Linux Audio

Check our new training course

Loading...
v3.1
   1/*
   2 * net-sysfs.c - network device class and attributes
   3 *
   4 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
   5 *
   6 *	This program is free software; you can redistribute it and/or
   7 *	modify it under the terms of the GNU General Public License
   8 *	as published by the Free Software Foundation; either version
   9 *	2 of the License, or (at your option) any later version.
  10 */
  11
  12#include <linux/capability.h>
  13#include <linux/kernel.h>
  14#include <linux/netdevice.h>
  15#include <linux/if_arp.h>
  16#include <linux/slab.h>
  17#include <linux/nsproxy.h>
  18#include <net/sock.h>
  19#include <net/net_namespace.h>
  20#include <linux/rtnetlink.h>
  21#include <linux/wireless.h>
  22#include <linux/vmalloc.h>
 
 
  23#include <net/wext.h>
  24
  25#include "net-sysfs.h"
  26
  27#ifdef CONFIG_SYSFS
  28static const char fmt_hex[] = "%#x\n";
  29static const char fmt_long_hex[] = "%#lx\n";
  30static const char fmt_dec[] = "%d\n";
  31static const char fmt_udec[] = "%u\n";
  32static const char fmt_ulong[] = "%lu\n";
  33static const char fmt_u64[] = "%llu\n";
  34
  35static inline int dev_isalive(const struct net_device *dev)
  36{
  37	return dev->reg_state <= NETREG_REGISTERED;
  38}
  39
  40/* use same locking rules as GIF* ioctl's */
  41static ssize_t netdev_show(const struct device *dev,
  42			   struct device_attribute *attr, char *buf,
  43			   ssize_t (*format)(const struct net_device *, char *))
  44{
  45	struct net_device *net = to_net_dev(dev);
  46	ssize_t ret = -EINVAL;
  47
  48	read_lock(&dev_base_lock);
  49	if (dev_isalive(net))
  50		ret = (*format)(net, buf);
  51	read_unlock(&dev_base_lock);
  52
  53	return ret;
  54}
  55
  56/* generate a show function for simple field */
  57#define NETDEVICE_SHOW(field, format_string)				\
  58static ssize_t format_##field(const struct net_device *net, char *buf)	\
  59{									\
  60	return sprintf(buf, format_string, net->field);			\
  61}									\
  62static ssize_t show_##field(struct device *dev,				\
  63			    struct device_attribute *attr, char *buf)	\
  64{									\
  65	return netdev_show(dev, attr, buf, format_##field);		\
  66}
  67
  68
  69/* use same locking and permission rules as SIF* ioctl's */
  70static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  71			    const char *buf, size_t len,
  72			    int (*set)(struct net_device *, unsigned long))
  73{
  74	struct net_device *net = to_net_dev(dev);
  75	char *endp;
  76	unsigned long new;
  77	int ret = -EINVAL;
  78
  79	if (!capable(CAP_NET_ADMIN))
  80		return -EPERM;
  81
  82	new = simple_strtoul(buf, &endp, 0);
  83	if (endp == buf)
  84		goto err;
  85
  86	if (!rtnl_trylock())
  87		return restart_syscall();
  88
  89	if (dev_isalive(net)) {
  90		if ((ret = (*set)(net, new)) == 0)
  91			ret = len;
  92	}
  93	rtnl_unlock();
  94 err:
  95	return ret;
  96}
  97
  98NETDEVICE_SHOW(dev_id, fmt_hex);
  99NETDEVICE_SHOW(addr_assign_type, fmt_dec);
 100NETDEVICE_SHOW(addr_len, fmt_dec);
 101NETDEVICE_SHOW(iflink, fmt_dec);
 102NETDEVICE_SHOW(ifindex, fmt_dec);
 103NETDEVICE_SHOW(type, fmt_dec);
 104NETDEVICE_SHOW(link_mode, fmt_dec);
 105
 106/* use same locking rules as GIFHWADDR ioctl's */
 107static ssize_t show_address(struct device *dev, struct device_attribute *attr,
 108			    char *buf)
 109{
 110	struct net_device *net = to_net_dev(dev);
 111	ssize_t ret = -EINVAL;
 112
 113	read_lock(&dev_base_lock);
 114	if (dev_isalive(net))
 115		ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
 116	read_unlock(&dev_base_lock);
 117	return ret;
 118}
 119
 120static ssize_t show_broadcast(struct device *dev,
 121			    struct device_attribute *attr, char *buf)
 122{
 123	struct net_device *net = to_net_dev(dev);
 124	if (dev_isalive(net))
 125		return sysfs_format_mac(buf, net->broadcast, net->addr_len);
 126	return -EINVAL;
 127}
 128
 129static ssize_t show_carrier(struct device *dev,
 130			    struct device_attribute *attr, char *buf)
 131{
 132	struct net_device *netdev = to_net_dev(dev);
 133	if (netif_running(netdev)) {
 134		return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 135	}
 136	return -EINVAL;
 137}
 138
 139static ssize_t show_speed(struct device *dev,
 140			  struct device_attribute *attr, char *buf)
 141{
 142	struct net_device *netdev = to_net_dev(dev);
 143	int ret = -EINVAL;
 144
 145	if (!rtnl_trylock())
 146		return restart_syscall();
 147
 148	if (netif_running(netdev)) {
 149		struct ethtool_cmd cmd;
 150		if (!dev_ethtool_get_settings(netdev, &cmd))
 151			ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
 152	}
 153	rtnl_unlock();
 154	return ret;
 155}
 156
 157static ssize_t show_duplex(struct device *dev,
 158			   struct device_attribute *attr, char *buf)
 159{
 160	struct net_device *netdev = to_net_dev(dev);
 161	int ret = -EINVAL;
 162
 163	if (!rtnl_trylock())
 164		return restart_syscall();
 165
 166	if (netif_running(netdev)) {
 167		struct ethtool_cmd cmd;
 168		if (!dev_ethtool_get_settings(netdev, &cmd))
 169			ret = sprintf(buf, "%s\n",
 170				      cmd.duplex ? "full" : "half");
 171	}
 172	rtnl_unlock();
 173	return ret;
 174}
 175
 176static ssize_t show_dormant(struct device *dev,
 177			    struct device_attribute *attr, char *buf)
 178{
 179	struct net_device *netdev = to_net_dev(dev);
 180
 181	if (netif_running(netdev))
 182		return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 183
 184	return -EINVAL;
 185}
 186
 187static const char *const operstates[] = {
 188	"unknown",
 189	"notpresent", /* currently unused */
 190	"down",
 191	"lowerlayerdown",
 192	"testing", /* currently unused */
 193	"dormant",
 194	"up"
 195};
 196
 197static ssize_t show_operstate(struct device *dev,
 198			      struct device_attribute *attr, char *buf)
 199{
 200	const struct net_device *netdev = to_net_dev(dev);
 201	unsigned char operstate;
 202
 203	read_lock(&dev_base_lock);
 204	operstate = netdev->operstate;
 205	if (!netif_running(netdev))
 206		operstate = IF_OPER_DOWN;
 207	read_unlock(&dev_base_lock);
 208
 209	if (operstate >= ARRAY_SIZE(operstates))
 210		return -EINVAL; /* should not happen */
 211
 212	return sprintf(buf, "%s\n", operstates[operstate]);
 213}
 214
 215/* read-write attributes */
 216NETDEVICE_SHOW(mtu, fmt_dec);
 217
 218static int change_mtu(struct net_device *net, unsigned long new_mtu)
 219{
 220	return dev_set_mtu(net, (int) new_mtu);
 221}
 222
 223static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
 224			 const char *buf, size_t len)
 225{
 226	return netdev_store(dev, attr, buf, len, change_mtu);
 227}
 228
 229NETDEVICE_SHOW(flags, fmt_hex);
 230
 231static int change_flags(struct net_device *net, unsigned long new_flags)
 232{
 233	return dev_change_flags(net, (unsigned) new_flags);
 234}
 235
 236static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
 237			   const char *buf, size_t len)
 238{
 239	return netdev_store(dev, attr, buf, len, change_flags);
 240}
 241
 242NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
 243
 244static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
 245{
 246	net->tx_queue_len = new_len;
 247	return 0;
 248}
 249
 250static ssize_t store_tx_queue_len(struct device *dev,
 251				  struct device_attribute *attr,
 252				  const char *buf, size_t len)
 253{
 254	return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 255}
 256
 257static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
 258			     const char *buf, size_t len)
 259{
 260	struct net_device *netdev = to_net_dev(dev);
 261	size_t count = len;
 262	ssize_t ret;
 263
 264	if (!capable(CAP_NET_ADMIN))
 265		return -EPERM;
 266
 267	/* ignore trailing newline */
 268	if (len >  0 && buf[len - 1] == '\n')
 269		--count;
 270
 271	if (!rtnl_trylock())
 272		return restart_syscall();
 273	ret = dev_set_alias(netdev, buf, count);
 274	rtnl_unlock();
 275
 276	return ret < 0 ? ret : len;
 277}
 278
 279static ssize_t show_ifalias(struct device *dev,
 280			    struct device_attribute *attr, char *buf)
 281{
 282	const struct net_device *netdev = to_net_dev(dev);
 283	ssize_t ret = 0;
 284
 285	if (!rtnl_trylock())
 286		return restart_syscall();
 287	if (netdev->ifalias)
 288		ret = sprintf(buf, "%s\n", netdev->ifalias);
 289	rtnl_unlock();
 290	return ret;
 291}
 292
 293NETDEVICE_SHOW(group, fmt_dec);
 294
 295static int change_group(struct net_device *net, unsigned long new_group)
 296{
 297	dev_set_group(net, (int) new_group);
 298	return 0;
 299}
 300
 301static ssize_t store_group(struct device *dev, struct device_attribute *attr,
 302			 const char *buf, size_t len)
 303{
 304	return netdev_store(dev, attr, buf, len, change_group);
 305}
 306
 307static struct device_attribute net_class_attributes[] = {
 308	__ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
 309	__ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
 310	__ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
 311	__ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
 312	__ATTR(iflink, S_IRUGO, show_iflink, NULL),
