2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/if_tunnel.h>
137 #include <linux/if_pppox.h>
138 #include <linux/ppp_defs.h>
139 #include <linux/net_tstamp.h>
141 #include "net-sysfs.h"
143 /* Instead of increasing this, you should create a hash table. */
144 #define MAX_GRO_SKBS 8
146 /* This should be increased if a protocol with a bigger head is added. */
147 #define GRO_MAX_HEAD (MAX_HEADER + 128)
150 * The list of packet types we will receive (as opposed to discard)
151 * and the routines to invoke.
153 * Why 16. Because with 16 the only overlap we get on a hash of the
154 * low nibble of the protocol value is RARP/SNAP/X.25.
156 * NOTE: That is no longer true with the addition of VLAN tags. Not
157 * sure which should go first, but I bet it won't make much
158 * difference if we are running VLANs. The good news is that
159 * this protocol won't be in the list unless compiled in, so
160 * the average user (w/out VLANs) will not be adversely affected.
177 #define PTYPE_HASH_SIZE (16)
178 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
180 static DEFINE_SPINLOCK(ptype_lock);
181 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
182 static struct list_head ptype_all __read_mostly; /* Taps */
185 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
188 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
190 * Writers must hold the rtnl semaphore while they loop through the
191 * dev_base_head list, and hold dev_base_lock for writing when they do the
192 * actual updates. This allows pure readers to access the list even
193 * while a writer is preparing to update it.
195 * To put it another way, dev_base_lock is held for writing only to
196 * protect against pure readers; the rtnl semaphore provides the
197 * protection against other writers.
199 * See, for example usages, register_netdevice() and
200 * unregister_netdevice(), which must be called with the rtnl
203 DEFINE_RWLOCK(dev_base_lock);
204 EXPORT_SYMBOL(dev_base_lock);
206 static inline void dev_base_seq_inc(struct net *net)
208 while (++net->dev_base_seq == 0);
211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
213 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
214 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
217 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
219 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
222 static inline void rps_lock(struct softnet_data *sd)
225 spin_lock(&sd->input_pkt_queue.lock);
229 static inline void rps_unlock(struct softnet_data *sd)
232 spin_unlock(&sd->input_pkt_queue.lock);
236 /* Device list insertion */
237 static int list_netdevice(struct net_device *dev)
239 struct net *net = dev_net(dev);
243 write_lock_bh(&dev_base_lock);
244 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
245 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
246 hlist_add_head_rcu(&dev->index_hlist,
247 dev_index_hash(net, dev->ifindex));
248 write_unlock_bh(&dev_base_lock);
250 dev_base_seq_inc(net);
255 /* Device list removal
256 * caller must respect a RCU grace period before freeing/reusing dev
258 static void unlist_netdevice(struct net_device *dev)
262 /* Unlink dev from the device chain */
263 write_lock_bh(&dev_base_lock);
264 list_del_rcu(&dev->dev_list);
265 hlist_del_rcu(&dev->name_hlist);
266 hlist_del_rcu(&dev->index_hlist);
267 write_unlock_bh(&dev_base_lock);
269 dev_base_seq_inc(dev_net(dev));
276 static RAW_NOTIFIER_HEAD(netdev_chain);
279 * Device drivers call our routines to queue packets here. We empty the
280 * queue in the local softnet handler.
283 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
284 EXPORT_PER_CPU_SYMBOL(softnet_data);
286 #ifdef CONFIG_LOCKDEP
288 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
289 * according to dev->type
291 static const unsigned short netdev_lock_type[] =
292 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
293 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
294 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
295 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
296 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
297 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
298 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
299 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
300 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
301 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
302 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
303 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
304 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
305 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
306 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
307 ARPHRD_VOID, ARPHRD_NONE};
309 static const char *const netdev_lock_name[] =
310 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
311 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
312 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
313 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
314 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
315 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
316 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
317 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
318 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
319 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
320 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
321 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
322 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
323 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
324 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
325 "_xmit_VOID", "_xmit_NONE"};
327 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
328 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
330 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
334 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
335 if (netdev_lock_type[i] == dev_type)
337 /* the last key is used by default */
338 return ARRAY_SIZE(netdev_lock_type) - 1;
341 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
342 unsigned short dev_type)
346 i = netdev_lock_pos(dev_type);
347 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
348 netdev_lock_name[i]);
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
355 i = netdev_lock_pos(dev->type);
356 lockdep_set_class_and_name(&dev->addr_list_lock,
357 &netdev_addr_lock_key[i],
358 netdev_lock_name[i]);
361 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
362 unsigned short dev_type)
365 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
370 /*******************************************************************************
372 Protocol management and registration routines
374 *******************************************************************************/
377 * Add a protocol ID to the list. Now that the input handler is
378 * smarter we can dispense with all the messy stuff that used to be
381 * BEWARE!!! Protocol handlers, mangling input packets,
382 * MUST BE last in hash buckets and checking protocol handlers
383 * MUST start from promiscuous ptype_all chain in net_bh.
384 * It is true now, do not change it.
385 * Explanation follows: if protocol handler, mangling packet, will
386 * be the first on list, it is not able to sense, that packet
387 * is cloned and should be copied-on-write, so that it will
388 * change it and subsequent readers will get broken packet.
392 static inline struct list_head *ptype_head(const struct packet_type *pt)
394 if (pt->type == htons(ETH_P_ALL))
397 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
401 * dev_add_pack - add packet handler
402 * @pt: packet type declaration
404 * Add a protocol handler to the networking stack. The passed &packet_type
405 * is linked into kernel lists and may not be freed until it has been
406 * removed from the kernel lists.
408 * This call does not sleep therefore it can not
409 * guarantee all CPU's that are in middle of receiving packets
410 * will see the new packet type (until the next received packet).
413 void dev_add_pack(struct packet_type *pt)
415 struct list_head *head = ptype_head(pt);
417 spin_lock(&ptype_lock);
418 list_add_rcu(&pt->list, head);
419 spin_unlock(&ptype_lock);
421 EXPORT_SYMBOL(dev_add_pack);
424 * __dev_remove_pack - remove packet handler
425 * @pt: packet type declaration
427 * Remove a protocol handler that was previously added to the kernel
428 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
429 * from the kernel lists and can be freed or reused once this function
432 * The packet type might still be in use by receivers
433 * and must not be freed until after all the CPU's have gone
434 * through a quiescent state.
436 void __dev_remove_pack(struct packet_type *pt)
438 struct list_head *head = ptype_head(pt);
439 struct packet_type *pt1;
441 spin_lock(&ptype_lock);
443 list_for_each_entry(pt1, head, list) {
445 list_del_rcu(&pt->list);
450 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
452 spin_unlock(&ptype_lock);
454 EXPORT_SYMBOL(__dev_remove_pack);
457 * dev_remove_pack - remove packet handler
458 * @pt: packet type declaration
460 * Remove a protocol handler that was previously added to the kernel
461 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
462 * from the kernel lists and can be freed or reused once this function
465 * This call sleeps to guarantee that no CPU is looking at the packet
468 void dev_remove_pack(struct packet_type *pt)
470 __dev_remove_pack(pt);
474 EXPORT_SYMBOL(dev_remove_pack);
476 /******************************************************************************
478 Device Boot-time Settings Routines
480 *******************************************************************************/
482 /* Boot time configuration table */
483 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
486 * netdev_boot_setup_add - add new setup entry
487 * @name: name of the device
488 * @map: configured settings for the device
490 * Adds new setup entry to the dev_boot_setup list. The function
491 * returns 0 on error and 1 on success. This is a generic routine to
494 static int netdev_boot_setup_add(char *name, struct ifmap *map)
496 struct netdev_boot_setup *s;
500 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
501 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
502 memset(s[i].name, 0, sizeof(s[i].name));
503 strlcpy(s[i].name, name, IFNAMSIZ);
504 memcpy(&s[i].map, map, sizeof(s[i].map));
509 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
513 * netdev_boot_setup_check - check boot time settings
514 * @dev: the netdevice
516 * Check boot time settings for the device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found, 1 if they are.
521 int netdev_boot_setup_check(struct net_device *dev)
523 struct netdev_boot_setup *s = dev_boot_setup;
526 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
527 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
528 !strcmp(dev->name, s[i].name)) {
529 dev->irq = s[i].map.irq;
530 dev->base_addr = s[i].map.base_addr;
531 dev->mem_start = s[i].map.mem_start;
532 dev->mem_end = s[i].map.mem_end;
538 EXPORT_SYMBOL(netdev_boot_setup_check);
542 * netdev_boot_base - get address from boot time settings
543 * @prefix: prefix for network device
544 * @unit: id for network device
546 * Check boot time settings for the base address of device.
