1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * net-sysfs.c - network device class and attributes 4 * 5 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org> 6 */ 7 8 #include <linux/capability.h> 9 #include <linux/kernel.h> 10 #include <linux/netdevice.h> 11 #include <linux/if_arp.h> 12 #include <linux/slab.h> 13 #include <linux/sched/signal.h> 14 #include <linux/sched/isolation.h> 15 #include <linux/nsproxy.h> 16 #include <net/sock.h> 17 #include <net/net_namespace.h> 18 #include <linux/rtnetlink.h> 19 #include <linux/vmalloc.h> 20 #include <linux/export.h> 21 #include <linux/jiffies.h> 22 #include <linux/pm_runtime.h> 23 #include <linux/of.h> 24 #include <linux/of_net.h> 25 #include <linux/cpu.h> 26 #include <net/netdev_rx_queue.h> 27 28 #include "dev.h" 29 #include "net-sysfs.h" 30 31 #ifdef CONFIG_SYSFS 32 static const char fmt_hex[] = "%#x\n"; 33 static const char fmt_dec[] = "%d\n"; 34 static const char fmt_ulong[] = "%lu\n"; 35 static const char fmt_u64[] = "%llu\n"; 36 37 /* Caller holds RTNL or RCU */ 38 static inline int dev_isalive(const struct net_device *dev) 39 { 40 return READ_ONCE(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 rcu_read_lock(); 52 if (dev_isalive(ndev)) 53 ret = (*format)(ndev, buf); 54 rcu_read_unlock(); 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 sysfs_emit(buf, format_string, READ_ONCE(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; 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 sysfs_emit(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 sysfs_emit(buf, fmt_dec, READ_ONCE(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 (READ_ONCE(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 (dev_get_mac_address()) */ 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 down_read(&dev_addr_sem); 153 154 rcu_read_lock(); 155 if (dev_isalive(ndev)) 156 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len); 157 rcu_read_unlock(); 158 159 up_read(&dev_addr_sem); 160 return ret; 161 } 162 static DEVICE_ATTR_RO(address); 163 164 static ssize_t broadcast_show(struct device *dev, 165 struct device_attribute *attr, char *buf) 166 { 167 struct net_device *ndev = to_net_dev(dev); 168 int ret = -EINVAL; 169 170 rcu_read_lock(); 171 if (dev_isalive(ndev)) 172 ret = sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len); 173 rcu_read_unlock(); 174 return ret; 175 } 176 static DEVICE_ATTR_RO(broadcast); 177 178 static int change_carrier(struct net_device *dev, unsigned long new_carrier) 179 { 180 if (!netif_running(dev)) 181 return -EINVAL; 182 return dev_change_carrier(dev, (bool)new_carrier); 183 } 184 185 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr, 186 const char *buf, size_t len) 187 { 188 struct net_device *netdev = to_net_dev(dev); 189 190 /* The check is also done in change_carrier; this helps returning early 191 * without hitting the trylock/restart in netdev_store. 192 */ 193 if (!netdev->netdev_ops->ndo_change_carrier) 194 return -EOPNOTSUPP; 195 196 return netdev_store(dev, attr, buf, len, change_carrier); 197 } 198 199 static ssize_t carrier_show(struct device *dev, 200 struct device_attribute *attr, char *buf) 201 { 202 struct net_device *netdev = to_net_dev(dev); 203 int ret = -EINVAL; 204 205 if (!rtnl_trylock()) 206 return restart_syscall(); 207 208 if (netif_running(netdev)) { 209 /* Synchronize carrier state with link watch, 210 * see also rtnl_getlink(). 