 313	__ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
 314	__ATTR(type, S_IRUGO, show_type, NULL),
 315	__ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
 316	__ATTR(address, S_IRUGO, show_address, NULL),
 317	__ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
 318	__ATTR(carrier, S_IRUGO, show_carrier, NULL),
 319	__ATTR(speed, S_IRUGO, show_speed, NULL),
 320	__ATTR(duplex, S_IRUGO, show_duplex, NULL),
 321	__ATTR(dormant, S_IRUGO, show_dormant, NULL),
 322	__ATTR(operstate, S_IRUGO, show_operstate, NULL),
 323	__ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
 324	__ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
 325	__ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
 326	       store_tx_queue_len),
 327	__ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
 328	{}
 329};
 330
 331/* Show a given an attribute in the statistics group */
 332static ssize_t netstat_show(const struct device *d,
 333			    struct device_attribute *attr, char *buf,
 334			    unsigned long offset)
 335{
 336	struct net_device *dev = to_net_dev(d);
 337	ssize_t ret = -EINVAL;
 338
 339	WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 340			offset % sizeof(u64) != 0);
 341
 342	read_lock(&dev_base_lock);
 343	if (dev_isalive(dev)) {
 344		struct rtnl_link_stats64 temp;
 345		const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 346
 347		ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 348	}
 349	read_unlock(&dev_base_lock);
 350	return ret;
 351}
 352
 353/* generate a read-only statistics attribute */
 354#define NETSTAT_ENTRY(name)						\
 355static ssize_t show_##name(struct device *d,				\
 356			   struct device_attribute *attr, char *buf) 	\
 357{									\
 358	return netstat_show(d, attr, buf,				\
 359			    offsetof(struct rtnl_link_stats64, name));	\
 360}									\
 361static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 362
 363NETSTAT_ENTRY(rx_packets);
 364NETSTAT_ENTRY(tx_packets);
 365NETSTAT_ENTRY(rx_bytes);
 366NETSTAT_ENTRY(tx_bytes);
 367NETSTAT_ENTRY(rx_errors);
 368NETSTAT_ENTRY(tx_errors);
 369NETSTAT_ENTRY(rx_dropped);
 370NETSTAT_ENTRY(tx_dropped);
 371NETSTAT_ENTRY(multicast);
 372NETSTAT_ENTRY(collisions);
 373NETSTAT_ENTRY(rx_length_errors);
 374NETSTAT_ENTRY(rx_over_errors);
 375NETSTAT_ENTRY(rx_crc_errors);
 376NETSTAT_ENTRY(rx_frame_errors);
 377NETSTAT_ENTRY(rx_fifo_errors);
 378NETSTAT_ENTRY(rx_missed_errors);
 379NETSTAT_ENTRY(tx_aborted_errors);
 380NETSTAT_ENTRY(tx_carrier_errors);
 381NETSTAT_ENTRY(tx_fifo_errors);
 382NETSTAT_ENTRY(tx_heartbeat_errors);
 383NETSTAT_ENTRY(tx_window_errors);
 384NETSTAT_ENTRY(rx_compressed);
 385NETSTAT_ENTRY(tx_compressed);
 386
 387static struct attribute *netstat_attrs[] = {
 388	&dev_attr_rx_packets.attr,
 389	&dev_attr_tx_packets.attr,
 390	&dev_attr_rx_bytes.attr,
 391	&dev_attr_tx_bytes.attr,
 392	&dev_attr_rx_errors.attr,
 393	&dev_attr_tx_errors.attr,
 394	&dev_attr_rx_dropped.attr,
 395	&dev_attr_tx_dropped.attr,
 396	&dev_attr_multicast.attr,
 397	&dev_attr_collisions.attr,
 398	&dev_attr_rx_length_errors.attr,
 399	&dev_attr_rx_over_errors.attr,
 400	&dev_attr_rx_crc_errors.attr,
 401	&dev_attr_rx_frame_errors.attr,
 402	&dev_attr_rx_fifo_errors.attr,
 403	&dev_attr_rx_missed_errors.attr,
 404	&dev_attr_tx_aborted_errors.attr,
 405	&dev_attr_tx_carrier_errors.attr,
 406	&dev_attr_tx_fifo_errors.attr,
 407	&dev_attr_tx_heartbeat_errors.attr,
 408	&dev_attr_tx_window_errors.attr,
 409	&dev_attr_rx_compressed.attr,
 410	&dev_attr_tx_compressed.attr,
 411	NULL
 412};
 413
 414
 415static struct attribute_group netstat_group = {
 416	.name  = "statistics",
 417	.attrs  = netstat_attrs,
 418};
 419
 420#ifdef CONFIG_WIRELESS_EXT_SYSFS
 421/* helper function that does all the locking etc for wireless stats */
 422static ssize_t wireless_show(struct device *d, char *buf,
 423			     ssize_t (*format)(const struct iw_statistics *,
 424					       char *))
 425{
 426	struct net_device *dev = to_net_dev(d);
 427	const struct iw_statistics *iw;
 428	ssize_t ret = -EINVAL;
 429
 430	if (!rtnl_trylock())
 431		return restart_syscall();
 432	if (dev_isalive(dev)) {
 433		iw = get_wireless_stats(dev);
 434		if (iw)
 435			ret = (*format)(iw, buf);
 436	}
 437	rtnl_unlock();
 438
 439	return ret;
 440}
 441
 442/* show function template for wireless fields */
 443#define WIRELESS_SHOW(name, field, format_string)			\
 444static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
 445{									\
 446	return sprintf(buf, format_string, iw->field);			\
 447}									\
 448static ssize_t show_iw_##name(struct device *d,				\
 449			      struct device_attribute *attr, char *buf)	\
 450{									\
 451	return wireless_show(d, buf, format_iw_##name);			\
 452}									\
 453static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
 454
 455WIRELESS_SHOW(status, status, fmt_hex);
 456WIRELESS_SHOW(link, qual.qual, fmt_dec);
 457WIRELESS_SHOW(level, qual.level, fmt_dec);
 458WIRELESS_SHOW(noise, qual.noise, fmt_dec);
 459WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
 460WIRELESS_SHOW(crypt, discard.code, fmt_dec);
 461WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
 462WIRELESS_SHOW(misc, discard.misc, fmt_dec);
 463WIRELESS_SHOW(retries, discard.retries, fmt_dec);
 464WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
 465
 466static struct attribute *wireless_attrs[] = {
 467	&dev_attr_status.attr,
 468	&dev_attr_link.attr,
 469	&dev_attr_level.attr,
 470	&dev_attr_noise.attr,
 471	&dev_attr_nwid.attr,
 472	&dev_attr_crypt.attr,
 473	&dev_attr_fragment.attr,
 474	&dev_attr_retries.attr,
 475	&dev_attr_misc.attr,
 476	&dev_attr_beacon.attr,
 477	NULL
 478};
 479
 480static struct attribute_group wireless_group = {
 481	.name = "wireless",
 482	.attrs = wireless_attrs,
 483};
 484#endif
 485#endif /* CONFIG_SYSFS */
 486
 487#ifdef CONFIG_RPS
 488/*
 489 * RX queue sysfs structures and functions.
 490 */
 491struct rx_queue_attribute {
 492	struct attribute attr;
 493	ssize_t (*show)(struct netdev_rx_queue *queue,
 494	    struct rx_queue_attribute *attr, char *buf);
 495	ssize_t (*store)(struct netdev_rx_queue *queue,
 496	    struct rx_queue_attribute *attr, const char *buf, size_t len);
 497};
 498#define to_rx_queue_attr(_attr) container_of(_attr,		\
 499    struct rx_queue_attribute, attr)
 500
 501#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 502
 503static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 504				  char *buf)
 505{
 506	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 507	struct netdev_rx_queue *queue = to_rx_queue(kobj);
 508
 509	if (!attribute->show)
 510		return -EIO;
 511
 512	return attribute->show(queue, attribute, buf);
 513}
 514
 515static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 516				   const char *buf, size_t count)
 517{
 518	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 519	struct netdev_rx_queue *queue = to_rx_queue(kobj);
 520
 521	if (!attribute->store)
 522		return -EIO;
 523
 524	return attribute->store(queue, attribute, buf, count);
 525}
 526
 527static const struct sysfs_ops rx_queue_sysfs_ops = {
 528	.show = rx_queue_attr_show,
 529	.store = rx_queue_attr_store,
 530};
 531
 532static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 533			    struct rx_queue_attribute *attribute, char *buf)
 534{
 535	struct rps_map *map;
 536	cpumask_var_t mask;
 537	size_t len = 0;
 538	int i;
 539
 540	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 541		return -ENOMEM;
 542
 543	rcu_read_lock();
 544	map = rcu_dereference(queue->rps_map);
 545	if (map)
 546		for (i = 0; i < map->len; i++)
 547			cpumask_set_cpu(map->cpus[i], mask);
 548
 549	len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 550	if (PAGE_SIZE - len < 3) {
 551		rcu_read_unlock();
 552		free_cpumask_var(mask);
 553		return -EINVAL;
 554	}
 555	rcu_read_unlock();
 556
 557	free_cpumask_var(mask);
 558	len += sprintf(buf + len, "\n");
 559	return len;
 560}
 561
 562static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 563		      struct rx_queue_attribute *attribute,
 564		      const char *buf, size_t len)
 565{
 566	struct rps_map *old_map, *map;
 567	cpumask_var_t mask;
 568	int err, cpu, i;
 569	static DEFINE_SPINLOCK(rps_map_lock);
 570
 571	if (!capable(CAP_NET_ADMIN))
 572		return -EPERM;
 573
 574	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 575		return -ENOMEM;
 576
 577	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 578	if (err) {
 579		free_cpumask_var(mask);
 580		return err;
 581	}
 582
 583	map = kzalloc(max_t(unsigned,
 584	    RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 585	    GFP_KERNEL);
 586	if (!map) {
 587		free_cpumask_var(mask);
 588		return -ENOMEM;
 589	}
 590
 591	i = 0;
 592	for_each_cpu_and(cpu, mask, cpu_online_mask)
 593		map->cpus[i++] = cpu;
 594
 595	if (i)
 596		map->len = i;
 597	else {
 598		kfree(map);
 599		map = NULL;
 600	}
 601
 602	spin_lock(&rps_map_lock);
 603	old_map = rcu_dereference_protected(queue->rps_map,
 604					    lockdep_is_held(&rps_map_lock));
 605	rcu_assign_pointer(queue->rps_map, map);
 606	spin_unlock(&rps_map_lock);
 607
 608	if (old_map)
 