547 * The found settings are set for the device to be used
548 * later in the device probing.
549 * Returns 0 if no settings found.
551 unsigned long netdev_boot_base(const char *prefix, int unit)
553 const struct netdev_boot_setup *s = dev_boot_setup;
557 sprintf(name, "%s%d", prefix, unit);
560 * If device already registered then return base of 1
561 * to indicate not to probe for this interface
563 if (__dev_get_by_name(&init_net, name))
566 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
567 if (!strcmp(name, s[i].name))
568 return s[i].map.base_addr;
573 * Saves at boot time configured settings for any netdevice.
575 int __init netdev_boot_setup(char *str)
580 str = get_options(str, ARRAY_SIZE(ints), ints);
585 memset(&map, 0, sizeof(map));
589 map.base_addr = ints[2];
591 map.mem_start = ints[3];
593 map.mem_end = ints[4];
595 /* Add new entry to the list */
596 return netdev_boot_setup_add(str, &map);
599 __setup("netdev=", netdev_boot_setup);
601 /*******************************************************************************
603 Device Interface Subroutines
605 *******************************************************************************/
608 * __dev_get_by_name - find a device by its name
609 * @net: the applicable net namespace
610 * @name: name to find
612 * Find an interface by name. Must be called under RTNL semaphore
613 * or @dev_base_lock. If the name is found a pointer to the device
614 * is returned. If the name is not found then %NULL is returned. The
615 * reference counters are not incremented so the caller must be
616 * careful with locks.
619 struct net_device *__dev_get_by_name(struct net *net, const char *name)
621 struct hlist_node *p;
622 struct net_device *dev;
623 struct hlist_head *head = dev_name_hash(net, name);
625 hlist_for_each_entry(dev, p, head, name_hlist)
626 if (!strncmp(dev->name, name, IFNAMSIZ))
631 EXPORT_SYMBOL(__dev_get_by_name);
634 * dev_get_by_name_rcu - find a device by its name
635 * @net: the applicable net namespace
636 * @name: name to find
638 * Find an interface by name.
639 * If the name is found a pointer to the device is returned.
640 * If the name is not found then %NULL is returned.
641 * The reference counters are not incremented so the caller must be
642 * careful with locks. The caller must hold RCU lock.
645 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
647 struct hlist_node *p;
648 struct net_device *dev;
649 struct hlist_head *head = dev_name_hash(net, name);
651 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
652 if (!strncmp(dev->name, name, IFNAMSIZ))
657 EXPORT_SYMBOL(dev_get_by_name_rcu);
660 * dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. This can be called from any
665 * context and does its own locking. The returned handle has
666 * the usage count incremented and the caller must use dev_put() to
667 * release it when it is no longer needed. %NULL is returned if no
668 * matching device is found.
671 struct net_device *dev_get_by_name(struct net *net, const char *name)
673 struct net_device *dev;
676 dev = dev_get_by_name_rcu(net, name);
682 EXPORT_SYMBOL(dev_get_by_name);
685 * __dev_get_by_index - find a device by its ifindex
686 * @net: the applicable net namespace
687 * @ifindex: index of device
689 * Search for an interface by index. Returns %NULL if the device
690 * is not found or a pointer to the device. The device has not
691 * had its reference counter increased so the caller must be careful
692 * about locking. The caller must hold either the RTNL semaphore
696 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
698 struct hlist_node *p;
699 struct net_device *dev;
700 struct hlist_head *head = dev_index_hash(net, ifindex);
702 hlist_for_each_entry(dev, p, head, index_hlist)
703 if (dev->ifindex == ifindex)
708 EXPORT_SYMBOL(__dev_get_by_index);
711 * dev_get_by_index_rcu - find a device by its ifindex
712 * @net: the applicable net namespace
713 * @ifindex: index of device
715 * Search for an interface by index. Returns %NULL if the device
716 * is not found or a pointer to the device. The device has not
717 * had its reference counter increased so the caller must be careful
718 * about locking. The caller must hold RCU lock.
721 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
723 struct hlist_node *p;
724 struct net_device *dev;
725 struct hlist_head *head = dev_index_hash(net, ifindex);
727 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
728 if (dev->ifindex == ifindex)
733 EXPORT_SYMBOL(dev_get_by_index_rcu);
737 * dev_get_by_index - find a device by its ifindex
738 * @net: the applicable net namespace
739 * @ifindex: index of device
741 * Search for an interface by index. Returns NULL if the device
742 * is not found or a pointer to the device. The device returned has
743 * had a reference added and the pointer is safe until the user calls
744 * dev_put to indicate they have finished with it.
747 struct net_device *dev_get_by_index(struct net *net, int ifindex)
749 struct net_device *dev;
752 dev = dev_get_by_index_rcu(net, ifindex);
758 EXPORT_SYMBOL(dev_get_by_index);
761 * dev_getbyhwaddr_rcu - find a device by its hardware address
762 * @net: the applicable net namespace
763 * @type: media type of device
764 * @ha: hardware address
766 * Search for an interface by MAC address. Returns NULL if the device
767 * is not found or a pointer to the device.
768 * The caller must hold RCU or RTNL.
769 * The returned device has not had its ref count increased
770 * and the caller must therefore be careful about locking
774 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
777 struct net_device *dev;
779 for_each_netdev_rcu(net, dev)
780 if (dev->type == type &&
781 !memcmp(dev->dev_addr, ha, dev->addr_len))
786 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
788 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
790 struct net_device *dev;
793 for_each_netdev(net, dev)
794 if (dev->type == type)
799 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
801 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
803 struct net_device *dev, *ret = NULL;
806 for_each_netdev_rcu(net, dev)
807 if (dev->type == type) {
815 EXPORT_SYMBOL(dev_getfirstbyhwtype);
818 * dev_get_by_flags_rcu - find any device with given flags
819 * @net: the applicable net namespace
820 * @if_flags: IFF_* values
821 * @mask: bitmask of bits in if_flags to check
823 * Search for any interface with the given flags. Returns NULL if a device
824 * is not found or a pointer to the device. Must be called inside
825 * rcu_read_lock(), and result refcount is unchanged.
828 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
831 struct net_device *dev, *ret;
834 for_each_netdev_rcu(net, dev) {
835 if (((dev->flags ^ if_flags) & mask) == 0) {
842 EXPORT_SYMBOL(dev_get_by_flags_rcu);
845 * dev_valid_name - check if name is okay for network device
848 * Network device names need to be valid file names to
849 * to allow sysfs to work. We also disallow any kind of
852 int dev_valid_name(const char *name)
856 if (strlen(name) >= IFNAMSIZ)
858 if (!strcmp(name, ".") || !strcmp(name, ".."))
862 if (*name == '/' || *name == ':' || isspace(*name))
868 EXPORT_SYMBOL(dev_valid_name);
871 * __dev_alloc_name - allocate a name for a device
872 * @net: network namespace to allocate the device name in
873 * @name: name format string
874 * @buf: scratch buffer and result name string
876 * Passed a format string - eg "lt%d" it will try and find a suitable
877 * id. It scans list of devices to build up a free map, then chooses
878 * the first empty slot. The caller must hold the dev_base or rtnl lock
879 * while allocating the name and adding the device in order to avoid
881 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
882 * Returns the number of the unit assigned or a negative errno code.
885 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
889 const int max_netdevices = 8*PAGE_SIZE;
890 unsigned long *inuse;
891 struct net_device *d;
893 p = strnchr(name, IFNAMSIZ-1, '%');
896 * Verify the string as this thing may have come from
897 * the user. There must be either one "%d" and no other "%"
900 if (p[1] != 'd' || strchr(p + 2, '%'))
903 /* Use one page as a bit array of possible slots */
904 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
908 for_each_netdev(net, d) {
909 if (!sscanf(d->name, name, &i))
911 if (i < 0 || i >= max_netdevices)
914 /* avoid cases where sscanf is not exact inverse of printf */
915 snprintf(buf, IFNAMSIZ, name, i);
916 if (!strncmp(buf, d->name, IFNAMSIZ))
920 i = find_first_zero_bit(inuse, max_netdevices);
921 free_page((unsigned long) inuse);
925 snprintf(buf, IFNAMSIZ, name, i);
926 if (!__dev_get_by_name(net, buf))
929 /* It is possible to run out of possible slots
930 * when the name is long and there isn't enough space left
931 * for the digits, or if all bits are used.
