Linux Audio

Check our new training course

Loading...
v3.5.6
   1/*
   2 * 	NET3	Protocol independent device support routines.
   3 *
   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.
   8 *
   9 *	Derived from the non IP parts of dev.c 1.0.19
  10 * 		Authors:	Ross Biro
  11 *				Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12 *				Mark Evans, <evansmp@uhura.aston.ac.uk>
  13 *
  14 *	Additional Authors:
  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>
  21 *
  22 *	Changes:
  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
  34 *					drivers
  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
  44 *					call a packet.
  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
  50 *					changes.
  51 *		Rudi Cilibrasi	:	Pass the right thing to
  52 *					set_mac_address()
  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
  58 *					1 device.
  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
  66 *					the backlog queue.
  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
  73 */
  74
  75#include <asm/uaccess.h>
 
  76#include <linux/bitops.h>
  77#include <linux/capability.h>
  78#include <linux/cpu.h>
  79#include <linux/types.h>
  80#include <linux/kernel.h>
  81#include <linux/hash.h>
  82#include <linux/slab.h>
  83#include <linux/sched.h>
  84#include <linux/mutex.h>
  85#include <linux/string.h>
  86#include <linux/mm.h>
  87#include <linux/socket.h>
  88#include <linux/sockios.h>
  89#include <linux/errno.h>
  90#include <linux/interrupt.h>
  91#include <linux/if_ether.h>
  92#include <linux/netdevice.h>
  93#include <linux/etherdevice.h>
  94#include <linux/ethtool.h>
  95#include <linux/notifier.h>
  96#include <linux/skbuff.h>
  97#include <net/net_namespace.h>
  98#include <net/sock.h>
  99#include <linux/rtnetlink.h>
 100#include <linux/proc_fs.h>
 101#include <linux/seq_file.h>
 102#include <linux/stat.h>
 103#include <net/dst.h>
 104#include <net/pkt_sched.h>
 105#include <net/checksum.h>
 106#include <net/xfrm.h>
 107#include <linux/highmem.h>
 108#include <linux/init.h>
 109#include <linux/kmod.h>
 110#include <linux/module.h>
 111#include <linux/netpoll.h>
 112#include <linux/rcupdate.h>
 113#include <linux/delay.h>
 114#include <net/wext.h>
 115#include <net/iw_handler.h>
 116#include <asm/current.h>
 117#include <linux/audit.h>
 118#include <linux/dmaengine.h>
 119#include <linux/err.h>
 120#include <linux/ctype.h>
 121#include <linux/if_arp.h>
 122#include <linux/if_vlan.h>
 123#include <linux/ip.h>
 124#include <net/ip.h>
 125#include <linux/ipv6.h>
 126#include <linux/in.h>
 127#include <linux/jhash.h>
 128#include <linux/random.h>
 129#include <trace/events/napi.h>
 130#include <trace/events/net.h>
 131#include <trace/events/skb.h>
 132#include <linux/pci.h>
 133#include <linux/inetdevice.h>
 134#include <linux/cpu_rmap.h>
 135#include <linux/net_tstamp.h>
 136#include <linux/static_key.h>
 137#include <net/flow_keys.h>
 138
 139#include "net-sysfs.h"
 140
 141/* Instead of increasing this, you should create a hash table. */
 142#define MAX_GRO_SKBS 8
 143
 144/* This should be increased if a protocol with a bigger head is added. */
 145#define GRO_MAX_HEAD (MAX_HEADER + 128)
 146
 147/*
 148 *	The list of packet types we will receive (as opposed to discard)
 149 *	and the routines to invoke.
 150 *
 151 *	Why 16. Because with 16 the only overlap we get on a hash of the
 152 *	low nibble of the protocol value is RARP/SNAP/X.25.
 153 *
 154 *      NOTE:  That is no longer true with the addition of VLAN tags.  Not
 155 *             sure which should go first, but I bet it won't make much
 156 *             difference if we are running VLANs.  The good news is that
 157 *             this protocol won't be in the list unless compiled in, so
 158 *             the average user (w/out VLANs) will not be adversely affected.
 159 *             --BLG
 160 *
 161 *		0800	IP
 162 *		8100    802.1Q VLAN
 163 *		0001	802.3
 164 *		0002	AX.25
 165 *		0004	802.2
 166 *		8035	RARP
 167 *		0005	SNAP
 168 *		0805	X.25
 169 *		0806	ARP
 170 *		8137	IPX
 171 *		0009	Localtalk
 172 *		86DD	IPv6
 173 */
 174
 175#define PTYPE_HASH_SIZE	(16)
 176#define PTYPE_HASH_MASK	(PTYPE_HASH_SIZE - 1)
 177
 178static DEFINE_SPINLOCK(ptype_lock);
 179static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
 180static struct list_head ptype_all __read_mostly;	/* Taps */
 181
 182/*
 183 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
 184 * semaphore.
 185 *
 186 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
 187 *
 188 * Writers must hold the rtnl semaphore while they loop through the
 189 * dev_base_head list, and hold dev_base_lock for writing when they do the
 190 * actual updates.  This allows pure readers to access the list even
 191 * while a writer is preparing to update it.
 192 *
 193 * To put it another way, dev_base_lock is held for writing only to
 194 * protect against pure readers; the rtnl semaphore provides the
 195 * protection against other writers.
 196 *
 197 * See, for example usages, register_netdevice() and
 198 * unregister_netdevice(), which must be called with the rtnl
 199 * semaphore held.
 200 */
 201DEFINE_RWLOCK(dev_base_lock);
 202EXPORT_SYMBOL(dev_base_lock);
 203
 204static inline void dev_base_seq_inc(struct net *net)
 205{
 206	while (++net->dev_base_seq == 0);
 207}
 208
 209static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
 210{
 211	unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
 212
 213	return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
 214}
 215
 216static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
 217{
 218	return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
 219}
 220
 221static inline void rps_lock(struct softnet_data *sd)
 222{
 223#ifdef CONFIG_RPS
 224	spin_lock(&sd->input_pkt_queue.lock);
 225#endif
 226}
 227
 228static inline void rps_unlock(struct softnet_data *sd)
 229{
 230#ifdef CONFIG_RPS
 231	spin_unlock(&sd->input_pkt_queue.lock);
 232#endif
 233}
 234
 235/* Device list insertion */
 236static int list_netdevice(struct net_device *dev)
 237{
 238	struct net *net = dev_net(dev);
 239
 240	ASSERT_RTNL();
 241
 242	write_lock_bh(&dev_base_lock);
 243	list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
 244	hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
 245	hlist_add_head_rcu(&dev->index_hlist,
 246			   dev_index_hash(net, dev->ifindex));
 247	write_unlock_bh(&dev_base_lock);
 248
 249	dev_base_seq_inc(net);
 250
 251	return 0;
 252}
 253
 254/* Device list removal
 255 * caller must respect a RCU grace period before freeing/reusing dev
 256 */
 257static void unlist_netdevice(struct net_device *dev)
 258{
 259	ASSERT_RTNL();
 260
 261	/* Unlink dev from the device chain */
 262	write_lock_bh(&dev_base_lock);
 263	list_del_rcu(&dev->dev_list);
 264	hlist_del_rcu(&dev->name_hlist);
 265	hlist_del_rcu(&dev->index_hlist);
 266	write_unlock_bh(&dev_base_lock);
 267
 268	dev_base_seq_inc(dev_net(dev));
 269}
 270
 271/*
 272 *	Our notifier list
 273 */
 274
 275static RAW_NOTIFIER_HEAD(netdev_chain);
 276
 277/*
 278 *	Device drivers call our routines to queue packets here. We empty the
 279 *	queue in the local softnet handler.
 280 */
 281
 282DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
 283EXPORT_PER_CPU_SYMBOL(softnet_data);
 284
 285#ifdef CONFIG_LOCKDEP
 286/*
 287 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
 288 * according to dev->type
 289 */
 290static const unsigned short netdev_lock_type[] =
 291	{ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
 292	 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
 293	 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
 294	 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
 295	 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
 296	 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
 297	 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
 298	 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
 299	 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
 300	 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
 301	 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
 302	 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
 303	 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
 304	 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
 305	 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
 
 306
 307static const char *const netdev_lock_name[] =
 308	{"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
 309	 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
 310	 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
 311	 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
 312	 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
 313	 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
 314	 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
 315	 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
 316	 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
 317	 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
 318	 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
 319	 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
 320	 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
 321	 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
 322	 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
 
 323
 324static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
 325static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
 326
 327static inline unsigned short netdev_lock_pos(unsigned short dev_type)
 328{
 329	int i;
 330
 331	for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
 332		if (netdev_lock_type[i] == dev_type)
 333			return i;
 334	/* the last key is used by default */
 335	return ARRAY_SIZE(netdev_lock_type) - 1;
 336}
 337
 338static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
 339						 unsigned short dev_type)
 340{
 341	int i;
 342
 343	i = netdev_lock_pos(dev_type);
 344	lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
 345				   netdev_lock_name[i]);
 346}
 347
 348static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
 349{
 350	int i;
 351
 352	i = netdev_lock_pos(dev->type);
 353	lockdep_set_class_and_name(&dev->addr_list_lock,
 354				   &netdev_addr_lock_key[i],
 355				   netdev_lock_name[i]);
 356}
 357#else
 358static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
 359						 unsigned short dev_type)
 360{
 361}
 362static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
 363{
 364}
 365#endif
 366
 367/*******************************************************************************
 368
 369		Protocol management and registration routines
 370
 371*******************************************************************************/
 372
 373/*
 374 *	Add a protocol ID to the list. Now that the input handler is
 375 *	smarter we can dispense with all the messy stuff that used to be
 376 *	here.
 377 *
 378 *	BEWARE!!! Protocol handlers, mangling input packets,
 379 *	MUST BE last in hash buckets and checking protocol handlers
 380 *	MUST start from promiscuous ptype_all chain in net_bh.
 381 *	It is true now, do not change it.
 382 *	Explanation follows: if protocol handler, mangling packet, will
 383 *	be the first on list, it is not able to sense, that packet
 384 *	is cloned and should be copied-on-write, so that it will
 385 *	change it and subsequent readers will get broken packet.
 386 *							--ANK (980803)
 387 */
 388
 389static inline struct list_head *ptype_head(const struct packet_type *pt)
 390{
 391	if (pt->type == htons(ETH_P_ALL))
 392		return &ptype_all;
 393	else
 394		return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
 395}
 396
 397/**
 398 *	dev_add_pack - add packet handler
 399 *	@pt: packet type declaration
 400 *
 401 *	Add a protocol handler to the networking stack. The passed &packet_type
 402 *	is linked into kernel lists and may not be freed until it has been
 403 *	removed from the kernel lists.
 404 *
 405 *	This call does not sleep therefore it can not
 406 *	guarantee all CPU's that are in middle of receiving packets
 407 *	will see the new packet type (until the next received packet).
 408 */
 409
 410void dev_add_pack(struct packet_type *pt)
 411{
 412	struct list_head *head = ptype_head(pt);
 413
 414	spin_lock(&ptype_lock);
 415	list_add_rcu(&pt->list, head);
 416	spin_unlock(&ptype_lock);
 417}
 418EXPORT_SYMBOL(dev_add_pack);
 419
 420/**
 421 *	__dev_remove_pack	 - remove packet handler
 422 *	@pt: packet type declaration
 423 *
 424 *	Remove a protocol handler that was previously added to the kernel
 425 *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
 426 *	from the kernel lists and can be freed or reused once this function
 427 *	returns.
 428 *
 429 *      The packet type might still be in use by receivers
 430 *	and must not be freed until after all the CPU's have gone
 431 *	through a quiescent state.
 432 */
 433void __dev_remove_pack(struct packet_type *pt)
 434{
 435	struct list_head *head = ptype_head(pt);
 436	struct packet_type *pt1;
 437
 438	spin_lock(&ptype_lock);
 439
 440	list_for_each_entry(pt1, head, list) {
 441		if (pt == pt1) {
 442			list_del_rcu(&pt->list);
 443			goto out;
 444		}
 445	}
 446
 447	pr_warn("dev_remove_pack: %p not found\n", pt);
 448out:
 449	spin_unlock(&ptype_lock);
 450}
 451EXPORT_SYMBOL(__dev_remove_pack);
 452
 453/**
 454 *	dev_remove_pack	 - remove packet handler
 455 *	@pt: packet type declaration
 456 *
 457 *	Remove a protocol handler that was previously added to the kernel
 458 *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
 459 *	from the kernel lists and can be freed or reused once this function
 460 *	returns.
 461 *
 462 *	This call sleeps to guarantee that no CPU is looking at the packet
 463 *	type after return.
 464 */
 465void dev_remove_pack(struct packet_type *pt)
 466{
 467	__dev_remove_pack(pt);
 468
 469	synchronize_net();
 470}
 471EXPORT_SYMBOL(dev_remove_pack);
 472
 473/******************************************************************************
 474
 475		      Device Boot-time Settings Routines
 476
 477*******************************************************************************/
 478
 479/* Boot time configuration table */
 480static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
 481
 482/**
 483 *	netdev_boot_setup_add	- add new setup entry
 484 *	@name: name of the device
 485 *	@map: configured settings for the device
 486 *
 487 *	Adds new setup entry to the dev_boot_setup list.  The function
 488 *	returns 0 on error and 1 on success.  This is a generic routine to
 489 *	all netdevices.
 490 */
 491static int netdev_boot_setup_add(char *name, struct ifmap *map)
 492{
 493	struct netdev_boot_setup *s;
 494	int i;
 495
 496	s = dev_boot_setup;
 497	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
 498		if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
 499			memset(s[i].name, 0, sizeof(s[i].name));
 500			strlcpy(s[i].name, name, IFNAMSIZ);
 501			memcpy(&s[i].map, map, sizeof(s[i].map));
 502			break;
 503		}
 504	}
 505
 506	return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
 507}
 508
 509/**
 510 *	netdev_boot_setup_check	- check boot time settings
 511 *	@dev: the netdevice
 512 *
 513 * 	Check boot time settings for the device.
 514 *	The found settings are set for the device to be used
 515 *	later in the device probing.
 516 *	Returns 0 if no settings found, 1 if they are.
 517 */
 518int netdev_boot_setup_check(struct net_device *dev)
 519{
 520	struct netdev_boot_setup *s = dev_boot_setup;
 521	int i;
 522
 523	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
 524		if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
 525		    !strcmp(dev->name, s[i].name)) {
 526			dev->irq 	= s[i].map.irq;
 527			dev->base_addr 	= s[i].map.base_addr;
 528			dev->mem_start 	= s[i].map.mem_start;
 529			dev->mem_end 	= s[i].map.mem_end;
 530			return 1;
 531		}
 532	}
 533	return 0;
 534}
 535EXPORT_SYMBOL(netdev_boot_setup_check);
 536
 537
 538/**
 539 *	netdev_boot_base	- get address from boot time settings
 540 *	@prefix: prefix for network device
 541 *	@unit: id for network device
 542 *
 543 * 	Check boot time settings for the base address of device.
 544 *	The found settings are set for the device to be used
 545 *	later in the device probing.
 546 *	Returns 0 if no settings found.
 547 */
 548unsigned long netdev_boot_base(const char *prefix, int unit)
 549{
 550	const struct netdev_boot_setup *s = dev_boot_setup;
 551	char name[IFNAMSIZ];
 552	int i;
 553
 554	sprintf(name, "%s%d", prefix, unit);
 555
 556	/*
 557	 * If device already registered then return base of 1
 558	 * to indicate not to probe for this interface
 559	 */
 560	if (__dev_get_by_name(&init_net, name))
 561		return 1;
 562
 563	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
 564		if (!strcmp(name, s[i].name))
 565			return s[i].map.base_addr;
 566	return 0;
 567}
 568
 569/*
 570 * Saves at boot time configured settings for any netdevice.
 571 */
 572int __init netdev_boot_setup(char *str)
 573{
 574	int ints[5];
 575	struct ifmap map;
 576
 577	str = get_options(str, ARRAY_SIZE(ints), ints);
 578	if (!str || !*str)
 579		return 0;
 580
 581	/* Save settings */
 582	memset(&map, 0, sizeof(map));
 583	if (ints[0] > 0)
 584		map.irq = ints[1];
 585	if (ints[0] > 1)
 586		map.base_addr = ints[2];
 587	if (ints[0] > 2)
 588		map.mem_start = ints[3];
 589	if (ints[0] > 3)
 590		map.mem_end = ints[4];
 591
 592	/* Add new entry to the list */
 593	return netdev_boot_setup_add(str, &map);
 594}
 595
 596__setup("netdev=", netdev_boot_setup);
 597
 598/*******************************************************************************
 599
 600			    Device Interface Subroutines
 601
 602*******************************************************************************/
 603
 604/**
 605 *	__dev_get_by_name	- find a device by its name
 606 *	@net: the applicable net namespace
 607 *	@name: name to find
 608 *
 609 *	Find an interface by name. Must be called under RTNL semaphore
 610 *	or @dev_base_lock. If the name is found a pointer to the device
 611 *	is returned. If the name is not found then %NULL is returned. The
 612 *	reference counters are not incremented so the caller must be
 613 *	careful with locks.
 614 */
 615
 616struct net_device *__dev_get_by_name(struct net *net, const char *name)
 617{
 618	struct hlist_node *p;
 619	struct net_device *dev;
 620	struct hlist_head *head = dev_name_hash(net, name);
 621
 622	hlist_for_each_entry(dev, p, head, name_hlist)
 623		if (!strncmp(dev->name, name, IFNAMSIZ))
 624			return dev;
 625
 626	return NULL;
 627}
 628EXPORT_SYMBOL(__dev_get_by_name);
 629
 630/**
 631 *	dev_get_by_name_rcu	- find a device by its name
 632 *	@net: the applicable net namespace
 633 *	@name: name to find
 634 *
 635 *	Find an interface by name.
 636 *	If the name is found a pointer to the device is returned.
 637 * 	If the name is not found then %NULL is returned.
 638 *	The reference counters are not incremented so the caller must be
 639 *	careful with locks. The caller must hold RCU lock.
 640 */
 641
 642struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
 643{
 644	struct hlist_node *p;
 645	struct net_device *dev;
 646	struct hlist_head *head = dev_name_hash(net, name);
 647
 648	hlist_for_each_entry_rcu(dev, p, head, name_hlist)
 649		if (!strncmp(dev->name, name, IFNAMSIZ))
 650			return dev;
 651
 652	return NULL;
 653}
 654EXPORT_SYMBOL(dev_get_by_name_rcu);
 655
 656/**
 657 *	dev_get_by_name		- find a device by its name
 658 *	@net: the applicable net namespace
 659 *	@name: name to find
 660 *
 661 *	Find an interface by name. This can be called from any
 662 *	context and does its own locking. The returned handle has
 663 *	the usage count incremented and the caller must use dev_put() to
 664 *	release it when it is no longer needed. %NULL is returned if no
 665 *	matching device is found.
 666 */
 667
 668struct net_device *dev_get_by_name(struct net *net, const char *name)
 669{
 670	struct net_device *dev;
 671
 672	rcu_read_lock();
 673	dev = dev_get_by_name_rcu(net, name);
 674	if (dev)
 675		dev_hold(dev);
 676	rcu_read_unlock();
 677	return dev;
 678}
 679EXPORT_SYMBOL(dev_get_by_name);
 680
 681/**
 682 *	__dev_get_by_index - find a device by its ifindex
 683 *	@net: the applicable net namespace
 684 *	@ifindex: index of device
 685 *
 686 *	Search for an interface by index. Returns %NULL if the device
 687 *	is not found or a pointer to the device. The device has not
 688 *	had its reference counter increased so the caller must be careful
 689 *	about locking. The caller must hold either the RTNL semaphore
 690 *	or @dev_base_lock.
 691 */
 692
 693struct net_device *__dev_get_by_index(struct net *net, int ifindex)
 694{
 695	struct hlist_node *p;
 696	struct net_device *dev;
 697	struct hlist_head *head = dev_index_hash(net, ifindex);
 698
 699	hlist_for_each_entry(dev, p, head, index_hlist)
 700		if (dev->ifindex == ifindex)
 701			return dev;
 702
 703	return NULL;
 704}
 705EXPORT_SYMBOL(__dev_get_by_index);
 706
 707/**
 708 *	dev_get_by_index_rcu - find a device by its ifindex
 709 *	@net: the applicable net namespace
 710 *	@ifindex: index of device
 711 *
 712 *	Search for an interface by index. Returns %NULL if the device
 713 *	is not found or a pointer to the device. The device has not
 714 *	had its reference counter increased so the caller must be careful
 715 *	about locking. The caller must hold RCU lock.
 716 */
 717
 718struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
 719{
 720	struct hlist_node *p;
 721	struct net_device *dev;
 722	struct hlist_head *head = dev_index_hash(net, ifindex);
 723
 724	hlist_for_each_entry_rcu(dev, p, head, index_hlist)
 725		if (dev->ifindex == ifindex)
 726			return dev;
 727
 728	return NULL;
 729}
 730EXPORT_SYMBOL(dev_get_by_index_rcu);
 731
 732
 733/**
 734 *	dev_get_by_index - find a device by its ifindex
 735 *	@net: the applicable net namespace
 736 *	@ifindex: index of device
 737 *
 738 *	Search for an interface by index. Returns NULL if the device
 739 *	is not found or a pointer to the device. The device returned has
 740 *	had a reference added and the pointer is safe until the user calls
 741 *	dev_put to indicate they have finished with it.
 742 */
 743
 744struct net_device *dev_get_by_index(struct net *net, int ifindex)
 745{
 746	struct net_device *dev;
 747
 748	rcu_read_lock();
 749	dev = dev_get_by_index_rcu(net, ifindex);
 750	if (dev)
 751		dev_hold(dev);
 752	rcu_read_unlock();
 753	return dev;
 754}
 755EXPORT_SYMBOL(dev_get_by_index);
 756
 757/**
 758 *	dev_getbyhwaddr_rcu - find a device by its hardware address
 759 *	@net: the applicable net namespace
 760 *	@type: media type of device
 761 *	@ha: hardware address
 762 *
 763 *	Search for an interface by MAC address. Returns NULL if the device
 764 *	is not found or a pointer to the device.
 765 *	The caller must hold RCU or RTNL.
 766 *	The returned device has not had its ref count increased
 767 *	and the caller must therefore be careful about locking
 768 *
 769 */
 770
 771struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
 772				       const char *ha)
 773{
 774	struct net_device *dev;
 775
 776	for_each_netdev_rcu(net, dev)
 777		if (dev->type == type &&
 778		    !memcmp(dev->dev_addr, ha, dev->addr_len))
 779			return dev;
 780
 781	return NULL;
 782}
 783EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
 784
 785struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
 786{
 787	struct net_device *dev;
 788
 789	ASSERT_RTNL();
 790	for_each_netdev(net, dev)
 791		if (dev->type == type)
 792			return dev;
 793
 794	return NULL;
 795}
 796EXPORT_SYMBOL(__dev_getfirstbyhwtype);
 797
 798struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
 799{
 800	struct net_device *dev, *ret = NULL;
 801
 802	rcu_read_lock();
 803	for_each_netdev_rcu(net, dev)
 804		if (dev->type == type) {
 805			dev_hold(dev);
 806			ret = dev;
 807			break;
 808		}
 809	rcu_read_unlock();
 810	return ret;
 811}
 812EXPORT_SYMBOL(dev_getfirstbyhwtype);
 813
 814/**
 815 *	dev_get_by_flags_rcu - find any device with given flags
 816 *	@net: the applicable net namespace
 817 *	@if_flags: IFF_* values
 818 *	@mask: bitmask of bits in if_flags to check
 819 *
 820 *	Search for any interface with the given flags. Returns NULL if a device
 821 *	is not found or a pointer to the device. Must be called inside
 822 *	rcu_read_lock(), and result refcount is unchanged.
 823 */
 824
 825struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
 826				    unsigned short mask)
 827{
 828	struct net_device *dev, *ret;
 829
 830	ret = NULL;
 831	for_each_netdev_rcu(net, dev) {
 832		if (((dev->flags ^ if_flags) & mask) == 0) {
 833			ret = dev;
 834			break;
 835		}
 836	}
 837	return ret;
 838}
 839EXPORT_SYMBOL(dev_get_by_flags_rcu);
 840
 841/**
 842 *	dev_valid_name - check if name is okay for network device
 843 *	@name: name string
 844 *
 845 *	Network device names need to be valid file names to
 846 *	to allow sysfs to work.  We also disallow any kind of
 847 *	whitespace.
 848 */
 849bool dev_valid_name(const char *name)
 850{
 851	if (*name == '\0')
 852		return false;
 853	if (strlen(name) >= IFNAMSIZ)
 854		return false;
 855	if (!strcmp(name, ".") || !strcmp(name, ".."))
 856		return false;
 857
 858	while (*name) {
 859		if (*name == '/' || isspace(*name))
 860			return false;
 861		name++;
 862	}
 863	return true;
 864}
 865EXPORT_SYMBOL(dev_valid_name);
 866
 867/**
 868 *	__dev_alloc_name - allocate a name for a device
 869 *	@net: network namespace to allocate the device name in
 870 *	@name: name format string
 871 *	@buf:  scratch buffer and result name string
 872 *
 873 *	Passed a format string - eg "lt%d" it will try and find a suitable
 874 *	id. It scans list of devices to build up a free map, then chooses
 875 *	the first empty slot. The caller must hold the dev_base or rtnl lock
 876 *	while allocating the name and adding the device in order to avoid
 877 *	duplicates.
 878 *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
 879 *	Returns the number of the unit assigned or a negative errno code.
 880 */
 881
 882static int __dev_alloc_name(struct net *net, const char *name, char *buf)
 883{
 884	int i = 0;
 885	const char *p;
 886	const int max_netdevices = 8*PAGE_SIZE;
 887	unsigned long *inuse;
 888	struct net_device *d;
 889
 890	p = strnchr(name, IFNAMSIZ-1, '%');
 891	if (p) {
 892		/*
 893		 * Verify the string as this thing may have come from
 894		 * the user.  There must be either one "%d" and no other "%"
 895		 * characters.
 896		 */
 897		if (p[1] != 'd' || strchr(p + 2, '%'))
 898			return -EINVAL;
 899
 900		/* Use one page as a bit array of possible slots */
 901		inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
 902		if (!inuse)
 903			return -ENOMEM;
 904
 905		for_each_netdev(net, d) {
 906			if (!sscanf(d->name, name, &i))
 907				continue;
 908			if (i < 0 || i >= max_netdevices)
 909				continue;
 910
 911			/*  avoid cases where sscanf is not exact inverse of printf */
 912			snprintf(buf, IFNAMSIZ, name, i);
 913			if (!strncmp(buf, d->name, IFNAMSIZ))
 914				set_bit(i, inuse);
 915		}
 916
 917		i = find_first_zero_bit(inuse, max_netdevices);
 918		free_page((unsigned long) inuse);
 919	}
 920
 921	if (buf != name)
 922		snprintf(buf, IFNAMSIZ, name, i);
 923	if (!__dev_get_by_name(net, buf))
 924		return i;
 925
 926	/* It is possible to run out of possible slots
 927	 * when the name is long and there isn't enough space left
 928	 * for the digits, or if all bits are used.
 929	 */
 930	return -ENFILE;
 931}
 932
 933/**
 934 *	dev_alloc_name - allocate a name for a device
 935 *	@dev: device
 936 *	@name: name format string
 937 *
 938 *	Passed a format string - eg "lt%d" it will try and find a suitable
 939 *	id. It scans list of devices to build up a free map, then chooses
 940 *	the first empty slot. The caller must hold the dev_base or rtnl lock
 941 *	while allocating the name and adding the device in order to avoid
 942 *	duplicates.
 943 *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
 944 *	Returns the number of the unit assigned or a negative errno code.
 945 */
 946
 947int dev_alloc_name(struct net_device *dev, const char *name)
 948{
 949	char buf[IFNAMSIZ];
 950	struct net *net;
 951	int ret;
 952
 953	BUG_ON(!dev_net(dev));
 954	net = dev_net(dev);
 955	ret = __dev_alloc_name(net, name, buf);
 956	if (ret >= 0)
 957		strlcpy(dev->name, buf, IFNAMSIZ);
 958	return ret;
 959}
 960EXPORT_SYMBOL(dev_alloc_name);
 961
 962static int dev_get_valid_name(struct net_device *dev, const char *name)
 963{
 964	struct net *net;
 965
 966	BUG_ON(!dev_net(dev));
 967	net = dev_net(dev);
 968
 969	if (!dev_valid_name(name))
 970		return -EINVAL;
 971
 972	if (strchr(name, '%'))
 973		return dev_alloc_name(dev, name);
 974	else if (__dev_get_by_name(net, name))
 975		return -EEXIST;
 976	else if (dev->name != name)
 977		strlcpy(dev->name, name, IFNAMSIZ);
 978
 979	return 0;
 980}
 981
 982/**
 983 *	dev_change_name - change name of a device
 984 *	@dev: device
 985 *	@newname: name (or format string) must be at least IFNAMSIZ
 986 *
 987 *	Change name of a device, can pass format strings "eth%d".
 988 *	for wildcarding.
 989 */
 990int dev_change_name(struct net_device *dev, const char *newname)
 991{
 992	char oldname[IFNAMSIZ];
 993	int err = 0;
 994	int ret;
 995	struct net *net;
 996
 997	ASSERT_RTNL();
 998	BUG_ON(!dev_net(dev));
 999
1000	net = dev_net(dev);
1001	if (dev->flags & IFF_UP)
1002		return -EBUSY;
1003
1004	if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1005		return 0;
1006
1007	memcpy(oldname, dev->name, IFNAMSIZ);
1008
1009	err = dev_get_valid_name(dev, newname);
1010	if (err < 0)
1011		return err;
1012
1013rollback:
1014	ret = device_rename(&dev->dev, dev->name);
1015	if (ret) {
1016		memcpy(dev->name, oldname, IFNAMSIZ);
1017		return ret;
1018	}
1019
1020	write_lock_bh(&dev_base_lock);
1021	hlist_del_rcu(&dev->name_hlist);
1022	write_unlock_bh(&dev_base_lock);
1023
1024	synchronize_rcu();
1025
1026	write_lock_bh(&dev_base_lock);
1027	hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1028	write_unlock_bh(&dev_base_lock);
1029
1030	ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1031	ret = notifier_to_errno(ret);
1032
1033	if (ret) {
1034		/* err >= 0 after dev_alloc_name() or stores the first errno */
1035		if (err >= 0) {
1036			err = ret;
1037			memcpy(dev->name, oldname, IFNAMSIZ);
1038			goto rollback;
1039		} else {
1040			pr_err("%s: name change rollback failed: %d\n",
 
1041			       dev->name, ret);
1042		}
1043	}
1044
1045	return err;
1046}
1047
1048/**
1049 *	dev_set_alias - change ifalias of a device
1050 *	@dev: device
1051 *	@alias: name up to IFALIASZ
1052 *	@len: limit of bytes to copy from info
1053 *
1054 *	Set ifalias for a device,
1055 */
1056int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1057{
1058	char *new_ifalias;
1059
1060	ASSERT_RTNL();
1061
1062	if (len >= IFALIASZ)
1063		return -EINVAL;
1064
1065	if (!len) {
1066		if (dev->ifalias) {
1067			kfree(dev->ifalias);
1068			dev->ifalias = NULL;
1069		}
1070		return 0;
1071	}
1072
1073	new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1074	if (!new_ifalias)
1075		return -ENOMEM;
1076	dev->ifalias = new_ifalias;
1077
1078	strlcpy(dev->ifalias, alias, len+1);
1079	return len;
1080}
1081
1082
1083/**
1084 *	netdev_features_change - device changes features
1085 *	@dev: device to cause notification
1086 *
1087 *	Called to indicate a device has changed features.
1088 */
1089void netdev_features_change(struct net_device *dev)
1090{
1091	call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1092}
1093EXPORT_SYMBOL(netdev_features_change);
1094
1095/**
1096 *	netdev_state_change - device changes state
1097 *	@dev: device to cause notification
1098 *
1099 *	Called to indicate a device has changed state. This function calls
1100 *	the notifier chains for netdev_chain and sends a NEWLINK message
1101 *	to the routing socket.
1102 */
1103void netdev_state_change(struct net_device *dev)
1104{
1105	if (dev->flags & IFF_UP) {
1106		call_netdevice_notifiers(NETDEV_CHANGE, dev);
1107		rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1108	}
1109}
1110EXPORT_SYMBOL(netdev_state_change);
1111
1112int netdev_bonding_change(struct net_device *dev, unsigned long event)
1113{
1114	return call_netdevice_notifiers(event, dev);
1115}
1116EXPORT_SYMBOL(netdev_bonding_change);
1117
1118/**
1119 *	dev_load 	- load a network module
1120 *	@net: the applicable net namespace
1121 *	@name: name of interface
1122 *
1123 *	If a network interface is not present and the process has suitable
1124 *	privileges this function loads the module. If module loading is not
1125 *	available in this kernel then it becomes a nop.
1126 */
1127
1128void dev_load(struct net *net, const char *name)
1129{
1130	struct net_device *dev;
1131	int no_module;
1132
1133	rcu_read_lock();
1134	dev = dev_get_by_name_rcu(net, name);
1135	rcu_read_unlock();
1136
1137	no_module = !dev;
1138	if (no_module && capable(CAP_NET_ADMIN))
1139		no_module = request_module("netdev-%s", name);
1140	if (no_module && capable(CAP_SYS_MODULE)) {
1141		if (!request_module("%s", name))
1142			pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated).  Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1143				name);
 
1144	}
1145}
1146EXPORT_SYMBOL(dev_load);
1147
1148static int __dev_open(struct net_device *dev)
1149{
1150	const struct net_device_ops *ops = dev->netdev_ops;
1151	int ret;
1152
1153	ASSERT_RTNL();
1154
1155	if (!netif_device_present(dev))
1156		return -ENODEV;
1157
1158	ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1159	ret = notifier_to_errno(ret);
1160	if (ret)
1161		return ret;
1162
1163	set_bit(__LINK_STATE_START, &dev->state);
1164
1165	if (ops->ndo_validate_addr)
1166		ret = ops->ndo_validate_addr(dev);
1167
1168	if (!ret && ops->ndo_open)
1169		ret = ops->ndo_open(dev);
1170
1171	if (ret)
1172		clear_bit(__LINK_STATE_START, &dev->state);
1173	else {
1174		dev->flags |= IFF_UP;
1175		net_dmaengine_get();
1176		dev_set_rx_mode(dev);
1177		dev_activate(dev);
1178		add_device_randomness(dev->dev_addr, dev->addr_len);
1179	}
1180
1181	return ret;
1182}
1183
1184/**
1185 *	dev_open	- prepare an interface for use.
1186 *	@dev:	device to open
1187 *
1188 *	Takes a device from down to up state. The device's private open
1189 *	function is invoked and then the multicast lists are loaded. Finally
1190 *	the device is moved into the up state and a %NETDEV_UP message is
1191 *	sent to the netdev notifier chain.
1192 *
1193 *	Calling this function on an active interface is a nop. On a failure
1194 *	a negative errno code is returned.
1195 */
1196int dev_open(struct net_device *dev)
1197{
1198	int ret;
1199
1200	if (dev->flags & IFF_UP)
1201		return 0;
1202
1203	ret = __dev_open(dev);
1204	if (ret < 0)
1205		return ret;
1206
1207	rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1208	call_netdevice_notifiers(NETDEV_UP, dev);
1209
1210	return ret;
1211}
1212EXPORT_SYMBOL(dev_open);
1213
1214static int __dev_close_many(struct list_head *head)
1215{
1216	struct net_device *dev;
1217
1218	ASSERT_RTNL();
1219	might_sleep();
1220
1221	list_for_each_entry(dev, head, unreg_list) {
1222		call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1223
1224		clear_bit(__LINK_STATE_START, &dev->state);
1225
1226		/* Synchronize to scheduled poll. We cannot touch poll list, it
1227		 * can be even on different cpu. So just clear netif_running().
1228		 *
1229		 * dev->stop() will invoke napi_disable() on all of it's
1230		 * napi_struct instances on this device.
1231		 */
1232		smp_mb__after_clear_bit(); /* Commit netif_running(). */
1233	}
1234
1235	dev_deactivate_many(head);
1236
1237	list_for_each_entry(dev, head, unreg_list) {
1238		const struct net_device_ops *ops = dev->netdev_ops;
1239
1240		/*
1241		 *	Call the device specific close. This cannot fail.
1242		 *	Only if device is UP
1243		 *
1244		 *	We allow it to be called even after a DETACH hot-plug
1245		 *	event.
1246		 */
1247		if (ops->ndo_stop)
1248			ops->ndo_stop(dev);
1249
1250		dev->flags &= ~IFF_UP;
1251		net_dmaengine_put();
1252	}
1253
1254	return 0;
1255}
1256
1257static int __dev_close(struct net_device *dev)
1258{
1259	int retval;
1260	LIST_HEAD(single);
1261
1262	list_add(&dev->unreg_list, &single);
1263	retval = __dev_close_many(&single);
1264	list_del(&single);
1265	return retval;
1266}
1267
1268static int dev_close_many(struct list_head *head)
1269{
1270	struct net_device *dev, *tmp;
1271	LIST_HEAD(tmp_list);
1272
1273	list_for_each_entry_safe(dev, tmp, head, unreg_list)
1274		if (!(dev->flags & IFF_UP))
1275			list_move(&dev->unreg_list, &tmp_list);
1276
1277	__dev_close_many(head);
1278
1279	list_for_each_entry(dev, head, unreg_list) {
1280		rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1281		call_netdevice_notifiers(NETDEV_DOWN, dev);
1282	}
1283
1284	/* rollback_registered_many needs the complete original list */
1285	list_splice(&tmp_list, head);
1286	return 0;
1287}
1288
1289/**
1290 *	dev_close - shutdown an interface.
1291 *	@dev: device to shutdown
1292 *
1293 *	This function moves an active device into down state. A
1294 *	%NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1295 *	is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1296 *	chain.
1297 */
1298int dev_close(struct net_device *dev)
1299{
1300	if (dev->flags & IFF_UP) {
1301		LIST_HEAD(single);
1302
1303		list_add(&dev->unreg_list, &single);
1304		dev_close_many(&single);
1305		list_del(&single);
1306	}
1307	return 0;
1308}
1309EXPORT_SYMBOL(dev_close);
1310
1311
1312/**
1313 *	dev_disable_lro - disable Large Receive Offload on a device
1314 *	@dev: device
1315 *
1316 *	Disable Large Receive Offload (LRO) on a net device.  Must be
1317 *	called under RTNL.  This is needed if received packets may be
1318 *	forwarded to another interface.
1319 */
1320void dev_disable_lro(struct net_device *dev)
1321{
 
 
1322	/*
1323	 * If we're trying to disable lro on a vlan device
1324	 * use the underlying physical device instead
1325	 */
1326	if (is_vlan_dev(dev))
1327		dev = vlan_dev_real_dev(dev);
1328
1329	dev->wanted_features &= ~NETIF_F_LRO;
1330	netdev_update_features(dev);
 
 
1331
 
 
 
 
1332	if (unlikely(dev->features & NETIF_F_LRO))
1333		netdev_WARN(dev, "failed to disable LRO!\n");
1334}
1335EXPORT_SYMBOL(dev_disable_lro);
1336
1337
1338static int dev_boot_phase = 1;
1339
1340/**
1341 *	register_netdevice_notifier - register a network notifier block
1342 *	@nb: notifier
1343 *
1344 *	Register a notifier to be called when network device events occur.
1345 *	The notifier passed is linked into the kernel structures and must
1346 *	not be reused until it has been unregistered. A negative errno code
1347 *	is returned on a failure.
1348 *
1349 * 	When registered all registration and up events are replayed
1350 *	to the new notifier to allow device to have a race free
1351 *	view of the network device list.
1352 */
1353
1354int register_netdevice_notifier(struct notifier_block *nb)
1355{
1356	struct net_device *dev;
1357	struct net_device *last;
1358	struct net *net;
1359	int err;
1360
1361	rtnl_lock();
1362	err = raw_notifier_chain_register(&netdev_chain, nb);
1363	if (err)
1364		goto unlock;
1365	if (dev_boot_phase)
1366		goto unlock;
1367	for_each_net(net) {
1368		for_each_netdev(net, dev) {
1369			err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1370			err = notifier_to_errno(err);
1371			if (err)
1372				goto rollback;
1373
1374			if (!(dev->flags & IFF_UP))
1375				continue;
1376
1377			nb->notifier_call(nb, NETDEV_UP, dev);
1378		}
1379	}
1380
1381unlock:
1382	rtnl_unlock();
1383	return err;
1384
1385rollback:
1386	last = dev;
1387	for_each_net(net) {
1388		for_each_netdev(net, dev) {
1389			if (dev == last)
1390				goto outroll;
1391
1392			if (dev->flags & IFF_UP) {
1393				nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1394				nb->notifier_call(nb, NETDEV_DOWN, dev);
1395			}
1396			nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1397			nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1398		}
1399	}
1400
1401outroll:
1402	raw_notifier_chain_unregister(&netdev_chain, nb);
1403	goto unlock;
1404}
1405EXPORT_SYMBOL(register_netdevice_notifier);
1406
1407/**
1408 *	unregister_netdevice_notifier - unregister a network notifier block
1409 *	@nb: notifier
1410 *
1411 *	Unregister a notifier previously registered by
1412 *	register_netdevice_notifier(). The notifier is unlinked into the
1413 *	kernel structures and may then be reused. A negative errno code
1414 *	is returned on a failure.
1415 *
1416 * 	After unregistering unregister and down device events are synthesized
1417 *	for all devices on the device list to the removed notifier to remove
1418 *	the need for special case cleanup code.
1419 */
1420
1421int unregister_netdevice_notifier(struct notifier_block *nb)
1422{
1423	struct net_device *dev;
1424	struct net *net;
1425	int err;
1426
1427	rtnl_lock();
1428	err = raw_notifier_chain_unregister(&netdev_chain, nb);
1429	if (err)
1430		goto unlock;
1431
1432	for_each_net(net) {
1433		for_each_netdev(net, dev) {
1434			if (dev->flags & IFF_UP) {
1435				nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1436				nb->notifier_call(nb, NETDEV_DOWN, dev);
1437			}
1438			nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1439			nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1440		}
1441	}
1442unlock:
1443	rtnl_unlock();
1444	return err;
1445}
1446EXPORT_SYMBOL(unregister_netdevice_notifier);
1447
1448/**
1449 *	call_netdevice_notifiers - call all network notifier blocks
1450 *      @val: value passed unmodified to notifier function
1451 *      @dev: net_device pointer passed unmodified to notifier function
1452 *
1453 *	Call all network notifier blocks.  Parameters and return value
1454 *	are as for raw_notifier_call_chain().
1455 */
1456
1457int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1458{
1459	ASSERT_RTNL();
1460	return raw_notifier_call_chain(&netdev_chain, val, dev);
1461}
1462EXPORT_SYMBOL(call_netdevice_notifiers);
1463
1464static struct static_key netstamp_needed __read_mostly;
1465#ifdef HAVE_JUMP_LABEL
1466/* We are not allowed to call static_key_slow_dec() from irq context
1467 * If net_disable_timestamp() is called from irq context, defer the
1468 * static_key_slow_dec() calls.
1469 */
1470static atomic_t netstamp_needed_deferred;
1471#endif
1472
1473void net_enable_timestamp(void)
1474{
1475#ifdef HAVE_JUMP_LABEL
1476	int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1477
1478	if (deferred) {
1479		while (--deferred)
1480			static_key_slow_dec(&netstamp_needed);
1481		return;
1482	}
1483#endif
1484	WARN_ON(in_interrupt());
1485	static_key_slow_inc(&netstamp_needed);
1486}
1487EXPORT_SYMBOL(net_enable_timestamp);
1488
1489void net_disable_timestamp(void)
1490{
1491#ifdef HAVE_JUMP_LABEL
1492	if (in_interrupt()) {
1493		atomic_inc(&netstamp_needed_deferred);
1494		return;
1495	}
1496#endif
1497	static_key_slow_dec(&netstamp_needed);
1498}
1499EXPORT_SYMBOL(net_disable_timestamp);
1500
1501static inline void net_timestamp_set(struct sk_buff *skb)
1502{
1503	skb->tstamp.tv64 = 0;
1504	if (static_key_false(&netstamp_needed))
1505		__net_timestamp(skb);
 
 
1506}
1507
1508#define net_timestamp_check(COND, SKB)			\
1509	if (static_key_false(&netstamp_needed)) {		\
1510		if ((COND) && !(SKB)->tstamp.tv64)	\
1511			__net_timestamp(SKB);		\
1512	}						\
1513
1514static int net_hwtstamp_validate(struct ifreq *ifr)
1515{
1516	struct hwtstamp_config cfg;
1517	enum hwtstamp_tx_types tx_type;
1518	enum hwtstamp_rx_filters rx_filter;
1519	int tx_type_valid = 0;
1520	int rx_filter_valid = 0;
1521
1522	if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1523		return -EFAULT;
1524
1525	if (cfg.flags) /* reserved for future extensions */
1526		return -EINVAL;
1527
1528	tx_type = cfg.tx_type;
1529	rx_filter = cfg.rx_filter;
1530
1531	switch (tx_type) {
1532	case HWTSTAMP_TX_OFF:
1533	case HWTSTAMP_TX_ON:
1534	case HWTSTAMP_TX_ONESTEP_SYNC:
1535		tx_type_valid = 1;
1536		break;
1537	}
1538
1539	switch (rx_filter) {
1540	case HWTSTAMP_FILTER_NONE:
1541	case HWTSTAMP_FILTER_ALL:
1542	case HWTSTAMP_FILTER_SOME:
1543	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1544	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1545	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1546	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1547	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1548	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1549	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1550	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1551	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1552	case HWTSTAMP_FILTER_PTP_V2_EVENT:
1553	case HWTSTAMP_FILTER_PTP_V2_SYNC:
1554	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1555		rx_filter_valid = 1;
1556		break;
1557	}
1558
1559	if (!tx_type_valid || !rx_filter_valid)
1560		return -ERANGE;
1561
1562	return 0;
1563}
1564
1565static inline bool is_skb_forwardable(struct net_device *dev,
1566				      struct sk_buff *skb)
1567{
1568	unsigned int len;
1569
1570	if (!(dev->flags & IFF_UP))
1571		return false;
1572
1573	len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1574	if (skb->len <= len)
1575		return true;
1576
1577	/* if TSO is enabled, we don't care about the length as the packet
1578	 * could be forwarded without being segmented before
1579	 */
1580	if (skb_is_gso(skb))
1581		return true;
1582
1583	return false;
1584}
1585
1586/**
1587 * dev_forward_skb - loopback an skb to another netif
1588 *
1589 * @dev: destination network device
1590 * @skb: buffer to forward
1591 *
1592 * return values:
1593 *	NET_RX_SUCCESS	(no congestion)
1594 *	NET_RX_DROP     (packet was dropped, but freed)
1595 *
1596 * dev_forward_skb can be used for injecting an skb from the
1597 * start_xmit function of one device into the receive queue
1598 * of another device.
1599 *
1600 * The receiving device may be in another namespace, so
1601 * we have to clear all information in the skb that could
1602 * impact namespace isolation.
1603 */
1604int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1605{
1606	if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1607		if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1608			atomic_long_inc(&dev->rx_dropped);
1609			kfree_skb(skb);
1610			return NET_RX_DROP;
1611		}
1612	}
1613
1614	skb_orphan(skb);
1615	nf_reset(skb);
1616
1617	if (unlikely(!is_skb_forwardable(dev, skb))) {
1618		atomic_long_inc(&dev->rx_dropped);
1619		kfree_skb(skb);
1620		return NET_RX_DROP;
1621	}
1622	skb->skb_iif = 0;
1623	skb->dev = dev;
1624	skb_dst_drop(skb);
1625	skb->tstamp.tv64 = 0;
1626	skb->pkt_type = PACKET_HOST;
1627	skb->protocol = eth_type_trans(skb, dev);
1628	skb->mark = 0;
1629	secpath_reset(skb);
1630	nf_reset(skb);
1631	return netif_rx(skb);
1632}
1633EXPORT_SYMBOL_GPL(dev_forward_skb);
1634
1635static inline int deliver_skb(struct sk_buff *skb,
1636			      struct packet_type *pt_prev,
1637			      struct net_device *orig_dev)
1638{
1639	atomic_inc(&skb->users);
1640	return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1641}
1642
1643static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1644{
1645	if (ptype->af_packet_priv == NULL)
1646		return false;
1647
1648	if (ptype->id_match)
1649		return ptype->id_match(ptype, skb->sk);
1650	else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1651		return true;
1652
1653	return false;
1654}
1655
1656/*
1657 *	Support routine. Sends outgoing frames to any network
1658 *	taps currently in use.
1659 */
1660
1661static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1662{
1663	struct packet_type *ptype;
1664	struct sk_buff *skb2 = NULL;
1665	struct packet_type *pt_prev = NULL;
1666
1667	rcu_read_lock();
1668	list_for_each_entry_rcu(ptype, &ptype_all, list) {
1669		/* Never send packets back to the socket
1670		 * they originated from - MvS (miquels@drinkel.ow.org)
1671		 */
1672		if ((ptype->dev == dev || !ptype->dev) &&
1673		    (!skb_loop_sk(ptype, skb))) {
 