 
 609		kfree_rcu(old_map, rcu);
 610
 
 611	free_cpumask_var(mask);
 612	return len;
 613}
 614
 615static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 616					   struct rx_queue_attribute *attr,
 617					   char *buf)
 618{
 619	struct rps_dev_flow_table *flow_table;
 620	unsigned int val = 0;
 621
 622	rcu_read_lock();
 623	flow_table = rcu_dereference(queue->rps_flow_table);
 624	if (flow_table)
 625		val = flow_table->mask + 1;
 626	rcu_read_unlock();
 627
 628	return sprintf(buf, "%u\n", val);
 629}
 630
 631static void rps_dev_flow_table_release_work(struct work_struct *work)
 632{
 633	struct rps_dev_flow_table *table = container_of(work,
 634	    struct rps_dev_flow_table, free_work);
 635
 636	vfree(table);
 637}
 638
 639static void rps_dev_flow_table_release(struct rcu_head *rcu)
 640{
 641	struct rps_dev_flow_table *table = container_of(rcu,
 642	    struct rps_dev_flow_table, rcu);
 643
 644	INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
 645	schedule_work(&table->free_work);
 646}
 647
 648static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 649				     struct rx_queue_attribute *attr,
 650				     const char *buf, size_t len)
 651{
 652	unsigned int count;
 653	char *endp;
 654	struct rps_dev_flow_table *table, *old_table;
 655	static DEFINE_SPINLOCK(rps_dev_flow_lock);
 