211 */ 212 linkwatch_sync_dev(netdev); 213 214 ret = sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev)); 215 } 216 rtnl_unlock(); 217 218 return ret; 219 } 220 static DEVICE_ATTR_RW(carrier); 221 222 static ssize_t speed_show(struct device *dev, 223 struct device_attribute *attr, char *buf) 224 { 225 struct net_device *netdev = to_net_dev(dev); 226 int ret = -EINVAL; 227 228 /* The check is also done in __ethtool_get_link_ksettings; this helps 229 * returning early without hitting the trylock/restart below. 230 */ 231 if (!netdev->ethtool_ops->get_link_ksettings) 232 return ret; 233 234 if (!rtnl_trylock()) 235 return restart_syscall(); 236 237 if (netif_running(netdev) && netif_device_present(netdev)) { 238 struct ethtool_link_ksettings cmd; 239 240 if (!__ethtool_get_link_ksettings(netdev, &cmd)) 241 ret = sysfs_emit(buf, fmt_dec, cmd.base.speed); 242 } 243 rtnl_unlock(); 244 return ret; 245 } 246 static DEVICE_ATTR_RO(speed); 247 248 static ssize_t duplex_show(struct device *dev, 249 struct device_attribute *attr, char *buf) 250 { 251 struct net_device *netdev = to_net_dev(dev); 252 int ret = -EINVAL; 253 254 /* The check is also done in __ethtool_get_link_ksettings; this helps 255 * returning early without hitting the trylock/restart below. 256 */ 257 if (!netdev->ethtool_ops->get_link_ksettings) 258 return ret; 259 260 if (!rtnl_trylock()) 261 return restart_syscall(); 262 263 if (netif_running(netdev)) { 264 struct ethtool_link_ksettings cmd; 265 266 if (!__ethtool_get_link_ksettings(netdev, &cmd)) { 267 const char *duplex; 268 269 switch (cmd.base.duplex) { 270 case DUPLEX_HALF: 271 duplex = "half"; 272 break; 273 case DUPLEX_FULL: 274 duplex = "full"; 275 break; 276 default: 277 duplex = "unknown"; 278 break; 279 } 280 ret = sysfs_emit(buf, "%s\n", duplex); 281 } 282 } 283 rtnl_unlock(); 284 return ret; 285 } 286 static DEVICE_ATTR_RO(duplex); 287 288 static ssize_t testing_show(struct device *dev, 289 struct device_attribute *attr, char *buf) 290 { 291 struct net_device *netdev = to_net_dev(dev); 292 293 if (netif_running(netdev)) 294 return sysfs_emit(buf, fmt_dec, !!netif_testing(netdev)); 295 296 return -EINVAL; 297 } 298 static DEVICE_ATTR_RO(testing); 299 300 static ssize_t dormant_show(struct device *dev, 301 struct device_attribute *attr, char *buf) 302 { 303 struct net_device *netdev = to_net_dev(dev); 304 305 if (netif_running(netdev)) 306 return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev)); 307 308 return -EINVAL; 309 } 310 static DEVICE_ATTR_RO(dormant); 311 312 static const char *const operstates[] = { 313 "unknown", 314 "notpresent", /* currently unused */ 315 "down", 316 "lowerlayerdown", 317 "testing", 318 "dormant", 319 "up" 320 }; 321 322 static ssize_t operstate_show(struct device *dev, 323 struct device_attribute *attr, char *buf) 324 { 325 const struct net_device *netdev = to_net_dev(dev); 326 unsigned char operstate; 327 328 operstate = READ_ONCE(netdev->operstate); 329 if (!netif_running(netdev)) 330 operstate = IF_OPER_DOWN; 331 332 if (operstate >= ARRAY_SIZE(operstates)) 333 return -EINVAL; /* should not happen */ 334 335 return sysfs_emit(buf, "%s\n", operstates[operstate]); 336 } 337 static DEVICE_ATTR_RO(operstate); 338 339 static ssize_t carrier_changes_show(struct device *dev, 340 struct device_attribute *attr, 341 char *buf) 342 { 343 struct net_device *netdev = to_net_dev(dev); 344 345 return sysfs_emit(buf, fmt_dec, 346 atomic_read(&netdev->carrier_up_count) + 347 atomic_read(&netdev->carrier_down_count)); 348 } 349 static DEVICE_ATTR_RO(carrier_changes); 350 351 static ssize_t carrier_up_count_show(struct device *dev, 352 struct device_attribute *attr, 353 char *buf) 354 { 355 struct net_device *netdev = to_net_dev(dev); 356 357 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_up_count)); 358 } 359 static DEVICE_ATTR_RO(carrier_up_count); 