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_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); 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 switchdev_attr attr = { 504 .orig_dev = netdev, 505 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID, 506 .flags = SWITCHDEV_F_NO_RECURSE, 507 }; 508 509 ret = switchdev_port_attr_get(netdev, &attr); 510 if (!ret) 511 ret = sprintf(buf, "%*phN\n", attr.u.ppid.id_len, 512 attr.u.ppid.id); 513 } 514 rtnl_unlock(); 515 516 return ret; 517 } 518 static DEVICE_ATTR_RO(phys_switch_id); 519 520 static struct attribute *net_class_attrs[] __ro_after_init = { 521 &dev_attr_netdev_group.attr, 522 &dev_attr_type.attr, 523 &dev_attr_dev_id.attr, 524 &dev_attr_dev_port.attr, 525 &dev_attr_iflink.attr, 526 &dev_attr_ifindex.attr, 527 &dev_attr_name_assign_type.attr, 528 &dev_attr_addr_assign_type.attr, 529 &dev_attr_addr_len.attr, 530 &dev_attr_link_mode.attr, 531 &dev_attr_address.attr, 532 &dev_attr_broadcast.attr, 533 &dev_attr_speed.attr, 534 &dev_attr_duplex.attr, 535 &dev_attr_dormant.attr, 536 &dev_attr_operstate.attr, 537 &dev_attr_carrier_changes.attr, 538 &dev_attr_ifalias.attr, 539 &dev_attr_carrier.attr, 540 &dev_attr_mtu.attr, 541 &dev_attr_flags.attr, 542 &dev_attr_tx_queue_len.attr, 543 &dev_attr_gro_flush_timeout.attr, 544 &dev_attr_phys_port_id.attr, 545 &dev_attr_phys_port_name.attr, 546 &dev_attr_phys_switch_id.attr, 547 &dev_attr_proto_down.attr, 548 &dev_attr_carrier_up_count.attr, 549 &dev_attr_carrier_down_count.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, 0644, 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, 0644, 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 void rx_queue_get_ownership(struct kobject *kobj, 909 kuid_t *uid, kgid_t *gid) 910 { 911 const struct net *net = rx_queue_namespace(kobj); 912 913 net_ns_get_ownership(net, uid, gid); 914 } 915 916 static struct kobj_type rx_queue_ktype __ro_after_init = { 917 .sysfs_ops = &rx_queue_sysfs_ops, 918 .release = rx_queue_release, 919 .default_attrs = rx_queue_default_attrs, 920 .namespace = rx_queue_namespace, 921 .get_ownership = rx_queue_get_ownership, 922 }; 923 924 static int rx_queue_add_kobject(struct net_device *dev, int index) 925 { 926 struct netdev_rx_queue *queue = dev->_rx + index; 927 struct kobject *kobj = &queue->kobj; 928 int error = 0; 929 930 kobj->kset = dev->queues_kset; 931 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL, 932 "rx-%u", index); 933 if (error) 934 return error; 935 936 if (dev->sysfs_rx_queue_group) { 937 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group); 938 if (error) { 939 kobject_put(kobj); 940 return error; 941 } 942 } 943 944 kobject_uevent(kobj, KOBJ_ADD); 945 dev_hold(queue->dev); 946 947 return error; 948 } 949 #endif /* CONFIG_SYSFS */ 950 951 int 952 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num) 953 { 954 #ifdef CONFIG_SYSFS 955 int i; 956 int error = 0; 957 958 #ifndef CONFIG_RPS 959 if (!dev->sysfs_rx_queue_group) 960 return 0; 961 #endif 962 for (i = old_num; i < new_num; i++) { 963 error = rx_queue_add_kobject(dev, i); 964 if (error) { 965 new_num = old_num; 966 break; 967 } 968 } 969 970 while (--i >= new_num) { 971 struct kobject *kobj = &dev->_rx[i].kobj; 972 973 if (!refcount_read(&dev_net(dev)->count)) 974 kobj->uevent_suppress = 1; 975 if (dev->sysfs_rx_queue_group) 976 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group); 977 kobject_put(kobj); 978 } 979 980 return error; 981 #else 982 return 0; 983 #endif 984 } 985 986 #ifdef CONFIG_SYSFS 987 /* 988 * netdev_queue sysfs structures and functions. 