 656
 657	if (!capable(CAP_NET_ADMIN))
 658		return -EPERM;
 659
 660	count = simple_strtoul(buf, &endp, 0);
 661	if (endp == buf)
 662		return -EINVAL;
 663
 664	if (count) {
 665		int i;
 666
 667		if (count > 1<<30) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 668			/* Enforce a limit to prevent overflow */
 669			return -EINVAL;
 670		}
 671		count = roundup_pow_of_two(count);
 672		table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(count));
 673		if (!table)
 674			return -ENOMEM;
 675
 676		table->mask = count - 1;
 677		for (i = 0; i < count; i++)
 678			table->flows[i].cpu = RPS_NO_CPU;
 679	} else
 680		table = NULL;
 681
 682	spin_lock(&rps_dev_flow_lock);
 683	old_table = rcu_dereference_protected(queue->rps_flow_table,
 684					      lockdep_is_held(&rps_dev_flow_lock));
 685	rcu_assign_pointer(queue->rps_flow_table, table);
 686	spin_unlock(&rps_dev_flow_lock);
 687
 688	if (old_table)
 689		call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 690
 691	return len;
 692}
 693
 694static struct rx_queue_attribute rps_cpus_attribute =
 695	__ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 696
 697
 698static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 699	__ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 700	    show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 701
 702static struct attribute *rx_queue_default_attrs[] = {
 703	&rps_cpus_attribute.attr,
 704	&rps_dev_flow_table_cnt_attribute.attr,
 705	NULL
 706};
 707
 708static void rx_queue_release(struct kobject *kobj)
 709{
 710	struct netdev_rx_queue *queue = to_rx_queue(kobj);
 711	struct rps_map *map;
 712	struct rps_dev_flow_table *flow_table;
 713
 714
 715	map = rcu_dereference_raw(queue->rps_map);
 716	if (map) {
 717		RCU_INIT_POINTER(queue->rps_map, NULL);
 718		kfree_rcu(map, rcu);
 719	}
 720
 721	flow_table = rcu_dereference_raw(queue->rps_flow_table);
 722	if (flow_table) {
 723		RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 724		call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 725	}
 726
 727	memset(kobj, 0, sizeof(*kobj));
 728	dev_put(queue->dev);
 729}
 730
 731static struct kobj_type rx_queue_ktype = {
 732	.sysfs_ops = &rx_queue_sysfs_ops,
 733	.release = rx_queue_release,
 734	.default_attrs = rx_queue_default_attrs,
 735};
 736
 737static int rx_queue_add_kobject(struct net_device *net, int index)
 738{
 739	struct netdev_rx_queue *queue = net->_rx + index;
 740	struct kobject *kobj = &queue->kobj;
 741	int error = 0;
 742
 743	kobj->kset = net->queues_kset;
 744	error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 745	    "rx-%u", index);
 746	if (error) {
 747		kobject_put(kobj);
 748		return error;
 749	}
 750
 751	kobject_uevent(kobj, KOBJ_ADD);
 752	dev_hold(queue->dev);
 753
 754	return error;
 755}
 756#endif /* CONFIG_RPS */
 757
 758int
 759net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
 760{
 761#ifdef CONFIG_RPS
 762	int i;
 763	int error = 0;
 764
 765	for (i = old_num; i < new_num; i++) {
 766		error = rx_queue_add_kobject(net, i);
 767		if (error) {
 768			new_num = old_num;
 769			break;
 770		}
 771	}
 772
 773	while (--i >= new_num)
 774		kobject_put(&net->_rx[i].kobj);
 775
 776	return error;
 777#else
 778	return 0;
 779#endif
 780}
 781
 782#ifdef CONFIG_XPS
 783/*
 784 * netdev_queue sysfs structures and functions.
 785 */
 786struct netdev_queue_attribute {
 787	struct attribute attr;
 788	ssize_t (*show)(struct netdev_queue *queue,
 789	    struct netdev_queue_attribute *attr, char *buf);
 790	ssize_t (*store)(struct netdev_queue *queue,
 791	    struct netdev_queue_attribute *attr, const char *buf, size_t len);
 792};
 793#define to_netdev_queue_attr(_attr) container_of(_attr,		\
 794    struct netdev_queue_attribute, attr)
 795
 796#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 797
 798static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 799				      struct attribute *attr, char *buf)
 800{
 801	struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 802	struct netdev_queue *queue = to_netdev_queue(kobj);
 803
 804	if (!attribute->show)
 805		return -EIO;
 806
 807	return attribute->show(queue, attribute, buf);
 808}
 809
 810static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 811				       struct attribute *attr,
 812				       const char *buf, size_t count)
 813{
 814	struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 815	struct netdev_queue *queue = to_netdev_queue(kobj);
 816
 817	if (!attribute->store)
 818		return -EIO;
 819
 820	return attribute->store(queue, attribute, buf, count);
 821}
 822
 823static const struct sysfs_ops netdev_queue_sysfs_ops = {
 824	.show = netdev_queue_attr_show,
 825	.store = netdev_queue_attr_store,
 826};
 827
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 828static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 829{
 830	struct net_device *dev = queue->dev;
 831	int i;
 832
 833	for (i = 0; i < dev->num_tx_queues; i++)
 834		if (queue == &dev->_tx[i])
 835			break;
 836
 837	BUG_ON(i >= dev->num_tx_queues);
 838
 839	return i;
 840}
 841
 842
 843static ssize_t show_xps_map(struct netdev_queue *queue,
 844			    struct netdev_queue_attribute *attribute, char *buf)
 845{
 846	struct net_device *dev = queue->dev;
 847	struct xps_dev_maps *dev_maps;
 848	cpumask_var_t mask;
 849	unsigned long index;
 850	size_t len = 0;
 851	int i;
 852
 853	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 854		return -ENOMEM;
 855
 856	index = get_netdev_queue_index(queue);
 857
 858	rcu_read_lock();
 859	dev_maps = rcu_dereference(dev->xps_maps);
 860	if (dev_maps) {
 861		for_each_possible_cpu(i) {
 862			struct xps_map *map =
 863			    rcu_dereference(dev_maps->cpu_map[i]);
 864			if (map) {
 865				int j;
 866				for (j = 0; j < map->len; j++) {
 867					if (map->queues[j] == index) {
 868						cpumask_set_cpu(i, mask);
 869						break;
 870					}
 871				}
 872			}
 873		}
 874	}
 875	rcu_read_unlock();
 876
 877	len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 878	if (PAGE_SIZE - len < 3) {
 879		free_cpumask_var(mask);
 880		return -EINVAL;
 881	}
 882
 883	free_cpumask_var(mask);
 884	len += sprintf(buf + len, "\n");
 885	return len;
 886}
 887
 888static DEFINE_MUTEX(xps_map_mutex);
 889#define xmap_dereference(P)		\
 890	rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
 891
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 892static ssize_t store_xps_map(struct netdev_queue *queue,
 893		      struct netdev_queue_attribute *attribute,
 894		      const char *buf, size_t len)
 895{
 896	struct net_device *dev = queue->dev;
 897	cpumask_var_t mask;
 898	int err, i, cpu, pos, map_len, alloc_len, need_set;
 899	unsigned long index;
 900	struct xps_map *map, *new_map;
 901	struct xps_dev_maps *dev_maps, *new_dev_maps;
 902	int nonempty = 0;
 903	int numa_node = -2;
 904
 905	if (!capable(CAP_NET_ADMIN))
 906		return -EPERM;
 907
 908	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 909		return -ENOMEM;
 910
 911	index = get_netdev_queue_index(queue);
 912
 913	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 914	if (err) {
 915		free_cpumask_var(mask);
 916		return err;
 917	}
 918
 919	new_dev_maps = kzalloc(max_t(unsigned,
 920	    XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
 921	if (!new_dev_maps) {
 922		free_cpumask_var(mask);
 923		return -ENOMEM;
 924	}
 925
 926	mutex_lock(&xps_map_mutex);
 927
 928	dev_maps = xmap_dereference(dev->xps_maps);
 929
 930	for_each_possible_cpu(cpu) {
 931		map = dev_maps ?
 932			xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
 933		new_map = map;
 934		if (map) {
 935			for (pos = 0; pos < map->len; pos++)
 936				if (map->queues[pos] == index)
 937					break;
 938			map_len = map->len;
 939			alloc_len = map->alloc_len;
 940		} else
 941			pos = map_len = alloc_len = 0;
 942
 943		need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu);
 944#ifdef CONFIG_NUMA
 945		if (need_set) {
 946			if (numa_node == -2)
 947				numa_node = cpu_to_node(cpu);
 948			else if (numa_node != cpu_to_node(cpu))
 949				numa_node = -1;
 950		}
 951#endif
 952		if (need_set && pos >= map_len) {
 953			/* Need to add queue to this CPU's map */
 954			if (map_len >= alloc_len) {
 955				alloc_len = alloc_len ?
 956				    2 * alloc_len : XPS_MIN_MAP_ALLOC;
 957				new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
 958						       GFP_KERNEL,
 959						       cpu_to_node(cpu));
 960				if (!new_map)
 961					goto error;
 962				new_map->alloc_len = alloc_len;
 963				for (i = 0; i < map_len; i++)
 964					new_map->queues[i] = map->queues[i];
 965				new_map->len = map_len;
 966			}
 967			new_map->queues[new_map->len++] = index;
 968		} else if (!need_set && pos < map_len) {
 969			/* Need to remove queue from this CPU's map */
 970			if (map_len > 1)
 971				new_map->queues[pos] =
 972				    new_map->queues[--new_map->len];
 973			else
 974				new_map = NULL;
 975		}
 976		RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
 977	}
 978
 979	/* Cleanup old maps */
 980	for_each_possible_cpu(cpu) {
 981		map = dev_maps ?
 982			xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
 983		if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
 984			kfree_rcu(map, rcu);
 985		if (new_dev_maps->cpu_map[cpu])
 986			nonempty = 1;
 987	}
 988
 989	if (nonempty)
 990		rcu_assign_pointer(dev->xps_maps, new_dev_maps);
 991	else {
 992		kfree(new_dev_maps);
 993		rcu_assign_pointer(dev->xps_maps, NULL);
 994	}
 995
 996	if (dev_maps)
 997		kfree_rcu(dev_maps, rcu);
 998
 999	netdev_queue_numa_node_write(queue, (numa_node >= 0) ? numa_node :
1000					    NUMA_NO_NODE);
1001
1002	mutex_unlock(&xps_map_mutex);
1003
1004	free_cpumask_var(mask);
1005	return len;
1006
1007error:
1008	mutex_unlock(&xps_map_mutex);
1009
1010	if (new_dev_maps)
1011		for_each_possible_cpu(i)
1012			kfree(rcu_dereference_protected(
1013				new_dev_maps->cpu_map[i],
1014				1));
1015	kfree(new_dev_maps);
1016	free_cpumask_var(mask);
1017	return -ENOMEM;
1018}
1019
1020static struct netdev_queue_attribute xps_cpus_attribute =
1021    __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
 
