ifsioc(net, &ifr, cmd);
5203 if (!ret && copy_to_user(arg, &ifr,
5204 sizeof(struct ifreq)))
5208 /* Take care of Wireless Extensions */
5209 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5210 return wext_handle_ioctl(net, &ifr, cmd, arg);
5217 * dev_new_index - allocate an ifindex
5218 * @net: the applicable net namespace
5220 * Returns a suitable unique value for a new device interface
5221 * number. The caller must hold the rtnl semaphore or the
5222 * dev_base_lock to be sure it remains unique.
5224 static int dev_new_index(struct net *net)
5230 if (!__dev_get_by_index(net, ifindex))
5235 /* Delayed registration/unregisteration */
5236 static LIST_HEAD(net_todo_list);
5238 static void net_set_todo(struct net_device *dev)
5240 list_add_tail(&dev->todo_list, &net_todo_list);
5243 static void rollback_registered_many(struct list_head *head)
5245 struct net_device *dev, *tmp;
5247 BUG_ON(dev_boot_phase);
5250 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5251 /* Some devices call without registering
5252 * for initialization unwind. Remove those
5253 * devices and proceed with the remaining.
5255 if (dev->reg_state == NETREG_UNINITIALIZED) {
5256 pr_debug("unregister_netdevice: device %s/%p never "
5257 "was registered\n", dev->name, dev);
5260 list_del(&dev->unreg_list);
5263 dev->dismantle = true;
5264 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5267 /* If device is running, close it first. */
5268 dev_close_many(head);
5270 list_for_each_entry(dev, head, unreg_list) {
5271 /* And unlink it from device chain. */
5272 unlist_netdevice(dev);
5274 dev->reg_state = NETREG_UNREGISTERING;
5279 list_for_each_entry(dev, head, unreg_list) {
5280 /* Shutdown queueing discipline. */
5284 /* Notify protocols, that we are about to destroy
5285 this device. They should clean all the things.
5287 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5289 if (!dev->rtnl_link_ops ||
5290 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5291 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5294 * Flush the unicast and multicast chains
5299 if (dev->netdev_ops->ndo_uninit)
5300 dev->netdev_ops->ndo_uninit(dev);
5302 /* Notifier chain MUST detach us from master device. */
5303 WARN_ON(dev->master);
5305 /* Remove entries from kobject tree */
5306 netdev_unregister_kobject(dev);
5309 /* Process any work delayed until the end of the batch */
5310 dev = list_first_entry(head, struct net_device, unreg_list);
5311 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5315 list_for_each_entry(dev, head, unreg_list)
5319 static void rollback_registered(struct net_device *dev)
5323 list_add(&dev->unreg_list, &single);
5324 rollback_registered_many(&single);
5328 static u32 netdev_fix_features(struct net_device *dev, u32 features)
5330 /* Fix illegal checksum combinations */
5331 if ((features & NETIF_F_HW_CSUM) &&
5332 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5333 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5334 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5337 if ((features & NETIF_F_NO_CSUM) &&
5338 (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5339 netdev_warn(dev, "mixed no checksumming and other settings.\n");
5340 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5343 /* Fix illegal SG+CSUM combinations. */
5344 if ((features & NETIF_F_SG) &&
5345 !(features & NETIF_F_ALL_CSUM)) {
5347 "Dropping NETIF_F_SG since no checksum feature.\n");
5348 features &= ~NETIF_F_SG;
5351 /* TSO requires that SG is present as well. */
5352 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5353 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5354 features &= ~NETIF_F_ALL_TSO;
5357 /* TSO ECN requires that TSO is present as well. */
5358 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5359 features &= ~NETIF_F_TSO_ECN;
5361 /* Software GSO depends on SG. */
5362 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5363 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5364 features &= ~NETIF_F_GSO;
5367 /* UFO needs SG and checksumming */
5368 if (features & NETIF_F_UFO) {
5369 /* maybe split UFO into V4 and V6? */
5370 if (!((features & NETIF_F_GEN_CSUM) ||
5371 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5372 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5374 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5375 features &= ~NETIF_F_UFO;
5378 if (!(features & NETIF_F_SG)) {
5380 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5381 features &= ~NETIF_F_UFO;
5388 int __netdev_update_features(struct net_device *dev)
5395 features = netdev_get_wanted_features(dev);
5397 if (dev->netdev_ops->ndo_fix_features)
5398 features = dev->netdev_ops->ndo_fix_features(dev, features);
5400 /* driver might be less strict about feature dependencies */
5401 features = netdev_fix_features(dev, features);
5403 if (dev->features == features)
5406 netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5407 dev->features, features);
5409 if (dev->netdev_ops->ndo_set_features)
5410 err = dev->netdev_ops->ndo_set_features(dev, features);
5412 if (unlikely(err < 0)) {
5414 "set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5415 err, features, dev->features);
5420 dev->features = features;
5426 * netdev_update_features - recalculate device features
5427 * @dev: the device to check
5429 * Recalculate dev->features set and send notifications if it
5430 * has changed. Should be called after driver or hardware dependent
5431 * conditions might have changed that influence the features.
