1 /* 2 * net-sysfs.c - network device class and attributes 3 * 4 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org> 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 */ 11 12 #include <linux/capability.h> 13 #include <linux/kernel.h> 14 #include <linux/netdevice.h> 15 #include <linux/if_arp.h> 16 #include <linux/slab.h> 17 #include <linux/sched/signal.h> 18 #include <linux/nsproxy.h> 19 #include <net/sock.h> 20 #include <net/net_namespace.h> 21 #include <linux/rtnetlink.h> 22 #include <linux/vmalloc.h> 23 #include <linux/export.h> 24 #include <linux/jiffies.h> 25 #include <linux/pm_runtime.h> 26 #include <linux/of.h> 27 #include <linux/of_net.h> 28 #include <linux/cpu.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_up_count) + 299 atomic_read(&netdev->carrier_down_count)); 300 } 301 static DEVICE_ATTR_RO(carrier_changes); 302 303 static ssize_t carrier_up_count_show(struct device *dev, 304 struct device_attribute *attr, 305 char *buf) 306 { 307 struct net_device *netdev = to_net_dev(dev); 308 309 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count)); 310 } 311 static DEVICE_ATTR_RO(carrier_up_count); 312 313 static ssize_t carrier_down_count_show(struct device *dev, 314 struct device_attribute *attr, 315 char *buf) 316 { 317 struct net_device *netdev = to_net_dev(dev); 318 319 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count)); 320 } 321 static DEVICE_ATTR_RO(carrier_down_count); 322 323 /* read-write attributes */ 324 325 static int change_mtu(struct net_device *dev, unsigned long new_mtu) 326 { 327 return dev_set_mtu(dev, (int)new_mtu); 328 } 329 330 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr, 331 const char *buf, size_t len) 332 { 333 return netdev_store(dev, attr, buf, len, change_mtu); 334 } 335 NETDEVICE_SHOW_RW(mtu, fmt_dec); 336 337 static int change_flags(struct net_device *dev, unsigned long new_flags) 338 { 339 return dev_change_flags(dev, (unsigned int)new_flags, NULL); 340 } 341 342 static ssize_t flags_store(struct device *dev, struct device_attribute *attr, 343 const char *buf, size_t len) 344 { 345 return netdev_store(dev, attr, buf, len, change_flags); 346 } 347 NETDEVICE_SHOW_RW(flags, fmt_hex); 348 349 static ssize_t tx_queue_len_store(struct device *dev, 350 struct device_attribute *attr, 351 const char *buf, size_t len) 352 { 353 if (!capable(CAP_NET_ADMIN)) 354 return -EPERM; 355 356 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len); 357 } 358 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec); 359 360 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val) 361 { 362 dev->gro_flush_timeout = val; 363 return 0; 364 } 365 366 static ssize_t gro_flush_timeout_store(struct device *dev, 367 struct device_attribute *attr, 368 const char *buf, size_t len) 369 { 370 if (!capable(CAP_NET_ADMIN)) 371 return -EPERM; 372 373 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout); 374 } 375 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong); 376 377 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr, 378 const char *buf, size_t len) 379 { 380 struct net_device *netdev = to_net_dev(dev); 381 struct net *net = dev_net(netdev); 382 size_t count = len; 383 ssize_t ret = 0; 384 385 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 386 return -EPERM; 387 388 /* ignore trailing newline */ 389 if (len > 0 && buf[len - 1] == '\n') 390 --count; 391 392 if (!rtnl_trylock()) 393 return restart_syscall(); 394 395 if (dev_isalive(netdev)) { 396 ret = dev_set_alias(netdev, buf, count); 397 if (ret < 0) 398 goto err; 399 ret = len; 400 netdev_state_change(netdev); 401 } 402 err: 403 rtnl_unlock(); 404 405 return ret; 406 } 407 408 static ssize_t ifalias_show(struct device *dev, 409 struct device_attribute *attr, char *buf) 410 { 411 const struct net_device *netdev = to_net_dev(dev); 412 char tmp[IFALIASZ]; 413 ssize_t ret = 0; 414 415 ret = dev_get_alias(netdev, tmp, sizeof(tmp)); 416 if (ret > 0) 417 ret = sprintf(buf, "%s\n", tmp); 418 return ret; 419 } 420 static DEVICE_ATTR_RW(ifalias); 421 422 static int change_group(struct net_device *dev, unsigned long new_group) 423 { 424 dev_set_group(dev, (int)new_group); 425 return 0; 426 } 427 428 static ssize_t group_store(struct device *dev, struct device_attribute *attr, 429 const char *buf, size_t len) 430 { 431 return netdev_store(dev, attr, buf, len, change_group); 432 } 433 NETDEVICE_SHOW(group, fmt_dec); 434 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store); 435 436 static int change_proto_down(struct net_device *dev, unsigned long proto_down) 437 { 438 return dev_change_proto_down(dev, (bool)proto_down); 439 } 440 441 static ssize_t proto_down_store(struct device *dev, 442 struct device_attribute *attr, 443 const char *buf, size_t len) 444 { 445 return netdev_store(dev, attr, buf, len, change_proto_down); 446 } 447 NETDEVICE_SHOW_RW(proto_down, fmt_dec); 448 449 static ssize_t phys_port_id_show(struct device *dev, 450 struct device_attribute *attr, char *buf) 451 { 452 