1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * net-sysfs.c - network device class and attributes
   4 *
   5 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
   6 */
   7
   8#include <linux/capability.h>
   9#include <linux/kernel.h>
  10#include <linux/netdevice.h>
  11#include <linux/if_arp.h>
  12#include <linux/slab.h>
  13#include <linux/sched/signal.h>
  14#include <linux/sched/isolation.h>
  15#include <linux/nsproxy.h>
  16#include <net/sock.h>
  17#include <net/net_namespace.h>
  18#include <linux/rtnetlink.h>
  19#include <linux/vmalloc.h>
  20#include <linux/export.h>
  21#include <linux/jiffies.h>
  22#include <linux/pm_runtime.h>
  23#include <linux/of.h>
  24#include <linux/of_net.h>
  25#include <linux/cpu.h>
 
  26
  27#include "dev.h"
  28#include "net-sysfs.h"
  29
  30#ifdef CONFIG_SYSFS
  31static const char fmt_hex[] = "%#x\n";
  32static const char fmt_dec[] = "%d\n";
  33static const char fmt_ulong[] = "%lu\n";
  34static const char fmt_u64[] = "%llu\n";
  35
  36/* Caller holds RTNL or dev_base_lock */
  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 *ndev = to_net_dev(dev);
  48	ssize_t ret = -EINVAL;
  49
  50	read_lock(&dev_base_lock);
  51	if (dev_isalive(ndev))
  52		ret = (*format)(ndev, 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 *dev, char *buf)	\
  61{									\
  62	return sysfs_emit(buf, format_string, dev->field);		\
  63}									\
  64static ssize_t field##_show(struct device *dev,				\
  65			    struct device_attribute *attr, char *buf)	\
  66{									\
  67	return netdev_show(dev, attr, buf, format_##field);		\
  68}									\
  69
  70#define NETDEVICE_SHOW_RO(field, format_string)				\
  71NETDEVICE_SHOW(field, format_string);					\
  72static DEVICE_ATTR_RO(field)
  73
  74#define NETDEVICE_SHOW_RW(field, format_string)				\
  75NETDEVICE_SHOW(field, format_string);					\
  76static DEVICE_ATTR_RW(field)
  77
  78/* use same locking and permission rules as SIF* ioctl's */
  79static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
  80			    const char *buf, size_t len,
  81			    int (*set)(struct net_device *, unsigned long))
  82{
  83	struct net_device *netdev = to_net_dev(dev);
  84	struct net *net = dev_net(netdev);
  85	unsigned long new;
  86	int ret;
  87
  88	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
  89		return -EPERM;
  90
  91	ret = kstrtoul(buf, 0, &new);
  92	if (ret)
  93		goto err;
  94
  95	if (!rtnl_trylock())
  96		return restart_syscall();
  97
  98	if (dev_isalive(netdev)) {
  99		ret = (*set)(netdev, new);
 100		if (ret == 0)
 101			ret = len;
 102	}
 103	rtnl_unlock();
 104 err:
 105	return ret;
 106}
 107
 108NETDEVICE_SHOW_RO(dev_id, fmt_hex);
 109NETDEVICE_SHOW_RO(dev_port, fmt_dec);
 110NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
 111NETDEVICE_SHOW_RO(addr_len, fmt_dec);
 112NETDEVICE_SHOW_RO(ifindex, fmt_dec);
 113NETDEVICE_SHOW_RO(type, fmt_dec);
 114NETDEVICE_SHOW_RO(link_mode, fmt_dec);
 115
 116static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
 117			   char *buf)
 118{
 119	struct net_device *ndev = to_net_dev(dev);
 120
 121	return sysfs_emit(buf, fmt_dec, dev_get_iflink(ndev));
 122}
 123static DEVICE_ATTR_RO(iflink);
 124
 125static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
 126{
 127	return sysfs_emit(buf, fmt_dec, dev->name_assign_type);
 128}
 129
 130static ssize_t name_assign_type_show(struct device *dev,
 131				     struct device_attribute *attr,
 132				     char *buf)
 133{
 134	struct net_device *ndev = to_net_dev(dev);
 135	ssize_t ret = -EINVAL;
 136
 137	if (ndev->name_assign_type != NET_NAME_UNKNOWN)
 138		ret = netdev_show(dev, attr, buf, format_name_assign_type);
 139
 140	return ret;
 141}
 142static DEVICE_ATTR_RO(name_assign_type);
 143
 144/* use same locking rules as GIFHWADDR ioctl's */
 145static ssize_t address_show(struct device *dev, struct device_attribute *attr,
 146			    char *buf)
 147{
 148	struct net_device *ndev = to_net_dev(dev);
 149	ssize_t ret = -EINVAL;
 150
 151	read_lock(&dev_base_lock);
 152	if (dev_isalive(ndev))
 153		ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
 154	read_unlock(&dev_base_lock);
 155	return ret;
 156}
 157static DEVICE_ATTR_RO(address);
 158
 159static ssize_t broadcast_show(struct device *dev,
 160			      struct device_attribute *attr, char *buf)
 161{
 162	struct net_device *ndev = to_net_dev(dev);
 163
 164	if (dev_isalive(ndev))
 165		return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
 166	return -EINVAL;
 167}
 168static DEVICE_ATTR_RO(broadcast);
 169
 170static int change_carrier(struct net_device *dev, unsigned long new_carrier)
 171{
 172	if (!netif_running(dev))
 173		return -EINVAL;
 174	return dev_change_carrier(dev, (bool)new_carrier);
 175}
 176
 177static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
 178			     const char *buf, size_t len)
 179{
 180	struct net_device *netdev = to_net_dev(dev);
 181
 182	/* The check is also done in change_carrier; this helps returning early
 183	 * without hitting the trylock/restart in netdev_store.
 184	 */
 185	if (!netdev->netdev_ops->ndo_change_carrier)
 186		return -EOPNOTSUPP;
 187
 188	return netdev_store(dev, attr, buf, len, change_carrier);
 189}
 190
 191static ssize_t carrier_show(struct device *dev,
 192			    struct device_attribute *attr, char *buf)
 193{
 194	struct net_device *netdev = to_net_dev(dev);
 
 195
 196	if (netif_running(netdev))
 197		return sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev));
 198
 199	return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 200}
 201static DEVICE_ATTR_RW(carrier);
 202
 203static ssize_t speed_show(struct device *dev,
 204			  struct device_attribute *attr, char *buf)
 205{
 206	struct net_device *netdev = to_net_dev(dev);
 207	int ret = -EINVAL;
 208
 209	/* The check is also done in __ethtool_get_link_ksettings; this helps
 210	 * returning early without hitting the trylock/restart below.
 211	 */
 212	if (!netdev->ethtool_ops->get_link_ksettings)
 213		return ret;
 214
 215	if (!rtnl_trylock())
 216		return restart_syscall();
 217
 218	if (netif_running(netdev) && netif_device_present(netdev)) {
 219		struct ethtool_link_ksettings cmd;
 220
 221		if (!__ethtool_get_link_ksettings(netdev, &cmd))
 222			ret = sysfs_emit(buf, fmt_dec, cmd.base.speed);
 223	}
 224	rtnl_unlock();
 225	return ret;
 226}
 227static DEVICE_ATTR_RO(speed);
 228
 229static ssize_t duplex_show(struct device *dev,
 230			   struct device_attribute *attr, char *buf)
 231{
 232	struct net_device *netdev = to_net_dev(dev);
 233	int ret = -EINVAL;
 234
 235	/* The check is also done in __ethtool_get_link_ksettings; this helps
 236	 * returning early without hitting the trylock/restart below.