1022
1023static struct attribute *netdev_queue_default_attrs[] = {
 
 
1024	&xps_cpus_attribute.attr,
 
1025	NULL
1026};
1027
1028static void netdev_queue_release(struct kobject *kobj)
1029{
1030	struct netdev_queue *queue = to_netdev_queue(kobj);
1031	struct net_device *dev = queue->dev;
1032	struct xps_dev_maps *dev_maps;
1033	struct xps_map *map;
1034	unsigned long index;
1035	int i, pos, nonempty = 0;
1036
1037	index = get_netdev_queue_index(queue);
1038
1039	mutex_lock(&xps_map_mutex);
1040	dev_maps = xmap_dereference(dev->xps_maps);
1041
1042	if (dev_maps) {
1043		for_each_possible_cpu(i) {
1044			map = xmap_dereference(dev_maps->cpu_map[i]);
1045			if (!map)
1046				continue;
1047
1048			for (pos = 0; pos < map->len; pos++)
1049				if (map->queues[pos] == index)
1050					break;
1051
1052			if (pos < map->len) {
1053				if (map->len > 1)
1054					map->queues[pos] =
1055					    map->queues[--map->len];
1056				else {
1057					RCU_INIT_POINTER(dev_maps->cpu_map[i],
1058					    NULL);
1059					kfree_rcu(map, rcu);
1060					map = NULL;
1061				}
1062			}
1063			if (map)
1064				nonempty = 1;
1065		}
1066
1067		if (!nonempty) {
1068			RCU_INIT_POINTER(dev->xps_maps, NULL);
1069			kfree_rcu(dev_maps, rcu);
1070		}
1071	}
1072
1073	mutex_unlock(&xps_map_mutex);
1074
1075	memset(kobj, 0, sizeof(*kobj));
1076	dev_put(queue->dev);
1077}
1078
1079static struct kobj_type netdev_queue_ktype = {
1080	.sysfs_ops = &netdev_queue_sysfs_ops,
1081	.release = netdev_queue_release,
1082	.default_attrs = netdev_queue_default_attrs,
1083};
1084
1085static int netdev_queue_add_kobject(struct net_device *net, int index)
1086{
1087	struct netdev_queue *queue = net->_tx + index;
1088	struct kobject *kobj = &queue->kobj;
1089	int error = 0;
1090
1091	kobj->kset = net->queues_kset;
1092	error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1093	    "tx-%u", index);
1094	if (error) {
1095		kobject_put(kobj);
1096		return error;
1097	}
 
 
 
 
1098
1099	kobject_uevent(kobj, KOBJ_ADD);
1100	dev_hold(queue->dev);
1101
 
 
 
1102	return error;
1103}
1104#endif /* CONFIG_XPS */
1105
1106int
1107netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1108{
1109#ifdef CONFIG_XPS
1110	int i;
1111	int error = 0;
1112
1113	for (i = old_num; i < new_num; i++) {
1114		error = netdev_queue_add_kobject(net, i);
1115		if (error) {
1116			new_num = old_num;
1117			break;
1118		}
1119	}
1120
1121	while (--i >= new_num)
1122		kobject_put(&net->_tx[i].kobj);
 
 
 
 
 
 
1123
1124	return error;
1125#else
1126	return 0;
1127#endif
1128}
1129
1130static int register_queue_kobjects(struct net_device *net)
1131{
1132	int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1133
1134#if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1135	net->queues_kset = kset_create_and_add("queues",
1136	    NULL, &net->dev.kobj);
1137	if (!net->queues_kset)
1138		return -ENOMEM;
1139#endif
1140
1141#ifdef CONFIG_RPS
1142	real_rx = net->real_num_rx_queues;
1143#endif
1144	real_tx = net->real_num_tx_queues;
1145
1146	error = net_rx_queue_update_kobjects(net, 0, real_rx);
1147	if (error)
1148		goto error;
1149	rxq = real_rx;
1150
1151	error = netdev_queue_update_kobjects(net, 0, real_tx);
1152	if (error)
1153		goto error;
1154	txq = real_tx;
1155
1156	return 0;
1157
1158error:
1159	netdev_queue_update_kobjects(net, txq, 0);
1160	net_rx_queue_update_kobjects(net, rxq, 0);
1161	return error;
1162}
1163
1164static void remove_queue_kobjects(struct net_device *net)
1165{
1166	int real_rx = 0, real_tx = 0;
1167
1168#ifdef CONFIG_RPS
1169	real_rx = net->real_num_rx_queues;
1170#endif
1171	real_tx = net->real_num_tx_queues;
1172
1173	net_rx_queue_update_kobjects(net, real_rx, 0);
1174	netdev_queue_update_kobjects(net, real_tx, 0);
1175#if defined(CONFIG_RPS) || defined(CONFIG_XPS)
1176	kset_unregister(net->queues_kset);
1177#endif
1178}
1179
1180static void *net_grab_current_ns(void)
1181{
1182	struct net *ns = current->nsproxy->net_ns;
1183#ifdef CONFIG_NET_NS
1184	if (ns)
1185		atomic_inc(&ns->passive);
1186#endif
1187	return ns;
1188}
1189
1190static const void *net_initial_ns(void)
1191{
1192	return &init_net;
1193}
1194
1195static const void *net_netlink_ns(struct sock *sk)
1196{
1197	return sock_net(sk);
1198}
1199
1200struct kobj_ns_type_operations net_ns_type_operations = {
1201	.type = KOBJ_NS_TYPE_NET,
1202	.grab_current_ns = net_grab_current_ns,
1203	.netlink_ns = net_netlink_ns,
1204	.initial_ns = net_initial_ns,
1205	.drop_ns = net_drop_ns,
1206};
1207EXPORT_SYMBOL_GPL(net_ns_type_operations);
1208
1209#ifdef CONFIG_HOTPLUG
1210static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1211{
1212	struct net_device *dev = to_net_dev(d);
1213	int retval;
1214
1215	/* pass interface to uevent. */
1216	retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1217	if (retval)
1218		goto exit;
1219
1220	/* pass ifindex to uevent.
1221	 * ifindex is useful as it won't change (interface name may change)
1222	 * and is what RtNetlink uses natively. */
1223	retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1224
1225exit:
1226	return retval;
1227}
1228#endif
1229
1230/*
1231 *	netdev_release -- destroy and free a dead device.
1232 *	Called when last reference to device kobject is gone.
1233 */
1234static void netdev_release(struct device *d)
1235{
1236	struct net_device *dev = to_net_dev(d);
1237
1238	BUG_ON(dev->reg_state != NETREG_RELEASED);
1239
1240	kfree(dev->ifalias);
1241	kfree((char *)dev - dev->padded);
1242}
1243
1244static const void *net_namespace(struct device *d)
1245{
1246	struct net_device *dev;
1247	dev = container_of(d, struct net_device, dev);
1248	return dev_net(dev);
1249}
1250
1251static struct class net_class = {
1252	.name = "net",
1253	.dev_release = netdev_release,
1254#ifdef CONFIG_SYSFS
1255	.dev_attrs = net_class_attributes,
1256#endif /* CONFIG_SYSFS */
1257#ifdef CONFIG_HOTPLUG
1258	.dev_uevent = netdev_uevent,
1259#endif
1260	.ns_type = &net_ns_type_operations,
1261	.namespace = net_namespace,
1262};
1263
1264/* Delete sysfs entries but hold kobject reference until after all
1265 * netdev references are gone.
1266 */
1267void netdev_unregister_kobject(struct net_device * net)
1268{
1269	struct device *dev = &(net->dev);
1270
1271	kobject_get(&dev->kobj);
1272
1273	remove_queue_kobjects(net);
1274
1275	device_del(dev);
1276}
1277
1278/* Create sysfs entries for network device. */
1279int netdev_register_kobject(struct net_device *net)
1280{
1281	struct device *dev = &(net->dev);
1282	const struct attribute_group **groups = net->sysfs_groups;
1283	int error = 0;
1284
1285	device_initialize(dev);
1286	dev->class = &net_class;
1287	dev->platform_data = net;
1288	dev->groups = groups;
1289
1290	dev_set_name(dev, "%s", net->name);
1291
1292#ifdef CONFIG_SYSFS
1293	/* Allow for a device specific group */
1294	if (*groups)
1295		groups++;
1296
1297	*groups++ = &netstat_group;
1298#ifdef CONFIG_WIRELESS_EXT_SYSFS
1299	if (net->ieee80211_ptr)
1300		*groups++ = &wireless_group;
1301#ifdef CONFIG_WIRELESS_EXT
1302	else if (net->wireless_handlers)
1303		*groups++ = &wireless_group;
1304#endif
1305#endif
1306#endif /* CONFIG_SYSFS */
1307
1308	error = device_add(dev);
1309	if (error)
1310		return error;
1311
1312	error = register_queue_kobjects(net);
1313	if (error) {
1314		device_del(dev);
1315		return error;
1316	}
1317
1318	return error;
1319}
1320
1321int netdev_class_create_file(struct class_attribute *class_attr)
1322{
1323	return class_create_file(&net_class, class_attr);
1324}
1325EXPORT_SYMBOL(netdev_class_create_file);
1326
1327void netdev_class_remove_file(struct class_attribute *class_attr)
1328{
1329	class_remove_file(&net_class, class_attr);
1330}
1331EXPORT_SYMBOL(netdev_class_remove_file);
1332
1333int netdev_kobject_init(void)
1334{
1335	kobj_ns_type_register(&net_ns_type_operations);
1336	return class_register(&net_class);
1337}
v3.5.6
   1/*
   2 * net-sysfs.c - network device class and attributes
   3 *
   4 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
   5 *
   6 *	This program is free software; you can redistribute it and/or
   7 *	modify it under the terms of the GNU General Public License
   8 *	as published by the Free Software Foundation; either version
   9 *	2 of the License, or (at your option) any later version.
  10 */
  11
  12#include <linux/capability.h>
  13#include <linux/kernel.h>
  14#include <linux/netdevice.h>
  15#include <linux/if_arp.h>
  16#include <linux/slab.h>
  17#include <linux/nsproxy.h>
  18#include <net/sock.h>
  19#include <net/net_namespace.h>
  20#include <linux/rtnetlink.h>
  21#include <linux/wireless.h>
  22#include <linux/vmalloc.h>
  23#include <linux/export.h>
  24#include <linux/jiffies.h>
  25#include <net/wext.h>
  26
  27#include "net-sysfs.h"
  28
  29#ifdef CONFIG_SYSFS
  30static const char fmt_hex[] = "%#x\n";
  31static const char fmt_long_hex[] = "%#lx\n";
  32static const char fmt_dec[] = "%d\n";
  33static const char fmt_udec[] = "%u\n";
  34static const char fmt_ulong[] = "%lu\n";
  35static const char fmt_u64[] = "%llu\n";
  36
  37static inline int dev_isalive(const struct net_device *dev)
  38{
  39	return dev->reg_state <= NETREG_REGISTERED;
  40}
  41
  42/* use same locking rules as GIF* ioctl's */
  43static ssize_t netdev_show(const struct device *dev,
  44			   struct device_attribute *attr, char *buf,
  45			   ssize_t (*format)(const struct net_device *, char *))
  46{
  47	struct net_device *net = to_net_dev(dev);
  48	ssize_t ret = -EINVAL;
  49
  50	read_lock(&dev_base_lock);
  51	if (dev_isalive(net))
  52		ret = (*format)(net, buf);
  53	read_unlock(&dev_base_lock);
  54
  55	return ret;
  56}
  57
  58/* generate a show function for simple field */
  59#define NETDEVICE_SHOW(field, format_string)				\
  60static ssize_t format_##field(const struct net_device *net, char *buf)	\
  61{									\
  62	return sprintf(buf, format_string, net->field);			\
  63}									\
  64static ssize_t show_##field(struct device *dev,				\
  65			    struct device_attribute *attr, char *buf)	\
  66{									\
  67	return netdev_show(dev, attr, buf, format_##field);		\
  68}
  69
  70
  71/* use same locking and permission rules as SIF* ioctl's */
  72static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  73			    const char *buf, size_t len,
  74			    int (*set)(struct net_device *, unsigned long))
  75{
  76	struct net_device *net = to_net_dev(dev);
 