1674			if (pt_prev) {
1675				deliver_skb(skb2, pt_prev, skb->dev);
1676				pt_prev = ptype;
1677				continue;
1678			}
1679
1680			skb2 = skb_clone(skb, GFP_ATOMIC);
1681			if (!skb2)
1682				break;
1683
1684			net_timestamp_set(skb2);
1685
1686			/* skb->nh should be correctly
1687			   set by sender, so that the second statement is
1688			   just protection against buggy protocols.
1689			 */
1690			skb_reset_mac_header(skb2);
1691
1692			if (skb_network_header(skb2) < skb2->data ||
1693			    skb2->network_header > skb2->tail) {
1694				net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1695						     ntohs(skb2->protocol),
1696						     dev->name);
 
 
1697				skb_reset_network_header(skb2);
1698			}
1699
1700			skb2->transport_header = skb2->network_header;
1701			skb2->pkt_type = PACKET_OUTGOING;
1702			pt_prev = ptype;
1703		}
1704	}
1705	if (pt_prev)
1706		pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1707	rcu_read_unlock();
1708}
1709
1710/* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1711 * @dev: Network device
1712 * @txq: number of queues available
1713 *
1714 * If real_num_tx_queues is changed the tc mappings may no longer be
1715 * valid. To resolve this verify the tc mapping remains valid and if
1716 * not NULL the mapping. With no priorities mapping to this
1717 * offset/count pair it will no longer be used. In the worst case TC0
1718 * is invalid nothing can be done so disable priority mappings. If is
1719 * expected that drivers will fix this mapping if they can before
1720 * calling netif_set_real_num_tx_queues.
1721 */
1722static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1723{
1724	int i;
1725	struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1726
1727	/* If TC0 is invalidated disable TC mapping */
1728	if (tc->offset + tc->count > txq) {
1729		pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
 
 
1730		dev->num_tc = 0;
1731		return;
1732	}
1733
1734	/* Invalidated prio to tc mappings set to TC0 */
1735	for (i = 1; i < TC_BITMASK + 1; i++) {
1736		int q = netdev_get_prio_tc_map(dev, i);
1737
1738		tc = &dev->tc_to_txq[q];
1739		if (tc->offset + tc->count > txq) {
1740			pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1741				i, q);
 
 
 
1742			netdev_set_prio_tc_map(dev, i, 0);
1743		}
1744	}
1745}
1746
1747/*
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.
1750 */
1751int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1752{
1753	int rc;
1754
1755	if (txq < 1 || txq > dev->num_tx_queues)
1756		return -EINVAL;
1757
1758	if (dev->reg_state == NETREG_REGISTERED ||
1759	    dev->reg_state == NETREG_UNREGISTERING) {
1760		ASSERT_RTNL();
1761
1762		rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1763						  txq);
1764		if (rc)
1765			return rc;
1766
1767		if (dev->num_tc)
1768			netif_setup_tc(dev, txq);
1769
1770		if (txq < dev->real_num_tx_queues)
1771			qdisc_reset_all_tx_gt(dev, txq);
1772	}
1773
1774	dev->real_num_tx_queues = txq;
1775	return 0;
1776}
1777EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1778
1779#ifdef CONFIG_RPS
1780/**
1781 *	netif_set_real_num_rx_queues - set actual number of RX queues used
1782 *	@dev: Network device
1783 *	@rxq: Actual number of RX queues
1784 *
1785 *	This must be called either with the rtnl_lock held or before
1786 *	registration of the net device.  Returns 0 on success, or a
1787 *	negative error code.  If called before registration, it always
1788 *	succeeds.
1789 */
1790int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1791{
1792	int rc;
1793
1794	if (rxq < 1 || rxq > dev->num_rx_queues)
1795		return -EINVAL;
1796
1797	if (dev->reg_state == NETREG_REGISTERED) {
1798		ASSERT_RTNL();
1799
1800		rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1801						  rxq);
1802		if (rc)
1803			return rc;
1804	}
1805
1806	dev->real_num_rx_queues = rxq;
1807	return 0;
1808}
1809EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1810#endif
1811
1812static inline void __netif_reschedule(struct Qdisc *q)
1813{
1814	struct softnet_data *sd;
1815	unsigned long flags;
1816
1817	local_irq_save(flags);
1818	sd = &__get_cpu_var(softnet_data);
1819	q->next_sched = NULL;
1820	*sd->output_queue_tailp = q;
1821	sd->output_queue_tailp = &q->next_sched;
1822	raise_softirq_irqoff(NET_TX_SOFTIRQ);
1823	local_irq_restore(flags);
1824}
1825
1826void __netif_schedule(struct Qdisc *q)
1827{
1828	if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1829		__netif_reschedule(q);
1830}
1831EXPORT_SYMBOL(__netif_schedule);
1832
1833void dev_kfree_skb_irq(struct sk_buff *skb)
1834{
1835	if (atomic_dec_and_test(&skb->users)) {
1836		struct softnet_data *sd;
1837		unsigned long flags;
1838
1839		local_irq_save(flags);
1840		sd = &__get_cpu_var(softnet_data);
1841		skb->next = sd->completion_queue;
1842		sd->completion_queue = skb;
1843		raise_softirq_irqoff(NET_TX_SOFTIRQ);
1844		local_irq_restore(flags);
1845	}
1846}
1847EXPORT_SYMBOL(dev_kfree_skb_irq);
1848
1849void dev_kfree_skb_any(struct sk_buff *skb)
1850{
1851	if (in_irq() || irqs_disabled())
1852		dev_kfree_skb_irq(skb);
1853	else
1854		dev_kfree_skb(skb);
1855}
1856EXPORT_SYMBOL(dev_kfree_skb_any);
1857
1858
1859/**
1860 * netif_device_detach - mark device as removed
1861 * @dev: network device
1862 *
1863 * Mark device as removed from system and therefore no longer available.
1864 */
1865void netif_device_detach(struct net_device *dev)
1866{
1867	if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1868	    netif_running(dev)) {
1869		netif_tx_stop_all_queues(dev);
1870	}
1871}
1872EXPORT_SYMBOL(netif_device_detach);
1873
1874/**
1875 * netif_device_attach - mark device as attached
1876 * @dev: network device
1877 *
1878 * Mark device as attached from system and restart if needed.
1879 */
1880void netif_device_attach(struct net_device *dev)
1881{
1882	if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1883	    netif_running(dev)) {
1884		netif_tx_wake_all_queues(dev);
1885		__netdev_watchdog_up(dev);
1886	}
1887}
1888EXPORT_SYMBOL(netif_device_attach);
1889
1890static void skb_warn_bad_offload(const struct sk_buff *skb)
 
 
 
 
 
 
 
 
 
 
1891{
1892	static const netdev_features_t null_features = 0;
1893	struct net_device *dev = skb->dev;
1894	const char *driver = "";
1895
1896	if (dev && dev->dev.parent)
1897		driver = dev_driver_string(dev->dev.parent);
1898
1899	WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1900	     "gso_type=%d ip_summed=%d\n",
1901	     driver, dev ? &dev->features : &null_features,
1902	     skb->sk ? &skb->sk->sk_route_caps : &null_features,
1903	     skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1904	     skb_shinfo(skb)->gso_type, skb->ip_summed);
 
1905}
 
 
1906
1907/*
1908 * Invalidate hardware checksum when packet is to be mangled, and
1909 * complete checksum manually on outgoing path.
1910 */
1911int skb_checksum_help(struct sk_buff *skb)
1912{
1913	__wsum csum;
1914	int ret = 0, offset;
1915
1916	if (skb->ip_summed == CHECKSUM_COMPLETE)
1917		goto out_set_summed;
1918
1919	if (unlikely(skb_shinfo(skb)->gso_size)) {
1920		skb_warn_bad_offload(skb);
1921		return -EINVAL;
1922	}
1923
1924	offset = skb_checksum_start_offset(skb);
1925	BUG_ON(offset >= skb_headlen(skb));
1926	csum = skb_checksum(skb, offset, skb->len - offset, 0);
1927
1928	offset += skb->csum_offset;
1929	BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1930
1931	if (skb_cloned(skb) &&
1932	    !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1933		ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1934		if (ret)
1935			goto out;
1936	}
1937
1938	*(__sum16 *)(skb->data + offset) = csum_fold(csum);
1939out_set_summed:
1940	skb->ip_summed = CHECKSUM_NONE;
1941out:
1942	return ret;
1943}
1944EXPORT_SYMBOL(skb_checksum_help);
1945
1946/**
1947 *	skb_gso_segment - Perform segmentation on skb.
1948 *	@skb: buffer to segment
1949 *	@features: features for the output path (see dev->features)
1950 *
1951 *	This function segments the given skb and returns a list of segments.
1952 *
1953 *	It may return NULL if the skb requires no segmentation.  This is
1954 *	only possible when GSO is used for verifying header integrity.
1955 */
1956struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1957	netdev_features_t features)
1958{
1959	struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1960	struct packet_type *ptype;
1961	__be16 type = skb->protocol;
1962	int vlan_depth = ETH_HLEN;
1963	int err;
1964
1965	while (type == htons(ETH_P_8021Q)) {
1966		struct vlan_hdr *vh;
1967
1968		if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1969			return ERR_PTR(-EINVAL);
1970
1971		vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1972		type = vh->h_vlan_encapsulated_proto;
1973		vlan_depth += VLAN_HLEN;
1974	}
1975
1976	skb_reset_mac_header(skb);
1977	skb->mac_len = skb->network_header - skb->mac_header;
1978	__skb_pull(skb, skb->mac_len);
1979
1980	if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1981		skb_warn_bad_offload(skb);
 
 
 
 
 
 
 
 
 
1982
1983		if (skb_header_cloned(skb) &&
1984		    (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1985			return ERR_PTR(err);
1986	}
1987
1988	rcu_read_lock();
1989	list_for_each_entry_rcu(ptype,
1990			&ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1991		if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1992			if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1993				err = ptype->gso_send_check(skb);
1994				segs = ERR_PTR(err);
1995				if (err || skb_gso_ok(skb, features))
1996					break;
1997				__skb_push(skb, (skb->data -
1998						 skb_network_header(skb)));
1999			}
2000			segs = ptype->gso_segment(skb, features);
2001			break;
2002		}
2003	}
2004	rcu_read_unlock();
2005
2006	__skb_push(skb, skb->data - skb_mac_header(skb));
2007
2008	return segs;
2009}
2010EXPORT_SYMBOL(skb_gso_segment);
2011
2012/* Take action when hardware reception checksum errors are detected. */
2013#ifdef CONFIG_BUG
2014void netdev_rx_csum_fault(struct net_device *dev)
2015{
2016	if (net_ratelimit()) {
2017		pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
 
2018		dump_stack();
2019	}
2020}
2021EXPORT_SYMBOL(netdev_rx_csum_fault);
2022#endif
2023
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.
2027 */
2028
2029static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2030{
2031#ifdef CONFIG_HIGHMEM
2032	int i;
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)))
2037				return 1;
2038		}
2039	}
2040
2041	if (PCI_DMA_BUS_IS_PHYS) {
2042		struct device *pdev = dev->dev.parent;
2043
2044		if (!pdev)
2045			return 0;
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)
2050				return 1;
2051		}
2052	}
2053#endif
2054	return 0;
2055}
2056
2057struct dev_gso_cb {
2058	void (*destructor)(struct sk_buff *skb);
2059};
2060
2061#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2062
2063static void dev_gso_skb_destructor(struct sk_buff *skb)
2064{
2065	struct dev_gso_cb *cb;
2066
2067	do {
2068		struct sk_buff *nskb = skb->next;
2069
2070		skb->next = nskb->next;
2071		nskb->next = NULL;
2072		kfree_skb(nskb);
2073	} while (skb->next);
2074
2075	cb = DEV_GSO_CB(skb);
2076	if (cb->destructor)
2077		cb->destructor(skb);
2078}
2079
2080/**
2081 *	dev_gso_segment - Perform emulated hardware segmentation on skb.
2082 *	@skb: buffer to segment
2083 *	@features: device features as applicable to this skb
2084 *
2085 *	This function segments the given skb and stores the list of segments
2086 *	in skb->next.
2087 */
2088static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2089{
2090	struct sk_buff *segs;
2091
2092	segs = skb_gso_segment(skb, features);
2093
2094	/* Verifying header integrity only. */
2095	if (!segs)
2096		return 0;
2097
2098	if (IS_ERR(segs))
2099		return PTR_ERR(segs);
2100
2101	skb->next = segs;
2102	DEV_GSO_CB(skb)->destructor = skb->destructor;
2103	skb->destructor = dev_gso_skb_destructor;
2104
2105	return 0;
2106}
2107
2108static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2109{
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)));
2117}
2118
2119static netdev_features_t harmonize_features(struct sk_buff *skb,
2120	__be16 protocol, netdev_features_t features)
2121{
2122	if (!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;
2127	}
2128
2129	return features;
2130}
2131
2132netdev_features_t netif_skb_features(struct sk_buff *skb)
2133{
2134	__be16 protocol = skb->protocol;
2135	netdev_features_t features = skb->dev->features;
2136
2137	if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2138		features &= ~NETIF_F_GSO_MASK;
2139
2140	if (protocol == htons(ETH_P_8021Q)) {
2141		struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2142		protocol = veh->h_vlan_encapsulated_proto;
2143	} else if (!vlan_tx_tag_present(skb)) {
2144		return harmonize_features(skb, protocol, features);
2145	}
2146
2147	features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2148
2149	if (protocol != htons(ETH_P_8021Q)) {
2150		return harmonize_features(skb, protocol, features);
2151	} else {
2152		features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2153				NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2154		return harmonize_features(skb, protocol, features);
2155	}
2156}
2157EXPORT_SYMBOL(netif_skb_features);
2158
2159/*
2160 * Returns true if either:
2161 *	1. skb has frag_list and the device doesn't support FRAGLIST, or
2162 *	2. skb is fragmented and the device does not support SG, or if
2163 *	   at least one of fragments is in highmem and device does not
2164 *	   support DMA from it.
2165 */
2166static inline int skb_needs_linearize(struct sk_buff *skb,
2167				      int features)
2168{
2169	return skb_is_nonlinear(skb) &&
2170			((skb_has_frag_list(skb) &&
2171				!(features & NETIF_F_FRAGLIST)) ||
2172			(skb_shinfo(skb)->nr_frags &&
2173				!(features & NETIF_F_SG)));
2174}
2175
2176int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2177			struct netdev_queue *txq)
2178{
2179	const struct net_device_ops *ops = dev->netdev_ops;
2180	int rc = NETDEV_TX_OK;
2181	unsigned int skb_len;
2182
2183	if (likely(!skb->next)) {
2184		netdev_features_t features;
2185
2186		/*
2187		 * If device doesn't need skb->dst, release it right now while
2188		 * its hot in this cpu cache
2189		 */
2190		if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2191			skb_dst_drop(skb);
2192
2193		if (!list_empty(&ptype_all))
2194			dev_queue_xmit_nit(skb, dev);
2195
 
 
2196		features = netif_skb_features(skb);
2197
2198		if (vlan_tx_tag_present(skb) &&
2199		    !(features & NETIF_F_HW_VLAN_TX)) {
2200			skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2201			if (unlikely(!skb))
2202				goto out;
2203
2204			skb->vlan_tci = 0;
2205		}
2206
2207		if (netif_needs_gso(skb, features)) {
2208			if (unlikely(dev_gso_segment(skb, features)))
2209				goto out_kfree_skb;
2210			if (skb->next)
2211				goto gso;
2212		} else {
2213			if (skb_needs_linearize(skb, features) &&
2214			    __skb_linearize(skb))
2215				goto out_kfree_skb;
2216
2217			/* If packet is not checksummed and device does not
2218			 * support checksumming for this protocol, complete
2219			 * checksumming here.
2220			 */
2221			if (skb->ip_summed == CHECKSUM_PARTIAL) {
2222				skb_set_transport_header(skb,
2223					skb_checksum_start_offset(skb));
2224				if (!(features & NETIF_F_ALL_CSUM) &&
2225				     skb_checksum_help(skb))
2226					goto out_kfree_skb;
2227			}
2228		}
2229
2230		skb_len = skb->len;
2231		rc = ops->ndo_start_xmit(skb, dev);
2232		trace_net_dev_xmit(skb, rc, dev, skb_len);
2233		if (rc == NETDEV_TX_OK)
2234			txq_trans_update(txq);
2235		return rc;
2236	}
2237
2238gso:
2239	do {
2240		struct sk_buff *nskb = skb->next;
2241
2242		skb->next = nskb->next;
2243		nskb->next = NULL;
2244
2245		/*
2246		 * If device doesn't need nskb->dst, release it right now while
2247		 * its hot in this cpu cache
2248		 */
2249		if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2250			skb_dst_drop(nskb);
2251
2252		skb_len = nskb->len;
2253		rc = ops->ndo_start_xmit(nskb, dev);
2254		trace_net_dev_xmit(nskb, rc, dev, skb_len);
2255		if (unlikely(rc != NETDEV_TX_OK)) {
2256			if (rc & ~NETDEV_TX_MASK)
2257				goto out_kfree_gso_skb;
2258			nskb->next = skb->next;
2259			skb->next = nskb;
2260			return rc;
2261		}
2262		txq_trans_update(txq);
2263		if (unlikely(netif_xmit_stopped(txq) && skb->next))
2264			return NETDEV_TX_BUSY;
2265	} while (skb->next);
2266
2267out_kfree_gso_skb:
2268	if (likely(skb->next == NULL))
2269		skb->destructor = DEV_GSO_CB(skb)->destructor;
2270out_kfree_skb:
2271	kfree_skb(skb);
2272out:
2273	return rc;
2274}
2275
2276static u32 hashrnd __read_mostly;
2277
2278/*
2279 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2280 * to be used as a distribution range.
2281 */
2282u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2283		  unsigned int num_tx_queues)
2284{
2285	u32 hash;
2286	u16 qoffset = 0;
2287	u16 qcount = num_tx_queues;
2288
2289	if (skb_rx_queue_recorded(skb)) {
2290		hash = skb_get_rx_queue(skb);
2291		while (unlikely(hash >= num_tx_queues))
2292			hash -= num_tx_queues;
2293		return hash;
2294	}
2295
2296	if (dev->num_tc) {
2297		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2298		qoffset = dev->tc_to_txq[tc].offset;
2299		qcount = dev->tc_to_txq[tc].count;
2300	}
2301
2302	if (skb->sk && skb->sk->sk_hash)
2303		hash = skb->sk->sk_hash;
2304	else
2305		hash = (__force u16) skb->protocol;
2306	hash = jhash_1word(hash, hashrnd);
2307
2308	return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2309}
2310EXPORT_SYMBOL(__skb_tx_hash);
2311
2312static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2313{
2314	if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2315		net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2316				     dev->name, queue_index,
2317				     dev->real_num_tx_queues);
 
 
2318		return 0;
2319	}
2320	return queue_index;
2321}
2322
2323static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2324{
2325#ifdef CONFIG_XPS
2326	struct xps_dev_maps *dev_maps;
2327	struct xps_map *map;
2328	int queue_index = -1;
2329
2330	rcu_read_lock();
2331	dev_maps = rcu_dereference(dev->xps_maps);
2332	if (dev_maps) {
2333		map = rcu_dereference(
2334		    dev_maps->cpu_map[raw_smp_processor_id()]);
2335		if (map) {
2336			if (map->len == 1)
2337				queue_index = map->queues[0];
2338			else {
2339				u32 hash;
2340				if (skb->sk && skb->sk->sk_hash)
2341					hash = skb->sk->sk_hash;
2342				else
2343					hash = (__force u16) skb->protocol ^
2344					    skb->rxhash;
2345				hash = jhash_1word(hash, hashrnd);
2346				queue_index = map->queues[
2347				    ((u64)hash * map->len) >> 32];
2348			}
2349			if (unlikely(queue_index >= dev->real_num_tx_queues))
2350				queue_index = -1;
2351		}
2352	}
2353	rcu_read_unlock();
2354
2355	return queue_index;
2356#else
2357	return -1;
2358#endif
2359}
2360
2361static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2362					struct sk_buff *skb)
2363{
2364	int queue_index;
2365	const struct net_device_ops *ops = dev->netdev_ops;
2366
2367	if (dev->real_num_tx_queues == 1)
2368		queue_index = 0;
2369	else if (ops->ndo_select_queue) {
2370		queue_index = ops->ndo_select_queue(dev, skb);
2371		queue_index = dev_cap_txqueue(dev, queue_index);
2372	} else {
2373		struct sock *sk = skb->sk;
2374		queue_index = sk_tx_queue_get(sk);
2375
2376		if (queue_index < 0 || skb->ooo_okay ||
2377		    queue_index >= dev->real_num_tx_queues) {
2378			int old_index = queue_index;
2379
2380			queue_index = get_xps_queue(dev, skb);
2381			if (queue_index < 0)
2382				queue_index = skb_tx_hash(dev, skb);
2383
2384			if (queue_index != old_index && sk) {
2385				struct dst_entry *dst =
2386				    rcu_dereference_check(sk->sk_dst_cache, 1);
2387
2388				if (dst && skb_dst(skb) == dst)
2389					sk_tx_queue_set(sk, queue_index);
2390			}
2391		}
2392	}
2393
2394	skb_set_queue_mapping(skb, queue_index);
2395	return netdev_get_tx_queue(dev, queue_index);
2396}
2397
2398static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2399				 struct net_device *dev,
2400				 struct netdev_queue *txq)
2401{
2402	spinlock_t *root_lock = qdisc_lock(q);
2403	bool contended;
2404	int rc;
2405
2406	qdisc_skb_cb(skb)->pkt_len = skb->len;
2407	qdisc_calculate_pkt_len(skb, q);
2408	/*
2409	 * Heuristic to force contended enqueues to serialize on a
2410	 * separate lock before trying to get qdisc main lock.
2411	 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2412	 * and dequeue packets faster.
2413	 */
2414	contended = qdisc_is_running(q);
2415	if (unlikely(contended))
2416		spin_lock(&q->busylock);
2417
2418	spin_lock(root_lock);
2419	if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2420		kfree_skb(skb);
2421		rc = NET_XMIT_DROP;
2422	} else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2423		   qdisc_run_begin(q)) {
2424		/*
2425		 * This is a work-conserving queue; there are no old skbs
2426		 * waiting to be sent out; and the qdisc is not running -
2427		 * xmit the skb directly.
2428		 */
2429		if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2430			skb_dst_force(skb);
2431
2432		qdisc_bstats_update(q, skb);
2433
2434		if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2435			if (unlikely(contended)) {
2436				spin_unlock(&q->busylock);
2437				contended = false;
2438			}
2439			__qdisc_run(q);
2440		} else
2441			qdisc_run_end(q);
2442
2443		rc = NET_XMIT_SUCCESS;
2444	} else {
2445		skb_dst_force(skb);
2446		rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2447		if (qdisc_run_begin(q)) {
2448			if (unlikely(contended)) {
2449				spin_unlock(&q->busylock);
2450				contended = false;
2451			}
2452			__qdisc_run(q);
2453		}
2454	}
2455	spin_unlock(root_lock);
2456	if (unlikely(contended))
2457		spin_unlock(&q->busylock);
2458	return rc;
2459}
2460
2461#if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2462static void skb_update_prio(struct sk_buff *skb)
2463{
2464	struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2465
2466	if (!skb->priority && skb->sk && map) {
2467		unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2468
2469		if (prioidx < map->priomap_len)
2470			skb->priority = map->priomap[prioidx];
2471	}
2472}
2473#else
2474#define skb_update_prio(skb)
2475#endif
2476
2477static DEFINE_PER_CPU(int, xmit_recursion);
2478#define RECURSION_LIMIT 10
2479
2480/**
2481 *	dev_queue_xmit - transmit a buffer
2482 *	@skb: buffer to transmit
2483 *
2484 *	Queue a buffer for transmission to a network device. The caller must
2485 *	have set the device and priority and built the buffer before calling
2486 *	this function. The function can be called from an interrupt.
2487 *
2488 *	A negative errno code is returned on a failure. A success does not
2489 *	guarantee the frame will be transmitted as it may be dropped due
2490 *	to congestion or traffic shaping.
2491 *
2492 * -----------------------------------------------------------------------------------
2493 *      I notice this method can also return errors from the queue disciplines,
2494 *      including NET_XMIT_DROP, which is a positive value.  So, errors can also
2495 *      be positive.
2496 *
2497 *      Regardless of the return value, the skb is consumed, so it is currently
2498 *      difficult to retry a send to this method.  (You can bump the ref count
2499 *      before sending to hold a reference for retry if you are careful.)
2500 *
2501 *      When calling this method, interrupts MUST be enabled.  This is because
2502 *      the BH enable code must have IRQs enabled so that it will not deadlock.
2503 *          --BLG
2504 */
2505int dev_queue_xmit(struct sk_buff *skb)
2506{
2507	struct net_device *dev = skb->dev;
2508	struct netdev_queue *txq;
2509	struct Qdisc *q;
2510	int rc = -ENOMEM;
2511
2512	/* Disable soft irqs for various locks below. Also
2513	 * stops preemption for RCU.
2514	 */
2515	rcu_read_lock_bh();
2516
2517	skb_update_prio(skb);
2518
2519	txq = dev_pick_tx(dev, skb);
2520	q = rcu_dereference_bh(txq->qdisc);
2521
2522#ifdef CONFIG_NET_CLS_ACT
2523	skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2524#endif
2525	trace_net_dev_queue(skb);
2526	if (q->enqueue) {
2527		rc = __dev_xmit_skb(skb, q, dev, txq);
2528		goto out;
2529	}
2530
2531	/* The device has no queue. Common case for software devices:
2532	   loopback, all the sorts of tunnels...
2533
2534	   Really, it is unlikely that netif_tx_lock protection is necessary
2535	   here.  (f.e. loopback and IP tunnels are clean ignoring statistics
2536	   counters.)
2537	   However, it is possible, that they rely on protection
2538	   made by us here.
2539
2540	   Check this and shot the lock. It is not prone from deadlocks.
2541	   Either shot noqueue qdisc, it is even simpler 8)
2542	 */
2543	if (dev->flags & IFF_UP) {
2544		int cpu = smp_processor_id(); /* ok because BHs are off */
2545
2546		if (txq->xmit_lock_owner != cpu) {
2547
2548			if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2549				goto recursion_alert;
2550
2551			HARD_TX_LOCK(dev, txq, cpu);
2552
2553			if (!netif_xmit_stopped(txq)) {
2554				__this_cpu_inc(xmit_recursion);
2555				rc = dev_hard_start_xmit(skb, dev, txq);
2556				__this_cpu_dec(xmit_recursion);
2557				if (dev_xmit_complete(rc)) {
2558					HARD_TX_UNLOCK(dev, txq);
2559					goto out;
2560				}
2561			}
2562			HARD_TX_UNLOCK(dev, txq);
2563			net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2564					     dev->name);
 
2565		} else {
2566			/* Recursion is detected! It is possible,
2567			 * unfortunately
2568			 */
2569recursion_alert:
2570			net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2571					     dev->name);
 
2572		}
2573	}
2574
2575	rc = -ENETDOWN;
2576	rcu_read_unlock_bh();
2577
2578	kfree_skb(skb);
2579	return rc;
2580out:
2581	rcu_read_unlock_bh();
2582	return rc;
2583}
2584EXPORT_SYMBOL(dev_queue_xmit);
2585
2586
2587/*=======================================================================
2588			Receiver routines
2589  =======================================================================*/
2590
2591int netdev_max_backlog __read_mostly = 1000;
2592int netdev_tstamp_prequeue __read_mostly = 1;
2593int netdev_budget __read_mostly = 300;
2594int weight_p __read_mostly = 64;            /* old backlog weight */
2595
2596/* Called with irq disabled */
2597static inline void ____napi_schedule(struct softnet_data *sd,
2598				     struct napi_struct *napi)
2599{
2600	list_add_tail(&napi->poll_list, &sd->poll_list);
2601	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
2602}
2603
2604/*
2605 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2606 * and src/dst port numbers.  Sets rxhash in skb to non-zero hash value
2607 * on success, zero indicates no valid hash.  Also, sets l4_rxhash in skb
2608 * if hash is a canonical 4-tuple hash over transport ports.
2609 */
2610void __skb_get_rxhash(struct sk_buff *skb)
2611{
2612	struct flow_keys keys;
2613	u32 hash;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2614
2615	if (!skb_flow_dissect(skb, &keys))
2616		return;
 
 
 
 
 
 
 
 
 
 
2617
2618	if (keys.ports) {
2619		if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2620			swap(keys.port16[0], keys.port16[1]);
2621		skb->l4_rxhash = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2622	}
2623
2624	/* get a consistent hash (same value on both flow directions) */
2625	if ((__force u32)keys.dst < (__force u32)keys.src)
2626		swap(keys.dst, keys.src);
2627
2628	hash = jhash_3words((__force u32)keys.dst,
2629			    (__force u32)keys.src,
2630			    (__force u32)keys.ports, hashrnd);
2631	if (!hash)
2632		hash = 1;
2633
2634	skb->rxhash = hash;
 
2635}
2636EXPORT_SYMBOL(__skb_get_rxhash);
2637
2638#ifdef CONFIG_RPS
2639
2640/* One global table that all flow-based protocols share. */
2641struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2642EXPORT_SYMBOL(rps_sock_flow_table);
2643
2644struct static_key rps_needed __read_mostly;
2645
2646static struct rps_dev_flow *
2647set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2648	    struct rps_dev_flow *rflow, u16 next_cpu)
2649{
2650	if (next_cpu != RPS_NO_CPU) {
 
 
 
2651#ifdef CONFIG_RFS_ACCEL
2652		struct netdev_rx_queue *rxqueue;
2653		struct rps_dev_flow_table *flow_table;
2654		struct rps_dev_flow *old_rflow;
2655		u32 flow_id;
2656		u16 rxq_index;
2657		int rc;
2658
2659		/* Should we steer this flow to a different hardware queue? */
2660		if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2661		    !(dev->features & NETIF_F_NTUPLE))
2662			goto out;
2663		rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2664		if (rxq_index == skb_get_rx_queue(skb))
2665			goto out;
2666
2667		rxqueue = dev->_rx + rxq_index;
2668		flow_table = rcu_dereference(rxqueue->rps_flow_table);
2669		if (!flow_table)
2670			goto out;
2671		flow_id = skb->rxhash & flow_table->mask;
2672		rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2673							rxq_index, flow_id);
2674		if (rc < 0)
2675			goto out;
2676		old_rflow = rflow;
2677		rflow = &flow_table->flows[flow_id];
 
2678		rflow->filter = rc;
2679		if (old_rflow->filter == rflow->filter)
2680			old_rflow->filter = RPS_NO_FILTER;
2681	out:
2682#endif
2683		rflow->last_qtail =
2684			per_cpu(softnet_data, next_cpu).input_queue_head;
2685	}
2686
2687	rflow->cpu = next_cpu;
2688	return rflow;
2689}
2690
2691/*
2692 * get_rps_cpu is called from netif_receive_skb and returns the target
2693 * CPU from the RPS map of the receiving queue for a given skb.
2694 * rcu_read_lock must be held on entry.
2695 */
2696static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2697		       struct rps_dev_flow **rflowp)
2698{
2699	struct netdev_rx_queue *rxqueue;
2700	struct rps_map *map;
2701	struct rps_dev_flow_table *flow_table;
2702	struct rps_sock_flow_table *sock_flow_table;
2703	int cpu = -1;
2704	u16 tcpu;
2705
2706	if (skb_rx_queue_recorded(skb)) {
2707		u16 index = skb_get_rx_queue(skb);
2708		if (unlikely(index >= dev->real_num_rx_queues)) {
2709			WARN_ONCE(dev->real_num_rx_queues > 1,
2710				  "%s received packet on queue %u, but number "
2711				  "of RX queues is %u\n",
2712				  dev->name, index, dev->real_num_rx_queues);
2713			goto done;
2714		}
2715		rxqueue = dev->_rx + index;
2716	} else
2717		rxqueue = dev->_rx;
2718
2719	map = rcu_dereference(rxqueue->rps_map);
2720	if (map) {
2721		if (map->len == 1 &&
2722		    !rcu_access_pointer(rxqueue->rps_flow_table)) {
2723			tcpu = map->cpus[0];
2724			if (cpu_online(tcpu))
2725				cpu = tcpu;
2726			goto done;
2727		}
2728	} else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2729		goto done;
2730	}
2731
2732	skb_reset_network_header(skb);
2733	if (!skb_get_rxhash(skb))
2734		goto done;
2735
2736	flow_table = rcu_dereference(rxqueue->rps_flow_table);
2737	sock_flow_table = rcu_dereference(rps_sock_flow_table);
2738	if (flow_table && sock_flow_table) {
2739		u16 next_cpu;
2740		struct rps_dev_flow *rflow;
2741
2742		rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2743		tcpu = rflow->cpu;
2744
2745		next_cpu = sock_flow_table->ents[skb->rxhash &
2746		    sock_flow_table->mask];
2747
2748		/*
2749		 * If the desired CPU (where last recvmsg was done) is
2750		 * different from current CPU (one in the rx-queue flow
2751		 * table entry), switch if one of the following holds:
2752		 *   - Current CPU is unset (equal to RPS_NO_CPU).
2753		 *   - Current CPU is offline.
2754		 *   - The current CPU's queue tail has advanced beyond the
2755		 *     last packet that was enqueued using this table entry.
2756		 *     This guarantees that all previous packets for the flow
2757		 *     have been dequeued, thus preserving in order delivery.
2758		 */
2759		if (unlikely(tcpu != next_cpu) &&
2760		    (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2761		     ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2762		      rflow->last_qtail)) >= 0))
2763			rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2764
2765		if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2766			*rflowp = rflow;
2767			cpu = tcpu;
2768			goto done;
2769		}
2770	}
2771
2772	if (map) {
2773		tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2774
2775		if (cpu_online(tcpu)) {
2776			cpu = tcpu;
2777			goto done;
2778		}
2779	}
2780
2781done:
2782	return cpu;
2783}
2784
2785#ifdef CONFIG_RFS_ACCEL
2786
2787/**
2788 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2789 * @dev: Device on which the filter was set
2790 * @rxq_index: RX queue index
2791 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2792 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2793 *
2794 * Drivers that implement ndo_rx_flow_steer() should periodically call
2795 * this function for each installed filter and remove the filters for
2796 * which it returns %true.
2797 */
2798bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2799			 u32 flow_id, u16 filter_id)
2800{
2801	struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2802	struct rps_dev_flow_table *flow_table;
2803	struct rps_dev_flow *rflow;
2804	bool expire = true;
2805	int cpu;
2806
2807	rcu_read_lock();
2808	flow_table = rcu_dereference(rxqueue->rps_flow_table);
2809	if (flow_table && flow_id <= flow_table->mask) {
2810		rflow = &flow_table->flows[flow_id];
2811		cpu = ACCESS_ONCE(rflow->cpu);
2812		if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2813		    ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2814			   rflow->last_qtail) <
2815		     (int)(10 * flow_table->mask)))
2816			expire = false;
2817	}
2818	rcu_read_unlock();
2819	return expire;
2820}
2821EXPORT_SYMBOL(rps_may_expire_flow);
2822
2823#endif /* CONFIG_RFS_ACCEL */
2824
2825/* Called from hardirq (IPI) context */
2826static void rps_trigger_softirq(void *data)
2827{
2828	struct softnet_data *sd = data;
2829
2830	____napi_schedule(sd, &sd->backlog);
2831	sd->received_rps++;
2832}
2833
2834#endif /* CONFIG_RPS */
2835
2836/*
2837 * Check if this softnet_data structure is another cpu one
2838 * If yes, queue it to our IPI list and return 1
2839 * If no, return 0
2840 */
2841static int rps_ipi_queued(struct softnet_data *sd)
2842{
2843#ifdef CONFIG_RPS
2844	struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2845
2846	if (sd != mysd) {
2847		sd->rps_ipi_next = mysd->rps_ipi_list;
2848		mysd->rps_ipi_list = sd;
2849
2850		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
2851		return 1;
2852	}
2853#endif /* CONFIG_RPS */
2854	return 0;
2855}
2856
2857/*
2858 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2859 * queue (may be a remote CPU queue).
2860 */
2861static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2862			      unsigned int *qtail)
2863{
2864	struct softnet_data *sd;
2865	unsigned long flags;
2866
2867	sd = &per_cpu(softnet_data, cpu);
2868
2869	local_irq_save(flags);
2870
2871	rps_lock(sd);
2872	if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2873		if (skb_queue_len(&sd->input_pkt_queue)) {
2874enqueue:
2875			__skb_queue_tail(&sd->input_pkt_queue, skb);
2876			input_queue_tail_incr_save(sd, qtail);
2877			rps_unlock(sd);
2878			local_irq_restore(flags);
2879			return NET_RX_SUCCESS;
2880		}
2881
2882		/* Schedule NAPI for backlog device
2883		 * We can use non atomic operation since we own the queue lock
2884		 */
2885		if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2886			if (!rps_ipi_queued(sd))
2887				____napi_schedule(sd, &sd->backlog);
2888		}
2889		goto enqueue;
2890	}
2891
2892	sd->dropped++;
2893	rps_unlock(sd);
2894
2895	local_irq_restore(flags);
2896
2897	atomic_long_inc(&skb->dev->rx_dropped);
2898	kfree_skb(skb);
2899	return NET_RX_DROP;
2900}
2901
2902/**
2903 *	netif_rx	-	post buffer to the network code
2904 *	@skb: buffer to post
2905 *
2906 *	This function receives a packet from a device driver and queues it for
2907 *	the upper (protocol) levels to process.  It always succeeds. The buffer
2908 *	may be dropped during processing for congestion control or by the
2909 *	protocol layers.
2910 *
2911 *	return values:
2912 *	NET_RX_SUCCESS	(no congestion)
2913 *	NET_RX_DROP     (packet was dropped)
2914 *
2915 */
2916
2917int netif_rx(struct sk_buff *skb)
2918{
2919	int ret;
2920
2921	/* if netpoll wants it, pretend we never saw it */
2922	if (netpoll_rx(skb))
2923		return NET_RX_DROP;
2924
2925	net_timestamp_check(netdev_tstamp_prequeue, skb);
 
2926
2927	trace_netif_rx(skb);
2928#ifdef CONFIG_RPS
2929	if (static_key_false(&rps_needed)) {
2930		struct rps_dev_flow voidflow, *rflow = &voidflow;
2931		int cpu;
2932
2933		preempt_disable();
2934		rcu_read_lock();
2935
2936		cpu = get_rps_cpu(skb->dev, skb, &rflow);
2937		if (cpu < 0)
2938			cpu = smp_processor_id();
2939
2940		ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2941
2942		rcu_read_unlock();
2943		preempt_enable();
2944	} else
2945#endif
2946	{
2947		unsigned int qtail;
2948		ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2949		put_cpu();
2950	}
 
2951	return ret;
2952}
2953EXPORT_SYMBOL(netif_rx);
2954
2955int netif_rx_ni(struct sk_buff *skb)
2956{
2957	int err;
2958
2959	preempt_disable();
2960	err = netif_rx(skb);
2961	if (local_softirq_pending())
2962		do_softirq();
2963	preempt_enable();
2964
2965	return err;
2966}
2967EXPORT_SYMBOL(netif_rx_ni);
2968
2969static void net_tx_action(struct softirq_action *h)
2970{
2971	struct softnet_data *sd = &__get_cpu_var(softnet_data);
2972
2973	if (sd->completion_queue) {
2974		struct sk_buff *clist;
2975
2976		local_irq_disable();
2977		clist = sd->completion_queue;
2978		sd->completion_queue = NULL;
2979		local_irq_enable();
2980
2981		while (clist) {
2982			struct sk_buff *skb = clist;
2983			clist = clist->next;
2984
2985			WARN_ON(atomic_read(&skb->users));
2986			trace_kfree_skb(skb, net_tx_action);
2987			__kfree_skb(skb);
2988		}
2989	}
2990
2991	if (sd->output_queue) {
2992		struct Qdisc *head;
2993
2994		local_irq_disable();
2995		head = sd->output_queue;
2996		sd->output_queue = NULL;
2997		sd->output_queue_tailp = &sd->output_queue;
2998		local_irq_enable();
2999
3000		while (head) {
3001			struct Qdisc *q = head;
3002			spinlock_t *root_lock;
3003
3004			head = head->next_sched;
3005
3006			root_lock = qdisc_lock(q);
3007			if (spin_trylock(root_lock)) {
3008				smp_mb__before_clear_bit();
3009				clear_bit(__QDISC_STATE_SCHED,
3010					  &q->state);
3011				qdisc_run(q);
3012				spin_unlock(root_lock);
3013			} else {
3014				if (!test_bit(__QDISC_STATE_DEACTIVATED,
3015					      &q->state)) {
3016					__netif_reschedule(q);
3017				} else {
3018					smp_mb__before_clear_bit();
3019					clear_bit(__QDISC_STATE_SCHED,
3020						  &q->state);
3021				}
3022			}
3023		}
3024	}
3025}
3026
3027#if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3028    (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3029/* This hook is defined here for ATM LANE */
3030int (*br_fdb_test_addr_hook)(struct net_device *dev,
3031			     unsigned char *addr) __read_mostly;
3032EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3033#endif
3034
3035#ifdef CONFIG_NET_CLS_ACT
3036/* TODO: Maybe we should just force sch_ingress to be compiled in
3037 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3038 * a compare and 2 stores extra right now if we dont have it on
3039 * but have CONFIG_NET_CLS_ACT
3040 * NOTE: This doesn't stop any functionality; if you dont have
3041 * the ingress scheduler, you just can't add policies on ingress.
3042 *
3043 */
3044static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3045{
3046	struct net_device *dev = skb->dev;
3047	u32 ttl = G_TC_RTTL(skb->tc_verd);
3048	int result = TC_ACT_OK;
3049	struct Qdisc *q;
3050
3051	if (unlikely(MAX_RED_LOOP < ttl++)) {
3052		net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3053				     skb->skb_iif, dev->ifindex);
 
3054		return TC_ACT_SHOT;
3055	}
3056
3057	skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3058	skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3059
3060	q = rxq->qdisc;
3061	if (q != &noop_qdisc) {
3062		spin_lock(qdisc_lock(q));
3063		if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3064			result = qdisc_enqueue_root(skb, q);
3065		spin_unlock(qdisc_lock(q));
3066	}
3067
3068	return result;
3069}
3070
3071static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3072					 struct packet_type **pt_prev,
3073					 int *ret, struct net_device *orig_dev)
3074{
3075	struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3076
3077	if (!rxq || rxq->qdisc == &noop_qdisc)
3078		goto out;
3079
3080	if (*pt_prev) {
3081		*ret = deliver_skb(skb, *pt_prev, orig_dev);
3082		*pt_prev = NULL;
3083	}
3084
3085	switch (ing_filter(skb, rxq)) {
3086	case TC_ACT_SHOT:
3087	case TC_ACT_STOLEN:
3088		kfree_skb(skb);
3089		return NULL;
3090	}
3091
3092out:
3093	skb->tc_verd = 0;
3094	return skb;
3095}
3096#endif
3097
3098/**
3099 *	netdev_rx_handler_register - register receive handler
3100 *	@dev: device to register a handler for
3101 *	@rx_handler: receive handler to register
3102 *	@rx_handler_data: data pointer that is used by rx handler
3103 *
3104 *	Register a receive hander for a device. This handler will then be
3105 *	called from __netif_receive_skb. A negative errno code is returned
3106 *	on a failure.
3107 *
3108 *	The caller must hold the rtnl_mutex.
3109 *
3110 *	For a general description of rx_handler, see enum rx_handler_result.
3111 */
3112int netdev_rx_handler_register(struct net_device *dev,
3113			       rx_handler_func_t *rx_handler,
3114			       void *rx_handler_data)
3115{
3116	ASSERT_RTNL();
3117
3118	if (dev->rx_handler)
3119		return -EBUSY;
3120
3121	rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3122	rcu_assign_pointer(dev->rx_handler, rx_handler);
3123
3124	return 0;
3125}
3126EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3127
3128/**
3129 *	netdev_rx_handler_unregister - unregister receive handler
3130 *	@dev: device to unregister a handler from
3131 *
3132 *	Unregister a receive hander from a device.
3133 *
3134 *	The caller must hold the rtnl_mutex.
3135 */
3136void netdev_rx_handler_unregister(struct net_device *dev)
3137{
3138
3139	ASSERT_RTNL();
3140	RCU_INIT_POINTER(dev->rx_handler, NULL);
3141	RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3142}
3143EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3144
3145static int __netif_receive_skb(struct sk_buff *skb)
3146{
3147	struct packet_type *ptype, *pt_prev;
3148	rx_handler_func_t *rx_handler;
3149	struct net_device *orig_dev;
3150	struct net_device *null_or_dev;
3151	bool deliver_exact = false;
3152	int ret = NET_RX_DROP;
3153	__be16 type;
3154
3155	net_timestamp_check(!netdev_tstamp_prequeue, skb);
 
3156
3157	trace_netif_receive_skb(skb);
3158
3159	/* if we've gotten here through NAPI, check netpoll */
3160	if (netpoll_receive_skb(skb))
3161		return NET_RX_DROP;
3162
3163	if (!skb->skb_iif)
3164		skb->skb_iif = skb->dev->ifindex;
3165	orig_dev = skb->dev;
3166
3167	skb_reset_network_header(skb);
3168	skb_reset_transport_header(skb);
3169	skb_reset_mac_len(skb);
3170
3171	pt_prev = NULL;
3172
3173	rcu_read_lock();
3174
3175another_round:
3176
3177	__this_cpu_inc(softnet_data.processed);
3178
3179	if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3180		skb = vlan_untag(skb);
3181		if (unlikely(!skb))
3182			goto out;
3183	}
3184
3185#ifdef CONFIG_NET_CLS_ACT
3186	if (skb->tc_verd & TC_NCLS) {
3187		skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3188		goto ncls;
3189	}
3190#endif
3191
3192	list_for_each_entry_rcu(ptype, &ptype_all, list) {
3193		if (!ptype->dev || ptype->dev == skb->dev) {
3194			if (pt_prev)
3195				ret = deliver_skb(skb, pt_prev, orig_dev);
3196			pt_prev = ptype;
3197		}
3198	}
3199
3200#ifdef CONFIG_NET_CLS_ACT
3201	skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3202	if (!skb)
3203		goto out;
3204ncls:
3205#endif
3206
3207	rx_handler = rcu_dereference(skb->dev->rx_handler);
3208	if (vlan_tx_tag_present(skb)) {
3209		if (pt_prev) {
3210			ret = deliver_skb(skb, pt_prev, orig_dev);
3211			pt_prev = NULL;
3212		}
3213		if (vlan_do_receive(&skb, !rx_handler))
3214			goto another_round;
3215		else if (unlikely(!skb))
3216			goto out;
3217	}
3218
3219	if (rx_handler) {
3220		if (pt_prev) {
3221			ret = deliver_skb(skb, pt_prev, orig_dev);
3222			pt_prev = NULL;
3223		}
3224		switch (rx_handler(&skb)) {
3225		case RX_HANDLER_CONSUMED:
3226			goto out;
3227		case RX_HANDLER_ANOTHER:
3228			goto another_round;
3229		case RX_HANDLER_EXACT:
3230			deliver_exact = true;
3231		case RX_HANDLER_PASS:
3232			break;
3233		default:
3234			BUG();
3235		}
3236	}
3237
 
 
 
 
 
 
 
 
 
 
 
 
3238	/* deliver only exact match when indicated */
3239	null_or_dev = deliver_exact ? skb->dev : NULL;
3240
3241	type = skb->protocol;
3242	list_for_each_entry_rcu(ptype,
3243			&ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3244		if (ptype->type == type &&
3245		    (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3246		     ptype->dev == orig_dev)) {
3247			if (pt_prev)
3248				ret = deliver_skb(skb, pt_prev, orig_dev);
3249			pt_prev = ptype;
3250		}
3251	}
3252
3253	if (pt_prev) {
3254		ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3255	} else {
3256		atomic_long_inc(&skb->dev->rx_dropped);
3257		kfree_skb(skb);
3258		/* Jamal, now you will not able to escape explaining
3259		 * me how you were going to use this. :-)
3260		 */
3261		ret = NET_RX_DROP;
3262	}
3263
3264out:
3265	rcu_read_unlock();
3266	return ret;
3267}
3268
3269/**
3270 *	netif_receive_skb - process receive buffer from network
3271 *	@skb: buffer to process
3272 *
3273 *	netif_receive_skb() is the main receive data processing function.
3274 *	It always succeeds. The buffer may be dropped during processing
3275 *	for congestion control or by the protocol layers.
3276 *
3277 *	This function may only be called from softirq context and interrupts
3278 *	should be enabled.
3279 *
3280 *	Return values (usually ignored):
3281 *	NET_RX_SUCCESS: no congestion
3282 *	NET_RX_DROP: packet was dropped
3283 */
3284int netif_receive_skb(struct sk_buff *skb)
3285{
3286	net_timestamp_check(netdev_tstamp_prequeue, skb);
 
3287
3288	if (skb_defer_rx_timestamp(skb))
3289		return NET_RX_SUCCESS;
3290
3291#ifdef CONFIG_RPS
3292	if (static_key_false(&rps_needed)) {
3293		struct rps_dev_flow voidflow, *rflow = &voidflow;
3294		int cpu, ret;
3295
3296		rcu_read_lock();
3297
3298		cpu = get_rps_cpu(skb->dev, skb, &rflow);
3299
3300		if (cpu >= 0) {
3301			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3302			rcu_read_unlock();
3303			return ret;
 
 
3304		}
3305		rcu_read_unlock();
 
3306	}
3307#endif
3308	return __netif_receive_skb(skb);
 
