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 <net/switchdev.h> 16 #include <linux/if_arp.h> 17 #include <linux/slab.h> 18 #include <linux/sched/signal.h> 19 #include <linux/nsproxy.h> 20 #include <net/sock.h> 21 #include <net/net_namespace.h> 22 #include <linux/rtnetlink.h> 23 #include <linux/vmalloc.h> 24 #include <linux/export.h> 25 #include <linux/jiffies.h> 26 #include <linux/pm_runtime.h> 27 #include <linux/of.h> 28 #include <linux/of_net.h> 29 30 #include "net-sysfs.h" 31 32 #ifdef CONFIG_SYSFS 33 static const char fmt_hex[] = "%#x\n"; 34 static const char fmt_dec[] = "%d\n"; 35 static const char fmt_ulong[] = "%lu\n"; 36 static const char fmt_u64[] = "%llu\n"; 37 38 static inline int dev_isalive(const struct net_device *dev) 39 { 40 return dev->reg_state <= NETREG_REGISTERED; 41 } 42 43 /* use same locking rules as GIF* ioctl's */ 44 static ssize_t netdev_show(const struct device *dev, 45 struct device_attribute *attr, char *buf, 46 ssize_t (*format)(const struct net_device *, char *)) 47 { 48 struct net_device *ndev = to_net_dev(dev); 49 ssize_t ret = -EINVAL; 50 51 read_lock(&dev_base_lock); 52 if (dev_isalive(ndev)) 53 ret = (*format)(ndev, buf); 54 read_unlock(&dev_base_lock); 55 56 return ret; 57 } 58 59 /* generate a show function for simple field */ 60 #define NETDEVICE_SHOW(field, format_string) \ 61 static ssize_t format_##field(const struct net_device *dev, char *buf) \ 62 { \ 63 return sprintf(buf, format_string, dev->field); \ 64 } \ 65 static ssize_t field##_show(struct device *dev, \ 66 struct device_attribute *attr, char *buf) \ 67 { \ 68 return netdev_show(dev, attr, buf, format_##field); \ 69 } \ 70 71 #define NETDEVICE_SHOW_RO(field, format_string) \ 72 NETDEVICE_SHOW(field, format_string); \ 73 static DEVICE_ATTR_RO(field) 74 75 #define NETDEVICE_SHOW_RW(field, format_string) \ 76 NETDEVICE_SHOW(field, format_string); \ 77 static DEVICE_ATTR_RW(field) 78 79 /* use same locking and permission rules as SIF* ioctl's */ 80 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr, 81 const char *buf, size_t len, 82 int (*set)(struct net_device *, unsigned long)) 83 { 84 struct net_device *netdev = to_net_dev(dev); 85 struct net *net = dev_net(netdev); 86 unsigned long new; 87 int ret = -EINVAL; 88 89 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 90 return -EPERM; 91 92 ret = kstrtoul(buf, 0, &new); 93 if (ret) 94 goto err; 95 96 if (!rtnl_trylock()) 97 return restart_syscall(); 98 99 if (dev_isalive(netdev)) { 100 ret = (*set)(netdev, new); 101 if (ret == 0) 102 ret = len; 103 } 104 rtnl_unlock(); 105 err: 106 return ret; 107 } 108 109 NETDEVICE_SHOW_RO(dev_id, fmt_hex); 110 NETDEVICE_SHOW_RO(dev_port, fmt_dec); 111 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec); 112 NETDEVICE_SHOW_RO(addr_len, fmt_dec); 113 NETDEVICE_SHOW_RO(ifindex, fmt_dec); 114 NETDEVICE_SHOW_RO(type, fmt_dec); 115 NETDEVICE_SHOW_RO(link_mode, fmt_dec); 116 117 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr, 118 char *buf) 119 { 120 struct net_device *ndev = to_net_dev(dev); 121 122 return sprintf(buf, fmt_dec, dev_get_iflink(ndev)); 123 } 124 static DEVICE_ATTR_RO(iflink); 125 126 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf) 127 { 128 return sprintf(buf, fmt_dec, dev->name_assign_type); 129 } 130 131 static ssize_t name_assign_type_show(struct device *dev, 132 struct device_attribute *attr, 133 char *buf) 134 { 135 struct net_device *ndev = to_net_dev(dev); 136 ssize_t ret = -EINVAL; 137 138 if (ndev->name_assign_type != NET_NAME_UNKNOWN) 139 ret = netdev_show(dev, attr, buf, format_name_assign_type); 140 141 return ret; 142 } 143 static DEVICE_ATTR_RO(name_assign_type); 144 145 /* use same locking rules as GIFHWADDR ioctl's */ 146 static ssize_t address_show(struct device *dev, struct device_attribute *attr, 147 char *buf) 148 { 149 struct net_device *ndev = to_net_dev(dev); 150 ssize_t ret = -EINVAL; 151 152 read_lock(&dev_base_lock); 153 if (dev_isalive(ndev)) 154 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len); 155 read_unlock(&dev_base_lock); 156 return ret; 157 } 158 static DEVICE_ATTR_RO(address); 159 160 static ssize_t broadcast_show(struct device *dev, 161 struct device_attribute *attr, char *buf) 162 { 163 struct net_device *ndev = to_net_dev(dev); 164 165 if (dev_isalive(ndev)) 166 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len); 167 return -EINVAL; 168 } 169 static DEVICE_ATTR_RO(broadcast); 170 171 static int change_carrier(struct net_device *dev, unsigned long new_carrier) 172 { 173 if (!netif_running(dev)) 174 return -EINVAL; 175 return dev_change_carrier(dev, (bool)new_carrier); 176 } 177 178 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr, 179 const char *buf, size_t len) 180 { 181 return netdev_store(dev, attr, buf, len, change_carrier); 182 } 183 184 static ssize_t carrier_show(struct device *dev, 185 struct device_attribute *attr, char *buf) 186 { 187 struct net_device *netdev = to_net_dev(dev); 188 189 if (netif_running(netdev)) 190 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev)); 191 192 return -EINVAL; 193 } 194 static DEVICE_ATTR_RW(carrier); 195 196 static ssize_t speed_show(struct device *dev, 197 struct device_attribute *attr, char *buf) 198 { 199 struct net_device *netdev = to_net_dev(dev); 200 int ret = -EINVAL; 201 202 if (!rtnl_trylock()) 203 return restart_syscall(); 204 205 if (netif_running(netdev)) { 206 struct ethtool_link_ksettings cmd; 207 208 if (!