5433 void netdev_update_features(struct net_device *dev)
5435 if (__netdev_update_features(dev))
5436 netdev_features_change(dev);
5438 EXPORT_SYMBOL(netdev_update_features);
5441 * netdev_change_features - recalculate device features
5442 * @dev: the device to check
5444 * Recalculate dev->features set and send notifications even
5445 * if they have not changed. Should be called instead of
5446 * netdev_update_features() if also dev->vlan_features might
5447 * have changed to allow the changes to be propagated to stacked
5450 void netdev_change_features(struct net_device *dev)
5452 __netdev_update_features(dev);
5453 netdev_features_change(dev);
5455 EXPORT_SYMBOL(netdev_change_features);
5458 * netif_stacked_transfer_operstate - transfer operstate
5459 * @rootdev: the root or lower level device to transfer state from
5460 * @dev: the device to transfer operstate to
5462 * Transfer operational state from root to device. This is normally
5463 * called when a stacking relationship exists between the root
5464 * device and the device(a leaf device).
5466 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5467 struct net_device *dev)
5469 if (rootdev->operstate == IF_OPER_DORMANT)
5470 netif_dormant_on(dev);
5472 netif_dormant_off(dev);
5474 if (netif_carrier_ok(rootdev)) {
5475 if (!netif_carrier_ok(dev))
5476 netif_carrier_on(dev);
5478 if (netif_carrier_ok(dev))
5479 netif_carrier_off(dev);
5482 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5485 static int netif_alloc_rx_queues(struct net_device *dev)
5487 unsigned int i, count = dev->num_rx_queues;
5488 struct netdev_rx_queue *rx;
5492 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5494 pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5499 for (i = 0; i < count; i++)
5505 static void netdev_init_one_queue(struct net_device *dev,
5506 struct netdev_queue *queue, void *_unused)
5508 /* Initialize queue lock */
5509 spin_lock_init(&queue->_xmit_lock);
5510 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5511 queue->xmit_lock_owner = -1;
5512 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5516 static int netif_alloc_netdev_queues(struct net_device *dev)
5518 unsigned int count = dev->num_tx_queues;
5519 struct netdev_queue *tx;
5523 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5525 pr_err("netdev: Unable to allocate %u tx queues.\n",
5531 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5532 spin_lock_init(&dev->tx_global_lock);
5538 * register_netdevice - register a network device
5539 * @dev: device to register
5541 * Take a completed network device structure and add it to the kernel
5542 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5543 * chain. 0 is returned on success. A negative errno code is returned
5544 * on a failure to set up the device, or if the name is a duplicate.
5546 * Callers must hold the rtnl semaphore. You may want
5547 * register_netdev() instead of this.
5550 * The locking appears insufficient to guarantee two parallel registers
5551 * will not get the same name.