struct net_device *netdev = to_net_dev(dev); 453 ssize_t ret = -EINVAL; 454 455 if (!rtnl_trylock()) 456 return restart_syscall(); 457 458 if (dev_isalive(netdev)) { 459 struct netdev_phys_item_id ppid; 460 461 ret = dev_get_phys_port_id(netdev, &ppid); 462 if (!ret) 463 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id); 464 } 465 rtnl_unlock(); 466 467 return ret; 468 } 469 static DEVICE_ATTR_RO(phys_port_id); 470 471 static ssize_t phys_port_name_show(struct device *dev, 472 struct device_attribute *attr, char *buf) 473 { 474 struct net_device *netdev = to_net_dev(dev); 475 ssize_t ret = -EINVAL; 476 477 if (!rtnl_trylock()) 478 return restart_syscall(); 479 480 if (dev_isalive(netdev)) { 481 char name[IFNAMSIZ]; 482 483 ret = dev_get_phys_port_name(netdev, name, sizeof(name)); 484 if (!ret) 485 ret = sprintf(buf, "%s\n", name); 486 } 487 rtnl_unlock(); 488 489 return ret; 490 } 491 static DEVICE_ATTR_RO(phys_port_name); 492 493 static ssize_t phys_switch_id_show(struct device *dev, 494 struct device_attribute *attr, char *buf) 495 { 496 struct net_device *netdev = to_net_dev(dev); 497 ssize_t ret = -EINVAL; 498 499 if (!rtnl_trylock()) 500 return restart_syscall(); 501 502 if (dev_isalive(netdev)) { 503 struct netdev_phys_item_id ppid = { }; 504 505 ret = dev_get_port_parent_id(netdev, &ppid, false); 506 if (!ret) 507 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id); 508 } 509 rtnl_unlock(); 510 511 return ret; 512 } 513 static DEVICE_ATTR_RO(phys_switch_id); 514 515 static struct attribute *net_class_attrs[] __ro_after_init = { 516 &dev_attr_netdev_group.attr, 517 &dev_attr_type.attr, 518 &dev_attr_dev_id.attr, 519 &dev_attr_dev_port.attr, 520 &dev_attr_iflink.attr, 521 &dev_attr_ifindex.attr, 522 &dev_attr_name_assign_type.attr, 523 &dev_attr_addr_assign_type.attr, 524 &dev_attr_addr_len.attr, 525 &dev_attr_link_mode.attr, 526 &dev_attr_address.attr, 527 &dev_attr_broadcast.attr, 528 &dev_attr_speed.attr, 529 &dev_attr_duplex.attr, 530 &dev_attr_dormant.attr, 531 &dev_attr_operstate.attr, 532 &dev_attr_carrier_changes.attr, 533 &dev_attr_ifalias.attr, 534 &dev_attr_carrier.attr, 535 &dev_attr_mtu.attr, 536 &dev_attr_flags.attr, 537 &dev_attr_tx_queue_len.attr, 538 &dev_attr_gro_flush_timeout.attr, 539 &dev_attr_phys_port_id.attr, 540 &dev_attr_phys_port_name.attr, 541 &dev_attr_phys_switch_id.attr, 542 &dev_attr_proto_down.attr, 543 &dev_attr_carrier_up_count.attr, 544 &dev_attr_carrier_down_count.attr, 545 NULL, 546 }; 547 ATTRIBUTE_GROUPS(net_class); 548 549 /* Show a given an attribute in the statistics group */ 550 static ssize_t netstat_show(const struct device *d, 551 struct device_attribute *attr, char *buf, 552 unsigned long offset) 553 { 554 struct net_device *dev = to_net_dev(d); 555 ssize_t ret = -EINVAL; 556 557 WARN_ON(offset > sizeof(struct rtnl_link_stats64) || 558 offset % sizeof(u64) != 0); 559 560 read_lock(&dev_base_lock); 561 if (dev_isalive(dev)) { 562 struct rtnl_link_stats64 temp; 563 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); 564 565 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset)); 566 } 567 read_unlock(&dev_base_lock); 568 return ret; 569 } 570 571 /* generate a read-only statistics attribute */ 572 #define NETSTAT_ENTRY(name) \ 573 static ssize_t name##_show(struct device *d, \ 574 struct device_attribute *attr, char *buf) \ 575 { \ 576 return netstat_show(d, attr, buf, \ 577 offsetof(struct rtnl_link_stats64, name)); \ 578 } \ 579 static DEVICE_ATTR_RO(name) 580 581 NETSTAT_ENTRY(rx_packets); 582 NETSTAT_ENTRY(tx_packets); 583 NETSTAT_ENTRY(rx_bytes); 584 NETSTAT_ENTRY(tx_bytes); 585 NETSTAT_ENTRY(rx_errors); 586 NETSTAT_ENTRY(tx_errors); 587 NETSTAT_ENTRY(rx_dropped); 588 NETSTAT_ENTRY(tx_dropped); 589 NETSTAT_ENTRY(multicast); 590 NETSTAT_ENTRY(collisions); 591 NETSTAT_ENTRY(rx_length_errors); 592 NETSTAT_ENTRY(rx_over_errors); 593 NETSTAT_ENTRY(rx_crc_errors); 594 NETSTAT_ENTRY(rx_frame_errors); 595 NETSTAT_ENTRY(rx_fifo_errors); 596 NETSTAT_ENTRY(rx_missed_errors); 597 NETSTAT_ENTRY(tx_aborted_errors); 598 NETSTAT_ENTRY(tx_carrier_errors); 599 NETSTAT_ENTRY(tx_fifo_errors); 600 NETSTAT_ENTRY(tx_heartbeat_errors); 601 NETSTAT_ENTRY(tx_window_errors); 602 NETSTAT_ENTRY(rx_compressed); 603 NETSTAT_ENTRY(tx_compressed); 604 NETSTAT_ENTRY(rx_nohandler); 605 606 static struct attribute *netstat_attrs[] __ro_after_init = { 607 &dev_attr_rx_packets.attr, 608 &dev_attr_tx_packets.attr, 609 &dev_attr_rx_bytes.attr, 610 &dev_attr_tx_bytes.attr, 611 &dev_attr_rx_errors.attr, 612 &dev_attr_tx_errors.attr, 613 &dev_attr_rx_dropped.attr, 614 &dev_attr_tx_dropped.attr, 615 &dev_attr_multicast.attr, 616 &dev_attr_collisions.attr, 617 &dev_attr_rx_length_errors.attr, 618 &dev_attr_rx_over_errors.attr, 619 &dev_attr_rx_crc_errors.attr, 620 &dev_attr_rx_frame_errors.attr, 621 &dev_attr_rx_fifo_errors.attr, 622 &dev_attr_rx_missed_errors.attr, 623 &dev_attr_tx_aborted_errors.attr, 624 &dev_attr_tx_carrier_errors.attr, 625 &dev_attr_tx_fifo_errors.attr, 