3309}
3310EXPORT_SYMBOL(netif_receive_skb);
3311
3312/* Network device is going away, flush any packets still pending
3313 * Called with irqs disabled.
3314 */
3315static void flush_backlog(void *arg)
3316{
3317	struct net_device *dev = arg;
3318	struct softnet_data *sd = &__get_cpu_var(softnet_data);
3319	struct sk_buff *skb, *tmp;
3320
3321	rps_lock(sd);
3322	skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3323		if (skb->dev == dev) {
3324			__skb_unlink(skb, &sd->input_pkt_queue);
3325			kfree_skb(skb);
3326			input_queue_head_incr(sd);
3327		}
3328	}
3329	rps_unlock(sd);
3330
3331	skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3332		if (skb->dev == dev) {
3333			__skb_unlink(skb, &sd->process_queue);
3334			kfree_skb(skb);
3335			input_queue_head_incr(sd);
3336		}
3337	}
3338}
3339
3340static int napi_gro_complete(struct sk_buff *skb)
3341{
3342	struct packet_type *ptype;
3343	__be16 type = skb->protocol;
3344	struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3345	int err = -ENOENT;
3346
3347	if (NAPI_GRO_CB(skb)->count == 1) {
3348		skb_shinfo(skb)->gso_size = 0;
3349		goto out;
3350	}
3351
3352	rcu_read_lock();
3353	list_for_each_entry_rcu(ptype, head, list) {
3354		if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3355			continue;
3356
3357		err = ptype->gro_complete(skb);
3358		break;
3359	}
3360	rcu_read_unlock();
3361
3362	if (err) {
3363		WARN_ON(&ptype->list == head);
3364		kfree_skb(skb);
3365		return NET_RX_SUCCESS;
3366	}
3367
3368out:
3369	return netif_receive_skb(skb);
3370}
3371
3372inline void napi_gro_flush(struct napi_struct *napi)
3373{
3374	struct sk_buff *skb, *next;
3375
3376	for (skb = napi->gro_list; skb; skb = next) {
3377		next = skb->next;
3378		skb->next = NULL;
3379		napi_gro_complete(skb);
3380	}
3381
3382	napi->gro_count = 0;
3383	napi->gro_list = NULL;
3384}
3385EXPORT_SYMBOL(napi_gro_flush);
3386
3387enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3388{
3389	struct sk_buff **pp = NULL;
3390	struct packet_type *ptype;
3391	__be16 type = skb->protocol;
3392	struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3393	int same_flow;
3394	int mac_len;
3395	enum gro_result ret;
3396
3397	if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3398		goto normal;
3399
3400	if (skb_is_gso(skb) || skb_has_frag_list(skb))
3401		goto normal;
3402
3403	rcu_read_lock();
3404	list_for_each_entry_rcu(ptype, head, list) {
3405		if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3406			continue;
3407
3408		skb_set_network_header(skb, skb_gro_offset(skb));
3409		mac_len = skb->network_header - skb->mac_header;
3410		skb->mac_len = mac_len;
3411		NAPI_GRO_CB(skb)->same_flow = 0;
3412		NAPI_GRO_CB(skb)->flush = 0;
3413		NAPI_GRO_CB(skb)->free = 0;
3414
3415		pp = ptype->gro_receive(&napi->gro_list, skb);
3416		break;
3417	}
3418	rcu_read_unlock();
3419
3420	if (&ptype->list == head)
3421		goto normal;
3422
3423	same_flow = NAPI_GRO_CB(skb)->same_flow;
3424	ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3425
3426	if (pp) {
3427		struct sk_buff *nskb = *pp;
3428
3429		*pp = nskb->next;
3430		nskb->next = NULL;
3431		napi_gro_complete(nskb);
3432		napi->gro_count--;
3433	}
3434
3435	if (same_flow)
3436		goto ok;
3437
3438	if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3439		goto normal;
3440
3441	napi->gro_count++;
3442	NAPI_GRO_CB(skb)->count = 1;
3443	skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3444	skb->next = napi->gro_list;
3445	napi->gro_list = skb;
3446	ret = GRO_HELD;
3447
3448pull:
3449	if (skb_headlen(skb) < skb_gro_offset(skb)) {
3450		int grow = skb_gro_offset(skb) - skb_headlen(skb);
3451
3452		BUG_ON(skb->end - skb->tail < grow);
3453
3454		memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3455
3456		skb->tail += grow;
3457		skb->data_len -= grow;
3458
3459		skb_shinfo(skb)->frags[0].page_offset += grow;
3460		skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3461
3462		if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3463			skb_frag_unref(skb, 0);
3464			memmove(skb_shinfo(skb)->frags,
3465				skb_shinfo(skb)->frags + 1,
3466				--skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3467		}
3468	}
3469
3470ok:
3471	return ret;
3472
3473normal:
3474	ret = GRO_NORMAL;
3475	goto pull;
3476}
3477EXPORT_SYMBOL(dev_gro_receive);
3478
3479static inline gro_result_t
3480__napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3481{
3482	struct sk_buff *p;
3483	unsigned int maclen = skb->dev->hard_header_len;
3484
3485	for (p = napi->gro_list; p; p = p->next) {
3486		unsigned long diffs;
3487
3488		diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3489		diffs |= p->vlan_tci ^ skb->vlan_tci;
3490		if (maclen == ETH_HLEN)
3491			diffs |= compare_ether_header(skb_mac_header(p),
3492						      skb_gro_mac_header(skb));
3493		else if (!diffs)
3494			diffs = memcmp(skb_mac_header(p),
3495				       skb_gro_mac_header(skb),
3496				       maclen);
3497		NAPI_GRO_CB(p)->same_flow = !diffs;
3498		NAPI_GRO_CB(p)->flush = 0;
3499	}
3500
3501	return dev_gro_receive(napi, skb);
3502}
3503
3504gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3505{
3506	switch (ret) {
3507	case GRO_NORMAL:
3508		if (netif_receive_skb(skb))
3509			ret = GRO_DROP;
3510		break;
3511
3512	case GRO_DROP:
3513		kfree_skb(skb);
3514		break;
3515
3516	case GRO_MERGED_FREE:
3517		if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3518			kmem_cache_free(skbuff_head_cache, skb);
3519		else
3520			__kfree_skb(skb);
3521		break;
3522
3523	case GRO_HELD:
3524	case GRO_MERGED:
3525		break;
3526	}
3527
3528	return ret;
3529}
3530EXPORT_SYMBOL(napi_skb_finish);
3531
3532void skb_gro_reset_offset(struct sk_buff *skb)
3533{
3534	NAPI_GRO_CB(skb)->data_offset = 0;
3535	NAPI_GRO_CB(skb)->frag0 = NULL;
3536	NAPI_GRO_CB(skb)->frag0_len = 0;
3537
3538	if (skb->mac_header == skb->tail &&
3539	    !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3540		NAPI_GRO_CB(skb)->frag0 =
3541			skb_frag_address(&skb_shinfo(skb)->frags[0]);
3542		NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
 
3543	}
3544}
3545EXPORT_SYMBOL(skb_gro_reset_offset);
3546
3547gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3548{
3549	skb_gro_reset_offset(skb);
3550
3551	return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3552}
3553EXPORT_SYMBOL(napi_gro_receive);
3554
3555static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3556{
3557	__skb_pull(skb, skb_headlen(skb));
3558	/* restore the reserve we had after netdev_alloc_skb_ip_align() */
3559	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3560	skb->vlan_tci = 0;
3561	skb->dev = napi->dev;
3562	skb->skb_iif = 0;
3563
3564	napi->skb = skb;
3565}
3566
3567struct sk_buff *napi_get_frags(struct napi_struct *napi)
3568{
3569	struct sk_buff *skb = napi->skb;
3570
3571	if (!skb) {
3572		skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3573		if (skb)
3574			napi->skb = skb;
3575	}
3576	return skb;
3577}
3578EXPORT_SYMBOL(napi_get_frags);
3579
3580gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3581			       gro_result_t ret)
3582{
3583	switch (ret) {
3584	case GRO_NORMAL:
3585	case GRO_HELD:
3586		skb->protocol = eth_type_trans(skb, skb->dev);
3587
3588		if (ret == GRO_HELD)
3589			skb_gro_pull(skb, -ETH_HLEN);
3590		else if (netif_receive_skb(skb))
3591			ret = GRO_DROP;
3592		break;
3593
3594	case GRO_DROP:
3595	case GRO_MERGED_FREE:
3596		napi_reuse_skb(napi, skb);
3597		break;
3598
3599	case GRO_MERGED:
3600		break;
3601	}
3602
3603	return ret;
3604}
3605EXPORT_SYMBOL(napi_frags_finish);
3606
3607static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3608{
3609	struct sk_buff *skb = napi->skb;
3610	struct ethhdr *eth;
3611	unsigned int hlen;
3612	unsigned int off;
3613
3614	napi->skb = NULL;
3615
3616	skb_reset_mac_header(skb);
3617	skb_gro_reset_offset(skb);
3618
3619	off = skb_gro_offset(skb);
3620	hlen = off + sizeof(*eth);
3621	eth = skb_gro_header_fast(skb, off);
3622	if (skb_gro_header_hard(skb, hlen)) {
3623		eth = skb_gro_header_slow(skb, hlen, off);
3624		if (unlikely(!eth)) {
3625			napi_reuse_skb(napi, skb);
3626			skb = NULL;
3627			goto out;
3628		}
3629	}
3630
3631	skb_gro_pull(skb, sizeof(*eth));
3632
3633	/*
3634	 * This works because the only protocols we care about don't require
3635	 * special handling.  We'll fix it up properly at the end.
3636	 */
3637	skb->protocol = eth->h_proto;
3638
3639out:
3640	return skb;
3641}
 
3642
3643gro_result_t napi_gro_frags(struct napi_struct *napi)
3644{
3645	struct sk_buff *skb = napi_frags_skb(napi);
3646
3647	if (!skb)
3648		return GRO_DROP;
3649
3650	return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3651}
3652EXPORT_SYMBOL(napi_gro_frags);
3653
3654/*
3655 * net_rps_action sends any pending IPI's for rps.
3656 * Note: called with local irq disabled, but exits with local irq enabled.
3657 */
3658static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3659{
3660#ifdef CONFIG_RPS
3661	struct softnet_data *remsd = sd->rps_ipi_list;
3662
3663	if (remsd) {
3664		sd->rps_ipi_list = NULL;
3665
3666		local_irq_enable();
3667
3668		/* Send pending IPI's to kick RPS processing on remote cpus. */
3669		while (remsd) {
3670			struct softnet_data *next = remsd->rps_ipi_next;
3671
3672			if (cpu_online(remsd->cpu))
3673				__smp_call_function_single(remsd->cpu,
3674							   &remsd->csd, 0);
3675			remsd = next;
3676		}
3677	} else
3678#endif
3679		local_irq_enable();
3680}
3681
3682static int process_backlog(struct napi_struct *napi, int quota)
3683{
3684	int work = 0;
3685	struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3686
3687#ifdef CONFIG_RPS
3688	/* Check if we have pending ipi, its better to send them now,
3689	 * not waiting net_rx_action() end.
3690	 */
3691	if (sd->rps_ipi_list) {
3692		local_irq_disable();
3693		net_rps_action_and_irq_enable(sd);
3694	}
3695#endif
3696	napi->weight = weight_p;
3697	local_irq_disable();
3698	while (work < quota) {
3699		struct sk_buff *skb;
3700		unsigned int qlen;
3701
3702		while ((skb = __skb_dequeue(&sd->process_queue))) {
3703			local_irq_enable();
3704			__netif_receive_skb(skb);
3705			local_irq_disable();
3706			input_queue_head_incr(sd);
3707			if (++work >= quota) {
3708				local_irq_enable();
3709				return work;
3710			}
3711		}
3712
3713		rps_lock(sd);
3714		qlen = skb_queue_len(&sd->input_pkt_queue);
3715		if (qlen)
3716			skb_queue_splice_tail_init(&sd->input_pkt_queue,
3717						   &sd->process_queue);
3718
3719		if (qlen < quota - work) {
3720			/*
3721			 * Inline a custom version of __napi_complete().
3722			 * only current cpu owns and manipulates this napi,
3723			 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3724			 * we can use a plain write instead of clear_bit(),
3725			 * and we dont need an smp_mb() memory barrier.
3726			 */
3727			list_del(&napi->poll_list);
3728			napi->state = 0;
3729
3730			quota = work + qlen;
3731		}
3732		rps_unlock(sd);
3733	}
3734	local_irq_enable();
3735
3736	return work;
3737}
3738
3739/**
3740 * __napi_schedule - schedule for receive
3741 * @n: entry to schedule
3742 *
3743 * The entry's receive function will be scheduled to run
3744 */
3745void __napi_schedule(struct napi_struct *n)
3746{
3747	unsigned long flags;
3748
3749	local_irq_save(flags);
3750	____napi_schedule(&__get_cpu_var(softnet_data), n);
3751	local_irq_restore(flags);
3752}
3753EXPORT_SYMBOL(__napi_schedule);
3754
3755void __napi_complete(struct napi_struct *n)
3756{
3757	BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3758	BUG_ON(n->gro_list);
3759
3760	list_del(&n->poll_list);
3761	smp_mb__before_clear_bit();
3762	clear_bit(NAPI_STATE_SCHED, &n->state);
3763}
3764EXPORT_SYMBOL(__napi_complete);
3765
3766void napi_complete(struct napi_struct *n)
3767{
3768	unsigned long flags;
3769
3770	/*
3771	 * don't let napi dequeue from the cpu poll list
3772	 * just in case its running on a different cpu
3773	 */
3774	if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3775		return;
3776
3777	napi_gro_flush(n);
3778	local_irq_save(flags);
3779	__napi_complete(n);
3780	local_irq_restore(flags);
3781}
3782EXPORT_SYMBOL(napi_complete);
3783
3784void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3785		    int (*poll)(struct napi_struct *, int), int weight)
3786{
3787	INIT_LIST_HEAD(&napi->poll_list);
3788	napi->gro_count = 0;
3789	napi->gro_list = NULL;
3790	napi->skb = NULL;
3791	napi->poll = poll;
3792	napi->weight = weight;
3793	list_add(&napi->dev_list, &dev->napi_list);
3794	napi->dev = dev;
3795#ifdef CONFIG_NETPOLL
3796	spin_lock_init(&napi->poll_lock);
3797	napi->poll_owner = -1;
3798#endif
3799	set_bit(NAPI_STATE_SCHED, &napi->state);
3800}
3801EXPORT_SYMBOL(netif_napi_add);
3802
3803void netif_napi_del(struct napi_struct *napi)
3804{
3805	struct sk_buff *skb, *next;
3806
3807	list_del_init(&napi->dev_list);
3808	napi_free_frags(napi);
3809
3810	for (skb = napi->gro_list; skb; skb = next) {
3811		next = skb->next;
3812		skb->next = NULL;
3813		kfree_skb(skb);
3814	}
3815
3816	napi->gro_list = NULL;
3817	napi->gro_count = 0;
3818}
3819EXPORT_SYMBOL(netif_napi_del);
3820
3821static void net_rx_action(struct softirq_action *h)
3822{
3823	struct softnet_data *sd = &__get_cpu_var(softnet_data);
3824	unsigned long time_limit = jiffies + 2;
3825	int budget = netdev_budget;
3826	void *have;
3827
3828	local_irq_disable();
3829
3830	while (!list_empty(&sd->poll_list)) {
3831		struct napi_struct *n;
3832		int work, weight;
3833
3834		/* If softirq window is exhuasted then punt.
3835		 * Allow this to run for 2 jiffies since which will allow
3836		 * an average latency of 1.5/HZ.
3837		 */
3838		if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3839			goto softnet_break;
3840
3841		local_irq_enable();
3842
3843		/* Even though interrupts have been re-enabled, this
3844		 * access is safe because interrupts can only add new
3845		 * entries to the tail of this list, and only ->poll()
3846		 * calls can remove this head entry from the list.
3847		 */
3848		n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3849
3850		have = netpoll_poll_lock(n);
3851
3852		weight = n->weight;
3853
3854		/* This NAPI_STATE_SCHED test is for avoiding a race
3855		 * with netpoll's poll_napi().  Only the entity which
3856		 * obtains the lock and sees NAPI_STATE_SCHED set will
3857		 * actually make the ->poll() call.  Therefore we avoid
3858		 * accidentally calling ->poll() when NAPI is not scheduled.
3859		 */
3860		work = 0;
3861		if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3862			work = n->poll(n, weight);
3863			trace_napi_poll(n);
3864		}
3865
3866		WARN_ON_ONCE(work > weight);
3867
3868		budget -= work;
3869
3870		local_irq_disable();
3871
3872		/* Drivers must not modify the NAPI state if they
3873		 * consume the entire weight.  In such cases this code
3874		 * still "owns" the NAPI instance and therefore can
3875		 * move the instance around on the list at-will.
3876		 */
3877		if (unlikely(work == weight)) {
3878			if (unlikely(napi_disable_pending(n))) {
3879				local_irq_enable();
3880				napi_complete(n);
3881				local_irq_disable();
3882			} else
3883				list_move_tail(&n->poll_list, &sd->poll_list);
3884		}
3885
3886		netpoll_poll_unlock(have);
3887	}
3888out:
3889	net_rps_action_and_irq_enable(sd);
3890
3891#ifdef CONFIG_NET_DMA
3892	/*
3893	 * There may not be any more sk_buffs coming right now, so push
3894	 * any pending DMA copies to hardware
3895	 */
3896	dma_issue_pending_all();
3897#endif
3898
3899	return;
3900
3901softnet_break:
3902	sd->time_squeeze++;
3903	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
3904	goto out;
3905}
3906
3907static gifconf_func_t *gifconf_list[NPROTO];
3908
3909/**
3910 *	register_gifconf	-	register a SIOCGIF handler
3911 *	@family: Address family
3912 *	@gifconf: Function handler
3913 *
3914 *	Register protocol dependent address dumping routines. The handler
3915 *	that is passed must not be freed or reused until it has been replaced
3916 *	by another handler.
3917 */
3918int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3919{
3920	if (family >= NPROTO)
3921		return -EINVAL;
3922	gifconf_list[family] = gifconf;
3923	return 0;
3924}
3925EXPORT_SYMBOL(register_gifconf);
3926
3927
3928/*
3929 *	Map an interface index to its name (SIOCGIFNAME)
3930 */
3931
3932/*
3933 *	We need this ioctl for efficient implementation of the
3934 *	if_indextoname() function required by the IPv6 API.  Without
3935 *	it, we would have to search all the interfaces to find a
3936 *	match.  --pb
3937 */
3938
3939static int dev_ifname(struct net *net, struct ifreq __user *arg)
3940{
3941	struct net_device *dev;
3942	struct ifreq ifr;
3943
3944	/*
3945	 *	Fetch the caller's info block.
3946	 */
3947
3948	if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3949		return -EFAULT;
3950
3951	rcu_read_lock();
3952	dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3953	if (!dev) {
3954		rcu_read_unlock();
3955		return -ENODEV;
3956	}
3957
3958	strcpy(ifr.ifr_name, dev->name);
3959	rcu_read_unlock();
3960
3961	if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3962		return -EFAULT;
3963	return 0;
3964}
3965
3966/*
3967 *	Perform a SIOCGIFCONF call. This structure will change
3968 *	size eventually, and there is nothing I can do about it.
3969 *	Thus we will need a 'compatibility mode'.
3970 */
3971
3972static int dev_ifconf(struct net *net, char __user *arg)
3973{
3974	struct ifconf ifc;
3975	struct net_device *dev;
3976	char __user *pos;
3977	int len;
3978	int total;
3979	int i;
3980
3981	/*
3982	 *	Fetch the caller's info block.
3983	 */
3984
3985	if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3986		return -EFAULT;
3987
3988	pos = ifc.ifc_buf;
3989	len = ifc.ifc_len;
3990
3991	/*
3992	 *	Loop over the interfaces, and write an info block for each.
3993	 */
3994
3995	total = 0;
3996	for_each_netdev(net, dev) {
3997		for (i = 0; i < NPROTO; i++) {
3998			if (gifconf_list[i]) {
3999				int done;
4000				if (!pos)
4001					done = gifconf_list[i](dev, NULL, 0);
4002				else
4003					done = gifconf_list[i](dev, pos + total,
4004							       len - total);
4005				if (done < 0)
4006					return -EFAULT;
4007				total += done;
4008			}
4009		}
4010	}
4011
4012	/*
4013	 *	All done.  Write the updated control block back to the caller.
4014	 */
4015	ifc.ifc_len = total;
4016
4017	/*
4018	 * 	Both BSD and Solaris return 0 here, so we do too.
4019	 */
4020	return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4021}
4022
4023#ifdef CONFIG_PROC_FS
4024
4025#define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4026
4027#define get_bucket(x) ((x) >> BUCKET_SPACE)
4028#define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4029#define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4030
4031static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4032{
4033	struct net *net = seq_file_net(seq);
4034	struct net_device *dev;
4035	struct hlist_node *p;
4036	struct hlist_head *h;
4037	unsigned int count = 0, offset = get_offset(*pos);
4038
4039	h = &net->dev_name_head[get_bucket(*pos)];
4040	hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4041		if (++count == offset)
4042			return dev;
4043	}
4044
4045	return NULL;
4046}
4047
4048static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4049{
4050	struct net_device *dev;
4051	unsigned int bucket;
4052
4053	do {
4054		dev = dev_from_same_bucket(seq, pos);
4055		if (dev)
4056			return dev;
4057
4058		bucket = get_bucket(*pos) + 1;
4059		*pos = set_bucket_offset(bucket, 1);
4060	} while (bucket < NETDEV_HASHENTRIES);
4061
4062	return NULL;
4063}
4064
4065/*
4066 *	This is invoked by the /proc filesystem handler to display a device
4067 *	in detail.
4068 */
4069void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4070	__acquires(RCU)
4071{
 
 
 
 
4072	rcu_read_lock();
4073	if (!*pos)
4074		return SEQ_START_TOKEN;
4075
4076	if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4077		return NULL;
 
 
4078
4079	return dev_from_bucket(seq, pos);
4080}
4081
4082void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4083{
 
 
 
 
 
 
 
4084	++*pos;
4085	return dev_from_bucket(seq, pos);
4086}
4087
4088void dev_seq_stop(struct seq_file *seq, void *v)
4089	__releases(RCU)
4090{
4091	rcu_read_unlock();
4092}
4093
4094static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4095{
4096	struct rtnl_link_stats64 temp;
4097	const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4098
4099	seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4100		   "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4101		   dev->name, stats->rx_bytes, stats->rx_packets,
4102		   stats->rx_errors,
4103		   stats->rx_dropped + stats->rx_missed_errors,
4104		   stats->rx_fifo_errors,
4105		   stats->rx_length_errors + stats->rx_over_errors +
4106		    stats->rx_crc_errors + stats->rx_frame_errors,
4107		   stats->rx_compressed, stats->multicast,
4108		   stats->tx_bytes, stats->tx_packets,
4109		   stats->tx_errors, stats->tx_dropped,
4110		   stats->tx_fifo_errors, stats->collisions,
4111		   stats->tx_carrier_errors +
4112		    stats->tx_aborted_errors +
4113		    stats->tx_window_errors +
4114		    stats->tx_heartbeat_errors,
4115		   stats->tx_compressed);
4116}
4117
4118/*
4119 *	Called from the PROCfs module. This now uses the new arbitrary sized
4120 *	/proc/net interface to create /proc/net/dev
4121 */
4122static int dev_seq_show(struct seq_file *seq, void *v)
4123{
4124	if (v == SEQ_START_TOKEN)
4125		seq_puts(seq, "Inter-|   Receive                            "
4126			      "                    |  Transmit\n"
4127			      " face |bytes    packets errs drop fifo frame "
4128			      "compressed multicast|bytes    packets errs "
4129			      "drop fifo colls carrier compressed\n");
4130	else
4131		dev_seq_printf_stats(seq, v);
4132	return 0;
4133}
4134
4135static struct softnet_data *softnet_get_online(loff_t *pos)
4136{
4137	struct softnet_data *sd = NULL;
4138
4139	while (*pos < nr_cpu_ids)
4140		if (cpu_online(*pos)) {
4141			sd = &per_cpu(softnet_data, *pos);
4142			break;
4143		} else
4144			++*pos;
4145	return sd;
4146}
4147
4148static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4149{
4150	return softnet_get_online(pos);
4151}
4152
4153static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4154{
4155	++*pos;
4156	return softnet_get_online(pos);
4157}
4158
4159static void softnet_seq_stop(struct seq_file *seq, void *v)
4160{
4161}
4162
4163static int softnet_seq_show(struct seq_file *seq, void *v)
4164{
4165	struct softnet_data *sd = v;
4166
4167	seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4168		   sd->processed, sd->dropped, sd->time_squeeze, 0,
4169		   0, 0, 0, 0, /* was fastroute */
4170		   sd->cpu_collision, sd->received_rps);
4171	return 0;
4172}
4173
4174static const struct seq_operations dev_seq_ops = {
4175	.start = dev_seq_start,
4176	.next  = dev_seq_next,
4177	.stop  = dev_seq_stop,
4178	.show  = dev_seq_show,
4179};
4180
4181static int dev_seq_open(struct inode *inode, struct file *file)
4182{
4183	return seq_open_net(inode, file, &dev_seq_ops,
4184			    sizeof(struct seq_net_private));
4185}
4186
4187static const struct file_operations dev_seq_fops = {
4188	.owner	 = THIS_MODULE,
4189	.open    = dev_seq_open,
4190	.read    = seq_read,
4191	.llseek  = seq_lseek,
4192	.release = seq_release_net,
4193};
4194
4195static const struct seq_operations softnet_seq_ops = {
4196	.start = softnet_seq_start,
4197	.next  = softnet_seq_next,
4198	.stop  = softnet_seq_stop,
4199	.show  = softnet_seq_show,
4200};
4201
4202static int softnet_seq_open(struct inode *inode, struct file *file)
4203{
4204	return seq_open(file, &softnet_seq_ops);
4205}
4206
4207static const struct file_operations softnet_seq_fops = {
4208	.owner	 = THIS_MODULE,
4209	.open    = softnet_seq_open,
4210	.read    = seq_read,
4211	.llseek  = seq_lseek,
4212	.release = seq_release,
4213};
4214
4215static void *ptype_get_idx(loff_t pos)
4216{
4217	struct packet_type *pt = NULL;
4218	loff_t i = 0;
4219	int t;
4220
4221	list_for_each_entry_rcu(pt, &ptype_all, list) {
4222		if (i == pos)
4223			return pt;
4224		++i;
4225	}
4226
4227	for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4228		list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4229			if (i == pos)
4230				return pt;
4231			++i;
4232		}
4233	}
4234	return NULL;
4235}
4236
4237static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4238	__acquires(RCU)
4239{
4240	rcu_read_lock();
4241	return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4242}
4243
4244static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4245{
4246	struct packet_type *pt;
4247	struct list_head *nxt;
4248	int hash;
4249
4250	++*pos;
4251	if (v == SEQ_START_TOKEN)
4252		return ptype_get_idx(0);
4253
4254	pt = v;
4255	nxt = pt->list.next;
4256	if (pt->type == htons(ETH_P_ALL)) {
4257		if (nxt != &ptype_all)
4258			goto found;
4259		hash = 0;
4260		nxt = ptype_base[0].next;
4261	} else
4262		hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4263
4264	while (nxt == &ptype_base[hash]) {
4265		if (++hash >= PTYPE_HASH_SIZE)
4266			return NULL;
4267		nxt = ptype_base[hash].next;
4268	}
4269found:
4270	return list_entry(nxt, struct packet_type, list);
4271}
4272
4273static void ptype_seq_stop(struct seq_file *seq, void *v)
4274	__releases(RCU)
4275{
4276	rcu_read_unlock();
4277}
4278
4279static int ptype_seq_show(struct seq_file *seq, void *v)
4280{
4281	struct packet_type *pt = v;
4282
4283	if (v == SEQ_START_TOKEN)
4284		seq_puts(seq, "Type Device      Function\n");
4285	else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4286		if (pt->type == htons(ETH_P_ALL))
4287			seq_puts(seq, "ALL ");
4288		else
4289			seq_printf(seq, "%04x", ntohs(pt->type));
4290
4291		seq_printf(seq, " %-8s %pF\n",
4292			   pt->dev ? pt->dev->name : "", pt->func);
4293	}
4294
4295	return 0;
4296}
4297
4298static const struct seq_operations ptype_seq_ops = {
4299	.start = ptype_seq_start,
4300	.next  = ptype_seq_next,
4301	.stop  = ptype_seq_stop,
4302	.show  = ptype_seq_show,
4303};
4304
4305static int ptype_seq_open(struct inode *inode, struct file *file)
4306{
4307	return seq_open_net(inode, file, &ptype_seq_ops,
4308			sizeof(struct seq_net_private));
4309}
4310
4311static const struct file_operations ptype_seq_fops = {
4312	.owner	 = THIS_MODULE,
4313	.open    = ptype_seq_open,
4314	.read    = seq_read,
4315	.llseek  = seq_lseek,
4316	.release = seq_release_net,
4317};
4318
4319
4320static int __net_init dev_proc_net_init(struct net *net)
4321{
4322	int rc = -ENOMEM;
4323
4324	if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4325		goto out;
4326	if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4327		goto out_dev;
4328	if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4329		goto out_softnet;
4330
4331	if (wext_proc_init(net))
4332		goto out_ptype;
4333	rc = 0;
4334out:
4335	return rc;
4336out_ptype:
4337	proc_net_remove(net, "ptype");
4338out_softnet:
4339	proc_net_remove(net, "softnet_stat");
4340out_dev:
4341	proc_net_remove(net, "dev");
4342	goto out;
4343}
4344
4345static void __net_exit dev_proc_net_exit(struct net *net)
4346{
4347	wext_proc_exit(net);
4348
4349	proc_net_remove(net, "ptype");
4350	proc_net_remove(net, "softnet_stat");
4351	proc_net_remove(net, "dev");
4352}
4353
4354static struct pernet_operations __net_initdata dev_proc_ops = {
4355	.init = dev_proc_net_init,
4356	.exit = dev_proc_net_exit,
4357};
4358
4359static int __init dev_proc_init(void)
4360{
4361	return register_pernet_subsys(&dev_proc_ops);
4362}
4363#else
4364#define dev_proc_init() 0
4365#endif	/* CONFIG_PROC_FS */
4366
4367
4368/**
4369 *	netdev_set_master	-	set up master pointer
4370 *	@slave: slave device
4371 *	@master: new master device
4372 *
4373 *	Changes the master device of the slave. Pass %NULL to break the
4374 *	bonding. The caller must hold the RTNL semaphore. On a failure
4375 *	a negative errno code is returned. On success the reference counts
4376 *	are adjusted and the function returns zero.
4377 */
4378int netdev_set_master(struct net_device *slave, struct net_device *master)
4379{
4380	struct net_device *old = slave->master;
4381
4382	ASSERT_RTNL();
4383
4384	if (master) {
4385		if (old)
4386			return -EBUSY;
4387		dev_hold(master);
4388	}
4389
4390	slave->master = master;
4391
4392	if (old)
4393		dev_put(old);
4394	return 0;
4395}
4396EXPORT_SYMBOL(netdev_set_master);
4397
4398/**
4399 *	netdev_set_bond_master	-	set up bonding master/slave pair
4400 *	@slave: slave device
4401 *	@master: new master device
4402 *
4403 *	Changes the master device of the slave. Pass %NULL to break the
4404 *	bonding. The caller must hold the RTNL semaphore. On a failure
4405 *	a negative errno code is returned. On success %RTM_NEWLINK is sent
4406 *	to the routing socket and the function returns zero.
4407 */
4408int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4409{
4410	int err;
4411
4412	ASSERT_RTNL();
4413
4414	err = netdev_set_master(slave, master);
4415	if (err)
4416		return err;
4417	if (master)
4418		slave->flags |= IFF_SLAVE;
4419	else
4420		slave->flags &= ~IFF_SLAVE;
4421
4422	rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4423	return 0;
4424}
4425EXPORT_SYMBOL(netdev_set_bond_master);
4426
4427static void dev_change_rx_flags(struct net_device *dev, int flags)
4428{
4429	const struct net_device_ops *ops = dev->netdev_ops;
4430
4431	if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4432		ops->ndo_change_rx_flags(dev, flags);
4433}
4434
4435static int __dev_set_promiscuity(struct net_device *dev, int inc)
4436{
4437	unsigned int old_flags = dev->flags;
4438	uid_t uid;
4439	gid_t gid;
4440
4441	ASSERT_RTNL();
4442
4443	dev->flags |= IFF_PROMISC;
4444	dev->promiscuity += inc;
4445	if (dev->promiscuity == 0) {
4446		/*
4447		 * Avoid overflow.
4448		 * If inc causes overflow, untouch promisc and return error.
4449		 */
4450		if (inc < 0)
4451			dev->flags &= ~IFF_PROMISC;
4452		else {
4453			dev->promiscuity -= inc;
4454			pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4455				dev->name);
 
4456			return -EOVERFLOW;
4457		}
4458	}
4459	if (dev->flags != old_flags) {
4460		pr_info("device %s %s promiscuous mode\n",
4461			dev->name,
4462			dev->flags & IFF_PROMISC ? "entered" : "left");
4463		if (audit_enabled) {
4464			current_uid_gid(&uid, &gid);
4465			audit_log(current->audit_context, GFP_ATOMIC,
4466				AUDIT_ANOM_PROMISCUOUS,
4467				"dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4468				dev->name, (dev->flags & IFF_PROMISC),
4469				(old_flags & IFF_PROMISC),
4470				audit_get_loginuid(current),
4471				uid, gid,
4472				audit_get_sessionid(current));
4473		}
4474
4475		dev_change_rx_flags(dev, IFF_PROMISC);
4476	}
4477	return 0;
4478}
4479
4480/**
4481 *	dev_set_promiscuity	- update promiscuity count on a device
4482 *	@dev: device
4483 *	@inc: modifier
4484 *
4485 *	Add or remove promiscuity from a device. While the count in the device
4486 *	remains above zero the interface remains promiscuous. Once it hits zero
4487 *	the device reverts back to normal filtering operation. A negative inc
4488 *	value is used to drop promiscuity on the device.
4489 *	Return 0 if successful or a negative errno code on error.
4490 */
4491int dev_set_promiscuity(struct net_device *dev, int inc)
4492{
4493	unsigned int old_flags = dev->flags;
4494	int err;
4495
4496	err = __dev_set_promiscuity(dev, inc);
4497	if (err < 0)
4498		return err;
4499	if (dev->flags != old_flags)
4500		dev_set_rx_mode(dev);
4501	return err;
4502}
4503EXPORT_SYMBOL(dev_set_promiscuity);
4504
4505/**
4506 *	dev_set_allmulti	- update allmulti count on a device
4507 *	@dev: device
4508 *	@inc: modifier
4509 *
4510 *	Add or remove reception of all multicast frames to a device. While the
4511 *	count in the device remains above zero the interface remains listening
4512 *	to all interfaces. Once it hits zero the device reverts back to normal
4513 *	filtering operation. A negative @inc value is used to drop the counter
4514 *	when releasing a resource needing all multicasts.
4515 *	Return 0 if successful or a negative errno code on error.
4516 */
4517
4518int dev_set_allmulti(struct net_device *dev, int inc)
4519{
4520	unsigned int old_flags = dev->flags;
4521
4522	ASSERT_RTNL();
4523
4524	dev->flags |= IFF_ALLMULTI;
4525	dev->allmulti += inc;
4526	if (dev->allmulti == 0) {
4527		/*
4528		 * Avoid overflow.
4529		 * If inc causes overflow, untouch allmulti and return error.
4530		 */
4531		if (inc < 0)
4532			dev->flags &= ~IFF_ALLMULTI;
4533		else {
4534			dev->allmulti -= inc;
4535			pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4536				dev->name);
 
4537			return -EOVERFLOW;
4538		}
4539	}
4540	if (dev->flags ^ old_flags) {
4541		dev_change_rx_flags(dev, IFF_ALLMULTI);
4542		dev_set_rx_mode(dev);
4543	}
4544	return 0;
4545}
4546EXPORT_SYMBOL(dev_set_allmulti);
4547
4548/*
4549 *	Upload unicast and multicast address lists to device and
4550 *	configure RX filtering. When the device doesn't support unicast
4551 *	filtering it is put in promiscuous mode while unicast addresses
4552 *	are present.
4553 */
4554void __dev_set_rx_mode(struct net_device *dev)
4555{
4556	const struct net_device_ops *ops = dev->netdev_ops;
4557
4558	/* dev_open will call this function so the list will stay sane. */
4559	if (!(dev->flags&IFF_UP))
4560		return;
4561
4562	if (!netif_device_present(dev))
4563		return;
4564
4565	if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
 
 
4566		/* Unicast addresses changes may only happen under the rtnl,
4567		 * therefore calling __dev_set_promiscuity here is safe.
4568		 */
4569		if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4570			__dev_set_promiscuity(dev, 1);
4571			dev->uc_promisc = true;
4572		} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4573			__dev_set_promiscuity(dev, -1);
4574			dev->uc_promisc = false;
4575		}
4576	}
4577
4578	if (ops->ndo_set_rx_mode)
4579		ops->ndo_set_rx_mode(dev);
 
4580}
4581
4582void dev_set_rx_mode(struct net_device *dev)
4583{
4584	netif_addr_lock_bh(dev);
4585	__dev_set_rx_mode(dev);
4586	netif_addr_unlock_bh(dev);
4587}
4588
4589/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4590 *	dev_get_flags - get flags reported to userspace
4591 *	@dev: device
4592 *
4593 *	Get the combination of flag bits exported through APIs to userspace.
4594 */
4595unsigned int dev_get_flags(const struct net_device *dev)
4596{
4597	unsigned int flags;
4598
4599	flags = (dev->flags & ~(IFF_PROMISC |
4600				IFF_ALLMULTI |
4601				IFF_RUNNING |
4602				IFF_LOWER_UP |
4603				IFF_DORMANT)) |
4604		(dev->gflags & (IFF_PROMISC |
4605				IFF_ALLMULTI));
4606
4607	if (netif_running(dev)) {
4608		if (netif_oper_up(dev))
4609			flags |= IFF_RUNNING;
4610		if (netif_carrier_ok(dev))
4611			flags |= IFF_LOWER_UP;
4612		if (netif_dormant(dev))
4613			flags |= IFF_DORMANT;
4614	}
4615
4616	return flags;
4617}
4618EXPORT_SYMBOL(dev_get_flags);
4619
4620int __dev_change_flags(struct net_device *dev, unsigned int flags)
4621{
4622	unsigned int old_flags = dev->flags;
4623	int ret;
4624
4625	ASSERT_RTNL();
4626
4627	/*
4628	 *	Set the flags on our device.
4629	 */
4630
4631	dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4632			       IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4633			       IFF_AUTOMEDIA)) |
4634		     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4635				    IFF_ALLMULTI));
4636
4637	/*
4638	 *	Load in the correct multicast list now the flags have changed.
4639	 */
4640
4641	if ((old_flags ^ flags) & IFF_MULTICAST)
4642		dev_change_rx_flags(dev, IFF_MULTICAST);
4643
4644	dev_set_rx_mode(dev);
4645
4646	/*
4647	 *	Have we downed the interface. We handle IFF_UP ourselves
4648	 *	according to user attempts to set it, rather than blindly
4649	 *	setting it.
4650	 */
4651
4652	ret = 0;
4653	if ((old_flags ^ flags) & IFF_UP) {	/* Bit is different  ? */
4654		ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4655
4656		if (!ret)
4657			dev_set_rx_mode(dev);
4658	}
4659
4660	if ((flags ^ dev->gflags) & IFF_PROMISC) {
4661		int inc = (flags & IFF_PROMISC) ? 1 : -1;
4662
4663		dev->gflags ^= IFF_PROMISC;
4664		dev_set_promiscuity(dev, inc);
4665	}
4666
4667	/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4668	   is important. Some (broken) drivers set IFF_PROMISC, when
4669	   IFF_ALLMULTI is requested not asking us and not reporting.
4670	 */
4671	if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4672		int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4673
4674		dev->gflags ^= IFF_ALLMULTI;
4675		dev_set_allmulti(dev, inc);
4676	}
4677
4678	return ret;
4679}
4680
4681void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4682{
4683	unsigned int changes = dev->flags ^ old_flags;
4684
4685	if (changes & IFF_UP) {
4686		if (dev->flags & IFF_UP)
4687			call_netdevice_notifiers(NETDEV_UP, dev);
4688		else
4689			call_netdevice_notifiers(NETDEV_DOWN, dev);
4690	}
4691
4692	if (dev->flags & IFF_UP &&
4693	    (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4694		call_netdevice_notifiers(NETDEV_CHANGE, dev);
4695}
4696
4697/**
4698 *	dev_change_flags - change device settings
4699 *	@dev: device
4700 *	@flags: device state flags
4701 *
4702 *	Change settings on device based state flags. The flags are
4703 *	in the userspace exported format.
4704 */
4705int dev_change_flags(struct net_device *dev, unsigned int flags)
4706{
4707	int ret;
4708	unsigned int changes, old_flags = dev->flags;
4709
4710	ret = __dev_change_flags(dev, flags);
4711	if (ret < 0)
4712		return ret;
4713
4714	changes = old_flags ^ dev->flags;
4715	if (changes)
4716		rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4717
4718	__dev_notify_flags(dev, old_flags);
4719	return ret;
4720}
4721EXPORT_SYMBOL(dev_change_flags);
4722
4723/**
4724 *	dev_set_mtu - Change maximum transfer unit
4725 *	@dev: device
4726 *	@new_mtu: new transfer unit
4727 *
4728 *	Change the maximum transfer size of the network device.
4729 */
4730int dev_set_mtu(struct net_device *dev, int new_mtu)
4731{
4732	const struct net_device_ops *ops = dev->netdev_ops;
4733	int err;
4734
4735	if (new_mtu == dev->mtu)
4736		return 0;
4737
4738	/*	MTU must be positive.	 */
4739	if (new_mtu < 0)
4740		return -EINVAL;
4741
4742	if (!netif_device_present(dev))
4743		return -ENODEV;
4744
4745	err = 0;
4746	if (ops->ndo_change_mtu)
4747		err = ops->ndo_change_mtu(dev, new_mtu);
4748	else
4749		dev->mtu = new_mtu;
4750
4751	if (!err && dev->flags & IFF_UP)
4752		call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4753	return err;
4754}
4755EXPORT_SYMBOL(dev_set_mtu);
4756
4757/**
4758 *	dev_set_group - Change group this device belongs to
4759 *	@dev: device
4760 *	@new_group: group this device should belong to
4761 */
4762void dev_set_group(struct net_device *dev, int new_group)
4763{
4764	dev->group = new_group;
4765}
4766EXPORT_SYMBOL(dev_set_group);
4767
4768/**
4769 *	dev_set_mac_address - Change Media Access Control Address
4770 *	@dev: device
4771 *	@sa: new address
4772 *
4773 *	Change the hardware (MAC) address of the device
4774 */
4775int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4776{
4777	const struct net_device_ops *ops = dev->netdev_ops;
4778	int err;
4779
4780	if (!ops->ndo_set_mac_address)
4781		return -EOPNOTSUPP;
4782	if (sa->sa_family != dev->type)
4783		return -EINVAL;
4784	if (!netif_device_present(dev))
4785		return -ENODEV;
4786	err = ops->ndo_set_mac_address(dev, sa);
4787	if (!err)
4788		call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4789	add_device_randomness(dev->dev_addr, dev->addr_len);
4790	return err;
4791}
4792EXPORT_SYMBOL(dev_set_mac_address);
4793
4794/*
4795 *	Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4796 */
4797static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4798{
4799	int err;
4800	struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4801
4802	if (!dev)
4803		return -ENODEV;
4804
4805	switch (cmd) {
4806	case SIOCGIFFLAGS:	/* Get interface flags */
4807		ifr->ifr_flags = (short) dev_get_flags(dev);
4808		return 0;
4809
4810	case SIOCGIFMETRIC:	/* Get the metric on the interface
4811				   (currently unused) */
4812		ifr->ifr_metric = 0;
4813		return 0;
4814
4815	case SIOCGIFMTU:	/* Get the MTU of a device */
4816		ifr->ifr_mtu = dev->mtu;
4817		return 0;
4818
4819	case SIOCGIFHWADDR:
4820		if (!dev->addr_len)
4821			memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4822		else
4823			memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4824			       min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4825		ifr->ifr_hwaddr.sa_family = dev->type;
4826		return 0;
4827
4828	case SIOCGIFSLAVE:
4829		err = -EINVAL;
4830		break;
4831
4832	case SIOCGIFMAP:
4833		ifr->ifr_map.mem_start = dev->mem_start;
4834		ifr->ifr_map.mem_end   = dev->mem_end;
4835		ifr->ifr_map.base_addr = dev->base_addr;
4836		ifr->ifr_map.irq       = dev->irq;
4837		ifr->ifr_map.dma       = dev->dma;
4838		ifr->ifr_map.port      = dev->if_port;
4839		return 0;
4840
4841	case SIOCGIFINDEX:
4842		ifr->ifr_ifindex = dev->ifindex;
4843		return 0;
4844
4845	case SIOCGIFTXQLEN:
4846		ifr->ifr_qlen = dev->tx_queue_len;
4847		return 0;
4848
4849	default:
4850		/* dev_ioctl() should ensure this case
4851		 * is never reached
4852		 */
4853		WARN_ON(1);
4854		err = -ENOTTY;
4855		break;
4856
4857	}
4858	return err;
4859}
4860
4861/*
4862 *	Perform the SIOCxIFxxx calls, inside rtnl_lock()
4863 */
4864static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4865{
4866	int err;
4867	struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4868	const struct net_device_ops *ops;
4869
4870	if (!dev)
4871		return -ENODEV;
4872
4873	ops = dev->netdev_ops;
4874
4875	switch (cmd) {
4876	case SIOCSIFFLAGS:	/* Set interface flags */
4877		return dev_change_flags(dev, ifr->ifr_flags);
4878
4879	case SIOCSIFMETRIC:	/* Set the metric on the interface
4880				   (currently unused) */
4881		return -EOPNOTSUPP;
4882
4883	case SIOCSIFMTU:	/* Set the MTU of a device */
4884		return dev_set_mtu(dev, ifr->ifr_mtu);
4885
4886	case SIOCSIFHWADDR:
4887		return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4888
4889	case SIOCSIFHWBROADCAST:
4890		if (ifr->ifr_hwaddr.sa_family != dev->type)
4891			return -EINVAL;
4892		memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4893		       min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4894		call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4895		return 0;
4896
4897	case SIOCSIFMAP:
4898		if (ops->ndo_set_config) {
4899			if (!netif_device_present(dev))
4900				return -ENODEV;
4901			return ops->ndo_set_config(dev, &ifr->ifr_map);
4902		}
4903		return -EOPNOTSUPP;
4904
4905	case SIOCADDMULTI:
4906		if (!ops->ndo_set_rx_mode ||
4907		    ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4908			return -EINVAL;
4909		if (!netif_device_present(dev))
4910			return -ENODEV;
4911		return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4912
4913	case SIOCDELMULTI:
4914		if (!ops->ndo_set_rx_mode ||
4915		    ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4916			return -EINVAL;
4917		if (!netif_device_present(dev))
4918			return -ENODEV;
4919		return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4920
4921	case SIOCSIFTXQLEN:
4922		if (ifr->ifr_qlen < 0)
4923			return -EINVAL;
4924		dev->tx_queue_len = ifr->ifr_qlen;
4925		return 0;
4926
4927	case SIOCSIFNAME:
4928		ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4929		return dev_change_name(dev, ifr->ifr_newname);
4930
4931	case SIOCSHWTSTAMP:
4932		err = net_hwtstamp_validate(ifr);
4933		if (err)
4934			return err;
4935		/* fall through */
4936
4937	/*
4938	 *	Unknown or private ioctl
4939	 */
4940	default:
4941		if ((cmd >= SIOCDEVPRIVATE &&
4942		    cmd <= SIOCDEVPRIVATE + 15) ||
4943		    cmd == SIOCBONDENSLAVE ||
4944		    cmd == SIOCBONDRELEASE ||
4945		    cmd == SIOCBONDSETHWADDR ||
4946		    cmd == SIOCBONDSLAVEINFOQUERY ||
4947		    cmd == SIOCBONDINFOQUERY ||
4948		    cmd == SIOCBONDCHANGEACTIVE ||
4949		    cmd == SIOCGMIIPHY ||
4950		    cmd == SIOCGMIIREG ||
4951		    cmd == SIOCSMIIREG ||
4952		    cmd == SIOCBRADDIF ||
4953		    cmd == SIOCBRDELIF ||
4954		    cmd == SIOCSHWTSTAMP ||
4955		    cmd == SIOCWANDEV) {
4956			err = -EOPNOTSUPP;
4957			if (ops->ndo_do_ioctl) {
4958				if (netif_device_present(dev))
4959					err = ops->ndo_do_ioctl(dev, ifr, cmd);
4960				else
4961					err = -ENODEV;
4962			}
4963		} else
4964			err = -EINVAL;
4965
4966	}
4967	return err;
4968}
4969
4970/*
4971 *	This function handles all "interface"-type I/O control requests. The actual
4972 *	'doing' part of this is dev_ifsioc above.
4973 */
4974
4975/**
4976 *	dev_ioctl	-	network device ioctl
4977 *	@net: the applicable net namespace
4978 *	@cmd: command to issue
4979 *	@arg: pointer to a struct ifreq in user space
4980 *
4981 *	Issue ioctl functions to devices. This is normally called by the
4982 *	user space syscall interfaces but can sometimes be useful for
4983 *	other purposes. The return value is the return from the syscall if
4984 *	positive or a negative errno code on error.
4985 */
4986
4987int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4988{
4989	struct ifreq ifr;
4990	int ret;
4991	char *colon;
4992
4993	/* One special case: SIOCGIFCONF takes ifconf argument
4994	   and requires shared lock, because it sleeps writing
4995	   to user space.
4996	 */
4997
4998	if (cmd == SIOCGIFCONF) {
4999		rtnl_lock();
5000		ret = dev_ifconf(net, (char __user *) arg);
5001		rtnl_unlock();
5002		return ret;
5003	}
5004	if (cmd == SIOCGIFNAME)
5005		return dev_ifname(net, (struct ifreq __user *)arg);
5006
5007	if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5008		return -EFAULT;
5009
5010	ifr.ifr_name[IFNAMSIZ-1] = 0;
5011
5012	colon = strchr(ifr.ifr_name, ':');
5013	if (colon)
5014		*colon = 0;
5015
5016	/*
5017	 *	See which interface the caller is talking about.
5018	 */
5019
5020	switch (cmd) {
5021	/*
5022	 *	These ioctl calls:
5023	 *	- can be done by all.
5024	 *	- atomic and do not require locking.
5025	 *	- return a value
5026	 */
5027	case SIOCGIFFLAGS:
5028	case SIOCGIFMETRIC:
5029	case SIOCGIFMTU:
5030	case SIOCGIFHWADDR:
5031	case SIOCGIFSLAVE:
5032	case SIOCGIFMAP:
5033	case SIOCGIFINDEX:
5034	case SIOCGIFTXQLEN:
5035		dev_load(net, ifr.ifr_name);
5036		rcu_read_lock();
5037		ret = dev_ifsioc_locked(net, &ifr, cmd);
5038		rcu_read_unlock();
5039		if (!ret) {
5040			if (colon)
5041				*colon = ':';
5042			if (copy_to_user(arg, &ifr,
5043					 sizeof(struct ifreq)))
5044				ret = -EFAULT;
5045		}
5046		return ret;
5047
5048	case SIOCETHTOOL:
5049		dev_load(net, ifr.ifr_name);
5050		rtnl_lock();
5051		ret = dev_ethtool(net, &ifr);
5052		rtnl_unlock();
5053		if (!ret) {
5054			if (colon)
5055				*colon = ':';
5056			if (copy_to_user(arg, &ifr,
5057					 sizeof(struct ifreq)))
5058				ret = -EFAULT;
5059		}
5060		return ret;
5061
5062	/*
5063	 *	These ioctl calls:
5064	 *	- require superuser power.
5065	 *	- require strict serialization.
5066	 *	- return a value
5067	 */
5068	case SIOCGMIIPHY:
5069	case SIOCGMIIREG:
5070	case SIOCSIFNAME:
5071		if (!capable(CAP_NET_ADMIN))
5072			return -EPERM;
5073		dev_load(net, ifr.ifr_name);
5074		rtnl_lock();
5075		ret = dev_ifsioc(net, &ifr, cmd);
5076		rtnl_unlock();
5077		if (!ret) {
5078			if (colon)
5079				*colon = ':';
5080			if (copy_to_user(arg, &ifr,
5081					 sizeof(struct ifreq)))
5082				ret = -EFAULT;
5083		}
5084		return ret;
5085
5086	/*
5087	 *	These ioctl calls:
5088	 *	- require superuser power.
5089	 *	- require strict serialization.
5090	 *	- do not return a value
5091	 */
5092	case SIOCSIFFLAGS:
5093	case SIOCSIFMETRIC:
5094	case SIOCSIFMTU:
5095	case SIOCSIFMAP:
5096	case SIOCSIFHWADDR:
5097	case SIOCSIFSLAVE:
5098	case SIOCADDMULTI:
5099	case SIOCDELMULTI:
5100	case SIOCSIFHWBROADCAST:
5101	case SIOCSIFTXQLEN:
5102	case SIOCSMIIREG:
5103	case SIOCBONDENSLAVE:
5104	case SIOCBONDRELEASE:
5105	case SIOCBONDSETHWADDR:
5106	case SIOCBONDCHANGEACTIVE:
5107	case SIOCBRADDIF:
5108	case SIOCBRDELIF:
5109	case SIOCSHWTSTAMP:
5110		if (!capable(CAP_NET_ADMIN))
5111			return -EPERM;
5112		/* fall through */
5113	case SIOCBONDSLAVEINFOQUERY:
5114	case SIOCBONDINFOQUERY:
5115		dev_load(net, ifr.ifr_name);
5116		rtnl_lock();
5117		ret = dev_ifsioc(net, &ifr, cmd);
5118		rtnl_unlock();
5119		return ret;
5120
5121	case SIOCGIFMEM:
5122		/* Get the per device memory space. We can add this but
5123		 * currently do not support it */
5124	case SIOCSIFMEM:
5125		/* Set the per device memory buffer space.
5126		 * Not applicable in our case */
5127	case SIOCSIFLINK:
5128		return -ENOTTY;
5129
5130	/*
5131	 *	Unknown or private ioctl.
5132	 */
5133	default:
5134		if (cmd == SIOCWANDEV ||
5135		    (cmd >= SIOCDEVPRIVATE &&
5136		     cmd <= SIOCDEVPRIVATE + 15)) {
5137			dev_load(net, ifr.ifr_name);
5138			rtnl_lock();
5139			ret = dev_ifsioc(net, &ifr, cmd);
5140			rtnl_unlock();
5141			if (!ret && copy_to_user(arg, &ifr,
5142						 sizeof(struct ifreq)))
5143				ret = -EFAULT;
5144			return ret;
5145		}
5146		/* Take care of Wireless Extensions */
5147		if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5148			return wext_handle_ioctl(net, &ifr, cmd, arg);
5149		return -ENOTTY;
5150	}
5151}
5152
5153
5154/**
5155 *	dev_new_index	-	allocate an ifindex
5156 *	@net: the applicable net namespace
5157 *
5158 *	Returns a suitable unique value for a new device interface
5159 *	number.  The caller must hold the rtnl semaphore or the
5160 *	dev_base_lock to be sure it remains unique.
5161 */
5162static int dev_new_index(struct net *net)
5163{
5164	static int ifindex;
5165	for (;;) {
5166		if (++ifindex <= 0)
5167			ifindex = 1;
5168		if (!__dev_get_by_index(net, ifindex))
5169			return ifindex;
5170	}
5171}
5172
5173/* Delayed registration/unregisteration */
5174static LIST_HEAD(net_todo_list);
5175
5176static void net_set_todo(struct net_device *dev)
5177{
5178	list_add_tail(&dev->todo_list, &net_todo_list);
5179}
5180
5181static void rollback_registered_many(struct list_head *head)
5182{
5183	struct net_device *dev, *tmp;
5184
5185	BUG_ON(dev_boot_phase);
5186	ASSERT_RTNL();
5187
5188	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5189		/* Some devices call without registering
5190		 * for initialization unwind. Remove those
5191		 * devices and proceed with the remaining.
5192		 */
5193		if (dev->reg_state == NETREG_UNINITIALIZED) {
5194			pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5195				 dev->name, dev);
5196
5197			WARN_ON(1);
5198			list_del(&dev->unreg_list);
5199			continue;
5200		}
5201		dev->dismantle = true;
5202		BUG_ON(dev->reg_state != NETREG_REGISTERED);
5203	}
5204
5205	/* If device is running, close it first. */
5206	dev_close_many(head);
5207
5208	list_for_each_entry(dev, head, unreg_list) {
5209		/* And unlink it from device chain. */
5210		unlist_netdevice(dev);
5211
5212		dev->reg_state = NETREG_UNREGISTERING;
5213	}
5214
5215	synchronize_net();
5216
5217	list_for_each_entry(dev, head, unreg_list) {
5218		/* Shutdown queueing discipline. */
5219		dev_shutdown(dev);
5220
5221
5222		/* Notify protocols, that we are about to destroy
5223		   this device. They should clean all the things.
5224		*/
5225		call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5226
5227		if (!dev->rtnl_link_ops ||
5228		    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5229			rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5230
5231		/*
5232		 *	Flush the unicast and multicast chains
5233		 */
5234		dev_uc_flush(dev);
5235		dev_mc_flush(dev);
5236
5237		if (dev->netdev_ops->ndo_uninit)
5238			dev->netdev_ops->ndo_uninit(dev);
5239
5240		/* Notifier chain MUST detach us from master device. */
5241		WARN_ON(dev->master);
5242
5243		/* Remove entries from kobject tree */
5244		netdev_unregister_kobject(dev);
5245	}
5246
5247	/* Process any work delayed until the end of the batch */
5248	dev = list_first_entry(head, struct net_device, unreg_list);
5249	call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5250
5251	synchronize_net();
5252
5253	list_for_each_entry(dev, head, unreg_list)
5254		dev_put(dev);
5255}
5256
5257static void rollback_registered(struct net_device *dev)
5258{
5259	LIST_HEAD(single);
5260
5261	list_add(&dev->unreg_list, &single);
5262	rollback_registered_many(&single);
5263	list_del(&single);
5264}
5265
5266static netdev_features_t netdev_fix_features(struct net_device *dev,
5267	netdev_features_t features)
5268{
5269	/* Fix illegal checksum combinations */
5270	if ((features & NETIF_F_HW_CSUM) &&
5271	    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5272		netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5273		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5274	}
5275
 