__ethtool_get_link_ksettings(netdev, &cmd)) 209 ret = sprintf(buf, fmt_dec, cmd.base.speed); 210 } 211 rtnl_unlock(); 212 return ret; 213 } 214 static DEVICE_ATTR_RO(speed); 215 216 static ssize_t duplex_show(struct device *dev, 217 struct device_attribute *attr, char *buf) 218 { 219 struct net_device *netdev = to_net_dev(dev); 220 int ret = -EINVAL; 221 222 if (!rtnl_trylock()) 223 return restart_syscall(); 224 225 if (netif_running(netdev)) { 226 struct ethtool_link_ksettings cmd; 227 228 if (!__ethtool_get_link_ksettings(netdev, &cmd)) { 229 const char *duplex; 230 231 switch (cmd.base.duplex) { 232 case DUPLEX_HALF: 233 duplex = "half"; 234 break; 235 case DUPLEX_FULL: 236 duplex = "full"; 237 break; 238 default: 239 duplex = "unknown"; 240 break; 241 } 242 ret = sprintf(buf, "%s\n", duplex); 243 } 244 } 245 rtnl_unlock(); 246 return ret; 247 } 248 static DEVICE_ATTR_RO(duplex); 249 250 static ssize_t dormant_show(struct device *dev, 251 struct device_attribute *attr, char *buf) 252 { 253 struct net_device *netdev = to_net_dev(dev); 254 255 if (netif_running(netdev)) 256 return sprintf(buf, fmt_dec, !!netif_dormant(netdev)); 257 258 return -EINVAL; 259 } 260 static DEVICE_ATTR_RO(dormant); 261 262 static const char *const operstates[] = { 263 "unknown", 264 "notpresent", /* currently unused */ 265 "down", 266 "lowerlayerdown", 267 "testing", /* currently unused */ 268 "dormant", 269 "up" 270 }; 271 272 static ssize_t operstate_show(struct device *dev, 273 struct device_attribute *attr, char *buf) 274 { 275 const struct net_device *netdev = to_net_dev(dev); 276 unsigned char operstate; 277 278 read_lock(&dev_base_lock); 279 operstate = netdev->operstate; 280 if (!netif_running(netdev)) 281 operstate = IF_OPER_DOWN; 282 read_unlock(&dev_base_lock); 283 284 if (operstate >= ARRAY_SIZE(operstates)) 285 return -EINVAL; /* should not happen */ 286 287 return sprintf(buf, "%s\n", operstates[operstate]); 288 } 289 static DEVICE_ATTR_RO(operstate); 290 291 static ssize_t carrier_changes_show(struct device *dev, 292 struct device_attribute *attr, 293 char *buf) 294 { 295 struct net_device *netdev = to_net_dev(dev); 296 297 return sprintf(buf, fmt_dec, 298 atomic_read(&netdev->carrier_changes)); 299 } 300 static DEVICE_ATTR_RO(carrier_changes); 301 302 /* read-write attributes */ 303 304 static int change_mtu(struct net_device *dev, unsigned long new_mtu) 305 { 306 return dev_set_mtu(dev, (int)new_mtu); 307 } 308 309 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr, 310 const char *buf, size_t len) 311 { 312 return netdev_store(dev, attr, buf, len, change_mtu); 313 } 314 NETDEVICE_SHOW_RW(mtu, fmt_dec); 315 316 static int change_flags(struct net_device *dev, unsigned long new_flags) 317 { 318 return dev_change_flags(dev, (unsigned int)new_flags); 319 } 320 321 static ssize_t flags_store(struct device *dev, struct device_attribute *attr, 322 const char *buf, size_t len) 323 { 324 return netdev_store(dev, attr, buf, len, change_flags); 325 } 326 NETDEVICE_SHOW_RW(flags, fmt_hex); 327 328 static int change_tx_queue_len(struct net_device *dev, unsigned long new_len) 329 { 330 unsigned int orig_len = dev->tx_queue_len; 331 int res; 332 333 if (new_len != (unsigned int)new_len) 334 return -ERANGE; 335 336 if (new_len != orig_len) { 337 dev->tx_queue_len = new_len; 338 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); 339 res = notifier_to_errno(res); 340 if (res) { 341 netdev_err(dev, 342 "refused to change device tx_queue_len\n"); 343 dev->tx_queue_len = orig_len; 344 return -EFAULT; 345 } 346 } 347 348 return 0; 349 } 350 351 static ssize_t tx_queue_len_store(struct device *dev, 352 struct device_attribute *attr, 353 const char *buf, size_t len) 354 { 355 if (!capable(CAP_NET_ADMIN)) 356 return -EPERM; 357 358 return netdev_store(dev, attr, buf, len, change_tx_queue_len); 359 } 360 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec); 361 362 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val) 363 { 364 dev->gro_flush_timeout = val; 365 return 0; 366 } 367 368 static ssize_t gro_flush_timeout_store(struct device *dev, 369 struct device_attribute *attr, 370 const char *buf, size_t len) 371 { 372 if (!capable(CAP_NET_ADMIN)) 373 return -EPERM; 374 375 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout); 376 } 377 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong); 378 379 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr, 380 const char *buf, size_t len) 381 { 382 struct net_device *netdev = to_net_dev(dev); 383 struct net *net = dev_net(netdev); 384 size_t count = len; 385 ssize_t ret = 0; 386 387 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 388 return -EPERM; 389 390 /* ignore trailing newline */ 391 if (len > 0 && buf[len - 1] == '\n') 392 --count; 393 394 if (!rtnl_trylock()) 395 return restart_syscall(); 396 397 if (dev_isalive(netdev)) { 398 ret = dev_set_alias(netdev, buf, count); 399 if (ret < 0) 400 goto err; 401 ret = len; 402 netdev_state_change(netdev); 403 } 404 err: 405 rtnl_unlock(); 406 407 return ret; 408 } 409 410 static ssize_t ifalias_show(struct device *dev, 411 struct device_attribute *attr, char *buf) 412 { 413 const struct net_device *netdev = to_net_dev(dev); 414 char tmp[IFALIASZ]; 415 ssize_t ret = 0; 