360 361 static ssize_t carrier_down_count_show(struct device *dev, 362 struct device_attribute *attr, 363 char *buf) 364 { 365 struct net_device *netdev = to_net_dev(dev); 366 367 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_down_count)); 368 } 369 static DEVICE_ATTR_RO(carrier_down_count); 370 371 /* read-write attributes */ 372 373 static int change_mtu(struct net_device *dev, unsigned long new_mtu) 374 { 375 return dev_set_mtu(dev, (int)new_mtu); 376 } 377 378 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr, 379 const char *buf, size_t len) 380 { 381 return netdev_store(dev, attr, buf, len, change_mtu); 382 } 383 NETDEVICE_SHOW_RW(mtu, fmt_dec); 384 385 static int change_flags(struct net_device *dev, unsigned long new_flags) 386 { 387 return dev_change_flags(dev, (unsigned int)new_flags, NULL); 388 } 389 390 static ssize_t flags_store(struct device *dev, struct device_attribute *attr, 391 const char *buf, size_t len) 392 { 393 return netdev_store(dev, attr, buf, len, change_flags); 394 } 395 NETDEVICE_SHOW_RW(flags, fmt_hex); 396 397 static ssize_t tx_queue_len_store(struct device *dev, 398 struct device_attribute *attr, 399 const char *buf, size_t len) 400 { 401 if (!capable(CAP_NET_ADMIN)) 402 return -EPERM; 403 404 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len); 405 } 406 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec); 407 408 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val) 409 { 410 WRITE_ONCE(dev->gro_flush_timeout, val); 411 return 0; 412 } 413 414 static ssize_t gro_flush_timeout_store(struct device *dev, 415 struct device_attribute *attr, 416 const char *buf, size_t len) 417 { 418 if (!capable(CAP_NET_ADMIN)) 419 return -EPERM; 420 421 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout); 422 } 423 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong); 424 425 static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val) 426 { 427 WRITE_ONCE(dev->napi_defer_hard_irqs, val); 428 return 0; 429 } 430 431 static ssize_t napi_defer_hard_irqs_store(struct device *dev, 432 struct device_attribute *attr, 433 const char *buf, size_t len) 434 { 435 if (!capable(CAP_NET_ADMIN)) 436 return -EPERM; 437 438 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs); 439 } 440 NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec); 441 442 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr, 443 const char *buf, size_t len) 444 { 445 struct net_device *netdev = to_net_dev(dev); 446 struct net *net = dev_net(netdev); 447 size_t count = len; 448 ssize_t ret = 0; 449 450 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 451 return -EPERM; 452 453 /* ignore trailing newline */ 454 if (len > 0 && buf[len - 1] == '\n') 455 --count; 456 457 if (!rtnl_trylock()) 458 return restart_syscall(); 459 460 if (dev_isalive(netdev)) { 461 ret = dev_set_alias(netdev, buf, count); 462 if (ret < 0) 463 goto err; 464 ret = len; 465 netdev_state_change(netdev); 466 } 467 err: 468 rtnl_unlock(); 469 470 return ret; 471 } 472 473 static ssize_t ifalias_show(struct device *dev, 474 struct device_attribute *attr, char *buf) 475 { 476 const struct net_device *netdev = to_net_dev(dev); 477 char tmp[IFALIASZ]; 478 ssize_t ret = 0; 479 480 ret = dev_get_alias(netdev, tmp, sizeof(tmp)); 481 if (ret > 0) 482 ret = sysfs_emit(buf, "%s\n", tmp); 483 return ret; 484 } 485 static DEVICE_ATTR_RW(ifalias); 486 487 static int change_group(struct net_device *dev, unsigned long new_group) 488 { 489 dev_set_group(dev, (int)new_group); 490 return 0; 491 } 492 493 static ssize_t group_store(struct device *dev, struct device_attribute *attr, 494 