626 &dev_attr_tx_heartbeat_errors.attr, 627 &dev_attr_tx_window_errors.attr, 628 &dev_attr_rx_compressed.attr, 629 &dev_attr_tx_compressed.attr, 630 &dev_attr_rx_nohandler.attr, 631 NULL 632 }; 633 634 static const struct attribute_group netstat_group = { 635 .name = "statistics", 636 .attrs = netstat_attrs, 637 }; 638 639 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211) 640 static struct attribute *wireless_attrs[] = { 641 NULL 642 }; 643 644 static const struct attribute_group wireless_group = { 645 .name = "wireless", 646 .attrs = wireless_attrs, 647 }; 648 #endif 649 650 #else /* CONFIG_SYSFS */ 651 #define net_class_groups NULL 652 #endif /* CONFIG_SYSFS */ 653 654 #ifdef CONFIG_SYSFS 655 #define to_rx_queue_attr(_attr) \ 656 container_of(_attr, struct rx_queue_attribute, attr) 657 658 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj) 659 660 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr, 661 char *buf) 662 { 663 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr); 664 struct netdev_rx_queue *queue = to_rx_queue(kobj); 665 666 if (!attribute->show) 667 return -EIO; 668 669 return attribute->show(queue, buf); 670 } 671 672 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr, 673 const char *buf, size_t count) 674 { 675 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr); 676 struct netdev_rx_queue *queue = to_rx_queue(kobj); 677 678 if (!attribute->store) 679 return -EIO; 680 681 return attribute->store(queue, buf, count); 682 } 683 684 static const struct sysfs_ops rx_queue_sysfs_ops = { 685 .show = rx_queue_attr_show, 686 .store = rx_queue_attr_store, 687 }; 688 689 #ifdef CONFIG_RPS 690 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf) 691 { 692 struct rps_map *map; 693 cpumask_var_t mask; 694 int i, len; 695 696 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 697 return -ENOMEM; 698 699 rcu_read_lock(); 700 map = rcu_dereference(queue->rps_map); 701 if (map) 702 for (i = 0; i < map->len; i++) 703 cpumask_set_cpu(map->cpus[i], mask); 704 705 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask)); 706 rcu_read_unlock(); 707 free_cpumask_var(mask); 708 709 return len < PAGE_SIZE ? len : -EINVAL; 710 } 711 712 static ssize_t store_rps_map(struct netdev_rx_queue *queue, 713 const char *buf, size_t len) 714 { 715 struct rps_map *old_map, *map; 716 cpumask_var_t mask; 717 int err, cpu, i; 718 static DEFINE_MUTEX(rps_map_mutex); 719 720 if (!capable(CAP_NET_ADMIN)) 721 return -EPERM; 722 723 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 724 return -ENOMEM; 725 726 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits); 727 if (err) { 728 free_cpumask_var(mask); 729 return err; 730 } 731 732 map = kzalloc(max_t(unsigned int, 733 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES), 734 GFP_KERNEL); 735 if (!map) { 736 free_cpumask_var(mask); 737 return -ENOMEM; 738 } 739 740 i = 0; 741 for_each_cpu_and(cpu, mask, cpu_online_mask) 742 map->cpus[i++] = cpu; 743 744 if (i) { 745 map->len = i; 746 } else { 747 kfree(map); 748 map = NULL; 749 } 750 751 mutex_lock(&rps_map_mutex); 752 old_map = rcu_dereference_protected(queue->rps_map, 753 mutex_is_locked(&rps_map_mutex)); 754 rcu_assign_pointer(queue->rps_map, map); 755 756 if (map) 757 static_key_slow_inc(&rps_needed); 758 if (old_map) 759 static_key_slow_dec(&rps_needed); 760 761 mutex_unlock(&rps_map_mutex); 762 763 if (old_map) 764 kfree_rcu(old_map, rcu); 765 766 free_cpumask_var(mask); 767 return len; 768 } 769 770 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue, 771 char *buf) 772 { 773 struct rps_dev_flow_table *flow_table; 774 unsigned long val = 0; 775 776 rcu_read_lock(); 777 flow_table = rcu_dereference(queue->rps_flow_table); 778 if (flow_table) 779 val = (unsigned long)flow_table->mask + 1; 780 rcu_read_unlock(); 781 782 return sprintf(buf, "%lu\n", val); 783 } 784 785 static void rps_dev_flow_table_release(struct rcu_head *rcu) 786 { 787 struct rps_dev_flow_table *table = container_of(rcu, 788 struct rps_dev_flow_table, rcu); 789 vfree(table); 790 } 791 792 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue, 793 const char *buf, size_t len) 794 { 795 unsigned long mask, count; 796 struct rps_dev_flow_table *table, *old_table; 797 static DEFINE_SPINLOCK(rps_dev_flow_lock); 798 int rc; 799 800 if (!capable(CAP_NET_ADMIN)) 801 return -EPERM; 802 803 rc = kstrtoul(buf, 0, &count); 804 if (rc < 0) 805 return rc; 806 807 if (count) { 808 mask = count - 1; 809 /* mask = roundup_pow_of_two(count) - 1; 810 * without overflows... 