416 417 ret = dev_get_alias(netdev, tmp, sizeof(tmp)); 418 if (ret > 0) 419 ret = sprintf(buf, "%s\n", tmp); 420 return ret; 421 } 422 static DEVICE_ATTR_RW(ifalias); 423 424 static int change_group(struct net_device *dev, unsigned long new_group) 425 { 426 dev_set_group(dev, (int)new_group); 427 return 0; 428 } 429 430 static ssize_t group_store(struct device *dev, struct device_attribute *attr, 431 const char *buf, size_t len) 432 { 433 return netdev_store(dev, attr, buf, len, change_group); 434 } 435 NETDEVICE_SHOW(group, fmt_dec); 436 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store); 437 438 static int change_proto_down(struct net_device *dev, unsigned long proto_down) 439 { 440 return dev_change_proto_down(dev, (bool)proto_down); 441 } 442 443 static ssize_t proto_down_store(struct device *dev, 444 struct device_attribute *attr, 445 const char *buf, size_t len) 446 { 447 return netdev_store(dev, attr, buf, len, change_proto_down); 448 } 449 NETDEVICE_SHOW_RW(proto_down, fmt_dec); 450 451 static ssize_t phys_port_id_show(struct device *dev, 452 struct device_attribute *attr, char *buf) 453 { 454 struct net_device *netdev = to_net_dev(dev); 455 ssize_t ret = -EINVAL; 456 457 if (!rtnl_trylock()) 458 return restart_syscall(); 459 460 if (dev_isalive(netdev)) { 461 struct netdev_phys_item_id ppid; 462 463 ret = dev_get_phys_port_id(netdev, &ppid); 464 if (!ret) 465 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id); 466 } 467 rtnl_unlock(); 468 469 return ret; 470 } 471 static DEVICE_ATTR_RO(phys_port_id); 472 473 static ssize_t phys_port_name_show(struct device *dev, 474 struct device_attribute *attr, char *buf) 475 { 476 struct net_device *netdev = to_net_dev(dev); 477 ssize_t ret = -EINVAL; 478 479 if (!rtnl_trylock()) 480 return restart_syscall(); 481 482 if (dev_isalive(netdev)) { 483 char name[IFNAMSIZ]; 484 485 ret = dev_get_phys_port_name(netdev, name, sizeof(name)); 486 if (!ret) 487 ret = sprintf(buf, "%s\n", name); 488 } 489 rtnl_unlock(); 490 491 return ret; 492 } 493 static DEVICE_ATTR_RO(phys_port_name); 494 495 static ssize_t phys_switch_id_show(struct device *dev, 496 struct device_attribute *attr, char *buf) 497 { 498 struct net_device *netdev = to_net_dev(dev); 499 ssize_t ret = -EINVAL; 500 501 if (!rtnl_trylock()) 502 return restart_syscall(); 503 504 if (dev_isalive(netdev)) { 505 struct switchdev_attr attr = { 506 .orig_dev = netdev, 507 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID, 508 .flags = SWITCHDEV_F_NO_RECURSE, 509 }; 510 511 ret = switchdev_port_attr_get(netdev, &attr); 512 if (!ret) 513 ret = sprintf(buf, "%*phN\n", attr.u.ppid.id_len, 514 attr.u.ppid.id); 515 } 516 rtnl_unlock(); 517 518 return ret; 519 } 520 static DEVICE_ATTR_RO(phys_switch_id); 521 522 static struct attribute *net_class_attrs[] __ro_after_init = { 523 &dev_attr_netdev_group.attr, 524 &dev_attr_type.attr, 525 &dev_attr_dev_id.attr, 526 &dev_attr_dev_port.attr, 527 &dev_attr_iflink.attr, 528 &dev_attr_ifindex.attr, 529 &dev_attr_name_assign_type.attr, 530 &dev_attr_addr_assign_type.attr, 531 &dev_attr_addr_len.attr, 532 &dev_attr_link_mode.attr, 533 &dev_attr_address.attr, 534 &dev_attr_broadcast.attr, 535 &dev_attr_speed.attr, 536 &dev_attr_duplex.attr, 537 &dev_attr_dormant.attr, 538 &dev_attr_operstate.attr, 539 &dev_attr_carrier_changes.attr, 540 &dev_attr_ifalias.attr, 541 &dev_attr_carrier.attr, 542 &dev_attr_mtu.attr, 543 &dev_attr_flags.attr, 544 &dev_attr_tx_queue_len.attr, 545 &dev_attr_gro_flush_timeout.attr, 546 &dev_attr_phys_port_id.attr, 547 &dev_attr_phys_port_name.attr, 548 &dev_attr_phys_switch_id.attr, 549 &dev_attr_proto_down.attr, 550 NULL, 551 }; 552 ATTRIBUTE_GROUPS(net_class); 553 554 /* Show a given an attribute in the statistics group */ 555 static ssize_t netstat_show(const struct device *d, 556 struct device_attribute *attr, char *buf, 557 unsigned long offset) 558 { 559 struct net_device *dev = to_net_dev(d); 560 ssize_t ret = -EINVAL; 561 562 WARN_ON(offset > sizeof(struct rtnl_link_stats64) || 563 offset % sizeof(u64) != 0); 564 565 read_lock(&dev_base_lock); 566 if (dev_isalive(dev)) { 567 struct rtnl_link_stats64 temp; 568 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); 569 570 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset)); 571 } 572 read_unlock(&dev_base_lock); 573 return ret; 574 } 575 576 /* generate a read-only statistics attribute */ 577 #define NETSTAT_ENTRY(name) \ 578 static ssize_t name##_show(struct device *d, \ 579 struct device_attribute *attr, char *buf) \ 580 { \ 581 return netstat_show(d, attr, buf, \ 582 offsetof(struct rtnl_link_stats64, name)); \ 583 } \ 584 static DEVICE_ATTR_RO(name) 585 586 NETSTAT_ENTRY(rx_packets); 587 NETSTAT_ENTRY(tx_packets); 588 NETSTAT_ENTRY(rx_bytes); 589 NETSTAT_ENTRY(tx_bytes); 590 NETSTAT_ENTRY(rx_errors); 591 NETSTAT_ENTRY(tx_errors); 592 NETSTAT_ENTRY(rx_dropped); 593 NETSTAT_ENTRY(tx_dropped); 594 NETSTAT_ENTRY(multicast); 595 NETSTAT_ENTRY(collisions); 596 NETSTAT_ENTRY(rx_length_errors); 597 NETSTAT_ENTRY(rx_over_errors); 598 NETSTAT_ENTRY(rx_crc_errors); 599 NETSTAT_ENTRY(rx_frame_errors); 600 NETSTAT_ENTRY(rx_fifo_errors); 601 NETSTAT_ENTRY(rx_missed_errors); 602 NETSTAT_ENTRY(tx_aborted_errors); 603 NETSTAT_ENTRY(tx_carrier_errors); 604 NETSTAT_ENTRY(tx_fifo_errors); 605 NETSTAT_ENTRY(tx_heartbeat_errors); 606 