 237	 */
 238	if (!netdev->ethtool_ops->get_link_ksettings)
 239		return ret;
 240
 241	if (!rtnl_trylock())
 242		return restart_syscall();
 243
 244	if (netif_running(netdev)) {
 245		struct ethtool_link_ksettings cmd;
 246
 247		if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
 248			const char *duplex;
 249
 250			switch (cmd.base.duplex) {
 251			case DUPLEX_HALF:
 252				duplex = "half";
 253				break;
 254			case DUPLEX_FULL:
 255				duplex = "full";
 256				break;
 257			default:
 258				duplex = "unknown";
 259				break;
 260			}
 261			ret = sysfs_emit(buf, "%s\n", duplex);
 262		}
 263	}
 264	rtnl_unlock();
 265	return ret;
 266}
 267static DEVICE_ATTR_RO(duplex);
 268
 269static ssize_t testing_show(struct device *dev,
 270			    struct device_attribute *attr, char *buf)
 271{
 272	struct net_device *netdev = to_net_dev(dev);
 273
 274	if (netif_running(netdev))
 275		return sysfs_emit(buf, fmt_dec, !!netif_testing(netdev));
 276
 277	return -EINVAL;
 278}
 279static DEVICE_ATTR_RO(testing);
 280
 281static ssize_t dormant_show(struct device *dev,
 282			    struct device_attribute *attr, char *buf)
 283{
 284	struct net_device *netdev = to_net_dev(dev);
 285
 286	if (netif_running(netdev))
 287		return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev));
 288
 289	return -EINVAL;
 290}
 291static DEVICE_ATTR_RO(dormant);
 292
 293static const char *const operstates[] = {
 294	"unknown",
 295	"notpresent", /* currently unused */
 296	"down",
 297	"lowerlayerdown",
 298	"testing",
 299	"dormant",
 300	"up"
 301};
 302
 303static ssize_t operstate_show(struct device *dev,
 304			      struct device_attribute *attr, char *buf)
 305{
 306	const struct net_device *netdev = to_net_dev(dev);
 307	unsigned char operstate;
 308
 309	read_lock(&dev_base_lock);
 310	operstate = netdev->operstate;
 311	if (!netif_running(netdev))
 312		operstate = IF_OPER_DOWN;
 313	read_unlock(&dev_base_lock);
 314
 315	if (operstate >= ARRAY_SIZE(operstates))
 316		return -EINVAL; /* should not happen */
 317
 318	return sysfs_emit(buf, "%s\n", operstates[operstate]);
 319}
 320static DEVICE_ATTR_RO(operstate);
 321
 322static ssize_t carrier_changes_show(struct device *dev,
 323				    struct device_attribute *attr,
 324				    char *buf)
 325{
 326	struct net_device *netdev = to_net_dev(dev);
 327
 328	return sysfs_emit(buf, fmt_dec,
 329			  atomic_read(&netdev->carrier_up_count) +
 330			  atomic_read(&netdev->carrier_down_count));
 331}
 332static DEVICE_ATTR_RO(carrier_changes);
 333
 334static ssize_t carrier_up_count_show(struct device *dev,
 335				     struct device_attribute *attr,
 336				     char *buf)
 337{
 338	struct net_device *netdev = to_net_dev(dev);
 339
 340	return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
 341}
 342static DEVICE_ATTR_RO(carrier_up_count);
 343
 344static ssize_t carrier_down_count_show(struct device *dev,
 345				       struct device_attribute *attr,
 346				       char *buf)
 347{
 348	struct net_device *netdev = to_net_dev(dev);
 349
 350	return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
 351}
 352static DEVICE_ATTR_RO(carrier_down_count);
 353
 354/* read-write attributes */
 355
 356static int change_mtu(struct net_device *dev, unsigned long new_mtu)
 357{
 358	return dev_set_mtu(dev, (int)new_mtu);
 359}
 360
 361static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
 362			 const char *buf, size_t len)
 363{
 364	return netdev_store(dev, attr, buf, len, change_mtu);
 365}
 366NETDEVICE_SHOW_RW(mtu, fmt_dec);
 367
 368static int change_flags(struct net_device *dev, unsigned long new_flags)
 369{
 370	return dev_change_flags(dev, (unsigned int)new_flags, NULL);
 371}
 372
 373static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
 374			   const char *buf, size_t len)
 375{
 376	return netdev_store(dev, attr, buf, len, change_flags);
 377}
 378NETDEVICE_SHOW_RW(flags, fmt_hex);
 379
 380static ssize_t tx_queue_len_store(struct device *dev,
 381				  struct device_attribute *attr,
 382				  const char *buf, size_t len)
 383{
 384	if (!capable(CAP_NET_ADMIN))
 385		return -EPERM;
 386
 387	return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
 388}
 389NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
 390
 391static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
 392{
 393	WRITE_ONCE(dev->gro_flush_timeout, val);
 394	return 0;
 395}
 396
 397static ssize_t gro_flush_timeout_store(struct device *dev,
 398				       struct device_attribute *attr,
 399				       const char *buf, size_t len)
 400{
 401	if (!capable(CAP_NET_ADMIN))
 402		return -EPERM;
 403
 404	return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
 405}
 406NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
 407
 408static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
 409{
 410	WRITE_ONCE(dev->napi_defer_hard_irqs, val);
 411	return 0;
 412}
 413
 414static ssize_t napi_defer_hard_irqs_store(struct device *dev,
 415					  struct device_attribute *attr,
 416					  const char *buf, size_t len)
 417{
 418	if (!capable(CAP_NET_ADMIN))
 419		return -EPERM;
 420
 421	return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
 422}
 423NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
 424
 425static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
 426			     const char *buf, size_t len)
 427{
 428	struct net_device *netdev = to_net_dev(dev);
 429	struct net *net = dev_net(netdev);
 430	size_t count = len;
 431	ssize_t ret = 0;
 432
 433	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 434		return -EPERM;
 435
 436	/* ignore trailing newline */
 437	if (len >  0 && buf[len - 1] == '\n')
 438		--count;
 439
 440	if (!rtnl_trylock())
 441		return restart_syscall();
 442
 443	if (dev_isalive(netdev)) {
 444		ret = dev_set_alias(netdev, buf, count);
 445		if (ret < 0)
 446			goto err;
 447		ret = len;
 448		netdev_state_change(netdev);
 449	}
 450err:
 451	rtnl_unlock();
 452
 453	return ret;
 454}
 455
 456static ssize_t ifalias_show(struct device *dev,
 457			    struct device_attribute *attr, char *buf)
 458{
 459	const struct net_device *netdev = to_net_dev(dev);
 460	char tmp[IFALIASZ];
 461	ssize_t ret = 0;
 462
 463	ret = dev_get_alias(netdev, tmp, sizeof(tmp));
 464	if (ret > 0)
 465		ret = sysfs_emit(buf, "%s\n", tmp);
 466	return ret;
 467}
 468static DEVICE_ATTR_RW(ifalias);
 469
 470static int change_group(struct net_device *dev, unsigned long new_group)
 471{
 472	dev_set_group(dev, (int)new_group);
 473	return 0;
 474}
 475
 476static ssize_t group_store(struct device *dev, struct device_attribute *attr,
 477			   const char *buf, size_t len)
 478{
 479	return netdev_store(dev, attr, buf, len, change_group);
 480}
 481NETDEVICE_SHOW(group, fmt_dec);
 482static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
 483
 484static int change_proto_down(struct net_device *dev, unsigned long proto_down)
 485{
 486	return dev_change_proto_down(dev, (bool)proto_down);
 487}
 488
 489static ssize_t proto_down_store(struct device *dev,
 490				struct device_attribute *attr,
 491				const char *buf, size_t len)
 492{
 493	return netdev_store(dev, attr, buf, len, change_proto_down);
 494}
 495NETDEVICE_SHOW_RW(proto_down, fmt_dec);
 496
 497static ssize_t phys_port_id_show(struct device *dev,
 498				 struct device_attribute *attr, char *buf)
 499{
 500	struct net_device *netdev = to_net_dev(dev);
 501	ssize_t ret = -EINVAL;
 502
 503	/* The check is also done in dev_get_phys_port_id; this helps returning
 504	 * early without hitting the trylock/restart below.