5554 int register_netdevice(struct net_device *dev)
5557 struct net *net = dev_net(dev);
5559 BUG_ON(dev_boot_phase);
5564 /* When net_device's are persistent, this will be fatal. */
5565 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5568 spin_lock_init(&dev->addr_list_lock);
5569 netdev_set_addr_lockdep_class(dev);
5573 ret = dev_get_valid_name(dev, dev->name);
5577 /* Init, if this function is available */
5578 if (dev->netdev_ops->ndo_init) {
5579 ret = dev->netdev_ops->ndo_init(dev);
5587 dev->ifindex = dev_new_index(net);
5588 if (dev->iflink == -1)
5589 dev->iflink = dev->ifindex;
5591 /* Transfer changeable features to wanted_features and enable
5592 * software offloads (GSO and GRO).
5594 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5595 dev->features |= NETIF_F_SOFT_FEATURES;
5596 dev->wanted_features = dev->features & dev->hw_features;
5598 /* Turn on no cache copy if HW is doing checksum */
5599 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5600 if ((dev->features & NETIF_F_ALL_CSUM) &&
5601 !(dev->features & NETIF_F_NO_CSUM)) {
5602 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5603 dev->features |= NETIF_F_NOCACHE_COPY;
5606 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5608 dev->vlan_features |= NETIF_F_HIGHDMA;
5610 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5611 ret = notifier_to_errno(ret);
5615 ret = netdev_register_kobject(dev);
5618 dev->reg_state = NETREG_REGISTERED;
5620 __netdev_update_features(dev);
5623 * Default initial state at registry is that the
5624 * device is present.
5627 set_bit(__LINK_STATE_PRESENT, &dev->state);
5629 dev_init_scheduler(dev);
5631 list_netdevice(dev);
5632 add_device_randomness(dev->dev_addr, dev->addr_len);
5634 /* Notify protocols, that a new device appeared. */
5635 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5636 ret = notifier_to_errno(ret);
5638 rollback_registered(dev);
5639 dev->reg_state = NETREG_UNREGISTERED;
5642 * Prevent userspace races by waiting until the network
5643 * device is fully setup before sending notifications.
5645 if (!dev->rtnl_link_ops ||
5646 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5647 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5653 if (dev->netdev_ops->ndo_uninit)
5654 dev->netdev_ops->ndo_uninit(dev);
5657 EXPORT_SYMBOL(register_netdevice);
5660 * init_dummy_netdev - init a dummy network device for NAPI
5661 * @dev: device to init
5663 * This takes a network device structure and initialize the minimum
5664 * amount of fields so it can be used to schedule NAPI polls without
5665 * registering a full blown interface. This is to be used by drivers
5666 * that need to tie several hardware interfaces to a single NAPI
5667 * poll scheduler due to HW limitations.
5669 int init_dummy_netdev(struct net_device *dev)
5671 /* Clear everything. Note we don't initialize spinlocks
5672 * are they aren't supposed to be taken by any of the
5673 * NAPI code and this dummy netdev is supposed to be
5674 * only ever used for NAPI polls
5676 memset(dev, 0, sizeof(struct net_device));
5678 /* make sure we BUG if trying to hit standard
5679 * register/unregister code path
5681 dev->reg_state = NETREG_DUMMY;
5683 /* NAPI wants this */
5684 INIT_LIST_HEAD(&dev->napi_list);
5686 /* a dummy interface is started by default */
5687 set_bit(__LINK_STATE_PRESENT, &dev->state);
5688 set_bit(__LINK_STATE_START, &dev->state);
5690 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5691 * because users of this 'device' dont need to change
5697 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5701 * register_netdev - register a network device
5702 * @dev: device to register
5704 * Take a completed network device structure and add it to the kernel
5705 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5706 * chain. 0 is returned on success. A negative errno code is returned
5707 * on a failure to set up the device, or if the name is a duplicate.
5709 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5710 * and expands the device name if you passed a format string to
5713 int register_netdev(struct net_device *dev)
5718 err = register_netdevice(dev);
5722 EXPORT_SYMBOL(register_netdev);
5724 int netdev_refcnt_read(const struct net_device *dev)
5728 for_each_possible_cpu(i)
5729 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5732 EXPORT_SYMBOL(netdev_refcnt_read);
5735 * netdev_wait_allrefs - wait until all references are gone.
5737 * This is called when unregistering network devices.
5739 * Any protocol or device that holds a reference should register
5740 * for netdevice notification, and cleanup and put back the
5741 * reference if they receive an UNREGISTER event.