989 */ 990 struct netdev_queue_attribute { 991 struct attribute attr; 992 ssize_t (*show)(struct netdev_queue *queue, char *buf); 993 ssize_t (*store)(struct netdev_queue *queue, 994 const char *buf, size_t len); 995 }; 996 #define to_netdev_queue_attr(_attr) \ 997 container_of(_attr, struct netdev_queue_attribute, attr) 998 999 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj) 1000 1001 static ssize_t netdev_queue_attr_show(struct kobject *kobj, 1002 struct attribute *attr, char *buf) 1003 { 1004 const struct netdev_queue_attribute *attribute 1005 = to_netdev_queue_attr(attr); 1006 struct netdev_queue *queue = to_netdev_queue(kobj); 1007 1008 if (!attribute->show) 1009 return -EIO; 1010 1011 return attribute->show(queue, buf); 1012 } 1013 1014 static ssize_t netdev_queue_attr_store(struct kobject *kobj, 1015 struct attribute *attr, 1016 const char *buf, size_t count) 1017 { 1018 const struct netdev_queue_attribute *attribute 1019 = to_netdev_queue_attr(attr); 1020 struct netdev_queue *queue = to_netdev_queue(kobj); 1021 1022 if (!attribute->store) 1023 return -EIO; 1024 1025 return attribute->store(queue, buf, count); 1026 } 1027 1028 static const struct sysfs_ops netdev_queue_sysfs_ops = { 1029 .show = netdev_queue_attr_show, 1030 .store = netdev_queue_attr_store, 1031 }; 1032 1033 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf) 1034 { 1035 unsigned long trans_timeout; 1036 1037 spin_lock_irq(&queue->_xmit_lock); 1038 trans_timeout = queue->trans_timeout; 1039 spin_unlock_irq(&queue->_xmit_lock); 1040 1041 return sprintf(buf, "%lu", trans_timeout); 1042 } 1043 1044 static unsigned int get_netdev_queue_index(struct netdev_queue *queue) 1045 { 1046 struct net_device *dev = queue->dev; 1047 unsigned int i; 1048 1049 i = queue - dev->_tx; 1050 BUG_ON(i >= dev->num_tx_queues); 1051 1052 return i; 1053 } 1054 1055 static ssize_t traffic_class_show(struct netdev_queue *queue, 1056 char *buf) 1057 { 1058 struct net_device *dev = queue->dev; 1059 int index; 1060 int tc; 1061 1062 if (!netif_is_multiqueue(dev)) 1063 return -ENOENT; 1064 1065 index = get_netdev_queue_index(queue); 1066 1067 /* If queue belongs to subordinate dev use its TC mapping */ 1068 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 1069 1070 tc = netdev_txq_to_tc(dev, index); 1071 if (tc < 0) 1072 return -EINVAL; 1073 1074 /* We can report the traffic class one of two ways: 1075 * Subordinate device traffic classes are reported with the traffic 1076 * class first, and then the subordinate class so for example TC0 on 1077 * subordinate device 2 will be reported as "0-2". If the queue 1078 * belongs to the root device it will be reported with just the 1079 * traffic class, so just "0" for TC 0 for example. 1080 */ 1081 return dev->num_tc < 0 ? sprintf(buf, "%u%d\n", tc, dev->num_tc) : 1082 sprintf(buf, "%u\n", tc); 1083 } 1084 1085 #ifdef CONFIG_XPS 1086 static ssize_t tx_maxrate_show(struct netdev_queue *queue, 1087 char *buf) 1088 { 1089 return sprintf(buf, "%lu\n", queue->tx_maxrate); 1090 } 1091 1092 static ssize_t tx_maxrate_store(struct netdev_queue *queue, 1093 const char *buf, size_t len) 1094 { 1095 struct net_device *dev = queue->dev; 1096 int err, index = get_netdev_queue_index(queue); 1097 u32 rate = 0; 1098 1099 if (!capable(CAP_NET_ADMIN)) 1100 return -EPERM; 1101 1102 err = kstrtou32(buf, 10, &rate); 1103 if (err < 