 505	 */
 506	if (!netdev->netdev_ops->ndo_get_phys_port_id)
 507		return -EOPNOTSUPP;
 508
 509	if (!rtnl_trylock())
 510		return restart_syscall();
 511
 512	if (dev_isalive(netdev)) {
 513		struct netdev_phys_item_id ppid;
 514
 515		ret = dev_get_phys_port_id(netdev, &ppid);
 516		if (!ret)
 517			ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
 518	}
 519	rtnl_unlock();
 520
 521	return ret;
 522}
 523static DEVICE_ATTR_RO(phys_port_id);
 524
 525static ssize_t phys_port_name_show(struct device *dev,
 526				   struct device_attribute *attr, char *buf)
 527{
 528	struct net_device *netdev = to_net_dev(dev);
 529	ssize_t ret = -EINVAL;
 530
 531	/* The checks are also done in dev_get_phys_port_name; this helps
 532	 * returning early without hitting the trylock/restart below.
 533	 */
 534	if (!netdev->netdev_ops->ndo_get_phys_port_name &&
 535	    !netdev->devlink_port)
 536		return -EOPNOTSUPP;
 537
 538	if (!rtnl_trylock())
 539		return restart_syscall();
 540
 541	if (dev_isalive(netdev)) {
 542		char name[IFNAMSIZ];
 543
 544		ret = dev_get_phys_port_name(netdev, name, sizeof(name));
 545		if (!ret)
 546			ret = sysfs_emit(buf, "%s\n", name);
 547	}
 548	rtnl_unlock();
 549
 550	return ret;
 551}
 552static DEVICE_ATTR_RO(phys_port_name);
 553
 554static ssize_t phys_switch_id_show(struct device *dev,
 555				   struct device_attribute *attr, char *buf)
 556{
 557	struct net_device *netdev = to_net_dev(dev);
 558	ssize_t ret = -EINVAL;
 559
 560	/* The checks are also done in dev_get_phys_port_name; this helps
 561	 * returning early without hitting the trylock/restart below. This works
 562	 * because recurse is false when calling dev_get_port_parent_id.
 563	 */
 564	if (!netdev->netdev_ops->ndo_get_port_parent_id &&
 565	    !netdev->devlink_port)
 566		return -EOPNOTSUPP;
 567
 568	if (!rtnl_trylock())
 569		return restart_syscall();
 570
 571	if (dev_isalive(netdev)) {
 572		struct netdev_phys_item_id ppid = { };
 573
 574		ret = dev_get_port_parent_id(netdev, &ppid, false);
 575		if (!ret)
 576			ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
 577	}
 578	rtnl_unlock();
 579
 580	return ret;
 581}
 582static DEVICE_ATTR_RO(phys_switch_id);
 583
 584static ssize_t threaded_show(struct device *dev,
 585			     struct device_attribute *attr, char *buf)
 586{
 587	struct net_device *netdev = to_net_dev(dev);
 588	ssize_t ret = -EINVAL;
 589
 590	if (!rtnl_trylock())
 591		return restart_syscall();
 592
 593	if (dev_isalive(netdev))
 594		ret = sysfs_emit(buf, fmt_dec, netdev->threaded);
 595
 596	rtnl_unlock();
 597	return ret;
 598}
 599
 600static int modify_napi_threaded(struct net_device *dev, unsigned long val)
 601{
 602	int ret;
 603
 604	if (list_empty(&dev->napi_list))
 605		return -EOPNOTSUPP;
 606
 607	if (val != 0 && val != 1)
 608		return -EOPNOTSUPP;
 609
 610	ret = dev_set_threaded(dev, val);
 611
 612	return ret;
 613}
 614
 615static ssize_t threaded_store(struct device *dev,
 616			      struct device_attribute *attr,
 617			      const char *buf, size_t len)
 618{
 619	return netdev_store(dev, attr, buf, len, modify_napi_threaded);
 620}
 621static DEVICE_ATTR_RW(threaded);
 622
 623static struct attribute *net_class_attrs[] __ro_after_init = {
 624	&dev_attr_netdev_group.attr,
 625	&dev_attr_type.attr,
 626	&dev_attr_dev_id.attr,
 627	&dev_attr_dev_port.attr,
 628	&dev_attr_iflink.attr,
 629	&dev_attr_ifindex.attr,
 630	&dev_attr_name_assign_type.attr,
 631	&dev_attr_addr_assign_type.attr,
 632	&dev_attr_addr_len.attr,
 633	&dev_attr_link_mode.attr,
 634	&dev_attr_address.attr,
 635	&dev_attr_broadcast.attr,
 636	&dev_attr_speed.attr,
 637	&dev_attr_duplex.attr,
 638	&dev_attr_dormant.attr,
 639	&dev_attr_testing.attr,
 640	&dev_attr_operstate.attr,
 641	&dev_attr_carrier_changes.attr,
 642	&dev_attr_ifalias.attr,
 643	&dev_attr_carrier.attr,
 644	&dev_attr_mtu.attr,
 645	&dev_attr_flags.attr,
 646	&dev_attr_tx_queue_len.attr,
 647	&dev_attr_gro_flush_timeout.attr,
 648	&dev_attr_napi_defer_hard_irqs.attr,
 649	&dev_attr_phys_port_id.attr,
 650	&dev_attr_phys_port_name.attr,
 651	&dev_attr_phys_switch_id.attr,
 652	&dev_attr_proto_down.attr,
 653	&dev_attr_carrier_up_count.attr,
 654	&dev_attr_carrier_down_count.attr,
 655	&dev_attr_threaded.attr,
 656	NULL,
 657};
 658ATTRIBUTE_GROUPS(net_class);
 659
 660/* Show a given an attribute in the statistics group */
 661static ssize_t netstat_show(const struct device *d,
 662			    struct device_attribute *attr, char *buf,
 663			    unsigned long offset)
 664{
 665	struct net_device *dev = to_net_dev(d);
 666	ssize_t ret = -EINVAL;
 667
 668	WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
 669		offset % sizeof(u64) != 0);
 670
 671	read_lock(&dev_base_lock);
 672	if (dev_isalive(dev)) {
 673		struct rtnl_link_stats64 temp;
 674		const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
 675
 676		ret = sysfs_emit(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
 677	}
 678	read_unlock(&dev_base_lock);
 679	return ret;
 680}
 681
 682/* generate a read-only statistics attribute */
 683#define NETSTAT_ENTRY(name)						\
 684static ssize_t name##_show(struct device *d,				\
 685			   struct device_attribute *attr, char *buf)	\
 686{									\
 687	return netstat_show(d, attr, buf,				\
 688			    offsetof(struct rtnl_link_stats64, name));	\
 689}									\
 690static DEVICE_ATTR_RO(name)
 691
 692NETSTAT_ENTRY(rx_packets);
 693NETSTAT_ENTRY(tx_packets);
 694NETSTAT_ENTRY(rx_bytes);
 695NETSTAT_ENTRY(tx_bytes);
 696NETSTAT_ENTRY(rx_errors);
 697NETSTAT_ENTRY(tx_errors);
 698NETSTAT_ENTRY(rx_dropped);
 699NETSTAT_ENTRY(tx_dropped);
 700NETSTAT_ENTRY(multicast);
 701NETSTAT_ENTRY(collisions);
 702NETSTAT_ENTRY(rx_length_errors);
 703NETSTAT_ENTRY(rx_over_errors);
 704NETSTAT_ENTRY(rx_crc_errors);
 705NETSTAT_ENTRY(rx_frame_errors);
 706NETSTAT_ENTRY(rx_fifo_errors);
 707NETSTAT_ENTRY(rx_missed_errors);
 708NETSTAT_ENTRY(tx_aborted_errors);
 709NETSTAT_ENTRY(tx_carrier_errors);
 710NETSTAT_ENTRY(tx_fifo_errors);
 711NETSTAT_ENTRY(tx_heartbeat_errors);
 712NETSTAT_ENTRY(tx_window_errors);
 713NETSTAT_ENTRY(rx_compressed);