937 * dev_alloc_name - allocate a name for a device
939 * @name: name format string
941 * Passed a format string - eg "lt%d" it will try and find a suitable
942 * id. It scans list of devices to build up a free map, then chooses
943 * the first empty slot. The caller must hold the dev_base or rtnl lock
944 * while allocating the name and adding the device in order to avoid
946 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
947 * Returns the number of the unit assigned or a negative errno code.
950 int dev_alloc_name(struct net_device *dev, const char *name)
956 BUG_ON(!dev_net(dev));
958 ret = __dev_alloc_name(net, name, buf);
960 strlcpy(dev->name, buf, IFNAMSIZ);
963 EXPORT_SYMBOL(dev_alloc_name);
965 static int dev_get_valid_name(struct net_device *dev, const char *name)
969 BUG_ON(!dev_net(dev));
972 if (!dev_valid_name(name))
975 if (strchr(name, '%'))
976 return dev_alloc_name(dev, name);
977 else if (__dev_get_by_name(net, name))
979 else if (dev->name != name)
980 strlcpy(dev->name, name, IFNAMSIZ);
986 * dev_change_name - change name of a device
988 * @newname: name (or format string) must be at least IFNAMSIZ
990 * Change name of a device, can pass format strings "eth%d".
993 int dev_change_name(struct net_device *dev, const char *newname)
995 char oldname[IFNAMSIZ];
1001 BUG_ON(!dev_net(dev));
1004 if (dev->flags & IFF_UP)
1007 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1010 memcpy(oldname, dev->name, IFNAMSIZ);
1012 err = dev_get_valid_name(dev, newname);
1017 ret = device_rename(&dev->dev, dev->name);
1019 memcpy(dev->name, oldname, IFNAMSIZ);
1023 write_lock_bh(&dev_base_lock);
1024 hlist_del_rcu(&dev->name_hlist);
1025 write_unlock_bh(&dev_base_lock);
1029 write_lock_bh(&dev_base_lock);
1030 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1031 write_unlock_bh(&dev_base_lock);
1033 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1034 ret = notifier_to_errno(ret);
1037 /* err >= 0 after dev_alloc_name() or stores the first errno */
1040 memcpy(dev->name, oldname, IFNAMSIZ);
1044 "%s: name change rollback failed: %d.\n",
1053 * dev_set_alias - change ifalias of a device
1055 * @alias: name up to IFALIASZ
1056 * @len: limit of bytes to copy from info
1058 * Set ifalias for a device,
1060 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1066 if (len >= IFALIASZ)
1071 kfree(dev->ifalias);
1072 dev->ifalias = NULL;
1077 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1080 dev->ifalias = new_ifalias;
1082 strlcpy(dev->ifalias, alias, len+1);
1088 * netdev_features_change - device changes features
1089 * @dev: device to cause notification
1091 * Called to indicate a device has changed features.
1093 void netdev_features_change(struct net_device *dev)
1095 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1097 EXPORT_SYMBOL(netdev_features_change);
1100 * netdev_state_change - device changes state
1101 * @dev: device to cause notification
1103 * Called to indicate a device has changed state. This function calls
1104 * the notifier chains for netdev_chain and sends a NEWLINK message
1105 * to the routing socket.
1107 void netdev_state_change(struct net_device *dev)
1109 if (dev->flags & IFF_UP) {
1110 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1111 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1114 EXPORT_SYMBOL(netdev_state_change);
1116 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1118 return call_netdevice_notifiers(event, dev);
1120 EXPORT_SYMBOL(netdev_bonding_change);
1123 * dev_load - load a network module
1124 * @net: the applicable net namespace
1125 * @name: name of interface
1127 * If a network interface is not present and the process has suitable
1128 * privileges this function loads the module. If module loading is not
1129 * available in this kernel then it becomes a nop.
1132 void dev_load(struct net *net, const char *name)
1134 struct net_device *dev;
1138 dev = dev_get_by_name_rcu(net, name);
1142 if (no_module && capable(CAP_NET_ADMIN))
1143 no_module = request_module("netdev-%s", name);
1144 if (no_module && capable(CAP_SYS_MODULE)) {
1145 if (!request_module("%s", name))
1146 pr_err("Loading kernel module for a network device "
1147 "with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s "
1151 EXPORT_SYMBOL(dev_load);
1153 static int __dev_open(struct net_device *dev)
1155 const struct net_device_ops *ops = dev->netdev_ops;
1160 if (!netif_device_present(dev))
1163 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1164 ret = notifier_to_errno(ret);
1168 set_bit(__LINK_STATE_START, &dev->state);
1170 if (ops->ndo_validate_addr)
1171 ret = ops->ndo_validate_addr(dev);
1173 if (!ret && ops->ndo_open)
1174 ret = ops->ndo_open(dev);
1177 clear_bit(__LINK_STATE_START, &dev->state);
1179 dev->flags |= IFF_UP;
1180 net_dmaengine_get();
1181 dev_set_rx_mode(dev);
1183 add_device_randomness(dev->dev_addr, dev->addr_len);
1190 * dev_open - prepare an interface for use.
1191 * @dev: device to open
1193 * Takes a device from down to up state. The device's private open
1194 * function is invoked and then the multicast lists are loaded. Finally
1195 * the device is moved into the up state and a %NETDEV_UP message is
1196 * sent to the netdev notifier chain.
1198 * Calling this function on an active interface is a nop. On a failure
1199 * a negative errno code is returned.
1201 int dev_open(struct net_device *dev)
1205 if (dev->flags & IFF_UP)
1208 ret = __dev_open(dev);
1212 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1213 call_netdevice_notifiers(NETDEV_UP, dev);
1217 EXPORT_SYMBOL(dev_open);
1219 static int __dev_close_many(struct list_head *head)
1221 struct net_device *dev;
1226 list_for_each_entry(dev, head, unreg_list) {
1227 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1229 clear_bit(__LINK_STATE_START, &dev->state);
1231 /* Synchronize to scheduled poll. We cannot touch poll list, it
1232 * can be even on different cpu. So just clear netif_running().
1234 * dev->stop() will invoke napi_disable() on all of it's
1235 * napi_struct instances on this device.
1237 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1240 dev_deactivate_many(head);
1242 list_for_each_entry(dev, head, unreg_list) {
1243 const struct net_device_ops *ops = dev->netdev_ops;
1246 * Call the device specific close. This cannot fail.
1247 * Only if device is UP
1249 * We allow it to be called even after a DETACH hot-plug
1255 dev->flags &= ~IFF_UP;
1256 net_dmaengine_put();
1262 static int __dev_close(struct net_device *dev)
1267 list_add(&dev->unreg_list, &single);
1268 retval = __dev_close_many(&single);
1273 static int dev_close_many(struct list_head *head)
1275 struct net_device *dev, *tmp;
1276 LIST_HEAD(tmp_list);
1278 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1279 if (!(dev->flags & IFF_UP))
1280 list_move(&dev->unreg_list, &tmp_list);
1282 __dev_close_many(head);
1284 list_for_each_entry(dev, head, unreg_list) {
1285 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1286 call_netdevice_notifiers(NETDEV_DOWN, dev);
1289 /* rollback_registered_many needs the complete original list */
1290 list_splice(&tmp_list, head);
1295 * dev_close - shutdown an interface.
1296 * @dev: device to shutdown
1298 * This function moves an active device into down state. A
1299 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1300 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1303 int dev_close(struct net_device *dev)
1305 if (dev->flags & IFF_UP) {
1308 list_add(&dev->unreg_list, &single);
1309 dev_close_many(&single);
1314 EXPORT_SYMBOL(dev_close);
1318 * dev_disable_lro - disable Large Receive Offload on a device
1321 * Disable Large Receive Offload (LRO) on a net device. Must be
1322 * called under RTNL. This is needed if received packets may be
1323 * forwarded to another interface.
1325 void dev_disable_lro(struct net_device *dev)
1330 * If we're trying to disable lro on a vlan device
1331 * use the underlying physical device instead
1333 if (is_vlan_dev(dev))
1334 dev = vlan_dev_real_dev(dev);
1336 if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1337 flags = dev->ethtool_ops->get_flags(dev);
1339 flags = ethtool_op_get_flags(dev);
1341 if (!(flags & ETH_FLAG_LRO))
1344 __ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1345 if (unlikely(dev->features & NETIF_F_LRO))
1346 netdev_WARN(dev, "failed to disable LRO!\n");
1348 EXPORT_SYMBOL(dev_disable_lro);
1351 static int dev_boot_phase = 1;
1354 * register_netdevice_notifier - register a network notifier block
1357 * Register a notifier to be called when network device events occur.
1358 * The notifier passed is linked into the kernel structures and must
1359 * not be reused until it has been unregistered. A negative errno code
1360 * is returned on a failure.