0) 1104 return err; 1105 1106 if (!rtnl_trylock()) 1107 return restart_syscall(); 1108 1109 err = -EOPNOTSUPP; 1110 if (dev->netdev_ops->ndo_set_tx_maxrate) 1111 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate); 1112 1113 rtnl_unlock(); 1114 if (!err) { 1115 queue->tx_maxrate = rate; 1116 return len; 1117 } 1118 return err; 1119 } 1120 1121 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init 1122 = __ATTR_RW(tx_maxrate); 1123 #endif 1124 1125 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init 1126 = __ATTR_RO(tx_timeout); 1127 1128 static struct netdev_queue_attribute queue_traffic_class __ro_after_init 1129 = __ATTR_RO(traffic_class); 1130 1131 #ifdef CONFIG_BQL 1132 /* 1133 * Byte queue limits sysfs structures and functions. 1134 */ 1135 static ssize_t bql_show(char *buf, unsigned int value) 1136 { 1137 return sprintf(buf, "%u\n", value); 1138 } 1139 1140 static ssize_t bql_set(const char *buf, const size_t count, 1141 unsigned int *pvalue) 1142 { 1143 unsigned int value; 1144 int err; 1145 1146 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) { 1147 value = DQL_MAX_LIMIT; 1148 } else { 1149 err = kstrtouint(buf, 10, &value); 1150 if (err < 0) 1151 return err; 1152 if (value > DQL_MAX_LIMIT) 1153 return -EINVAL; 1154 } 1155 1156 *pvalue = value; 1157 1158 return count; 1159 } 1160 1161 static ssize_t bql_show_hold_time(struct netdev_queue *queue, 1162 char *buf) 1163 { 1164 struct dql *dql = &queue->dql; 1165 1166 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time)); 1167 } 1168 1169 static ssize_t bql_set_hold_time(struct netdev_queue *queue, 1170 const char *buf, size_t len) 1171 { 1172 struct dql *dql = &queue->dql; 1173 unsigned int value; 1174 int err; 1175 1176 err = kstrtouint(buf, 10, &value); 1177 if (err < 0) 1178 return err; 1179 1180 dql->slack_hold_time = msecs_to_jiffies(value); 1181 1182 return len; 1183 } 1184 1185 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init 1186 = __ATTR(hold_time, 0644, 1187 bql_show_hold_time, bql_set_hold_time); 1188 1189 static ssize_t bql_show_inflight(struct netdev_queue *queue, 1190 char *buf) 1191 { 1192 struct dql *dql = &queue->dql; 1193 1194 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed); 1195 } 1196 1197 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init = 1198 __ATTR(inflight, 0444, bql_show_inflight, NULL); 1199 1200 #define BQL_ATTR(NAME, FIELD) \ 1201 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \ 1202 char *buf) \ 1203 { \ 1204 return bql_show(buf, queue->dql.FIELD); \ 1205 } \ 1206 \ 1207 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \ 1208 const char *buf, size_t len) \ 1209 { \ 1210 return bql_set(buf, len, &queue->dql.FIELD); \ 1211 } \ 1212 \ 1213 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \ 1214 = __ATTR(NAME, 0644, \ 1215 bql_show_ ## NAME, bql_set_ ## NAME) 1216 1217 BQL_ATTR(limit, limit); 1218 BQL_ATTR(limit_max, max_limit); 1219 BQL_ATTR(limit_min, min_limit); 1220 1221 static struct attribute *dql_attrs[] __ro_after_init = { 1222 &bql_limit_attribute.attr, 1223 &bql_limit_max_attribute.attr, 1224 &bql_limit_min_attribute.attr, 1225 &bql_hold_time_attribute.attr, 1226 &bql_inflight_attribute.attr, 1227 NULL 1228 }; 1229 1230 static const struct attribute_group dql_group = { 1231 .name = "byte_queue_limits", 1232 .attrs = dql_attrs, 1233 }; 1234 #endif /* CONFIG_BQL */ 1235 1236 #ifdef CONFIG_XPS 1237 static ssize_t xps_cpus_show(struct netdev_queue *queue, 1238 char *buf) 