 
 
 
 
 
5276	/* Fix illegal SG+CSUM combinations. */
5277	if ((features & NETIF_F_SG) &&
5278	    !(features & NETIF_F_ALL_CSUM)) {
5279		netdev_dbg(dev,
5280			"Dropping NETIF_F_SG since no checksum feature.\n");
5281		features &= ~NETIF_F_SG;
5282	}
5283
5284	/* TSO requires that SG is present as well. */
5285	if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5286		netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5287		features &= ~NETIF_F_ALL_TSO;
5288	}
5289
5290	/* TSO ECN requires that TSO is present as well. */
5291	if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5292		features &= ~NETIF_F_TSO_ECN;
5293
5294	/* Software GSO depends on SG. */
5295	if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5296		netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5297		features &= ~NETIF_F_GSO;
5298	}
5299
5300	/* UFO needs SG and checksumming */
5301	if (features & NETIF_F_UFO) {
5302		/* maybe split UFO into V4 and V6? */
5303		if (!((features & NETIF_F_GEN_CSUM) ||
5304		    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5305			    == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5306			netdev_dbg(dev,
5307				"Dropping NETIF_F_UFO since no checksum offload features.\n");
5308			features &= ~NETIF_F_UFO;
5309		}
5310
5311		if (!(features & NETIF_F_SG)) {
5312			netdev_dbg(dev,
5313				"Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5314			features &= ~NETIF_F_UFO;
5315		}
5316	}
5317
5318	return features;
5319}
5320
5321int __netdev_update_features(struct net_device *dev)
5322{
5323	netdev_features_t features;
5324	int err = 0;
5325
5326	ASSERT_RTNL();
5327
5328	features = netdev_get_wanted_features(dev);
5329
5330	if (dev->netdev_ops->ndo_fix_features)
5331		features = dev->netdev_ops->ndo_fix_features(dev, features);
5332
5333	/* driver might be less strict about feature dependencies */
5334	features = netdev_fix_features(dev, features);
5335
5336	if (dev->features == features)
5337		return 0;
5338
5339	netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5340		&dev->features, &features);
5341
5342	if (dev->netdev_ops->ndo_set_features)
5343		err = dev->netdev_ops->ndo_set_features(dev, features);
5344
5345	if (unlikely(err < 0)) {
5346		netdev_err(dev,
5347			"set_features() failed (%d); wanted %pNF, left %pNF\n",
5348			err, &features, &dev->features);
5349		return -1;
5350	}
5351
5352	if (!err)
5353		dev->features = features;
5354
5355	return 1;
5356}
5357
5358/**
5359 *	netdev_update_features - recalculate device features
5360 *	@dev: the device to check
5361 *
5362 *	Recalculate dev->features set and send notifications if it
5363 *	has changed. Should be called after driver or hardware dependent
5364 *	conditions might have changed that influence the features.
5365 */
5366void netdev_update_features(struct net_device *dev)
5367{
5368	if (__netdev_update_features(dev))
5369		netdev_features_change(dev);
5370}
5371EXPORT_SYMBOL(netdev_update_features);
5372
5373/**
5374 *	netdev_change_features - recalculate device features
5375 *	@dev: the device to check
5376 *
5377 *	Recalculate dev->features set and send notifications even
5378 *	if they have not changed. Should be called instead of
5379 *	netdev_update_features() if also dev->vlan_features might
5380 *	have changed to allow the changes to be propagated to stacked
5381 *	VLAN devices.
5382 */
5383void netdev_change_features(struct net_device *dev)
5384{
5385	__netdev_update_features(dev);
5386	netdev_features_change(dev);
5387}
5388EXPORT_SYMBOL(netdev_change_features);
5389
5390/**
5391 *	netif_stacked_transfer_operstate -	transfer operstate
5392 *	@rootdev: the root or lower level device to transfer state from
5393 *	@dev: the device to transfer operstate to
5394 *
5395 *	Transfer operational state from root to device. This is normally
5396 *	called when a stacking relationship exists between the root
5397 *	device and the device(a leaf device).
5398 */
5399void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5400					struct net_device *dev)
5401{
5402	if (rootdev->operstate == IF_OPER_DORMANT)
5403		netif_dormant_on(dev);
5404	else
5405		netif_dormant_off(dev);
5406
5407	if (netif_carrier_ok(rootdev)) {
5408		if (!netif_carrier_ok(dev))
5409			netif_carrier_on(dev);
5410	} else {
5411		if (netif_carrier_ok(dev))
5412			netif_carrier_off(dev);
5413	}
5414}
5415EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5416
5417#ifdef CONFIG_RPS
5418static int netif_alloc_rx_queues(struct net_device *dev)
5419{
5420	unsigned int i, count = dev->num_rx_queues;
5421	struct netdev_rx_queue *rx;
5422
5423	BUG_ON(count < 1);
5424
5425	rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5426	if (!rx) {
5427		pr_err("netdev: Unable to allocate %u rx queues\n", count);
5428		return -ENOMEM;
5429	}
5430	dev->_rx = rx;
5431
5432	for (i = 0; i < count; i++)
5433		rx[i].dev = dev;
5434	return 0;
5435}
5436#endif
5437
5438static void netdev_init_one_queue(struct net_device *dev,
5439				  struct netdev_queue *queue, void *_unused)
5440{
5441	/* Initialize queue lock */
5442	spin_lock_init(&queue->_xmit_lock);
5443	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5444	queue->xmit_lock_owner = -1;
5445	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5446	queue->dev = dev;
5447#ifdef CONFIG_BQL
5448	dql_init(&queue->dql, HZ);
5449#endif
5450}
5451
5452static int netif_alloc_netdev_queues(struct net_device *dev)
5453{
5454	unsigned int count = dev->num_tx_queues;
5455	struct netdev_queue *tx;
5456
5457	BUG_ON(count < 1);
5458
5459	tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5460	if (!tx) {
5461		pr_err("netdev: Unable to allocate %u tx queues\n", count);
 
5462		return -ENOMEM;
5463	}
5464	dev->_tx = tx;
5465
5466	netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5467	spin_lock_init(&dev->tx_global_lock);
5468
5469	return 0;
5470}
5471
5472/**
5473 *	register_netdevice	- register a network device
5474 *	@dev: device to register
5475 *
5476 *	Take a completed network device structure and add it to the kernel
5477 *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5478 *	chain. 0 is returned on success. A negative errno code is returned
5479 *	on a failure to set up the device, or if the name is a duplicate.
5480 *
5481 *	Callers must hold the rtnl semaphore. You may want
5482 *	register_netdev() instead of this.
5483 *
5484 *	BUGS:
5485 *	The locking appears insufficient to guarantee two parallel registers
5486 *	will not get the same name.
5487 */
5488
5489int register_netdevice(struct net_device *dev)
5490{
5491	int ret;
5492	struct net *net = dev_net(dev);
5493
5494	BUG_ON(dev_boot_phase);
5495	ASSERT_RTNL();
5496
5497	might_sleep();
5498
5499	/* When net_device's are persistent, this will be fatal. */
5500	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5501	BUG_ON(!net);
5502
5503	spin_lock_init(&dev->addr_list_lock);
5504	netdev_set_addr_lockdep_class(dev);
5505
5506	dev->iflink = -1;
5507
5508	ret = dev_get_valid_name(dev, dev->name);
5509	if (ret < 0)
5510		goto out;
5511
5512	/* Init, if this function is available */
5513	if (dev->netdev_ops->ndo_init) {
5514		ret = dev->netdev_ops->ndo_init(dev);
5515		if (ret) {
5516			if (ret > 0)
5517				ret = -EIO;
5518			goto out;
5519		}
5520	}
5521
5522	dev->ifindex = dev_new_index(net);
5523	if (dev->iflink == -1)
5524		dev->iflink = dev->ifindex;
5525
5526	/* Transfer changeable features to wanted_features and enable
5527	 * software offloads (GSO and GRO).
5528	 */
5529	dev->hw_features |= NETIF_F_SOFT_FEATURES;
5530	dev->features |= NETIF_F_SOFT_FEATURES;
5531	dev->wanted_features = dev->features & dev->hw_features;
5532
5533	/* Turn on no cache copy if HW is doing checksum */
5534	if (!(dev->flags & IFF_LOOPBACK)) {
5535		dev->hw_features |= NETIF_F_NOCACHE_COPY;
5536		if (dev->features & NETIF_F_ALL_CSUM) {
5537			dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5538			dev->features |= NETIF_F_NOCACHE_COPY;
5539		}
5540	}
5541
5542	/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5543	 */
5544	dev->vlan_features |= NETIF_F_HIGHDMA;
5545
5546	ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5547	ret = notifier_to_errno(ret);
5548	if (ret)
5549		goto err_uninit;
5550
5551	ret = netdev_register_kobject(dev);
5552	if (ret)
5553		goto err_uninit;
5554	dev->reg_state = NETREG_REGISTERED;
5555
5556	__netdev_update_features(dev);
5557
5558	/*
5559	 *	Default initial state at registry is that the
5560	 *	device is present.
5561	 */
5562
5563	set_bit(__LINK_STATE_PRESENT, &dev->state);
5564
5565	dev_init_scheduler(dev);
5566	dev_hold(dev);
5567	list_netdevice(dev);
5568	add_device_randomness(dev->dev_addr, dev->addr_len);
5569
5570	/* Notify protocols, that a new device appeared. */
5571	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5572	ret = notifier_to_errno(ret);
5573	if (ret) {
5574		rollback_registered(dev);
5575		dev->reg_state = NETREG_UNREGISTERED;
5576	}
5577	/*
5578	 *	Prevent userspace races by waiting until the network
5579	 *	device is fully setup before sending notifications.
5580	 */
5581	if (!dev->rtnl_link_ops ||
5582	    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5583		rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5584
5585out:
5586	return ret;
5587
5588err_uninit:
5589	if (dev->netdev_ops->ndo_uninit)
5590		dev->netdev_ops->ndo_uninit(dev);
5591	goto out;
5592}
5593EXPORT_SYMBOL(register_netdevice);
5594
5595/**
5596 *	init_dummy_netdev	- init a dummy network device for NAPI
5597 *	@dev: device to init
5598 *
5599 *	This takes a network device structure and initialize the minimum
5600 *	amount of fields so it can be used to schedule NAPI polls without
5601 *	registering a full blown interface. This is to be used by drivers
5602 *	that need to tie several hardware interfaces to a single NAPI
5603 *	poll scheduler due to HW limitations.
5604 */
5605int init_dummy_netdev(struct net_device *dev)
5606{
5607	/* Clear everything. Note we don't initialize spinlocks
5608	 * are they aren't supposed to be taken by any of the
5609	 * NAPI code and this dummy netdev is supposed to be
5610	 * only ever used for NAPI polls
5611	 */
5612	memset(dev, 0, sizeof(struct net_device));
5613
5614	/* make sure we BUG if trying to hit standard
5615	 * register/unregister code path
5616	 */
5617	dev->reg_state = NETREG_DUMMY;
5618
5619	/* NAPI wants this */
5620	INIT_LIST_HEAD(&dev->napi_list);
5621
5622	/* a dummy interface is started by default */
5623	set_bit(__LINK_STATE_PRESENT, &dev->state);
5624	set_bit(__LINK_STATE_START, &dev->state);
5625
5626	/* Note : We dont allocate pcpu_refcnt for dummy devices,
5627	 * because users of this 'device' dont need to change
5628	 * its refcount.
5629	 */
5630
5631	return 0;
5632}
5633EXPORT_SYMBOL_GPL(init_dummy_netdev);
5634
5635
5636/**
5637 *	register_netdev	- register a network device
5638 *	@dev: device to register
5639 *
5640 *	Take a completed network device structure and add it to the kernel
5641 *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5642 *	chain. 0 is returned on success. A negative errno code is returned
5643 *	on a failure to set up the device, or if the name is a duplicate.
5644 *
5645 *	This is a wrapper around register_netdevice that takes the rtnl semaphore
5646 *	and expands the device name if you passed a format string to
5647 *	alloc_netdev.
5648 */
5649int register_netdev(struct net_device *dev)
5650{
5651	int err;
5652
5653	rtnl_lock();
5654	err = register_netdevice(dev);
5655	rtnl_unlock();
5656	return err;
5657}
5658EXPORT_SYMBOL(register_netdev);
5659
5660int netdev_refcnt_read(const struct net_device *dev)
5661{
5662	int i, refcnt = 0;
5663
5664	for_each_possible_cpu(i)
5665		refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5666	return refcnt;
5667}
5668EXPORT_SYMBOL(netdev_refcnt_read);
5669
5670/*
5671 * netdev_wait_allrefs - wait until all references are gone.
5672 *
5673 * This is called when unregistering network devices.
5674 *
5675 * Any protocol or device that holds a reference should register
5676 * for netdevice notification, and cleanup and put back the
5677 * reference if they receive an UNREGISTER event.
5678 * We can get stuck here if buggy protocols don't correctly
5679 * call dev_put.
5680 */
5681static void netdev_wait_allrefs(struct net_device *dev)
5682{
5683	unsigned long rebroadcast_time, warning_time;
5684	int refcnt;
5685
5686	linkwatch_forget_dev(dev);
5687
5688	rebroadcast_time = warning_time = jiffies;
5689	refcnt = netdev_refcnt_read(dev);
5690
5691	while (refcnt != 0) {
5692		if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5693			rtnl_lock();
5694
5695			/* Rebroadcast unregister notification */
5696			call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5697			/* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5698			 * should have already handle it the first time */
5699
5700			if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5701				     &dev->state)) {
5702				/* We must not have linkwatch events
5703				 * pending on unregister. If this
5704				 * happens, we simply run the queue
5705				 * unscheduled, resulting in a noop
5706				 * for this device.
5707				 */
5708				linkwatch_run_queue();
5709			}
5710
5711			__rtnl_unlock();
5712
5713			rebroadcast_time = jiffies;
5714		}
5715
5716		msleep(250);
5717
5718		refcnt = netdev_refcnt_read(dev);
5719
5720		if (time_after(jiffies, warning_time + 10 * HZ)) {
5721			pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5722				 dev->name, refcnt);
 
 
5723			warning_time = jiffies;
5724		}
5725	}
5726}
5727
5728/* The sequence is:
5729 *
5730 *	rtnl_lock();
5731 *	...
5732 *	register_netdevice(x1);
5733 *	register_netdevice(x2);
5734 *	...
5735 *	unregister_netdevice(y1);
5736 *	unregister_netdevice(y2);
5737 *      ...
5738 *	rtnl_unlock();
5739 *	free_netdev(y1);
5740 *	free_netdev(y2);
5741 *
5742 * We are invoked by rtnl_unlock().
5743 * This allows us to deal with problems:
5744 * 1) We can delete sysfs objects which invoke hotplug
5745 *    without deadlocking with linkwatch via keventd.
5746 * 2) Since we run with the RTNL semaphore not held, we can sleep
5747 *    safely in order to wait for the netdev refcnt to drop to zero.
5748 *
5749 * We must not return until all unregister events added during
5750 * the interval the lock was held have been completed.
5751 */
5752void netdev_run_todo(void)
5753{
5754	struct list_head list;
5755
5756	/* Snapshot list, allow later requests */
5757	list_replace_init(&net_todo_list, &list);
5758
5759	__rtnl_unlock();
5760
5761	/* Wait for rcu callbacks to finish before attempting to drain
5762	 * the device list.  This usually avoids a 250ms wait.
5763	 */
5764	if (!list_empty(&list))
5765		rcu_barrier();
5766
5767	while (!list_empty(&list)) {
5768		struct net_device *dev
5769			= list_first_entry(&list, struct net_device, todo_list);
5770		list_del(&dev->todo_list);
5771
5772		if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5773			pr_err("network todo '%s' but state %d\n",
5774			       dev->name, dev->reg_state);
5775			dump_stack();
5776			continue;
5777		}
5778
5779		dev->reg_state = NETREG_UNREGISTERED;
5780
5781		on_each_cpu(flush_backlog, dev, 1);
5782
5783		netdev_wait_allrefs(dev);
5784
5785		/* paranoia */
5786		BUG_ON(netdev_refcnt_read(dev));
5787		WARN_ON(rcu_access_pointer(dev->ip_ptr));
5788		WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5789		WARN_ON(dev->dn_ptr);
5790
5791		if (dev->destructor)
5792			dev->destructor(dev);
5793
5794		/* Free network device */
5795		kobject_put(&dev->dev.kobj);
5796	}
5797}
5798
5799/* Convert net_device_stats to rtnl_link_stats64.  They have the same
5800 * fields in the same order, with only the type differing.
5801 */
5802void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5803			     const struct net_device_stats *netdev_stats)
5804{
5805#if BITS_PER_LONG == 64
5806	BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5807	memcpy(stats64, netdev_stats, sizeof(*stats64));
5808#else
5809	size_t i, n = sizeof(*stats64) / sizeof(u64);
5810	const unsigned long *src = (const unsigned long *)netdev_stats;
5811	u64 *dst = (u64 *)stats64;
5812
5813	BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5814		     sizeof(*stats64) / sizeof(u64));
5815	for (i = 0; i < n; i++)
5816		dst[i] = src[i];
5817#endif
5818}
5819EXPORT_SYMBOL(netdev_stats_to_stats64);
5820
5821/**
5822 *	dev_get_stats	- get network device statistics
5823 *	@dev: device to get statistics from
5824 *	@storage: place to store stats
5825 *
5826 *	Get network statistics from device. Return @storage.
5827 *	The device driver may provide its own method by setting
5828 *	dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5829 *	otherwise the internal statistics structure is used.
5830 */
5831struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5832					struct rtnl_link_stats64 *storage)
5833{
5834	const struct net_device_ops *ops = dev->netdev_ops;
5835
5836	if (ops->ndo_get_stats64) {
5837		memset(storage, 0, sizeof(*storage));
5838		ops->ndo_get_stats64(dev, storage);
5839	} else if (ops->ndo_get_stats) {
5840		netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5841	} else {
5842		netdev_stats_to_stats64(storage, &dev->stats);
5843	}
5844	storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5845	return storage;
5846}
5847EXPORT_SYMBOL(dev_get_stats);
5848
5849struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5850{
5851	struct netdev_queue *queue = dev_ingress_queue(dev);
5852
5853#ifdef CONFIG_NET_CLS_ACT
5854	if (queue)
5855		return queue;
5856	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5857	if (!queue)
5858		return NULL;
5859	netdev_init_one_queue(dev, queue, NULL);
5860	queue->qdisc = &noop_qdisc;
5861	queue->qdisc_sleeping = &noop_qdisc;
5862	rcu_assign_pointer(dev->ingress_queue, queue);
5863#endif
5864	return queue;
5865}
5866
5867/**
5868 *	alloc_netdev_mqs - allocate network device
5869 *	@sizeof_priv:	size of private data to allocate space for
5870 *	@name:		device name format string
5871 *	@setup:		callback to initialize device
5872 *	@txqs:		the number of TX subqueues to allocate
5873 *	@rxqs:		the number of RX subqueues to allocate
5874 *
5875 *	Allocates a struct net_device with private data area for driver use
5876 *	and performs basic initialization.  Also allocates subquue structs
5877 *	for each queue on the device.
5878 */
5879struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5880		void (*setup)(struct net_device *),
5881		unsigned int txqs, unsigned int rxqs)
5882{
5883	struct net_device *dev;
5884	size_t alloc_size;
5885	struct net_device *p;
5886
5887	BUG_ON(strlen(name) >= sizeof(dev->name));
5888
5889	if (txqs < 1) {
5890		pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
 
5891		return NULL;
5892	}
5893
5894#ifdef CONFIG_RPS
5895	if (rxqs < 1) {
5896		pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
 
5897		return NULL;
5898	}
5899#endif
5900
5901	alloc_size = sizeof(struct net_device);
5902	if (sizeof_priv) {
5903		/* ensure 32-byte alignment of private area */
5904		alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5905		alloc_size += sizeof_priv;
5906	}
5907	/* ensure 32-byte alignment of whole construct */
5908	alloc_size += NETDEV_ALIGN - 1;
5909
5910	p = kzalloc(alloc_size, GFP_KERNEL);
5911	if (!p) {
5912		pr_err("alloc_netdev: Unable to allocate device\n");
5913		return NULL;
5914	}
5915
5916	dev = PTR_ALIGN(p, NETDEV_ALIGN);
5917	dev->padded = (char *)dev - (char *)p;
5918
5919	dev->pcpu_refcnt = alloc_percpu(int);
5920	if (!dev->pcpu_refcnt)
5921		goto free_p;
5922
5923	if (dev_addr_init(dev))
5924		goto free_pcpu;
5925
5926	dev_mc_init(dev);
5927	dev_uc_init(dev);
5928
5929	dev_net_set(dev, &init_net);
5930
5931	dev->gso_max_size = GSO_MAX_SIZE;
5932	dev->gso_max_segs = GSO_MAX_SEGS;
5933
5934	INIT_LIST_HEAD(&dev->napi_list);
5935	INIT_LIST_HEAD(&dev->unreg_list);
5936	INIT_LIST_HEAD(&dev->link_watch_list);
5937	dev->priv_flags = IFF_XMIT_DST_RELEASE;
5938	setup(dev);
5939
5940	dev->num_tx_queues = txqs;
5941	dev->real_num_tx_queues = txqs;
5942	if (netif_alloc_netdev_queues(dev))
5943		goto free_all;
5944
5945#ifdef CONFIG_RPS
5946	dev->num_rx_queues = rxqs;
5947	dev->real_num_rx_queues = rxqs;
5948	if (netif_alloc_rx_queues(dev))
5949		goto free_all;
5950#endif
5951
5952	strcpy(dev->name, name);
5953	dev->group = INIT_NETDEV_GROUP;
5954	return dev;
5955
5956free_all:
5957	free_netdev(dev);
5958	return NULL;
5959
5960free_pcpu:
5961	free_percpu(dev->pcpu_refcnt);
5962	kfree(dev->_tx);
5963#ifdef CONFIG_RPS
5964	kfree(dev->_rx);
5965#endif
5966
5967free_p:
5968	kfree(p);
5969	return NULL;
5970}
5971EXPORT_SYMBOL(alloc_netdev_mqs);
5972
5973/**
5974 *	free_netdev - free network device
5975 *	@dev: device
5976 *
5977 *	This function does the last stage of destroying an allocated device
5978 * 	interface. The reference to the device object is released.
5979 *	If this is the last reference then it will be freed.
5980 */
5981void free_netdev(struct net_device *dev)
5982{
5983	struct napi_struct *p, *n;
5984
5985	release_net(dev_net(dev));
5986
5987	kfree(dev->_tx);
5988#ifdef CONFIG_RPS
5989	kfree(dev->_rx);
5990#endif
5991
5992	kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5993
5994	/* Flush device addresses */
5995	dev_addr_flush(dev);
5996
5997	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5998		netif_napi_del(p);
5999
6000	free_percpu(dev->pcpu_refcnt);
6001	dev->pcpu_refcnt = NULL;
6002
6003	/*  Compatibility with error handling in drivers */
6004	if (dev->reg_state == NETREG_UNINITIALIZED) {
6005		kfree((char *)dev - dev->padded);
6006		return;
6007	}
6008
6009	BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6010	dev->reg_state = NETREG_RELEASED;
6011
6012	/* will free via device release */
6013	put_device(&dev->dev);
6014}
6015EXPORT_SYMBOL(free_netdev);
6016
6017/**
6018 *	synchronize_net -  Synchronize with packet receive processing
6019 *
6020 *	Wait for packets currently being received to be done.
6021 *	Does not block later packets from starting.
6022 */
6023void synchronize_net(void)
6024{
6025	might_sleep();
6026	if (rtnl_is_locked())
6027		synchronize_rcu_expedited();
6028	else
6029		synchronize_rcu();
6030}
6031EXPORT_SYMBOL(synchronize_net);
6032
6033/**
6034 *	unregister_netdevice_queue - remove device from the kernel
6035 *	@dev: device
6036 *	@head: list
6037 *
6038 *	This function shuts down a device interface and removes it
6039 *	from the kernel tables.
6040 *	If head not NULL, device is queued to be unregistered later.
6041 *
6042 *	Callers must hold the rtnl semaphore.  You may want
6043 *	unregister_netdev() instead of this.
6044 */
6045
6046void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6047{
6048	ASSERT_RTNL();
6049
6050	if (head) {
6051		list_move_tail(&dev->unreg_list, head);
6052	} else {
6053		rollback_registered(dev);
6054		/* Finish processing unregister after unlock */
6055		net_set_todo(dev);
6056	}
6057}
6058EXPORT_SYMBOL(unregister_netdevice_queue);
6059
6060/**
6061 *	unregister_netdevice_many - unregister many devices
6062 *	@head: list of devices
6063 */
6064void unregister_netdevice_many(struct list_head *head)
6065{
6066	struct net_device *dev;
6067
6068	if (!list_empty(head)) {
6069		rollback_registered_many(head);
6070		list_for_each_entry(dev, head, unreg_list)
6071			net_set_todo(dev);
6072	}
6073}
6074EXPORT_SYMBOL(unregister_netdevice_many);
6075
6076/**
6077 *	unregister_netdev - remove device from the kernel
6078 *	@dev: device
6079 *
6080 *	This function shuts down a device interface and removes it
6081 *	from the kernel tables.
6082 *
6083 *	This is just a wrapper for unregister_netdevice that takes
6084 *	the rtnl semaphore.  In general you want to use this and not
6085 *	unregister_netdevice.
6086 */
6087void unregister_netdev(struct net_device *dev)
6088{
6089	rtnl_lock();
6090	unregister_netdevice(dev);
6091	rtnl_unlock();
6092}
6093EXPORT_SYMBOL(unregister_netdev);
6094
6095/**
6096 *	dev_change_net_namespace - move device to different nethost namespace
6097 *	@dev: device
6098 *	@net: network namespace
6099 *	@pat: If not NULL name pattern to try if the current device name
6100 *	      is already taken in the destination network namespace.
6101 *
6102 *	This function shuts down a device interface and moves it
6103 *	to a new network namespace. On success 0 is returned, on
6104 *	a failure a netagive errno code is returned.
6105 *
6106 *	Callers must hold the rtnl semaphore.
6107 */
6108
6109int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6110{
6111	int err;
6112
6113	ASSERT_RTNL();
6114
6115	/* Don't allow namespace local devices to be moved. */
6116	err = -EINVAL;
6117	if (dev->features & NETIF_F_NETNS_LOCAL)
6118		goto out;
6119
6120	/* Ensure the device has been registrered */
6121	err = -EINVAL;
6122	if (dev->reg_state != NETREG_REGISTERED)
6123		goto out;
6124
6125	/* Get out if there is nothing todo */
6126	err = 0;
6127	if (net_eq(dev_net(dev), net))
6128		goto out;
6129
6130	/* Pick the destination device name, and ensure
6131	 * we can use it in the destination network namespace.
6132	 */
6133	err = -EEXIST;
6134	if (__dev_get_by_name(net, dev->name)) {
6135		/* We get here if we can't use the current device name */
6136		if (!pat)
6137			goto out;
6138		if (dev_get_valid_name(dev, pat) < 0)
6139			goto out;
6140	}
6141
6142	/*
6143	 * And now a mini version of register_netdevice unregister_netdevice.
6144	 */
6145
6146	/* If device is running close it first. */
6147	dev_close(dev);
6148
6149	/* And unlink it from device chain */
6150	err = -ENODEV;
6151	unlist_netdevice(dev);
6152
6153	synchronize_net();
6154
6155	/* Shutdown queueing discipline. */
6156	dev_shutdown(dev);
6157
6158	/* Notify protocols, that we are about to destroy
6159	   this device. They should clean all the things.
6160
6161	   Note that dev->reg_state stays at NETREG_REGISTERED.
6162	   This is wanted because this way 8021q and macvlan know
6163	   the device is just moving and can keep their slaves up.
6164	*/
6165	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6166	call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6167	rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6168
6169	/*
6170	 *	Flush the unicast and multicast chains
6171	 */
6172	dev_uc_flush(dev);
6173	dev_mc_flush(dev);
6174
6175	/* Actually switch the network namespace */
6176	dev_net_set(dev, net);
6177
6178	/* If there is an ifindex conflict assign a new one */
6179	if (__dev_get_by_index(net, dev->ifindex)) {
6180		int iflink = (dev->iflink == dev->ifindex);
6181		dev->ifindex = dev_new_index(net);
6182		if (iflink)
6183			dev->iflink = dev->ifindex;
6184	}
6185
6186	/* Fixup kobjects */
6187	err = device_rename(&dev->dev, dev->name);
6188	WARN_ON(err);
6189
6190	/* Add the device back in the hashes */
6191	list_netdevice(dev);
6192
6193	/* Notify protocols, that a new device appeared. */
6194	call_netdevice_notifiers(NETDEV_REGISTER, dev);
6195
6196	/*
6197	 *	Prevent userspace races by waiting until the network
6198	 *	device is fully setup before sending notifications.
6199	 */
6200	rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6201
6202	synchronize_net();
6203	err = 0;
6204out:
6205	return err;
6206}
6207EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6208
6209static int dev_cpu_callback(struct notifier_block *nfb,
6210			    unsigned long action,
6211			    void *ocpu)
6212{
6213	struct sk_buff **list_skb;
6214	struct sk_buff *skb;
6215	unsigned int cpu, oldcpu = (unsigned long)ocpu;
6216	struct softnet_data *sd, *oldsd;
6217
6218	if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6219		return NOTIFY_OK;
6220
6221	local_irq_disable();
6222	cpu = smp_processor_id();
6223	sd = &per_cpu(softnet_data, cpu);
6224	oldsd = &per_cpu(softnet_data, oldcpu);
6225
6226	/* Find end of our completion_queue. */
6227	list_skb = &sd->completion_queue;
6228	while (*list_skb)
6229		list_skb = &(*list_skb)->next;
6230	/* Append completion queue from offline CPU. */
6231	*list_skb = oldsd->completion_queue;
6232	oldsd->completion_queue = NULL;
6233
6234	/* Append output queue from offline CPU. */
6235	if (oldsd->output_queue) {
6236		*sd->output_queue_tailp = oldsd->output_queue;
6237		sd->output_queue_tailp = oldsd->output_queue_tailp;
6238		oldsd->output_queue = NULL;
6239		oldsd->output_queue_tailp = &oldsd->output_queue;
6240	}
6241	/* Append NAPI poll list from offline CPU. */
6242	if (!list_empty(&oldsd->poll_list)) {
6243		list_splice_init(&oldsd->poll_list, &sd->poll_list);
6244		raise_softirq_irqoff(NET_RX_SOFTIRQ);
6245	}
6246
6247	raise_softirq_irqoff(NET_TX_SOFTIRQ);
6248	local_irq_enable();
6249
6250	/* Process offline CPU's input_pkt_queue */
6251	while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6252		netif_rx(skb);
6253		input_queue_head_incr(oldsd);
6254	}
6255	while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6256		netif_rx(skb);
6257		input_queue_head_incr(oldsd);
6258	}
6259
6260	return NOTIFY_OK;
6261}
6262
6263
6264/**
6265 *	netdev_increment_features - increment feature set by one
6266 *	@all: current feature set
6267 *	@one: new feature set
6268 *	@mask: mask feature set
6269 *
6270 *	Computes a new feature set after adding a device with feature set
6271 *	@one to the master device with current feature set @all.  Will not
6272 *	enable anything that is off in @mask. Returns the new feature set.
6273 */
6274netdev_features_t netdev_increment_features(netdev_features_t all,
6275	netdev_features_t one, netdev_features_t mask)
6276{
6277	if (mask & NETIF_F_GEN_CSUM)
6278		mask |= NETIF_F_ALL_CSUM;
6279	mask |= NETIF_F_VLAN_CHALLENGED;
6280
6281	all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6282	all &= one | ~NETIF_F_ALL_FOR_ALL;
6283
 
 
 
 
6284	/* If one device supports hw checksumming, set for all. */
6285	if (all & NETIF_F_GEN_CSUM)
6286		all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6287
6288	return all;
6289}
6290EXPORT_SYMBOL(netdev_increment_features);
6291
6292static struct hlist_head *netdev_create_hash(void)
6293{
6294	int i;
6295	struct hlist_head *hash;
6296
6297	hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6298	if (hash != NULL)
6299		for (i = 0; i < NETDEV_HASHENTRIES; i++)
6300			INIT_HLIST_HEAD(&hash[i]);
6301
6302	return hash;
6303}
6304
6305/* Initialize per network namespace state */
6306static int __net_init netdev_init(struct net *net)
6307{
6308	if (net != &init_net)
6309		INIT_LIST_HEAD(&net->dev_base_head);
6310
6311	net->dev_name_head = netdev_create_hash();
6312	if (net->dev_name_head == NULL)
6313		goto err_name;
6314
6315	net->dev_index_head = netdev_create_hash();
6316	if (net->dev_index_head == NULL)
6317		goto err_idx;
6318
6319	return 0;
6320
6321err_idx:
6322	kfree(net->dev_name_head);
6323err_name:
6324	return -ENOMEM;
6325}
6326
6327/**
6328 *	netdev_drivername - network driver for the device
6329 *	@dev: network device
6330 *
6331 *	Determine network driver for device.
6332 */
6333const char *netdev_drivername(const struct net_device *dev)
6334{
6335	const struct device_driver *driver;
6336	const struct device *parent;
6337	const char *empty = "";
6338
6339	parent = dev->dev.parent;
6340	if (!parent)
6341		return empty;
6342
6343	driver = parent->driver;
6344	if (driver && driver->name)
6345		return driver->name;
6346	return empty;
6347}
6348
6349int __netdev_printk(const char *level, const struct net_device *dev,
6350			   struct va_format *vaf)
6351{
6352	int r;
6353
6354	if (dev && dev->dev.parent)
6355		r = dev_printk(level, dev->dev.parent, "%s: %pV",
6356			       netdev_name(dev), vaf);
6357	else if (dev)
6358		r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6359	else
6360		r = printk("%s(NULL net_device): %pV", level, vaf);
6361
6362	return r;
6363}
6364EXPORT_SYMBOL(__netdev_printk);
6365
6366int netdev_printk(const char *level, const struct net_device *dev,
6367		  const char *format, ...)
6368{
6369	struct va_format vaf;
6370	va_list args;
6371	int r;
6372
6373	va_start(args, format);
6374
6375	vaf.fmt = format;
6376	vaf.va = &args;
6377
6378	r = __netdev_printk(level, dev, &vaf);
6379	va_end(args);
6380
6381	return r;
6382}
6383EXPORT_SYMBOL(netdev_printk);
6384
6385#define define_netdev_printk_level(func, level)			\
6386int func(const struct net_device *dev, const char *fmt, ...)	\
6387{								\
6388	int r;							\
6389	struct va_format vaf;					\
6390	va_list args;						\
6391								\
6392	va_start(args, fmt);					\
6393								\
6394	vaf.fmt = fmt;						\
6395	vaf.va = &args;						\
6396								\
6397	r = __netdev_printk(level, dev, &vaf);			\
6398	va_end(args);						\
6399								\
6400	return r;						\
6401}								\
6402EXPORT_SYMBOL(func);
6403
6404define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6405define_netdev_printk_level(netdev_alert, KERN_ALERT);
6406define_netdev_printk_level(netdev_crit, KERN_CRIT);
6407define_netdev_printk_level(netdev_err, KERN_ERR);
6408define_netdev_printk_level(netdev_warn, KERN_WARNING);
6409define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6410define_netdev_printk_level(netdev_info, KERN_INFO);
6411
6412static void __net_exit netdev_exit(struct net *net)
6413{
6414	kfree(net->dev_name_head);
6415	kfree(net->dev_index_head);
6416}
6417
6418static struct pernet_operations __net_initdata netdev_net_ops = {
6419	.init = netdev_init,
6420	.exit = netdev_exit,
6421};
6422
6423static void __net_exit default_device_exit(struct net *net)
6424{
6425	struct net_device *dev, *aux;
6426	/*
6427	 * Push all migratable network devices back to the
6428	 * initial network namespace
6429	 */
6430	rtnl_lock();
6431	for_each_netdev_safe(net, dev, aux) {
6432		int err;
6433		char fb_name[IFNAMSIZ];
6434
6435		/* Ignore unmoveable devices (i.e. loopback) */
6436		if (dev->features & NETIF_F_NETNS_LOCAL)
6437			continue;
6438
6439		/* Leave virtual devices for the generic cleanup */
6440		if (dev->rtnl_link_ops)
6441			continue;
6442
6443		/* Push remaining network devices to init_net */
6444		snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6445		err = dev_change_net_namespace(dev, &init_net, fb_name);
6446		if (err) {
6447			pr_emerg("%s: failed to move %s to init_net: %d\n",
6448				 __func__, dev->name, err);
6449			BUG();
6450		}
6451	}
6452	rtnl_unlock();
6453}
6454
6455static void __net_exit default_device_exit_batch(struct list_head *net_list)
6456{
6457	/* At exit all network devices most be removed from a network
6458	 * namespace.  Do this in the reverse order of registration.
6459	 * Do this across as many network namespaces as possible to
6460	 * improve batching efficiency.
6461	 */
6462	struct net_device *dev;
6463	struct net *net;
6464	LIST_HEAD(dev_kill_list);
6465
6466	rtnl_lock();
6467	list_for_each_entry(net, net_list, exit_list) {
6468		for_each_netdev_reverse(net, dev) {
6469			if (dev->rtnl_link_ops)
6470				dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6471			else
6472				unregister_netdevice_queue(dev, &dev_kill_list);
6473		}
6474	}
6475	unregister_netdevice_many(&dev_kill_list);
6476	list_del(&dev_kill_list);
6477	rtnl_unlock();
6478}
6479
6480static struct pernet_operations __net_initdata default_device_ops = {
6481	.exit = default_device_exit,
6482	.exit_batch = default_device_exit_batch,
6483};
6484
6485/*
6486 *	Initialize the DEV module. At boot time this walks the device list and
6487 *	unhooks any devices that fail to initialise (normally hardware not
6488 *	present) and leaves us with a valid list of present and active devices.
6489 *
6490 */
6491
6492/*
6493 *       This is called single threaded during boot, so no need
6494 *       to take the rtnl semaphore.
6495 */
6496static int __init net_dev_init(void)
6497{
6498	int i, rc = -ENOMEM;
6499
6500	BUG_ON(!dev_boot_phase);
6501
6502	if (dev_proc_init())
6503		goto out;
6504
6505	if (netdev_kobject_init())
6506		goto out;
6507
6508	INIT_LIST_HEAD(&ptype_all);
6509	for (i = 0; i < PTYPE_HASH_SIZE; i++)
6510		INIT_LIST_HEAD(&ptype_base[i]);
6511
6512	if (register_pernet_subsys(&netdev_net_ops))
6513		goto out;
6514
6515	/*
6516	 *	Initialise the packet receive queues.
6517	 */
6518
6519	for_each_possible_cpu(i) {
6520		struct softnet_data *sd = &per_cpu(softnet_data, i);
6521
6522		memset(sd, 0, sizeof(*sd));
6523		skb_queue_head_init(&sd->input_pkt_queue);
6524		skb_queue_head_init(&sd->process_queue);
6525		sd->completion_queue = NULL;
6526		INIT_LIST_HEAD(&sd->poll_list);
6527		sd->output_queue = NULL;
6528		sd->output_queue_tailp = &sd->output_queue;
6529#ifdef CONFIG_RPS
6530		sd->csd.func = rps_trigger_softirq;
6531		sd->csd.info = sd;
6532		sd->csd.flags = 0;
6533		sd->cpu = i;
6534#endif
6535
6536		sd->backlog.poll = process_backlog;
6537		sd->backlog.weight = weight_p;
6538		sd->backlog.gro_list = NULL;
6539		sd->backlog.gro_count = 0;
6540	}
6541
6542	dev_boot_phase = 0;
6543
6544	/* The loopback device is special if any other network devices
6545	 * is present in a network namespace the loopback device must
6546	 * be present. Since we now dynamically allocate and free the
6547	 * loopback device ensure this invariant is maintained by
6548	 * keeping the loopback device as the first device on the
6549	 * list of network devices.  Ensuring the loopback devices
6550	 * is the first device that appears and the last network device
6551	 * that disappears.
6552	 */
6553	if (register_pernet_device(&loopback_net_ops))
6554		goto out;
6555
6556	if (register_pernet_device(&default_device_ops))
6557		goto out;
6558
6559	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6560	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6561
6562	hotcpu_notifier(dev_cpu_callback, 0);
6563	dst_init();
6564	dev_mcast_init();
6565	rc = 0;
6566out:
6567	return rc;
6568}
6569
6570subsys_initcall(net_dev_init);
6571
6572static int __init initialize_hashrnd(void)
6573{
6574	get_random_bytes(&hashrnd, sizeof(hashrnd));
6575	return 0;
6576}
6577
6578late_initcall_sync(initialize_hashrnd);
6579
v3.1
   1/*
   2 * 	NET3	Protocol independent device support routines.
   3 *
   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.
   8 *
   9 *	Derived from the non IP parts of dev.c 1.0.19
  10 * 		Authors:	Ross Biro
  11 *				Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12 *				Mark Evans, <evansmp@uhura.aston.ac.uk>
  13 *
  14 *	Additional Authors:
  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>
  21 *
  22 *	Changes:
  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
  34 *					drivers
  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
  44 *					call a packet.
  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
  50 *					changes.
  51 *		Rudi Cilibrasi	:	Pass the right thing to
  52 *					set_mac_address()
  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
  58 *					1 device.
  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
  66 *					the backlog queue.
  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
  73 */
  74
  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>
  87#include <linux/mm.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>
  99#include <net/sock.h>
 100#include <linux/rtnetlink.h>
 101#include <linux/proc_fs.h>
 102#include <linux/seq_file.h>
 103#include <linux/stat.h>
 104#include <net/dst.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>
 125#include <net/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
 137#include "net-sysfs.h"
 138
 139/* Instead of increasing this, you should create a hash table. */
 140#define MAX_GRO_SKBS 8
 141
 142/* This should be increased if a protocol with a bigger head is added. */
 143#define GRO_MAX_HEAD (MAX_HEADER + 128)
 144
 145/*
 146 *	The list of packet types we will receive (as opposed to discard)
 147 *	and the routines to invoke.
 148 *
 149 *	Why 16. Because with 16 the only overlap we get on a hash of the
 150 *	low nibble of the protocol value is RARP/SNAP/X.25.
 151 *
 152 *      NOTE:  That is no longer true with the addition of VLAN tags.  Not
 153 *             sure which should go first, but I bet it won't make much
 154 *             difference if we are running VLANs.  The good news is that
 155 *             this protocol won't be in the list unless compiled in, so
 156 *             the average user (w/out VLANs) will not be adversely affected.
 157 *             --BLG
 158 *
 159 *		0800	IP
 160 *		8100    802.1Q VLAN
 161 *		0001	802.3
 162 *		0002	AX.25
 163 *		0004	802.2
 164 *		8035	RARP
 165 *		0005	SNAP
 166 *		0805	X.25
 167 *		0806	ARP
 168 *		8137	IPX
 169 *		0009	Localtalk
 170 *		86DD	IPv6
 171 */
 172
 173#define PTYPE_HASH_SIZE	(16)
 174#define PTYPE_HASH_MASK	(PTYPE_HASH_SIZE - 1)
 175
 176static DEFINE_SPINLOCK(ptype_lock);
 177static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
 178static struct list_head ptype_all __read_mostly;	/* Taps */
 179
 180/*
 181 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
 182 * semaphore.
 183 *
 184 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
 185 *
 186 * Writers must hold the rtnl semaphore while they loop through the
 187 * dev_base_head list, and hold dev_base_lock for writing when they do the
 188 * actual updates.  This allows pure readers to access the list even
 189 * while a writer is preparing to update it.
 190 *
 191 * To put it another way, dev_base_lock is held for writing only to
 192 * protect against pure readers; the rtnl semaphore provides the
 193 * protection against other writers.
 194 *
 195 * See, for example usages, register_netdevice() and
 196 * unregister_netdevice(), which must be called with the rtnl
 197 * semaphore held.
 198 */
 199DEFINE_RWLOCK(dev_base_lock);
 200EXPORT_SYMBOL(dev_base_lock);
 201
 202static inline void dev_base_seq_inc(struct net *net)
 203{
 204	while (++net->dev_base_seq == 0);
 205}
 206
 207static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
 208{
 209	unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
 