1362 * When registered all registration and up events are replayed
1363 * to the new notifier to allow device to have a race free
1364 * view of the network device list.
1367 int register_netdevice_notifier(struct notifier_block *nb)
1369 struct net_device *dev;
1370 struct net_device *last;
1375 err = raw_notifier_chain_register(&netdev_chain, nb);
1381 for_each_netdev(net, dev) {
1382 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1383 err = notifier_to_errno(err);
1387 if (!(dev->flags & IFF_UP))
1390 nb->notifier_call(nb, NETDEV_UP, dev);
1401 for_each_netdev(net, dev) {
1405 if (dev->flags & IFF_UP) {
1406 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1407 nb->notifier_call(nb, NETDEV_DOWN, dev);
1409 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1410 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1415 raw_notifier_chain_unregister(&netdev_chain, nb);
1418 EXPORT_SYMBOL(register_netdevice_notifier);
1421 * unregister_netdevice_notifier - unregister a network notifier block
1424 * Unregister a notifier previously registered by
1425 * register_netdevice_notifier(). The notifier is unlinked into the
1426 * kernel structures and may then be reused. A negative errno code
1427 * is returned on a failure.
1429 * After unregistering unregister and down device events are synthesized
1430 * for all devices on the device list to the removed notifier to remove
1431 * the need for special case cleanup code.
1434 int unregister_netdevice_notifier(struct notifier_block *nb)
1436 struct net_device *dev;
1441 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1446 for_each_netdev(net, dev) {
1447 if (dev->flags & IFF_UP) {
1448 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1449 nb->notifier_call(nb, NETDEV_DOWN, dev);
1451 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1452 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1459 EXPORT_SYMBOL(unregister_netdevice_notifier);
1462 * call_netdevice_notifiers - call all network notifier blocks
1463 * @val: value passed unmodified to notifier function
1464 * @dev: net_device pointer passed unmodified to notifier function
1466 * Call all network notifier blocks. Parameters and return value
1467 * are as for raw_notifier_call_chain().
1470 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1473 return raw_notifier_call_chain(&netdev_chain, val, dev);
1475 EXPORT_SYMBOL(call_netdevice_notifiers);
1477 /* When > 0 there are consumers of rx skb time stamps */
1478 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1480 void net_enable_timestamp(void)
1482 atomic_inc(&netstamp_needed);
1484 EXPORT_SYMBOL(net_enable_timestamp);
1486 void net_disable_timestamp(void)
1488 atomic_dec(&netstamp_needed);
1490 EXPORT_SYMBOL(net_disable_timestamp);
1492 static inline void net_timestamp_set(struct sk_buff *skb)
1494 if (atomic_read(&netstamp_needed))
1495 __net_timestamp(skb);
1497 skb->tstamp.tv64 = 0;
1500 static inline void net_timestamp_check(struct sk_buff *skb)
1502 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1503 __net_timestamp(skb);
1506 static int net_hwtstamp_validate(struct ifreq *ifr)
1508 struct hwtstamp_config cfg;
1509 enum hwtstamp_tx_types tx_type;
1510 enum hwtstamp_rx_filters rx_filter;
1511 int tx_type_valid = 0;
1512 int rx_filter_valid = 0;
1514 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1517 if (cfg.flags) /* reserved for future extensions */
1520 tx_type = cfg.tx_type;
1521 rx_filter = cfg.rx_filter;
1524 case HWTSTAMP_TX_OFF:
1525 case HWTSTAMP_TX_ON:
1526 case HWTSTAMP_TX_ONESTEP_SYNC:
1531 switch (rx_filter) {
1532 case HWTSTAMP_FILTER_NONE:
1533 case HWTSTAMP_FILTER_ALL:
1534 case HWTSTAMP_FILTER_SOME:
1535 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1536 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1537 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1538 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1539 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1540 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1541 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1542 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1543 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1544 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1545 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1546 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1547 rx_filter_valid = 1;
1551 if (!tx_type_valid || !rx_filter_valid)
1557 static inline bool is_skb_forwardable(struct net_device *dev,
1558 struct sk_buff *skb)
1562 if (!(dev->flags & IFF_UP))
1565 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1566 if (skb->len <= len)
1569 /* if TSO is enabled, we don't care about the length as the packet
1570 * could be forwarded without being segmented before
1572 if (skb_is_gso(skb))
1579 * dev_forward_skb - loopback an skb to another netif
1581 * @dev: destination network device
1582 * @skb: buffer to forward
1585 * NET_RX_SUCCESS (no congestion)
1586 * NET_RX_DROP (packet was dropped, but freed)
1588 * dev_forward_skb can be used for injecting an skb from the
1589 * start_xmit function of one device into the receive queue
1590 * of another device.
1592 * The receiving device may be in another namespace, so
1593 * we have to clear all information in the skb that could
1594 * impact namespace isolation.
1596 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1598 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1599 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1600 atomic_long_inc(&dev->rx_dropped);
1609 if (unlikely(!is_skb_forwardable(dev, skb))) {
1610 atomic_long_inc(&dev->rx_dropped);
1616 skb->tstamp.tv64 = 0;
1617 skb->pkt_type = PACKET_HOST;
1618 skb->protocol = eth_type_trans(skb, dev);
1619 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1623 nf_reset_trace(skb);
1624 return netif_rx(skb);
1626 EXPORT_SYMBOL_GPL(dev_forward_skb);
1628 static inline int deliver_skb(struct sk_buff *skb,
1629 struct packet_type *pt_prev,
1630 struct net_device *orig_dev)
1632 atomic_inc(&skb->users);
1633 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1636 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1638 if (!ptype->af_packet_priv || !skb->sk)
1641 if (ptype->id_match)
1642 return ptype->id_match(ptype, skb->sk);
1643 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1650 * Support routine. Sends outgoing frames to any network
1651 * taps currently in use.
1654 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1656 struct packet_type *ptype;
1657 struct sk_buff *skb2 = NULL;
1658 struct packet_type *pt_prev = NULL;
1661 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1662 /* Never send packets back to the socket
1663 * they originated from - MvS (miquels@drinkel.ow.org)
1665 if ((ptype->dev == dev || !ptype->dev) &&
1666 (!skb_loop_sk(ptype, skb))) {
1668 deliver_skb(skb2, pt_prev, skb->dev);
1673 skb2 = skb_clone(skb, GFP_ATOMIC);
1677 net_timestamp_set(skb2);
1679 /* skb->nh should be correctly
1680 set by sender, so that the second statement is
1681 just protection against buggy protocols.
1683 skb_reset_mac_header(skb2);
1685 if (skb_network_header(skb2) < skb2->data ||
1686 skb2->network_header > skb2->tail) {
1687 if (net_ratelimit())
1688 printk(KERN_CRIT "protocol %04x is "
1690 ntohs(skb2->protocol),
1692 skb_reset_network_header(skb2);
1695 skb2->transport_header = skb2->network_header;
1696 skb2->pkt_type = PACKET_OUTGOING;
1701 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1705 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1706 * @dev: Network device
1707 * @txq: number of queues available
1709 * If real_num_tx_queues is changed the tc mappings may no longer be
1710 * valid. To resolve this verify the tc mapping remains valid and if
1711 * not NULL the mapping. With no priorities mapping to this
1712 * offset/count pair it will no longer be used. In the worst case TC0
1713 * is invalid nothing can be done so disable priority mappings. If is
1714 * expected that drivers will fix this mapping if they can before
1715 * calling netif_set_real_num_tx_queues.