NETSTAT_ENTRY(tx_window_errors); 607 NETSTAT_ENTRY(rx_compressed); 608 NETSTAT_ENTRY(tx_compressed); 609 NETSTAT_ENTRY(rx_nohandler); 610 611 static struct attribute *netstat_attrs[] __ro_after_init = { 612 &dev_attr_rx_packets.attr, 613 &dev_attr_tx_packets.attr, 614 &dev_attr_rx_bytes.attr, 615 &dev_attr_tx_bytes.attr, 616 &dev_attr_rx_errors.attr, 617 &dev_attr_tx_errors.attr, 618 &dev_attr_rx_dropped.attr, 619 &dev_attr_tx_dropped.attr, 620 &dev_attr_multicast.attr, 621 &dev_attr_collisions.attr, 622 &dev_attr_rx_length_errors.attr, 623 &dev_attr_rx_over_errors.attr, 624 &dev_attr_rx_crc_errors.attr, 625 &dev_attr_rx_frame_errors.attr, 626 &dev_attr_rx_fifo_errors.attr, 627 &dev_attr_rx_missed_errors.attr, 628 &dev_attr_tx_aborted_errors.attr, 629 &dev_attr_tx_carrier_errors.attr, 630 &dev_attr_tx_fifo_errors.attr, 631 &dev_attr_tx_heartbeat_errors.attr, 632 &dev_attr_tx_window_errors.attr, 633 &dev_attr_rx_compressed.attr, 634 &dev_attr_tx_compressed.attr, 635 &dev_attr_rx_nohandler.attr, 636 NULL 637 }; 638 639 static const struct attribute_group netstat_group = { 640 .name = "statistics", 641 .attrs = netstat_attrs, 642 }; 643 644 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211) 645 static struct attribute *wireless_attrs[] = { 646 NULL 647 }; 648 649 static const struct attribute_group wireless_group = { 650 .name = "wireless", 651 .attrs = wireless_attrs, 652 }; 653 #endif 654 655 #else /* CONFIG_SYSFS */ 656 #define net_class_groups NULL 657 #endif /* CONFIG_SYSFS */ 658 659 #ifdef CONFIG_SYSFS 660 #define to_rx_queue_attr(_attr) \ 661 container_of(_attr, struct rx_queue_attribute, attr) 662 663 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj) 664 665 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr, 666 char *buf) 667 { 668 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr); 669 struct netdev_rx_queue *queue = to_rx_queue(kobj); 670 671 if (!attribute->show) 672 return -EIO; 673 674 return attribute->show(queue, buf); 675 } 676 677 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr, 678 const char *buf, size_t count) 679 { 680 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr); 681 struct netdev_rx_queue *queue = to_rx_queue(kobj); 682 683 if (!attribute->store) 684 return -EIO; 685 686 return attribute->store(queue, buf, count); 687 } 688 689 static const struct sysfs_ops rx_queue_sysfs_ops = { 690 .show = rx_queue_attr_show, 691 .store = rx_queue_attr_store, 692 }; 693 694 #ifdef CONFIG_RPS 695 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf) 696 { 697 struct rps_map *map; 698 cpumask_var_t mask; 699 int i, len; 700 701 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 702 return -ENOMEM; 703 704 rcu_read_lock(); 705 map = rcu_dereference(queue->rps_map); 706 if (map) 707 for (i = 0; i < map->len; i++) 708 cpumask_set_cpu(map->cpus[i], mask); 709 710 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask)); 711 rcu_read_unlock(); 712 free_cpumask_var(mask); 713 714 return len < PAGE_SIZE ? len : -EINVAL; 715 } 716 717 static ssize_t store_rps_map(struct netdev_rx_queue *queue, 718 const char *buf, size_t len) 719 { 720 struct rps_map *old_map, *map; 721 cpumask_var_t mask; 722 int err, cpu, i; 723 static DEFINE_MUTEX(rps_map_mutex); 724 725 if (!capable(CAP_NET_ADMIN)) 726 return -EPERM; 727 728 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 729 return -ENOMEM; 730 731 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits); 732 if (err) { 733 free_cpumask_var(mask); 734 return err; 735 } 736 737 map = kzalloc(max_t(unsigned int, 738 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES), 739 GFP_KERNEL); 740 if (!map) { 741 free_cpumask_var(mask); 742 return -ENOMEM; 743 } 744 745 i = 0; 746 for_each_cpu_and(cpu, mask, cpu_online_mask) 747 map->cpus[i++] = cpu; 748 749 if (i) { 750 map->len = i; 751 } else { 752 kfree(map); 753 map = NULL; 754 } 755 756 mutex_lock(&rps_map_mutex); 757 old_map = rcu_dereference_protected(queue->rps_map, 758 mutex_is_locked(&rps_map_mutex)); 759 rcu_assign_pointer(queue->rps_map, map); 760 761 if (map) 762 static_key_slow_inc(&rps_needed); 763 if (old_map) 764 static_key_slow_dec(&rps_needed); 765 766 mutex_unlock(&rps_map_mutex); 767 768 if (old_map) 769 kfree_rcu(old_map, rcu); 770 771 free_cpumask_var(mask); 772 return len; 773 } 774 775 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue, 776 char *buf) 777 { 778 struct rps_dev_flow_table *flow_table; 779 unsigned long val = 0; 780 781 rcu_read_lock(); 782 flow_table = rcu_dereference(queue->rps_flow_table); 783 if (flow_table) 784 val = (unsigned long)flow_table->mask + 1; 785 rcu_read_unlock(); 786 787 return sprintf(buf, "%lu\n", val); 788 } 789 790 static void rps_dev_flow_table_release(struct rcu_head *rcu) 791 { 792 struct rps_dev_flow_table *table = container_of(rcu, 793 struct rps_dev_flow_table, rcu); 794 vfree(table); 795 } 796 797 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue, 798 const char *buf, size_t len) 799 { 800 unsigned long mask, count; 801 struct rps_dev_flow_table *table, *old_table; 802 static DEFINE_SPINLOCK(rps_dev_flow_lock); 803 int rc; 804 805 if (!capable(CAP_NET_ADMIN)) 806 return -EPERM; 807 808 rc = kstrtoul(buf, 0, &count); 809 if (rc < 0) 810 return rc; 811 812 if (count) { 813 mask = count - 1; 814 /* mask = roundup_pow_of_two(count) - 1; 815 * without overflows... 