 210	return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
 211}
 212
 213static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
 214{
 215	return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
 216}
 217
 218static inline void rps_lock(struct softnet_data *sd)
 219{
 220#ifdef CONFIG_RPS
 221	spin_lock(&sd->input_pkt_queue.lock);
 222#endif
 223}
 224
 225static inline void rps_unlock(struct softnet_data *sd)
 226{
 227#ifdef CONFIG_RPS
 228	spin_unlock(&sd->input_pkt_queue.lock);
 229#endif
 230}
 231
 232/* Device list insertion */
 233static int list_netdevice(struct net_device *dev)
 234{
 235	struct net *net = dev_net(dev);
 236
 237	ASSERT_RTNL();
 238
 239	write_lock_bh(&dev_base_lock);
 240	list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
 241	hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
 242	hlist_add_head_rcu(&dev->index_hlist,
 243			   dev_index_hash(net, dev->ifindex));
 244	write_unlock_bh(&dev_base_lock);
 245
 246	dev_base_seq_inc(net);
 247
 248	return 0;
 249}
 250
 251/* Device list removal
 252 * caller must respect a RCU grace period before freeing/reusing dev
 253 */
 254static void unlist_netdevice(struct net_device *dev)
 255{
 256	ASSERT_RTNL();
 257
 258	/* Unlink dev from the device chain */
 259	write_lock_bh(&dev_base_lock);
 260	list_del_rcu(&dev->dev_list);
 261	hlist_del_rcu(&dev->name_hlist);
 262	hlist_del_rcu(&dev->index_hlist);
 263	write_unlock_bh(&dev_base_lock);
 264
 265	dev_base_seq_inc(dev_net(dev));
 266}
 267
 268/*
 269 *	Our notifier list
 270 */
 271
 272static RAW_NOTIFIER_HEAD(netdev_chain);
 273
 274/*
 275 *	Device drivers call our routines to queue packets here. We empty the
 276 *	queue in the local softnet handler.
 277 */
 278
 279DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
 280EXPORT_PER_CPU_SYMBOL(softnet_data);
 281
 282#ifdef CONFIG_LOCKDEP
 283/*
 284 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
 285 * according to dev->type
 286 */
 287static const unsigned short netdev_lock_type[] =
 288	{ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
 289	 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
 290	 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
 291	 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
 292	 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
 293	 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
 294	 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
 295	 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
 296	 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
 297	 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
 298	 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
 299	 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
 300	 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
 301	 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
 302	 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
 303	 ARPHRD_VOID, ARPHRD_NONE};
 304
 305static const char *const netdev_lock_name[] =
 306	{"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
 307	 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
 308	 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
 309	 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
 310	 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
 311	 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
 312	 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
 313	 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
 314	 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
 315	 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
 316	 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
 317	 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
 318	 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
 319	 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
 320	 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
 321	 "_xmit_VOID", "_xmit_NONE"};
 322
 323static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
 324static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
 325
 326static inline unsigned short netdev_lock_pos(unsigned short dev_type)
 327{
 328	int i;
 329
 330	for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
 331		if (netdev_lock_type[i] == dev_type)
 332			return i;
 333	/* the last key is used by default */
 334	return ARRAY_SIZE(netdev_lock_type) - 1;
 335}
 336
 337static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
 338						 unsigned short dev_type)
 339{
 340	int i;
 341
 342	i = netdev_lock_pos(dev_type);
 343	lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
 344				   netdev_lock_name[i]);
 345}
 346
 347static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
 348{
 349	int i;
 350
 351	i = netdev_lock_pos(dev->type);
 352	lockdep_set_class_and_name(&dev->addr_list_lock,
 353				   &netdev_addr_lock_key[i],
 354				   netdev_lock_name[i]);
 355}
 356#else
 357static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
 358						 unsigned short dev_type)
 359{
 360}
 361static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
 362{
 363}
 364#endif
 365
 366/*******************************************************************************
 367
 368		Protocol management and registration routines
 369
 370*******************************************************************************/
 371
 372/*
 373 *	Add a protocol ID to the list. Now that the input handler is
 374 *	smarter we can dispense with all the messy stuff that used to be
 375 *	here.
 376 *
 377 *	BEWARE!!! Protocol handlers, mangling input packets,
 378 *	MUST BE last in hash buckets and checking protocol handlers
 379 *	MUST start from promiscuous ptype_all chain in net_bh.
 380 *	It is true now, do not change it.
 381 *	Explanation follows: if protocol handler, mangling packet, will
 382 *	be the first on list, it is not able to sense, that packet
 383 *	is cloned and should be copied-on-write, so that it will
 384 *	change it and subsequent readers will get broken packet.
 385 *							--ANK (980803)
 386 */
 387
 388static inline struct list_head *ptype_head(const struct packet_type *pt)
 389{
 390	if (pt->type == htons(ETH_P_ALL))
 391		return &ptype_all;
 392	else
 393		return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
 394}
 395
 396/**
 397 *	dev_add_pack - add packet handler
 398 *	@pt: packet type declaration
 399 *
 400 *	Add a protocol handler to the networking stack. The passed &packet_type
 401 *	is linked into kernel lists and may not be freed until it has been
 402 *	removed from the kernel lists.
 403 *
 404 *	This call does not sleep therefore it can not
 405 *	guarantee all CPU's that are in middle of receiving packets
 406 *	will see the new packet type (until the next received packet).
 407 */
 408
 409void dev_add_pack(struct packet_type *pt)
 410{
 411	struct list_head *head = ptype_head(pt);
 412
 413	spin_lock(&ptype_lock);
 414	list_add_rcu(&pt->list, head);
 415	spin_unlock(&ptype_lock);
 416}
 417EXPORT_SYMBOL(dev_add_pack);
 418
 419/**
 420 *	__dev_remove_pack	 - remove packet handler
 421 *	@pt: packet type declaration
 422 *
 423 *	Remove a protocol handler that was previously added to the kernel
 424 *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
 425 *	from the kernel lists and can be freed or reused once this function
 426 *	returns.
 427 *
 428 *      The packet type might still be in use by receivers
 429 *	and must not be freed until after all the CPU's have gone
 430 *	through a quiescent state.
 431 */
 432void __dev_remove_pack(struct packet_type *pt)
 433{
 434	struct list_head *head = ptype_head(pt);
 435	struct packet_type *pt1;
 436
 437	spin_lock(&ptype_lock);
 438
 439	list_for_each_entry(pt1, head, list) {
 440		if (pt == pt1) {
 441			list_del_rcu(&pt->list);
 442			goto out;
 443		}
 444	}
 445
 446	printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
 447out:
 448	spin_unlock(&ptype_lock);
 449}
 450EXPORT_SYMBOL(__dev_remove_pack);
 451
 452/**
 453 *	dev_remove_pack	 - remove packet handler
 454 *	@pt: packet type declaration
 455 *
 456 *	Remove a protocol handler that was previously added to the kernel
 457 *	protocol handlers by dev_add_pack(). The passed &packet_type is removed
 458 *	from the kernel lists and can be freed or reused once this function
 459 *	returns.
 460 *
 461 *	This call sleeps to guarantee that no CPU is looking at the packet
 462 *	type after return.
 463 */
 464void dev_remove_pack(struct packet_type *pt)
 465{
 466	__dev_remove_pack(pt);
 467
 468	synchronize_net();
 469}
 470EXPORT_SYMBOL(dev_remove_pack);
 471
 472/******************************************************************************
 473
 474		      Device Boot-time Settings Routines
 475
 476*******************************************************************************/
 477
 478/* Boot time configuration table */
 479static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
 480
 481/**
 482 *	netdev_boot_setup_add	- add new setup entry
 483 *	@name: name of the device
 484 *	@map: configured settings for the device
 485 *
 486 *	Adds new setup entry to the dev_boot_setup list.  The function
 487 *	returns 0 on error and 1 on success.  This is a generic routine to
 488 *	all netdevices.
 489 */
 490static int netdev_boot_setup_add(char *name, struct ifmap *map)
 491{
 492	struct netdev_boot_setup *s;
 493	int i;
 494
 495	s = dev_boot_setup;
 496	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
 497		if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
 498			memset(s[i].name, 0, sizeof(s[i].name));
 499			strlcpy(s[i].name, name, IFNAMSIZ);
 500			memcpy(&s[i].map, map, sizeof(s[i].map));
 501			break;
 502		}
 503	}
 504
 505	return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
 506}
 507
 508/**
 509 *	netdev_boot_setup_check	- check boot time settings
 510 *	@dev: the netdevice
 511 *
 512 * 	Check boot time settings for the device.
 513 *	The found settings are set for the device to be used
 514 *	later in the device probing.
 515 *	Returns 0 if no settings found, 1 if they are.
 516 */
 517int netdev_boot_setup_check(struct net_device *dev)
 518{
 519	struct netdev_boot_setup *s = dev_boot_setup;
 520	int i;
 521
 522	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
 523		if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
 524		    !strcmp(dev->name, s[i].name)) {
 525			dev->irq 	= s[i].map.irq;
 526			dev->base_addr 	= s[i].map.base_addr;
 527			dev->mem_start 	= s[i].map.mem_start;
 528			dev->mem_end 	= s[i].map.mem_end;
 529			return 1;
 530		}
 531	}
 532	return 0;
 533}
 534EXPORT_SYMBOL(netdev_boot_setup_check);
 535
 536
 537/**
 538 *	netdev_boot_base	- get address from boot time settings
 539 *	@prefix: prefix for network device
 540 *	@unit: id for network device
 541 *
 542 * 	Check boot time settings for the base address of device.
 543 *	The found settings are set for the device to be used
 544 *	later in the device probing.
 545 *	Returns 0 if no settings found.
 546 */
 547unsigned long netdev_boot_base(const char *prefix, int unit)
 548{
 549	const struct netdev_boot_setup *s = dev_boot_setup;
 550	char name[IFNAMSIZ];
 551	int i;
 552
 553	sprintf(name, "%s%d", prefix, unit);
 554
 555	/*
 556	 * If device already registered then return base of 1
 557	 * to indicate not to probe for this interface
 558	 */
 559	if (__dev_get_by_name(&init_net, name))
 560		return 1;
 561
 562	for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
 563		if (!strcmp(name, s[i].name))
 564			return s[i].map.base_addr;
 565	return 0;
 566}
 567
 568/*
 569 * Saves at boot time configured settings for any netdevice.
 570 */
 571int __init netdev_boot_setup(char *str)
 572{
 573	int ints[5];
 574	struct ifmap map;
 575
 576	str = get_options(str, ARRAY_SIZE(ints), ints);
 577	if (!str || !*str)
 578		return 0;
 579
 580	/* Save settings */
 581	memset(&map, 0, sizeof(map));
 582	if (ints[0] > 0)
 583		map.irq = ints[1];
 584	if (ints[0] > 1)
 585		map.base_addr = ints[2];
 586	if (ints[0] > 2)
 587		map.mem_start = ints[3];
 588	if (ints[0] > 3)
 589		map.mem_end = ints[4];
 590
 591	/* Add new entry to the list */
 592	return netdev_boot_setup_add(str, &map);
 593}
 594
 595__setup("netdev=", netdev_boot_setup);
 596
 597/*******************************************************************************
 598
 599			    Device Interface Subroutines
 600
 601*******************************************************************************/
 602
 603/**
 604 *	__dev_get_by_name	- find a device by its name
 605 *	@net: the applicable net namespace
 606 *	@name: name to find
 607 *
 608 *	Find an interface by name. Must be called under RTNL semaphore
 609 *	or @dev_base_lock. If the name is found a pointer to the device
 610 *	is returned. If the name is not found then %NULL is returned. The
 611 *	reference counters are not incremented so the caller must be
 612 *	careful with locks.
 613 */
 614
 615struct net_device *__dev_get_by_name(struct net *net, const char *name)
 616{
 617	struct hlist_node *p;
 618	struct net_device *dev;
 619	struct hlist_head *head = dev_name_hash(net, name);
 620
 621	hlist_for_each_entry(dev, p, head, name_hlist)
 622		if (!strncmp(dev->name, name, IFNAMSIZ))
 623			return dev;
 624
 625	return NULL;
 626}
 627EXPORT_SYMBOL(__dev_get_by_name);
 628
 629/**
 630 *	dev_get_by_name_rcu	- find a device by its name
 631 *	@net: the applicable net namespace
 632 *	@name: name to find
 633 *
 634 *	Find an interface by name.
 635 *	If the name is found a pointer to the device is returned.
 636 * 	If the name is not found then %NULL is returned.
 637 *	The reference counters are not incremented so the caller must be
 638 *	careful with locks. The caller must hold RCU lock.
 639 */
 640
 641struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
 642{
 643	struct hlist_node *p;
 644	struct net_device *dev;
 645	struct hlist_head *head = dev_name_hash(net, name);
 646
 647	hlist_for_each_entry_rcu(dev, p, head, name_hlist)
 648		if (!strncmp(dev->name, name, IFNAMSIZ))
 649			return dev;
 650
 651	return NULL;
 652}
 653EXPORT_SYMBOL(dev_get_by_name_rcu);
 654
 655/**
 656 *	dev_get_by_name		- find a device by its name
 657 *	@net: the applicable net namespace
 658 *	@name: name to find
 659 *
 660 *	Find an interface by name. This can be called from any
 661 *	context and does its own locking. The returned handle has
 662 *	the usage count incremented and the caller must use dev_put() to
 663 *	release it when it is no longer needed. %NULL is returned if no
 664 *	matching device is found.
 665 */
 666
 667struct net_device *dev_get_by_name(struct net *net, const char *name)
 668{
 669	struct net_device *dev;
 670
 671	rcu_read_lock();
 672	dev = dev_get_by_name_rcu(net, name);
 673	if (dev)
 674		dev_hold(dev);
 675	rcu_read_unlock();
 676	return dev;
 677}
 678EXPORT_SYMBOL(dev_get_by_name);
 679
 680/**
 681 *	__dev_get_by_index - find a device by its ifindex
 682 *	@net: the applicable net namespace
 683 *	@ifindex: index of device
 684 *
 685 *	Search for an interface by index. Returns %NULL if the device
 686 *	is not found or a pointer to the device. The device has not
 687 *	had its reference counter increased so the caller must be careful
 688 *	about locking. The caller must hold either the RTNL semaphore
 689 *	or @dev_base_lock.
 690 */
 691
 692struct net_device *__dev_get_by_index(struct net *net, int ifindex)
 693{
 694	struct hlist_node *p;
 695	struct net_device *dev;
 696	struct hlist_head *head = dev_index_hash(net, ifindex);
 697
 698	hlist_for_each_entry(dev, p, head, index_hlist)
 699		if (dev->ifindex == ifindex)
 700			return dev;
 701
 702	return NULL;
 703}
 704EXPORT_SYMBOL(__dev_get_by_index);
 705
 706/**
 707 *	dev_get_by_index_rcu - find a device by its ifindex
 708 *	@net: the applicable net namespace
 709 *	@ifindex: index of device
 710 *
 711 *	Search for an interface by index. Returns %NULL if the device
 712 *	is not found or a pointer to the device. The device has not
 713 *	had its reference counter increased so the caller must be careful
 714 *	about locking. The caller must hold RCU lock.
 715 */
 716
 717struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
 718{
 719	struct hlist_node *p;
 720	struct net_device *dev;
 721	struct hlist_head *head = dev_index_hash(net, ifindex);
 722
 723	hlist_for_each_entry_rcu(dev, p, head, index_hlist)
 724		if (dev->ifindex == ifindex)
 725			return dev;
 726
 727	return NULL;
 728}
 729EXPORT_SYMBOL(dev_get_by_index_rcu);
 730
 731
 732/**
 733 *	dev_get_by_index - find a device by its ifindex
 734 *	@net: the applicable net namespace
 735 *	@ifindex: index of device
 736 *
 737 *	Search for an interface by index. Returns NULL if the device
 738 *	is not found or a pointer to the device. The device returned has
 739 *	had a reference added and the pointer is safe until the user calls
 740 *	dev_put to indicate they have finished with it.
 741 */
 742
 743struct net_device *dev_get_by_index(struct net *net, int ifindex)
 744{
 745	struct net_device *dev;
 746
 747	rcu_read_lock();
 748	dev = dev_get_by_index_rcu(net, ifindex);
 749	if (dev)
 750		dev_hold(dev);
 751	rcu_read_unlock();
 752	return dev;
 753}
 754EXPORT_SYMBOL(dev_get_by_index);
 755
 756/**
 757 *	dev_getbyhwaddr_rcu - find a device by its hardware address
 758 *	@net: the applicable net namespace
 759 *	@type: media type of device
 760 *	@ha: hardware address
 761 *
 762 *	Search for an interface by MAC address. Returns NULL if the device
 763 *	is not found or a pointer to the device.
 764 *	The caller must hold RCU or RTNL.
 765 *	The returned device has not had its ref count increased
 766 *	and the caller must therefore be careful about locking
 767 *
 768 */
 769
 770struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
 771				       const char *ha)
 772{
 773	struct net_device *dev;
 774
 775	for_each_netdev_rcu(net, dev)
 776		if (dev->type == type &&
 777		    !memcmp(dev->dev_addr, ha, dev->addr_len))
 778			return dev;
 779
 780	return NULL;
 781}
 782EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
 783
 784struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
 785{
 786	struct net_device *dev;
 787
 788	ASSERT_RTNL();
 789	for_each_netdev(net, dev)
 790		if (dev->type == type)
 791			return dev;
 792
 793	return NULL;
 794}
 795EXPORT_SYMBOL(__dev_getfirstbyhwtype);
 796
 797struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
 798{
 799	struct net_device *dev, *ret = NULL;
 800
 801	rcu_read_lock();
 802	for_each_netdev_rcu(net, dev)
 803		if (dev->type == type) {
 804			dev_hold(dev);
 805			ret = dev;
 806			break;
 807		}
 808	rcu_read_unlock();
 809	return ret;
 810}
 811EXPORT_SYMBOL(dev_getfirstbyhwtype);
 812
 813/**
 814 *	dev_get_by_flags_rcu - find any device with given flags
 815 *	@net: the applicable net namespace
 816 *	@if_flags: IFF_* values
 817 *	@mask: bitmask of bits in if_flags to check
 818 *
 819 *	Search for any interface with the given flags. Returns NULL if a device
 820 *	is not found or a pointer to the device. Must be called inside
 821 *	rcu_read_lock(), and result refcount is unchanged.
 822 */
 823
 824struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
 825				    unsigned short mask)
 826{
 827	struct net_device *dev, *ret;
 828
 829	ret = NULL;
 830	for_each_netdev_rcu(net, dev) {
 831		if (((dev->flags ^ if_flags) & mask) == 0) {
 832			ret = dev;
 833			break;
 834		}
 835	}
 836	return ret;
 837}
 838EXPORT_SYMBOL(dev_get_by_flags_rcu);
 839
 840/**
 841 *	dev_valid_name - check if name is okay for network device
 842 *	@name: name string
 843 *
 844 *	Network device names need to be valid file names to
 845 *	to allow sysfs to work.  We also disallow any kind of
 846 *	whitespace.
 847 */
 848int dev_valid_name(const char *name)
 849{
 850	if (*name == '\0')
 851		return 0;
 852	if (strlen(name) >= IFNAMSIZ)
 853		return 0;
 854	if (!strcmp(name, ".") || !strcmp(name, ".."))
 855		return 0;
 856
 857	while (*name) {
 858		if (*name == '/' || isspace(*name))
 859			return 0;
 860		name++;
 861	}
 862	return 1;
 863}
 864EXPORT_SYMBOL(dev_valid_name);
 865
 866/**
 867 *	__dev_alloc_name - allocate a name for a device
 868 *	@net: network namespace to allocate the device name in
 869 *	@name: name format string
 870 *	@buf:  scratch buffer and result name string
 871 *
 872 *	Passed a format string - eg "lt%d" it will try and find a suitable
 873 *	id. It scans list of devices to build up a free map, then chooses
 874 *	the first empty slot. The caller must hold the dev_base or rtnl lock
 875 *	while allocating the name and adding the device in order to avoid
 876 *	duplicates.
 877 *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
 878 *	Returns the number of the unit assigned or a negative errno code.
 879 */
 880
 881static int __dev_alloc_name(struct net *net, const char *name, char *buf)
 882{
 883	int i = 0;
 884	const char *p;
 885	const int max_netdevices = 8*PAGE_SIZE;
 886	unsigned long *inuse;
 887	struct net_device *d;
 888
 889	p = strnchr(name, IFNAMSIZ-1, '%');
 890	if (p) {
 891		/*
 892		 * Verify the string as this thing may have come from
 893		 * the user.  There must be either one "%d" and no other "%"
 894		 * characters.
 895		 */
 896		if (p[1] != 'd' || strchr(p + 2, '%'))
 897			return -EINVAL;
 898
 899		/* Use one page as a bit array of possible slots */
 900		inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
 901		if (!inuse)
 902			return -ENOMEM;
 903
 904		for_each_netdev(net, d) {
 905			if (!sscanf(d->name, name, &i))
 906				continue;
 907			if (i < 0 || i >= max_netdevices)
 908				continue;
 909
 910			/*  avoid cases where sscanf is not exact inverse of printf */
 911			snprintf(buf, IFNAMSIZ, name, i);
 912			if (!strncmp(buf, d->name, IFNAMSIZ))
 913				set_bit(i, inuse);
 914		}
 915
 916		i = find_first_zero_bit(inuse, max_netdevices);
 917		free_page((unsigned long) inuse);
 918	}
 919
 920	if (buf != name)
 921		snprintf(buf, IFNAMSIZ, name, i);
 922	if (!__dev_get_by_name(net, buf))
 923		return i;
 924
 925	/* It is possible to run out of possible slots
 926	 * when the name is long and there isn't enough space left
 927	 * for the digits, or if all bits are used.
 928	 */
 929	return -ENFILE;
 930}
 931
 932/**
 933 *	dev_alloc_name - allocate a name for a device
 934 *	@dev: device
 935 *	@name: name format string
 936 *
 937 *	Passed a format string - eg "lt%d" it will try and find a suitable
 938 *	id. It scans list of devices to build up a free map, then chooses
 939 *	the first empty slot. The caller must hold the dev_base or rtnl lock
 940 *	while allocating the name and adding the device in order to avoid
 941 *	duplicates.
 942 *	Limited to bits_per_byte * page size devices (ie 32K on most platforms).
 943 *	Returns the number of the unit assigned or a negative errno code.
 944 */
 945
 946int dev_alloc_name(struct net_device *dev, const char *name)
 947{
 948	char buf[IFNAMSIZ];
 949	struct net *net;
 950	int ret;
 951
 952	BUG_ON(!dev_net(dev));
 953	net = dev_net(dev);
 954	ret = __dev_alloc_name(net, name, buf);
 955	if (ret >= 0)
 956		strlcpy(dev->name, buf, IFNAMSIZ);
 957	return ret;
 958}
 959EXPORT_SYMBOL(dev_alloc_name);
 960
 961static int dev_get_valid_name(struct net_device *dev, const char *name)
 962{
 963	struct net *net;
 964
 965	BUG_ON(!dev_net(dev));
 966	net = dev_net(dev);
 967
 968	if (!dev_valid_name(name))
 969		return -EINVAL;
 970
 971	if (strchr(name, '%'))
 972		return dev_alloc_name(dev, name);
 973	else if (__dev_get_by_name(net, name))
 974		return -EEXIST;
 975	else if (dev->name != name)
 976		strlcpy(dev->name, name, IFNAMSIZ);
 977
 978	return 0;
 979}
 980
 981/**
 982 *	dev_change_name - change name of a device
 983 *	@dev: device
 984 *	@newname: name (or format string) must be at least IFNAMSIZ
 985 *
 986 *	Change name of a device, can pass format strings "eth%d".
 987 *	for wildcarding.
 988 */
 989int dev_change_name(struct net_device *dev, const char *newname)
 990{
 991	char oldname[IFNAMSIZ];
 992	int err = 0;
 993	int ret;
 994	struct net *net;
 995
 996	ASSERT_RTNL();
 997	BUG_ON(!dev_net(dev));
 998
 999	net = dev_net(dev);
1000	if (dev->flags & IFF_UP)
1001		return -EBUSY;
1002
1003	if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1004		return 0;
1005
1006	memcpy(oldname, dev->name, IFNAMSIZ);
1007
1008	err = dev_get_valid_name(dev, newname);
1009	if (err < 0)
1010		return err;
1011
1012rollback:
1013	ret = device_rename(&dev->dev, dev->name);
1014	if (ret) {
1015		memcpy(dev->name, oldname, IFNAMSIZ);
1016		return ret;
1017	}
1018
1019	write_lock_bh(&dev_base_lock);
1020	hlist_del_rcu(&dev->name_hlist);
1021	write_unlock_bh(&dev_base_lock);
1022
1023	synchronize_rcu();
1024
1025	write_lock_bh(&dev_base_lock);
1026	hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1027	write_unlock_bh(&dev_base_lock);
1028
1029	ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1030	ret = notifier_to_errno(ret);
1031
1032	if (ret) {
1033		/* err >= 0 after dev_alloc_name() or stores the first errno */
1034		if (err >= 0) {
1035			err = ret;
1036			memcpy(dev->name, oldname, IFNAMSIZ);
1037			goto rollback;
1038		} else {
1039			printk(KERN_ERR
1040			       "%s: name change rollback failed: %d.\n",
1041			       dev->name, ret);
1042		}
1043	}
1044
1045	return err;
1046}
1047
1048/**
1049 *	dev_set_alias - change ifalias of a device
1050 *	@dev: device
1051 *	@alias: name up to IFALIASZ
1052 *	@len: limit of bytes to copy from info
1053 *
1054 *	Set ifalias for a device,
1055 */
1056int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1057{
 
 
1058	ASSERT_RTNL();
1059
1060	if (len >= IFALIASZ)
1061		return -EINVAL;
1062
1063	if (!len) {
1064		if (dev->ifalias) {
1065			kfree(dev->ifalias);
1066			dev->ifalias = NULL;
1067		}
1068		return 0;
1069	}
1070
1071	dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1072	if (!dev->ifalias)
1073		return -ENOMEM;
 
1074
1075	strlcpy(dev->ifalias, alias, len+1);
1076	return len;
1077}
1078
1079
1080/**
1081 *	netdev_features_change - device changes features
1082 *	@dev: device to cause notification
1083 *
1084 *	Called to indicate a device has changed features.
1085 */
1086void netdev_features_change(struct net_device *dev)
1087{
1088	call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1089}
1090EXPORT_SYMBOL(netdev_features_change);
1091
1092/**
1093 *	netdev_state_change - device changes state
1094 *	@dev: device to cause notification
1095 *
1096 *	Called to indicate a device has changed state. This function calls
1097 *	the notifier chains for netdev_chain and sends a NEWLINK message
1098 *	to the routing socket.
1099 */
1100void netdev_state_change(struct net_device *dev)
1101{
1102	if (dev->flags & IFF_UP) {
1103		call_netdevice_notifiers(NETDEV_CHANGE, dev);
1104		rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1105	}
1106}
1107EXPORT_SYMBOL(netdev_state_change);
1108
1109int netdev_bonding_change(struct net_device *dev, unsigned long event)
1110{
1111	return call_netdevice_notifiers(event, dev);
1112}
1113EXPORT_SYMBOL(netdev_bonding_change);
1114
1115/**
1116 *	dev_load 	- load a network module
1117 *	@net: the applicable net namespace
1118 *	@name: name of interface
1119 *
1120 *	If a network interface is not present and the process has suitable
1121 *	privileges this function loads the module. If module loading is not
1122 *	available in this kernel then it becomes a nop.
1123 */
1124
1125void dev_load(struct net *net, const char *name)
1126{
1127	struct net_device *dev;
1128	int no_module;
1129
1130	rcu_read_lock();
1131	dev = dev_get_by_name_rcu(net, name);
1132	rcu_read_unlock();
1133
1134	no_module = !dev;
1135	if (no_module && capable(CAP_NET_ADMIN))
1136		no_module = request_module("netdev-%s", name);
1137	if (no_module && capable(CAP_SYS_MODULE)) {
1138		if (!request_module("%s", name))
1139			pr_err("Loading kernel module for a network device "
1140"with CAP_SYS_MODULE (deprecated).  Use CAP_NET_ADMIN and alias netdev-%s "
1141"instead\n", name);
1142	}
1143}
1144EXPORT_SYMBOL(dev_load);
1145
1146static int __dev_open(struct net_device *dev)
1147{
1148	const struct net_device_ops *ops = dev->netdev_ops;
1149	int ret;
1150
1151	ASSERT_RTNL();
1152
1153	if (!netif_device_present(dev))
1154		return -ENODEV;
1155
1156	ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1157	ret = notifier_to_errno(ret);
1158	if (ret)
1159		return ret;
1160
1161	set_bit(__LINK_STATE_START, &dev->state);
1162
1163	if (ops->ndo_validate_addr)
1164		ret = ops->ndo_validate_addr(dev);
1165
1166	if (!ret && ops->ndo_open)
1167		ret = ops->ndo_open(dev);
1168
1169	if (ret)
1170		clear_bit(__LINK_STATE_START, &dev->state);
1171	else {
1172		dev->flags |= IFF_UP;
1173		net_dmaengine_get();
1174		dev_set_rx_mode(dev);
1175		dev_activate(dev);
 
1176	}
1177
1178	return ret;
1179}
1180
1181/**
1182 *	dev_open	- prepare an interface for use.
1183 *	@dev:	device to open
1184 *
1185 *	Takes a device from down to up state. The device's private open
1186 *	function is invoked and then the multicast lists are loaded. Finally
1187 *	the device is moved into the up state and a %NETDEV_UP message is
1188 *	sent to the netdev notifier chain.
1189 *
1190 *	Calling this function on an active interface is a nop. On a failure
1191 *	a negative errno code is returned.
1192 */
1193int dev_open(struct net_device *dev)
1194{
1195	int ret;
1196
1197	if (dev->flags & IFF_UP)
1198		return 0;
1199
1200	ret = __dev_open(dev);
1201	if (ret < 0)
1202		return ret;
1203
1204	rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1205	call_netdevice_notifiers(NETDEV_UP, dev);
1206
1207	return ret;
1208}
1209EXPORT_SYMBOL(dev_open);
1210
1211static int __dev_close_many(struct list_head *head)
1212{
1213	struct net_device *dev;
1214
1215	ASSERT_RTNL();
1216	might_sleep();
1217
1218	list_for_each_entry(dev, head, unreg_list) {
1219		call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1220
1221		clear_bit(__LINK_STATE_START, &dev->state);
1222
1223		/* Synchronize to scheduled poll. We cannot touch poll list, it
1224		 * can be even on different cpu. So just clear netif_running().
1225		 *
1226		 * dev->stop() will invoke napi_disable() on all of it's
1227		 * napi_struct instances on this device.
1228		 */
1229		smp_mb__after_clear_bit(); /* Commit netif_running(). */
1230	}
1231
1232	dev_deactivate_many(head);
1233
1234	list_for_each_entry(dev, head, unreg_list) {
1235		const struct net_device_ops *ops = dev->netdev_ops;
1236
1237		/*
1238		 *	Call the device specific close. This cannot fail.
1239		 *	Only if device is UP
1240		 *
1241		 *	We allow it to be called even after a DETACH hot-plug
1242		 *	event.
1243		 */
1244		if (ops->ndo_stop)
1245			ops->ndo_stop(dev);
1246
1247		dev->flags &= ~IFF_UP;
1248		net_dmaengine_put();
1249	}
1250
1251	return 0;
1252}
1253
1254static int __dev_close(struct net_device *dev)
1255{
1256	int retval;
1257	LIST_HEAD(single);
1258
1259	list_add(&dev->unreg_list, &single);
1260	retval = __dev_close_many(&single);
1261	list_del(&single);
1262	return retval;
1263}
1264
1265static int dev_close_many(struct list_head *head)
1266{
1267	struct net_device *dev, *tmp;
1268	LIST_HEAD(tmp_list);
1269
1270	list_for_each_entry_safe(dev, tmp, head, unreg_list)
1271		if (!(dev->flags & IFF_UP))
1272			list_move(&dev->unreg_list, &tmp_list);
1273
1274	__dev_close_many(head);
1275
1276	list_for_each_entry(dev, head, unreg_list) {
1277		rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1278		call_netdevice_notifiers(NETDEV_DOWN, dev);
1279	}
1280
1281	/* rollback_registered_many needs the complete original list */
1282	list_splice(&tmp_list, head);
1283	return 0;
1284}
1285
1286/**
1287 *	dev_close - shutdown an interface.
1288 *	@dev: device to shutdown
1289 *
1290 *	This function moves an active device into down state. A
1291 *	%NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1292 *	is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1293 *	chain.
1294 */
1295int dev_close(struct net_device *dev)
1296{
1297	if (dev->flags & IFF_UP) {
1298		LIST_HEAD(single);
1299
1300		list_add(&dev->unreg_list, &single);
1301		dev_close_many(&single);
1302		list_del(&single);
1303	}
1304	return 0;
1305}
1306EXPORT_SYMBOL(dev_close);
1307
1308
1309/**
1310 *	dev_disable_lro - disable Large Receive Offload on a device
1311 *	@dev: device
1312 *
1313 *	Disable Large Receive Offload (LRO) on a net device.  Must be
1314 *	called under RTNL.  This is needed if received packets may be
1315 *	forwarded to another interface.
1316 */
1317void dev_disable_lro(struct net_device *dev)
1318{
1319	u32 flags;
1320
1321	/*
1322	 * If we're trying to disable lro on a vlan device
1323	 * use the underlying physical device instead
1324	 */
1325	if (is_vlan_dev(dev))
1326		dev = vlan_dev_real_dev(dev);
1327
1328	if (dev->ethtool_ops && dev->ethtool_ops->get_flags)
1329		flags = dev->ethtool_ops->get_flags(dev);
1330	else
1331		flags = ethtool_op_get_flags(dev);
1332
1333	if (!(flags & ETH_FLAG_LRO))
1334		return;
1335
1336	__ethtool_set_flags(dev, flags & ~ETH_FLAG_LRO);
1337	if (unlikely(dev->features & NETIF_F_LRO))
1338		netdev_WARN(dev, "failed to disable LRO!\n");
1339}
1340EXPORT_SYMBOL(dev_disable_lro);
1341
1342
1343static int dev_boot_phase = 1;
1344
1345/**
1346 *	register_netdevice_notifier - register a network notifier block
1347 *	@nb: notifier
1348 *
1349 *	Register a notifier to be called when network device events occur.
1350 *	The notifier passed is linked into the kernel structures and must
1351 *	not be reused until it has been unregistered. A negative errno code
1352 *	is returned on a failure.
1353 *
1354 * 	When registered all registration and up events are replayed
1355 *	to the new notifier to allow device to have a race free
1356 *	view of the network device list.
1357 */
1358
1359int register_netdevice_notifier(struct notifier_block *nb)
1360{
1361	struct net_device *dev;
1362	struct net_device *last;
1363	struct net *net;
1364	int err;
1365
1366	rtnl_lock();
1367	err = raw_notifier_chain_register(&netdev_chain, nb);
1368	if (err)
1369		goto unlock;
1370	if (dev_boot_phase)
1371		goto unlock;
1372	for_each_net(net) {
1373		for_each_netdev(net, dev) {
1374			err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1375			err = notifier_to_errno(err);
1376			if (err)
1377				goto rollback;
1378
1379			if (!(dev->flags & IFF_UP))
1380				continue;
1381
1382			nb->notifier_call(nb, NETDEV_UP, dev);
1383		}
1384	}
1385
1386unlock:
1387	rtnl_unlock();
1388	return err;
1389
1390rollback:
1391	last = dev;
1392	for_each_net(net) {
1393		for_each_netdev(net, dev) {
1394			if (dev == last)
1395				break;
1396
1397			if (dev->flags & IFF_UP) {
1398				nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1399				nb->notifier_call(nb, NETDEV_DOWN, dev);
1400			}
1401			nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1402			nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1403		}
1404	}
1405
 
1406	raw_notifier_chain_unregister(&netdev_chain, nb);
1407	goto unlock;
1408}
1409EXPORT_SYMBOL(register_netdevice_notifier);
1410
1411/**
1412 *	unregister_netdevice_notifier - unregister a network notifier block
1413 *	@nb: notifier
1414 *
1415 *	Unregister a notifier previously registered by
1416 *	register_netdevice_notifier(). The notifier is unlinked into the
1417 *	kernel structures and may then be reused. A negative errno code
1418 *	is returned on a failure.
 
 
 
 
1419 */
1420
1421int unregister_netdevice_notifier(struct notifier_block *nb)
1422{
 
 
1423	int err;
1424
1425	rtnl_lock();
1426	err = raw_notifier_chain_unregister(&netdev_chain, nb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1427	rtnl_unlock();
1428	return err;
1429}
1430EXPORT_SYMBOL(unregister_netdevice_notifier);
1431
1432/**
1433 *	call_netdevice_notifiers - call all network notifier blocks
1434 *      @val: value passed unmodified to notifier function
1435 *      @dev: net_device pointer passed unmodified to notifier function
1436 *
1437 *	Call all network notifier blocks.  Parameters and return value
1438 *	are as for raw_notifier_call_chain().
1439 */
1440
1441int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1442{
1443	ASSERT_RTNL();
1444	return raw_notifier_call_chain(&netdev_chain, val, dev);
1445}
1446EXPORT_SYMBOL(call_netdevice_notifiers);
1447
1448/* When > 0 there are consumers of rx skb time stamps */
1449static atomic_t netstamp_needed = ATOMIC_INIT(0);
 
 
 
 
 
 
1450
1451void net_enable_timestamp(void)
1452{
1453	atomic_inc(&netstamp_needed);
 
 
 
 
 
 
 
 
 
 
1454}
1455EXPORT_SYMBOL(net_enable_timestamp);
1456
1457void net_disable_timestamp(void)
1458{
1459	atomic_dec(&netstamp_needed);
 
 
 
 
 
 
1460}
1461EXPORT_SYMBOL(net_disable_timestamp);
1462
1463static inline void net_timestamp_set(struct sk_buff *skb)
1464{
1465	if (atomic_read(&netstamp_needed))
 
1466		__net_timestamp(skb);
1467	else
1468		skb->tstamp.tv64 = 0;
1469}
1470
1471static inline void net_timestamp_check(struct sk_buff *skb)
1472{
1473	if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1474		__net_timestamp(skb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1475}
1476
1477static inline bool is_skb_forwardable(struct net_device *dev,
1478				      struct sk_buff *skb)
1479{
1480	unsigned int len;
1481
1482	if (!(dev->flags & IFF_UP))
1483		return false;
1484
1485	len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1486	if (skb->len <= len)
1487		return true;
1488
1489	/* if TSO is enabled, we don't care about the length as the packet
1490	 * could be forwarded without being segmented before
1491	 */
1492	if (skb_is_gso(skb))
1493		return true;
1494
1495	return false;
1496}
1497
1498/**
1499 * dev_forward_skb - loopback an skb to another netif
1500 *
1501 * @dev: destination network device
1502 * @skb: buffer to forward
1503 *
1504 * return values:
1505 *	NET_RX_SUCCESS	(no congestion)
1506 *	NET_RX_DROP     (packet was dropped, but freed)
1507 *
1508 * dev_forward_skb can be used for injecting an skb from the
1509 * start_xmit function of one device into the receive queue
1510 * of another device.
1511 *
1512 * The receiving device may be in another namespace, so
1513 * we have to clear all information in the skb that could
1514 * impact namespace isolation.
1515 */
1516int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1517{
1518	if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1519		if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1520			atomic_long_inc(&dev->rx_dropped);
1521			kfree_skb(skb);
1522			return NET_RX_DROP;
1523		}
1524	}
1525
1526	skb_orphan(skb);
1527	nf_reset(skb);
1528
1529	if (unlikely(!is_skb_forwardable(dev, skb))) {
1530		atomic_long_inc(&dev->rx_dropped);
1531		kfree_skb(skb);
1532		return NET_RX_DROP;
1533	}
1534	skb_set_dev(skb, dev);
 
 
1535	skb->tstamp.tv64 = 0;
1536	skb->pkt_type = PACKET_HOST;
1537	skb->protocol = eth_type_trans(skb, dev);
 
 
 
1538	return netif_rx(skb);
1539}
1540EXPORT_SYMBOL_GPL(dev_forward_skb);
1541
1542static inline int deliver_skb(struct sk_buff *skb,
1543			      struct packet_type *pt_prev,
1544			      struct net_device *orig_dev)
1545{
1546	atomic_inc(&skb->users);
1547	return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1548}
1549
 
 
 
 
 
 
 
 
 
 
 
 
 
1550/*
1551 *	Support routine. Sends outgoing frames to any network
1552 *	taps currently in use.
1553 */
1554
1555static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1556{
1557	struct packet_type *ptype;
1558	struct sk_buff *skb2 = NULL;
1559	struct packet_type *pt_prev = NULL;
1560
1561	rcu_read_lock();
1562	list_for_each_entry_rcu(ptype, &ptype_all, list) {
1563		/* Never send packets back to the socket
1564		 * they originated from - MvS (miquels@drinkel.ow.org)
1565		 */
1566		if ((ptype->dev == dev || !ptype->dev) &&
1567		    (ptype->af_packet_priv == NULL ||
1568		     (struct sock *)ptype->af_packet_priv != skb->sk)) {
1569			if (pt_prev) {
1570				deliver_skb(skb2, pt_prev, skb->dev);
1571				pt_prev = ptype;
1572				continue;
1573			}
1574
1575			skb2 = skb_clone(skb, GFP_ATOMIC);
1576			if (!skb2)
1577				break;
1578
1579			net_timestamp_set(skb2);
1580
1581			/* skb->nh should be correctly
1582			   set by sender, so that the second statement is
1583			   just protection against buggy protocols.
1584			 */
1585			skb_reset_mac_header(skb2);
1586
1587			if (skb_network_header(skb2) < skb2->data ||
1588			    skb2->network_header > skb2->tail) {
1589				if (net_ratelimit())
1590					printk(KERN_CRIT "protocol %04x is "
1591					       "buggy, dev %s\n",
1592					       ntohs(skb2->protocol),
1593					       dev->name);
1594				skb_reset_network_header(skb2);
1595			}
1596
1597			skb2->transport_header = skb2->network_header;
1598			skb2->pkt_type = PACKET_OUTGOING;
1599			pt_prev = ptype;
1600		}
1601	}
1602	if (pt_prev)
1603		pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1604	rcu_read_unlock();
1605}
1606
1607/* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1608 * @dev: Network device
1609 * @txq: number of queues available
1610 *
1611 * If real_num_tx_queues is changed the tc mappings may no longer be
1612 * valid. To resolve this verify the tc mapping remains valid and if
1613 * not NULL the mapping. With no priorities mapping to this
1614 * offset/count pair it will no longer be used. In the worst case TC0
1615 * is invalid nothing can be done so disable priority mappings. If is
1616 * expected that drivers will fix this mapping if they can before
1617 * calling netif_set_real_num_tx_queues.
1618 */
1619static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1620{
1621	int i;
1622	struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1623
1624	/* If TC0 is invalidated disable TC mapping */
1625	if (tc->offset + tc->count > txq) {
1626		pr_warning("Number of in use tx queues changed "
1627			   "invalidating tc mappings. Priority "
1628			   "traffic classification disabled!\n");
1629		dev->num_tc = 0;
1630		return;
1631	}
1632
1633	/* Invalidated prio to tc mappings set to TC0 */
1634	for (i = 1; i < TC_BITMASK + 1; i++) {
1635		int q = netdev_get_prio_tc_map(dev, i);
1636
1637		tc = &dev->tc_to_txq[q];
1638		if (tc->offset + tc->count > txq) {
1639			pr_warning("Number of in use tx queues "
1640				   "changed. Priority %i to tc "
1641				   "mapping %i is no longer valid "
1642				   "setting map to 0\n",
1643				   i, q);
1644			netdev_set_prio_tc_map(dev, i, 0);
1645		}
1646	}
1647}
1648
1649/*
1650 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1651 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1652 */
1653int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1654{
1655	int rc;
1656
1657	if (txq < 1 || txq > dev->num_tx_queues)
1658		return -EINVAL;
1659
1660	if (dev->reg_state == NETREG_REGISTERED ||
1661	    dev->reg_state == NETREG_UNREGISTERING) {
1662		ASSERT_RTNL();
1663
1664		rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1665						  txq);
1666		if (rc)
1667			return rc;
1668
1669		if (dev->num_tc)
1670			netif_setup_tc(dev, txq);
1671
1672		if (txq < dev->real_num_tx_queues)
1673			qdisc_reset_all_tx_gt(dev, txq);
1674	}
1675
1676	dev->real_num_tx_queues = txq;
1677	return 0;
1678}
1679EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1680
1681#ifdef CONFIG_RPS
1682/**
1683 *	netif_set_real_num_rx_queues - set actual number of RX queues used
1684 *	@dev: Network device
1685 *	@rxq: Actual number of RX queues
1686 *
1687 *	This must be called either with the rtnl_lock held or before
1688 *	registration of the net device.  Returns 0 on success, or a
1689 *	negative error code.  If called before registration, it always
1690 *	succeeds.
1691 */
1692int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1693{
1694	int rc;
1695
1696	if (rxq < 1 || rxq > dev->num_rx_queues)
1697		return -EINVAL;
1698
1699	if (dev->reg_state == NETREG_REGISTERED) {
1700		ASSERT_RTNL();
1701
1702		rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1703						  rxq);
1704		if (rc)
1705			return rc;
1706	}
1707
1708	dev->real_num_rx_queues = rxq;
1709	return 0;
1710}
1711EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1712#endif
1713
1714static inline void __netif_reschedule(struct Qdisc *q)
1715{
1716	struct softnet_data *sd;
1717	unsigned long flags;
1718
1719	local_irq_save(flags);
1720	sd = &__get_cpu_var(softnet_data);
1721	q->next_sched = NULL;
1722	*sd->output_queue_tailp = q;
1723	sd->output_queue_tailp = &q->next_sched;
1724	raise_softirq_irqoff(NET_TX_SOFTIRQ);
1725	local_irq_restore(flags);
1726}
1727
1728void __netif_schedule(struct Qdisc *q)
1729{
1730	if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1731		__netif_reschedule(q);
1732}
1733EXPORT_SYMBOL(__netif_schedule);
1734
1735void dev_kfree_skb_irq(struct sk_buff *skb)
1736{
1737	if (atomic_dec_and_test(&skb->users)) {
1738		struct softnet_data *sd;
1739		unsigned long flags;
1740
1741		local_irq_save(flags);
1742		sd = &__get_cpu_var(softnet_data);
1743		skb->next = sd->completion_queue;
1744		sd->completion_queue = skb;
1745		raise_softirq_irqoff(NET_TX_SOFTIRQ);
1746		local_irq_restore(flags);
1747	}
1748}
1749EXPORT_SYMBOL(dev_kfree_skb_irq);
1750
1751void dev_kfree_skb_any(struct sk_buff *skb)
1752{
1753	if (in_irq() || irqs_disabled())
1754		dev_kfree_skb_irq(skb);
1755	else
1756		dev_kfree_skb(skb);
1757}
1758EXPORT_SYMBOL(dev_kfree_skb_any);
1759
1760
1761/**
1762 * netif_device_detach - mark device as removed
1763 * @dev: network device
1764 *
1765 * Mark device as removed from system and therefore no longer available.
1766 */
1767void netif_device_detach(struct net_device *dev)
1768{
1769	if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1770	    netif_running(dev)) {
1771		netif_tx_stop_all_queues(dev);
1772	}
1773}
1774EXPORT_SYMBOL(netif_device_detach);
1775
1776/**
1777 * netif_device_attach - mark device as attached
1778 * @dev: network device
1779 *
1780 * Mark device as attached from system and restart if needed.
1781 */
1782void netif_device_attach(struct net_device *dev)
1783{
1784	if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1785	    netif_running(dev)) {
1786		netif_tx_wake_all_queues(dev);
1787		__netdev_watchdog_up(dev);
1788	}
1789}
1790EXPORT_SYMBOL(netif_device_attach);
1791
1792/**
1793 * skb_dev_set -- assign a new device to a buffer
1794 * @skb: buffer for the new device
1795 * @dev: network device
1796 *
1797 * If an skb is owned by a device already, we have to reset
1798 * all data private to the namespace a device belongs to
1799 * before assigning it a new device.
1800 */
1801#ifdef CONFIG_NET_NS
1802void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1803{
1804	skb_dst_drop(skb);
1805	if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1806		secpath_reset(skb);
1807		nf_reset(skb);
1808		skb_init_secmark(skb);
1809		skb->mark = 0;
1810		skb->priority = 0;
1811		skb->nf_trace = 0;
1812		skb->ipvs_property = 0;
1813#ifdef CONFIG_NET_SCHED
1814		skb->tc_index = 0;
1815#endif
1816	}
1817	skb->dev = dev;
1818}
1819EXPORT_SYMBOL(skb_set_dev);
1820#endif /* CONFIG_NET_NS */
1821
1822/*
1823 * Invalidate hardware checksum when packet is to be mangled, and
1824 * complete checksum manually on outgoing path.
1825 */
1826int skb_checksum_help(struct sk_buff *skb)
1827{
1828	__wsum csum;
1829	int ret = 0, offset;
1830
1831	if (skb->ip_summed == CHECKSUM_COMPLETE)
1832		goto out_set_summed;
1833
1834	if (unlikely(skb_shinfo(skb)->gso_size)) {
1835		/* Let GSO fix up the checksum. */
1836		goto out_set_summed;
1837	}
1838
1839	offset = skb_checksum_start_offset(skb);
1840	BUG_ON(offset >= skb_headlen(skb));
1841	csum = skb_checksum(skb, offset, skb->len - offset, 0);
1842
1843	offset += skb->csum_offset;
1844	BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1845
1846	if (skb_cloned(skb) &&
1847	    !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1848		ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1849		if (ret)
1850			goto out;
1851	}
1852
1853	*(__sum16 *)(skb->data + offset) = csum_fold(csum);
1854out_set_summed:
1855	skb->ip_summed = CHECKSUM_NONE;
1856out:
1857	return ret;
1858}
1859EXPORT_SYMBOL(skb_checksum_help);
1860
1861/**
1862 *	skb_gso_segment - Perform segmentation on skb.
1863 *	@skb: buffer to segment
1864 *	@features: features for the output path (see dev->features)
1865 *
1866 *	This function segments the given skb and returns a list of segments.
1867 *
1868 *	It may return NULL if the skb requires no segmentation.  This is
1869 *	only possible when GSO is used for verifying header integrity.
1870 */
1871struct sk_buff *skb_gso_segment(struct sk_buff *skb, u32 features)
 