1239 { 1240 struct net_device *dev = queue->dev; 1241 int cpu, len, num_tc = 1, tc = 0; 1242 struct xps_dev_maps *dev_maps; 1243 cpumask_var_t mask; 1244 unsigned long index; 1245 1246 if (!netif_is_multiqueue(dev)) 1247 return -ENOENT; 1248 1249 index = get_netdev_queue_index(queue); 1250 1251 if (dev->num_tc) { 1252 /* Do not allow XPS on subordinate device directly */ 1253 num_tc = dev->num_tc; 1254 if (num_tc < 0) 1255 return -EINVAL; 1256 1257 /* If queue belongs to subordinate dev use its map */ 1258 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 1259 1260 tc = netdev_txq_to_tc(dev, index); 1261 if (tc < 0) 1262 return -EINVAL; 1263 } 1264 1265 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 1266 return -ENOMEM; 1267 1268 rcu_read_lock(); 1269 dev_maps = rcu_dereference(dev->xps_cpus_map); 1270 if (dev_maps) { 1271 for_each_possible_cpu(cpu) { 1272 int i, tci = cpu * num_tc + tc; 1273 struct xps_map *map; 1274 1275 map = rcu_dereference(dev_maps->attr_map[tci]); 1276 if (!map) 1277 continue; 1278 1279 for (i = map->len; i--;) { 1280 if (map->queues[i] == index) { 1281 cpumask_set_cpu(cpu, mask); 1282 break; 1283 } 1284 } 1285 } 1286 } 1287 rcu_read_unlock(); 1288 1289 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask)); 1290 free_cpumask_var(mask); 1291 return len < PAGE_SIZE ? len : -EINVAL; 1292 } 1293 1294 static ssize_t xps_cpus_store(struct netdev_queue *queue, 1295 const char *buf, size_t len) 1296 { 1297 struct net_device *dev = queue->dev; 1298 unsigned long index; 1299 cpumask_var_t mask; 1300 int err; 1301 1302 if (!netif_is_multiqueue(dev)) 1303 return -ENOENT; 1304 1305 if (!capable(CAP_NET_ADMIN)) 1306 return -EPERM; 1307 1308 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 1309 return -ENOMEM; 1310 1311 index = get_netdev_queue_index(queue); 1312 1313 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits); 1314 if (err) { 1315 free_cpumask_var(mask); 1316 return err; 1317 } 1318 1319 err = netif_set_xps_queue(dev, mask, index); 1320 1321 free_cpumask_var(mask); 1322 1323 return err ? : len; 1324 } 1325 1326 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init 1327 = __ATTR_RW(xps_cpus); 1328 1329 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf) 1330 { 1331 struct net_device *dev = queue->dev; 1332 struct xps_dev_maps *dev_maps; 1333 unsigned long *mask, index; 1334 int j, len, num_tc = 1, tc = 0; 1335 1336 index = get_netdev_queue_index(queue); 1337 1338 if (dev->num_tc) { 1339 num_tc = dev->num_tc; 1340 tc = netdev_txq_to_tc(dev, index); 1341 if (tc < 0) 1342 return -EINVAL; 1343 } 1344 mask = kcalloc(BITS_TO_LONGS(dev->num_rx_queues), sizeof(long), 1345 GFP_KERNEL); 1346 if (!mask) 1347 return -ENOMEM; 1348 1349 rcu_read_lock(); 1350 dev_maps = rcu_dereference(dev->xps_rxqs_map); 1351 if (!dev_maps) 1352 goto out_no_maps; 1353 1354 for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues), 1355 j < dev->num_rx_queues;) { 1356 int i, tci = j * num_tc + tc; 1357 struct xps_map *map; 1358 1359 map = rcu_dereference(dev_maps->attr_map[tci]); 1360 if (!map) 1361 continue; 1362 1363 for (i = map->len; i--;) { 1364 if (map->queues[i] == index) { 1365 set_bit(j, mask); 1366 break; 1367 } 1368 } 1369 } 1370 out_no_maps: 1371 rcu_read_unlock(); 1372 1373 len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues); 1374 kfree(mask); 1375 1376 return len < PAGE_SIZE ? len : -EINVAL; 1377 } 1378 1379 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf, 1380 size_t len) 