5742 * We can get stuck here if buggy protocols don't correctly
5745 static void netdev_wait_allrefs(struct net_device *dev)
5747 unsigned long rebroadcast_time, warning_time;
5750 linkwatch_forget_dev(dev);
5752 rebroadcast_time = warning_time = jiffies;
5753 refcnt = netdev_refcnt_read(dev);
5755 while (refcnt != 0) {
5756 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5759 /* Rebroadcast unregister notification */
5760 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5761 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5762 * should have already handle it the first time */
5764 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5766 /* We must not have linkwatch events
5767 * pending on unregister. If this
5768 * happens, we simply run the queue
5769 * unscheduled, resulting in a noop
5772 linkwatch_run_queue();
5777 rebroadcast_time = jiffies;
5782 refcnt = netdev_refcnt_read(dev);
5784 if (time_after(jiffies, warning_time + 10 * HZ)) {
5785 printk(KERN_EMERG "unregister_netdevice: "
5786 "waiting for %s to become free. Usage "
5789 warning_time = jiffies;
5798 * register_netdevice(x1);
5799 * register_netdevice(x2);
5801 * unregister_netdevice(y1);
5802 * unregister_netdevice(y2);
5808 * We are invoked by rtnl_unlock().
5809 * This allows us to deal with problems:
5810 * 1) We can delete sysfs objects which invoke hotplug
5811 * without deadlocking with linkwatch via keventd.
5812 * 2) Since we run with the RTNL semaphore not held, we can sleep
5813 * safely in order to wait for the netdev refcnt to drop to zero.
5815 * We must not return until all unregister events added during
5816 * the interval the lock was held have been completed.
5818 void netdev_run_todo(void)
5820 struct list_head list;
5822 /* Snapshot list, allow later requests */
5823 list_replace_init(&net_todo_list, &list);
5827 /* Wait for rcu callbacks to finish before attempting to drain
5828 * the device list. This usually avoids a 250ms wait.
5830 if (!list_empty(&list))
5833 while (!list_empty(&list)) {
5834 struct net_device *dev
5835 = list_first_entry(&list, struct net_device, todo_list);
5836 list_del(&dev->todo_list);
5838 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5839 printk(KERN_ERR "network todo '%s' but state %d\n",
5840 dev->name, dev->reg_state);
5845 dev->reg_state = NETREG_UNREGISTERED;
5847 on_each_cpu(flush_backlog, dev, 1);
5849 netdev_wait_allrefs(dev);
5852 BUG_ON(netdev_refcnt_read(dev));
5853 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5854 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5855 WARN_ON(dev->dn_ptr);
5857 if (dev->destructor)
5858 dev->destructor(dev);
5860 /* Free network device */
5861 kobject_put(&dev->dev.kobj);
5865 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5866 * fields in the same order, with only the type differing.
5868 static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5869 const struct net_device_stats *netdev_stats)
5871 #if BITS_PER_LONG == 64
5872 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5873 memcpy(stats64, netdev_stats, sizeof(*stats64));
5875 size_t i, n = sizeof(*stats64) / sizeof(u64);
5876 const unsigned long *src = (const unsigned long *)netdev_stats;
5877 u64 *dst = (u64 *)stats64;
5879 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5880 sizeof(*stats64) / sizeof(u64));
5881 for (i = 0; i < n; i++)
5887 * dev_get_stats - get network device statistics
5888 * @dev: device to get statistics from
5889 * @storage: place to store stats
5891 * Get network statistics from device. Return @storage.
5892 * The device driver may provide its own method by setting
5893 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5894 * otherwise the internal statistics structure is used.