  77	unsigned long new;
  78	int ret = -EINVAL;
  79
  80	if (!capable(CAP_NET_ADMIN))
  81		return -EPERM;
  82
  83	ret = kstrtoul(buf, 0, &new);
  84	if (ret)
  85		goto err;
  86
  87	if (!rtnl_trylock())
  88		return restart_syscall();
  89
  90	if (dev_isalive(net)) {
  91		if ((ret = (*set)(net, new)) == 0)
  92			ret = len;
  93	}
  94	rtnl_unlock();
  95 err:
  96	return ret;
  97}
  98
  99NETDEVICE_SHOW(dev_id, fmt_hex);
 100NETDEVICE_SHOW(addr_assign_type, fmt_dec);
 101NETDEVICE_SHOW(addr_len, fmt_dec);
 102NETDEVICE_SHOW(iflink, fmt_dec);
 103NETDEVICE_SHOW(ifindex, fmt_dec);
 104NETDEVICE_SHOW(type, fmt_dec);
 105NETDEVICE_SHOW(link_mode, fmt_dec);
 106
 107/* use same locking rules as GIFHWADDR ioctl's */
 108static ssize_t show_address(struct device *dev, struct device_attribute *attr,
 109			    char *buf)
 110{
 111	struct net_device *net = to_net_dev(dev);
 112	ssize_t ret = -EINVAL;
 113
 114	read_lock(&dev_base_lock);
 115	if (dev_isalive(net))
 116		ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
 117	read_unlock(&dev_base_lock);
 118	return ret;
 119}
 120
 121static ssize_t show_broadcast(struct device *dev,
 122			    struct device_attribute *attr, char *buf)
 123{
 124	struct net_device *net = to_net_dev(dev);
 125	if (dev_isalive(net))
 126		return sysfs_format_mac(buf, net->broadcast, net->addr_len);
 127	return -EINVAL;
 128}
 129
 130static ssize_t show_carrier(struct device *dev,
 131			    struct device_attribute *attr, char *buf)
 132{
 133	struct net_device *netdev = to_net_dev(dev);
 134	if (netif_running(netdev)) {
 135		return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
 136	}
 137	return -EINVAL;
 138}
 139
 140static ssize_t show_speed(struct device *dev,
 141			  struct device_attribute *attr, char *buf)
 142{
 143	struct net_device *netdev = to_net_dev(dev);
 144	int ret = -EINVAL;
 145
 146	if (!rtnl_trylock())
 147		return restart_syscall();
 148
 149	if (netif_running(netdev)) {
 150		struct ethtool_cmd cmd;
 151		if (!__ethtool_get_settings(netdev, &cmd))
 152			ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
 153	}
 154	rtnl_unlock();
 155	return ret;
 156}
 157
 158static ssize_t show_duplex(struct device *dev,
 159			   struct device_attribute *attr, char *buf)
 160{
 161	struct net_device *netdev = to_net_dev(dev);
 162	int ret = -EINVAL;
 163
 164	if (!rtnl_trylock())
 165		return restart_syscall();
 166
 167	if (netif_running(netdev)) {
 168		struct ethtool_cmd cmd;
 169		if (!__ethtool_get_settings(netdev, &cmd))
 170			ret = sprintf(buf, "%s\n",
 171				      cmd.duplex ? "full" : "half");
 172	}
 173	rtnl_unlock();
 174	return ret;
 175}
 176
 177static ssize_t show_dormant(struct device *dev,
 178			    struct device_attribute *attr, char *buf)
 179{
 180	struct net_device *netdev = to_net_dev(dev);
 181
 182	if (netif_running(netdev))
 183		return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
 184
 185	return -EINVAL;
 186}
 187
 188static const char *const operstates[] = {
 189	"unknown",
 190	"notpresent", /* currently unused */
 191	"down",
 192	"lowerlayerdown",
 193	"testing", /* currently unused */
 194	"dormant",
 195	"up"
 196};
 197
 198static ssize_t show_operstate(struct device *dev,
 199			      struct device_attribute *attr, char *buf)
 200{
 201	const struct net_device *netdev = to_net_dev(dev);
 202	unsigned char operstate;
 203
 204	read_lock(&dev_base_lock);
 205	operstate = netdev->operstate;
 206	if (!netif_running(netdev))
 207		operstate = IF_OPER_DOWN;
 208	read_unlock(&dev_base_lock);
 209
 210	if (operstate >= ARRAY_SIZE(operstates))
 211		return -EINVAL; /* should not happen */
 212
 213	return sprintf(buf, "%s\n", operstates[operstate]);
 214}
 215
 216/* read-write attributes */
 217NETDEVICE_SHOW(mtu, fmt_dec);
 218
 219static int change_mtu(struct net_device *net, unsigned long new_mtu)
 220{
 221	return dev_set_mtu(net, (int) new_mtu);
 222}
 223
 224static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
 225			 const char *buf, size_t len)
 226{
 227	return netdev_store(dev, attr, buf, len, change_mtu);
 228}
 229
 230NETDEVICE_SHOW(flags, fmt_hex);
 231
 232static int change_flags(struct net_device *net, unsigned long new_flags)
 233{
 234	return dev_change_flags(net, (unsigned int) new_flags);
 235}
 236
 237static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
 238			   const char *buf, size_t len)
 239{
 240	return netdev_store(dev, attr, buf, len, change_flags);
 241}
 242
 243NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
 244
 245static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
 246{
 247	net->tx_queue_len = new_len;
 248	return 0;
 249}
 250
 251static ssize_t store_tx_queue_len(struct device *dev,
 252				  struct device_attribute *attr,
 253				  const char *buf, size_t len)
 254{
 255	return netdev_store(dev, attr, buf, len, change_tx_queue_len);
 256}
 257
 258static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
 259			     const char *buf, size_t len)
 260{
 261	struct net_device *netdev = to_net_dev(dev);
 262	size_t count = len;
 263	ssize_t ret;
 264
 265	if (!capable(CAP_NET_ADMIN))
 266		return -EPERM;
 267
 268	/* ignore trailing newline */
 269	if (len >  0 && buf[len - 1] == '\n')
 270		--count;
 271
 272	if (!rtnl_trylock())
 273		return restart_syscall();
 274	ret = dev_set_alias(netdev, buf, count);
 275	rtnl_unlock();
 276
 277	return ret < 0 ? ret : len;
 278}
 279
 280static ssize_t show_ifalias(struct device *dev,
 281			    struct device_attribute *attr, char *buf)
 282{
 283	const struct net_device *netdev = to_net_dev(dev);
 284	ssize_t ret = 0;
 285
 286	if (!rtnl_trylock())
 287		return restart_syscall();
 288	if (netdev->ifalias)
 289		ret = sprintf(buf, "%s\n", netdev->ifalias);
 290	rtnl_unlock();
 291	return ret;
 292}
 293
 294NETDEVICE_SHOW(group, fmt_dec);
 295
 296static int change_group(struct net_device *net, unsigned long new_group)
 297{
 298	dev_set_group(net, (int) new_group);
 299	return 0;
 300}
 301
 302static ssize_t store_group(struct device *dev, struct device_attribute *attr,
 303			 const char *buf, size_t len)
 304{
 305	return netdev_store(dev, attr, buf, len, change_group);
 306}
 307
 308static struct device_attribute net_class_attributes[] = {
 309	__ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
 310	__ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
 311	__ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
 312	__ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
 313	__ATTR(iflink, S_IRUGO, show_iflink, NULL),
 314	__ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
 315	__ATTR(type, S_IRUGO, show_type, NULL),
 316	__ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
 317	__ATTR(address, S_IRUGO, show_address, NULL),
 318	__ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
 319	__ATTR(carrier, S_IRUGO, show_carrier, NULL),
 320	__ATTR(speed, S_IRUGO, show_speed, NULL),
 321	__ATTR(duplex, S_IRUGO, show_duplex, NULL),
 322	__ATTR(dormant, S_IRUGO, show_dormant, NULL),
 323	__ATTR(operstate, S_IRUGO, show_operstate, NULL),
 324	__ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
 325	__ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
 326	__ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
 327	       store_tx_queue_len),
 328	__ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
 329	{}
 330};
 331
 332/* Show a given an attribute in the statistics group */
 333static ssize_t netstat_show(const struct device *d,
 334			    struct device_attribute *attr, char *buf,
 335			    unsigned long offset)
 336{
 337	struct net_device *dev = to_net_dev(d);
 338	ssize_t ret = -EINVAL;
 339
 340	WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 341			offset % sizeof(u64) != 0);
 342
 343	read_lock(&dev_base_lock);
 344	if (dev_isalive(dev)) {
 345		struct rtnl_link_stats64 temp;
 346		const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 347
 348		ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
 349	}
 350	read_unlock(&dev_base_lock);
 351	return ret;
 352}
 353
 354/* generate a read-only statistics attribute */
 355#define NETSTAT_ENTRY(name)						\
 356static ssize_t show_##name(struct device *d,				\
 357			   struct device_attribute *attr, char *buf) 	\
 358{									\
 359	return netstat_show(d, attr, buf,				\
 360			    offsetof(struct rtnl_link_stats64, name));	\
 361}									\
 362static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 363
 364NETSTAT_ENTRY(rx_packets);
 365NETSTAT_ENTRY(tx_packets);
 366NETSTAT_ENTRY(rx_bytes);
 367NETSTAT_ENTRY(tx_bytes);
 368NETSTAT_ENTRY(rx_errors);
 369NETSTAT_ENTRY(tx_errors);
 370NETSTAT_ENTRY(rx_dropped);
 371NETSTAT_ENTRY(tx_dropped);
 372NETSTAT_ENTRY(multicast);
 373NETSTAT_ENTRY(collisions);
 374NETSTAT_ENTRY(rx_length_errors);
 375NETSTAT_ENTRY(rx_over_errors);
 376NETSTAT_ENTRY(rx_crc_errors);
 377NETSTAT_ENTRY(rx_frame_errors);
 378NETSTAT_ENTRY(rx_fifo_errors);
 379NETSTAT_ENTRY(rx_missed_errors);
 380NETSTAT_ENTRY(tx_aborted_errors);
 381NETSTAT_ENTRY(tx_carrier_errors);
 382NETSTAT_ENTRY(tx_fifo_errors);
 383NETSTAT_ENTRY(tx_heartbeat_errors);
 384NETSTAT_ENTRY(tx_window_errors);
 385NETSTAT_ENTRY(rx_compressed);
 386NETSTAT_ENTRY(tx_compressed);
 387
 388static struct attribute *netstat_attrs[] = {
 389	&dev_attr_rx_packets.attr,
 390	&dev_attr_tx_packets.attr,
 391	&dev_attr_rx_bytes.attr,
 392	&dev_attr_tx_bytes.attr,
 393	&dev_attr_rx_errors.attr,
 394	&dev_attr_tx_errors.attr,
 395	&dev_attr_rx_dropped.attr,
 396	&dev_attr_tx_dropped.attr,
 397	&dev_attr_multicast.attr,
 398	&dev_attr_collisions.attr,
 399	&dev_attr_rx_length_errors.attr,
 400	&dev_attr_rx_over_errors.attr,
 401	&dev_attr_rx_crc_errors.attr,
 402	&dev_attr_rx_frame_errors.attr,
 403	&dev_attr_rx_fifo_errors.attr,
 404	&dev_attr_rx_missed_errors.attr,
 405	&dev_attr_tx_aborted_errors.attr,
 406	&dev_attr_tx_carrier_errors.attr,
 407	&dev_attr_tx_fifo_errors.attr,
 408	&dev_attr_tx_heartbeat_errors.attr,
 409	&dev_attr_tx_window_errors.attr,
 410	&dev_attr_rx_compressed.attr,
 411	&dev_attr_tx_compressed.attr,
 412	NULL
 413};
 414
 415
 416static struct attribute_group netstat_group = {
 417	.name  = "statistics",
 418	.attrs  = netstat_attrs,
 419};
 420
 421#ifdef CONFIG_WIRELESS_EXT_SYSFS
 422/* helper function that does all the locking etc for wireless stats */
 423static ssize_t wireless_show(struct device *d, char *buf,
 424			     ssize_t (*format)(const struct iw_statistics *,
 425					       char *))
 426{
 427	struct net_device *dev = to_net_dev(d);
 428	const struct iw_statistics *iw;
 429	ssize_t ret = -EINVAL;
 430
 431	if (!rtnl_trylock())
 432		return restart_syscall();
 433	if (dev_isalive(dev)) {
 434		iw = get_wireless_stats(dev);
 435		if (iw)
 436			ret = (*format)(iw, buf);
 437	}
 438	rtnl_unlock();
 439
 440	return ret;
 441}
 442
 443/* show function template for wireless fields */
 444#define WIRELESS_SHOW(name, field, format_string)			\
 445static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
 446{									\
 447	return sprintf(buf, format_string, iw->field);			\
 448}									\
 449static ssize_t show_iw_##name(struct device *d,				\
 450			      struct device_attribute *attr, char *buf)	\
 451{									\
 452	return wireless_show(d, buf, format_iw_##name);			\
 453}									\
 454static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
 455
 456WIRELESS_SHOW(status, status, fmt_hex);
 457WIRELESS_SHOW(link, qual.qual, fmt_dec);
 458WIRELESS_SHOW(level, qual.level, fmt_dec);
 459WIRELESS_SHOW(noise, qual.noise, fmt_dec);
 460WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
 461WIRELESS_SHOW(crypt, discard.code, fmt_dec);
 462WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
 463WIRELESS_SHOW(misc, discard.misc, fmt_dec);
 464WIRELESS_SHOW(retries, discard.retries, fmt_dec);
 465WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
 466
 467static struct attribute *wireless_attrs[] = {
 468	&dev_attr_status.attr,
 469	&dev_attr_link.attr,
 470	&dev_attr_level.attr,
 471	&dev_attr_noise.attr,
 472	&dev_attr_nwid.attr,
 473	&dev_attr_crypt.attr,
 474	&dev_attr_fragment.attr,
 475	&dev_attr_retries.attr,
 476	&dev_attr_misc.attr,
 477	&dev_attr_beacon.attr,
 478	NULL
 479};
 480
 481static struct attribute_group wireless_group = {
 482	.name = "wireless",
 483	.attrs = wireless_attrs,
 484};
 485#endif
 486#endif /* CONFIG_SYSFS */
 487
 488#ifdef CONFIG_RPS
 489/*
 490 * RX queue sysfs structures and functions.
 491 */
 492struct rx_queue_attribute {
 493	struct attribute attr;
 494	ssize_t (*show)(struct netdev_rx_queue *queue,
 495	    struct rx_queue_attribute *attr, char *buf);
 496	ssize_t (*store)(struct netdev_rx_queue *queue,
 497	    struct rx_queue_attribute *attr, const char *buf, size_t len);
 498};
 499#define to_rx_queue_attr(_attr) container_of(_attr,		\
 500    struct rx_queue_attribute, attr)
 501
 502#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 503
 504static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 505				  char *buf)
 506{
 507	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 508	struct netdev_rx_queue *queue = to_rx_queue(kobj);
 509
 510	if (!attribute->show)
 511		return -EIO;
 512
 513	return attribute->show(queue, attribute, buf);
 514}
 515
 516static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 517				   const char *buf, size_t count)
 518{
 519	struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 520	struct netdev_rx_queue *queue = to_rx_queue(kobj);
 521
 522	if (!attribute->store)
 523		return -EIO;
 524
 525	return attribute->store(queue, attribute, buf, count);
 526}
 527
 528static const struct sysfs_ops rx_queue_sysfs_ops = {
 529	.show = rx_queue_attr_show,
 530	.store = rx_queue_attr_store,
 531};
 532
 533static ssize_t show_rps_map(struct netdev_rx_queue *queue,
 534			    struct rx_queue_attribute *attribute, char *buf)
 535{
 536	struct rps_map *map;
 537	cpumask_var_t mask;
 538	size_t len = 0;
 539	int i;
 540
 541	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 542		return -ENOMEM;
 543
 544	rcu_read_lock();
 545	map = rcu_dereference(queue->rps_map);
 546	if (map)
 547		for (i = 0; i < map->len; i++)
 548			cpumask_set_cpu(map->cpus[i], mask);
 549
 550	len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
 551	if (PAGE_SIZE - len < 3) {
 552		rcu_read_unlock();
 553		free_cpumask_var(mask);
 554		return -EINVAL;
 555	}
 556	rcu_read_unlock();
 557
 558	free_cpumask_var(mask);
 559	len += sprintf(buf + len, "\n");
 560	return len;
 561}
 562
 563static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 564		      struct rx_queue_attribute *attribute,
 565		      const char *buf, size_t len)
 566{
 567	struct rps_map *old_map, *map;
 568	cpumask_var_t mask;
 569	int err, cpu, i;
 570	static DEFINE_SPINLOCK(rps_map_lock);
 571
 572	if (!capable(CAP_NET_ADMIN))
 573		return -EPERM;
 574
 575	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 576		return -ENOMEM;
 577
 578	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 579	if (err) {
 580		free_cpumask_var(mask);
 581		return err;
 582	}
 583
 584	map = kzalloc(max_t(unsigned int,
 585	    RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 586	    GFP_KERNEL);
 587	if (!map) {
 588		free_cpumask_var(mask);
 589		return -ENOMEM;
 590	}
 591
 592	i = 0;
 593	for_each_cpu_and(cpu, mask, cpu_online_mask)
 594		map->cpus[i++] = cpu;
 595
 596	if (i)
 597		map->len = i;
 598	else {
 599		kfree(map);
 600		map = NULL;
 601	}
 602
 603	spin_lock(&rps_map_lock);
 604	old_map = rcu_dereference_protected(queue->rps_map,
 605					    lockdep_is_held(&rps_map_lock));
 606	rcu_assign_pointer(queue->rps_map, map);
 607	spin_unlock(&rps_map_lock);
 608
 609	if (map)
 610		static_key_slow_inc(&rps_needed);
 611	if (old_map) {
 612		kfree_rcu(old_map, rcu);
 613		static_key_slow_dec(&rps_needed);
 614	}
 615	free_cpumask_var(mask);
 616	return len;
 617}
 618
 619static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 620					   struct rx_queue_attribute *attr,
 621					   char *buf)
 622{
 623	struct rps_dev_flow_table *flow_table;
 624	unsigned long val = 0;
 625
 626	rcu_read_lock();
 627	flow_table = rcu_dereference(queue->rps_flow_table);
 628	if (flow_table)
 629		val = (unsigned long)flow_table->mask + 1;
 630	rcu_read_unlock();
 631
 632	return sprintf(buf, "%lu\n", val);
 633}
 634
 635static void rps_dev_flow_table_release_work(struct work_struct *work)
 636{
 637	struct rps_dev_flow_table *table = container_of(work,
 638	    struct rps_dev_flow_table, free_work);
 639
 640	vfree(table);
 641}
 642
 643static void rps_dev_flow_table_release(struct rcu_head *rcu)
 644{
 645	struct rps_dev_flow_table *table = container_of(rcu,
 646	    struct rps_dev_flow_table, rcu);
 647
 648	INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
 649	schedule_work(&table->free_work);
 650}
 651
 652static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 653				     struct rx_queue_attribute *attr,
 654				     const char *buf, size_t len)
 655{
 656	unsigned long mask, count;
 