 714NETSTAT_ENTRY(tx_compressed);
 715NETSTAT_ENTRY(rx_nohandler);
 716
 717static struct attribute *netstat_attrs[] __ro_after_init = {
 718	&dev_attr_rx_packets.attr,
 719	&dev_attr_tx_packets.attr,
 720	&dev_attr_rx_bytes.attr,
 721	&dev_attr_tx_bytes.attr,
 722	&dev_attr_rx_errors.attr,
 723	&dev_attr_tx_errors.attr,
 724	&dev_attr_rx_dropped.attr,
 725	&dev_attr_tx_dropped.attr,
 726	&dev_attr_multicast.attr,
 727	&dev_attr_collisions.attr,
 728	&dev_attr_rx_length_errors.attr,
 729	&dev_attr_rx_over_errors.attr,
 730	&dev_attr_rx_crc_errors.attr,
 731	&dev_attr_rx_frame_errors.attr,
 732	&dev_attr_rx_fifo_errors.attr,
 733	&dev_attr_rx_missed_errors.attr,
 734	&dev_attr_tx_aborted_errors.attr,
 735	&dev_attr_tx_carrier_errors.attr,
 736	&dev_attr_tx_fifo_errors.attr,
 737	&dev_attr_tx_heartbeat_errors.attr,
 738	&dev_attr_tx_window_errors.attr,
 739	&dev_attr_rx_compressed.attr,
 740	&dev_attr_tx_compressed.attr,
 741	&dev_attr_rx_nohandler.attr,
 742	NULL
 743};
 744
 745static const struct attribute_group netstat_group = {
 746	.name  = "statistics",
 747	.attrs  = netstat_attrs,
 748};
 749
 750static struct attribute *wireless_attrs[] = {
 751	NULL
 752};
 753
 754static const struct attribute_group wireless_group = {
 755	.name = "wireless",
 756	.attrs = wireless_attrs,
 757};
 758
 759static bool wireless_group_needed(struct net_device *ndev)
 760{
 761#if IS_ENABLED(CONFIG_CFG80211)
 762	if (ndev->ieee80211_ptr)
 763		return true;
 764#endif
 765#if IS_ENABLED(CONFIG_WIRELESS_EXT)
 766	if (ndev->wireless_handlers)
 767		return true;
 768#endif
 769	return false;
 770}
 771
 772#else /* CONFIG_SYSFS */
 773#define net_class_groups	NULL
 774#endif /* CONFIG_SYSFS */
 775
 776#ifdef CONFIG_SYSFS
 777#define to_rx_queue_attr(_attr) \
 778	container_of(_attr, struct rx_queue_attribute, attr)
 779
 780#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
 781
 782static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
 783				  char *buf)
 784{
 785	const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 786	struct netdev_rx_queue *queue = to_rx_queue(kobj);
 787
 788	if (!attribute->show)
 789		return -EIO;
 790
 791	return attribute->show(queue, buf);
 792}
 793
 794static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
 795				   const char *buf, size_t count)
 796{
 797	const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
 798	struct netdev_rx_queue *queue = to_rx_queue(kobj);
 799
 800	if (!attribute->store)
 801		return -EIO;
 802
 803	return attribute->store(queue, buf, count);
 804}
 805
 806static const struct sysfs_ops rx_queue_sysfs_ops = {
 807	.show = rx_queue_attr_show,
 808	.store = rx_queue_attr_store,
 809};
 810
 811#ifdef CONFIG_RPS
 812static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
 813{
 814	struct rps_map *map;
 815	cpumask_var_t mask;
 816	int i, len;
 817
 818	if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
 819		return -ENOMEM;
 820
 821	rcu_read_lock();
 822	map = rcu_dereference(queue->rps_map);
 823	if (map)
 824		for (i = 0; i < map->len; i++)
 825			cpumask_set_cpu(map->cpus[i], mask);
 826
 827	len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask));
 828	rcu_read_unlock();
 829	free_cpumask_var(mask);
 830
 831	return len < PAGE_SIZE ? len : -EINVAL;
 832}
 833
 834static ssize_t store_rps_map(struct netdev_rx_queue *queue,
 835			     const char *buf, size_t len)
 836{
 837	struct rps_map *old_map, *map;
 838	cpumask_var_t mask;
 839	int err, cpu, i;
 840	static DEFINE_MUTEX(rps_map_mutex);
 841
 842	if (!capable(CAP_NET_ADMIN))
 843		return -EPERM;
 844
 845	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
 846		return -ENOMEM;
 847
 848	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
 849	if (err) {
 850		free_cpumask_var(mask);
 851		return err;
 852	}
 853
 854	if (!cpumask_empty(mask)) {
 855		cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
 856		cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
 857		if (cpumask_empty(mask)) {
 858			free_cpumask_var(mask);
 859			return -EINVAL;
 860		}
 861	}
 862
 863	map = kzalloc(max_t(unsigned int,
 864			    RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
 865		      GFP_KERNEL);
 866	if (!map) {
 867		free_cpumask_var(mask);
 868		return -ENOMEM;
 869	}
 870
 871	i = 0;
 872	for_each_cpu_and(cpu, mask, cpu_online_mask)
 873		map->cpus[i++] = cpu;
 874
 875	if (i) {
 876		map->len = i;
 877	} else {
 878		kfree(map);
 879		map = NULL;
 880	}
 881
 882	mutex_lock(&rps_map_mutex);
 883	old_map = rcu_dereference_protected(queue->rps_map,
 884					    mutex_is_locked(&rps_map_mutex));
 885	rcu_assign_pointer(queue->rps_map, map);
 886
 887	if (map)
 888		static_branch_inc(&rps_needed);
 889	if (old_map)
 890		static_branch_dec(&rps_needed);
 891
 892	mutex_unlock(&rps_map_mutex);
 893
 894	if (old_map)
 895		kfree_rcu(old_map, rcu);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 896
 
 897	free_cpumask_var(mask);
 898	return len;
 899}
 900
 901static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 902					   char *buf)
 903{
 904	struct rps_dev_flow_table *flow_table;
 905	unsigned long val = 0;
 906
 907	rcu_read_lock();
 908	flow_table = rcu_dereference(queue->rps_flow_table);
 909	if (flow_table)
 910		val = (unsigned long)flow_table->mask + 1;
 911	rcu_read_unlock();
 912
 913	return sysfs_emit(buf, "%lu\n", val);
 914}
 915
 916static void rps_dev_flow_table_release(struct rcu_head *rcu)
 917{
 918	struct rps_dev_flow_table *table = container_of(rcu,
 919	    struct rps_dev_flow_table, rcu);
 920	vfree(table);
 921}
 922
 923static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
 924					    const char *buf, size_t len)
 925{
 926	unsigned long mask, count;
 927	struct rps_dev_flow_table *table, *old_table;
 928	static DEFINE_SPINLOCK(rps_dev_flow_lock);
 929	int rc;
 930
 931	if (!capable(CAP_NET_ADMIN))
 932		return -EPERM;
 933
 934	rc = kstrtoul(buf, 0, &count);
 935	if (rc < 0)
 936		return rc;
 937
 938	if (count) {
 939		mask = count - 1;
 940		/* mask = roundup_pow_of_two(count) - 1;
 941		 * without overflows...