5896 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5897 struct rtnl_link_stats64 *storage)
5899 const struct net_device_ops *ops = dev->netdev_ops;
5901 if (ops->ndo_get_stats64) {
5902 memset(storage, 0, sizeof(*storage));
5903 ops->ndo_get_stats64(dev, storage);
5904 } else if (ops->ndo_get_stats) {
5905 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5907 netdev_stats_to_stats64(storage, &dev->stats);
5909 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5912 EXPORT_SYMBOL(dev_get_stats);
5914 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5916 struct netdev_queue *queue = dev_ingress_queue(dev);
5918 #ifdef CONFIG_NET_CLS_ACT
5921 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5924 netdev_init_one_queue(dev, queue, NULL);
5925 queue->qdisc = &noop_qdisc;
5926 queue->qdisc_sleeping = &noop_qdisc;
5927 rcu_assign_pointer(dev->ingress_queue, queue);
5933 * alloc_netdev_mqs - allocate network device
5934 * @sizeof_priv: size of private data to allocate space for
5935 * @name: device name format string
5936 * @setup: callback to initialize device
5937 * @txqs: the number of TX subqueues to allocate
5938 * @rxqs: the number of RX subqueues to allocate
5940 * Allocates a struct net_device with private data area for driver use
5941 * and performs basic initialization. Also allocates subquue structs
5942 * for each queue on the device.
5944 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5945 void (*setup)(struct net_device *),
5946 unsigned int txqs, unsigned int rxqs)
5948 struct net_device *dev;
5950 struct net_device *p;
5952 BUG_ON(strlen(name) >= sizeof(dev->name));
5955 pr_err("alloc_netdev: Unable to allocate device "
5956 "with zero queues.\n");
5962 pr_err("alloc_netdev: Unable to allocate device "
5963 "with zero RX queues.\n");
5968 alloc_size = sizeof(struct net_device);
5970 /* ensure 32-byte alignment of private area */
5971 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5972 alloc_size += sizeof_priv;
5974 /* ensure 32-byte alignment of whole construct */
5975 alloc_size += NETDEV_ALIGN - 1;
5977 p = kzalloc(alloc_size, GFP_KERNEL);
5979 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5983 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5984 dev->padded = (char *)dev - (char *)p;
5986 dev->pcpu_refcnt = alloc_percpu(int);
5987 if (!dev->pcpu_refcnt)
5990 if (dev_addr_init(dev))
5996 dev_net_set(dev, &init_net);
5998 dev->gso_max_size = GSO_MAX_SIZE;
5999 dev->gso_max_segs = GSO_MAX_SEGS;
6001 INIT_LIST_HEAD(&dev->napi_list);
6002 INIT_LIST_HEAD(&dev->unreg_list);
6003 INIT_LIST_HEAD(&dev->link_watch_list);
6004 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6007 dev->num_tx_queues = txqs;
6008 dev->real_num_tx_queues = txqs;
6009 if (netif_alloc_netdev_queues(dev))
6013 dev->num_rx_queues = rxqs;
6014 dev->real_num_rx_queues = rxqs;
6015 if (netif_alloc_rx_queues(dev))
6019 strcpy(dev->name, name);
6020 dev->group = INIT_NETDEV_GROUP;
6028 free_percpu(dev->pcpu_refcnt);
6038 EXPORT_SYMBOL(alloc_netdev_mqs);
6041 * free_netdev - free network device
6044 * This function does the last stage of destroying an allocated device
6045 * interface. The reference to the device object is released.
6046 * If this is the last reference then it will be freed.
6048 void free_netdev(struct net_device *dev)
6050 struct napi_struct *p, *n;
6052 release_net(dev_net(dev));
6059 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6061 /* Flush device addresses */
6062 dev_addr_flush(dev);
6064 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6067 free_percpu(dev->pcpu_refcnt);
6068 dev->pcpu_refcnt = NULL;
6070 /* Compatibility with error handling in drivers */
6071 if (dev->reg_state == NETREG_UNINITIALIZED) {
6072 kfree((char *)dev - dev->padded);
6076 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6077 dev->reg_state = NETREG_RELEASED;
6079 /* will free via device release */
6080 put_device(&dev->dev);
6082 EXPORT_SYMBOL(free_netdev);
6085 * synchronize_net - Synchronize with packet receive processing
6087 * Wait for packets currently being received to be done.
6088 * Does not block later packets from starting.
6090 void synchronize_net(void)
6093 if (rtnl_is_locked())
6094 synchronize_rcu_expedited();
6098 EXPORT_SYMBOL(synchronize_net);
6101 * unregister_netdevice_queue - remove device from the kernel
6105 * This function shuts down a device interface and removes it
6106 * from the kernel tables.