1717 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1720 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1722 /* If TC0 is invalidated disable TC mapping */
1723 if (tc->offset + tc->count > txq) {
1724 pr_warning("Number of in use tx queues changed "
1725 "invalidating tc mappings. Priority "
1726 "traffic classification disabled!\n");
1731 /* Invalidated prio to tc mappings set to TC0 */
1732 for (i = 1; i < TC_BITMASK + 1; i++) {
1733 int q = netdev_get_prio_tc_map(dev, i);
1735 tc = &dev->tc_to_txq[q];
1736 if (tc->offset + tc->count > txq) {
1737 pr_warning("Number of in use tx queues "
1738 "changed. Priority %i to tc "
1739 "mapping %i is no longer valid "
1740 "setting map to 0\n",
1742 netdev_set_prio_tc_map(dev, i, 0);
1748 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1749 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1751 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1756 disabling = txq < dev->real_num_tx_queues;
1758 if (txq < 1 || txq > dev->num_tx_queues)
1761 if (dev->reg_state == NETREG_REGISTERED ||
1762 dev->reg_state == NETREG_UNREGISTERING) {
1765 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1771 netif_setup_tc(dev, txq);
1773 dev->real_num_tx_queues = txq;
1777 qdisc_reset_all_tx_gt(dev, txq);
1780 dev->real_num_tx_queues = txq;
1785 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1789 * netif_set_real_num_rx_queues - set actual number of RX queues used
1790 * @dev: Network device
1791 * @rxq: Actual number of RX queues
1793 * This must be called either with the rtnl_lock held or before
1794 * registration of the net device. Returns 0 on success, or a
1795 * negative error code. If called before registration, it always
1798 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1802 if (rxq < 1 || rxq > dev->num_rx_queues)
1805 if (dev->reg_state == NETREG_REGISTERED) {
1808 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1814 dev->real_num_rx_queues = rxq;
1817 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1820 static inline void __netif_reschedule(struct Qdisc *q)
1822 struct softnet_data *sd;
1823 unsigned long flags;
1825 local_irq_save(flags);
1826 sd = &__get_cpu_var(softnet_data);
1827 q->next_sched = NULL;
1828 *sd->output_queue_tailp = q;
1829 sd->output_queue_tailp = &q->next_sched;
1830 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1831 local_irq_restore(flags);
1834 void __netif_schedule(struct Qdisc *q)
1836 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1837 __netif_reschedule(q);
1839 EXPORT_SYMBOL(__netif_schedule);
1841 void dev_kfree_skb_irq(struct sk_buff *skb)
1843 if (atomic_dec_and_test(&skb->users)) {
1844 struct softnet_data *sd;
1845 unsigned long flags;
1847 local_irq_save(flags);
1848 sd = &__get_cpu_var(softnet_data);
1849 skb->next = sd->completion_queue;
1850 sd->completion_queue = skb;
1851 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1852 local_irq_restore(flags);
1855 EXPORT_SYMBOL(dev_kfree_skb_irq);
1857 void dev_kfree_skb_any(struct sk_buff *skb)
1859 if (in_irq() || irqs_disabled())
1860 dev_kfree_skb_irq(skb);
1864 EXPORT_SYMBOL(dev_kfree_skb_any);
1868 * netif_device_detach - mark device as removed
1869 * @dev: network device
1871 * Mark device as removed from system and therefore no longer available.
1873 void netif_device_detach(struct net_device *dev)
1875 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1876 netif_running(dev)) {
1877 netif_tx_stop_all_queues(dev);
1880 EXPORT_SYMBOL(netif_device_detach);
1883 * netif_device_attach - mark device as attached
1884 * @dev: network device
1886 * Mark device as attached from system and restart if needed.
1888 void netif_device_attach(struct net_device *dev)
1890 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1891 netif_running(dev)) {
1892 netif_tx_wake_all_queues(dev);
1893 __netdev_watchdog_up(dev);
1896 EXPORT_SYMBOL(netif_device_attach);
1899 * Invalidate hardware checksum when packet is to be mangled, and
1900 * complete checksum manually on outgoing path.
1902 int skb_checksum_help(struct sk_buff *skb)
1905 int ret = 0, offset;
1907 if (skb->ip_summed == CHECKSUM_COMPLETE)
1908 goto out_set_summed;
1910 if (unlikely(skb_shinfo(skb)->gso_size)) {
1911 /* Let GSO fix up the checksum. */
1912 goto out_set_summed;
1915 offset = skb_checksum_start_offset(skb);
1916 BUG_ON(offset >= skb_headlen(skb));
1917 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1919 offset += skb->csum_offset;
1920 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1922 if (skb_cloned(skb) &&
1923 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1924 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1929 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
1931 skb->ip_summed = CHECKSUM_NONE;
1935 EXPORT_SYMBOL(skb_checksum_help);
1938 * skb_gso_segment - Perform segmentation on skb.
1939 * @skb: buffer to segment
1940 * @features: features for the output path (see dev->features)
1942 * This function segments the given skb and returns a list of segments.
1944 * It may return NULL if the skb requires no segmentation. This is
1945 * only possible when GSO is used for verifying header integrity.
1947 struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
1949 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1950 struct packet_type *ptype;
1951 __be16 type = skb->protocol;
1952 int vlan_depth = ETH_HLEN;
1955 while (type == htons(ETH_P_8021Q)) {
1956 struct vlan_hdr *vh;
1958 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1959 return ERR_PTR(-EINVAL);
1961 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1962 type = vh->h_vlan_encapsulated_proto;
1963 vlan_depth += VLAN_HLEN;
1966 skb_reset_mac_header(skb);
1967 skb->mac_len = skb->network_header - skb->mac_header;
1968 __skb_pull(skb, skb->mac_len);
1970 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1971 struct net_device *dev = skb->dev;
1972 struct ethtool_drvinfo info = {};
1974 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1975 dev->ethtool_ops->get_drvinfo(dev, &info);
1977 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1978 info.driver, dev ? dev->features : 0L,
1979 skb->sk ? skb->sk->sk_route_caps : 0L,
1980 skb->len, skb->data_len, skb->ip_summed);
1982 if (skb_header_cloned(skb) &&
1983 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1984 return ERR_PTR(err);
1988 list_for_each_entry_rcu(ptype,
1989 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1990 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1991 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1992 err = ptype->gso_send_check(skb);
1993 segs = ERR_PTR(err);
1994 if (err || skb_gso_ok(skb, features))
1996 __skb_push(skb, (skb->data -
1997 skb_network_header(skb)));
1999 segs = ptype->gso_segment(skb, features);
2005 __skb_push(skb, skb->data - skb_mac_header(skb));
2009 EXPORT_SYMBOL(skb_gso_segment);
2011 /* Take action when hardware reception checksum errors are detected. */
2013 void netdev_rx_csum_fault(struct net_device *dev)
2015 if (net_ratelimit()) {
2016 printk(KERN_ERR "%s: hw csum failure.\n",
2017 dev ? dev->name : "<unknown>");
2021 EXPORT_SYMBOL(netdev_rx_csum_fault);
2024 /* Actually, we should eliminate this check as soon as we know, that:
2025 * 1. IOMMU is present and allows to map all the memory.
2026 * 2. No high memory really exists on this machine.
2029 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2031 #ifdef CONFIG_HIGHMEM
2033 if (!(dev->features & NETIF_F_HIGHDMA)) {
2034 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2035 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2036 if (PageHighMem(skb_frag_page(frag)))
2041 if (PCI_DMA_BUS_IS_PHYS) {
2042 struct device *pdev = dev->dev.parent;
2046 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2047 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2048 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2049 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2058 void (*destructor)(struct sk_buff *skb);
2061 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2063 static void dev_gso_skb_destructor(struct sk_buff *skb)
2065 struct dev_gso_cb *cb;
2068 struct sk_buff *nskb = skb->next;
2070 skb->next = nskb->next;
2073 } while (skb->next);
2075 cb = DEV_GSO_CB(skb);
2077 cb->destructor(skb);
2081 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2082 * @skb: buffer to segment
2083 * @features: device features as applicable to this skb
2085 * This function segments the given skb and stores the list of segments
2088 static int dev_gso_segment(struct sk_buff *skb, int features)
2090 struct sk_buff *segs;
2092 segs = skb_gso_segment(skb, features);
2094 /* Verifying header integrity only. */
2099 return PTR_ERR(segs);
2102 DEV_GSO_CB(skb)->destructor = skb->destructor;
2103 skb->destructor = dev_gso_skb_destructor;
2108 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2110 return ((features & NETIF_F_GEN_CSUM) ||
2111 ((features & NETIF_F_V4_CSUM) &&
2112 protocol == htons(ETH_P_IP)) ||
2113 ((features & NETIF_F_V6_CSUM) &&
2114 protocol == htons(ETH_P_IPV6)) ||
2115 ((features & NETIF_F_FCOE_CRC) &&
2116 protocol == htons(ETH_P_FCOE)));
2119 static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
2121 if (skb->ip_summed != CHECKSUM_NONE &&
2122 !can_checksum_protocol(features, protocol)) {
2123 features &= ~NETIF_F_ALL_CSUM;
2124 features &= ~NETIF_F_SG;
2125 } else if (illegal_highdma(skb->dev, skb)) {
2126 features &= ~NETIF_F_SG;
2132 u32 netif_skb_features(struct sk_buff *skb)
2134 __be16 protocol = skb->protocol;
2135 u32 features = skb->dev->features;
2137 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2138 features &= ~NETIF_F_GSO_MASK;
2140 if (!vlan_tx_tag_present(skb)) {
2141 if (unlikely(protocol == htons(ETH_P_8021Q))) {
2142 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2143 protocol = veh->h_vlan_encapsulated_proto;
2145 return harmonize_features(skb, protocol, features);
2149 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2151 if (protocol != htons(ETH_P_8021Q)) {
2152 return harmonize_features(skb, protocol, features);
2154 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2155 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2156 return harmonize_features(skb, protocol, features);
2159 EXPORT_SYMBOL(netif_skb_features);
2162 * Returns true if either:
2163 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2164 * 2. skb is fragmented and the device does not support SG, or if
2165 * at least one of fragments is in highmem and device does not
2166 * support DMA from it.