811 */ 812 while ((mask | (mask >> 1)) != mask) 813 mask |= (mask >> 1); 814 /* On 64 bit arches, must check mask fits in table->mask (u32), 815 * and on 32bit arches, must check 816 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow. 817 */ 818 #if BITS_PER_LONG > 32 819 if (mask > (unsigned long)(u32)mask) 820 return -EINVAL; 821 #else 822 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1)) 823 / sizeof(struct rps_dev_flow)) { 824 /* Enforce a limit to prevent overflow */ 825 return -EINVAL; 826 } 827 #endif 828 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1)); 829 if (!table) 830 return -ENOMEM; 831 832 table->mask = mask; 833 for (count = 0; count <= mask; count++) 834 table->flows[count].cpu = RPS_NO_CPU; 835 } else { 836 table = NULL; 837 } 838 839 spin_lock(&rps_dev_flow_lock); 840 old_table = rcu_dereference_protected(queue->rps_flow_table, 841 lockdep_is_held(&rps_dev_flow_lock)); 842 rcu_assign_pointer(queue->rps_flow_table, table); 843 spin_unlock(&rps_dev_flow_lock); 844 845 if (old_table) 846 call_rcu(&old_table->rcu, rps_dev_flow_table_release); 847 848 return len; 849 } 850 851 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init 852 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map); 853 854 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init 855 = __ATTR(rps_flow_cnt, 0644, 856 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt); 857 #endif /* CONFIG_RPS */ 858 859 static struct attribute *rx_queue_default_attrs[] __ro_after_init = { 860 #ifdef CONFIG_RPS 861 &rps_cpus_attribute.attr, 862 &rps_dev_flow_table_cnt_attribute.attr, 863 #endif 864 NULL 865 }; 866 867 static void rx_queue_release(struct kobject *kobj) 868 { 869 struct netdev_rx_queue *queue = to_rx_queue(kobj); 870 #ifdef CONFIG_RPS 871 struct rps_map *map; 872 struct rps_dev_flow_table *flow_table; 873 874 map = rcu_dereference_protected(queue->rps_map, 1); 875 if (map) { 876 RCU_INIT_POINTER(queue->rps_map, NULL); 877 kfree_rcu(map, rcu); 878 } 879 880 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1); 881 if (flow_table) { 882 RCU_INIT_POINTER(queue->rps_flow_table, NULL); 883 call_rcu(&flow_table->rcu, rps_dev_flow_table_release); 884 } 885 #endif 886 887 memset(kobj, 0, sizeof(*kobj)); 888 dev_put(queue->dev); 889 } 890 891 static const void *rx_queue_namespace(struct kobject *kobj) 892 { 893 struct netdev_rx_queue *queue = to_rx_queue(kobj); 894 struct device *dev = &queue->dev->dev; 895 const void *ns = NULL; 896 897 if (dev->class && dev->class->ns_type) 898 ns = dev->class->namespace(dev); 899 900 return ns; 901 } 902 903 static void rx_queue_get_ownership(struct kobject *kobj, 904 kuid_t *uid, kgid_t *gid) 905 { 906 const struct net *net = rx_queue_namespace(kobj); 907 908 net_ns_get_ownership(net, uid, gid); 909 } 910 911 static struct kobj_type rx_queue_ktype __ro_after_init = { 912 .sysfs_ops = &rx_queue_sysfs_ops, 913 .release = rx_queue_release, 914 .default_attrs = rx_queue_default_attrs, 915 .namespace = rx_queue_namespace, 916 .get_ownership = rx_queue_get_ownership, 917 }; 918 919 static int rx_queue_add_kobject(struct net_device *dev, int index) 920 { 921 struct netdev_rx_queue *queue = dev->_rx + index; 922 struct kobject *kobj = &queue->kobj; 923 int error = 0; 924 925 kobj->kset = dev->queues_kset; 926 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL, 927 "rx-%u", index); 928 if (error) 929 return error; 930 931 if (dev->sysfs_rx_queue_group) { 932 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group); 933 if (error) { 934 kobject_put(kobj); 935 return error; 936 } 937 } 938 939 kobject_uevent(kobj, KOBJ_ADD); 940 dev_hold(queue->dev); 941 942 return error; 943 } 944 #endif /* CONFIG_SYSFS */ 945 946 int 947 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num) 948 { 949 #ifdef CONFIG_SYSFS 950 int i; 951 int error = 0; 952 953 #ifndef CONFIG_RPS 954 if (!dev->sysfs_rx_queue_group) 955 return 0; 956 #endif 957 for (i = old_num; i < new_num; i++) { 958 error = rx_queue_add_kobject(dev, i); 959 if (error) { 960 new_num = old_num; 961 break; 962 } 963 } 964 965 while (--i >= new_num) { 966 struct kobject *kobj = &dev->_rx[i].kobj; 967 968 if (!refcount_read(&dev_net(dev)->count)) 969 kobj->uevent_suppress = 1; 970 if (dev->sysfs_rx_queue_group) 971 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group); 972 kobject_put(kobj); 973 } 974 975 return error; 976 #else 977 return 0; 978 #endif 979 } 980 981 #ifdef CONFIG_SYSFS 982 /* 983 * netdev_queue sysfs structures and functions. 984 */ 985 struct netdev_queue_attribute { 986 struct attribute attr; 987 ssize_t (*show)(struct netdev_queue *queue, char *buf); 988 ssize_t (*store)(struct netdev_queue *queue, 989 const char *buf, size_t len); 990 }; 991 #define to_netdev_queue_attr(_attr) \ 992 container_of(_attr, struct netdev_queue_attribute, attr) 993 994 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj) 995 996 static ssize_t netdev_queue_attr_show(struct kobject *kobj, 997 struct attribute *attr, char *buf) 998 { 999 const struct netdev_queue_attribute *attribute 1000 = to_netdev_queue_attr(attr); 1001 struct netdev_queue *queue = to_netdev_queue(kobj); 1002 1003 if (!attribute->show) 1004 return -EIO; 1005 1006 return attribute->show(queue, buf); 1007 } 1008 1009 static ssize_t netdev_queue_attr_store(struct kobject *kobj, 1010 struct attribute *attr, 1011 const char *buf, size_t count) 1012 { 1013 const struct netdev_queue_attribute *attribute 1014 = to_netdev_queue_attr(attr); 1015 struct netdev_queue *queue = to_netdev_queue(kobj); 1016 1017 if (!attribute->store) 1018 return -EIO; 1019 1020 return attribute->store(queue, buf, count); 1021 } 1022 1023 static const struct sysfs_ops netdev_queue_sysfs_ops = { 1024 .show = netdev_queue_attr_show, 1025 .store = netdev_queue_attr_store, 1026 }; 1027 1028 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf) 1029 { 1030 unsigned long trans_timeout; 1031 1032 spin_lock_irq(&queue->_xmit_lock); 1033 trans_timeout = queue->trans_timeout; 1034 spin_unlock_irq(&queue->_xmit_lock); 1035 1036 return sprintf(buf, "%lu", trans_timeout); 1037 } 1038 1039 static unsigned int get_netdev_queue_index(struct netdev_queue *queue) 1040 { 1041 struct net_device *dev = queue->dev; 1042 unsigned int i; 1043 1044 i = queue - dev->_tx; 1045 BUG_ON(i >= dev->num_tx_queues); 1046 1047 return i; 1048 } 1049 1050 static ssize_t traffic_class_show(struct netdev_queue *queue, 1051 char *buf) 1052 { 1053 struct net_device *dev = queue->dev; 1054 int index; 1055 int tc; 1056 1057 if (!netif_is_multiqueue(dev)) 1058 return -ENOENT; 1059 1060 index = get_netdev_queue_index(queue); 1061 1062 /* If queue belongs to subordinate dev use its TC mapping */ 1063 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 1064 1065 tc = netdev_txq_to_tc(dev, index); 1066 if (tc < 0) 1067 return -EINVAL; 1068 1069 /* We can report the traffic class one of two ways: 1070 * Subordinate device traffic classes are reported with the traffic 1071 * class first, and then the subordinate class so for example TC0 on 1072 * subordinate device 2 will be reported as "0-2". If the queue 1073 * belongs to the root device it will be reported with just the 1074 * traffic class, so just "0" for TC 0 for example. 1075 */ 1076 return dev->num_tc < 0 ? sprintf(buf, "%u%d\n", tc, dev->num_tc) : 1077 sprintf(buf, "%u\n", tc); 1078 } 1079 1080 #ifdef CONFIG_XPS 1081 static ssize_t tx_maxrate_show(struct netdev_queue *queue, 1082 char *buf) 1083 { 1084 return sprintf(buf, "%lu\n", queue->tx_maxrate); 1085 } 1086 1087 static ssize_t tx_maxrate_store(struct netdev_queue *queue, 1088 const char *buf, size_t len) 1089 { 1090 struct net_device *dev = queue->dev; 1091 int err, index = get_netdev_queue_index(queue); 1092 u32 rate = 0; 1093 1094 if (!capable(CAP_NET_ADMIN)) 1095 return -EPERM; 1096 1097 err = kstrtou32(buf, 10, &rate); 1098 if (err < 0) 1099 return err; 1100 1101 if (!rtnl_trylock()) 1102 return restart_syscall(); 1103 1104 err = -EOPNOTSUPP; 1105 if (dev->netdev_ops->ndo_set_tx_maxrate) 1106 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate); 1107 1108 rtnl_unlock(); 1109 if (!err) { 1110 queue->tx_maxrate = rate; 1111 return len; 1112 } 1113 return err; 1114 } 1115 1116 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init 1117 = __ATTR_RW(tx_maxrate); 1118 #endif 1119 1120 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init 1121 = __ATTR_RO(tx_timeout); 1122 1123 static struct netdev_queue_attribute queue_traffic_class __ro_after_init 1124 = __ATTR_RO(traffic_class); 1125 1126 #ifdef CONFIG_BQL 1127 /* 1128 * Byte queue limits sysfs structures and functions. 1129 */ 1130 static ssize_t bql_show(char *buf, unsigned int value) 1131 { 1132 return sprintf(buf, "%u\n", value); 1133 } 1134 1135 static ssize_t bql_set(const char *buf, const size_t count, 1136 unsigned int *pvalue) 1137 { 1138 unsigned int value; 1139 int err; 1140 1141 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) { 1142 value = DQL_MAX_LIMIT; 1143 } else { 1144 err = kstrtouint(buf, 10, &value); 1145 if (err < 0) 1146 return err; 1147 if (value > DQL_MAX_LIMIT) 1148 return -EINVAL; 1149 } 1150 1151 *pvalue = value; 1152 1153 return count; 1154 } 1155 1156 static ssize_t bql_show_hold_time(struct netdev_queue *queue, 1157 char *buf) 1158 { 1159 struct dql *dql = &queue->dql; 1160 1161 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time)); 1162 } 1163 1164 static ssize_t bql_set_hold_time(struct netdev_queue *queue, 1165 const char *buf, size_t len) 1166 { 1167 struct dql *dql = &queue->dql; 1168 unsigned int value; 1169 int err; 1170 1171 err = kstrtouint(buf, 10, &value); 1172 if (err < 0) 1173 return err; 1174 1175 dql->slack_hold_time = msecs_to_jiffies(value); 1176 1177 return len; 1178 } 1179 1180 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init 1181 = __ATTR(hold_time, 0644, 1182 bql_show_hold_time, bql_set_hold_time); 1183 1184 static ssize_t bql_show_inflight(struct netdev_queue *queue, 1185 char *buf) 1186 { 1187 struct dql *dql = &queue->dql; 1188 1189 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed); 1190 } 1191 1192 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init = 1193 __ATTR(inflight, 0444, bql_show_inflight, NULL); 1194 1195 #define BQL_ATTR(NAME, FIELD) \ 1196 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \ 1197 char *buf) \ 1198 { \ 1199 return bql_show(buf, queue->dql.FIELD); \ 1200 } \ 1201 \ 