816 */ 817 while ((mask | (mask >> 1)) != mask) 818 mask |= (mask >> 1); 819 /* On 64 bit arches, must check mask fits in table->mask (u32), 820 * and on 32bit arches, must check 821 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow. 822 */ 823 #if BITS_PER_LONG > 32 824 if (mask > (unsigned long)(u32)mask) 825 return -EINVAL; 826 #else 827 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1)) 828 / sizeof(struct rps_dev_flow)) { 829 /* Enforce a limit to prevent overflow */ 830 return -EINVAL; 831 } 832 #endif 833 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1)); 834 if (!table) 835 return -ENOMEM; 836 837 table->mask = mask; 838 for (count = 0; count <= mask; count++) 839 table->flows[count].cpu = RPS_NO_CPU; 840 } else { 841 table = NULL; 842 } 843 844 spin_lock(&rps_dev_flow_lock); 845 old_table = rcu_dereference_protected(queue->rps_flow_table, 846 lockdep_is_held(&rps_dev_flow_lock)); 847 rcu_assign_pointer(queue->rps_flow_table, table); 848 spin_unlock(&rps_dev_flow_lock); 849 850 if (old_table) 851 call_rcu(&old_table->rcu, rps_dev_flow_table_release); 852 853 return len; 854 } 855 856 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init 857 = __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map); 858 859 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init 860 = __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR, 861 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt); 862 #endif /* CONFIG_RPS */ 863 864 static struct attribute *rx_queue_default_attrs[] __ro_after_init = { 865 #ifdef CONFIG_RPS 866 &rps_cpus_attribute.attr, 867 &rps_dev_flow_table_cnt_attribute.attr, 868 #endif 869 NULL 870 }; 871 872 static void rx_queue_release(struct kobject *kobj) 873 { 874 struct netdev_rx_queue *queue = to_rx_queue(kobj); 875 #ifdef CONFIG_RPS 876 struct rps_map *map; 877 struct rps_dev_flow_table *flow_table; 878 879 map = rcu_dereference_protected(queue->rps_map, 1); 880 if (map) { 881 RCU_INIT_POINTER(queue->rps_map, NULL); 882 kfree_rcu(map, rcu); 883 } 884 885 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1); 886 if (flow_table) { 887 RCU_INIT_POINTER(queue->rps_flow_table, NULL); 888 call_rcu(&flow_table->rcu, rps_dev_flow_table_release); 889 } 890 #endif 891 892 memset(kobj, 0, sizeof(*kobj)); 893 dev_put(queue->dev); 894 } 895 896 static const void *rx_queue_namespace(struct kobject *kobj) 897 { 898 struct netdev_rx_queue *queue = to_rx_queue(kobj); 899 struct device *dev = &queue->dev->dev; 900 const void *ns = NULL; 901 902 if (dev->class && dev->class->ns_type) 903 ns = dev->class->namespace(dev); 904 905 return ns; 906 } 907 908 static struct kobj_type rx_queue_ktype __ro_after_init = { 909 .sysfs_ops = &rx_queue_sysfs_ops, 910 .release = rx_queue_release, 911 .default_attrs = rx_queue_default_attrs, 912 .namespace = rx_queue_namespace 913 }; 914 915 static int rx_queue_add_kobject(struct net_device *dev, int index) 916 { 917 struct netdev_rx_queue *queue = dev->_rx + index; 918 struct kobject *kobj = &queue->kobj; 919 int error = 0; 920 921 kobj->kset = dev->queues_kset; 922 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL, 923 "rx-%u", index); 924 if (error) 925 return error; 926 927 if (dev->sysfs_rx_queue_group) { 928 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group); 929 if (error) { 930 kobject_put(kobj); 931 return error; 932 } 933 } 934 935 kobject_uevent(kobj, KOBJ_ADD); 936 dev_hold(queue->dev); 937 938 return error; 939 } 940 #endif /* CONFIG_SYSFS */ 941 942 int 943 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num) 944 { 945 #ifdef CONFIG_SYSFS 946 int i; 947 int error = 0; 948 949 #ifndef CONFIG_RPS 950 if (!dev->sysfs_rx_queue_group) 951 return 0; 952 #endif 953 for (i = old_num; i < new_num; i++) { 954 error = rx_queue_add_kobject(dev, i); 955 if (error) { 956 new_num = old_num; 957 break; 958 } 959 } 960 961 while (--i >= new_num) { 962 struct kobject *kobj = &dev->_rx[i].kobj; 963 964 if (!atomic_read(&dev_net(dev)->count)) 965 kobj->uevent_suppress = 1; 966 if (dev->sysfs_rx_queue_group) 967 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group); 968 kobject_put(kobj); 969 } 970 971 return error; 972 #else 973 return 0; 974 #endif 975 } 976 977 #ifdef CONFIG_SYSFS 978 /* 979 * netdev_queue sysfs structures and functions. 980 */ 981 struct netdev_queue_attribute { 982 struct attribute attr; 983 ssize_t (*show)(struct netdev_queue *queue, char *buf); 984 ssize_t (*store)(struct netdev_queue *queue, 985 const char *buf, size_t len); 986 }; 987 #define to_netdev_queue_attr(_attr) \ 988 container_of(_attr, struct netdev_queue_attribute, attr) 989 990 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj) 991 992 static ssize_t netdev_queue_attr_show(struct kobject *kobj, 993 struct attribute *attr, char *buf) 994 { 995 const struct netdev_queue_attribute *attribute 996 = to_netdev_queue_attr(attr); 997 struct netdev_queue *queue = to_netdev_queue(kobj); 998 999 if (!attribute->show) 1000 return -EIO; 1001 1002 return attribute->show(queue, buf); 1003 } 1004 1005 static ssize_t netdev_queue_attr_store(struct kobject *kobj, 1006 struct attribute *attr, 1007 const char *buf, size_t count) 1008 { 1009 const struct netdev_queue_attribute *attribute 1010 = to_netdev_queue_attr(attr); 1011 struct netdev_queue *queue = to_netdev_queue(kobj); 1012 1013 if (!attribute->store) 