6107 * If head not NULL, device is queued to be unregistered later.
6109 * Callers must hold the rtnl semaphore. You may want
6110 * unregister_netdev() instead of this.
6113 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6118 list_move_tail(&dev->unreg_list, head);
6120 rollback_registered(dev);
6121 /* Finish processing unregister after unlock */
6125 EXPORT_SYMBOL(unregister_netdevice_queue);
6128 * unregister_netdevice_many - unregister many devices
6129 * @head: list of devices
6131 void unregister_netdevice_many(struct list_head *head)
6133 struct net_device *dev;
6135 if (!list_empty(head)) {
6136 rollback_registered_many(head);
6137 list_for_each_entry(dev, head, unreg_list)
6141 EXPORT_SYMBOL(unregister_netdevice_many);
6144 * unregister_netdev - remove device from the kernel
6147 * This function shuts down a device interface and removes it
6148 * from the kernel tables.
6150 * This is just a wrapper for unregister_netdevice that takes
6151 * the rtnl semaphore. In general you want to use this and not
6152 * unregister_netdevice.
6154 void unregister_netdev(struct net_device *dev)
6157 unregister_netdevice(dev);
6160 EXPORT_SYMBOL(unregister_netdev);
6163 * dev_change_net_namespace - move device to different nethost namespace
6165 * @net: network namespace
6166 * @pat: If not NULL name pattern to try if the current device name
6167 * is already taken in the destination network namespace.
6169 * This function shuts down a device interface and moves it
6170 * to a new network namespace. On success 0 is returned, on
6171 * a failure a netagive errno code is returned.
6173 * Callers must hold the rtnl semaphore.
6176 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6182 /* Don't allow namespace local devices to be moved. */
6184 if (dev->features & NETIF_F_NETNS_LOCAL)
6187 /* Ensure the device has been registrered */
6189 if (dev->reg_state != NETREG_REGISTERED)
6192 /* Get out if there is nothing todo */
6194 if (net_eq(dev_net(dev), net))
6197 /* Pick the destination device name, and ensure
6198 * we can use it in the destination network namespace.
6201 if (__dev_get_by_name(net, dev->name)) {
6202 /* We get here if we can't use the current device name */
6205 if (dev_get_valid_name(dev, pat) < 0)
6210 * And now a mini version of register_netdevice unregister_netdevice.
6213 /* If device is running close it first. */
6216 /* And unlink it from device chain */
6218 unlist_netdevice(dev);
6222 /* Shutdown queueing discipline. */
6225 /* Notify protocols, that we are about to destroy
6226 this device. They should clean all the things.
6228 Note that dev->reg_state stays at NETREG_REGISTERED.
6229 This is wanted because this way 8021q and macvlan know
6230 the device is just moving and can keep their slaves up.
6232 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6233 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6234 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6237 * Flush the unicast and multicast chains
6242 /* Actually switch the network namespace */
6243 dev_net_set(dev, net);
6245 /* If there is an ifindex conflict assign a new one */
6246 if (__dev_get_by_index(net, dev->ifindex)) {
6247 int iflink = (dev->iflink == dev->ifindex);
6248 dev->ifindex = dev_new_index(net);
6250 dev->iflink = dev->ifindex;
6253 /* Fixup kobjects */
6254 err = device_rename(&dev->dev, dev->name);
6257 /* Add the device back in the hashes */
6258 list_netdevice(dev);
6260 /* Notify protocols, that a new device appeared. */
6261 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6264 * Prevent userspace races by waiting until the network
6265 * device is fully setup before sending notifications.