1872{
1873	struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1874	struct packet_type *ptype;
1875	__be16 type = skb->protocol;
1876	int vlan_depth = ETH_HLEN;
1877	int err;
1878
1879	while (type == htons(ETH_P_8021Q)) {
1880		struct vlan_hdr *vh;
1881
1882		if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1883			return ERR_PTR(-EINVAL);
1884
1885		vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1886		type = vh->h_vlan_encapsulated_proto;
1887		vlan_depth += VLAN_HLEN;
1888	}
1889
1890	skb_reset_mac_header(skb);
1891	skb->mac_len = skb->network_header - skb->mac_header;
1892	__skb_pull(skb, skb->mac_len);
1893
1894	if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1895		struct net_device *dev = skb->dev;
1896		struct ethtool_drvinfo info = {};
1897
1898		if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1899			dev->ethtool_ops->get_drvinfo(dev, &info);
1900
1901		WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d ip_summed=%d\n",
1902		     info.driver, dev ? dev->features : 0L,
1903		     skb->sk ? skb->sk->sk_route_caps : 0L,
1904		     skb->len, skb->data_len, skb->ip_summed);
1905
1906		if (skb_header_cloned(skb) &&
1907		    (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1908			return ERR_PTR(err);
1909	}
1910
1911	rcu_read_lock();
1912	list_for_each_entry_rcu(ptype,
1913			&ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1914		if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1915			if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1916				err = ptype->gso_send_check(skb);
1917				segs = ERR_PTR(err);
1918				if (err || skb_gso_ok(skb, features))
1919					break;
1920				__skb_push(skb, (skb->data -
1921						 skb_network_header(skb)));
1922			}
1923			segs = ptype->gso_segment(skb, features);
1924			break;
1925		}
1926	}
1927	rcu_read_unlock();
1928
1929	__skb_push(skb, skb->data - skb_mac_header(skb));
1930
1931	return segs;
1932}
1933EXPORT_SYMBOL(skb_gso_segment);
1934
1935/* Take action when hardware reception checksum errors are detected. */
1936#ifdef CONFIG_BUG
1937void netdev_rx_csum_fault(struct net_device *dev)
1938{
1939	if (net_ratelimit()) {
1940		printk(KERN_ERR "%s: hw csum failure.\n",
1941			dev ? dev->name : "<unknown>");
1942		dump_stack();
1943	}
1944}
1945EXPORT_SYMBOL(netdev_rx_csum_fault);
1946#endif
1947
1948/* Actually, we should eliminate this check as soon as we know, that:
1949 * 1. IOMMU is present and allows to map all the memory.
1950 * 2. No high memory really exists on this machine.
1951 */
1952
1953static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1954{
1955#ifdef CONFIG_HIGHMEM
1956	int i;
1957	if (!(dev->features & NETIF_F_HIGHDMA)) {
1958		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1959			if (PageHighMem(skb_shinfo(skb)->frags[i].page))
 
1960				return 1;
 
1961	}
1962
1963	if (PCI_DMA_BUS_IS_PHYS) {
1964		struct device *pdev = dev->dev.parent;
1965
1966		if (!pdev)
1967			return 0;
1968		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1969			dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
 
1970			if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1971				return 1;
1972		}
1973	}
1974#endif
1975	return 0;
1976}
1977
1978struct dev_gso_cb {
1979	void (*destructor)(struct sk_buff *skb);
1980};
1981
1982#define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1983
1984static void dev_gso_skb_destructor(struct sk_buff *skb)
1985{
1986	struct dev_gso_cb *cb;
1987
1988	do {
1989		struct sk_buff *nskb = skb->next;
1990
1991		skb->next = nskb->next;
1992		nskb->next = NULL;
1993		kfree_skb(nskb);
1994	} while (skb->next);
1995
1996	cb = DEV_GSO_CB(skb);
1997	if (cb->destructor)
1998		cb->destructor(skb);
1999}
2000
2001/**
2002 *	dev_gso_segment - Perform emulated hardware segmentation on skb.
2003 *	@skb: buffer to segment
2004 *	@features: device features as applicable to this skb
2005 *
2006 *	This function segments the given skb and stores the list of segments
2007 *	in skb->next.
2008 */
2009static int dev_gso_segment(struct sk_buff *skb, int features)
2010{
2011	struct sk_buff *segs;
2012
2013	segs = skb_gso_segment(skb, features);
2014
2015	/* Verifying header integrity only. */
2016	if (!segs)
2017		return 0;
2018
2019	if (IS_ERR(segs))
2020		return PTR_ERR(segs);
2021
2022	skb->next = segs;
2023	DEV_GSO_CB(skb)->destructor = skb->destructor;
2024	skb->destructor = dev_gso_skb_destructor;
2025
2026	return 0;
2027}
2028
2029/*
2030 * Try to orphan skb early, right before transmission by the device.
2031 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2032 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2033 */
2034static inline void skb_orphan_try(struct sk_buff *skb)
2035{
2036	struct sock *sk = skb->sk;
2037
2038	if (sk && !skb_shinfo(skb)->tx_flags) {
2039		/* skb_tx_hash() wont be able to get sk.
2040		 * We copy sk_hash into skb->rxhash
2041		 */
2042		if (!skb->rxhash)
2043			skb->rxhash = sk->sk_hash;
2044		skb_orphan(skb);
2045	}
2046}
2047
2048static bool can_checksum_protocol(unsigned long features, __be16 protocol)
2049{
2050	return ((features & NETIF_F_GEN_CSUM) ||
2051		((features & NETIF_F_V4_CSUM) &&
2052		 protocol == htons(ETH_P_IP)) ||
2053		((features & NETIF_F_V6_CSUM) &&
2054		 protocol == htons(ETH_P_IPV6)) ||
2055		((features & NETIF_F_FCOE_CRC) &&
2056		 protocol == htons(ETH_P_FCOE)));
2057}
2058
2059static u32 harmonize_features(struct sk_buff *skb, __be16 protocol, u32 features)
 
2060{
2061	if (!can_checksum_protocol(features, protocol)) {
2062		features &= ~NETIF_F_ALL_CSUM;
2063		features &= ~NETIF_F_SG;
2064	} else if (illegal_highdma(skb->dev, skb)) {
2065		features &= ~NETIF_F_SG;
2066	}
2067
2068	return features;
2069}
2070
2071u32 netif_skb_features(struct sk_buff *skb)
2072{
2073	__be16 protocol = skb->protocol;
2074	u32 features = skb->dev->features;
 
 
 
2075
2076	if (protocol == htons(ETH_P_8021Q)) {
2077		struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2078		protocol = veh->h_vlan_encapsulated_proto;
2079	} else if (!vlan_tx_tag_present(skb)) {
2080		return harmonize_features(skb, protocol, features);
2081	}
2082
2083	features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2084
2085	if (protocol != htons(ETH_P_8021Q)) {
2086		return harmonize_features(skb, protocol, features);
2087	} else {
2088		features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2089				NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2090		return harmonize_features(skb, protocol, features);
2091	}
2092}
2093EXPORT_SYMBOL(netif_skb_features);
2094
2095/*
2096 * Returns true if either:
2097 *	1. skb has frag_list and the device doesn't support FRAGLIST, or
2098 *	2. skb is fragmented and the device does not support SG, or if
2099 *	   at least one of fragments is in highmem and device does not
2100 *	   support DMA from it.
2101 */
2102static inline int skb_needs_linearize(struct sk_buff *skb,
2103				      int features)
2104{
2105	return skb_is_nonlinear(skb) &&
2106			((skb_has_frag_list(skb) &&
2107				!(features & NETIF_F_FRAGLIST)) ||
2108			(skb_shinfo(skb)->nr_frags &&
2109				!(features & NETIF_F_SG)));
2110}
2111
2112int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2113			struct netdev_queue *txq)
2114{
2115	const struct net_device_ops *ops = dev->netdev_ops;
2116	int rc = NETDEV_TX_OK;
2117	unsigned int skb_len;
2118
2119	if (likely(!skb->next)) {
2120		u32 features;
2121
2122		/*
2123		 * If device doesn't need skb->dst, release it right now while
2124		 * its hot in this cpu cache
2125		 */
2126		if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2127			skb_dst_drop(skb);
2128
2129		if (!list_empty(&ptype_all))
2130			dev_queue_xmit_nit(skb, dev);
2131
2132		skb_orphan_try(skb);
2133
2134		features = netif_skb_features(skb);
2135
2136		if (vlan_tx_tag_present(skb) &&
2137		    !(features & NETIF_F_HW_VLAN_TX)) {
2138			skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2139			if (unlikely(!skb))
2140				goto out;
2141
2142			skb->vlan_tci = 0;
2143		}
2144
2145		if (netif_needs_gso(skb, features)) {
2146			if (unlikely(dev_gso_segment(skb, features)))
2147				goto out_kfree_skb;
2148			if (skb->next)
2149				goto gso;
2150		} else {
2151			if (skb_needs_linearize(skb, features) &&
2152			    __skb_linearize(skb))
2153				goto out_kfree_skb;
2154
2155			/* If packet is not checksummed and device does not
2156			 * support checksumming for this protocol, complete
2157			 * checksumming here.
2158			 */
2159			if (skb->ip_summed == CHECKSUM_PARTIAL) {
2160				skb_set_transport_header(skb,
2161					skb_checksum_start_offset(skb));
2162				if (!(features & NETIF_F_ALL_CSUM) &&
2163				     skb_checksum_help(skb))
2164					goto out_kfree_skb;
2165			}
2166		}
2167
2168		skb_len = skb->len;
2169		rc = ops->ndo_start_xmit(skb, dev);
2170		trace_net_dev_xmit(skb, rc, dev, skb_len);
2171		if (rc == NETDEV_TX_OK)
2172			txq_trans_update(txq);
2173		return rc;
2174	}
2175
2176gso:
2177	do {
2178		struct sk_buff *nskb = skb->next;
2179
2180		skb->next = nskb->next;
2181		nskb->next = NULL;
2182
2183		/*
2184		 * If device doesn't need nskb->dst, release it right now while
2185		 * its hot in this cpu cache
2186		 */
2187		if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2188			skb_dst_drop(nskb);
2189
2190		skb_len = nskb->len;
2191		rc = ops->ndo_start_xmit(nskb, dev);
2192		trace_net_dev_xmit(nskb, rc, dev, skb_len);
2193		if (unlikely(rc != NETDEV_TX_OK)) {
2194			if (rc & ~NETDEV_TX_MASK)
2195				goto out_kfree_gso_skb;
2196			nskb->next = skb->next;
2197			skb->next = nskb;
2198			return rc;
2199		}
2200		txq_trans_update(txq);
2201		if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
2202			return NETDEV_TX_BUSY;
2203	} while (skb->next);
2204
2205out_kfree_gso_skb:
2206	if (likely(skb->next == NULL))
2207		skb->destructor = DEV_GSO_CB(skb)->destructor;
2208out_kfree_skb:
2209	kfree_skb(skb);
2210out:
2211	return rc;
2212}
2213
2214static u32 hashrnd __read_mostly;
2215
2216/*
2217 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2218 * to be used as a distribution range.
2219 */
2220u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2221		  unsigned int num_tx_queues)
2222{
2223	u32 hash;
2224	u16 qoffset = 0;
2225	u16 qcount = num_tx_queues;
2226
2227	if (skb_rx_queue_recorded(skb)) {
2228		hash = skb_get_rx_queue(skb);
2229		while (unlikely(hash >= num_tx_queues))
2230			hash -= num_tx_queues;
2231		return hash;
2232	}
2233
2234	if (dev->num_tc) {
2235		u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2236		qoffset = dev->tc_to_txq[tc].offset;
2237		qcount = dev->tc_to_txq[tc].count;
2238	}
2239
2240	if (skb->sk && skb->sk->sk_hash)
2241		hash = skb->sk->sk_hash;
2242	else
2243		hash = (__force u16) skb->protocol ^ skb->rxhash;
2244	hash = jhash_1word(hash, hashrnd);
2245
2246	return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2247}
2248EXPORT_SYMBOL(__skb_tx_hash);
2249
2250static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2251{
2252	if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2253		if (net_ratelimit()) {
2254			pr_warning("%s selects TX queue %d, but "
2255				"real number of TX queues is %d\n",
2256				dev->name, queue_index, dev->real_num_tx_queues);
2257		}
2258		return 0;
2259	}
2260	return queue_index;
2261}
2262
2263static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2264{
2265#ifdef CONFIG_XPS
2266	struct xps_dev_maps *dev_maps;
2267	struct xps_map *map;
2268	int queue_index = -1;
2269
2270	rcu_read_lock();
2271	dev_maps = rcu_dereference(dev->xps_maps);
2272	if (dev_maps) {
2273		map = rcu_dereference(
2274		    dev_maps->cpu_map[raw_smp_processor_id()]);
2275		if (map) {
2276			if (map->len == 1)
2277				queue_index = map->queues[0];
2278			else {
2279				u32 hash;
2280				if (skb->sk && skb->sk->sk_hash)
2281					hash = skb->sk->sk_hash;
2282				else
2283					hash = (__force u16) skb->protocol ^
2284					    skb->rxhash;
2285				hash = jhash_1word(hash, hashrnd);
2286				queue_index = map->queues[
2287				    ((u64)hash * map->len) >> 32];
2288			}
2289			if (unlikely(queue_index >= dev->real_num_tx_queues))
2290				queue_index = -1;
2291		}
2292	}
2293	rcu_read_unlock();
2294
2295	return queue_index;
2296#else
2297	return -1;
2298#endif
2299}
2300
2301static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2302					struct sk_buff *skb)
2303{
2304	int queue_index;
2305	const struct net_device_ops *ops = dev->netdev_ops;
2306
2307	if (dev->real_num_tx_queues == 1)
2308		queue_index = 0;
2309	else if (ops->ndo_select_queue) {
2310		queue_index = ops->ndo_select_queue(dev, skb);
2311		queue_index = dev_cap_txqueue(dev, queue_index);
2312	} else {
2313		struct sock *sk = skb->sk;
2314		queue_index = sk_tx_queue_get(sk);
2315
2316		if (queue_index < 0 || skb->ooo_okay ||
2317		    queue_index >= dev->real_num_tx_queues) {
2318			int old_index = queue_index;
2319
2320			queue_index = get_xps_queue(dev, skb);
2321			if (queue_index < 0)
2322				queue_index = skb_tx_hash(dev, skb);
2323
2324			if (queue_index != old_index && sk) {
2325				struct dst_entry *dst =
2326				    rcu_dereference_check(sk->sk_dst_cache, 1);
2327
2328				if (dst && skb_dst(skb) == dst)
2329					sk_tx_queue_set(sk, queue_index);
2330			}
2331		}
2332	}
2333
2334	skb_set_queue_mapping(skb, queue_index);
2335	return netdev_get_tx_queue(dev, queue_index);
2336}
2337
2338static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2339				 struct net_device *dev,
2340				 struct netdev_queue *txq)
2341{
2342	spinlock_t *root_lock = qdisc_lock(q);
2343	bool contended;
2344	int rc;
2345
2346	qdisc_skb_cb(skb)->pkt_len = skb->len;
2347	qdisc_calculate_pkt_len(skb, q);
2348	/*
2349	 * Heuristic to force contended enqueues to serialize on a
2350	 * separate lock before trying to get qdisc main lock.
2351	 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2352	 * and dequeue packets faster.
2353	 */
2354	contended = qdisc_is_running(q);
2355	if (unlikely(contended))
2356		spin_lock(&q->busylock);
2357
2358	spin_lock(root_lock);
2359	if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2360		kfree_skb(skb);
2361		rc = NET_XMIT_DROP;
2362	} else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2363		   qdisc_run_begin(q)) {
2364		/*
2365		 * This is a work-conserving queue; there are no old skbs
2366		 * waiting to be sent out; and the qdisc is not running -
2367		 * xmit the skb directly.
2368		 */
2369		if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2370			skb_dst_force(skb);
2371
2372		qdisc_bstats_update(q, skb);
2373
2374		if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2375			if (unlikely(contended)) {
2376				spin_unlock(&q->busylock);
2377				contended = false;
2378			}
2379			__qdisc_run(q);
2380		} else
2381			qdisc_run_end(q);
2382
2383		rc = NET_XMIT_SUCCESS;
2384	} else {
2385		skb_dst_force(skb);
2386		rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2387		if (qdisc_run_begin(q)) {
2388			if (unlikely(contended)) {
2389				spin_unlock(&q->busylock);
2390				contended = false;
2391			}
2392			__qdisc_run(q);
2393		}
2394	}
2395	spin_unlock(root_lock);
2396	if (unlikely(contended))
2397		spin_unlock(&q->busylock);
2398	return rc;
2399}
2400
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2401static DEFINE_PER_CPU(int, xmit_recursion);
2402#define RECURSION_LIMIT 10
2403
2404/**
2405 *	dev_queue_xmit - transmit a buffer
2406 *	@skb: buffer to transmit
2407 *
2408 *	Queue a buffer for transmission to a network device. The caller must
2409 *	have set the device and priority and built the buffer before calling
2410 *	this function. The function can be called from an interrupt.
2411 *
2412 *	A negative errno code is returned on a failure. A success does not
2413 *	guarantee the frame will be transmitted as it may be dropped due
2414 *	to congestion or traffic shaping.
2415 *
2416 * -----------------------------------------------------------------------------------
2417 *      I notice this method can also return errors from the queue disciplines,
2418 *      including NET_XMIT_DROP, which is a positive value.  So, errors can also
2419 *      be positive.
2420 *
2421 *      Regardless of the return value, the skb is consumed, so it is currently
2422 *      difficult to retry a send to this method.  (You can bump the ref count
2423 *      before sending to hold a reference for retry if you are careful.)
2424 *
2425 *      When calling this method, interrupts MUST be enabled.  This is because
2426 *      the BH enable code must have IRQs enabled so that it will not deadlock.
2427 *          --BLG
2428 */
2429int dev_queue_xmit(struct sk_buff *skb)
2430{
2431	struct net_device *dev = skb->dev;
2432	struct netdev_queue *txq;
2433	struct Qdisc *q;
2434	int rc = -ENOMEM;
2435
2436	/* Disable soft irqs for various locks below. Also
2437	 * stops preemption for RCU.
2438	 */
2439	rcu_read_lock_bh();
2440
 
 
2441	txq = dev_pick_tx(dev, skb);
2442	q = rcu_dereference_bh(txq->qdisc);
2443
2444#ifdef CONFIG_NET_CLS_ACT
2445	skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2446#endif
2447	trace_net_dev_queue(skb);
2448	if (q->enqueue) {
2449		rc = __dev_xmit_skb(skb, q, dev, txq);
2450		goto out;
2451	}
2452
2453	/* The device has no queue. Common case for software devices:
2454	   loopback, all the sorts of tunnels...
2455
2456	   Really, it is unlikely that netif_tx_lock protection is necessary
2457	   here.  (f.e. loopback and IP tunnels are clean ignoring statistics
2458	   counters.)
2459	   However, it is possible, that they rely on protection
2460	   made by us here.
2461
2462	   Check this and shot the lock. It is not prone from deadlocks.
2463	   Either shot noqueue qdisc, it is even simpler 8)
2464	 */
2465	if (dev->flags & IFF_UP) {
2466		int cpu = smp_processor_id(); /* ok because BHs are off */
2467
2468		if (txq->xmit_lock_owner != cpu) {
2469
2470			if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2471				goto recursion_alert;
2472
2473			HARD_TX_LOCK(dev, txq, cpu);
2474
2475			if (!netif_tx_queue_stopped(txq)) {
2476				__this_cpu_inc(xmit_recursion);
2477				rc = dev_hard_start_xmit(skb, dev, txq);
2478				__this_cpu_dec(xmit_recursion);
2479				if (dev_xmit_complete(rc)) {
2480					HARD_TX_UNLOCK(dev, txq);
2481					goto out;
2482				}
2483			}
2484			HARD_TX_UNLOCK(dev, txq);
2485			if (net_ratelimit())
2486				printk(KERN_CRIT "Virtual device %s asks to "
2487				       "queue packet!\n", dev->name);
2488		} else {
2489			/* Recursion is detected! It is possible,
2490			 * unfortunately
2491			 */
2492recursion_alert:
2493			if (net_ratelimit())
2494				printk(KERN_CRIT "Dead loop on virtual device "
2495				       "%s, fix it urgently!\n", dev->name);
2496		}
2497	}
2498
2499	rc = -ENETDOWN;
2500	rcu_read_unlock_bh();
2501
2502	kfree_skb(skb);
2503	return rc;
2504out:
2505	rcu_read_unlock_bh();
2506	return rc;
2507}
2508EXPORT_SYMBOL(dev_queue_xmit);
2509
2510
2511/*=======================================================================
2512			Receiver routines
2513  =======================================================================*/
2514
2515int netdev_max_backlog __read_mostly = 1000;
2516int netdev_tstamp_prequeue __read_mostly = 1;
2517int netdev_budget __read_mostly = 300;
2518int weight_p __read_mostly = 64;            /* old backlog weight */
2519
2520/* Called with irq disabled */
2521static inline void ____napi_schedule(struct softnet_data *sd,
2522				     struct napi_struct *napi)
2523{
2524	list_add_tail(&napi->poll_list, &sd->poll_list);
2525	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
2526}
2527
2528/*
2529 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2530 * and src/dst port numbers. Returns a non-zero hash number on success
2531 * and 0 on failure.
 
2532 */
2533__u32 __skb_get_rxhash(struct sk_buff *skb)
2534{
2535	int nhoff, hash = 0, poff;
2536	const struct ipv6hdr *ip6;
2537	const struct iphdr *ip;
2538	u8 ip_proto;
2539	u32 addr1, addr2, ihl;
2540	union {
2541		u32 v32;
2542		u16 v16[2];
2543	} ports;
2544
2545	nhoff = skb_network_offset(skb);
2546
2547	switch (skb->protocol) {
2548	case __constant_htons(ETH_P_IP):
2549		if (!pskb_may_pull(skb, sizeof(*ip) + nhoff))
2550			goto done;
2551
2552		ip = (const struct iphdr *) (skb->data + nhoff);
2553		if (ip_is_fragment(ip))
2554			ip_proto = 0;
2555		else
2556			ip_proto = ip->protocol;
2557		addr1 = (__force u32) ip->saddr;
2558		addr2 = (__force u32) ip->daddr;
2559		ihl = ip->ihl;
2560		break;
2561	case __constant_htons(ETH_P_IPV6):
2562		if (!pskb_may_pull(skb, sizeof(*ip6) + nhoff))
2563			goto done;
2564
2565		ip6 = (const struct ipv6hdr *) (skb->data + nhoff);
2566		ip_proto = ip6->nexthdr;
2567		addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2568		addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2569		ihl = (40 >> 2);
2570		break;
2571	default:
2572		goto done;
2573	}
2574
2575	ports.v32 = 0;
2576	poff = proto_ports_offset(ip_proto);
2577	if (poff >= 0) {
2578		nhoff += ihl * 4 + poff;
2579		if (pskb_may_pull(skb, nhoff + 4)) {
2580			ports.v32 = * (__force u32 *) (skb->data + nhoff);
2581			if (ports.v16[1] < ports.v16[0])
2582				swap(ports.v16[0], ports.v16[1]);
2583		}
2584	}
2585
2586	/* get a consistent hash (same value on both flow directions) */
2587	if (addr2 < addr1)
2588		swap(addr1, addr2);
2589
2590	hash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
 
 
2591	if (!hash)
2592		hash = 1;
2593
2594done:
2595	return hash;
2596}
2597EXPORT_SYMBOL(__skb_get_rxhash);
2598
2599#ifdef CONFIG_RPS
2600
2601/* One global table that all flow-based protocols share. */
2602struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2603EXPORT_SYMBOL(rps_sock_flow_table);
2604
 
 
2605static struct rps_dev_flow *
2606set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2607	    struct rps_dev_flow *rflow, u16 next_cpu)
2608{
2609	u16 tcpu;
2610
2611	tcpu = rflow->cpu = next_cpu;
2612	if (tcpu != RPS_NO_CPU) {
2613#ifdef CONFIG_RFS_ACCEL
2614		struct netdev_rx_queue *rxqueue;
2615		struct rps_dev_flow_table *flow_table;
2616		struct rps_dev_flow *old_rflow;
2617		u32 flow_id;
2618		u16 rxq_index;
2619		int rc;
2620
2621		/* Should we steer this flow to a different hardware queue? */
2622		if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2623		    !(dev->features & NETIF_F_NTUPLE))
2624			goto out;
2625		rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2626		if (rxq_index == skb_get_rx_queue(skb))
2627			goto out;
2628
2629		rxqueue = dev->_rx + rxq_index;
2630		flow_table = rcu_dereference(rxqueue->rps_flow_table);
2631		if (!flow_table)
2632			goto out;
2633		flow_id = skb->rxhash & flow_table->mask;
2634		rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2635							rxq_index, flow_id);
2636		if (rc < 0)
2637			goto out;
2638		old_rflow = rflow;
2639		rflow = &flow_table->flows[flow_id];
2640		rflow->cpu = next_cpu;
2641		rflow->filter = rc;
2642		if (old_rflow->filter == rflow->filter)
2643			old_rflow->filter = RPS_NO_FILTER;
2644	out:
2645#endif
2646		rflow->last_qtail =
2647			per_cpu(softnet_data, tcpu).input_queue_head;
2648	}
2649
 
2650	return rflow;
2651}
2652
2653/*
2654 * get_rps_cpu is called from netif_receive_skb and returns the target
2655 * CPU from the RPS map of the receiving queue for a given skb.
2656 * rcu_read_lock must be held on entry.
2657 */
2658static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2659		       struct rps_dev_flow **rflowp)
2660{
2661	struct netdev_rx_queue *rxqueue;
2662	struct rps_map *map;
2663	struct rps_dev_flow_table *flow_table;
2664	struct rps_sock_flow_table *sock_flow_table;
2665	int cpu = -1;
2666	u16 tcpu;
2667
2668	if (skb_rx_queue_recorded(skb)) {
2669		u16 index = skb_get_rx_queue(skb);
2670		if (unlikely(index >= dev->real_num_rx_queues)) {
2671			WARN_ONCE(dev->real_num_rx_queues > 1,
2672				  "%s received packet on queue %u, but number "
2673				  "of RX queues is %u\n",
2674				  dev->name, index, dev->real_num_rx_queues);
2675			goto done;
2676		}
2677		rxqueue = dev->_rx + index;
2678	} else
2679		rxqueue = dev->_rx;
2680
2681	map = rcu_dereference(rxqueue->rps_map);
2682	if (map) {
2683		if (map->len == 1 &&
2684		    !rcu_dereference_raw(rxqueue->rps_flow_table)) {
2685			tcpu = map->cpus[0];
2686			if (cpu_online(tcpu))
2687				cpu = tcpu;
2688			goto done;
2689		}
2690	} else if (!rcu_dereference_raw(rxqueue->rps_flow_table)) {
2691		goto done;
2692	}
2693
2694	skb_reset_network_header(skb);
2695	if (!skb_get_rxhash(skb))
2696		goto done;
2697
2698	flow_table = rcu_dereference(rxqueue->rps_flow_table);
2699	sock_flow_table = rcu_dereference(rps_sock_flow_table);
2700	if (flow_table && sock_flow_table) {
2701		u16 next_cpu;
2702		struct rps_dev_flow *rflow;
2703
2704		rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2705		tcpu = rflow->cpu;
2706
2707		next_cpu = sock_flow_table->ents[skb->rxhash &
2708		    sock_flow_table->mask];
2709
2710		/*
2711		 * If the desired CPU (where last recvmsg was done) is
2712		 * different from current CPU (one in the rx-queue flow
2713		 * table entry), switch if one of the following holds:
2714		 *   - Current CPU is unset (equal to RPS_NO_CPU).
2715		 *   - Current CPU is offline.
2716		 *   - The current CPU's queue tail has advanced beyond the
2717		 *     last packet that was enqueued using this table entry.
2718		 *     This guarantees that all previous packets for the flow
2719		 *     have been dequeued, thus preserving in order delivery.
2720		 */
2721		if (unlikely(tcpu != next_cpu) &&
2722		    (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2723		     ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2724		      rflow->last_qtail)) >= 0))
2725			rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2726
2727		if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2728			*rflowp = rflow;
2729			cpu = tcpu;
2730			goto done;
2731		}
2732	}
2733
2734	if (map) {
2735		tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2736
2737		if (cpu_online(tcpu)) {
2738			cpu = tcpu;
2739			goto done;
2740		}
2741	}
2742
2743done:
2744	return cpu;
2745}
2746
2747#ifdef CONFIG_RFS_ACCEL
2748
2749/**
2750 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2751 * @dev: Device on which the filter was set
2752 * @rxq_index: RX queue index
2753 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2754 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2755 *
2756 * Drivers that implement ndo_rx_flow_steer() should periodically call
2757 * this function for each installed filter and remove the filters for
2758 * which it returns %true.
2759 */
2760bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2761			 u32 flow_id, u16 filter_id)
2762{
2763	struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2764	struct rps_dev_flow_table *flow_table;
2765	struct rps_dev_flow *rflow;
2766	bool expire = true;
2767	int cpu;
2768
2769	rcu_read_lock();
2770	flow_table = rcu_dereference(rxqueue->rps_flow_table);
2771	if (flow_table && flow_id <= flow_table->mask) {
2772		rflow = &flow_table->flows[flow_id];
2773		cpu = ACCESS_ONCE(rflow->cpu);
2774		if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2775		    ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2776			   rflow->last_qtail) <
2777		     (int)(10 * flow_table->mask)))
2778			expire = false;
2779	}
2780	rcu_read_unlock();
2781	return expire;
2782}
2783EXPORT_SYMBOL(rps_may_expire_flow);
2784
2785#endif /* CONFIG_RFS_ACCEL */
2786
2787/* Called from hardirq (IPI) context */
2788static void rps_trigger_softirq(void *data)
2789{
2790	struct softnet_data *sd = data;
2791
2792	____napi_schedule(sd, &sd->backlog);
2793	sd->received_rps++;
2794}
2795
2796#endif /* CONFIG_RPS */
2797
2798/*
2799 * Check if this softnet_data structure is another cpu one
2800 * If yes, queue it to our IPI list and return 1
2801 * If no, return 0
2802 */
2803static int rps_ipi_queued(struct softnet_data *sd)
2804{
2805#ifdef CONFIG_RPS
2806	struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2807
2808	if (sd != mysd) {
2809		sd->rps_ipi_next = mysd->rps_ipi_list;
2810		mysd->rps_ipi_list = sd;
2811
2812		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
2813		return 1;
2814	}
2815#endif /* CONFIG_RPS */
2816	return 0;
2817}
2818
2819/*
2820 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2821 * queue (may be a remote CPU queue).
2822 */
2823static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2824			      unsigned int *qtail)
2825{
2826	struct softnet_data *sd;
2827	unsigned long flags;
2828
2829	sd = &per_cpu(softnet_data, cpu);
2830
2831	local_irq_save(flags);
2832
2833	rps_lock(sd);
2834	if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2835		if (skb_queue_len(&sd->input_pkt_queue)) {
2836enqueue:
2837			__skb_queue_tail(&sd->input_pkt_queue, skb);
2838			input_queue_tail_incr_save(sd, qtail);
2839			rps_unlock(sd);
2840			local_irq_restore(flags);
2841			return NET_RX_SUCCESS;
2842		}
2843
2844		/* Schedule NAPI for backlog device
2845		 * We can use non atomic operation since we own the queue lock
2846		 */
2847		if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2848			if (!rps_ipi_queued(sd))
2849				____napi_schedule(sd, &sd->backlog);
2850		}
2851		goto enqueue;
2852	}
2853
2854	sd->dropped++;
2855	rps_unlock(sd);
2856
2857	local_irq_restore(flags);
2858
2859	atomic_long_inc(&skb->dev->rx_dropped);
2860	kfree_skb(skb);
2861	return NET_RX_DROP;
2862}
2863
2864/**
2865 *	netif_rx	-	post buffer to the network code
2866 *	@skb: buffer to post
2867 *
2868 *	This function receives a packet from a device driver and queues it for
2869 *	the upper (protocol) levels to process.  It always succeeds. The buffer
2870 *	may be dropped during processing for congestion control or by the
2871 *	protocol layers.
2872 *
2873 *	return values:
2874 *	NET_RX_SUCCESS	(no congestion)
2875 *	NET_RX_DROP     (packet was dropped)
2876 *
2877 */
2878
2879int netif_rx(struct sk_buff *skb)
2880{
2881	int ret;
2882
2883	/* if netpoll wants it, pretend we never saw it */
2884	if (netpoll_rx(skb))
2885		return NET_RX_DROP;
2886
2887	if (netdev_tstamp_prequeue)
2888		net_timestamp_check(skb);
2889
2890	trace_netif_rx(skb);
2891#ifdef CONFIG_RPS
2892	{
2893		struct rps_dev_flow voidflow, *rflow = &voidflow;
2894		int cpu;
2895
2896		preempt_disable();
2897		rcu_read_lock();
2898
2899		cpu = get_rps_cpu(skb->dev, skb, &rflow);
2900		if (cpu < 0)
2901			cpu = smp_processor_id();
2902
2903		ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2904
2905		rcu_read_unlock();
2906		preempt_enable();
2907	}
2908#else
2909	{
2910		unsigned int qtail;
2911		ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2912		put_cpu();
2913	}
2914#endif
2915	return ret;
2916}
2917EXPORT_SYMBOL(netif_rx);
2918
2919int netif_rx_ni(struct sk_buff *skb)
2920{
2921	int err;
2922
2923	preempt_disable();
2924	err = netif_rx(skb);
2925	if (local_softirq_pending())
2926		do_softirq();
2927	preempt_enable();
2928
2929	return err;
2930}
2931EXPORT_SYMBOL(netif_rx_ni);
2932
2933static void net_tx_action(struct softirq_action *h)
2934{
2935	struct softnet_data *sd = &__get_cpu_var(softnet_data);
2936
2937	if (sd->completion_queue) {
2938		struct sk_buff *clist;
2939
2940		local_irq_disable();
2941		clist = sd->completion_queue;
2942		sd->completion_queue = NULL;
2943		local_irq_enable();
2944
2945		while (clist) {
2946			struct sk_buff *skb = clist;
2947			clist = clist->next;
2948
2949			WARN_ON(atomic_read(&skb->users));
2950			trace_kfree_skb(skb, net_tx_action);
2951			__kfree_skb(skb);
2952		}
2953	}
2954
2955	if (sd->output_queue) {
2956		struct Qdisc *head;
2957
2958		local_irq_disable();
2959		head = sd->output_queue;
2960		sd->output_queue = NULL;
2961		sd->output_queue_tailp = &sd->output_queue;
2962		local_irq_enable();
2963
2964		while (head) {
2965			struct Qdisc *q = head;
2966			spinlock_t *root_lock;
2967
2968			head = head->next_sched;
2969
2970			root_lock = qdisc_lock(q);
2971			if (spin_trylock(root_lock)) {
2972				smp_mb__before_clear_bit();
2973				clear_bit(__QDISC_STATE_SCHED,
2974					  &q->state);
2975				qdisc_run(q);
2976				spin_unlock(root_lock);
2977			} else {
2978				if (!test_bit(__QDISC_STATE_DEACTIVATED,
2979					      &q->state)) {
2980					__netif_reschedule(q);
2981				} else {
2982					smp_mb__before_clear_bit();
2983					clear_bit(__QDISC_STATE_SCHED,
2984						  &q->state);
2985				}
2986			}
2987		}
2988	}
2989}
2990
2991#if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
2992    (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
2993/* This hook is defined here for ATM LANE */
2994int (*br_fdb_test_addr_hook)(struct net_device *dev,
2995			     unsigned char *addr) __read_mostly;
2996EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2997#endif
2998
2999#ifdef CONFIG_NET_CLS_ACT
3000/* TODO: Maybe we should just force sch_ingress to be compiled in
3001 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3002 * a compare and 2 stores extra right now if we dont have it on
3003 * but have CONFIG_NET_CLS_ACT
3004 * NOTE: This doesn't stop any functionality; if you dont have
3005 * the ingress scheduler, you just can't add policies on ingress.
3006 *
3007 */
3008static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3009{
3010	struct net_device *dev = skb->dev;
3011	u32 ttl = G_TC_RTTL(skb->tc_verd);
3012	int result = TC_ACT_OK;
3013	struct Qdisc *q;
3014
3015	if (unlikely(MAX_RED_LOOP < ttl++)) {
3016		if (net_ratelimit())
3017			pr_warning( "Redir loop detected Dropping packet (%d->%d)\n",
3018			       skb->skb_iif, dev->ifindex);
3019		return TC_ACT_SHOT;
3020	}
3021
3022	skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3023	skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3024
3025	q = rxq->qdisc;
3026	if (q != &noop_qdisc) {
3027		spin_lock(qdisc_lock(q));
3028		if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3029			result = qdisc_enqueue_root(skb, q);
3030		spin_unlock(qdisc_lock(q));
3031	}
3032
3033	return result;
3034}
3035
3036static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3037					 struct packet_type **pt_prev,
3038					 int *ret, struct net_device *orig_dev)
3039{
3040	struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3041
3042	if (!rxq || rxq->qdisc == &noop_qdisc)
3043		goto out;
3044
3045	if (*pt_prev) {
3046		*ret = deliver_skb(skb, *pt_prev, orig_dev);
3047		*pt_prev = NULL;
3048	}
3049
3050	switch (ing_filter(skb, rxq)) {
3051	case TC_ACT_SHOT:
3052	case TC_ACT_STOLEN:
3053		kfree_skb(skb);
3054		return NULL;
3055	}
3056
3057out:
3058	skb->tc_verd = 0;
3059	return skb;
3060}
3061#endif
3062
3063/**
3064 *	netdev_rx_handler_register - register receive handler
3065 *	@dev: device to register a handler for
3066 *	@rx_handler: receive handler to register
3067 *	@rx_handler_data: data pointer that is used by rx handler
3068 *
3069 *	Register a receive hander for a device. This handler will then be
3070 *	called from __netif_receive_skb. A negative errno code is returned
3071 *	on a failure.
3072 *
3073 *	The caller must hold the rtnl_mutex.
3074 *
3075 *	For a general description of rx_handler, see enum rx_handler_result.
3076 */
3077int netdev_rx_handler_register(struct net_device *dev,
3078			       rx_handler_func_t *rx_handler,
3079			       void *rx_handler_data)
3080{
3081	ASSERT_RTNL();
3082
3083	if (dev->rx_handler)
3084		return -EBUSY;
3085
3086	rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3087	rcu_assign_pointer(dev->rx_handler, rx_handler);
3088
3089	return 0;
3090}
3091EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3092
3093/**
3094 *	netdev_rx_handler_unregister - unregister receive handler
3095 *	@dev: device to unregister a handler from
3096 *
3097 *	Unregister a receive hander from a device.
3098 *
3099 *	The caller must hold the rtnl_mutex.
3100 */
3101void netdev_rx_handler_unregister(struct net_device *dev)
3102{
3103
3104	ASSERT_RTNL();
3105	rcu_assign_pointer(dev->rx_handler, NULL);
3106	rcu_assign_pointer(dev->rx_handler_data, NULL);
3107}
3108EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3109
3110static int __netif_receive_skb(struct sk_buff *skb)
3111{
3112	struct packet_type *ptype, *pt_prev;
3113	rx_handler_func_t *rx_handler;
3114	struct net_device *orig_dev;
3115	struct net_device *null_or_dev;
3116	bool deliver_exact = false;
3117	int ret = NET_RX_DROP;
3118	__be16 type;
3119
3120	if (!netdev_tstamp_prequeue)
3121		net_timestamp_check(skb);
3122
3123	trace_netif_receive_skb(skb);
3124
3125	/* if we've gotten here through NAPI, check netpoll */
3126	if (netpoll_receive_skb(skb))
3127		return NET_RX_DROP;
3128
3129	if (!skb->skb_iif)
3130		skb->skb_iif = skb->dev->ifindex;
3131	orig_dev = skb->dev;
3132
3133	skb_reset_network_header(skb);
3134	skb_reset_transport_header(skb);
3135	skb_reset_mac_len(skb);
3136
3137	pt_prev = NULL;
3138
3139	rcu_read_lock();
3140
3141another_round:
3142
3143	__this_cpu_inc(softnet_data.processed);
3144
3145	if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3146		skb = vlan_untag(skb);
3147		if (unlikely(!skb))
3148			goto out;
3149	}
3150
3151#ifdef CONFIG_NET_CLS_ACT
3152	if (skb->tc_verd & TC_NCLS) {
3153		skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3154		goto ncls;
3155	}
3156#endif
3157
3158	list_for_each_entry_rcu(ptype, &ptype_all, list) {
3159		if (!ptype->dev || ptype->dev == skb->dev) {
3160			if (pt_prev)
3161				ret = deliver_skb(skb, pt_prev, orig_dev);
3162			pt_prev = ptype;
3163		}
3164	}
3165
3166#ifdef CONFIG_NET_CLS_ACT
3167	skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3168	if (!skb)
3169		goto out;
3170ncls:
3171#endif
3172
3173	rx_handler = rcu_dereference(skb->dev->rx_handler);
 
 
 
 
 
 
 
 
 
 
 