1381 { 1382 struct net_device *dev = queue->dev; 1383 struct net *net = dev_net(dev); 1384 unsigned long *mask, index; 1385 int err; 1386 1387 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 1388 return -EPERM; 1389 1390 mask = kcalloc(BITS_TO_LONGS(dev->num_rx_queues), sizeof(long), 1391 GFP_KERNEL); 1392 if (!mask) 1393 return -ENOMEM; 1394 1395 index = get_netdev_queue_index(queue); 1396 1397 err = bitmap_parse(buf, len, mask, dev->num_rx_queues); 1398 if (err) { 1399 kfree(mask); 1400 return err; 1401 } 1402 1403 err = __netif_set_xps_queue(dev, mask, index, true); 1404 kfree(mask); 1405 return err ? : len; 1406 } 1407 1408 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init 1409 = __ATTR_RW(xps_rxqs); 1410 #endif /* CONFIG_XPS */ 1411 1412 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = { 1413 &queue_trans_timeout.attr, 1414 &queue_traffic_class.attr, 1415 #ifdef CONFIG_XPS 1416 &xps_cpus_attribute.attr, 1417 &xps_rxqs_attribute.attr, 1418 &queue_tx_maxrate.attr, 1419 #endif 1420 NULL 1421 }; 1422 1423 static void netdev_queue_release(struct kobject *kobj) 1424 { 1425 struct netdev_queue *queue = to_netdev_queue(kobj); 1426 1427 memset(kobj, 0, sizeof(*kobj)); 1428 dev_put(queue->dev); 1429 } 1430 1431 static const void *netdev_queue_namespace(struct kobject *kobj) 1432 { 1433 struct netdev_queue *queue = to_netdev_queue(kobj); 1434 struct device *dev = &queue->dev->dev; 1435 const void *ns = NULL; 1436 1437 if (dev->class && dev->class->ns_type) 1438 ns = dev->class->namespace(dev); 1439 1440 return ns; 1441 } 1442 1443 static void netdev_queue_get_ownership(struct kobject *kobj, 1444 kuid_t *uid, kgid_t *gid) 1445 { 1446 const struct net *net = netdev_queue_namespace(kobj); 1447 1448 net_ns_get_ownership(net, uid, gid); 1449 } 1450 1451 static struct kobj_type netdev_queue_ktype __ro_after_init = { 1452 .sysfs_ops = &netdev_queue_sysfs_ops, 1453 .release = netdev_queue_release, 1454 .default_attrs = netdev_queue_default_attrs, 1455 .namespace = netdev_queue_namespace, 1456 .get_ownership = netdev_queue_get_ownership, 1457 }; 1458 1459 static int netdev_queue_add_kobject(struct net_device *dev, int index) 1460 { 1461 struct netdev_queue *queue = dev->_tx + index; 1462 struct kobject *kobj = &queue->kobj; 1463 int error = 0; 1464 1465 kobj->kset = dev->queues_kset; 1466 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL, 1467 "tx-%u", index); 1468 if (error) 1469 return error; 1470 1471 #ifdef CONFIG_BQL 1472 error = sysfs_create_group(kobj, &dql_group); 1473 if (error) { 1474 kobject_put(kobj); 1475 return error; 1476 } 1477 #endif 1478 1479 kobject_uevent(kobj, KOBJ_ADD); 1480 dev_hold(queue->dev); 1481 1482 return 0; 1483 } 1484 #endif /* CONFIG_SYSFS */ 1485 1486 int 1487 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num) 1488 { 1489 #ifdef CONFIG_SYSFS 1490 int i; 1491 int error = 0; 1492 1493 for (i = old_num; i < new_num; i++) { 1494 error = netdev_queue_add_kobject(dev, i); 1495 if (error) { 1496 new_num = old_num; 1497 break; 1498 } 1499 } 1500 1501 while (--i >= new_num) { 1502 struct netdev_queue *queue = dev->_tx + i; 1503 1504 if (!refcount_read(&dev_net(dev)->count)) 1505 queue->kobj.uevent_suppress = 1; 1506 #ifdef CONFIG_BQL 1507 sysfs_remove_group(&queue->kobj, &dql_group); 1508 #endif 1509 kobject_put(&queue->kobj); 1510 } 1511 1512 return error; 1513 #else 1514 return 0; 1515 #endif /* CONFIG_SYSFS */ 1516 } 1517 1518 static int register_queue_kobjects(struct net_device *dev) 