6267 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6274 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6276 static int dev_cpu_callback(struct notifier_block *nfb,
6277 unsigned long action,
6280 struct sk_buff **list_skb;
6281 struct sk_buff *skb;
6282 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6283 struct softnet_data *sd, *oldsd;
6285 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6288 local_irq_disable();
6289 cpu = smp_processor_id();
6290 sd = &per_cpu(softnet_data, cpu);
6291 oldsd = &per_cpu(softnet_data, oldcpu);
6293 /* Find end of our completion_queue. */
6294 list_skb = &sd->completion_queue;
6296 list_skb = &(*list_skb)->next;
6297 /* Append completion queue from offline CPU. */
6298 *list_skb = oldsd->completion_queue;
6299 oldsd->completion_queue = NULL;
6301 /* Append output queue from offline CPU. */
6302 if (oldsd->output_queue) {
6303 *sd->output_queue_tailp = oldsd->output_queue;
6304 sd->output_queue_tailp = oldsd->output_queue_tailp;
6305 oldsd->output_queue = NULL;
6306 oldsd->output_queue_tailp = &oldsd->output_queue;
6308 /* Append NAPI poll list from offline CPU. */
6309 if (!list_empty(&oldsd->poll_list)) {
6310 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6311 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6314 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6317 /* Process offline CPU's input_pkt_queue */
6318 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6320 input_queue_head_incr(oldsd);
6322 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6324 input_queue_head_incr(oldsd);
6332 * netdev_increment_features - increment feature set by one
6333 * @all: current feature set
6334 * @one: new feature set
6335 * @mask: mask feature set
6337 * Computes a new feature set after adding a device with feature set
6338 * @one to the master device with current feature set @all. Will not
6339 * enable anything that is off in @mask. Returns the new feature set.
6341 u32 netdev_increment_features(u32 all, u32 one, u32 mask)
6343 if (mask & NETIF_F_GEN_CSUM)
6344 mask |= NETIF_F_ALL_CSUM;
6345 mask |= NETIF_F_VLAN_CHALLENGED;
6347 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6348 all &= one | ~NETIF_F_ALL_FOR_ALL;
6350 /* If device needs checksumming, downgrade to it. */
6351 if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6352 all &= ~NETIF_F_NO_CSUM;
6354 /* If one device supports hw checksumming, set for all. */
6355 if (all & NETIF_F_GEN_CSUM)
6356 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6360 EXPORT_SYMBOL(netdev_increment_features);
6362 static struct hlist_head *netdev_create_hash(void)
6365 struct hlist_head *hash;
6367 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6369 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6370 INIT_HLIST_HEAD(&hash[i]);
6375 /* Initialize per network namespace state */
6376 static int __net_init netdev_init(struct net *net)
6378 INIT_LIST_HEAD(&net->dev_base_head);
6380 net->dev_name_head = netdev_create_hash();
6381 if (net->dev_name_head == NULL)
6384 net->dev_index_head = netdev_create_hash();
6385 if (net->dev_index_head == NULL)
6391 kfree(net->dev_name_head);
6397 * netdev_drivername - network driver for the device
6398 * @dev: network device
6400 * Determine network driver for device.
6402 const char *netdev_drivername(const struct net_device *dev)
6404 const struct device_driver *driver;
6405 const struct device *parent;
6406 const char *empty = "";
6408 parent = dev->dev.parent;
6412 driver = parent->driver;
6413 if (driver && driver->name)
6414 return driver->name;
6418 int __netdev_printk(const char *level, const struct net_device *dev,
6419 struct va_format *vaf)
6423 if (dev && dev->dev.parent)
6424 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6425 netdev_name(dev), vaf);
6427 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6429 r = printk("%s(NULL net_device): %pV", level, vaf);
6433 EXPORT_SYMBOL(__netdev_printk);
6435 int netdev_printk(const char *level, const struct net_device *dev,
6436 const char *format, ...)