2168 static inline int skb_needs_linearize(struct sk_buff *skb,
2171 return skb_is_nonlinear(skb) &&
2172 ((skb_has_frag_list(skb) &&
2173 !(features & NETIF_F_FRAGLIST)) ||
2174 (skb_shinfo(skb)->nr_frags &&
2175 !(features & NETIF_F_SG)));
2178 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2179 struct netdev_queue *txq)
2181 const struct net_device_ops *ops = dev->netdev_ops;
2182 int rc = NETDEV_TX_OK;
2183 unsigned int skb_len;
2185 if (likely(!skb->next)) {
2189 * If device doesn't need skb->dst, release it right now while
2190 * its hot in this cpu cache
2192 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2195 if (!list_empty(&ptype_all))
2196 dev_queue_xmit_nit(skb, dev);
2198 features = netif_skb_features(skb);
2200 if (vlan_tx_tag_present(skb) &&
2201 !(features & NETIF_F_HW_VLAN_TX)) {
2202 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2209 if (netif_needs_gso(skb, features)) {
2210 if (unlikely(dev_gso_segment(skb, features)))
2215 if (skb_needs_linearize(skb, features) &&
2216 __skb_linearize(skb))
2219 /* If packet is not checksummed and device does not
2220 * support checksumming for this protocol, complete
2221 * checksumming here.
2223 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2224 skb_set_transport_header(skb,
2225 skb_checksum_start_offset(skb));
2226 if (!(features & NETIF_F_ALL_CSUM) &&
2227 skb_checksum_help(skb))
2233 rc = ops->ndo_start_xmit(skb, dev);
2234 trace_net_dev_xmit(skb, rc, dev, skb_len);
2235 if (rc == NETDEV_TX_OK)
2236 txq_trans_update(txq);
2242 struct sk_buff *nskb = skb->next;
2244 skb->next = nskb->next;
2248 * If device doesn't need nskb->dst, release it right now while
2249 * its hot in this cpu cache
2251 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2254 skb_len = nskb->len;
2255 rc = ops->ndo_start_xmit(nskb, dev);
2256 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2257 if (unlikely(rc != NETDEV_TX_OK)) {
2258 if (rc & ~NETDEV_TX_MASK)
2259 goto out_kfree_gso_skb;
2260 nskb->next = skb->next;
2264 txq_trans_update(txq);
2265 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2266 return NETDEV_TX_BUSY;
2267 } while (skb->next);
2270 if (likely(skb->next == NULL))
2271 skb->destructor = DEV_GSO_CB(skb)->destructor;
2278 static u32 hashrnd __read_mostly;
2281 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2282 * to be used as a distribution range.
2284 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2285 unsigned int num_tx_queues)
2289 u16 qcount = num_tx_queues;
2291 if (skb_rx_queue_recorded(skb)) {
2292 hash = skb_get_rx_queue(skb);
2293 while (unlikely(hash >= num_tx_queues))
2294 hash -= num_tx_queues;
2299 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2300 qoffset = dev->tc_to_txq[tc].offset;
2301 qcount = dev->tc_to_txq[tc].count;
2304 if (skb->sk && skb->sk->sk_hash)
2305 hash = skb->sk->sk_hash;
2307 hash = (__force u16) skb->protocol;
2308 hash = jhash_1word(hash, hashrnd);
2310 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2312 EXPORT_SYMBOL(__skb_tx_hash);
2314 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2316 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2317 if (net_ratelimit()) {
2318 pr_warning("%s selects TX queue %d, but "
2319 "real number of TX queues is %d\n",
2320 dev->name, queue_index, dev->real_num_tx_queues);
2327 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2330 struct xps_dev_maps *dev_maps;
2331 struct xps_map *map;
2332 int queue_index = -1;
2335 dev_maps = rcu_dereference(dev->xps_maps);
2337 map = rcu_dereference(
2338 dev_maps->cpu_map[raw_smp_processor_id()]);
2341 queue_index = map->queues[0];
2344 if (skb->sk && skb->sk->sk_hash)
2345 hash = skb->sk->sk_hash;
2347 hash = (__force u16) skb->protocol ^
2349 hash = jhash_1word(hash, hashrnd);
2350 queue_index = map->queues[
2351 ((u64)hash * map->len) >> 32];
2353 if (unlikely(queue_index >= dev->real_num_tx_queues))
2365 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2366 struct sk_buff *skb)
2369 const struct net_device_ops *ops = dev->netdev_ops;
2371 if (dev->real_num_tx_queues == 1)
2373 else if (ops->ndo_select_queue) {
2374 queue_index = ops->ndo_select_queue(dev, skb);
2375 queue_index = dev_cap_txqueue(dev, queue_index);
2377 struct sock *sk = skb->sk;
2378 queue_index = sk_tx_queue_get(sk);
2380 if (queue_index < 0 || skb->ooo_okay ||
2381 queue_index >= dev->real_num_tx_queues) {
2382 int old_index = queue_index;
2384 queue_index = get_xps_queue(dev, skb);
2385 if (queue_index < 0)
2386 queue_index = skb_tx_hash(dev, skb);
2388 if (queue_index != old_index && sk) {
2389 struct dst_entry *dst =
2390 rcu_dereference_check(sk->sk_dst_cache, 1);
2392 if (dst && skb_dst(skb) == dst)
2393 sk_tx_queue_set(sk, queue_index);
2398 skb_set_queue_mapping(skb, queue_index);
2399 return netdev_get_tx_queue(dev, queue_index);
2402 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2403 struct net_device *dev,
2404 struct netdev_queue *txq)
2406 spinlock_t *root_lock = qdisc_lock(q);
2410 qdisc_skb_cb(skb)->pkt_len = skb->len;
2411 qdisc_calculate_pkt_len(skb, q);
2413 * Heuristic to force contended enqueues to serialize on a
2414 * separate lock before trying to get qdisc main lock.
2415 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2416 * and dequeue packets faster.
2418 contended = qdisc_is_running(q);
2419 if (unlikely(contended))
2420 spin_lock(&q->busylock);
2422 spin_lock(root_lock);
2423 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2426 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2427 qdisc_run_begin(q)) {
2429 * This is a work-conserving queue; there are no old skbs
2430 * waiting to be sent out; and the qdisc is not running -
2431 * xmit the skb directly.
2433 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2436 qdisc_bstats_update(q, skb);
2438 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2439 if (unlikely(contended)) {
2440 spin_unlock(&q->busylock);
2447 rc = NET_XMIT_SUCCESS;
2450 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2451 if (qdisc_run_begin(q)) {
2452 if (unlikely(contended)) {
2453 spin_unlock(&q->busylock);
2459 spin_unlock(root_lock);
2460 if (unlikely(contended))
2461 spin_unlock(&q->busylock);
2465 static DEFINE_PER_CPU(int, xmit_recursion);
2466 #define RECURSION_LIMIT 10
2469 * dev_queue_xmit - transmit a buffer
2470 * @skb: buffer to transmit
2472 * Queue a buffer for transmission to a network device. The caller must
2473 * have set the device and priority and built the buffer before calling
2474 * this function. The function can be called from an interrupt.
2476 * A negative errno code is returned on a failure. A success does not
2477 * guarantee the frame will be transmitted as it may be dropped due
2478 * to congestion or traffic shaping.
2480 * -----------------------------------------------------------------------------------
2481 * I notice this method can also return errors from the queue disciplines,
2482 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2485 * Regardless of the return value, the skb is consumed, so it is currently
2486 * difficult to retry a send to this method. (You can bump the ref count
2487 * before sending to hold a reference for retry if you are careful.)
2489 * When calling this method, interrupts MUST be enabled. This is because
2490 * the BH enable code must have IRQs enabled so that it will not deadlock.
2493 int dev_queue_xmit(struct sk_buff *skb)
2495 struct net_device *dev = skb->dev;
2496 struct netdev_queue *txq;
2500 /* Disable soft irqs for various locks below. Also
2501 * stops preemption for RCU.