1202 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \ 1203 const char *buf, size_t len) \ 1204 { \ 1205 return bql_set(buf, len, &queue->dql.FIELD); \ 1206 } \ 1207 \ 1208 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \ 1209 = __ATTR(NAME, 0644, \ 1210 bql_show_ ## NAME, bql_set_ ## NAME) 1211 1212 BQL_ATTR(limit, limit); 1213 BQL_ATTR(limit_max, max_limit); 1214 BQL_ATTR(limit_min, min_limit); 1215 1216 static struct attribute *dql_attrs[] __ro_after_init = { 1217 &bql_limit_attribute.attr, 1218 &bql_limit_max_attribute.attr, 1219 &bql_limit_min_attribute.attr, 1220 &bql_hold_time_attribute.attr, 1221 &bql_inflight_attribute.attr, 1222 NULL 1223 }; 1224 1225 static const struct attribute_group dql_group = { 1226 .name = "byte_queue_limits", 1227 .attrs = dql_attrs, 1228 }; 1229 #endif /* CONFIG_BQL */ 1230 1231 #ifdef CONFIG_XPS 1232 static ssize_t xps_cpus_show(struct netdev_queue *queue, 1233 char *buf) 1234 { 1235 struct net_device *dev = queue->dev; 1236 int cpu, len, num_tc = 1, tc = 0; 1237 struct xps_dev_maps *dev_maps; 1238 cpumask_var_t mask; 1239 unsigned long index; 1240 1241 if (!netif_is_multiqueue(dev)) 1242 return -ENOENT; 1243 1244 index = get_netdev_queue_index(queue); 1245 1246 if (dev->num_tc) { 1247 /* Do not allow XPS on subordinate device directly */ 1248 num_tc = dev->num_tc; 1249 if (num_tc < 0) 1250 return -EINVAL; 1251 1252 /* If queue belongs to subordinate dev use its map */ 1253 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 1254 1255 tc = netdev_txq_to_tc(dev, index); 1256 if (tc < 0) 1257 return -EINVAL; 1258 } 1259 1260 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 1261 return -ENOMEM; 1262 1263 rcu_read_lock(); 1264 dev_maps = rcu_dereference(dev->xps_cpus_map); 1265 if (dev_maps) { 1266 for_each_possible_cpu(cpu) { 1267 int i, tci = cpu * num_tc + tc; 1268 struct xps_map *map; 1269 1270 map = rcu_dereference(dev_maps->attr_map[tci]); 1271 if (!map) 1272 continue; 1273 1274 for (i = map->len; i--;) { 1275 if (map->queues[i] == index) { 1276 cpumask_set_cpu(cpu, mask); 1277 break; 1278 } 1279 } 1280 } 1281 } 1282 rcu_read_unlock(); 1283 1284 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask)); 1285 free_cpumask_var(mask); 1286 return len < PAGE_SIZE ? len : -EINVAL; 1287 } 1288 1289 static ssize_t xps_cpus_store(struct netdev_queue *queue, 1290 const char *buf, size_t len) 1291 { 1292 struct net_device *dev = queue->dev; 1293 unsigned long index; 1294 cpumask_var_t mask; 1295 int err; 1296 1297 if (!netif_is_multiqueue(dev)) 1298 return -ENOENT; 1299 1300 if (!capable(CAP_NET_ADMIN)) 1301 return -EPERM; 1302 1303 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 1304 return -ENOMEM; 1305 1306 index = get_netdev_queue_index(queue); 1307 1308 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits); 1309 if (err) { 1310 free_cpumask_var(mask); 1311 return err; 1312 } 1313 1314 err = netif_set_xps_queue(dev, mask, index); 1315 1316 free_cpumask_var(mask); 1317 1318 return err ? : len; 1319 } 1320 1321 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init 1322 = __ATTR_RW(xps_cpus); 1323 1324 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf) 1325 { 1326 struct net_device *dev = queue->dev; 1327 struct xps_dev_maps *dev_maps; 1328 unsigned long *mask, index; 1329 int j, len, num_tc = 1, tc = 0; 1330 1331 index = get_netdev_queue_index(queue); 1332 1333 if (dev->num_tc) { 1334 num_tc = dev->num_tc; 1335 tc = netdev_txq_to_tc(dev, index); 1336 if (tc < 0) 1337 return -EINVAL; 1338 } 1339 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL); 1340 if (!mask) 1341 return -ENOMEM; 1342 1343 rcu_read_lock(); 1344 dev_maps = rcu_dereference(dev->xps_rxqs_map); 1345 if (!dev_maps) 1346 goto out_no_maps; 1347 1348 for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues), 1349 j < dev->num_rx_queues;) { 1350 int i, tci = j * num_tc + tc; 1351 struct xps_map *map; 1352 1353 map = rcu_dereference(dev_maps->attr_map[tci]); 1354 if (!map) 1355 continue; 1356 1357 for (i = map->len; i--;) { 1358 if (map->queues[i] == index) { 1359 set_bit(j, mask); 1360 break; 1361 } 1362 } 1363 } 1364 out_no_maps: 1365 rcu_read_unlock(); 1366 1367 len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues); 1368 bitmap_free(mask); 1369 1370 return len < PAGE_SIZE ? len : -EINVAL; 1371 } 1372 1373 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf, 1374 size_t len) 1375 { 1376 struct net_device *dev = queue->dev; 1377 struct net *net = dev_net(dev); 1378 unsigned long *mask, index; 1379 int err; 1380 1381 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 1382 return -EPERM; 1383 1384 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL); 1385 if (!mask) 1386 return -ENOMEM; 1387 1388 index = get_netdev_queue_index(queue); 1389 1390 err = bitmap_parse(buf, len, mask, dev->num_rx_queues); 1391 if (err) { 1392 bitmap_free(mask); 1393 return err; 1394 } 1395 1396 cpus_read_lock(); 1397 err = __netif_set_xps_queue(dev, mask, index, true); 1398 