1519 { 1520 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0; 1521 1522 #ifdef CONFIG_SYSFS 1523 dev->queues_kset = kset_create_and_add("queues", 1524 NULL, &dev->dev.kobj); 1525 if (!dev->queues_kset) 1526 return -ENOMEM; 1527 real_rx = dev->real_num_rx_queues; 1528 #endif 1529 real_tx = dev->real_num_tx_queues; 1530 1531 error = net_rx_queue_update_kobjects(dev, 0, real_rx); 1532 if (error) 1533 goto error; 1534 rxq = real_rx; 1535 1536 error = netdev_queue_update_kobjects(dev, 0, real_tx); 1537 if (error) 1538 goto error; 1539 txq = real_tx; 1540 1541 return 0; 1542 1543 error: 1544 netdev_queue_update_kobjects(dev, txq, 0); 1545 net_rx_queue_update_kobjects(dev, rxq, 0); 1546 return error; 1547 } 1548 1549 static void remove_queue_kobjects(struct net_device *dev) 1550 { 1551 int real_rx = 0, real_tx = 0; 1552 1553 #ifdef CONFIG_SYSFS 1554 real_rx = dev->real_num_rx_queues; 1555 #endif 1556 real_tx = dev->real_num_tx_queues; 1557 1558 net_rx_queue_update_kobjects(dev, real_rx, 0); 1559 netdev_queue_update_kobjects(dev, real_tx, 0); 1560 #ifdef CONFIG_SYSFS 1561 kset_unregister(dev->queues_kset); 1562 #endif 1563 } 1564 1565 static bool net_current_may_mount(void) 1566 { 1567 struct net *net = current->nsproxy->net_ns; 1568 1569 return ns_capable(net->user_ns, CAP_SYS_ADMIN); 1570 } 1571 1572 static void *net_grab_current_ns(void) 1573 { 1574 struct net *ns = current->nsproxy->net_ns; 1575 #ifdef CONFIG_NET_NS 1576 if (ns) 1577 refcount_inc(&ns->passive); 1578 #endif 1579 return ns; 1580 } 1581 1582 static const void *net_initial_ns(void) 1583 { 1584 return &init_net; 1585 } 1586 1587 static const void *net_netlink_ns(struct sock *sk) 1588 { 1589 return sock_net(sk); 1590 } 1591 1592 const struct kobj_ns_type_operations net_ns_type_operations = { 1593 .type = KOBJ_NS_TYPE_NET, 1594 .current_may_mount = net_current_may_mount, 1595 .grab_current_ns = net_grab_current_ns, 1596 .netlink_ns = net_netlink_ns, 1597 .initial_ns = net_initial_ns, 1598 .drop_ns = net_drop_ns, 1599 }; 1600 EXPORT_SYMBOL_GPL(net_ns_type_operations); 1601 1602 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env) 1603 { 1604 struct net_device *dev = to_net_dev(d); 1605 int retval; 1606 1607 /* pass interface to uevent. */ 1608 retval = add_uevent_var(env, "INTERFACE=%s", dev->name); 1609 if (retval) 1610 goto exit; 1611 1612 /* pass ifindex to uevent. 1613 * ifindex is useful as it won't change (interface name may change) 1614 * and is what RtNetlink uses natively. 1615 */ 1616 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex); 1617 1618 exit: 1619 return retval; 1620 } 1621 1622 /* 1623 * netdev_release -- destroy and free a dead device. 1624 * Called when last reference to device kobject is gone. 1625 */ 1626 static void netdev_release(struct device *d) 1627 { 1628 struct net_device *dev = to_net_dev(d); 1629 1630 BUG_ON(dev->reg_state != NETREG_RELEASED); 1631 1632 /* no need to wait for rcu grace period: 1633 * device is dead and about to be freed. 