2505 txq = dev_pick_tx(dev, skb);
2506 q = rcu_dereference_bh(txq->qdisc);
2508 #ifdef CONFIG_NET_CLS_ACT
2509 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2511 trace_net_dev_queue(skb);
2513 rc = __dev_xmit_skb(skb, q, dev, txq);
2517 /* The device has no queue. Common case for software devices:
2518 loopback, all the sorts of tunnels...
2520 Really, it is unlikely that netif_tx_lock protection is necessary
2521 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2523 However, it is possible, that they rely on protection
2526 Check this and shot the lock. It is not prone from deadlocks.
2527 Either shot noqueue qdisc, it is even simpler 8)
2529 if (dev->flags & IFF_UP) {
2530 int cpu = smp_processor_id(); /* ok because BHs are off */
2532 if (txq->xmit_lock_owner != cpu) {
2534 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2535 goto recursion_alert;
2537 HARD_TX_LOCK(dev, txq, cpu);
2539 if (!netif_tx_queue_stopped(txq)) {
2540 __this_cpu_inc(xmit_recursion);
2541 rc = dev_hard_start_xmit(skb, dev, txq);
2542 __this_cpu_dec(xmit_recursion);
2543 if (dev_xmit_complete(rc)) {
2544 HARD_TX_UNLOCK(dev, txq);
2548 HARD_TX_UNLOCK(dev, txq);
2549 if (net_ratelimit())
2550 printk(KERN_CRIT "Virtual device %s asks to "
2551 "queue packet!\n", dev->name);
2553 /* Recursion is detected! It is possible,
2557 if (net_ratelimit())
2558 printk(KERN_CRIT "Dead loop on virtual device "
2559 "%s, fix it urgently!\n", dev->name);
2564 rcu_read_unlock_bh();
2569 rcu_read_unlock_bh();
2572 EXPORT_SYMBOL(dev_queue_xmit);
2575 /*=======================================================================
2577 =======================================================================*/
2579 int netdev_max_backlog __read_mostly = 1000;
2580 int netdev_tstamp_prequeue __read_mostly = 1;
2581 int netdev_budget __read_mostly = 300;
2582 int weight_p __read_mostly = 64; /* old backlog weight */
2584 /* Called with irq disabled */
2585 static inline void ____napi_schedule(struct softnet_data *sd,
2586 struct napi_struct *napi)
2588 list_add_tail(&napi->poll_list, &sd->poll_list);
2589 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2593 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2594 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2595 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2596 * if hash is a canonical 4-tuple hash over transport ports.
2598 void __skb_get_rxhash(struct sk_buff *skb)
2600 int nhoff, hash = 0, poff;
2601 const struct ipv6hdr *ip6;
2602 const struct iphdr *ip;
2603 const struct vlan_hdr *vlan;
2612 nhoff = skb_network_offset(skb);
2613 proto = skb->protocol;
2617 case __constant_htons(ETH_P_IP):
2619 if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2622 ip = (const struct iphdr *) (skb->data + nhoff);
2625 if (ip_is_fragment(ip))
2628 ip_proto = ip->protocol;
2629 addr1 = (__force u32) ip->saddr;
2630 addr2 = (__force u32) ip->daddr;
2631 nhoff += ip->ihl * 4;
2633 case __constant_htons(ETH_P_IPV6):
2635 if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2638 ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2639 ip_proto = ip6->nexthdr;
2640 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2641 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2644 case __constant_htons(ETH_P_8021Q):
2645 if (!pskb_may_pull(skb, sizeof(*vlan) + nhoff))
2647 vlan = (const struct vlan_hdr *) (skb->data + nhoff);
2648 proto = vlan->h_vlan_encapsulated_proto;
2649 nhoff += sizeof(*vlan);
2651 case __constant_htons(ETH_P_PPP_SES):
2652 if (!pskb_may_pull(skb, PPPOE_SES_HLEN + nhoff))
2654 proto = *((__be16 *) (skb->data + nhoff +
2655 sizeof(struct pppoe_hdr)));
2656 nhoff += PPPOE_SES_HLEN;
2658 case __constant_htons(PPP_IP):
2660 case __constant_htons(PPP_IPV6):
2671 if (pskb_may_pull(skb, nhoff + 16)) {
2672 u8 *h = skb->data + nhoff;
2673 __be16 flags = *(__be16 *)h;
2676 * Only look inside GRE if version zero and no
2679 if (!(flags & (GRE_VERSION|GRE_ROUTING))) {
2680 proto = *(__be16 *)(h + 2);
2682 if (flags & GRE_CSUM)
2684 if (flags & GRE_KEY)
2686 if (flags & GRE_SEQ)
2699 poff = proto_ports_offset(ip_proto);
2702 if (pskb_may_pull(skb, nhoff + 4)) {
2703 ports.v32 = * (__force u32 *) (skb->data + nhoff);
2708 /* get a consistent hash (same value on both flow directions) */
2709 if (addr2 < addr1 ||
2711 ports.v16[1] < ports.v16[0])) {
2713 swap(ports.v16[0], ports.v16[1]);
2715 hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2722 EXPORT_SYMBOL(__skb_get_rxhash);
2726 /* One global table that all flow-based protocols share. */
2727 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2728 EXPORT_SYMBOL(rps_sock_flow_table);
2730 static struct rps_dev_flow *
2731 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2732 struct rps_dev_flow *rflow, u16 next_cpu)
2734 if (next_cpu != RPS_NO_CPU) {
2735 #ifdef CONFIG_RFS_ACCEL
2736 struct netdev_rx_queue *rxqueue;
2737 struct rps_dev_flow_table *flow_table;
2738 struct rps_dev_flow *old_rflow;
2743 /* Should we steer this flow to a different hardware queue? */
2744 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2745 !(dev->features & NETIF_F_NTUPLE))
2747 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2748 if (rxq_index == skb_get_rx_queue(skb))
2751 rxqueue = dev->_rx + rxq_index;
2752 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2755 flow_id = skb->rxhash & flow_table->mask;
2756 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2757 rxq_index, flow_id);
2761 rflow = &flow_table->flows[flow_id];
2763 if (old_rflow->filter == rflow->filter)
2764 old_rflow->filter = RPS_NO_FILTER;
2768 per_cpu(softnet_data, next_cpu).input_queue_head;
2771 rflow->cpu = next_cpu;
2776 * get_rps_cpu is called from netif_receive_skb and returns the target
2777 * CPU from the RPS map of the receiving queue for a given skb.
2778 * rcu_read_lock must be held on entry.
2780 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2781 struct rps_dev_flow **rflowp)
2783 struct netdev_rx_queue *rxqueue;
2784 struct rps_map *map;
2785 struct rps_dev_flow_table *flow_table;
2786 struct rps_sock_flow_table *sock_flow_table;
2790 if (skb_rx_queue_recorded(skb)) {
2791 u16 index = skb_get_rx_queue(skb);
2792 if (unlikely(index >= dev->real_num_rx_queues)) {
2793 WARN_ONCE(dev->real_num_rx_queues > 1,
2794 "%s received packet on queue %u, but number "
2795 "of RX queues is %u\n",
2796 dev->name, index, dev->real_num_rx_queues);
2799 rxqueue = dev->_rx + index;
2803 map = rcu_dereference(rxqueue->rps_map);
2805 if (map->len == 1 &&
2806 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2807 tcpu = map->cpus[0];
2808 if (cpu_online(tcpu))
2812 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2816 skb_reset_network_header(skb);
2817 if (!skb_get_rxhash(skb))
2820 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2821 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2822 if (flow_table && sock_flow_table) {
2824 struct rps_dev_flow *rflow;
2826 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2829 next_cpu = sock_flow_table->ents[skb->rxhash &
2830 sock_flow_table->mask];
2833 * If the desired CPU (where last recvmsg was done) is
2834 * different from current CPU (one in the rx-queue flow
2835 * table entry), switch if one of the following holds:
2836 * - Current CPU is unset (equal to RPS_NO_CPU).
2837 * - Current CPU is offline.
2838 * - The current CPU's queue tail has advanced beyond the
2839 * last packet that was enqueued using this table entry.
2840 * This guarantees that all previous packets for the flow
2841 * have been dequeued, thus preserving in order delivery.
2843 if (unlikely(tcpu != next_cpu) &&
2844 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2845 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2846 rflow->last_qtail)) >= 0)) {
2848 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2851 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2859 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2861 if (cpu_online(tcpu)) {
2871 #ifdef CONFIG_RFS_ACCEL
2874 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2875 * @dev: Device on which the filter was set
2876 * @rxq_index: RX queue index
2877 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2878 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2880 * Drivers that implement ndo_rx_flow_steer() should periodically call
2881 * this function for each installed filter and remove the filters for
2882 * which it returns %true.