 657	struct rps_dev_flow_table *table, *old_table;
 658	static DEFINE_SPINLOCK(rps_dev_flow_lock);
 659	int rc;
 660
 661	if (!capable(CAP_NET_ADMIN))
 662		return -EPERM;
 663
 664	rc = kstrtoul(buf, 0, &count);
 665	if (rc < 0)
 666		return rc;
 667
 668	if (count) {
 669		mask = count - 1;
 670		/* mask = roundup_pow_of_two(count) - 1;
 671		 * without overflows...
 672		 */
 673		while ((mask | (mask >> 1)) != mask)
 674			mask |= (mask >> 1);
 675		/* On 64 bit arches, must check mask fits in table->mask (u32),
 676		 * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
 677		 * doesnt overflow.
 678		 */
 679#if BITS_PER_LONG > 32
 680		if (mask > (unsigned long)(u32)mask)
 681			return -EINVAL;
 682#else
 683		if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 684				/ sizeof(struct rps_dev_flow)) {
 685			/* Enforce a limit to prevent overflow */
 686			return -EINVAL;
 687		}
 688#endif
 689		table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 690		if (!table)
 691			return -ENOMEM;
 692
 693		table->mask = mask;
 694		for (count = 0; count <= mask; count++)
 695			table->flows[count].cpu = RPS_NO_CPU;
 696	} else
 697		table = NULL;
 698
 699	spin_lock(&rps_dev_flow_lock);
 700	old_table = rcu_dereference_protected(queue->rps_flow_table,
 701					      lockdep_is_held(&rps_dev_flow_lock));
 702	rcu_assign_pointer(queue->rps_flow_table, table);
 703	spin_unlock(&rps_dev_flow_lock);
 704
 705	if (old_table)
 706		call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 707
 708	return len;
 709}
 710
 711static struct rx_queue_attribute rps_cpus_attribute =
 712	__ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
 713
 714
 715static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
 716	__ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
 717	    show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 718
 719static struct attribute *rx_queue_default_attrs[] = {
 720	&rps_cpus_attribute.attr,
 721	&rps_dev_flow_table_cnt_attribute.attr,
 722	NULL
 723};
 724
 725static void rx_queue_release(struct kobject *kobj)
 726{
 727	struct netdev_rx_queue *queue = to_rx_queue(kobj);
 728	struct rps_map *map;
 729	struct rps_dev_flow_table *flow_table;
 730
 731
 732	map = rcu_dereference_protected(queue->rps_map, 1);
 733	if (map) {
 734		RCU_INIT_POINTER(queue->rps_map, NULL);
 735		kfree_rcu(map, rcu);
 736	}
 737
 738	flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
 739	if (flow_table) {
 740		RCU_INIT_POINTER(queue->rps_flow_table, NULL);
 741		call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
 742	}
 743
 744	memset(kobj, 0, sizeof(*kobj));
 745	dev_put(queue->dev);
 746}
 747
 748static struct kobj_type rx_queue_ktype = {
 749	.sysfs_ops = &rx_queue_sysfs_ops,
 750	.release = rx_queue_release,
 751	.default_attrs = rx_queue_default_attrs,
 752};
 753
 754static int rx_queue_add_kobject(struct net_device *net, int index)
 755{
 756	struct netdev_rx_queue *queue = net->_rx + index;
 757	struct kobject *kobj = &queue->kobj;
 758	int error = 0;
 759
 760	kobj->kset = net->queues_kset;
 761	error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
 762	    "rx-%u", index);
 763	if (error) {
 764		kobject_put(kobj);
 765		return error;
 766	}
 767
 768	kobject_uevent(kobj, KOBJ_ADD);
 769	dev_hold(queue->dev);
 770
 771	return error;
 772}
 773#endif /* CONFIG_RPS */
 774
 775int
 776net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
 777{
 778#ifdef CONFIG_RPS
 779	int i;
 780	int error = 0;
 781
 782	for (i = old_num; i < new_num; i++) {
 783		error = rx_queue_add_kobject(net, i);
 784		if (error) {
 785			new_num = old_num;
 786			break;
 787		}
 788	}
 789
 790	while (--i >= new_num)
 791		kobject_put(&net->_rx[i].kobj);
 792
 793	return error;
 794#else
 795	return 0;
 796#endif
 797}
 798
 799#ifdef CONFIG_SYSFS
 800/*
 801 * netdev_queue sysfs structures and functions.
 802 */
 803struct netdev_queue_attribute {
 804	struct attribute attr;
 805	ssize_t (*show)(struct netdev_queue *queue,
 806	    struct netdev_queue_attribute *attr, char *buf);
 807	ssize_t (*store)(struct netdev_queue *queue,
 808	    struct netdev_queue_attribute *attr, const char *buf, size_t len);
 809};
 810#define to_netdev_queue_attr(_attr) container_of(_attr,		\
 811    struct netdev_queue_attribute, attr)
 812
 813#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
 814
 815static ssize_t netdev_queue_attr_show(struct kobject *kobj,
 816				      struct attribute *attr, char *buf)
 817{
 818	struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 819	struct netdev_queue *queue = to_netdev_queue(kobj);
 820
 821	if (!attribute->show)
 822		return -EIO;
 823
 824	return attribute->show(queue, attribute, buf);
 825}
 826
 827static ssize_t netdev_queue_attr_store(struct kobject *kobj,
 828				       struct attribute *attr,
 829				       const char *buf, size_t count)
 830{
 831	struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
 832	struct netdev_queue *queue = to_netdev_queue(kobj);
 833
 834	if (!attribute->store)
 835		return -EIO;
 836
 837	return attribute->store(queue, attribute, buf, count);
 838}
 839
 840static const struct sysfs_ops netdev_queue_sysfs_ops = {
 841	.show = netdev_queue_attr_show,
 842	.store = netdev_queue_attr_store,
 843};
 844
 845static ssize_t show_trans_timeout(struct netdev_queue *queue,
 846				  struct netdev_queue_attribute *attribute,
 847				  char *buf)
 848{
 849	unsigned long trans_timeout;
 850
 851	spin_lock_irq(&queue->_xmit_lock);
 852	trans_timeout = queue->trans_timeout;
 853	spin_unlock_irq(&queue->_xmit_lock);
 854
 855	return sprintf(buf, "%lu", trans_timeout);
 856}
 857
 858static struct netdev_queue_attribute queue_trans_timeout =
 859	__ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
 860
 861#ifdef CONFIG_BQL
 862/*
 863 * Byte queue limits sysfs structures and functions.
 864 */
 865static ssize_t bql_show(char *buf, unsigned int value)
 866{
 867	return sprintf(buf, "%u\n", value);
 868}
 869
 870static ssize_t bql_set(const char *buf, const size_t count,
 871		       unsigned int *pvalue)
 872{
 873	unsigned int value;
 874	int err;
 875
 876	if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
 877		value = DQL_MAX_LIMIT;
 878	else {
 879		err = kstrtouint(buf, 10, &value);
 880		if (err < 0)
 881			return err;
 882		if (value > DQL_MAX_LIMIT)
 883			return -EINVAL;
 884	}
 885
 886	*pvalue = value;
 887
 888	return count;
 889}
 890
 891static ssize_t bql_show_hold_time(struct netdev_queue *queue,
 892				  struct netdev_queue_attribute *attr,
 893				  char *buf)
 894{
 895	struct dql *dql = &queue->dql;
 896
 897	return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
 898}
 899
 900static ssize_t bql_set_hold_time(struct netdev_queue *queue,
 901				 struct netdev_queue_attribute *attribute,
 902				 const char *buf, size_t len)
 903{
 904	struct dql *dql = &queue->dql;
 905	unsigned int value;
 906	int err;
 907
 908	err = kstrtouint(buf, 10, &value);
 909	if (err < 0)
 910		return err;
 911
 912	dql->slack_hold_time = msecs_to_jiffies(value);
 913
 914	return len;
 915}
 916
 917static struct netdev_queue_attribute bql_hold_time_attribute =
 918	__ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
 919	    bql_set_hold_time);
 920
 921static ssize_t bql_show_inflight(struct netdev_queue *queue,
 922				 struct netdev_queue_attribute *attr,
 923				 char *buf)
 924{
 925	struct dql *dql = &queue->dql;
 926
 927	return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
 928}
 929
 930static struct netdev_queue_attribute bql_inflight_attribute =
 931	__ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
 932
 933#define BQL_ATTR(NAME, FIELD)						\
 934static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,		\
 935				 struct netdev_queue_attribute *attr,	\
 936				 char *buf)				\
 937{									\
 938	return bql_show(buf, queue->dql.FIELD);				\
 939}									\
 940									\
 941static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,		\
 942				struct netdev_queue_attribute *attr,	\
 943				const char *buf, size_t len)		\
 944{									\
 945	return bql_set(buf, len, &queue->dql.FIELD);			\
 946}									\
 947									\
 948static struct netdev_queue_attribute bql_ ## NAME ## _attribute =	\
 949	__ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME,		\
 950	    bql_set_ ## NAME);
 951
 952BQL_ATTR(limit, limit)
 953BQL_ATTR(limit_max, max_limit)
 954BQL_ATTR(limit_min, min_limit)
 955
 956static struct attribute *dql_attrs[] = {
 957	&bql_limit_attribute.attr,
 958	&bql_limit_max_attribute.attr,
 959	&bql_limit_min_attribute.attr,
 960	&bql_hold_time_attribute.attr,
 961	&bql_inflight_attribute.attr,
 962	NULL
 963};
 964
 965static struct attribute_group dql_group = {
 966	.name  = "byte_queue_limits",
 967	.attrs  = dql_attrs,
 968};
 969#endif /* CONFIG_BQL */
 970
 971#ifdef CONFIG_XPS
 972static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
 973{
 974	struct net_device *dev = queue->dev;
 975	int i;
 976
 977	for (i = 0; i < dev->num_tx_queues; i++)
 978		if (queue == &dev->_tx[i])
 979			break;
 980
 981	BUG_ON(i >= dev->num_tx_queues);
 982
 983	return i;
 984}
 985
 986
 987static ssize_t show_xps_map(struct netdev_queue *queue,
 988			    struct netdev_queue_attribute *attribute, char *buf)
 989{
 990	struct net_device *dev = queue->dev;
 991	struct xps_dev_maps *dev_maps;
 992	cpumask_var_t mask;
 993	unsigned long index;
 994	size_t len = 0;
 995	int i;
 996
 997	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 998		return -ENOMEM;
 999
1000	index = get_netdev_queue_index(queue);
1001
1002	rcu_read_lock();
1003	dev_maps = rcu_dereference(dev->xps_maps);
1004	if (dev_maps) {
1005		for_each_possible_cpu(i) {
1006			struct xps_map *map =
1007			    rcu_dereference(dev_maps->cpu_map[i]);
1008			if (map) {
1009				int j;
1010				for (j = 0; j < map->len; j++) {
1011					if (map->queues[j] == index) {
1012						cpumask_set_cpu(i, mask);
1013						break;
1014					}
1015				}
1016			}
1017		}
1018	}
1019	rcu_read_unlock();
1020
1021	len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
1022	if (PAGE_SIZE - len < 3) {
1023		free_cpumask_var(mask);
1024		return -EINVAL;
1025	}
1026
1027	free_cpumask_var(mask);
1028	len += sprintf(buf + len, "\n");
1029	return len;
1030}
1031
1032static DEFINE_MUTEX(xps_map_mutex);
1033#define xmap_dereference(P)		\
1034	rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1035
1036static void xps_queue_release(struct netdev_queue *queue)
1037{
1038	struct net_device *dev = queue->dev;
1039	struct xps_dev_maps *dev_maps;
1040	struct xps_map *map;
1041	unsigned long index;
1042	int i, pos, nonempty = 0;
1043
1044	index = get_netdev_queue_index(queue);
1045
1046	mutex_lock(&xps_map_mutex);
1047	dev_maps = xmap_dereference(dev->xps_maps);
1048
1049	if (dev_maps) {
1050		for_each_possible_cpu(i) {
1051			map = xmap_dereference(dev_maps->cpu_map[i]);
1052			if (!map)
1053				continue;
1054
1055			for (pos = 0; pos < map->len; pos++)
1056				if (map->queues[pos] == index)
1057					break;
1058
1059			if (pos < map->len) {
1060				if (map->len > 1)
1061					map->queues[pos] =
1062					    map->queues[--map->len];
1063				else {
1064					RCU_INIT_POINTER(dev_maps->cpu_map[i],
1065					    NULL);
1066					kfree_rcu(map, rcu);