1634 */ 1635 kfree(rcu_access_pointer(dev->ifalias)); 1636 netdev_freemem(dev); 1637 } 1638 1639 static const void *net_namespace(struct device *d) 1640 { 1641 struct net_device *dev = to_net_dev(d); 1642 1643 return dev_net(dev); 1644 } 1645 1646 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid) 1647 { 1648 struct net_device *dev = to_net_dev(d); 1649 const struct net *net = dev_net(dev); 1650 1651 net_ns_get_ownership(net, uid, gid); 1652 } 1653 1654 static struct class net_class __ro_after_init = { 1655 .name = "net", 1656 .dev_release = netdev_release, 1657 .dev_groups = net_class_groups, 1658 .dev_uevent = netdev_uevent, 1659 .ns_type = &net_ns_type_operations, 1660 .namespace = net_namespace, 1661 .get_ownership = net_get_ownership, 1662 }; 1663 1664 #ifdef CONFIG_OF_NET 1665 static int of_dev_node_match(struct device *dev, const void *data) 1666 { 1667 int ret = 0; 1668 1669 if (dev->parent) 1670 ret = dev->parent->of_node == data; 1671 1672 return ret == 0 ? dev->of_node == data : ret; 1673 } 1674 1675 /* 1676 * of_find_net_device_by_node - lookup the net device for the device node 1677 * @np: OF device node 1678 * 1679 * Looks up the net_device structure corresponding with the device node. 1680 * If successful, returns a pointer to the net_device with the embedded 1681 * struct device refcount incremented by one, or NULL on failure. The 1682 * refcount must be dropped when done with the net_device. 1683 */ 1684 struct net_device *of_find_net_device_by_node(struct device_node *np) 1685 { 1686 struct device *dev; 1687 1688 dev = class_find_device(&net_class, NULL, np, of_dev_node_match); 1689 if (!dev) 1690 return NULL; 1691 1692 return to_net_dev(dev); 1693 } 1694 EXPORT_SYMBOL(of_find_net_device_by_node); 1695 #endif 1696 1697 /* Delete sysfs entries but hold kobject reference until after all 1698 * netdev references are gone. 1699 */ 1700 void netdev_unregister_kobject(struct net_device *ndev) 1701 { 1702 struct device *dev = &ndev->dev; 1703 1704 if (!refcount_read(&dev_net(ndev)->count)) 1705 dev_set_uevent_suppress(dev, 1); 1706 1707 kobject_get(&dev->kobj); 1708 1709 remove_queue_kobjects(ndev); 1710 1711 pm_runtime_set_memalloc_noio(dev, false); 1712 1713 device_del(dev); 1714 } 1715 1716 /* Create sysfs entries for network device. */ 1717 int netdev_register_kobject(struct net_device *ndev) 1718 { 1719 struct device *dev = &ndev->dev; 1720 const struct attribute_group **groups = ndev->sysfs_groups; 1721 int error = 0; 1722 1723 device_initialize(dev); 1724 dev->class = &net_class; 1725 dev->platform_data = ndev; 1726 dev->groups = groups; 1727 1728 dev_set_name(dev, "%s", ndev->name); 1729 1730 #ifdef CONFIG_SYSFS 1731 /* Allow for a device specific group */ 1732 if (*groups) 1733 groups++; 1734 1735 *groups++ = &netstat_group; 1736 1737 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211) 1738 if (ndev->ieee80211_ptr) 1739 *groups++ = &wireless_group; 1740 #if IS_ENABLED(CONFIG_WIRELESS_EXT) 1741 else if (ndev->wireless_handlers) 1742 *groups++ = &wireless_group; 1743 #endif 1744 #endif 1745 #endif /* CONFIG_SYSFS */ 1746 1747 error = device_add(dev); 1748 if (error) 1749 return error; 1750 1751 error = register_queue_kobjects(ndev); 1752 if (error) { 1753 device_del(dev); 1754 return error; 1755 } 1756 1757 pm_runtime_set_memalloc_noio(dev, true); 1758 1759 return error; 1760 } 1761 1762 int netdev_class_create_file_ns(const struct class_attribute *class_attr, 1763 const void *ns) 1764 { 1765 return class_create_file_ns(&net_class, class_attr, ns); 1766 } 1767 EXPORT_SYMBOL(netdev_class_create_file_ns); 1768 1769 void netdev_class_remove_file_ns(const struct class_attribute *class_attr, 1770 const void *ns) 1771 { 1772 class_remove_file_ns(&net_class, class_attr, ns); 1773 } 1774 EXPORT_SYMBOL(netdev_class_remove_file_ns); 1775 1776 int __init netdev_kobject_init(void) 1777 { 1778 kobj_ns_type_register(&net_ns_type_operations); 1779 return class_register(&net_class); 1780 } 1781