1014 return -EIO; 1015 1016 return attribute->store(queue, buf, count); 1017 } 1018 1019 static const struct sysfs_ops netdev_queue_sysfs_ops = { 1020 .show = netdev_queue_attr_show, 1021 .store = netdev_queue_attr_store, 1022 }; 1023 1024 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf) 1025 { 1026 unsigned long trans_timeout; 1027 1028 spin_lock_irq(&queue->_xmit_lock); 1029 trans_timeout = queue->trans_timeout; 1030 spin_unlock_irq(&queue->_xmit_lock); 1031 1032 return sprintf(buf, "%lu", trans_timeout); 1033 } 1034 1035 static unsigned int get_netdev_queue_index(struct netdev_queue *queue) 1036 { 1037 struct net_device *dev = queue->dev; 1038 unsigned int i; 1039 1040 i = queue - dev->_tx; 1041 BUG_ON(i >= dev->num_tx_queues); 1042 1043 return i; 1044 } 1045 1046 static ssize_t traffic_class_show(struct netdev_queue *queue, 1047 char *buf) 1048 { 1049 struct net_device *dev = queue->dev; 1050 int index = get_netdev_queue_index(queue); 1051 int tc = netdev_txq_to_tc(dev, index); 1052 1053 if (tc < 0) 1054 return -EINVAL; 1055 1056 return sprintf(buf, "%u\n", tc); 1057 } 1058 1059 #ifdef CONFIG_XPS 1060 static ssize_t tx_maxrate_show(struct netdev_queue *queue, 1061 char *buf) 1062 { 1063 return sprintf(buf, "%lu\n", queue->tx_maxrate); 1064 } 1065 1066 static ssize_t tx_maxrate_store(struct netdev_queue *queue, 1067 const char *buf, size_t len) 1068 { 1069 struct net_device *dev = queue->dev; 1070 int err, index = get_netdev_queue_index(queue); 1071 u32 rate = 0; 1072 1073 err = kstrtou32(buf, 10, &rate); 1074 if (err < 0) 1075 return err; 1076 1077 if (!rtnl_trylock()) 1078 return restart_syscall(); 1079 1080 err = -EOPNOTSUPP; 1081 if (dev->netdev_ops->ndo_set_tx_maxrate) 1082 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate); 1083 1084 rtnl_unlock(); 1085 if (!err) { 1086 queue->tx_maxrate = rate; 1087 return len; 1088 } 1089 return err; 1090 } 1091 1092 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init 1093 = __ATTR_RW(tx_maxrate); 1094 #endif 1095 1096 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init 1097 = __ATTR_RO(tx_timeout); 1098 1099 static struct netdev_queue_attribute queue_traffic_class __ro_after_init 1100 = __ATTR_RO(traffic_class); 1101 1102 #ifdef CONFIG_BQL 1103 /* 1104 * Byte queue limits sysfs structures and functions. 1105 */ 1106 static ssize_t bql_show(char *buf, unsigned int value) 1107 { 1108 return sprintf(buf, "%u\n", value); 1109 } 1110 1111 static ssize_t bql_set(const char *buf, const size_t count, 1112 unsigned int *pvalue) 1113 { 1114 unsigned int value; 1115 int err; 1116 1117 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) { 1118 value = DQL_MAX_LIMIT; 1119 } else { 1120 err = kstrtouint(buf, 10, &value); 1121 if (err < 0) 1122 return err; 1123 if (value > DQL_MAX_LIMIT) 1124 return -EINVAL; 1125 } 1126 1127 *pvalue = value; 1128 1129 return count; 1130 } 1131 1132 static ssize_t bql_show_hold_time(struct netdev_queue *queue, 1133 char *buf) 1134 { 1135 struct dql *dql = &queue->dql; 1136 1137 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time)); 1138 } 1139 1140 static ssize_t bql_set_hold_time(struct netdev_queue *queue, 1141 const char *buf, size_t len) 1142 { 1143 struct dql *dql = &queue->dql; 1144 unsigned int value; 1145 int err; 1146 1147 err = kstrtouint(buf, 10, &value); 1148 if (err < 0) 1149 return err; 1150 1151 dql->slack_hold_time = msecs_to_jiffies(value); 1152 1153 return len; 1154 } 1155 1156 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init 1157 = __ATTR(hold_time, S_IRUGO | S_IWUSR, 1158 bql_show_hold_time, bql_set_hold_time); 1159 1160 static ssize_t bql_show_inflight(struct netdev_queue *queue, 1161 char *buf) 1162 { 1163 struct dql *dql = &queue->dql; 1164 1165 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed); 1166 } 1167 1168 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init = 1169 __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL); 1170 1171 #define BQL_ATTR(NAME, FIELD) \ 1172 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \ 1173 char *buf) \ 1174 { \ 1175 return bql_show(buf, queue->dql.FIELD); \ 1176 } \ 1177 \ 1178 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \ 1179 const char *buf, size_t len) \ 1180 { \ 1181 return bql_set(buf, len, &queue->dql.FIELD); \ 1182 } \ 1183 \ 1184 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \ 1185 = __ATTR(NAME, S_IRUGO | S_IWUSR, \ 1186 bql_show_ ## NAME, bql_set_ ## NAME) 1187 1188 BQL_ATTR(limit, limit); 1189 BQL_ATTR(limit_max, max_limit); 1190 BQL_ATTR(limit_min, min_limit); 1191 1192 static struct attribute *dql_attrs[] __ro_after_init = { 1193 &bql_limit_attribute.attr, 1194 &bql_limit_max_attribute.attr, 1195 &bql_limit_min_attribute.attr, 1196 &bql_hold_time_attribute.attr, 1197 &bql_inflight_attribute.attr, 1198 NULL 1199 }; 1200 1201 static const struct attribute_group dql_group = { 1202 .name = "byte_queue_limits", 1203 .attrs = dql_attrs, 1204 }; 1205 #endif /* CONFIG_BQL */ 1206 1207 #ifdef CONFIG_XPS 1208 static ssize_t xps_cpus_show(struct netdev_queue *queue, 1209 char *buf) 1210 { 1211 struct net_device *dev = queue->dev; 1212 int cpu, len, num_tc = 1, tc = 0; 1213 struct xps_dev_maps *dev_maps; 1214 cpumask_var_t mask; 1215 unsigned long index; 1216 1217 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 1218 return -ENOMEM; 1219 1220 index = get_netdev_queue_index(queue); 1221 1222 if (dev->num_tc) { 1223 num_tc = dev->num_tc; 1224 tc = netdev_txq_to_tc(dev, index); 1225 if (tc < 0) 1226 return -EINVAL; 1227 } 1228 1229 rcu_read_lock(); 1230 dev_maps = rcu_dereference(dev->xps_maps); 1231 if (dev_maps) { 1232 for_each_possible_cpu(cpu) { 1233 int i, tci = cpu * num_tc + tc; 1234 struct xps_map *map; 1235 1236 map = rcu_dereference(dev_maps->cpu_map[tci]); 1237 if (!map) 1238 continue; 1239 1240 for (i = map->len; i--;) { 1241 if (map->queues[i] == index) { 1242 cpumask_set_cpu(cpu, mask); 1243 break; 1244 } 1245 } 1246 } 1247 } 1248 rcu_read_unlock(); 1249 1250 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask)); 1251 free_cpumask_var(mask); 1252 return len < PAGE_SIZE ? len : -EINVAL; 1253 } 1254 1255 static ssize_t xps_cpus_store(struct netdev_queue *queue, 1256 const char *buf, size_t len) 1257 { 1258 struct net_device *dev = queue->dev; 1259 unsigned long index; 1260 cpumask_var_t mask; 1261 int err; 1262 1263 if (!capable(CAP_NET_ADMIN)) 1264 return -EPERM; 1265 1266 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 1267 return -ENOMEM; 1268 1269 index = get_netdev_queue_index(queue); 1270 1271 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits); 1272 if (err) { 1273 free_cpumask_var(mask); 1274 return err; 1275 } 1276 1277 err = netif_set_xps_queue(dev, mask, index); 1278 1279 free_cpumask_var(mask); 1280 1281 return err ? : len; 1282 } 1283 1284 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init 1285 = __ATTR_RW(xps_cpus); 1286 #endif /* CONFIG_XPS */ 1287 1288 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = { 1289 &queue_trans_timeout.attr, 1290 &queue_traffic_class.attr, 1291 #ifdef CONFIG_XPS 1292 &xps_cpus_attribute.attr, 1293 &queue_tx_maxrate.attr, 1294 #endif 1295 NULL 1296 }; 1297 1298 static void netdev_queue_release(struct kobject *kobj) 1299 { 1300 struct netdev_queue *queue = to_netdev_queue(kobj); 1301 1302 memset(kobj, 0, sizeof(*kobj)); 1303 dev_put(queue->dev); 1304 } 1305 1306 static const void *netdev_queue_namespace(struct kobject *kobj) 1307 { 1308 struct netdev_queue *queue = to_netdev_queue(kobj); 1309 struct device *dev = &queue->dev->dev; 1310 const void *ns = NULL; 1311 1312 if (dev->class && dev->class->ns_type) 1313 ns = dev->class->namespace(dev); 1314 1315 return ns; 1316 } 1317 1318 static struct kobj_type netdev_queue_ktype __ro_after_init = { 1319 .sysfs_ops = &netdev_queue_sysfs_ops, 1320 .release = netdev_queue_release, 1321 .default_attrs = netdev_queue_default_attrs, 1322 .namespace = netdev_queue_namespace, 1323 }; 1324 1325 static int netdev_queue_add_kobject(struct net_device *dev, int index) 1326 { 1327 struct netdev_queue *queue = dev->_tx + index; 1328 struct kobject *kobj = &queue->kobj; 1329 int error = 0; 1330 1331 kobj->kset = dev->queues_kset; 1332 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL, 1333 "tx-%u", index); 1334 if (error) 1335 return error; 1336 1337 #ifdef CONFIG_BQL 1338 error = sysfs_create_group(kobj, &dql_group); 1339 if (error) { 1340 kobject_put(kobj); 1341 return error; 1342 } 1343 #endif 1344 1345 kobject_uevent(kobj, KOBJ_ADD); 1346 dev_hold(queue->dev); 1347 1348 return 0; 1349 } 1350 #endif /* CONFIG_SYSFS */ 1351 1352 int 1353 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num) 1354 { 1355 #ifdef CONFIG_SYSFS 1356 int i; 1357 int error = 0; 1358 1359 for (i = old_num; i < new_num; i++) { 1360 error = netdev_queue_add_kobject(dev, i); 1361 if (error) { 1362 new_num = old_num; 1363 break; 1364 } 1365 } 1366 1367 while (--i >= new_num) { 1368 struct netdev_queue *queue = dev->_tx + i; 1369 1370 if (!atomic_read(&dev_net(dev)->count)) 1371 queue->kobj.uevent_suppress = 1; 1372 #ifdef CONFIG_BQL 1373 sysfs_remove_group(&queue->kobj, &dql_group); 1374 #endif 1375 kobject_put(&queue->kobj); 1376 } 1377 1378 return error; 1379 #else 1380 return 0; 1381 #endif /* CONFIG_SYSFS */ 1382 } 1383 1384 static int register_queue_kobjects(struct net_device *dev) 1385 { 1386 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0; 1387 1388 #ifdef CONFIG_SYSFS 1389 dev->queues_kset = kset_create_and_add("queues", 1390 NULL, &dev->dev.kobj); 1391 if (!dev->queues_kset) 1392 return -ENOMEM; 1393 real_rx = dev->real_num_rx_queues; 1394 #endif 1395 real_tx = dev->real_num_tx_queues; 1396 1397 error = net_rx_queue_update_kobjects(dev, 0, real_rx); 1398 if (error) 1399 goto error; 1400 rxq = real_rx; 1401 1402 error = netdev_queue_update_kobjects(dev, 0, real_tx); 1403 if (error) 1404 goto error; 1405 txq = real_tx; 1406 1407 return 0; 1408 1409 error: 1410 netdev_queue_update_kobjects(dev, txq, 0); 1411 net_rx_queue_update_kobjects(dev, rxq, 0); 1412 return error; 1413 } 1414 1415 static void remove_queue_kobjects(struct net_device *dev) 1416 { 1417 int real_rx = 0, real_tx = 0; 1418 1419 #ifdef CONFIG_SYSFS 1420 real_rx = dev->real_num_rx_queues; 1421 #endif 1422 real_tx = dev->real_num_tx_queues; 1423 1424 net_rx_queue_update_kobjects(dev, real_rx, 0); 1425 netdev_queue_update_kobjects(dev, real_tx, 0); 1426 #ifdef CONFIG_SYSFS 1427 kset_unregister(dev->queues_kset); 1428 #endif 1429 } 1430 1431 static bool net_current_may_mount(void) 1432 { 1433 struct net *net = current->nsproxy->net_ns; 1434 1435 return ns_capable(net->user_ns, CAP_SYS_ADMIN); 1436 } 1437 1438 static void *net_grab_current_ns(void) 