cpus_read_unlock(); 1399 1400 bitmap_free(mask); 1401 return err ? : len; 1402 } 1403 1404 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init 1405 = __ATTR_RW(xps_rxqs); 1406 #endif /* CONFIG_XPS */ 1407 1408 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = { 1409 &queue_trans_timeout.attr, 1410 &queue_traffic_class.attr, 1411 #ifdef CONFIG_XPS 1412 &xps_cpus_attribute.attr, 1413 &xps_rxqs_attribute.attr, 1414 &queue_tx_maxrate.attr, 1415 #endif 1416 NULL 1417 }; 1418 1419 static void netdev_queue_release(struct kobject *kobj) 1420 { 1421 struct netdev_queue *queue = to_netdev_queue(kobj); 1422 1423 memset(kobj, 0, sizeof(*kobj)); 1424 dev_put(queue->dev); 1425 } 1426 1427 static const void *netdev_queue_namespace(struct kobject *kobj) 1428 { 1429 struct netdev_queue *queue = to_netdev_queue(kobj); 1430 struct device *dev = &queue->dev->dev; 1431 const void *ns = NULL; 1432 1433 if (dev->class && dev->class->ns_type) 1434 ns = dev->class->namespace(dev); 1435 1436 return ns; 1437 } 1438 1439 static void netdev_queue_get_ownership(struct kobject *kobj, 1440 kuid_t *uid, kgid_t *gid) 1441 { 1442 const struct net *net = netdev_queue_namespace(kobj); 1443 1444 net_ns_get_ownership(net, uid, gid); 1445 } 1446 1447 static struct kobj_type netdev_queue_ktype __ro_after_init = { 1448 .sysfs_ops = &netdev_queue_sysfs_ops, 1449 .release = netdev_queue_release, 1450 .default_attrs = netdev_queue_default_attrs, 1451 .namespace = netdev_queue_namespace, 1452 .get_ownership = netdev_queue_get_ownership, 1453 }; 1454 1455 static int netdev_queue_add_kobject(struct net_device *dev, int index) 1456 { 1457 struct netdev_queue *queue = dev->_tx + index; 1458 struct kobject *kobj = &queue->kobj; 1459 int error = 0; 1460 1461 kobj->kset = dev->queues_kset; 1462 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL, 1463 "tx-%u", index); 1464 if (error) 1465 return error; 1466 1467 #ifdef CONFIG_BQL 1468 error = sysfs_create_group(kobj, &dql_group); 1469 if (error) { 1470 kobject_put(kobj); 1471 return error; 1472 } 1473 #endif 1474 1475 kobject_uevent(kobj, KOBJ_ADD); 1476 dev_hold(queue->dev); 1477 1478 return 0; 1479 } 1480 #endif /* CONFIG_SYSFS */ 1481 1482 int 1483 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num) 1484 { 1485 #ifdef CONFIG_SYSFS 1486 int i; 1487 int error = 0; 1488 1489 for (i = old_num; i < new_num; i++) { 1490 error = netdev_queue_add_kobject(dev, i); 1491 if (error) { 1492 new_num = old_num; 1493 break; 1494 } 1495 } 1496 1497 while (--i >= new_num) { 1498 struct netdev_queue *queue = dev->_tx + i; 1499 1500 if (!refcount_read(&dev_net(dev)->count)) 1501 queue->kobj.uevent_suppress = 1; 1502 #ifdef CONFIG_BQL 1503 sysfs_remove_group(&queue->kobj, &dql_group); 1504 #endif 1505 kobject_put(&queue->kobj); 1506 } 1507 1508 return error; 1509 #else 1510 return 0; 1511 #endif /* CONFIG_SYSFS */ 1512 } 1513 1514 static int register_queue_kobjects(struct net_device *dev) 1515 { 1516 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0; 1517 1518 #ifdef CONFIG_SYSFS 1519 dev->queues_kset = kset_create_and_add("queues", 1520 NULL, &dev->dev.kobj); 1521 if (!dev->queues_kset) 1522 return -ENOMEM; 1523 real_rx = dev->real_num_rx_queues; 1524 #endif 1525 real_tx = dev->real_num_tx_queues; 1526 1527 error = net_rx_queue_update_kobjects(dev, 0, real_rx); 1528 if (error) 1529 goto error; 1530 rxq = real_rx; 1531 1532 error = netdev_queue_update_kobjects(dev, 0, real_tx); 1533 if (error) 1534 goto error; 1535 txq = real_tx; 1536 1537 return 0; 1538 1539 error: 1540 netdev_queue_update_kobjects(dev, txq, 0); 1541 net_rx_queue_update_kobjects(dev, rxq, 0); 1542 #ifdef CONFIG_SYSFS 1543 kset_unregister(dev->queues_kset); 1544 #endif 1545 return error; 1546 } 1547 1548 static void remove_queue_kobjects(struct net_device *dev) 1549 { 1550 int real_rx = 0, real_tx = 0; 1551 1552 #ifdef CONFIG_SYSFS 1553 real_rx = dev->real_num_rx_queues; 1554 #endif 1555 real_tx = dev->real_num_tx_queues; 1556 1557 net_rx_queue_update_kobjects(dev, real_rx, 0); 1558 netdev_queue_update_kobjects(dev, real_tx, 0); 1559 #ifdef CONFIG_SYSFS 1560 kset_unregister(dev->queues_kset); 1561 #endif 1562 } 1563 1564 static bool net_current_may_mount(void) 1565 { 1566 struct net *net = current->nsproxy->net_ns; 1567 1568 return ns_capable(net->user_ns, CAP_SYS_ADMIN); 1569 } 1570 1571 static void *net_grab_current_ns(void) 1572 { 1573 struct net *ns = current->nsproxy->net_ns; 1574 #ifdef CONFIG_NET_NS 1575 if (ns) 1576 refcount_inc(&ns->passive); 1577 #endif 1578 return ns; 1579 } 1580 1581 static const void *net_initial_ns(void) 1582 { 1583 return &init_net; 1584 } 1585 1586 static const void *net_netlink_ns(struct sock *sk) 1587 { 1588 return sock_net(sk); 1589 } 1590 1591 const struct kobj_ns_type_operations net_ns_type_operations = { 1592 .type = KOBJ_NS_TYPE_NET, 1593 .current_may_mount = net_current_may_mount, 1594 .grab_current_ns = net_grab_current_ns, 1595 .netlink_ns = net_netlink_ns, 1596 .initial_ns = net_initial_ns, 1597 .drop_ns = net_drop_ns, 1598 }; 1599 EXPORT_SYMBOL_GPL(net_ns_type_operations); 1600 1601 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env) 1602 { 1603 struct net_device *dev = to_net_dev(d); 1604 int retval; 1605 1606 /* pass interface to uevent. */ 1607 retval = add_uevent_var(env, "INTERFACE=%s", dev->name); 1608 if (retval) 1609 goto exit; 1610 1611 /* pass ifindex to uevent. 