 942		 */
 943		while ((mask | (mask >> 1)) != mask)
 944			mask |= (mask >> 1);
 945		/* On 64 bit arches, must check mask fits in table->mask (u32),
 946		 * and on 32bit arches, must check
 947		 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
 948		 */
 949#if BITS_PER_LONG > 32
 950		if (mask > (unsigned long)(u32)mask)
 951			return -EINVAL;
 952#else
 953		if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
 954				/ sizeof(struct rps_dev_flow)) {
 955			/* Enforce a limit to prevent overflow */
 956			return -EINVAL;
 957		}
 958#endif
 959		table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
 960		if (!table)
 961			return -ENOMEM;
 962
 963		table->mask = mask;
 964		for (count = 0; count <= mask; count++)
 965			table->flows[count].cpu = RPS_NO_CPU;
 966	} else {
 967		table = NULL;
 968	}
 969
 970	spin_lock(&rps_dev_flow_lock);
 971	old_table = rcu_dereference_protected(queue->rps_flow_table,
 972					      lockdep_is_held(&rps_dev_flow_lock));
 973	rcu_assign_pointer(queue->rps_flow_table, table);
 974	spin_unlock(&rps_dev_flow_lock);
 975
 976	if (old_table)
 977		call_rcu(&old_table->rcu, rps_dev_flow_table_release);
 978
 979	return len;
 980}
 981
 982static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
 983	= __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
 984
 985static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
 986	= __ATTR(rps_flow_cnt, 0644,
 987		 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
 988#endif /* CONFIG_RPS */
 989
 990static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
 991#ifdef CONFIG_RPS
 992	&rps_cpus_attribute.attr,
 993	&rps_dev_flow_table_cnt_attribute.attr,
 994#endif
 995	NULL
 996};
 997ATTRIBUTE_GROUPS(rx_queue_default);
 998
 999static void rx_queue_release(struct kobject *kobj)
1000{
1001	struct netdev_rx_queue *queue = to_rx_queue(kobj);
1002#ifdef CONFIG_RPS
1003	struct rps_map *map;
1004	struct rps_dev_flow_table *flow_table;
1005
1006	map = rcu_dereference_protected(queue->rps_map, 1);
1007	if (map) {
1008		RCU_INIT_POINTER(queue->rps_map, NULL);
1009		kfree_rcu(map, rcu);
1010	}
1011
1012	flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1013	if (flow_table) {
1014		RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1015		call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1016	}
1017#endif
1018
1019	memset(kobj, 0, sizeof(*kobj));
1020	netdev_put(queue->dev, &queue->dev_tracker);
1021}
1022
1023static const void *rx_queue_namespace(const struct kobject *kobj)
1024{
1025	struct netdev_rx_queue *queue = to_rx_queue(kobj);
1026	struct device *dev = &queue->dev->dev;
1027	const void *ns = NULL;
1028
1029	if (dev->class && dev->class->ns_type)
1030		ns = dev->class->namespace(dev);
1031
1032	return ns;
1033}
1034
1035static void rx_queue_get_ownership(const struct kobject *kobj,
1036				   kuid_t *uid, kgid_t *gid)
1037{
1038	const struct net *net = rx_queue_namespace(kobj);
1039
1040	net_ns_get_ownership(net, uid, gid);
1041}
1042
1043static struct kobj_type rx_queue_ktype __ro_after_init = {
1044	.sysfs_ops = &rx_queue_sysfs_ops,
1045	.release = rx_queue_release,
1046	.default_groups = rx_queue_default_groups,
1047	.namespace = rx_queue_namespace,
1048	.get_ownership = rx_queue_get_ownership,
1049};
1050
 
 
 
 
 
 
 
 
 
 
 
 
1051static int rx_queue_add_kobject(struct net_device *dev, int index)
1052{
1053	struct netdev_rx_queue *queue = dev->_rx + index;
1054	struct kobject *kobj = &queue->kobj;
1055	int error = 0;
1056
1057	/* Kobject_put later will trigger rx_queue_release call which
1058	 * decreases dev refcount: Take that reference here
1059	 */
1060	netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1061
1062	kobj->kset = dev->queues_kset;
1063	error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1064				     "rx-%u", index);
1065	if (error)
1066		goto err;
1067
1068	if (dev->sysfs_rx_queue_group) {
1069		error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1070		if (error)
1071			goto err;
1072	}
1073
 
 
 
 
1074	kobject_uevent(kobj, KOBJ_ADD);
1075
1076	return error;
1077
1078err:
1079	kobject_put(kobj);
1080	return error;
1081}
1082
1083static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1084				 kgid_t kgid)
1085{
1086	struct netdev_rx_queue *queue = dev->_rx + index;
1087	struct kobject *kobj = &queue->kobj;
1088	int error;
1089
1090	error = sysfs_change_owner(kobj, kuid, kgid);
1091	if (error)
1092		return error;
1093
1094	if (dev->sysfs_rx_queue_group)
1095		error = sysfs_group_change_owner(
1096			kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1097
1098	return error;
1099}
1100#endif /* CONFIG_SYSFS */
1101
1102int
1103net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1104{
1105#ifdef CONFIG_SYSFS
1106	int i;
1107	int error = 0;
1108
1109#ifndef CONFIG_RPS
1110	if (!dev->sysfs_rx_queue_group)
1111		return 0;
1112#endif
1113	for (i = old_num; i < new_num; i++) {
1114		error = rx_queue_add_kobject(dev, i);
1115		if (error) {
1116			new_num = old_num;
1117			break;
1118		}
1119	}
1120
1121	while (--i >= new_num) {
1122		struct kobject *kobj = &dev->_rx[i].kobj;
1123
1124		if (!refcount_read(&dev_net(dev)->ns.count))
1125			kobj->uevent_suppress = 1;
1126		if (dev->sysfs_rx_queue_group)
1127			sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1128		kobject_put(kobj);
1129	}
1130
1131	return error;
1132#else
1133	return 0;
1134#endif
1135}
1136
1137static int net_rx_queue_change_owner(struct net_device *dev, int num,
1138				     kuid_t kuid, kgid_t kgid)
1139{
1140#ifdef CONFIG_SYSFS
1141	int error = 0;
1142	int i;
1143
1144#ifndef CONFIG_RPS
1145	if (!dev->sysfs_rx_queue_group)
1146		return 0;
1147#endif
1148	for (i = 0; i < num; i++) {
1149		error = rx_queue_change_owner(dev, i, kuid, kgid);
1150		if (error)
1151			break;
1152	}
1153
1154	return error;
1155#else
1156	return 0;
1157#endif
1158}
1159
1160#ifdef CONFIG_SYSFS
1161/*
1162 * netdev_queue sysfs structures and functions.
1163 */
1164struct netdev_queue_attribute {
1165	struct attribute attr;
1166	ssize_t (*show)(struct netdev_queue *queue, char *buf);
1167	ssize_t (*store)(struct netdev_queue *queue,
1168			 const char *buf, size_t len);
1169};
1170#define to_netdev_queue_attr(_attr) \
1171	container_of(_attr, struct netdev_queue_attribute, attr)
1172
1173#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1174
1175static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1176				      struct attribute *attr, char *buf)
1177{
1178	const struct netdev_queue_attribute *attribute
1179		= to_netdev_queue_attr(attr);
1180	struct netdev_queue *queue = to_netdev_queue(kobj);
1181
1182	if (!attribute->show)
1183		return -EIO;
1184
1185	return attribute->show(queue, buf);
1186}
1187
1188static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1189				       struct attribute *attr,
1190				       const char *buf, size_t count)
1191{
1192	const struct netdev_queue_attribute *attribute
1193		= to_netdev_queue_attr(attr);
1194	struct netdev_queue *queue = to_netdev_queue(kobj);
1195
1196	if (!attribute->store)
1197		return -EIO;
1198
1199	return attribute->store(queue, buf, count);
1200}
1201
1202static const struct sysfs_ops netdev_queue_sysfs_ops = {
1203	.show = netdev_queue_attr_show,
1204	.store = netdev_queue_attr_store,
1205};
1206
1207static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1208{
1209	unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1210
1211	return sysfs_emit(buf, fmt_ulong, trans_timeout);
1212}
1213
1214static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1215{
1216	struct net_device *dev = queue->dev;
1217	unsigned int i;
1218
1219	i = queue - dev->_tx;
1220	BUG_ON(i >= dev->num_tx_queues);
1221
1222	return i;
1223}
1224
1225static ssize_t traffic_class_show(struct netdev_queue *queue,
1226				  char *buf)
1227{
1228	struct net_device *dev = queue->dev;
1229	int num_tc, tc;
1230	int index;
1231
1232	if (!netif_is_multiqueue(dev))
1233		return -ENOENT;
1234
1235	if (!rtnl_trylock())
1236		return restart_syscall();
1237
1238	index = get_netdev_queue_index(queue);
1239
1240	/* If queue belongs to subordinate dev use its TC mapping */
1241	dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1242
1243	num_tc = dev->num_tc;
1244	tc = netdev_txq_to_tc(dev, index);
1245
1246	rtnl_unlock();
1247
1248	if (tc < 0)
1249		return -EINVAL;
1250
1251	/* We can report the traffic class one of two ways:
1252	 * Subordinate device traffic classes are reported with the traffic
1253	 * class first, and then the subordinate class so for example TC0 on
1254	 * subordinate device 2 will be reported as "0-2". If the queue
1255	 * belongs to the root device it will be reported with just the
1256	 * traffic class, so just "0" for TC 0 for example.