2884 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2885 u32 flow_id, u16 filter_id)
2887 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2888 struct rps_dev_flow_table *flow_table;
2889 struct rps_dev_flow *rflow;
2894 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2895 if (flow_table && flow_id <= flow_table->mask) {
2896 rflow = &flow_table->flows[flow_id];
2897 cpu = ACCESS_ONCE(rflow->cpu);
2898 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2899 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2900 rflow->last_qtail) <
2901 (int)(10 * flow_table->mask)))
2907 EXPORT_SYMBOL(rps_may_expire_flow);
2909 #endif /* CONFIG_RFS_ACCEL */
2911 /* Called from hardirq (IPI) context */
2912 static void rps_trigger_softirq(void *data)
2914 struct softnet_data *sd = data;
2916 ____napi_schedule(sd, &sd->backlog);
2920 #endif /* CONFIG_RPS */
2923 * Check if this softnet_data structure is another cpu one
2924 * If yes, queue it to our IPI list and return 1
2927 static int rps_ipi_queued(struct softnet_data *sd)
2930 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2933 sd->rps_ipi_next = mysd->rps_ipi_list;
2934 mysd->rps_ipi_list = sd;
2936 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2939 #endif /* CONFIG_RPS */
2944 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2945 * queue (may be a remote CPU queue).
2947 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2948 unsigned int *qtail)
2950 struct softnet_data *sd;
2951 unsigned long flags;
2953 sd = &per_cpu(softnet_data, cpu);
2955 local_irq_save(flags);
2958 if (!netif_running(skb->dev))
2960 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2961 if (skb_queue_len(&sd->input_pkt_queue)) {
2963 __skb_queue_tail(&sd->input_pkt_queue, skb);
2964 input_queue_tail_incr_save(sd, qtail);
2966 local_irq_restore(flags);
2967 return NET_RX_SUCCESS;
2970 /* Schedule NAPI for backlog device
2971 * We can use non atomic operation since we own the queue lock
2973 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2974 if (!rps_ipi_queued(sd))
2975 ____napi_schedule(sd, &sd->backlog);
2984 local_irq_restore(flags);
2986 atomic_long_inc(&skb->dev->rx_dropped);
2992 * netif_rx - post buffer to the network code
2993 * @skb: buffer to post
2995 * This function receives a packet from a device driver and queues it for
2996 * the upper (protocol) levels to process. It always succeeds. The buffer
2997 * may be dropped during processing for congestion control or by the
3001 * NET_RX_SUCCESS (no congestion)
3002 * NET_RX_DROP (packet was dropped)
3006 int netif_rx(struct sk_buff *skb)
3010 /* if netpoll wants it, pretend we never saw it */
3011 if (netpoll_rx(skb))
3014 if (netdev_tstamp_prequeue)
3015 net_timestamp_check(skb);
3017 trace_netif_rx(skb);
3020 struct rps_dev_flow voidflow, *rflow = &voidflow;
3026 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3028 cpu = smp_processor_id();
3030 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3038 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3044 EXPORT_SYMBOL(netif_rx);
3046 int netif_rx_ni(struct sk_buff *skb)
3051 err = netif_rx(skb);
3052 if (local_softirq_pending())
3058 EXPORT_SYMBOL(netif_rx_ni);
3060 static void net_tx_action(struct softirq_action *h)
3062 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3064 if (sd->completion_queue) {
3065 struct sk_buff *clist;
3067 local_irq_disable();
3068 clist = sd->completion_queue;
3069 sd->completion_queue = NULL;
3073 struct sk_buff *skb = clist;
3074 clist = clist->next;
3076 WARN_ON(atomic_read(&skb->users));
3077 trace_kfree_skb(skb, net_tx_action);
3082 if (sd->output_queue) {
3085 local_irq_disable();
3086 head = sd->output_queue;
3087 sd->output_queue = NULL;
3088 sd->output_queue_tailp = &sd->output_queue;
3092 struct Qdisc *q = head;
3093 spinlock_t *root_lock;
3095 head = head->next_sched;
3097 root_lock = qdisc_lock(q);
3098 if (spin_trylock(root_lock)) {
3099 smp_mb__before_clear_bit();
3100 clear_bit(__QDISC_STATE_SCHED,
3103 spin_unlock(root_lock);
3105 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3107 __netif_reschedule(q);
3109 smp_mb__before_clear_bit();
3110 clear_bit(__QDISC_STATE_SCHED,
3118 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3119 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3120 /* This hook is defined here for ATM LANE */
3121 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3122 unsigned char *addr) __read_mostly;
3123 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3126 #ifdef CONFIG_NET_CLS_ACT
3127 /* TODO: Maybe we should just force sch_ingress to be compiled in
3128 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3129 * a compare and 2 stores extra right now if we dont have it on
3130 * but have CONFIG_NET_CLS_ACT
3131 * NOTE: This doesn't stop any functionality; if you dont have
3132 * the ingress scheduler, you just can't add policies on ingress.
3135 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3137 struct net_device *dev = skb->dev;
3138 u32 ttl = G_TC_RTTL(skb->tc_verd);
3139 int result = TC_ACT_OK;
3142 if (unlikely(MAX_RED_LOOP < ttl++)) {
3143 if (net_ratelimit())
3144 pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3145 skb->skb_iif, dev->ifindex);
3149 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3150 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3153 if (q != &noop_qdisc) {
3154 spin_lock(qdisc_lock(q));
3155 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3156 result = qdisc_enqueue_root(skb, q);
3157 spin_unlock(qdisc_lock(q));
3163 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3164 struct packet_type **pt_prev,
3165 int *ret, struct net_device *orig_dev)
3167 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3169 if (!rxq || rxq->qdisc == &noop_qdisc)
3173 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3177 switch (ing_filter(skb, rxq)) {
3191 * netdev_rx_handler_register - register receive handler
3192 * @dev: device to register a handler for
3193 * @rx_handler: receive handler to register
3194 * @rx_handler_data: data pointer that is used by rx handler
3196 * Register a receive hander for a device. This handler will then be
3197 * called from __netif_receive_skb. A negative errno code is returned
3200 * The caller must hold the rtnl_mutex.
3202 * For a general description of rx_handler, see enum rx_handler_result.
3204 int netdev_rx_handler_register(struct net_device *dev,
3205 rx_handler_func_t *rx_handler,
3206 void *rx_handler_data)
3210 if (dev->rx_handler)
3213 /* Note: rx_handler_data must be set before rx_handler */
3214 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3215 rcu_assign_pointer(dev->rx_handler, rx_handler);
3219 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3222 * netdev_rx_handler_unregister - unregister receive handler
3223 * @dev: device to unregister a handler from
3225 * Unregister a receive hander from a device.
3227 * The caller must hold the rtnl_mutex.
3229 void netdev_rx_handler_unregister(struct net_device *dev)
3233 RCU_INIT_POINTER(dev->rx_handler, NULL);
3234 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3235 * section has a guarantee to see a non NULL rx_handler_data
3239 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3241 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3243 static int __netif_receive_skb(struct sk_buff *skb)
3245 struct packet_type *ptype, *pt_prev;
3246 rx_handler_func_t *rx_handler;
3247 struct net_device *orig_dev;
3248 struct net_device *null_or_dev;
3249 bool deliver_exact = false;
3250 int ret = NET_RX_DROP;
3253 if (!netdev_tstamp_prequeue)
3254 net_timestamp_check(skb);
3256 trace_netif_receive_skb(skb);
3258 /* if we've gotten here through NAPI, check netpoll */
3259 if (netpoll_receive_skb(skb))
3263 skb->skb_iif = skb->dev->ifindex;
3264 orig_dev = skb->dev;
3266 skb_reset_network_header(skb);
3267 skb_reset_transport_header(skb);
3268 skb_reset_mac_len(skb);
3274 __this_cpu_inc(softnet_data.processed);
3276 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3277 skb = vlan_untag(skb);
3282 #ifdef CONFIG_NET_CLS_ACT
3283 if (skb->tc_verd & TC_NCLS) {
3284 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3289 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3290 if (!ptype->dev || ptype->dev == skb->dev) {
3292 ret = deliver_skb(skb, pt_prev, orig_dev);
3297 #ifdef CONFIG_NET_CLS_ACT
3298 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3304 if (vlan_tx_tag_present(skb)) {
3306 ret = deliver_skb(skb, pt_prev, orig_dev);
3309 if (vlan_do_receive(&skb))
3311 else if (unlikely(!skb))
3315 rx_handler = rcu_dereference(skb->dev->rx_handler);