1612 * ifindex is useful as it won't change (interface name may change) 1613 * and is what RtNetlink uses natively. 1614 */ 1615 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex); 1616 1617 exit: 1618 return retval; 1619 } 1620 1621 /* 1622 * netdev_release -- destroy and free a dead device. 1623 * Called when last reference to device kobject is gone. 1624 */ 1625 static void netdev_release(struct device *d) 1626 { 1627 struct net_device *dev = to_net_dev(d); 1628 1629 BUG_ON(dev->reg_state != NETREG_RELEASED); 1630 1631 /* no need to wait for rcu grace period: 1632 * device is dead and about to be freed. 1633 */ 1634 kfree(rcu_access_pointer(dev->ifalias)); 1635 netdev_freemem(dev); 1636 } 1637 1638 static const void *net_namespace(struct device *d) 1639 { 1640 struct net_device *dev = to_net_dev(d); 1641 1642 return dev_net(dev); 1643 } 1644 1645 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid) 1646 { 1647 struct net_device *dev = to_net_dev(d); 1648 const struct net *net = dev_net(dev); 1649 1650 net_ns_get_ownership(net, uid, gid); 1651 } 1652 1653 static struct class net_class __ro_after_init = { 1654 .name = "net", 1655 .dev_release = netdev_release, 1656 .dev_groups = net_class_groups, 1657 .dev_uevent = netdev_uevent, 1658 .ns_type = &net_ns_type_operations, 1659 .namespace = net_namespace, 1660 .get_ownership = net_get_ownership, 1661 }; 1662 1663 #ifdef CONFIG_OF_NET 1664 static int of_dev_node_match(struct device *dev, const void *data) 1665 { 1666 int ret = 0; 1667 1668 if (dev->parent) 1669 ret = dev->parent->of_node == data; 1670 1671 return ret == 0 ? dev->of_node == data : ret; 1672 } 1673 1674 /* 1675 * of_find_net_device_by_node - lookup the net device for the device node 1676 * @np: OF device node 1677 * 1678 * Looks up the net_device structure corresponding with the device node. 1679 * If successful, returns a pointer to the net_device with the embedded 1680 * struct device refcount incremented by one, or NULL on failure. The 1681 * refcount must be dropped when done with the net_device. 1682 */ 1683 struct net_device *of_find_net_device_by_node(struct device_node *np) 1684 { 1685 struct device *dev; 1686 1687 dev = class_find_device(&net_class, NULL, np, of_dev_node_match); 1688 if (!dev) 1689 return NULL; 1690 1691 return to_net_dev(dev); 1692 } 1693 EXPORT_SYMBOL(of_find_net_device_by_node); 1694 #endif 1695 1696 /* Delete sysfs entries but hold kobject reference until after all 1697 * netdev references are gone. 1698 */ 1699 void netdev_unregister_kobject(struct net_device *ndev) 1700 { 1701 struct device *dev = &ndev->dev; 1702 1703 if (!refcount_read(&dev_net(ndev)->count)) 1704 dev_set_uevent_suppress(dev, 1); 1705 1706 kobject_get(&dev->kobj); 1707 1708 remove_queue_kobjects(ndev); 1709 1710 pm_runtime_set_memalloc_noio(dev, false); 1711 1712 device_del(dev); 1713 } 1714 1715 /* Create sysfs entries for network device. */ 1716 int netdev_register_kobject(struct net_device *ndev) 1717 { 1718 struct device *dev = &ndev->dev; 1719 const struct attribute_group **groups = ndev->sysfs_groups; 1720 int error = 0; 1721 1722 device_initialize(dev); 1723 dev->class = &net_class; 1724 dev->platform_data = ndev; 1725 dev->groups = groups; 1726 1727 dev_set_name(dev, "%s", ndev->name); 1728 1729 #ifdef CONFIG_SYSFS 1730 /* Allow for a device specific group */ 1731 if (*groups) 1732 groups++; 1733 1734 *groups++ = &netstat_group; 1735 1736 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211) 1737 if (ndev->ieee80211_ptr) 1738 *groups++ = &wireless_group; 1739 #if IS_ENABLED(CONFIG_WIRELESS_EXT) 1740 else if (ndev->wireless_handlers) 1741 *groups++ = &wireless_group; 1742 #endif 1743 #endif 1744 #endif /* CONFIG_SYSFS */ 1745 1746 error = device_add(dev); 1747 if (error) 1748 return error; 1749 1750 error = register_queue_kobjects(ndev); 1751 if (error) { 1752 device_del(dev); 1753 return error; 1754 } 1755 1756 pm_runtime_set_memalloc_noio(dev, true); 1757 1758 return error; 1759 } 1760 1761 int netdev_class_create_file_ns(const struct class_attribute *class_attr, 1762 const void *ns) 1763 { 1764 return class_create_file_ns(&net_class, class_attr, ns); 1765 } 1766 EXPORT_SYMBOL(netdev_class_create_file_ns); 1767 1768 void netdev_class_remove_file_ns(const struct class_attribute *class_attr, 1769 const void *ns) 1770 { 1771 class_remove_file_ns(&net_class, class_attr, ns); 1772 } 1773 EXPORT_SYMBOL(netdev_class_remove_file_ns); 1774 1775 int __init netdev_kobject_init(void) 1776 { 1777 kobj_ns_type_register(&net_ns_type_operations); 1778 return class_register(&net_class); 1779 } 1780