1257	 */
1258	return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) :
1259			    sysfs_emit(buf, "%d\n", tc);
1260}
1261
1262#ifdef CONFIG_XPS
1263static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1264			       char *buf)
1265{
1266	return sysfs_emit(buf, "%lu\n", queue->tx_maxrate);
1267}
1268
1269static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1270				const char *buf, size_t len)
1271{
1272	struct net_device *dev = queue->dev;
1273	int err, index = get_netdev_queue_index(queue);
1274	u32 rate = 0;
1275
1276	if (!capable(CAP_NET_ADMIN))
1277		return -EPERM;
1278
1279	/* The check is also done later; this helps returning early without
1280	 * hitting the trylock/restart below.
1281	 */
1282	if (!dev->netdev_ops->ndo_set_tx_maxrate)
1283		return -EOPNOTSUPP;
1284
1285	err = kstrtou32(buf, 10, &rate);
1286	if (err < 0)
1287		return err;
1288
1289	if (!rtnl_trylock())
1290		return restart_syscall();
1291
1292	err = -EOPNOTSUPP;
1293	if (dev->netdev_ops->ndo_set_tx_maxrate)
1294		err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1295
1296	rtnl_unlock();
1297	if (!err) {
1298		queue->tx_maxrate = rate;
1299		return len;
1300	}
1301	return err;
1302}
1303
1304static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1305	= __ATTR_RW(tx_maxrate);
1306#endif
1307
1308static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1309	= __ATTR_RO(tx_timeout);
1310
1311static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1312	= __ATTR_RO(traffic_class);
1313
1314#ifdef CONFIG_BQL
1315/*
1316 * Byte queue limits sysfs structures and functions.
1317 */
1318static ssize_t bql_show(char *buf, unsigned int value)
1319{
1320	return sysfs_emit(buf, "%u\n", value);
1321}
1322
1323static ssize_t bql_set(const char *buf, const size_t count,
1324		       unsigned int *pvalue)
1325{
1326	unsigned int value;
1327	int err;
1328
1329	if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1330		value = DQL_MAX_LIMIT;
1331	} else {
1332		err = kstrtouint(buf, 10, &value);
1333		if (err < 0)
1334			return err;
1335		if (value > DQL_MAX_LIMIT)
1336			return -EINVAL;
1337	}
1338
1339	*pvalue = value;
1340
1341	return count;
1342}
1343
1344static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1345				  char *buf)
1346{
1347	struct dql *dql = &queue->dql;
1348
1349	return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1350}
1351
1352static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1353				 const char *buf, size_t len)
1354{
1355	struct dql *dql = &queue->dql;
1356	unsigned int value;
1357	int err;
1358
1359	err = kstrtouint(buf, 10, &value);
1360	if (err < 0)
1361		return err;
1362
1363	dql->slack_hold_time = msecs_to_jiffies(value);
1364
1365	return len;
1366}
1367
1368static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1369	= __ATTR(hold_time, 0644,
1370		 bql_show_hold_time, bql_set_hold_time);
1371
1372static ssize_t bql_show_inflight(struct netdev_queue *queue,
1373				 char *buf)
1374{
1375	struct dql *dql = &queue->dql;
1376
1377	return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed);
1378}
1379
1380static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1381	__ATTR(inflight, 0444, bql_show_inflight, NULL);
1382
1383#define BQL_ATTR(NAME, FIELD)						\
1384static ssize_t bql_show_ ## NAME(struct netdev_queue *queue,		\
1385				 char *buf)				\
1386{									\
1387	return bql_show(buf, queue->dql.FIELD);				\
1388}									\
1389									\
1390static ssize_t bql_set_ ## NAME(struct netdev_queue *queue,		\
1391				const char *buf, size_t len)		\
1392{									\
1393	return bql_set(buf, len, &queue->dql.FIELD);			\
1394}									\
1395									\
1396static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1397	= __ATTR(NAME, 0644,				\
1398		 bql_show_ ## NAME, bql_set_ ## NAME)
1399
1400BQL_ATTR(limit, limit);
1401BQL_ATTR(limit_max, max_limit);
1402BQL_ATTR(limit_min, min_limit);
1403
1404static struct attribute *dql_attrs[] __ro_after_init = {
1405	&bql_limit_attribute.attr,
1406	&bql_limit_max_attribute.attr,
1407	&bql_limit_min_attribute.attr,
1408	&bql_hold_time_attribute.attr,
1409	&bql_inflight_attribute.attr,
1410	NULL
1411};
1412
1413static const struct attribute_group dql_group = {
1414	.name  = "byte_queue_limits",
1415	.attrs  = dql_attrs,
1416};
1417#endif /* CONFIG_BQL */
1418
1419#ifdef CONFIG_XPS
1420static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1421			      int tc, char *buf, enum xps_map_type type)
1422{
1423	struct xps_dev_maps *dev_maps;
1424	unsigned long *mask;
1425	unsigned int nr_ids;
1426	int j, len;
1427
1428	rcu_read_lock();
1429	dev_maps = rcu_dereference(dev->xps_maps[type]);
1430
1431	/* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1432	 * when dev_maps hasn't been allocated yet, to be backward compatible.
1433	 */
1434	nr_ids = dev_maps ? dev_maps->nr_ids :
1435		 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1436
1437	mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1438	if (!mask) {
1439		rcu_read_unlock();
1440		return -ENOMEM;
1441	}
1442
1443	if (!dev_maps || tc >= dev_maps->num_tc)
1444		goto out_no_maps;
1445
1446	for (j = 0; j < nr_ids; j++) {
1447		int i, tci = j * dev_maps->num_tc + tc;
1448		struct xps_map *map;
1449
1450		map = rcu_dereference(dev_maps->attr_map[tci]);
1451		if (!map)
1452			continue;
1453
1454		for (i = map->len; i--;) {
1455			if (map->queues[i] == index) {
1456				__set_bit(j, mask);
1457				break;
1458			}
1459		}
1460	}
1461out_no_maps:
1462	rcu_read_unlock();
1463
1464	len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1465	bitmap_free(mask);
1466
1467	return len < PAGE_SIZE ? len : -EINVAL;
1468}
1469
1470static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1471{
1472	struct net_device *dev = queue->dev;
1473	unsigned int index;
1474	int len, tc;
1475
1476	if (!netif_is_multiqueue(dev))
1477		return -ENOENT;
1478
1479	index = get_netdev_queue_index(queue);
1480
1481	if (!rtnl_trylock())
1482		return restart_syscall();
1483
1484	/* If queue belongs to subordinate dev use its map */
1485	dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1486
1487	tc = netdev_txq_to_tc(dev, index);
1488	if (tc < 0) {
1489		rtnl_unlock();
1490		return -EINVAL;
1491	}
1492
1493	/* Make sure the subordinate device can't be freed */
1494	get_device(&dev->dev);
1495	rtnl_unlock();
1496
1497	len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1498
1499	put_device(&dev->dev);
1500	return len;
1501}
1502
1503static ssize_t xps_cpus_store(struct netdev_queue *queue,
1504			      const char *buf, size_t len)
1505{
1506	struct net_device *dev = queue->dev;
1507	unsigned int index;
1508	cpumask_var_t mask;
1509	int err;
1510
1511	if (!netif_is_multiqueue(dev))
1512		return -ENOENT;
1513
1514	if (!capable(CAP_NET_ADMIN))
1515		return -EPERM;
1516
1517	if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1518		return -ENOMEM;
1519
1520	index = get_netdev_queue_index(queue);
1521
1522	err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1523	if (err) {
1524		free_cpumask_var(mask);
1525		return err;
1526	}
1527
1528	if (!rtnl_trylock()) {
1529		free_cpumask_var(mask);
1530		return restart_syscall();
1531	}
1532
1533	err = netif_set_xps_queue(dev, mask, index);
1534	rtnl_unlock();
1535
1536	free_cpumask_var(mask);
1537
1538	return err ? : len;
1539}
1540
1541static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1542	= __ATTR_RW(xps_cpus);
1543
1544static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1545{
1546	struct net_device *dev = queue->dev;
1547	unsigned int index;
1548	int tc;
1549
1550	index = get_netdev_queue_index(queue);
1551
1552	if (!rtnl_trylock())
1553		return restart_syscall();
1554
1555	tc = netdev_txq_to_tc(dev, index);
1556	rtnl_unlock();
1557	if (tc < 0)
1558		return -EINVAL;
1559
1560	return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1561}
1562
1563static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1564			      size_t len)
1565{
1566	struct net_device *dev = queue->dev;
1567	struct net *net = dev_net(dev);
1568	unsigned long *mask;
1569	unsigned int index;
1570	int err;
1571