1439 { 1440 struct net *ns = current->nsproxy->net_ns; 1441 #ifdef CONFIG_NET_NS 1442 if (ns) 1443 refcount_inc(&ns->passive); 1444 #endif 1445 return ns; 1446 } 1447 1448 static const void *net_initial_ns(void) 1449 { 1450 return &init_net; 1451 } 1452 1453 static const void *net_netlink_ns(struct sock *sk) 1454 { 1455 return sock_net(sk); 1456 } 1457 1458 const struct kobj_ns_type_operations net_ns_type_operations = { 1459 .type = KOBJ_NS_TYPE_NET, 1460 .current_may_mount = net_current_may_mount, 1461 .grab_current_ns = net_grab_current_ns, 1462 .netlink_ns = net_netlink_ns, 1463 .initial_ns = net_initial_ns, 1464 .drop_ns = net_drop_ns, 1465 }; 1466 EXPORT_SYMBOL_GPL(net_ns_type_operations); 1467 1468 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env) 1469 { 1470 struct net_device *dev = to_net_dev(d); 1471 int retval; 1472 1473 /* pass interface to uevent. */ 1474 retval = add_uevent_var(env, "INTERFACE=%s", dev->name); 1475 if (retval) 1476 goto exit; 1477 1478 /* pass ifindex to uevent. 1479 * ifindex is useful as it won't change (interface name may change) 1480 * and is what RtNetlink uses natively. 1481 */ 1482 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex); 1483 1484 exit: 1485 return retval; 1486 } 1487 1488 /* 1489 * netdev_release -- destroy and free a dead device. 1490 * Called when last reference to device kobject is gone. 1491 */ 1492 static void netdev_release(struct device *d) 1493 { 1494 struct net_device *dev = to_net_dev(d); 1495 1496 BUG_ON(dev->reg_state != NETREG_RELEASED); 1497 1498 /* no need to wait for rcu grace period: 1499 * device is dead and about to be freed. 1500 */ 1501 kfree(rcu_access_pointer(dev->ifalias)); 1502 netdev_freemem(dev); 1503 } 1504 1505 static const void *net_namespace(struct device *d) 1506 { 1507 struct net_device *dev = to_net_dev(d); 1508 1509 return dev_net(dev); 1510 } 1511 1512 static struct class net_class __ro_after_init = { 1513 .name = "net", 1514 .dev_release = netdev_release, 1515 .dev_groups = net_class_groups, 1516 .dev_uevent = netdev_uevent, 1517 .ns_type = &net_ns_type_operations, 1518 .namespace = net_namespace, 1519 }; 1520 1521 #ifdef CONFIG_OF_NET 1522 static int of_dev_node_match(struct device *dev, const void *data) 1523 { 1524 int ret = 0; 1525 1526 if (dev->parent) 1527 ret = dev->parent->of_node == data; 1528 1529 return ret == 0 ? dev->of_node == data : ret; 1530 } 1531 1532 /* 1533 * of_find_net_device_by_node - lookup the net device for the device node 1534 * @np: OF device node 1535 * 1536 * Looks up the net_device structure corresponding with the device node. 1537 * If successful, returns a pointer to the net_device with the embedded 1538 * struct device refcount incremented by one, or NULL on failure. The 1539 * refcount must be dropped when done with the net_device. 1540 */ 1541 struct net_device *of_find_net_device_by_node(struct device_node *np) 1542 { 1543 struct device *dev; 1544 1545 dev = class_find_device(&net_class, NULL, np, of_dev_node_match); 1546 if (!dev) 1547 return NULL; 1548 1549 return to_net_dev(dev); 1550 } 1551 EXPORT_SYMBOL(of_find_net_device_by_node); 1552 #endif 1553 1554 /* Delete sysfs entries but hold kobject reference until after all 1555 * netdev references are gone. 1556 */ 1557 void netdev_unregister_kobject(struct net_device *ndev) 1558 { 1559 struct device *dev = &ndev->dev; 1560 1561 if (!atomic_read(&dev_net(ndev)->count)) 1562 dev_set_uevent_suppress(dev, 1); 1563 1564 kobject_get(&dev->kobj); 1565 1566 remove_queue_kobjects(ndev); 1567 1568 pm_runtime_set_memalloc_noio(dev, false); 1569 1570 device_del(dev); 1571 } 1572 1573 /* Create sysfs entries for network device. */ 1574 int netdev_register_kobject(struct net_device *ndev) 1575 { 1576 struct device *dev = &ndev->dev; 1577 const struct attribute_group **groups = ndev->sysfs_groups; 1578 int error = 0; 1579 1580 device_initialize(dev); 1581 dev->class = &net_class; 1582 dev->platform_data = ndev; 1583 dev->groups = groups; 1584 1585 dev_set_name(dev, "%s", ndev->name); 1586 1587 #ifdef CONFIG_SYSFS 1588 /* Allow for a device specific group */ 1589 if (*groups) 1590 groups++; 1591 1592 *groups++ = &netstat_group; 1593 1594 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211) 1595 if (ndev->ieee80211_ptr) 1596 *groups++ = &wireless_group; 1597 #if IS_ENABLED(CONFIG_WIRELESS_EXT) 1598 else if (ndev->wireless_handlers) 1599 *groups++ = &wireless_group; 1600 #endif 1601 #endif 1602 #endif /* CONFIG_SYSFS */ 1603 1604 error = device_add(dev); 1605 if (error) 1606 return error; 1607 1608 error = register_queue_kobjects(ndev); 1609 if (error) { 1610 device_del(dev); 1611 return error; 1612 } 1613 1614 pm_runtime_set_memalloc_noio(dev, true); 1615 1616 return error; 1617 } 1618 1619 int netdev_class_create_file_ns(const struct class_attribute *class_attr, 1620 const void *ns) 1621 { 1622 return class_create_file_ns(&net_class, class_attr, ns); 1623 } 1624 EXPORT_SYMBOL(netdev_class_create_file_ns); 1625 1626 void netdev_class_remove_file_ns(const struct class_attribute *class_attr, 1627 const void *ns) 1628 { 1629 class_remove_file_ns(&net_class, class_attr, ns); 1630 } 1631 EXPORT_SYMBOL(netdev_class_remove_file_ns); 1632 1633 int __init netdev_kobject_init(void) 1634 { 1635 kobj_ns_type_register(&net_ns_type_operations); 1636 return class_register(&net_class); 1637 } 1638