3174	if (rx_handler) {
3175		if (pt_prev) {
3176			ret = deliver_skb(skb, pt_prev, orig_dev);
3177			pt_prev = NULL;
3178		}
3179		switch (rx_handler(&skb)) {
3180		case RX_HANDLER_CONSUMED:
3181			goto out;
3182		case RX_HANDLER_ANOTHER:
3183			goto another_round;
3184		case RX_HANDLER_EXACT:
3185			deliver_exact = true;
3186		case RX_HANDLER_PASS:
3187			break;
3188		default:
3189			BUG();
3190		}
3191	}
3192
3193	if (vlan_tx_tag_present(skb)) {
3194		if (pt_prev) {
3195			ret = deliver_skb(skb, pt_prev, orig_dev);
3196			pt_prev = NULL;
3197		}
3198		if (vlan_do_receive(&skb)) {
3199			ret = __netif_receive_skb(skb);
3200			goto out;
3201		} else if (unlikely(!skb))
3202			goto out;
3203	}
3204
3205	/* deliver only exact match when indicated */
3206	null_or_dev = deliver_exact ? skb->dev : NULL;
3207
3208	type = skb->protocol;
3209	list_for_each_entry_rcu(ptype,
3210			&ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3211		if (ptype->type == type &&
3212		    (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3213		     ptype->dev == orig_dev)) {
3214			if (pt_prev)
3215				ret = deliver_skb(skb, pt_prev, orig_dev);
3216			pt_prev = ptype;
3217		}
3218	}
3219
3220	if (pt_prev) {
3221		ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3222	} else {
3223		atomic_long_inc(&skb->dev->rx_dropped);
3224		kfree_skb(skb);
3225		/* Jamal, now you will not able to escape explaining
3226		 * me how you were going to use this. :-)
3227		 */
3228		ret = NET_RX_DROP;
3229	}
3230
3231out:
3232	rcu_read_unlock();
3233	return ret;
3234}
3235
3236/**
3237 *	netif_receive_skb - process receive buffer from network
3238 *	@skb: buffer to process
3239 *
3240 *	netif_receive_skb() is the main receive data processing function.
3241 *	It always succeeds. The buffer may be dropped during processing
3242 *	for congestion control or by the protocol layers.
3243 *
3244 *	This function may only be called from softirq context and interrupts
3245 *	should be enabled.
3246 *
3247 *	Return values (usually ignored):
3248 *	NET_RX_SUCCESS: no congestion
3249 *	NET_RX_DROP: packet was dropped
3250 */
3251int netif_receive_skb(struct sk_buff *skb)
3252{
3253	if (netdev_tstamp_prequeue)
3254		net_timestamp_check(skb);
3255
3256	if (skb_defer_rx_timestamp(skb))
3257		return NET_RX_SUCCESS;
3258
3259#ifdef CONFIG_RPS
3260	{
3261		struct rps_dev_flow voidflow, *rflow = &voidflow;
3262		int cpu, ret;
3263
3264		rcu_read_lock();
3265
3266		cpu = get_rps_cpu(skb->dev, skb, &rflow);
3267
3268		if (cpu >= 0) {
3269			ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3270			rcu_read_unlock();
3271		} else {
3272			rcu_read_unlock();
3273			ret = __netif_receive_skb(skb);
3274		}
3275
3276		return ret;
3277	}
3278#else
3279	return __netif_receive_skb(skb);
3280#endif
3281}
3282EXPORT_SYMBOL(netif_receive_skb);
3283
3284/* Network device is going away, flush any packets still pending
3285 * Called with irqs disabled.
3286 */
3287static void flush_backlog(void *arg)
3288{
3289	struct net_device *dev = arg;
3290	struct softnet_data *sd = &__get_cpu_var(softnet_data);
3291	struct sk_buff *skb, *tmp;
3292
3293	rps_lock(sd);
3294	skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3295		if (skb->dev == dev) {
3296			__skb_unlink(skb, &sd->input_pkt_queue);
3297			kfree_skb(skb);
3298			input_queue_head_incr(sd);
3299		}
3300	}
3301	rps_unlock(sd);
3302
3303	skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3304		if (skb->dev == dev) {
3305			__skb_unlink(skb, &sd->process_queue);
3306			kfree_skb(skb);
3307			input_queue_head_incr(sd);
3308		}
3309	}
3310}
3311
3312static int napi_gro_complete(struct sk_buff *skb)
3313{
3314	struct packet_type *ptype;
3315	__be16 type = skb->protocol;
3316	struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3317	int err = -ENOENT;
3318
3319	if (NAPI_GRO_CB(skb)->count == 1) {
3320		skb_shinfo(skb)->gso_size = 0;
3321		goto out;
3322	}
3323
3324	rcu_read_lock();
3325	list_for_each_entry_rcu(ptype, head, list) {
3326		if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3327			continue;
3328
3329		err = ptype->gro_complete(skb);
3330		break;
3331	}
3332	rcu_read_unlock();
3333
3334	if (err) {
3335		WARN_ON(&ptype->list == head);
3336		kfree_skb(skb);
3337		return NET_RX_SUCCESS;
3338	}
3339
3340out:
3341	return netif_receive_skb(skb);
3342}
3343
3344inline void napi_gro_flush(struct napi_struct *napi)
3345{
3346	struct sk_buff *skb, *next;
3347
3348	for (skb = napi->gro_list; skb; skb = next) {
3349		next = skb->next;
3350		skb->next = NULL;
3351		napi_gro_complete(skb);
3352	}
3353
3354	napi->gro_count = 0;
3355	napi->gro_list = NULL;
3356}
3357EXPORT_SYMBOL(napi_gro_flush);
3358
3359enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3360{
3361	struct sk_buff **pp = NULL;
3362	struct packet_type *ptype;
3363	__be16 type = skb->protocol;
3364	struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3365	int same_flow;
3366	int mac_len;
3367	enum gro_result ret;
3368
3369	if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3370		goto normal;
3371
3372	if (skb_is_gso(skb) || skb_has_frag_list(skb))
3373		goto normal;
3374
3375	rcu_read_lock();
3376	list_for_each_entry_rcu(ptype, head, list) {
3377		if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3378			continue;
3379
3380		skb_set_network_header(skb, skb_gro_offset(skb));
3381		mac_len = skb->network_header - skb->mac_header;
3382		skb->mac_len = mac_len;
3383		NAPI_GRO_CB(skb)->same_flow = 0;
3384		NAPI_GRO_CB(skb)->flush = 0;
3385		NAPI_GRO_CB(skb)->free = 0;
3386
3387		pp = ptype->gro_receive(&napi->gro_list, skb);
3388		break;
3389	}
3390	rcu_read_unlock();
3391
3392	if (&ptype->list == head)
3393		goto normal;
3394
3395	same_flow = NAPI_GRO_CB(skb)->same_flow;
3396	ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3397
3398	if (pp) {
3399		struct sk_buff *nskb = *pp;
3400
3401		*pp = nskb->next;
3402		nskb->next = NULL;
3403		napi_gro_complete(nskb);
3404		napi->gro_count--;
3405	}
3406
3407	if (same_flow)
3408		goto ok;
3409
3410	if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3411		goto normal;
3412
3413	napi->gro_count++;
3414	NAPI_GRO_CB(skb)->count = 1;
3415	skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3416	skb->next = napi->gro_list;
3417	napi->gro_list = skb;
3418	ret = GRO_HELD;
3419
3420pull:
3421	if (skb_headlen(skb) < skb_gro_offset(skb)) {
3422		int grow = skb_gro_offset(skb) - skb_headlen(skb);
3423
3424		BUG_ON(skb->end - skb->tail < grow);
3425
3426		memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3427
3428		skb->tail += grow;
3429		skb->data_len -= grow;
3430
3431		skb_shinfo(skb)->frags[0].page_offset += grow;
3432		skb_shinfo(skb)->frags[0].size -= grow;
3433
3434		if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3435			put_page(skb_shinfo(skb)->frags[0].page);
3436			memmove(skb_shinfo(skb)->frags,
3437				skb_shinfo(skb)->frags + 1,
3438				--skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3439		}
3440	}
3441
3442ok:
3443	return ret;
3444
3445normal:
3446	ret = GRO_NORMAL;
3447	goto pull;
3448}
3449EXPORT_SYMBOL(dev_gro_receive);
3450
3451static inline gro_result_t
3452__napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3453{
3454	struct sk_buff *p;
 
3455
3456	for (p = napi->gro_list; p; p = p->next) {
3457		unsigned long diffs;
3458
3459		diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3460		diffs |= p->vlan_tci ^ skb->vlan_tci;
3461		diffs |= compare_ether_header(skb_mac_header(p),
3462					      skb_gro_mac_header(skb));
 
 
 
 
 
3463		NAPI_GRO_CB(p)->same_flow = !diffs;
3464		NAPI_GRO_CB(p)->flush = 0;
3465	}
3466
3467	return dev_gro_receive(napi, skb);
3468}
3469
3470gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3471{
3472	switch (ret) {
3473	case GRO_NORMAL:
3474		if (netif_receive_skb(skb))
3475			ret = GRO_DROP;
3476		break;
3477
3478	case GRO_DROP:
 
 
 
3479	case GRO_MERGED_FREE:
3480		kfree_skb(skb);
 
 
 
3481		break;
3482
3483	case GRO_HELD:
3484	case GRO_MERGED:
3485		break;
3486	}
3487
3488	return ret;
3489}
3490EXPORT_SYMBOL(napi_skb_finish);
3491
3492void skb_gro_reset_offset(struct sk_buff *skb)
3493{
3494	NAPI_GRO_CB(skb)->data_offset = 0;
3495	NAPI_GRO_CB(skb)->frag0 = NULL;
3496	NAPI_GRO_CB(skb)->frag0_len = 0;
3497
3498	if (skb->mac_header == skb->tail &&
3499	    !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3500		NAPI_GRO_CB(skb)->frag0 =
3501			page_address(skb_shinfo(skb)->frags[0].page) +
3502			skb_shinfo(skb)->frags[0].page_offset;
3503		NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3504	}
3505}
3506EXPORT_SYMBOL(skb_gro_reset_offset);
3507
3508gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3509{
3510	skb_gro_reset_offset(skb);
3511
3512	return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3513}
3514EXPORT_SYMBOL(napi_gro_receive);
3515
3516static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3517{
3518	__skb_pull(skb, skb_headlen(skb));
3519	skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
 
3520	skb->vlan_tci = 0;
3521	skb->dev = napi->dev;
3522	skb->skb_iif = 0;
3523
3524	napi->skb = skb;
3525}
3526
3527struct sk_buff *napi_get_frags(struct napi_struct *napi)
3528{
3529	struct sk_buff *skb = napi->skb;
3530
3531	if (!skb) {
3532		skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3533		if (skb)
3534			napi->skb = skb;
3535	}
3536	return skb;
3537}
3538EXPORT_SYMBOL(napi_get_frags);
3539
3540gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3541			       gro_result_t ret)
3542{
3543	switch (ret) {
3544	case GRO_NORMAL:
3545	case GRO_HELD:
3546		skb->protocol = eth_type_trans(skb, skb->dev);
3547
3548		if (ret == GRO_HELD)
3549			skb_gro_pull(skb, -ETH_HLEN);
3550		else if (netif_receive_skb(skb))
3551			ret = GRO_DROP;
3552		break;
3553
3554	case GRO_DROP:
3555	case GRO_MERGED_FREE:
3556		napi_reuse_skb(napi, skb);
3557		break;
3558
3559	case GRO_MERGED:
3560		break;
3561	}
3562
3563	return ret;
3564}
3565EXPORT_SYMBOL(napi_frags_finish);
3566
3567struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3568{
3569	struct sk_buff *skb = napi->skb;
3570	struct ethhdr *eth;
3571	unsigned int hlen;
3572	unsigned int off;
3573
3574	napi->skb = NULL;
3575
3576	skb_reset_mac_header(skb);
3577	skb_gro_reset_offset(skb);
3578
3579	off = skb_gro_offset(skb);
3580	hlen = off + sizeof(*eth);
3581	eth = skb_gro_header_fast(skb, off);
3582	if (skb_gro_header_hard(skb, hlen)) {
3583		eth = skb_gro_header_slow(skb, hlen, off);
3584		if (unlikely(!eth)) {
3585			napi_reuse_skb(napi, skb);
3586			skb = NULL;
3587			goto out;
3588		}
3589	}
3590
3591	skb_gro_pull(skb, sizeof(*eth));
3592
3593	/*
3594	 * This works because the only protocols we care about don't require
3595	 * special handling.  We'll fix it up properly at the end.
3596	 */
3597	skb->protocol = eth->h_proto;
3598
3599out:
3600	return skb;
3601}
3602EXPORT_SYMBOL(napi_frags_skb);
3603
3604gro_result_t napi_gro_frags(struct napi_struct *napi)
3605{
3606	struct sk_buff *skb = napi_frags_skb(napi);
3607
3608	if (!skb)
3609		return GRO_DROP;
3610
3611	return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3612}
3613EXPORT_SYMBOL(napi_gro_frags);
3614
3615/*
3616 * net_rps_action sends any pending IPI's for rps.
3617 * Note: called with local irq disabled, but exits with local irq enabled.
3618 */
3619static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3620{
3621#ifdef CONFIG_RPS
3622	struct softnet_data *remsd = sd->rps_ipi_list;
3623
3624	if (remsd) {
3625		sd->rps_ipi_list = NULL;
3626
3627		local_irq_enable();
3628
3629		/* Send pending IPI's to kick RPS processing on remote cpus. */
3630		while (remsd) {
3631			struct softnet_data *next = remsd->rps_ipi_next;
3632
3633			if (cpu_online(remsd->cpu))
3634				__smp_call_function_single(remsd->cpu,
3635							   &remsd->csd, 0);
3636			remsd = next;
3637		}
3638	} else
3639#endif
3640		local_irq_enable();
3641}
3642
3643static int process_backlog(struct napi_struct *napi, int quota)
3644{
3645	int work = 0;
3646	struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3647
3648#ifdef CONFIG_RPS
3649	/* Check if we have pending ipi, its better to send them now,
3650	 * not waiting net_rx_action() end.
3651	 */
3652	if (sd->rps_ipi_list) {
3653		local_irq_disable();
3654		net_rps_action_and_irq_enable(sd);
3655	}
3656#endif
3657	napi->weight = weight_p;
3658	local_irq_disable();
3659	while (work < quota) {
3660		struct sk_buff *skb;
3661		unsigned int qlen;
3662
3663		while ((skb = __skb_dequeue(&sd->process_queue))) {
3664			local_irq_enable();
3665			__netif_receive_skb(skb);
3666			local_irq_disable();
3667			input_queue_head_incr(sd);
3668			if (++work >= quota) {
3669				local_irq_enable();
3670				return work;
3671			}
3672		}
3673
3674		rps_lock(sd);
3675		qlen = skb_queue_len(&sd->input_pkt_queue);
3676		if (qlen)
3677			skb_queue_splice_tail_init(&sd->input_pkt_queue,
3678						   &sd->process_queue);
3679
3680		if (qlen < quota - work) {
3681			/*
3682			 * Inline a custom version of __napi_complete().
3683			 * only current cpu owns and manipulates this napi,
3684			 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3685			 * we can use a plain write instead of clear_bit(),
3686			 * and we dont need an smp_mb() memory barrier.
3687			 */
3688			list_del(&napi->poll_list);
3689			napi->state = 0;
3690
3691			quota = work + qlen;
3692		}
3693		rps_unlock(sd);
3694	}
3695	local_irq_enable();
3696
3697	return work;
3698}
3699
3700/**
3701 * __napi_schedule - schedule for receive
3702 * @n: entry to schedule
3703 *
3704 * The entry's receive function will be scheduled to run
3705 */
3706void __napi_schedule(struct napi_struct *n)
3707{
3708	unsigned long flags;
3709
3710	local_irq_save(flags);
3711	____napi_schedule(&__get_cpu_var(softnet_data), n);
3712	local_irq_restore(flags);
3713}
3714EXPORT_SYMBOL(__napi_schedule);
3715
3716void __napi_complete(struct napi_struct *n)
3717{
3718	BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3719	BUG_ON(n->gro_list);
3720
3721	list_del(&n->poll_list);
3722	smp_mb__before_clear_bit();
3723	clear_bit(NAPI_STATE_SCHED, &n->state);
3724}
3725EXPORT_SYMBOL(__napi_complete);
3726
3727void napi_complete(struct napi_struct *n)
3728{
3729	unsigned long flags;
3730
3731	/*
3732	 * don't let napi dequeue from the cpu poll list
3733	 * just in case its running on a different cpu
3734	 */
3735	if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3736		return;
3737
3738	napi_gro_flush(n);
3739	local_irq_save(flags);
3740	__napi_complete(n);
3741	local_irq_restore(flags);
3742}
3743EXPORT_SYMBOL(napi_complete);
3744
3745void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3746		    int (*poll)(struct napi_struct *, int), int weight)
3747{
3748	INIT_LIST_HEAD(&napi->poll_list);
3749	napi->gro_count = 0;
3750	napi->gro_list = NULL;
3751	napi->skb = NULL;
3752	napi->poll = poll;
3753	napi->weight = weight;
3754	list_add(&napi->dev_list, &dev->napi_list);
3755	napi->dev = dev;
3756#ifdef CONFIG_NETPOLL
3757	spin_lock_init(&napi->poll_lock);
3758	napi->poll_owner = -1;
3759#endif
3760	set_bit(NAPI_STATE_SCHED, &napi->state);
3761}
3762EXPORT_SYMBOL(netif_napi_add);
3763
3764void netif_napi_del(struct napi_struct *napi)
3765{
3766	struct sk_buff *skb, *next;
3767
3768	list_del_init(&napi->dev_list);
3769	napi_free_frags(napi);
3770
3771	for (skb = napi->gro_list; skb; skb = next) {
3772		next = skb->next;
3773		skb->next = NULL;
3774		kfree_skb(skb);
3775	}
3776
3777	napi->gro_list = NULL;
3778	napi->gro_count = 0;
3779}
3780EXPORT_SYMBOL(netif_napi_del);
3781
3782static void net_rx_action(struct softirq_action *h)
3783{
3784	struct softnet_data *sd = &__get_cpu_var(softnet_data);
3785	unsigned long time_limit = jiffies + 2;
3786	int budget = netdev_budget;
3787	void *have;
3788
3789	local_irq_disable();
3790
3791	while (!list_empty(&sd->poll_list)) {
3792		struct napi_struct *n;
3793		int work, weight;
3794
3795		/* If softirq window is exhuasted then punt.
3796		 * Allow this to run for 2 jiffies since which will allow
3797		 * an average latency of 1.5/HZ.
3798		 */
3799		if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3800			goto softnet_break;
3801
3802		local_irq_enable();
3803
3804		/* Even though interrupts have been re-enabled, this
3805		 * access is safe because interrupts can only add new
3806		 * entries to the tail of this list, and only ->poll()
3807		 * calls can remove this head entry from the list.
3808		 */
3809		n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3810
3811		have = netpoll_poll_lock(n);
3812
3813		weight = n->weight;
3814
3815		/* This NAPI_STATE_SCHED test is for avoiding a race
3816		 * with netpoll's poll_napi().  Only the entity which
3817		 * obtains the lock and sees NAPI_STATE_SCHED set will
3818		 * actually make the ->poll() call.  Therefore we avoid
3819		 * accidentally calling ->poll() when NAPI is not scheduled.
3820		 */
3821		work = 0;
3822		if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3823			work = n->poll(n, weight);
3824			trace_napi_poll(n);
3825		}
3826
3827		WARN_ON_ONCE(work > weight);
3828
3829		budget -= work;
3830
3831		local_irq_disable();
3832
3833		/* Drivers must not modify the NAPI state if they
3834		 * consume the entire weight.  In such cases this code
3835		 * still "owns" the NAPI instance and therefore can
3836		 * move the instance around on the list at-will.
3837		 */
3838		if (unlikely(work == weight)) {
3839			if (unlikely(napi_disable_pending(n))) {
3840				local_irq_enable();
3841				napi_complete(n);
3842				local_irq_disable();
3843			} else
3844				list_move_tail(&n->poll_list, &sd->poll_list);
3845		}
3846
3847		netpoll_poll_unlock(have);
3848	}
3849out:
3850	net_rps_action_and_irq_enable(sd);
3851
3852#ifdef CONFIG_NET_DMA
3853	/*
3854	 * There may not be any more sk_buffs coming right now, so push
3855	 * any pending DMA copies to hardware
3856	 */
3857	dma_issue_pending_all();
3858#endif
3859
3860	return;
3861
3862softnet_break:
3863	sd->time_squeeze++;
3864	__raise_softirq_irqoff(NET_RX_SOFTIRQ);
3865	goto out;
3866}
3867
3868static gifconf_func_t *gifconf_list[NPROTO];
3869
3870/**
3871 *	register_gifconf	-	register a SIOCGIF handler
3872 *	@family: Address family
3873 *	@gifconf: Function handler
3874 *
3875 *	Register protocol dependent address dumping routines. The handler
3876 *	that is passed must not be freed or reused until it has been replaced
3877 *	by another handler.
3878 */
3879int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3880{
3881	if (family >= NPROTO)
3882		return -EINVAL;
3883	gifconf_list[family] = gifconf;
3884	return 0;
3885}
3886EXPORT_SYMBOL(register_gifconf);
3887
3888
3889/*
3890 *	Map an interface index to its name (SIOCGIFNAME)
3891 */
3892
3893/*
3894 *	We need this ioctl for efficient implementation of the
3895 *	if_indextoname() function required by the IPv6 API.  Without
3896 *	it, we would have to search all the interfaces to find a
3897 *	match.  --pb
3898 */
3899
3900static int dev_ifname(struct net *net, struct ifreq __user *arg)
3901{
3902	struct net_device *dev;
3903	struct ifreq ifr;
3904
3905	/*
3906	 *	Fetch the caller's info block.
3907	 */
3908
3909	if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3910		return -EFAULT;
3911
3912	rcu_read_lock();
3913	dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3914	if (!dev) {
3915		rcu_read_unlock();
3916		return -ENODEV;
3917	}
3918
3919	strcpy(ifr.ifr_name, dev->name);
3920	rcu_read_unlock();
3921
3922	if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3923		return -EFAULT;
3924	return 0;
3925}
3926
3927/*
3928 *	Perform a SIOCGIFCONF call. This structure will change
3929 *	size eventually, and there is nothing I can do about it.
3930 *	Thus we will need a 'compatibility mode'.
3931 */
3932
3933static int dev_ifconf(struct net *net, char __user *arg)
3934{
3935	struct ifconf ifc;
3936	struct net_device *dev;
3937	char __user *pos;
3938	int len;
3939	int total;
3940	int i;
3941
3942	/*
3943	 *	Fetch the caller's info block.
3944	 */
3945
3946	if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3947		return -EFAULT;
3948
3949	pos = ifc.ifc_buf;
3950	len = ifc.ifc_len;
3951
3952	/*
3953	 *	Loop over the interfaces, and write an info block for each.
3954	 */
3955
3956	total = 0;
3957	for_each_netdev(net, dev) {
3958		for (i = 0; i < NPROTO; i++) {
3959			if (gifconf_list[i]) {
3960				int done;
3961				if (!pos)
3962					done = gifconf_list[i](dev, NULL, 0);
3963				else
3964					done = gifconf_list[i](dev, pos + total,
3965							       len - total);
3966				if (done < 0)
3967					return -EFAULT;
3968				total += done;
3969			}
3970		}
3971	}
3972
3973	/*
3974	 *	All done.  Write the updated control block back to the caller.
3975	 */
3976	ifc.ifc_len = total;
3977
3978	/*
3979	 * 	Both BSD and Solaris return 0 here, so we do too.
3980	 */
3981	return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3982}
3983
3984#ifdef CONFIG_PROC_FS
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3985/*
3986 *	This is invoked by the /proc filesystem handler to display a device
3987 *	in detail.
3988 */
3989void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3990	__acquires(RCU)
3991{
3992	struct net *net = seq_file_net(seq);
3993	loff_t off;
3994	struct net_device *dev;
3995
3996	rcu_read_lock();
3997	if (!*pos)
3998		return SEQ_START_TOKEN;
3999
4000	off = 1;
4001	for_each_netdev_rcu(net, dev)
4002		if (off++ == *pos)
4003			return dev;
4004
4005	return NULL;
4006}
4007
4008void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4009{
4010	struct net_device *dev = v;
4011
4012	if (v == SEQ_START_TOKEN)
4013		dev = first_net_device_rcu(seq_file_net(seq));
4014	else
4015		dev = next_net_device_rcu(dev);
4016
4017	++*pos;
4018	return dev;
4019}
4020
4021void dev_seq_stop(struct seq_file *seq, void *v)
4022	__releases(RCU)
4023{
4024	rcu_read_unlock();
4025}
4026
4027static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4028{
4029	struct rtnl_link_stats64 temp;
4030	const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4031
4032	seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4033		   "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4034		   dev->name, stats->rx_bytes, stats->rx_packets,
4035		   stats->rx_errors,
4036		   stats->rx_dropped + stats->rx_missed_errors,
4037		   stats->rx_fifo_errors,
4038		   stats->rx_length_errors + stats->rx_over_errors +
4039		    stats->rx_crc_errors + stats->rx_frame_errors,
4040		   stats->rx_compressed, stats->multicast,
4041		   stats->tx_bytes, stats->tx_packets,
4042		   stats->tx_errors, stats->tx_dropped,
4043		   stats->tx_fifo_errors, stats->collisions,
4044		   stats->tx_carrier_errors +
4045		    stats->tx_aborted_errors +
4046		    stats->tx_window_errors +
4047		    stats->tx_heartbeat_errors,
4048		   stats->tx_compressed);
4049}
4050
4051/*
4052 *	Called from the PROCfs module. This now uses the new arbitrary sized
4053 *	/proc/net interface to create /proc/net/dev
4054 */
4055static int dev_seq_show(struct seq_file *seq, void *v)
4056{
4057	if (v == SEQ_START_TOKEN)
4058		seq_puts(seq, "Inter-|   Receive                            "
4059			      "                    |  Transmit\n"
4060			      " face |bytes    packets errs drop fifo frame "
4061			      "compressed multicast|bytes    packets errs "
4062			      "drop fifo colls carrier compressed\n");
4063	else
4064		dev_seq_printf_stats(seq, v);
4065	return 0;
4066}
4067
4068static struct softnet_data *softnet_get_online(loff_t *pos)
4069{
4070	struct softnet_data *sd = NULL;
4071
4072	while (*pos < nr_cpu_ids)
4073		if (cpu_online(*pos)) {
4074			sd = &per_cpu(softnet_data, *pos);
4075			break;
4076		} else
4077			++*pos;
4078	return sd;
4079}
4080
4081static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4082{
4083	return softnet_get_online(pos);
4084}
4085
4086static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4087{
4088	++*pos;
4089	return softnet_get_online(pos);
4090}
4091
4092static void softnet_seq_stop(struct seq_file *seq, void *v)
4093{
4094}
4095
4096static int softnet_seq_show(struct seq_file *seq, void *v)
4097{
4098	struct softnet_data *sd = v;
4099
4100	seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4101		   sd->processed, sd->dropped, sd->time_squeeze, 0,
4102		   0, 0, 0, 0, /* was fastroute */
4103		   sd->cpu_collision, sd->received_rps);
4104	return 0;
4105}
4106
4107static const struct seq_operations dev_seq_ops = {
4108	.start = dev_seq_start,
4109	.next  = dev_seq_next,
4110	.stop  = dev_seq_stop,
4111	.show  = dev_seq_show,
4112};
4113
4114static int dev_seq_open(struct inode *inode, struct file *file)
4115{
4116	return seq_open_net(inode, file, &dev_seq_ops,
4117			    sizeof(struct seq_net_private));
4118}
4119
4120static const struct file_operations dev_seq_fops = {
4121	.owner	 = THIS_MODULE,
4122	.open    = dev_seq_open,
4123	.read    = seq_read,
4124	.llseek  = seq_lseek,
4125	.release = seq_release_net,
4126};
4127
4128static const struct seq_operations softnet_seq_ops = {
4129	.start = softnet_seq_start,
4130	.next  = softnet_seq_next,
4131	.stop  = softnet_seq_stop,
4132	.show  = softnet_seq_show,
4133};
4134
4135static int softnet_seq_open(struct inode *inode, struct file *file)
4136{
4137	return seq_open(file, &softnet_seq_ops);
4138}
4139
4140static const struct file_operations softnet_seq_fops = {
4141	.owner	 = THIS_MODULE,
4142	.open    = softnet_seq_open,
4143	.read    = seq_read,
4144	.llseek  = seq_lseek,
4145	.release = seq_release,
4146};
4147
4148static void *ptype_get_idx(loff_t pos)
4149{
4150	struct packet_type *pt = NULL;
4151	loff_t i = 0;
4152	int t;
4153
4154	list_for_each_entry_rcu(pt, &ptype_all, list) {
4155		if (i == pos)
4156			return pt;
4157		++i;
4158	}
4159
4160	for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4161		list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4162			if (i == pos)
4163				return pt;
4164			++i;
4165		}
4166	}
4167	return NULL;
4168}
4169
4170static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4171	__acquires(RCU)
4172{
4173	rcu_read_lock();
4174	return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4175}
4176
4177static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4178{
4179	struct packet_type *pt;
4180	struct list_head *nxt;
4181	int hash;
4182
4183	++*pos;
4184	if (v == SEQ_START_TOKEN)
4185		return ptype_get_idx(0);
4186
4187	pt = v;
4188	nxt = pt->list.next;
4189	if (pt->type == htons(ETH_P_ALL)) {
4190		if (nxt != &ptype_all)
4191			goto found;
4192		hash = 0;
4193		nxt = ptype_base[0].next;
4194	} else
4195		hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4196
4197	while (nxt == &ptype_base[hash]) {
4198		if (++hash >= PTYPE_HASH_SIZE)
4199			return NULL;
4200		nxt = ptype_base[hash].next;
4201	}
4202found:
4203	return list_entry(nxt, struct packet_type, list);
4204}
4205
4206static void ptype_seq_stop(struct seq_file *seq, void *v)
4207	__releases(RCU)
4208{
4209	rcu_read_unlock();
4210}
4211
4212static int ptype_seq_show(struct seq_file *seq, void *v)
4213{
4214	struct packet_type *pt = v;
4215
4216	if (v == SEQ_START_TOKEN)
4217		seq_puts(seq, "Type Device      Function\n");
4218	else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4219		if (pt->type == htons(ETH_P_ALL))
4220			seq_puts(seq, "ALL ");
4221		else
4222			seq_printf(seq, "%04x", ntohs(pt->type));
4223
4224		seq_printf(seq, " %-8s %pF\n",
4225			   pt->dev ? pt->dev->name : "", pt->func);
4226	}
4227
4228	return 0;
4229}
4230
4231static const struct seq_operations ptype_seq_ops = {
4232	.start = ptype_seq_start,
4233	.next  = ptype_seq_next,
4234	.stop  = ptype_seq_stop,
4235	.show  = ptype_seq_show,
4236};
4237
4238static int ptype_seq_open(struct inode *inode, struct file *file)
4239{
4240	return seq_open_net(inode, file, &ptype_seq_ops,
4241			sizeof(struct seq_net_private));
4242}
4243
4244static const struct file_operations ptype_seq_fops = {
4245	.owner	 = THIS_MODULE,
4246	.open    = ptype_seq_open,
4247	.read    = seq_read,
4248	.llseek  = seq_lseek,
4249	.release = seq_release_net,
4250};
4251
4252
4253static int __net_init dev_proc_net_init(struct net *net)
4254{
4255	int rc = -ENOMEM;
4256
4257	if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4258		goto out;
4259	if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4260		goto out_dev;
4261	if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4262		goto out_softnet;
4263
4264	if (wext_proc_init(net))
4265		goto out_ptype;
4266	rc = 0;
4267out:
4268	return rc;
4269out_ptype:
4270	proc_net_remove(net, "ptype");
4271out_softnet:
4272	proc_net_remove(net, "softnet_stat");
4273out_dev:
4274	proc_net_remove(net, "dev");
4275	goto out;
4276}
4277
4278static void __net_exit dev_proc_net_exit(struct net *net)
4279{
4280	wext_proc_exit(net);
4281
4282	proc_net_remove(net, "ptype");
4283	proc_net_remove(net, "softnet_stat");
4284	proc_net_remove(net, "dev");
4285}
4286
4287static struct pernet_operations __net_initdata dev_proc_ops = {
4288	.init = dev_proc_net_init,
4289	.exit = dev_proc_net_exit,
4290};
4291
4292static int __init dev_proc_init(void)
4293{
4294	return register_pernet_subsys(&dev_proc_ops);
4295}
4296#else
4297#define dev_proc_init() 0
4298#endif	/* CONFIG_PROC_FS */
4299
4300
4301/**
4302 *	netdev_set_master	-	set up master pointer
4303 *	@slave: slave device
4304 *	@master: new master device
4305 *
4306 *	Changes the master device of the slave. Pass %NULL to break the
4307 *	bonding. The caller must hold the RTNL semaphore. On a failure
4308 *	a negative errno code is returned. On success the reference counts
4309 *	are adjusted and the function returns zero.
4310 */
4311int netdev_set_master(struct net_device *slave, struct net_device *master)
4312{
4313	struct net_device *old = slave->master;
4314
4315	ASSERT_RTNL();
4316
4317	if (master) {
4318		if (old)
4319			return -EBUSY;
4320		dev_hold(master);
4321	}
4322
4323	slave->master = master;
4324
4325	if (old)
4326		dev_put(old);
4327	return 0;
4328}
4329EXPORT_SYMBOL(netdev_set_master);
4330
4331/**
4332 *	netdev_set_bond_master	-	set up bonding master/slave pair
4333 *	@slave: slave device
4334 *	@master: new master device
4335 *
4336 *	Changes the master device of the slave. Pass %NULL to break the
4337 *	bonding. The caller must hold the RTNL semaphore. On a failure
4338 *	a negative errno code is returned. On success %RTM_NEWLINK is sent
4339 *	to the routing socket and the function returns zero.
4340 */
4341int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4342{
4343	int err;
4344
4345	ASSERT_RTNL();
4346
4347	err = netdev_set_master(slave, master);
4348	if (err)
4349		return err;
4350	if (master)
4351		slave->flags |= IFF_SLAVE;
4352	else
4353		slave->flags &= ~IFF_SLAVE;
4354
4355	rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4356	return 0;
4357}
4358EXPORT_SYMBOL(netdev_set_bond_master);
4359
4360static void dev_change_rx_flags(struct net_device *dev, int flags)
4361{
4362	const struct net_device_ops *ops = dev->netdev_ops;
4363
4364	if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4365		ops->ndo_change_rx_flags(dev, flags);
4366}
4367
4368static int __dev_set_promiscuity(struct net_device *dev, int inc)
4369{
4370	unsigned short old_flags = dev->flags;
4371	uid_t uid;
4372	gid_t gid;
4373
4374	ASSERT_RTNL();
4375
4376	dev->flags |= IFF_PROMISC;
4377	dev->promiscuity += inc;
4378	if (dev->promiscuity == 0) {
4379		/*
4380		 * Avoid overflow.
4381		 * If inc causes overflow, untouch promisc and return error.
4382		 */
4383		if (inc < 0)
4384			dev->flags &= ~IFF_PROMISC;
4385		else {
4386			dev->promiscuity -= inc;
4387			printk(KERN_WARNING "%s: promiscuity touches roof, "
4388				"set promiscuity failed, promiscuity feature "
4389				"of device might be broken.\n", dev->name);
4390			return -EOVERFLOW;
4391		}
4392	}
4393	if (dev->flags != old_flags) {
4394		printk(KERN_INFO "device %s %s promiscuous mode\n",
4395		       dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4396							       "left");
4397		if (audit_enabled) {
4398			current_uid_gid(&uid, &gid);
4399			audit_log(current->audit_context, GFP_ATOMIC,
4400				AUDIT_ANOM_PROMISCUOUS,
4401				"dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4402				dev->name, (dev->flags & IFF_PROMISC),
4403				(old_flags & IFF_PROMISC),
4404				audit_get_loginuid(current),
4405				uid, gid,
4406				audit_get_sessionid(current));
4407		}
4408
4409		dev_change_rx_flags(dev, IFF_PROMISC);
4410	}
4411	return 0;
4412}
4413
4414/**
4415 *	dev_set_promiscuity	- update promiscuity count on a device
4416 *	@dev: device
4417 *	@inc: modifier
4418 *
4419 *	Add or remove promiscuity from a device. While the count in the device
4420 *	remains above zero the interface remains promiscuous. Once it hits zero
4421 *	the device reverts back to normal filtering operation. A negative inc
4422 *	value is used to drop promiscuity on the device.
4423 *	Return 0 if successful or a negative errno code on error.
4424 */
4425int dev_set_promiscuity(struct net_device *dev, int inc)
4426{
4427	unsigned short old_flags = dev->flags;
4428	int err;
4429
4430	err = __dev_set_promiscuity(dev, inc);
4431	if (err < 0)
4432		return err;
4433	if (dev->flags != old_flags)
4434		dev_set_rx_mode(dev);
4435	return err;
4436}
4437EXPORT_SYMBOL(dev_set_promiscuity);
4438
4439/**
4440 *	dev_set_allmulti	- update allmulti count on a device
4441 *	@dev: device
4442 *	@inc: modifier
4443 *
4444 *	Add or remove reception of all multicast frames to a device. While the
4445 *	count in the device remains above zero the interface remains listening
4446 *	to all interfaces. Once it hits zero the device reverts back to normal
4447 *	filtering operation. A negative @inc value is used to drop the counter
4448 *	when releasing a resource needing all multicasts.
4449 *	Return 0 if successful or a negative errno code on error.
4450 */
4451
4452int dev_set_allmulti(struct net_device *dev, int inc)
4453{
4454	unsigned short old_flags = dev->flags;
4455
4456	ASSERT_RTNL();
4457
4458	dev->flags |= IFF_ALLMULTI;
4459	dev->allmulti += inc;
4460	if (dev->allmulti == 0) {
4461		/*
4462		 * Avoid overflow.
4463		 * If inc causes overflow, untouch allmulti and return error.
4464		 */
4465		if (inc < 0)
4466			dev->flags &= ~IFF_ALLMULTI;
4467		else {
4468			dev->allmulti -= inc;
4469			printk(KERN_WARNING "%s: allmulti touches roof, "
4470				"set allmulti failed, allmulti feature of "
4471				"device might be broken.\n", dev->name);
4472			return -EOVERFLOW;
4473		}
4474	}
4475	if (dev->flags ^ old_flags) {
4476		dev_change_rx_flags(dev, IFF_ALLMULTI);
4477		dev_set_rx_mode(dev);
4478	}
4479	return 0;
4480}
4481EXPORT_SYMBOL(dev_set_allmulti);
4482
4483/*
4484 *	Upload unicast and multicast address lists to device and
4485 *	configure RX filtering. When the device doesn't support unicast
4486 *	filtering it is put in promiscuous mode while unicast addresses
4487 *	are present.
4488 */
4489void __dev_set_rx_mode(struct net_device *dev)
4490{
4491	const struct net_device_ops *ops = dev->netdev_ops;
4492
4493	/* dev_open will call this function so the list will stay sane. */
4494	if (!(dev->flags&IFF_UP))
4495		return;
4496
4497	if (!netif_device_present(dev))
4498		return;
4499
4500	if (ops->ndo_set_rx_mode)
4501		ops->ndo_set_rx_mode(dev);
4502	else {
4503		/* Unicast addresses changes may only happen under the rtnl,
4504		 * therefore calling __dev_set_promiscuity here is safe.
4505		 */
4506		if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4507			__dev_set_promiscuity(dev, 1);
4508			dev->uc_promisc = true;
4509		} else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4510			__dev_set_promiscuity(dev, -1);
4511			dev->uc_promisc = false;
4512		}
 
4513
4514		if (ops->ndo_set_multicast_list)
4515			ops->ndo_set_multicast_list(dev);
4516	}
4517}
4518
4519void dev_set_rx_mode(struct net_device *dev)
4520{
4521	netif_addr_lock_bh(dev);
4522	__dev_set_rx_mode(dev);
4523	netif_addr_unlock_bh(dev);
4524}
4525
4526/**
4527 *	dev_ethtool_get_settings - call device's ethtool_ops::get_settings()
4528 *	@dev: device
4529 *	@cmd: memory area for ethtool_ops::get_settings() result
4530 *
4531 *      The cmd arg is initialized properly (cleared and
4532 *      ethtool_cmd::cmd field set to ETHTOOL_GSET).
4533 *
4534 *	Return device's ethtool_ops::get_settings() result value or
4535 *	-EOPNOTSUPP when device doesn't expose
4536 *	ethtool_ops::get_settings() operation.
4537 */
4538int dev_ethtool_get_settings(struct net_device *dev,
4539			     struct ethtool_cmd *cmd)
4540{
4541	if (!dev->ethtool_ops || !dev->ethtool_ops->get_settings)
4542		return -EOPNOTSUPP;
4543
4544	memset(cmd, 0, sizeof(struct ethtool_cmd));
4545	cmd->cmd = ETHTOOL_GSET;
4546	return dev->ethtool_ops->get_settings(dev, cmd);
4547}
4548EXPORT_SYMBOL(dev_ethtool_get_settings);
4549
4550/**
4551 *	dev_get_flags - get flags reported to userspace
4552 *	@dev: device
4553 *
4554 *	Get the combination of flag bits exported through APIs to userspace.
4555 */
4556unsigned dev_get_flags(const struct net_device *dev)
4557{
4558	unsigned flags;
4559
4560	flags = (dev->flags & ~(IFF_PROMISC |
4561				IFF_ALLMULTI |
4562				IFF_RUNNING |
4563				IFF_LOWER_UP |
4564				IFF_DORMANT)) |
4565		(dev->gflags & (IFF_PROMISC |
4566				IFF_ALLMULTI));
4567
4568	if (netif_running(dev)) {
4569		if (netif_oper_up(dev))
4570			flags |= IFF_RUNNING;
4571		if (netif_carrier_ok(dev))
4572			flags |= IFF_LOWER_UP;
4573		if (netif_dormant(dev))
4574			flags |= IFF_DORMANT;
4575	}
4576
4577	return flags;
4578}
4579EXPORT_SYMBOL(dev_get_flags);
4580
4581int __dev_change_flags(struct net_device *dev, unsigned int flags)
4582{
4583	int old_flags = dev->flags;
4584	int ret;
4585
4586	ASSERT_RTNL();
4587
4588	/*
4589	 *	Set the flags on our device.
4590	 */
4591
4592	dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4593			       IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4594			       IFF_AUTOMEDIA)) |
4595		     (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4596				    IFF_ALLMULTI));
4597
4598	/*
4599	 *	Load in the correct multicast list now the flags have changed.
4600	 */
4601
4602	if ((old_flags ^ flags) & IFF_MULTICAST)
4603		dev_change_rx_flags(dev, IFF_MULTICAST);
4604
4605	dev_set_rx_mode(dev);
4606
4607	/*
4608	 *	Have we downed the interface. We handle IFF_UP ourselves
4609	 *	according to user attempts to set it, rather than blindly
4610	 *	setting it.
4611	 */
4612
4613	ret = 0;
4614	if ((old_flags ^ flags) & IFF_UP) {	/* Bit is different  ? */
4615		ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4616
4617		if (!ret)
4618			dev_set_rx_mode(dev);
4619	}
4620
4621	if ((flags ^ dev->gflags) & IFF_PROMISC) {
4622		int inc = (flags & IFF_PROMISC) ? 1 : -1;
4623
4624		dev->gflags ^= IFF_PROMISC;
4625		dev_set_promiscuity(dev, inc);
4626	}
4627
4628	/* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4629	   is important. Some (broken) drivers set IFF_PROMISC, when
4630	   IFF_ALLMULTI is requested not asking us and not reporting.
4631	 */
4632	if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4633		int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4634
4635		dev->gflags ^= IFF_ALLMULTI;
4636		dev_set_allmulti(dev, inc);
4637	}
4638
4639	return ret;
4640}
4641
4642void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4643{
4644	unsigned int changes = dev->flags ^ old_flags;
4645
4646	if (changes & IFF_UP) {
4647		if (dev->flags & IFF_UP)
4648			call_netdevice_notifiers(NETDEV_UP, dev);
4649		else
4650			call_netdevice_notifiers(NETDEV_DOWN, dev);
4651	}
4652
4653	if (dev->flags & IFF_UP &&
4654	    (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4655		call_netdevice_notifiers(NETDEV_CHANGE, dev);
4656}
4657
4658/**
4659 *	dev_change_flags - change device settings
4660 *	@dev: device
4661 *	@flags: device state flags
4662 *
4663 *	Change settings on device based state flags. The flags are
4664 *	in the userspace exported format.
4665 */
4666int dev_change_flags(struct net_device *dev, unsigned flags)
4667{
4668	int ret, changes;
4669	int old_flags = dev->flags;
4670
4671	ret = __dev_change_flags(dev, flags);
4672	if (ret < 0)
4673		return ret;
4674
4675	changes = old_flags ^ dev->flags;
4676	if (changes)
4677		rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4678
4679	__dev_notify_flags(dev, old_flags);
4680	return ret;
4681}
4682EXPORT_SYMBOL(dev_change_flags);
4683
4684/**
4685 *	dev_set_mtu - Change maximum transfer unit
4686 *	@dev: device
4687 *	@new_mtu: new transfer unit
4688 *
4689 *	Change the maximum transfer size of the network device.
4690 */
4691int dev_set_mtu(struct net_device *dev, int new_mtu)
4692{
4693	const struct net_device_ops *ops = dev->netdev_ops;
4694	int err;
4695
4696	if (new_mtu == dev->mtu)
4697		return 0;
4698
4699	/*	MTU must be positive.	 */
4700	if (new_mtu < 0)
4701		return -EINVAL;
4702
4703	if (!netif_device_present(dev))
4704		return -ENODEV;
4705
4706	err = 0;
4707	if (ops->ndo_change_mtu)
4708		err = ops->ndo_change_mtu(dev, new_mtu);
4709	else
4710		dev->mtu = new_mtu;
4711
4712	if (!err && dev->flags & IFF_UP)
4713		call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4714	return err;
4715}
4716EXPORT_SYMBOL(dev_set_mtu);
4717
4718/**
4719 *	dev_set_group - Change group this device belongs to
4720 *	@dev: device
4721 *	@new_group: group this device should belong to
4722 */
4723void dev_set_group(struct net_device *dev, int new_group)
4724{
4725	dev->group = new_group;
4726}
4727EXPORT_SYMBOL(dev_set_group);
4728
4729/**
4730 *	dev_set_mac_address - Change Media Access Control Address
4731 *	@dev: device
4732 *	@sa: new address
4733 *
4734 *	Change the hardware (MAC) address of the device
4735 */
4736int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4737{
4738	const struct net_device_ops *ops = dev->netdev_ops;
4739	int err;
4740
4741	if (!ops->ndo_set_mac_address)
4742		return -EOPNOTSUPP;
4743	if (sa->sa_family != dev->type)
4744		return -EINVAL;
4745	if (!netif_device_present(dev))
4746		return -ENODEV;
4747	err = ops->ndo_set_mac_address(dev, sa);
4748	if (!err)
4749		call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
 
4750	return err;
4751}
4752EXPORT_SYMBOL(dev_set_mac_address);
4753
4754/*
4755 *	Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4756 */
4757static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4758{
4759	int err;
4760	struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4761
4762	if (!dev)
4763		return -ENODEV;
4764
4765	switch (cmd) {
4766	case SIOCGIFFLAGS:	/* Get interface flags */
4767		ifr->ifr_flags = (short) dev_get_flags(dev);
4768		return 0;
4769
4770	case SIOCGIFMETRIC:	/* Get the metric on the interface
4771				   (currently unused) */
4772		ifr->ifr_metric = 0;
4773		return 0;
4774
4775	case SIOCGIFMTU:	/* Get the MTU of a device */
4776		ifr->ifr_mtu = dev->mtu;
4777		return 0;
4778
4779	case SIOCGIFHWADDR:
4780		if (!dev->addr_len)
4781			memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4782		else
4783			memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4784			       min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4785		ifr->ifr_hwaddr.sa_family = dev->type;
4786		return 0;
4787
4788	case SIOCGIFSLAVE:
4789		err = -EINVAL;
4790		break;
4791
4792	case SIOCGIFMAP:
4793		ifr->ifr_map.mem_start = dev->mem_start;
4794		ifr->ifr_map.mem_end   = dev->mem_end;
4795		ifr->ifr_map.base_addr = dev->base_addr;
4796		ifr->ifr_map.irq       = dev->irq;
4797		ifr->ifr_map.dma       = dev->dma;
4798		ifr->ifr_map.port      = dev->if_port;
4799		return 0;
4800
4801	case SIOCGIFINDEX:
4802		ifr->ifr_ifindex = dev->ifindex;
4803		return 0;
4804
4805	case SIOCGIFTXQLEN:
4806		ifr->ifr_qlen = dev->tx_queue_len;
4807		return 0;
4808
4809	default:
4810		/* dev_ioctl() should ensure this case
4811		 * is never reached
4812		 */
4813		WARN_ON(1);
4814		err = -ENOTTY;
4815		break;
4816
4817	}
4818	return err;
4819}
4820
4821/*
4822 *	Perform the SIOCxIFxxx calls, inside rtnl_lock()
4823 */
4824static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4825{
4826	int err;
4827	struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4828	const struct net_device_ops *ops;
4829
4830	if (!dev)
4831		return -ENODEV;
4832
4833	ops = dev->netdev_ops;
4834
4835	switch (cmd) {
4836	case SIOCSIFFLAGS:	/* Set interface flags */
4837		return dev_change_flags(dev, ifr->ifr_flags);
4838
4839	case SIOCSIFMETRIC:	/* Set the metric on the interface
4840				   (currently unused) */
4841		return -EOPNOTSUPP;
4842
4843	case SIOCSIFMTU:	/* Set the MTU of a device */
4844		return dev_set_mtu(dev, ifr->ifr_mtu);
4845
4846	case SIOCSIFHWADDR:
4847		return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4848
4849	case SIOCSIFHWBROADCAST:
4850		if (ifr->ifr_hwaddr.sa_family != dev->type)
4851			return -EINVAL;
4852		memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4853		       min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4854		call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4855		return 0;
4856
4857	case SIOCSIFMAP:
4858		if (ops->ndo_set_config) {
4859			if (!netif_device_present(dev))
4860				return -ENODEV;
4861			return ops->ndo_set_config(dev, &ifr->ifr_map);
4862		}
4863		return -EOPNOTSUPP;
4864
4865	case SIOCADDMULTI:
4866		if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4867		    ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4868			return -EINVAL;
4869		if (!netif_device_present(dev))
4870			return -ENODEV;
4871		return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4872
4873	case SIOCDELMULTI:
4874		if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4875		    ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4876			return -EINVAL;
4877		if (!netif_device_present(dev))
4878			return -ENODEV;
4879		return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4880
4881	case SIOCSIFTXQLEN:
4882		if (ifr->ifr_qlen < 0)
4883			return -EINVAL;
4884		dev->tx_queue_len = ifr->ifr_qlen;
4885		return 0;
4886
4887	case SIOCSIFNAME:
4888		ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4889		return dev_change_name(dev, ifr->ifr_newname);
4890
 
 
 
 
 