6438 struct va_format vaf;
6442 va_start(args, format);
6447 r = __netdev_printk(level, dev, &vaf);
6452 EXPORT_SYMBOL(netdev_printk);
6454 #define define_netdev_printk_level(func, level) \
6455 int func(const struct net_device *dev, const char *fmt, ...) \
6458 struct va_format vaf; \
6461 va_start(args, fmt); \
6466 r = __netdev_printk(level, dev, &vaf); \
6471 EXPORT_SYMBOL(func);
6473 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6474 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6475 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6476 define_netdev_printk_level(netdev_err, KERN_ERR);
6477 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6478 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6479 define_netdev_printk_level(netdev_info, KERN_INFO);
6481 static void __net_exit netdev_exit(struct net *net)
6483 kfree(net->dev_name_head);
6484 kfree(net->dev_index_head);
6487 static struct pernet_operations __net_initdata netdev_net_ops = {
6488 .init = netdev_init,
6489 .exit = netdev_exit,
6492 static void __net_exit default_device_exit(struct net *net)
6494 struct net_device *dev, *aux;
6496 * Push all migratable network devices back to the
6497 * initial network namespace
6500 for_each_netdev_safe(net, dev, aux) {
6502 char fb_name[IFNAMSIZ];
6504 /* Ignore unmoveable devices (i.e. loopback) */
6505 if (dev->features & NETIF_F_NETNS_LOCAL)
6508 /* Leave virtual devices for the generic cleanup */
6509 if (dev->rtnl_link_ops)
6512 /* Push remaining network devices to init_net */
6513 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6514 err = dev_change_net_namespace(dev, &init_net, fb_name);
6516 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6517 __func__, dev->name, err);
6524 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6526 /* At exit all network devices most be removed from a network
6527 * namespace. Do this in the reverse order of registration.
6528 * Do this across as many network namespaces as possible to
6529 * improve batching efficiency.
6531 struct net_device *dev;
6533 LIST_HEAD(dev_kill_list);
6536 list_for_each_entry(net, net_list, exit_list) {
6537 for_each_netdev_reverse(net, dev) {
6538 if (dev->rtnl_link_ops)
6539 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6541 unregister_netdevice_queue(dev, &dev_kill_list);
6544 unregister_netdevice_many(&dev_kill_list);
6545 list_del(&dev_kill_list);
6549 static struct pernet_operations __net_initdata default_device_ops = {
6550 .exit = default_device_exit,
6551 .exit_batch = default_device_exit_batch,
6555 * Initialize the DEV module. At boot time this walks the device list and
6556 * unhooks any devices that fail to initialise (normally hardware not
6557 * present) and leaves us with a valid list of present and active devices.
6562 * This is called single threaded during boot, so no need
6563 * to take the rtnl semaphore.
6565 static int __init net_dev_init(void)
6567 int i, rc = -ENOMEM;
6569 BUG_ON(!dev_boot_phase);
6571 if (dev_proc_init())
6574 if (netdev_kobject_init())
6577 INIT_LIST_HEAD(&ptype_all);
6578 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6579 INIT_LIST_HEAD(&ptype_base[i]);
6581 if (register_pernet_subsys(&netdev_net_ops))
6585 * Initialise the packet receive queues.
6588 for_each_possible_cpu(i) {
6589 struct softnet_data *sd = &per_cpu(softnet_data, i);
6591 memset(sd, 0, sizeof(*sd));
6592 skb_queue_head_init(&sd->input_pkt_queue);
6593 skb_queue_head_init(&sd->process_queue);
6594 sd->completion_queue = NULL;
6595 INIT_LIST_HEAD(&sd->poll_list);
6596 sd->output_queue = NULL;
6597 sd->output_queue_tailp = &sd->output_queue;
6599 sd->csd.func = rps_trigger_softirq;
6605 sd->backlog.poll = process_backlog;
6606 sd->backlog.weight = weight_p;
6607 sd->backlog.gro_list = NULL;
6608 sd->backlog.gro_count = 0;
6613 /* The loopback device is special if any other network devices
6614 * is present in a network namespace the loopback device must
6615 * be present. Since we now dynamically allocate and free the
6616 * loopback device ensure this invariant is maintained by
6617 * keeping the loopback device as the first device on the
6618 * list of network devices. Ensuring the loopback devices
6619 * is the first device that appears and the last network device
6622 if (register_pernet_device(&loopback_net_ops))
6625 if (register_pernet_device(&default_device_ops))
6628 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6629 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6631 hotcpu_notifier(dev_cpu_callback, 0);
6639 subsys_initcall(net_dev_init);
6641 static int __init initialize_hashrnd(void)
6643 get_random_bytes(&hashrnd, sizeof(hashrnd));
6647 late_initcall_sync(initialize_hashrnd);