1067					map = NULL;
1068				}
1069			}
1070			if (map)
1071				nonempty = 1;
1072		}
1073
1074		if (!nonempty) {
1075			RCU_INIT_POINTER(dev->xps_maps, NULL);
1076			kfree_rcu(dev_maps, rcu);
1077		}
1078	}
1079	mutex_unlock(&xps_map_mutex);
1080}
1081
1082static ssize_t store_xps_map(struct netdev_queue *queue,
1083		      struct netdev_queue_attribute *attribute,
1084		      const char *buf, size_t len)
1085{
1086	struct net_device *dev = queue->dev;
1087	cpumask_var_t mask;
1088	int err, i, cpu, pos, map_len, alloc_len, need_set;
1089	unsigned long index;
1090	struct xps_map *map, *new_map;
1091	struct xps_dev_maps *dev_maps, *new_dev_maps;
1092	int nonempty = 0;
1093	int numa_node_id = -2;
1094
1095	if (!capable(CAP_NET_ADMIN))
1096		return -EPERM;
1097
1098	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1099		return -ENOMEM;
1100
1101	index = get_netdev_queue_index(queue);
1102
1103	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1104	if (err) {
1105		free_cpumask_var(mask);
1106		return err;
1107	}
1108
1109	new_dev_maps = kzalloc(max_t(unsigned int,
1110	    XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
1111	if (!new_dev_maps) {
1112		free_cpumask_var(mask);
1113		return -ENOMEM;
1114	}
1115
1116	mutex_lock(&xps_map_mutex);
1117
1118	dev_maps = xmap_dereference(dev->xps_maps);
1119
1120	for_each_possible_cpu(cpu) {
1121		map = dev_maps ?
1122			xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1123		new_map = map;
1124		if (map) {
1125			for (pos = 0; pos < map->len; pos++)
1126				if (map->queues[pos] == index)
1127					break;
1128			map_len = map->len;
1129			alloc_len = map->alloc_len;
1130		} else
1131			pos = map_len = alloc_len = 0;
1132
1133		need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu);
1134#ifdef CONFIG_NUMA
1135		if (need_set) {
1136			if (numa_node_id == -2)
1137				numa_node_id = cpu_to_node(cpu);
1138			else if (numa_node_id != cpu_to_node(cpu))
1139				numa_node_id = -1;
1140		}
1141#endif
1142		if (need_set && pos >= map_len) {
1143			/* Need to add queue to this CPU's map */
1144			if (map_len >= alloc_len) {
1145				alloc_len = alloc_len ?
1146				    2 * alloc_len : XPS_MIN_MAP_ALLOC;
1147				new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
1148						       GFP_KERNEL,
1149						       cpu_to_node(cpu));
1150				if (!new_map)
1151					goto error;
1152				new_map->alloc_len = alloc_len;
1153				for (i = 0; i < map_len; i++)
1154					new_map->queues[i] = map->queues[i];
1155				new_map->len = map_len;
1156			}
1157			new_map->queues[new_map->len++] = index;
1158		} else if (!need_set && pos < map_len) {
1159			/* Need to remove queue from this CPU's map */
1160			if (map_len > 1)
1161				new_map->queues[pos] =
1162				    new_map->queues[--new_map->len];
1163			else
1164				new_map = NULL;
1165		}
1166		RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
1167	}
1168
1169	/* Cleanup old maps */
1170	for_each_possible_cpu(cpu) {
1171		map = dev_maps ?
1172			xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1173		if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
1174			kfree_rcu(map, rcu);
1175		if (new_dev_maps->cpu_map[cpu])
1176			nonempty = 1;
1177	}
1178
1179	if (nonempty) {
1180		rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1181	} else {
1182		kfree(new_dev_maps);
1183		RCU_INIT_POINTER(dev->xps_maps, NULL);
1184	}
1185
1186	if (dev_maps)
1187		kfree_rcu(dev_maps, rcu);
1188
1189	netdev_queue_numa_node_write(queue, (numa_node_id >= 0) ? numa_node_id :
1190					    NUMA_NO_NODE);
1191
1192	mutex_unlock(&xps_map_mutex);
1193
1194	free_cpumask_var(mask);
1195	return len;
1196
1197error:
1198	mutex_unlock(&xps_map_mutex);
1199
1200	if (new_dev_maps)
1201		for_each_possible_cpu(i)
1202			kfree(rcu_dereference_protected(
1203				new_dev_maps->cpu_map[i],
1204				1));
1205	kfree(new_dev_maps);
1206	free_cpumask_var(mask);
1207	return -ENOMEM;
1208}
1209
1210static struct netdev_queue_attribute xps_cpus_attribute =
1211    __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1212#endif /* CONFIG_XPS */
1213
1214static struct attribute *netdev_queue_default_attrs[] = {
1215	&queue_trans_timeout.attr,
1216#ifdef CONFIG_XPS
1217	&xps_cpus_attribute.attr,
1218#endif
1219	NULL
1220};
1221
1222static void netdev_queue_release(struct kobject *kobj)
1223{
1224	struct netdev_queue *queue = to_netdev_queue(kobj);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1225
1226#ifdef CONFIG_XPS
1227	xps_queue_release(queue);
1228#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1229
1230	memset(kobj, 0, sizeof(*kobj));
1231	dev_put(queue->dev);
1232}
1233
1234static struct kobj_type netdev_queue_ktype = {
1235	.sysfs_ops = &netdev_queue_sysfs_ops,
1236	.release = netdev_queue_release,
1237	.default_attrs = netdev_queue_default_attrs,
1238};
1239
1240static int netdev_queue_add_kobject(struct net_device *net, int index)
1241{
1242	struct netdev_queue *queue = net->_tx + index;
1243	struct kobject *kobj = &queue->kobj;
1244	int error = 0;
1245
1246	kobj->kset = net->queues_kset;
1247	error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1248	    "tx-%u", index);
1249	if (error)
1250		goto exit;
1251
1252#ifdef CONFIG_BQL
1253	error = sysfs_create_group(kobj, &dql_group);
1254	if (error)
1255		goto exit;
1256#endif
1257
1258	kobject_uevent(kobj, KOBJ_ADD);
1259	dev_hold(queue->dev);
1260
1261	return 0;
1262exit:
1263	kobject_put(kobj);
1264	return error;
1265}
1266#endif /* CONFIG_SYSFS */
1267
1268int
1269netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1270{
1271#ifdef CONFIG_SYSFS
1272	int i;
1273	int error = 0;
1274
1275	for (i = old_num; i < new_num; i++) {
1276		error = netdev_queue_add_kobject(net, i);
1277		if (error) {
1278			new_num = old_num;
1279			break;
1280		}
1281	}
1282
1283	while (--i >= new_num) {
1284		struct netdev_queue *queue = net->_tx + i;
1285
1286#ifdef CONFIG_BQL
1287		sysfs_remove_group(&queue->kobj, &dql_group);
1288#endif
1289		kobject_put(&queue->kobj);
1290	}
1291
1292	return error;
1293#else
1294	return 0;
1295#endif /* CONFIG_SYSFS */
1296}
1297
1298static int register_queue_kobjects(struct net_device *net)
1299{
1300	int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1301
1302#ifdef CONFIG_SYSFS
1303	net->queues_kset = kset_create_and_add("queues",
1304	    NULL, &net->dev.kobj);
1305	if (!net->queues_kset)
1306		return -ENOMEM;
1307#endif
1308
1309#ifdef CONFIG_RPS
1310	real_rx = net->real_num_rx_queues;
1311#endif
1312	real_tx = net->real_num_tx_queues;
1313
1314	error = net_rx_queue_update_kobjects(net, 0, real_rx);
1315	if (error)
1316		goto error;
1317	rxq = real_rx;
1318
1319	error = netdev_queue_update_kobjects(net, 0, real_tx);
1320	if (error)
1321		goto error;
1322	txq = real_tx;
1323
1324	return 0;
1325
1326error:
1327	netdev_queue_update_kobjects(net, txq, 0);
1328	net_rx_queue_update_kobjects(net, rxq, 0);
1329	return error;
1330}
1331
1332static void remove_queue_kobjects(struct net_device *net)
1333{
1334	int real_rx = 0, real_tx = 0;
1335
1336#ifdef CONFIG_RPS
1337	real_rx = net->real_num_rx_queues;
1338#endif
1339	real_tx = net->real_num_tx_queues;
1340
1341	net_rx_queue_update_kobjects(net, real_rx, 0);
1342	netdev_queue_update_kobjects(net, real_tx, 0);
1343#ifdef CONFIG_SYSFS
1344	kset_unregister(net->queues_kset);
1345#endif
1346}
1347
1348static void *net_grab_current_ns(void)
1349{
1350	struct net *ns = current->nsproxy->net_ns;
1351#ifdef CONFIG_NET_NS
1352	if (ns)
1353		atomic_inc(&ns->passive);
1354#endif
1355	return ns;
1356}
1357
1358static const void *net_initial_ns(void)
1359{
1360	return &init_net;
1361}
1362
1363static const void *net_netlink_ns(struct sock *sk)
1364{
1365	return sock_net(sk);
1366}
1367
1368struct kobj_ns_type_operations net_ns_type_operations = {
1369	.type = KOBJ_NS_TYPE_NET,
1370	.grab_current_ns = net_grab_current_ns,
1371	.netlink_ns = net_netlink_ns,
1372	.initial_ns = net_initial_ns,
1373	.drop_ns = net_drop_ns,
1374};
1375EXPORT_SYMBOL_GPL(net_ns_type_operations);
1376
1377#ifdef CONFIG_HOTPLUG
1378static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1379{
1380	struct net_device *dev = to_net_dev(d);
1381	int retval;
1382
1383	/* pass interface to uevent. */
1384	retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1385	if (retval)
1386		goto exit;
1387
1388	/* pass ifindex to uevent.
1389	 * ifindex is useful as it won't change (interface name may change)
1390	 * and is what RtNetlink uses natively. */
1391	retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1392
1393exit:
1394	return retval;
1395}
1396#endif
1397
1398/*
1399 *	netdev_release -- destroy and free a dead device.
1400 *	Called when last reference to device kobject is gone.
1401 */
1402static void netdev_release(struct device *d)
1403{
1404	struct net_device *dev = to_net_dev(d);
1405
1406	BUG_ON(dev->reg_state != NETREG_RELEASED);
1407
1408	kfree(dev->ifalias);
1409	kfree((char *)dev - dev->padded);
1410}
1411
1412static const void *net_namespace(struct device *d)
1413{
1414	struct net_device *dev;
1415	dev = container_of(d, struct net_device, dev);
1416	return dev_net(dev);
1417}
1418
1419static struct class net_class = {
1420	.name = "net",
1421	.dev_release = netdev_release,
1422#ifdef CONFIG_SYSFS
1423	.dev_attrs = net_class_attributes,
1424#endif /* CONFIG_SYSFS */
1425#ifdef CONFIG_HOTPLUG
1426	.dev_uevent = netdev_uevent,
1427#endif
1428	.ns_type = &net_ns_type_operations,
1429	.namespace = net_namespace,
1430};
1431
1432/* Delete sysfs entries but hold kobject reference until after all
1433 * netdev references are gone.
1434 */
1435void netdev_unregister_kobject(struct net_device * net)
1436{
1437	struct device *dev = &(net->dev);
1438
1439	kobject_get(&dev->kobj);
1440
1441	remove_queue_kobjects(net);
1442
1443	device_del(dev);
1444}
1445
1446/* Create sysfs entries for network device. */
1447int netdev_register_kobject(struct net_device *net)
1448{
1449	struct device *dev = &(net->dev);
1450	const struct attribute_group **groups = net->sysfs_groups;
1451	int error = 0;
1452
1453	device_initialize(dev);
1454	dev->class = &net_class;
1455	dev->platform_data = net;
1456	dev->groups = groups;
1457
1458	dev_set_name(dev, "%s", net->name);
1459
1460#ifdef CONFIG_SYSFS
1461	/* Allow for a device specific group */
1462	if (*groups)
1463		groups++;
1464
1465	*groups++ = &netstat_group;
1466#ifdef CONFIG_WIRELESS_EXT_SYSFS
1467	if (net->ieee80211_ptr)
1468		*groups++ = &wireless_group;
1469#ifdef CONFIG_WIRELESS_EXT
1470	else if (net->wireless_handlers)
1471		*groups++ = &wireless_group;
1472#endif
1473#endif
1474#endif /* CONFIG_SYSFS */
1475
1476	error = device_add(dev);
1477	if (error)
1478		return error;
1479
1480	error = register_queue_kobjects(net);
1481	if (error) {
1482		device_del(dev);
1483		return error;
1484	}
1485
1486	return error;
1487}
1488
1489int netdev_class_create_file(struct class_attribute *class_attr)
1490{
1491	return class_create_file(&net_class, class_attr);
1492}
1493EXPORT_SYMBOL(netdev_class_create_file);
1494
1495void netdev_class_remove_file(struct class_attribute *class_attr)
1496{
1497	class_remove_file(&net_class, class_attr);
1498}
1499EXPORT_SYMBOL(netdev_class_remove_file);
1500
1501int netdev_kobject_init(void)
1502{
1503	kobj_ns_type_register(&net_ns_type_operations);
1504	return class_register(&net_class);
1505}