1572	if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1573		return -EPERM;
1574
1575	mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1576	if (!mask)
1577		return -ENOMEM;
1578
1579	index = get_netdev_queue_index(queue);
1580
1581	err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1582	if (err) {
1583		bitmap_free(mask);
1584		return err;
1585	}
1586
1587	if (!rtnl_trylock()) {
1588		bitmap_free(mask);
1589		return restart_syscall();
1590	}
1591
1592	cpus_read_lock();
1593	err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1594	cpus_read_unlock();
1595
1596	rtnl_unlock();
1597
1598	bitmap_free(mask);
1599	return err ? : len;
1600}
1601
1602static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1603	= __ATTR_RW(xps_rxqs);
1604#endif /* CONFIG_XPS */
1605
1606static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1607	&queue_trans_timeout.attr,
1608	&queue_traffic_class.attr,
1609#ifdef CONFIG_XPS
1610	&xps_cpus_attribute.attr,
1611	&xps_rxqs_attribute.attr,
1612	&queue_tx_maxrate.attr,
1613#endif
1614	NULL
1615};
1616ATTRIBUTE_GROUPS(netdev_queue_default);
1617
1618static void netdev_queue_release(struct kobject *kobj)
1619{
1620	struct netdev_queue *queue = to_netdev_queue(kobj);
1621
1622	memset(kobj, 0, sizeof(*kobj));
1623	netdev_put(queue->dev, &queue->dev_tracker);
1624}
1625
1626static const void *netdev_queue_namespace(const struct kobject *kobj)
1627{
1628	struct netdev_queue *queue = to_netdev_queue(kobj);
1629	struct device *dev = &queue->dev->dev;
1630	const void *ns = NULL;
1631
1632	if (dev->class && dev->class->ns_type)
1633		ns = dev->class->namespace(dev);
1634
1635	return ns;
1636}
1637
1638static void netdev_queue_get_ownership(const struct kobject *kobj,
1639				       kuid_t *uid, kgid_t *gid)
1640{
1641	const struct net *net = netdev_queue_namespace(kobj);
1642
1643	net_ns_get_ownership(net, uid, gid);
1644}
1645
1646static struct kobj_type netdev_queue_ktype __ro_after_init = {
1647	.sysfs_ops = &netdev_queue_sysfs_ops,
1648	.release = netdev_queue_release,
1649	.default_groups = netdev_queue_default_groups,
1650	.namespace = netdev_queue_namespace,
1651	.get_ownership = netdev_queue_get_ownership,
1652};
1653
1654static int netdev_queue_add_kobject(struct net_device *dev, int index)
1655{
1656	struct netdev_queue *queue = dev->_tx + index;
1657	struct kobject *kobj = &queue->kobj;
1658	int error = 0;
1659
1660	/* Kobject_put later will trigger netdev_queue_release call
1661	 * which decreases dev refcount: Take that reference here
1662	 */
1663	netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1664
1665	kobj->kset = dev->queues_kset;
1666	error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1667				     "tx-%u", index);
1668	if (error)
1669		goto err;
1670
1671#ifdef CONFIG_BQL
1672	error = sysfs_create_group(kobj, &dql_group);
1673	if (error)
1674		goto err;
1675#endif
1676
1677	kobject_uevent(kobj, KOBJ_ADD);
1678	return 0;
1679
1680err:
1681	kobject_put(kobj);
1682	return error;
1683}
1684
1685static int tx_queue_change_owner(struct net_device *ndev, int index,
1686				 kuid_t kuid, kgid_t kgid)
1687{
1688	struct netdev_queue *queue = ndev->_tx + index;
1689	struct kobject *kobj = &queue->kobj;
1690	int error;
1691
1692	error = sysfs_change_owner(kobj, kuid, kgid);
1693	if (error)
1694		return error;
1695
1696#ifdef CONFIG_BQL
1697	error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1698#endif
1699	return error;
1700}
1701#endif /* CONFIG_SYSFS */
1702
1703int
1704netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1705{
1706#ifdef CONFIG_SYSFS
1707	int i;
1708	int error = 0;
1709
1710	/* Tx queue kobjects are allowed to be updated when a device is being
1711	 * unregistered, but solely to remove queues from qdiscs. Any path
1712	 * adding queues should be fixed.
1713	 */
1714	WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1715	     "New queues can't be registered after device unregistration.");
1716
1717	for (i = old_num; i < new_num; i++) {
1718		error = netdev_queue_add_kobject(dev, i);
1719		if (error) {
1720			new_num = old_num;
1721			break;
1722		}
1723	}
1724
1725	while (--i >= new_num) {
1726		struct netdev_queue *queue = dev->_tx + i;
1727
1728		if (!refcount_read(&dev_net(dev)->ns.count))
1729			queue->kobj.uevent_suppress = 1;
1730#ifdef CONFIG_BQL
1731		sysfs_remove_group(&queue->kobj, &dql_group);
1732#endif
1733		kobject_put(&queue->kobj);
1734	}
1735
1736	return error;
1737#else
1738	return 0;
1739#endif /* CONFIG_SYSFS */
1740}
1741
1742static int net_tx_queue_change_owner(struct net_device *dev, int num,
1743				     kuid_t kuid, kgid_t kgid)
1744{
1745#ifdef CONFIG_SYSFS
1746	int error = 0;
1747	int i;
1748
1749	for (i = 0; i < num; i++) {
1750		error = tx_queue_change_owner(dev, i, kuid, kgid);
1751		if (error)
1752			break;
1753	}
1754
1755	return error;
1756#else
1757	return 0;
1758#endif /* CONFIG_SYSFS */
1759}
1760
1761static int register_queue_kobjects(struct net_device *dev)
1762{
1763	int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1764
1765#ifdef CONFIG_SYSFS
1766	dev->queues_kset = kset_create_and_add("queues",
1767					       NULL, &dev->dev.kobj);
1768	if (!dev->queues_kset)
1769		return -ENOMEM;
1770	real_rx = dev->real_num_rx_queues;
1771#endif
1772	real_tx = dev->real_num_tx_queues;
1773
1774	error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1775	if (error)
1776		goto error;
1777	rxq = real_rx;
1778
1779	error = netdev_queue_update_kobjects(dev, 0, real_tx);
1780	if (error)
1781		goto error;
1782	txq = real_tx;
1783
1784	return 0;
1785
1786error:
1787	netdev_queue_update_kobjects(dev, txq, 0);
1788	net_rx_queue_update_kobjects(dev, rxq, 0);
1789#ifdef CONFIG_SYSFS
1790	kset_unregister(dev->queues_kset);
1791#endif
1792	return error;
1793}
1794
1795static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1796{
1797	int error = 0, real_rx = 0, real_tx = 0;
1798
1799#ifdef CONFIG_SYSFS
1800	if (ndev->queues_kset) {
1801		error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1802		if (error)
1803			return error;
1804	}
1805	real_rx = ndev->real_num_rx_queues;
1806#endif
1807	real_tx = ndev->real_num_tx_queues;
1808
1809	error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1810	if (error)
1811		return error;
1812
1813	error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1814	if (error)
1815		return error;
1816
1817	return 0;
1818}
1819
1820static void remove_queue_kobjects(struct net_device *dev)
1821{
1822	int real_rx = 0, real_tx = 0;
1823
1824#ifdef CONFIG_SYSFS
1825	real_rx = dev->real_num_rx_queues;
1826#endif
1827	real_tx = dev->real_num_tx_queues;
1828
1829	net_rx_queue_update_kobjects(dev, real_rx, 0);
1830	netdev_queue_update_kobjects(dev, real_tx, 0);
1831
1832	dev->real_num_rx_queues = 0;
1833	dev->real_num_tx_queues = 0;
1834#ifdef CONFIG_SYSFS
1835	kset_unregister(dev->queues_kset);
1836#endif
1837}
1838
1839static bool net_current_may_mount(void)
1840{
1841	struct net *net = current->nsproxy->net_ns;
1842
1843	return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1844}
1845
1846static void *net_grab_current_ns(void)
1847{
1848	struct net *ns = current->nsproxy->net_ns;
1849#ifdef CONFIG_NET_NS
1850	if (ns)
1851		refcount_inc(&ns->passive);
1852#endif
1853	return ns;
1854}
1855
1856static const void *net_initial_ns(void)
1857{
1858	return &init_net;
1859}
1860
1861static const void *net_netlink_ns(struct sock *sk)
1862{
1863	return sock_net(sk);
1864}
1865
1866const struct kobj_ns_type_operations net_ns_type_operations = {
1867	.type = KOBJ_NS_TYPE_NET,
1868	.current_may_mount = net_current_may_mount,
1869	.grab_current_ns = net_grab_current_ns,
1870	.netlink_ns = net_netlink_ns,
1871	.initial_ns = net_initial_ns,
1872	.drop_ns = net_drop_ns,
1873};
1874EXPORT_SYMBOL_GPL(net_ns_type_operations);
1875
1876static int netdev_uevent(const struct device *d, struct kobj_uevent_env *env)
1877{
1878	const struct net_device *dev = to_net_dev(d);
1879	int retval;
1880
1881	/* pass interface to uevent. */
1882	retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1883	if (retval)
1884		goto exit;
1885
1886	/* pass ifindex to uevent.