 
4891	/*
4892	 *	Unknown or private ioctl
4893	 */
4894	default:
4895		if ((cmd >= SIOCDEVPRIVATE &&
4896		    cmd <= SIOCDEVPRIVATE + 15) ||
4897		    cmd == SIOCBONDENSLAVE ||
4898		    cmd == SIOCBONDRELEASE ||
4899		    cmd == SIOCBONDSETHWADDR ||
4900		    cmd == SIOCBONDSLAVEINFOQUERY ||
4901		    cmd == SIOCBONDINFOQUERY ||
4902		    cmd == SIOCBONDCHANGEACTIVE ||
4903		    cmd == SIOCGMIIPHY ||
4904		    cmd == SIOCGMIIREG ||
4905		    cmd == SIOCSMIIREG ||
4906		    cmd == SIOCBRADDIF ||
4907		    cmd == SIOCBRDELIF ||
4908		    cmd == SIOCSHWTSTAMP ||
4909		    cmd == SIOCWANDEV) {
4910			err = -EOPNOTSUPP;
4911			if (ops->ndo_do_ioctl) {
4912				if (netif_device_present(dev))
4913					err = ops->ndo_do_ioctl(dev, ifr, cmd);
4914				else
4915					err = -ENODEV;
4916			}
4917		} else
4918			err = -EINVAL;
4919
4920	}
4921	return err;
4922}
4923
4924/*
4925 *	This function handles all "interface"-type I/O control requests. The actual
4926 *	'doing' part of this is dev_ifsioc above.
4927 */
4928
4929/**
4930 *	dev_ioctl	-	network device ioctl
4931 *	@net: the applicable net namespace
4932 *	@cmd: command to issue
4933 *	@arg: pointer to a struct ifreq in user space
4934 *
4935 *	Issue ioctl functions to devices. This is normally called by the
4936 *	user space syscall interfaces but can sometimes be useful for
4937 *	other purposes. The return value is the return from the syscall if
4938 *	positive or a negative errno code on error.
4939 */
4940
4941int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4942{
4943	struct ifreq ifr;
4944	int ret;
4945	char *colon;
4946
4947	/* One special case: SIOCGIFCONF takes ifconf argument
4948	   and requires shared lock, because it sleeps writing
4949	   to user space.
4950	 */
4951
4952	if (cmd == SIOCGIFCONF) {
4953		rtnl_lock();
4954		ret = dev_ifconf(net, (char __user *) arg);
4955		rtnl_unlock();
4956		return ret;
4957	}
4958	if (cmd == SIOCGIFNAME)
4959		return dev_ifname(net, (struct ifreq __user *)arg);
4960
4961	if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4962		return -EFAULT;
4963
4964	ifr.ifr_name[IFNAMSIZ-1] = 0;
4965
4966	colon = strchr(ifr.ifr_name, ':');
4967	if (colon)
4968		*colon = 0;
4969
4970	/*
4971	 *	See which interface the caller is talking about.
4972	 */
4973
4974	switch (cmd) {
4975	/*
4976	 *	These ioctl calls:
4977	 *	- can be done by all.
4978	 *	- atomic and do not require locking.
4979	 *	- return a value
4980	 */
4981	case SIOCGIFFLAGS:
4982	case SIOCGIFMETRIC:
4983	case SIOCGIFMTU:
4984	case SIOCGIFHWADDR:
4985	case SIOCGIFSLAVE:
4986	case SIOCGIFMAP:
4987	case SIOCGIFINDEX:
4988	case SIOCGIFTXQLEN:
4989		dev_load(net, ifr.ifr_name);
4990		rcu_read_lock();
4991		ret = dev_ifsioc_locked(net, &ifr, cmd);
4992		rcu_read_unlock();
4993		if (!ret) {
4994			if (colon)
4995				*colon = ':';
4996			if (copy_to_user(arg, &ifr,
4997					 sizeof(struct ifreq)))
4998				ret = -EFAULT;
4999		}
5000		return ret;
5001
5002	case SIOCETHTOOL:
5003		dev_load(net, ifr.ifr_name);
5004		rtnl_lock();
5005		ret = dev_ethtool(net, &ifr);
5006		rtnl_unlock();
5007		if (!ret) {
5008			if (colon)
5009				*colon = ':';
5010			if (copy_to_user(arg, &ifr,
5011					 sizeof(struct ifreq)))
5012				ret = -EFAULT;
5013		}
5014		return ret;
5015
5016	/*
5017	 *	These ioctl calls:
5018	 *	- require superuser power.
5019	 *	- require strict serialization.
5020	 *	- return a value
5021	 */
5022	case SIOCGMIIPHY:
5023	case SIOCGMIIREG:
5024	case SIOCSIFNAME:
5025		if (!capable(CAP_NET_ADMIN))
5026			return -EPERM;
5027		dev_load(net, ifr.ifr_name);
5028		rtnl_lock();
5029		ret = dev_ifsioc(net, &ifr, cmd);
5030		rtnl_unlock();
5031		if (!ret) {
5032			if (colon)
5033				*colon = ':';
5034			if (copy_to_user(arg, &ifr,
5035					 sizeof(struct ifreq)))
5036				ret = -EFAULT;
5037		}
5038		return ret;
5039
5040	/*
5041	 *	These ioctl calls:
5042	 *	- require superuser power.
5043	 *	- require strict serialization.
5044	 *	- do not return a value
5045	 */
5046	case SIOCSIFFLAGS:
5047	case SIOCSIFMETRIC:
5048	case SIOCSIFMTU:
5049	case SIOCSIFMAP:
5050	case SIOCSIFHWADDR:
5051	case SIOCSIFSLAVE:
5052	case SIOCADDMULTI:
5053	case SIOCDELMULTI:
5054	case SIOCSIFHWBROADCAST:
5055	case SIOCSIFTXQLEN:
5056	case SIOCSMIIREG:
5057	case SIOCBONDENSLAVE:
5058	case SIOCBONDRELEASE:
5059	case SIOCBONDSETHWADDR:
5060	case SIOCBONDCHANGEACTIVE:
5061	case SIOCBRADDIF:
5062	case SIOCBRDELIF:
5063	case SIOCSHWTSTAMP:
5064		if (!capable(CAP_NET_ADMIN))
5065			return -EPERM;
5066		/* fall through */
5067	case SIOCBONDSLAVEINFOQUERY:
5068	case SIOCBONDINFOQUERY:
5069		dev_load(net, ifr.ifr_name);
5070		rtnl_lock();
5071		ret = dev_ifsioc(net, &ifr, cmd);
5072		rtnl_unlock();
5073		return ret;
5074
5075	case SIOCGIFMEM:
5076		/* Get the per device memory space. We can add this but
5077		 * currently do not support it */
5078	case SIOCSIFMEM:
5079		/* Set the per device memory buffer space.
5080		 * Not applicable in our case */
5081	case SIOCSIFLINK:
5082		return -ENOTTY;
5083
5084	/*
5085	 *	Unknown or private ioctl.
5086	 */
5087	default:
5088		if (cmd == SIOCWANDEV ||
5089		    (cmd >= SIOCDEVPRIVATE &&
5090		     cmd <= SIOCDEVPRIVATE + 15)) {
5091			dev_load(net, ifr.ifr_name);
5092			rtnl_lock();
5093			ret = dev_ifsioc(net, &ifr, cmd);
5094			rtnl_unlock();
5095			if (!ret && copy_to_user(arg, &ifr,
5096						 sizeof(struct ifreq)))
5097				ret = -EFAULT;
5098			return ret;
5099		}
5100		/* Take care of Wireless Extensions */
5101		if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5102			return wext_handle_ioctl(net, &ifr, cmd, arg);
5103		return -ENOTTY;
5104	}
5105}
5106
5107
5108/**
5109 *	dev_new_index	-	allocate an ifindex
5110 *	@net: the applicable net namespace
5111 *
5112 *	Returns a suitable unique value for a new device interface
5113 *	number.  The caller must hold the rtnl semaphore or the
5114 *	dev_base_lock to be sure it remains unique.
5115 */
5116static int dev_new_index(struct net *net)
5117{
5118	static int ifindex;
5119	for (;;) {
5120		if (++ifindex <= 0)
5121			ifindex = 1;
5122		if (!__dev_get_by_index(net, ifindex))
5123			return ifindex;
5124	}
5125}
5126
5127/* Delayed registration/unregisteration */
5128static LIST_HEAD(net_todo_list);
5129
5130static void net_set_todo(struct net_device *dev)
5131{
5132	list_add_tail(&dev->todo_list, &net_todo_list);
5133}
5134
5135static void rollback_registered_many(struct list_head *head)
5136{
5137	struct net_device *dev, *tmp;
5138
5139	BUG_ON(dev_boot_phase);
5140	ASSERT_RTNL();
5141
5142	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5143		/* Some devices call without registering
5144		 * for initialization unwind. Remove those
5145		 * devices and proceed with the remaining.
5146		 */
5147		if (dev->reg_state == NETREG_UNINITIALIZED) {
5148			pr_debug("unregister_netdevice: device %s/%p never "
5149				 "was registered\n", dev->name, dev);
5150
5151			WARN_ON(1);
5152			list_del(&dev->unreg_list);
5153			continue;
5154		}
5155		dev->dismantle = true;
5156		BUG_ON(dev->reg_state != NETREG_REGISTERED);
5157	}
5158
5159	/* If device is running, close it first. */
5160	dev_close_many(head);
5161
5162	list_for_each_entry(dev, head, unreg_list) {
5163		/* And unlink it from device chain. */
5164		unlist_netdevice(dev);
5165
5166		dev->reg_state = NETREG_UNREGISTERING;
5167	}
5168
5169	synchronize_net();
5170
5171	list_for_each_entry(dev, head, unreg_list) {
5172		/* Shutdown queueing discipline. */
5173		dev_shutdown(dev);
5174
5175
5176		/* Notify protocols, that we are about to destroy
5177		   this device. They should clean all the things.
5178		*/
5179		call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5180
5181		if (!dev->rtnl_link_ops ||
5182		    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5183			rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5184
5185		/*
5186		 *	Flush the unicast and multicast chains
5187		 */
5188		dev_uc_flush(dev);
5189		dev_mc_flush(dev);
5190
5191		if (dev->netdev_ops->ndo_uninit)
5192			dev->netdev_ops->ndo_uninit(dev);
5193
5194		/* Notifier chain MUST detach us from master device. */
5195		WARN_ON(dev->master);
5196
5197		/* Remove entries from kobject tree */
5198		netdev_unregister_kobject(dev);
5199	}
5200
5201	/* Process any work delayed until the end of the batch */
5202	dev = list_first_entry(head, struct net_device, unreg_list);
5203	call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5204
5205	rcu_barrier();
5206
5207	list_for_each_entry(dev, head, unreg_list)
5208		dev_put(dev);
5209}
5210
5211static void rollback_registered(struct net_device *dev)
5212{
5213	LIST_HEAD(single);
5214
5215	list_add(&dev->unreg_list, &single);
5216	rollback_registered_many(&single);
5217	list_del(&single);
5218}
5219
5220static u32 netdev_fix_features(struct net_device *dev, u32 features)
 
5221{
5222	/* Fix illegal checksum combinations */
5223	if ((features & NETIF_F_HW_CSUM) &&
5224	    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5225		netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5226		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5227	}
5228
5229	if ((features & NETIF_F_NO_CSUM) &&
5230	    (features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5231		netdev_warn(dev, "mixed no checksumming and other settings.\n");
5232		features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5233	}
5234
5235	/* Fix illegal SG+CSUM combinations. */
5236	if ((features & NETIF_F_SG) &&
5237	    !(features & NETIF_F_ALL_CSUM)) {
5238		netdev_dbg(dev,
5239			"Dropping NETIF_F_SG since no checksum feature.\n");
5240		features &= ~NETIF_F_SG;
5241	}
5242
5243	/* TSO requires that SG is present as well. */
5244	if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5245		netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5246		features &= ~NETIF_F_ALL_TSO;
5247	}
5248
5249	/* TSO ECN requires that TSO is present as well. */
5250	if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5251		features &= ~NETIF_F_TSO_ECN;
5252
5253	/* Software GSO depends on SG. */
5254	if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5255		netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5256		features &= ~NETIF_F_GSO;
5257	}
5258
5259	/* UFO needs SG and checksumming */
5260	if (features & NETIF_F_UFO) {
5261		/* maybe split UFO into V4 and V6? */
5262		if (!((features & NETIF_F_GEN_CSUM) ||
5263		    (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5264			    == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5265			netdev_dbg(dev,
5266				"Dropping NETIF_F_UFO since no checksum offload features.\n");
5267			features &= ~NETIF_F_UFO;
5268		}
5269
5270		if (!(features & NETIF_F_SG)) {
5271			netdev_dbg(dev,
5272				"Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5273			features &= ~NETIF_F_UFO;
5274		}
5275	}
5276
5277	return features;
5278}
5279
5280int __netdev_update_features(struct net_device *dev)
5281{
5282	u32 features;
5283	int err = 0;
5284
5285	ASSERT_RTNL();
5286
5287	features = netdev_get_wanted_features(dev);
5288
5289	if (dev->netdev_ops->ndo_fix_features)
5290		features = dev->netdev_ops->ndo_fix_features(dev, features);
5291
5292	/* driver might be less strict about feature dependencies */
5293	features = netdev_fix_features(dev, features);
5294
5295	if (dev->features == features)
5296		return 0;
5297
5298	netdev_dbg(dev, "Features changed: 0x%08x -> 0x%08x\n",
5299		dev->features, features);
5300
5301	if (dev->netdev_ops->ndo_set_features)
5302		err = dev->netdev_ops->ndo_set_features(dev, features);
5303
5304	if (unlikely(err < 0)) {
5305		netdev_err(dev,
5306			"set_features() failed (%d); wanted 0x%08x, left 0x%08x\n",
5307			err, features, dev->features);
5308		return -1;
5309	}
5310
5311	if (!err)
5312		dev->features = features;
5313
5314	return 1;
5315}
5316
5317/**
5318 *	netdev_update_features - recalculate device features
5319 *	@dev: the device to check
5320 *
5321 *	Recalculate dev->features set and send notifications if it
5322 *	has changed. Should be called after driver or hardware dependent
5323 *	conditions might have changed that influence the features.
5324 */
5325void netdev_update_features(struct net_device *dev)
5326{
5327	if (__netdev_update_features(dev))
5328		netdev_features_change(dev);
5329}
5330EXPORT_SYMBOL(netdev_update_features);
5331
5332/**
5333 *	netdev_change_features - recalculate device features
5334 *	@dev: the device to check
5335 *
5336 *	Recalculate dev->features set and send notifications even
5337 *	if they have not changed. Should be called instead of
5338 *	netdev_update_features() if also dev->vlan_features might
5339 *	have changed to allow the changes to be propagated to stacked
5340 *	VLAN devices.
5341 */
5342void netdev_change_features(struct net_device *dev)
5343{
5344	__netdev_update_features(dev);
5345	netdev_features_change(dev);
5346}
5347EXPORT_SYMBOL(netdev_change_features);
5348
5349/**
5350 *	netif_stacked_transfer_operstate -	transfer operstate
5351 *	@rootdev: the root or lower level device to transfer state from
5352 *	@dev: the device to transfer operstate to
5353 *
5354 *	Transfer operational state from root to device. This is normally
5355 *	called when a stacking relationship exists between the root
5356 *	device and the device(a leaf device).
5357 */
5358void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5359					struct net_device *dev)
5360{
5361	if (rootdev->operstate == IF_OPER_DORMANT)
5362		netif_dormant_on(dev);
5363	else
5364		netif_dormant_off(dev);
5365
5366	if (netif_carrier_ok(rootdev)) {
5367		if (!netif_carrier_ok(dev))
5368			netif_carrier_on(dev);
5369	} else {
5370		if (netif_carrier_ok(dev))
5371			netif_carrier_off(dev);
5372	}
5373}
5374EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5375
5376#ifdef CONFIG_RPS
5377static int netif_alloc_rx_queues(struct net_device *dev)
5378{
5379	unsigned int i, count = dev->num_rx_queues;
5380	struct netdev_rx_queue *rx;
5381
5382	BUG_ON(count < 1);
5383
5384	rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5385	if (!rx) {
5386		pr_err("netdev: Unable to allocate %u rx queues.\n", count);
5387		return -ENOMEM;
5388	}
5389	dev->_rx = rx;
5390
5391	for (i = 0; i < count; i++)
5392		rx[i].dev = dev;
5393	return 0;
5394}
5395#endif
5396
5397static void netdev_init_one_queue(struct net_device *dev,
5398				  struct netdev_queue *queue, void *_unused)
5399{
5400	/* Initialize queue lock */
5401	spin_lock_init(&queue->_xmit_lock);
5402	netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5403	queue->xmit_lock_owner = -1;
5404	netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5405	queue->dev = dev;
 
 
 
5406}
5407
5408static int netif_alloc_netdev_queues(struct net_device *dev)
5409{
5410	unsigned int count = dev->num_tx_queues;
5411	struct netdev_queue *tx;
5412
5413	BUG_ON(count < 1);
5414
5415	tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5416	if (!tx) {
5417		pr_err("netdev: Unable to allocate %u tx queues.\n",
5418		       count);
5419		return -ENOMEM;
5420	}
5421	dev->_tx = tx;
5422
5423	netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5424	spin_lock_init(&dev->tx_global_lock);
5425
5426	return 0;
5427}
5428
5429/**
5430 *	register_netdevice	- register a network device
5431 *	@dev: device to register
5432 *
5433 *	Take a completed network device structure and add it to the kernel
5434 *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5435 *	chain. 0 is returned on success. A negative errno code is returned
5436 *	on a failure to set up the device, or if the name is a duplicate.
5437 *
5438 *	Callers must hold the rtnl semaphore. You may want
5439 *	register_netdev() instead of this.
5440 *
5441 *	BUGS:
5442 *	The locking appears insufficient to guarantee two parallel registers
5443 *	will not get the same name.
5444 */
5445
5446int register_netdevice(struct net_device *dev)
5447{
5448	int ret;
5449	struct net *net = dev_net(dev);
5450
5451	BUG_ON(dev_boot_phase);
5452	ASSERT_RTNL();
5453
5454	might_sleep();
5455
5456	/* When net_device's are persistent, this will be fatal. */
5457	BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5458	BUG_ON(!net);
5459
5460	spin_lock_init(&dev->addr_list_lock);
5461	netdev_set_addr_lockdep_class(dev);
5462
5463	dev->iflink = -1;
5464
5465	ret = dev_get_valid_name(dev, dev->name);
5466	if (ret < 0)
5467		goto out;
5468
5469	/* Init, if this function is available */
5470	if (dev->netdev_ops->ndo_init) {
5471		ret = dev->netdev_ops->ndo_init(dev);
5472		if (ret) {
5473			if (ret > 0)
5474				ret = -EIO;
5475			goto out;
5476		}
5477	}
5478
5479	dev->ifindex = dev_new_index(net);
5480	if (dev->iflink == -1)
5481		dev->iflink = dev->ifindex;
5482
5483	/* Transfer changeable features to wanted_features and enable
5484	 * software offloads (GSO and GRO).
5485	 */
5486	dev->hw_features |= NETIF_F_SOFT_FEATURES;
5487	dev->features |= NETIF_F_SOFT_FEATURES;
5488	dev->wanted_features = dev->features & dev->hw_features;
5489
5490	/* Turn on no cache copy if HW is doing checksum */
5491	dev->hw_features |= NETIF_F_NOCACHE_COPY;
5492	if ((dev->features & NETIF_F_ALL_CSUM) &&
5493	    !(dev->features & NETIF_F_NO_CSUM)) {
5494		dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5495		dev->features |= NETIF_F_NOCACHE_COPY;
 
5496	}
5497
5498	/* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5499	 */
5500	dev->vlan_features |= NETIF_F_HIGHDMA;
5501
5502	ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5503	ret = notifier_to_errno(ret);
5504	if (ret)
5505		goto err_uninit;
5506
5507	ret = netdev_register_kobject(dev);
5508	if (ret)
5509		goto err_uninit;
5510	dev->reg_state = NETREG_REGISTERED;
5511
5512	__netdev_update_features(dev);
5513
5514	/*
5515	 *	Default initial state at registry is that the
5516	 *	device is present.
5517	 */
5518
5519	set_bit(__LINK_STATE_PRESENT, &dev->state);
5520
5521	dev_init_scheduler(dev);
5522	dev_hold(dev);
5523	list_netdevice(dev);
 
5524
5525	/* Notify protocols, that a new device appeared. */
5526	ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5527	ret = notifier_to_errno(ret);
5528	if (ret) {
5529		rollback_registered(dev);
5530		dev->reg_state = NETREG_UNREGISTERED;
5531	}
5532	/*
5533	 *	Prevent userspace races by waiting until the network
5534	 *	device is fully setup before sending notifications.
5535	 */
5536	if (!dev->rtnl_link_ops ||
5537	    dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5538		rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5539
5540out:
5541	return ret;
5542
5543err_uninit:
5544	if (dev->netdev_ops->ndo_uninit)
5545		dev->netdev_ops->ndo_uninit(dev);
5546	goto out;
5547}
5548EXPORT_SYMBOL(register_netdevice);
5549
5550/**
5551 *	init_dummy_netdev	- init a dummy network device for NAPI
5552 *	@dev: device to init
5553 *
5554 *	This takes a network device structure and initialize the minimum
5555 *	amount of fields so it can be used to schedule NAPI polls without
5556 *	registering a full blown interface. This is to be used by drivers
5557 *	that need to tie several hardware interfaces to a single NAPI
5558 *	poll scheduler due to HW limitations.
5559 */
5560int init_dummy_netdev(struct net_device *dev)
5561{
5562	/* Clear everything. Note we don't initialize spinlocks
5563	 * are they aren't supposed to be taken by any of the
5564	 * NAPI code and this dummy netdev is supposed to be
5565	 * only ever used for NAPI polls
5566	 */
5567	memset(dev, 0, sizeof(struct net_device));
5568
5569	/* make sure we BUG if trying to hit standard
5570	 * register/unregister code path
5571	 */
5572	dev->reg_state = NETREG_DUMMY;
5573
5574	/* NAPI wants this */
5575	INIT_LIST_HEAD(&dev->napi_list);
5576
5577	/* a dummy interface is started by default */
5578	set_bit(__LINK_STATE_PRESENT, &dev->state);
5579	set_bit(__LINK_STATE_START, &dev->state);
5580
5581	/* Note : We dont allocate pcpu_refcnt for dummy devices,
5582	 * because users of this 'device' dont need to change
5583	 * its refcount.
5584	 */
5585
5586	return 0;
5587}
5588EXPORT_SYMBOL_GPL(init_dummy_netdev);
5589
5590
5591/**
5592 *	register_netdev	- register a network device
5593 *	@dev: device to register
5594 *
5595 *	Take a completed network device structure and add it to the kernel
5596 *	interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5597 *	chain. 0 is returned on success. A negative errno code is returned
5598 *	on a failure to set up the device, or if the name is a duplicate.
5599 *
5600 *	This is a wrapper around register_netdevice that takes the rtnl semaphore
5601 *	and expands the device name if you passed a format string to
5602 *	alloc_netdev.
5603 */
5604int register_netdev(struct net_device *dev)
5605{
5606	int err;
5607
5608	rtnl_lock();
5609	err = register_netdevice(dev);
5610	rtnl_unlock();
5611	return err;
5612}
5613EXPORT_SYMBOL(register_netdev);
5614
5615int netdev_refcnt_read(const struct net_device *dev)
5616{
5617	int i, refcnt = 0;
5618
5619	for_each_possible_cpu(i)
5620		refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5621	return refcnt;
5622}
5623EXPORT_SYMBOL(netdev_refcnt_read);
5624
5625/*
5626 * netdev_wait_allrefs - wait until all references are gone.
5627 *
5628 * This is called when unregistering network devices.
5629 *
5630 * Any protocol or device that holds a reference should register
5631 * for netdevice notification, and cleanup and put back the
5632 * reference if they receive an UNREGISTER event.
5633 * We can get stuck here if buggy protocols don't correctly
5634 * call dev_put.
5635 */
5636static void netdev_wait_allrefs(struct net_device *dev)
5637{
5638	unsigned long rebroadcast_time, warning_time;
5639	int refcnt;
5640
5641	linkwatch_forget_dev(dev);
5642
5643	rebroadcast_time = warning_time = jiffies;
5644	refcnt = netdev_refcnt_read(dev);
5645
5646	while (refcnt != 0) {
5647		if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5648			rtnl_lock();
5649
5650			/* Rebroadcast unregister notification */
5651			call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5652			/* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5653			 * should have already handle it the first time */
5654
5655			if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5656				     &dev->state)) {
5657				/* We must not have linkwatch events
5658				 * pending on unregister. If this
5659				 * happens, we simply run the queue
5660				 * unscheduled, resulting in a noop
5661				 * for this device.
5662				 */
5663				linkwatch_run_queue();
5664			}
5665
5666			__rtnl_unlock();
5667
5668			rebroadcast_time = jiffies;
5669		}
5670
5671		msleep(250);
5672
5673		refcnt = netdev_refcnt_read(dev);
5674
5675		if (time_after(jiffies, warning_time + 10 * HZ)) {
5676			printk(KERN_EMERG "unregister_netdevice: "
5677			       "waiting for %s to become free. Usage "
5678			       "count = %d\n",
5679			       dev->name, refcnt);
5680			warning_time = jiffies;
5681		}
5682	}
5683}
5684
5685/* The sequence is:
5686 *
5687 *	rtnl_lock();
5688 *	...
5689 *	register_netdevice(x1);
5690 *	register_netdevice(x2);
5691 *	...
5692 *	unregister_netdevice(y1);
5693 *	unregister_netdevice(y2);
5694 *      ...
5695 *	rtnl_unlock();
5696 *	free_netdev(y1);
5697 *	free_netdev(y2);
5698 *
5699 * We are invoked by rtnl_unlock().
5700 * This allows us to deal with problems:
5701 * 1) We can delete sysfs objects which invoke hotplug
5702 *    without deadlocking with linkwatch via keventd.
5703 * 2) Since we run with the RTNL semaphore not held, we can sleep
5704 *    safely in order to wait for the netdev refcnt to drop to zero.
5705 *
5706 * We must not return until all unregister events added during
5707 * the interval the lock was held have been completed.
5708 */
5709void netdev_run_todo(void)
5710{
5711	struct list_head list;
5712
5713	/* Snapshot list, allow later requests */
5714	list_replace_init(&net_todo_list, &list);
5715
5716	__rtnl_unlock();
5717
 
 
 
 
 
 
5718	while (!list_empty(&list)) {
5719		struct net_device *dev
5720			= list_first_entry(&list, struct net_device, todo_list);
5721		list_del(&dev->todo_list);
5722
5723		if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5724			printk(KERN_ERR "network todo '%s' but state %d\n",
5725			       dev->name, dev->reg_state);
5726			dump_stack();
5727			continue;
5728		}
5729
5730		dev->reg_state = NETREG_UNREGISTERED;
5731
5732		on_each_cpu(flush_backlog, dev, 1);
5733
5734		netdev_wait_allrefs(dev);
5735
5736		/* paranoia */
5737		BUG_ON(netdev_refcnt_read(dev));
5738		WARN_ON(rcu_dereference_raw(dev->ip_ptr));
5739		WARN_ON(rcu_dereference_raw(dev->ip6_ptr));
5740		WARN_ON(dev->dn_ptr);
5741
5742		if (dev->destructor)
5743			dev->destructor(dev);
5744
5745		/* Free network device */
5746		kobject_put(&dev->dev.kobj);
5747	}
5748}
5749
5750/* Convert net_device_stats to rtnl_link_stats64.  They have the same
5751 * fields in the same order, with only the type differing.
5752 */
5753static void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5754				    const struct net_device_stats *netdev_stats)
5755{
5756#if BITS_PER_LONG == 64
5757        BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5758        memcpy(stats64, netdev_stats, sizeof(*stats64));
5759#else
5760	size_t i, n = sizeof(*stats64) / sizeof(u64);
5761	const unsigned long *src = (const unsigned long *)netdev_stats;
5762	u64 *dst = (u64 *)stats64;
5763
5764	BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5765		     sizeof(*stats64) / sizeof(u64));
5766	for (i = 0; i < n; i++)
5767		dst[i] = src[i];
5768#endif
5769}
 
5770
5771/**
5772 *	dev_get_stats	- get network device statistics
5773 *	@dev: device to get statistics from
5774 *	@storage: place to store stats
5775 *
5776 *	Get network statistics from device. Return @storage.
5777 *	The device driver may provide its own method by setting
5778 *	dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5779 *	otherwise the internal statistics structure is used.
5780 */
5781struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5782					struct rtnl_link_stats64 *storage)
5783{
5784	const struct net_device_ops *ops = dev->netdev_ops;
5785
5786	if (ops->ndo_get_stats64) {
5787		memset(storage, 0, sizeof(*storage));
5788		ops->ndo_get_stats64(dev, storage);
5789	} else if (ops->ndo_get_stats) {
5790		netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5791	} else {
5792		netdev_stats_to_stats64(storage, &dev->stats);
5793	}
5794	storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5795	return storage;
5796}
5797EXPORT_SYMBOL(dev_get_stats);
5798
5799struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5800{
5801	struct netdev_queue *queue = dev_ingress_queue(dev);
5802
5803#ifdef CONFIG_NET_CLS_ACT
5804	if (queue)
5805		return queue;
5806	queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5807	if (!queue)
5808		return NULL;
5809	netdev_init_one_queue(dev, queue, NULL);
5810	queue->qdisc = &noop_qdisc;
5811	queue->qdisc_sleeping = &noop_qdisc;
5812	rcu_assign_pointer(dev->ingress_queue, queue);
5813#endif
5814	return queue;
5815}
5816
5817/**
5818 *	alloc_netdev_mqs - allocate network device
5819 *	@sizeof_priv:	size of private data to allocate space for
5820 *	@name:		device name format string
5821 *	@setup:		callback to initialize device
5822 *	@txqs:		the number of TX subqueues to allocate
5823 *	@rxqs:		the number of RX subqueues to allocate
5824 *
5825 *	Allocates a struct net_device with private data area for driver use
5826 *	and performs basic initialization.  Also allocates subquue structs
5827 *	for each queue on the device.
5828 */
5829struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5830		void (*setup)(struct net_device *),
5831		unsigned int txqs, unsigned int rxqs)
5832{
5833	struct net_device *dev;
5834	size_t alloc_size;
5835	struct net_device *p;
5836
5837	BUG_ON(strlen(name) >= sizeof(dev->name));
5838
5839	if (txqs < 1) {
5840		pr_err("alloc_netdev: Unable to allocate device "
5841		       "with zero queues.\n");
5842		return NULL;
5843	}
5844
5845#ifdef CONFIG_RPS
5846	if (rxqs < 1) {
5847		pr_err("alloc_netdev: Unable to allocate device "
5848		       "with zero RX queues.\n");
5849		return NULL;
5850	}
5851#endif
5852
5853	alloc_size = sizeof(struct net_device);
5854	if (sizeof_priv) {
5855		/* ensure 32-byte alignment of private area */
5856		alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5857		alloc_size += sizeof_priv;
5858	}
5859	/* ensure 32-byte alignment of whole construct */
5860	alloc_size += NETDEV_ALIGN - 1;
5861
5862	p = kzalloc(alloc_size, GFP_KERNEL);
5863	if (!p) {
5864		printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5865		return NULL;
5866	}
5867
5868	dev = PTR_ALIGN(p, NETDEV_ALIGN);
5869	dev->padded = (char *)dev - (char *)p;
5870
5871	dev->pcpu_refcnt = alloc_percpu(int);
5872	if (!dev->pcpu_refcnt)
5873		goto free_p;
5874
5875	if (dev_addr_init(dev))
5876		goto free_pcpu;
5877
5878	dev_mc_init(dev);
5879	dev_uc_init(dev);
5880
5881	dev_net_set(dev, &init_net);
5882
5883	dev->gso_max_size = GSO_MAX_SIZE;
 
5884
5885	INIT_LIST_HEAD(&dev->napi_list);
5886	INIT_LIST_HEAD(&dev->unreg_list);
5887	INIT_LIST_HEAD(&dev->link_watch_list);
5888	dev->priv_flags = IFF_XMIT_DST_RELEASE;
5889	setup(dev);
5890
5891	dev->num_tx_queues = txqs;
5892	dev->real_num_tx_queues = txqs;
5893	if (netif_alloc_netdev_queues(dev))
5894		goto free_all;
5895
5896#ifdef CONFIG_RPS
5897	dev->num_rx_queues = rxqs;
5898	dev->real_num_rx_queues = rxqs;
5899	if (netif_alloc_rx_queues(dev))
5900		goto free_all;
5901#endif
5902
5903	strcpy(dev->name, name);
5904	dev->group = INIT_NETDEV_GROUP;
5905	return dev;
5906
5907free_all:
5908	free_netdev(dev);
5909	return NULL;
5910
5911free_pcpu:
5912	free_percpu(dev->pcpu_refcnt);
5913	kfree(dev->_tx);
5914#ifdef CONFIG_RPS
5915	kfree(dev->_rx);
5916#endif
5917
5918free_p:
5919	kfree(p);
5920	return NULL;
5921}
5922EXPORT_SYMBOL(alloc_netdev_mqs);
5923
5924/**
5925 *	free_netdev - free network device
5926 *	@dev: device
5927 *
5928 *	This function does the last stage of destroying an allocated device
5929 * 	interface. The reference to the device object is released.
5930 *	If this is the last reference then it will be freed.
5931 */
5932void free_netdev(struct net_device *dev)
5933{
5934	struct napi_struct *p, *n;
5935
5936	release_net(dev_net(dev));
5937
5938	kfree(dev->_tx);
5939#ifdef CONFIG_RPS
5940	kfree(dev->_rx);
5941#endif
5942
5943	kfree(rcu_dereference_raw(dev->ingress_queue));
5944
5945	/* Flush device addresses */
5946	dev_addr_flush(dev);
5947
5948	list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5949		netif_napi_del(p);
5950
5951	free_percpu(dev->pcpu_refcnt);
5952	dev->pcpu_refcnt = NULL;
5953
5954	/*  Compatibility with error handling in drivers */
5955	if (dev->reg_state == NETREG_UNINITIALIZED) {
5956		kfree((char *)dev - dev->padded);
5957		return;
5958	}
5959
5960	BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5961	dev->reg_state = NETREG_RELEASED;
5962
5963	/* will free via device release */
5964	put_device(&dev->dev);
5965}
5966EXPORT_SYMBOL(free_netdev);
5967
5968/**
5969 *	synchronize_net -  Synchronize with packet receive processing
5970 *
5971 *	Wait for packets currently being received to be done.
5972 *	Does not block later packets from starting.
5973 */
5974void synchronize_net(void)
5975{
5976	might_sleep();
5977	if (rtnl_is_locked())
5978		synchronize_rcu_expedited();
5979	else
5980		synchronize_rcu();
5981}
5982EXPORT_SYMBOL(synchronize_net);
5983
5984/**
5985 *	unregister_netdevice_queue - remove device from the kernel
5986 *	@dev: device
5987 *	@head: list
5988 *
5989 *	This function shuts down a device interface and removes it
5990 *	from the kernel tables.
5991 *	If head not NULL, device is queued to be unregistered later.
5992 *
5993 *	Callers must hold the rtnl semaphore.  You may want
5994 *	unregister_netdev() instead of this.
5995 */
5996
5997void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5998{
5999	ASSERT_RTNL();
6000
6001	if (head) {
6002		list_move_tail(&dev->unreg_list, head);
6003	} else {
6004		rollback_registered(dev);
6005		/* Finish processing unregister after unlock */
6006		net_set_todo(dev);
6007	}
6008}
6009EXPORT_SYMBOL(unregister_netdevice_queue);
6010
6011/**
6012 *	unregister_netdevice_many - unregister many devices
6013 *	@head: list of devices
6014 */
6015void unregister_netdevice_many(struct list_head *head)
6016{
6017	struct net_device *dev;
6018
6019	if (!list_empty(head)) {
6020		rollback_registered_many(head);
6021		list_for_each_entry(dev, head, unreg_list)
6022			net_set_todo(dev);
6023	}
6024}
6025EXPORT_SYMBOL(unregister_netdevice_many);
6026
6027/**
6028 *	unregister_netdev - remove device from the kernel
6029 *	@dev: device
6030 *
6031 *	This function shuts down a device interface and removes it
6032 *	from the kernel tables.
6033 *
6034 *	This is just a wrapper for unregister_netdevice that takes
6035 *	the rtnl semaphore.  In general you want to use this and not
6036 *	unregister_netdevice.
6037 */
6038void unregister_netdev(struct net_device *dev)
6039{
6040	rtnl_lock();
6041	unregister_netdevice(dev);
6042	rtnl_unlock();
6043}
6044EXPORT_SYMBOL(unregister_netdev);
6045
6046/**
6047 *	dev_change_net_namespace - move device to different nethost namespace
6048 *	@dev: device
6049 *	@net: network namespace
6050 *	@pat: If not NULL name pattern to try if the current device name
6051 *	      is already taken in the destination network namespace.
6052 *
6053 *	This function shuts down a device interface and moves it
6054 *	to a new network namespace. On success 0 is returned, on
6055 *	a failure a netagive errno code is returned.
6056 *
6057 *	Callers must hold the rtnl semaphore.
6058 */
6059
6060int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6061{
6062	int err;
6063
6064	ASSERT_RTNL();
6065
6066	/* Don't allow namespace local devices to be moved. */
6067	err = -EINVAL;
6068	if (dev->features & NETIF_F_NETNS_LOCAL)
6069		goto out;
6070
6071	/* Ensure the device has been registrered */
6072	err = -EINVAL;
6073	if (dev->reg_state != NETREG_REGISTERED)
6074		goto out;
6075
6076	/* Get out if there is nothing todo */
6077	err = 0;
6078	if (net_eq(dev_net(dev), net))
6079		goto out;
6080
6081	/* Pick the destination device name, and ensure
6082	 * we can use it in the destination network namespace.
6083	 */
6084	err = -EEXIST;
6085	if (__dev_get_by_name(net, dev->name)) {
6086		/* We get here if we can't use the current device name */
6087		if (!pat)
6088			goto out;
6089		if (dev_get_valid_name(dev, pat) < 0)
6090			goto out;
6091	}
6092
6093	/*
6094	 * And now a mini version of register_netdevice unregister_netdevice.
6095	 */
6096
6097	/* If device is running close it first. */
6098	dev_close(dev);
6099
6100	/* And unlink it from device chain */
6101	err = -ENODEV;
6102	unlist_netdevice(dev);
6103
6104	synchronize_net();
6105
6106	/* Shutdown queueing discipline. */
6107	dev_shutdown(dev);
6108
6109	/* Notify protocols, that we are about to destroy
6110	   this device. They should clean all the things.
6111
6112	   Note that dev->reg_state stays at NETREG_REGISTERED.
6113	   This is wanted because this way 8021q and macvlan know
6114	   the device is just moving and can keep their slaves up.
6115	*/
6116	call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6117	call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
 
6118
6119	/*
6120	 *	Flush the unicast and multicast chains
6121	 */
6122	dev_uc_flush(dev);
6123	dev_mc_flush(dev);
6124
6125	/* Actually switch the network namespace */
6126	dev_net_set(dev, net);
6127
6128	/* If there is an ifindex conflict assign a new one */
6129	if (__dev_get_by_index(net, dev->ifindex)) {
6130		int iflink = (dev->iflink == dev->ifindex);
6131		dev->ifindex = dev_new_index(net);
6132		if (iflink)
6133			dev->iflink = dev->ifindex;
6134	}
6135
6136	/* Fixup kobjects */
6137	err = device_rename(&dev->dev, dev->name);
6138	WARN_ON(err);
6139
6140	/* Add the device back in the hashes */
6141	list_netdevice(dev);
6142
6143	/* Notify protocols, that a new device appeared. */
6144	call_netdevice_notifiers(NETDEV_REGISTER, dev);
6145
6146	/*
6147	 *	Prevent userspace races by waiting until the network
6148	 *	device is fully setup before sending notifications.
6149	 */
6150	rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6151
6152	synchronize_net();
6153	err = 0;
6154out:
6155	return err;
6156}
6157EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6158
6159static int dev_cpu_callback(struct notifier_block *nfb,
6160			    unsigned long action,
6161			    void *ocpu)
6162{
6163	struct sk_buff **list_skb;
6164	struct sk_buff *skb;
6165	unsigned int cpu, oldcpu = (unsigned long)ocpu;
6166	struct softnet_data *sd, *oldsd;
6167
6168	if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6169		return NOTIFY_OK;
6170
6171	local_irq_disable();
6172	cpu = smp_processor_id();
6173	sd = &per_cpu(softnet_data, cpu);
6174	oldsd = &per_cpu(softnet_data, oldcpu);
6175
6176	/* Find end of our completion_queue. */
6177	list_skb = &sd->completion_queue;
6178	while (*list_skb)
6179		list_skb = &(*list_skb)->next;
6180	/* Append completion queue from offline CPU. */
6181	*list_skb = oldsd->completion_queue;
6182	oldsd->completion_queue = NULL;
6183
6184	/* Append output queue from offline CPU. */
6185	if (oldsd->output_queue) {
6186		*sd->output_queue_tailp = oldsd->output_queue;
6187		sd->output_queue_tailp = oldsd->output_queue_tailp;
6188		oldsd->output_queue = NULL;
6189		oldsd->output_queue_tailp = &oldsd->output_queue;
6190	}
6191	/* Append NAPI poll list from offline CPU. */
6192	if (!list_empty(&oldsd->poll_list)) {
6193		list_splice_init(&oldsd->poll_list, &sd->poll_list);
6194		raise_softirq_irqoff(NET_RX_SOFTIRQ);
6195	}
6196
6197	raise_softirq_irqoff(NET_TX_SOFTIRQ);
6198	local_irq_enable();
6199
6200	/* Process offline CPU's input_pkt_queue */
6201	while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6202		netif_rx(skb);
6203		input_queue_head_incr(oldsd);
6204	}
6205	while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6206		netif_rx(skb);
6207		input_queue_head_incr(oldsd);
6208	}
6209
6210	return NOTIFY_OK;
6211}
6212
6213
6214/**
6215 *	netdev_increment_features - increment feature set by one
6216 *	@all: current feature set
6217 *	@one: new feature set
6218 *	@mask: mask feature set
6219 *
6220 *	Computes a new feature set after adding a device with feature set
6221 *	@one to the master device with current feature set @all.  Will not
6222 *	enable anything that is off in @mask. Returns the new feature set.
6223 */
6224u32 netdev_increment_features(u32 all, u32 one, u32 mask)
 
6225{
6226	if (mask & NETIF_F_GEN_CSUM)
6227		mask |= NETIF_F_ALL_CSUM;
6228	mask |= NETIF_F_VLAN_CHALLENGED;
6229
6230	all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6231	all &= one | ~NETIF_F_ALL_FOR_ALL;
6232
6233	/* If device needs checksumming, downgrade to it. */
6234	if (all & (NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM))
6235		all &= ~NETIF_F_NO_CSUM;
6236
6237	/* If one device supports hw checksumming, set for all. */
6238	if (all & NETIF_F_GEN_CSUM)
6239		all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6240
6241	return all;
6242}
6243EXPORT_SYMBOL(netdev_increment_features);
6244
6245static struct hlist_head *netdev_create_hash(void)
6246{
6247	int i;
6248	struct hlist_head *hash;
6249
6250	hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6251	if (hash != NULL)
6252		for (i = 0; i < NETDEV_HASHENTRIES; i++)
6253			INIT_HLIST_HEAD(&hash[i]);
6254
6255	return hash;
6256}
6257
6258/* Initialize per network namespace state */
6259static int __net_init netdev_init(struct net *net)
6260{
6261	INIT_LIST_HEAD(&net->dev_base_head);
 
6262
6263	net->dev_name_head = netdev_create_hash();
6264	if (net->dev_name_head == NULL)
6265		goto err_name;
6266
6267	net->dev_index_head = netdev_create_hash();
6268	if (net->dev_index_head == NULL)
6269		goto err_idx;
6270
6271	return 0;
6272
6273err_idx:
6274	kfree(net->dev_name_head);
6275err_name:
6276	return -ENOMEM;
6277}
6278
6279/**
6280 *	netdev_drivername - network driver for the device
6281 *	@dev: network device
6282 *
6283 *	Determine network driver for device.
6284 */
6285const char *netdev_drivername(const struct net_device *dev)
6286{
6287	const struct device_driver *driver;
6288	const struct device *parent;
6289	const char *empty = "";
6290
6291	parent = dev->dev.parent;
6292	if (!parent)
6293		return empty;
6294
6295	driver = parent->driver;
6296	if (driver && driver->name)
6297		return driver->name;
6298	return empty;
6299}
6300
6301static int __netdev_printk(const char *level, const struct net_device *dev,
6302			   struct va_format *vaf)
6303{
6304	int r;
6305
6306	if (dev && dev->dev.parent)
6307		r = dev_printk(level, dev->dev.parent, "%s: %pV",
6308			       netdev_name(dev), vaf);
6309	else if (dev)
6310		r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6311	else
6312		r = printk("%s(NULL net_device): %pV", level, vaf);
6313
6314	return r;
6315}
 
6316
6317int netdev_printk(const char *level, const struct net_device *dev,
6318		  const char *format, ...)
6319{
6320	struct va_format vaf;
6321	va_list args;
6322	int r;
6323
6324	va_start(args, format);
6325
6326	vaf.fmt = format;
6327	vaf.va = &args;
6328
6329	r = __netdev_printk(level, dev, &vaf);
6330	va_end(args);
6331
6332	return r;
6333}
6334EXPORT_SYMBOL(netdev_printk);
6335
6336#define define_netdev_printk_level(func, level)			\
6337int func(const struct net_device *dev, const char *fmt, ...)	\
6338{								\
6339	int r;							\
6340	struct va_format vaf;					\
6341	va_list args;						\
6342								\
6343	va_start(args, fmt);					\
6344								\
6345	vaf.fmt = fmt;						\
6346	vaf.va = &args;						\
6347								\
6348	r = __netdev_printk(level, dev, &vaf);			\
6349	va_end(args);						\
6350								\
6351	return r;						\
6352}								\
6353EXPORT_SYMBOL(func);
6354
6355define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6356define_netdev_printk_level(netdev_alert, KERN_ALERT);
6357define_netdev_printk_level(netdev_crit, KERN_CRIT);
6358define_netdev_printk_level(netdev_err, KERN_ERR);
6359define_netdev_printk_level(netdev_warn, KERN_WARNING);
6360define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6361define_netdev_printk_level(netdev_info, KERN_INFO);
6362
6363static void __net_exit netdev_exit(struct net *net)
6364{
6365	kfree(net->dev_name_head);
6366	kfree(net->dev_index_head);
6367}
6368
6369static struct pernet_operations __net_initdata netdev_net_ops = {
6370	.init = netdev_init,
6371	.exit = netdev_exit,
6372};
6373
6374static void __net_exit default_device_exit(struct net *net)
6375{
6376	struct net_device *dev, *aux;
6377	/*
6378	 * Push all migratable network devices back to the
6379	 * initial network namespace
6380	 */
6381	rtnl_lock();
6382	for_each_netdev_safe(net, dev, aux) {
6383		int err;
6384		char fb_name[IFNAMSIZ];
6385
6386		/* Ignore unmoveable devices (i.e. loopback) */
6387		if (dev->features & NETIF_F_NETNS_LOCAL)
6388			continue;
6389
6390		/* Leave virtual devices for the generic cleanup */
6391		if (dev->rtnl_link_ops)
6392			continue;
6393
6394		/* Push remaining network devices to init_net */
6395		snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6396		err = dev_change_net_namespace(dev, &init_net, fb_name);
6397		if (err) {
6398			printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
6399				__func__, dev->name, err);
6400			BUG();
6401		}
6402	}
6403	rtnl_unlock();
6404}
6405
6406static void __net_exit default_device_exit_batch(struct list_head *net_list)
6407{
6408	/* At exit all network devices most be removed from a network
6409	 * namespace.  Do this in the reverse order of registration.
6410	 * Do this across as many network namespaces as possible to
6411	 * improve batching efficiency.
6412	 */
6413	struct net_device *dev;
6414	struct net *net;
6415	LIST_HEAD(dev_kill_list);
6416
6417	rtnl_lock();
6418	list_for_each_entry(net, net_list, exit_list) {
6419		for_each_netdev_reverse(net, dev) {
6420			if (dev->rtnl_link_ops)
6421				dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6422			else
6423				unregister_netdevice_queue(dev, &dev_kill_list);
6424		}
6425	}
6426	unregister_netdevice_many(&dev_kill_list);
6427	list_del(&dev_kill_list);
6428	rtnl_unlock();
6429}
6430
6431static struct pernet_operations __net_initdata default_device_ops = {
6432	.exit = default_device_exit,
6433	.exit_batch = default_device_exit_batch,
6434};
6435
6436/*
6437 *	Initialize the DEV module. At boot time this walks the device list and
6438 *	unhooks any devices that fail to initialise (normally hardware not
6439 *	present) and leaves us with a valid list of present and active devices.
6440 *
6441 */
6442
6443/*
6444 *       This is called single threaded during boot, so no need
6445 *       to take the rtnl semaphore.
6446 */
6447static int __init net_dev_init(void)
6448{
6449	int i, rc = -ENOMEM;
6450
6451	BUG_ON(!dev_boot_phase);
6452
6453	if (dev_proc_init())
6454		goto out;
6455
6456	if (netdev_kobject_init())
6457		goto out;
6458
6459	INIT_LIST_HEAD(&ptype_all);
6460	for (i = 0; i < PTYPE_HASH_SIZE; i++)
6461		INIT_LIST_HEAD(&ptype_base[i]);
6462
6463	if (register_pernet_subsys(&netdev_net_ops))
6464		goto out;
6465
6466	/*
6467	 *	Initialise the packet receive queues.
6468	 */
6469
6470	for_each_possible_cpu(i) {
6471		struct softnet_data *sd = &per_cpu(softnet_data, i);
6472
6473		memset(sd, 0, sizeof(*sd));
6474		skb_queue_head_init(&sd->input_pkt_queue);
6475		skb_queue_head_init(&sd->process_queue);
6476		sd->completion_queue = NULL;
6477		INIT_LIST_HEAD(&sd->poll_list);
6478		sd->output_queue = NULL;
6479		sd->output_queue_tailp = &sd->output_queue;
6480#ifdef CONFIG_RPS
6481		sd->csd.func = rps_trigger_softirq;
6482		sd->csd.info = sd;
6483		sd->csd.flags = 0;
6484		sd->cpu = i;
6485#endif
6486
6487		sd->backlog.poll = process_backlog;
6488		sd->backlog.weight = weight_p;
6489		sd->backlog.gro_list = NULL;
6490		sd->backlog.gro_count = 0;
6491	}
6492
6493	dev_boot_phase = 0;
6494
6495	/* The loopback device is special if any other network devices
6496	 * is present in a network namespace the loopback device must
6497	 * be present. Since we now dynamically allocate and free the
6498	 * loopback device ensure this invariant is maintained by
6499	 * keeping the loopback device as the first device on the
6500	 * list of network devices.  Ensuring the loopback devices
6501	 * is the first device that appears and the last network device
6502	 * that disappears.
6503	 */
6504	if (register_pernet_device(&loopback_net_ops))
6505		goto out;
6506
6507	if (register_pernet_device(&default_device_ops))
6508		goto out;
6509
6510	open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6511	open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6512
6513	hotcpu_notifier(dev_cpu_callback, 0);
6514	dst_init();
6515	dev_mcast_init();
6516	rc = 0;
6517out:
6518	return rc;
6519}
6520
6521subsys_initcall(net_dev_init);
6522
6523static int __init initialize_hashrnd(void)
6524{
6525	get_random_bytes(&hashrnd, sizeof(hashrnd));
6526	return 0;
6527}
6528
6529late_initcall_sync(initialize_hashrnd);
6530