1887	 * ifindex is useful as it won't change (interface name may change)
1888	 * and is what RtNetlink uses natively.
1889	 */
1890	retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1891
1892exit:
1893	return retval;
1894}
1895
1896/*
1897 *	netdev_release -- destroy and free a dead device.
1898 *	Called when last reference to device kobject is gone.
1899 */
1900static void netdev_release(struct device *d)
1901{
1902	struct net_device *dev = to_net_dev(d);
1903
1904	BUG_ON(dev->reg_state != NETREG_RELEASED);
1905
1906	/* no need to wait for rcu grace period:
1907	 * device is dead and about to be freed.
1908	 */
1909	kfree(rcu_access_pointer(dev->ifalias));
1910	netdev_freemem(dev);
1911}
1912
1913static const void *net_namespace(const struct device *d)
1914{
1915	const struct net_device *dev = to_net_dev(d);
1916
1917	return dev_net(dev);
1918}
1919
1920static void net_get_ownership(const struct device *d, kuid_t *uid, kgid_t *gid)
1921{
1922	const struct net_device *dev = to_net_dev(d);
1923	const struct net *net = dev_net(dev);
1924
1925	net_ns_get_ownership(net, uid, gid);
1926}
1927
1928static struct class net_class __ro_after_init = {
1929	.name = "net",
1930	.dev_release = netdev_release,
1931	.dev_groups = net_class_groups,
1932	.dev_uevent = netdev_uevent,
1933	.ns_type = &net_ns_type_operations,
1934	.namespace = net_namespace,
1935	.get_ownership = net_get_ownership,
1936};
1937
1938#ifdef CONFIG_OF
1939static int of_dev_node_match(struct device *dev, const void *data)
1940{
1941	for (; dev; dev = dev->parent) {
1942		if (dev->of_node == data)
1943			return 1;
1944	}
1945
1946	return 0;
1947}
1948
1949/*
1950 * of_find_net_device_by_node - lookup the net device for the device node
1951 * @np: OF device node
1952 *
1953 * Looks up the net_device structure corresponding with the device node.
1954 * If successful, returns a pointer to the net_device with the embedded
1955 * struct device refcount incremented by one, or NULL on failure. The
1956 * refcount must be dropped when done with the net_device.
1957 */
1958struct net_device *of_find_net_device_by_node(struct device_node *np)
1959{
1960	struct device *dev;
1961
1962	dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1963	if (!dev)
1964		return NULL;
1965
1966	return to_net_dev(dev);
1967}
1968EXPORT_SYMBOL(of_find_net_device_by_node);
1969#endif
1970
1971/* Delete sysfs entries but hold kobject reference until after all
1972 * netdev references are gone.
1973 */
1974void netdev_unregister_kobject(struct net_device *ndev)
1975{
1976	struct device *dev = &ndev->dev;
1977
1978	if (!refcount_read(&dev_net(ndev)->ns.count))
1979		dev_set_uevent_suppress(dev, 1);
1980
1981	kobject_get(&dev->kobj);
1982
1983	remove_queue_kobjects(ndev);
1984
1985	pm_runtime_set_memalloc_noio(dev, false);
1986
1987	device_del(dev);
1988}
1989
1990/* Create sysfs entries for network device. */
1991int netdev_register_kobject(struct net_device *ndev)
1992{
1993	struct device *dev = &ndev->dev;
1994	const struct attribute_group **groups = ndev->sysfs_groups;
1995	int error = 0;
1996
1997	device_initialize(dev);
1998	dev->class = &net_class;
1999	dev->platform_data = ndev;
2000	dev->groups = groups;
2001
2002	dev_set_name(dev, "%s", ndev->name);
2003
2004#ifdef CONFIG_SYSFS
2005	/* Allow for a device specific group */
2006	if (*groups)
2007		groups++;
2008
2009	*groups++ = &netstat_group;
2010
2011	if (wireless_group_needed(ndev))
2012		*groups++ = &wireless_group;
2013#endif /* CONFIG_SYSFS */
2014
2015	error = device_add(dev);
2016	if (error)
2017		return error;
2018
2019	error = register_queue_kobjects(ndev);
2020	if (error) {
2021		device_del(dev);
2022		return error;
2023	}
2024
2025	pm_runtime_set_memalloc_noio(dev, true);
2026
2027	return error;
2028}
2029
2030/* Change owner for sysfs entries when moving network devices across network
2031 * namespaces owned by different user namespaces.
2032 */
2033int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2034			const struct net *net_new)
2035{
2036	kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2037	kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2038	struct device *dev = &ndev->dev;
2039	int error;
2040
2041	net_ns_get_ownership(net_old, &old_uid, &old_gid);
2042	net_ns_get_ownership(net_new, &new_uid, &new_gid);
2043
2044	/* The network namespace was changed but the owning user namespace is
2045	 * identical so there's no need to change the owner of sysfs entries.
2046	 */
2047	if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2048		return 0;
2049
2050	error = device_change_owner(dev, new_uid, new_gid);
2051	if (error)
2052		return error;
2053
2054	error = queue_change_owner(ndev, new_uid, new_gid);
2055	if (error)
2056		return error;
2057
2058	return 0;
2059}
2060
2061int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2062				const void *ns)
2063{
2064	return class_create_file_ns(&net_class, class_attr, ns);
2065}
2066EXPORT_SYMBOL(netdev_class_create_file_ns);
2067
2068void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2069				 const void *ns)
2070{
2071	class_remove_file_ns(&net_class, class_attr, ns);
2072}
2073EXPORT_SYMBOL(netdev_class_remove_file_ns);
2074
2075int __init netdev_kobject_init(void)
2076{
2077	kobj_ns_type_register(&net_ns_type_operations);
2078	return class_register(&net_class);
2079}