const char *buf, size_t len) 495 { 496 return netdev_store(dev, attr, buf, len, change_group); 497 } 498 NETDEVICE_SHOW(group, fmt_dec); 499 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store); 500 501 static int change_proto_down(struct net_device *dev, unsigned long proto_down) 502 { 503 return dev_change_proto_down(dev, (bool)proto_down); 504 } 505 506 static ssize_t proto_down_store(struct device *dev, 507 struct device_attribute *attr, 508 const char *buf, size_t len) 509 { 510 return netdev_store(dev, attr, buf, len, change_proto_down); 511 } 512 NETDEVICE_SHOW_RW(proto_down, fmt_dec); 513 514 static ssize_t phys_port_id_show(struct device *dev, 515 struct device_attribute *attr, char *buf) 516 { 517 struct net_device *netdev = to_net_dev(dev); 518 ssize_t ret = -EINVAL; 519 520 /* The check is also done in dev_get_phys_port_id; this helps returning 521 * early without hitting the trylock/restart below. 522 */ 523 if (!netdev->netdev_ops->ndo_get_phys_port_id) 524 return -EOPNOTSUPP; 525 526 if (!rtnl_trylock()) 527 return restart_syscall(); 528 529 if (dev_isalive(netdev)) { 530 struct netdev_phys_item_id ppid; 531 532 ret = dev_get_phys_port_id(netdev, &ppid); 533 if (!ret) 534 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id); 535 } 536 rtnl_unlock(); 537 538 return ret; 539 } 540 static DEVICE_ATTR_RO(phys_port_id); 541 542 static ssize_t phys_port_name_show(struct device *dev, 543 struct device_attribute *attr, char *buf) 544 { 545 struct net_device *netdev = to_net_dev(dev); 546 ssize_t ret = -EINVAL; 547 548 /* The checks are also done in dev_get_phys_port_name; this helps 549 * returning early without hitting the trylock/restart below. 550 */ 551 if (!netdev->netdev_ops->ndo_get_phys_port_name && 552 !netdev->devlink_port) 553 return -EOPNOTSUPP; 554 555 if (!rtnl_trylock()) 556 return restart_syscall(); 557 558 if (dev_isalive(netdev)) { 559 char name[IFNAMSIZ]; 560 561 ret = dev_get_phys_port_name(netdev, name, sizeof(name)); 562 if (!ret) 563 ret = sysfs_emit(buf, "%s\n", name); 564 } 565 rtnl_unlock(); 566 567 return ret; 568 } 569 static DEVICE_ATTR_RO(phys_port_name); 570 571 static ssize_t phys_switch_id_show(struct device *dev, 572 struct device_attribute *attr, char *buf) 573 { 574 struct net_device *netdev = to_net_dev(dev); 575 ssize_t ret = -EINVAL; 576 577 /* The checks are also done in dev_get_phys_port_name; this helps 578 * returning early without hitting the trylock/restart below. This works 579 * because recurse is false when calling dev_get_port_parent_id. 580 */ 581 if (!netdev->netdev_ops->ndo_get_port_parent_id && 582 !netdev->devlink_port) 583 return -EOPNOTSUPP; 584 585 if (!rtnl_trylock()) 586 return restart_syscall(); 587 588 if (dev_isalive(netdev)) { 589 struct netdev_phys_item_id ppid = { }; 590 591 ret = dev_get_port_parent_id(netdev, &ppid, false); 592 if (!ret) 593 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id); 594 } 595 rtnl_unlock(); 596 597 return ret; 598 } 599 static DEVICE_ATTR_RO(phys_switch_id); 600 601 static ssize_t threaded_show(struct device *dev, 602 struct device_attribute *attr, char *buf) 603 { 604 struct net_device *netdev = to_net_dev(dev); 605 ssize_t ret = -EINVAL; 606 607 if (!rtnl_trylock()) 608 return restart_syscall(); 609 610 if (dev_isalive(netdev)) 611 ret = sysfs_emit(buf, fmt_dec, netdev->threaded); 612 613 rtnl_unlock(); 614 return ret; 615 } 616 617 static int modify_napi_threaded(struct net_device *dev, unsigned long val) 618 { 619 int ret; 620 621 if (list_empty(&dev->napi_list)) 622 return -EOPNOTSUPP; 623 624 if (val != 0 && val != 1) 625 return -EOPNOTSUPP; 626 627 ret = dev_set_threaded(dev, val); 628 629 return ret; 630 } 631 632 static ssize_t threaded_store(struct device *dev, 633 struct device_attribute *attr, 634 const char *buf, size_t len) 635 { 636 return netdev_store(dev, attr, buf, len, modify_napi_threaded); 637 } 638 static DEVICE_ATTR_RW(threaded); 639 640 static struct attribute *net_class_attrs[] __ro_after_init = { 641 &dev_attr_netdev_group.attr, 642 &dev_attr_type.attr, 643 &dev_attr_dev_id.attr, 644 &dev_attr_dev_port.attr, 645 &dev_attr_iflink.attr, 646 &dev_attr_ifindex.attr, 647 &dev_attr_name_assign_type.attr, 648 &dev_attr_addr_assign_type.attr, 649 &dev_attr_addr_len.attr, 650 &dev_attr_link_mode.attr, 651 &dev_attr_address.attr, 652 &dev_attr_broadcast.attr, 653 &dev_attr_speed.attr, 654 &dev_attr_duplex.attr, 655 &dev_attr_dormant.attr, 656 &dev_attr_testing.attr, 657 &dev_attr_operstate.attr, 658 &dev_attr_carrier_changes.attr, 659 &dev_attr_ifalias.attr, 660 &dev_attr_carrier.attr, 661 &dev_attr_mtu.attr, 662 &dev_attr_flags.attr, 663 &dev_attr_tx_queue_len.attr, 664 &dev_attr_gro_flush_timeout.attr, 665 &dev_attr_napi_defer_hard_irqs.attr, 666 &dev_attr_phys_port_id.attr, 667 &dev_attr_phys_port_name.attr, 668 &dev_attr_phys_switch_id.attr, 669 &dev_attr_proto_down.attr, 670 &dev_attr_carrier_up_count.attr, 671 &dev_attr_carrier_down_count.attr, 672 &dev_attr_threaded.attr, 673 NULL, 674 }; 675 ATTRIBUTE_GROUPS(net_class); 676 677 /* Show a given an attribute in the statistics group */ 678 static ssize_t netstat_show(const struct device *d, 679 struct device_attribute *attr, char *buf, 680 unsigned long offset) 681 { 682 struct net_device *dev = to_net_dev(d); 683 ssize_t ret = -EINVAL; 684 685 WARN_ON(offset > sizeof(struct rtnl_link_stats64) || 686 offset % sizeof(u64) != 0); 687 688 rcu_read_lock(); 689 if (dev_isalive(dev)) { 690 struct rtnl_link_stats64 temp; 691 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); 692 693 ret = sysfs_emit(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset)); 694 } 695 rcu_read_unlock(); 696 return ret; 697 } 698 699 /* generate a read-only statistics attribute */ 700 #define NETSTAT_ENTRY(name) \ 701 static ssize_t name##_show(struct device *d, \ 702 struct device_attribute *attr, char *buf) \ 703 { \ 704 return netstat_show(d, attr, buf, \ 705 offsetof(struct rtnl_link_stats64, name)); \ 706 } \ 707 static DEVICE_ATTR_RO(name) 708 709 NETSTAT_ENTRY(rx_packets); 710 NETSTAT_ENTRY(tx_packets); 711 NETSTAT_ENTRY(rx_bytes); 712 NETSTAT_ENTRY(tx_bytes); 713 NETSTAT_ENTRY(rx_errors); 714 NETSTAT_ENTRY(tx_errors); 715 NETSTAT_ENTRY(rx_dropped); 716 NETSTAT_ENTRY(tx_dropped); 717 NETSTAT_ENTRY(multicast); 718 NETSTAT_ENTRY(collisions); 719 NETSTAT_ENTRY(rx_length_errors); 720 NETSTAT_ENTRY(rx_over_errors); 721 NETSTAT_ENTRY(rx_crc_errors); 722 NETSTAT_ENTRY(rx_frame_errors); 723 NETSTAT_ENTRY(rx_fifo_errors); 724 NETSTAT_ENTRY(rx_missed_errors); 725 NETSTAT_ENTRY(tx_aborted_errors); 726 NETSTAT_ENTRY(tx_carrier_errors); 727 NETSTAT_ENTRY(tx_fifo_errors); 728 NETSTAT_ENTRY(tx_heartbeat_errors); 729 NETSTAT_ENTRY(tx_window_errors); 730 NETSTAT_ENTRY(rx_compressed); 731 NETSTAT_ENTRY(tx_compressed); 732 NETSTAT_ENTRY(rx_nohandler); 733 734 static struct attribute *netstat_attrs[] __ro_after_init = { 735 &dev_attr_rx_packets.attr, 736 &dev_attr_tx_packets.attr, 737 &dev_attr_rx_bytes.attr, 738 &dev_attr_tx_bytes.attr, 739 &dev_attr_rx_errors.attr, 740 &dev_attr_tx_errors.attr, 741 &dev_attr_rx_dropped.attr, 742 &dev_attr_tx_dropped.attr, 743 &dev_attr_multicast.attr, 744 &dev_attr_collisions.attr, 745 &dev_attr_rx_length_errors.attr, 746 &dev_attr_rx_over_errors.attr, 747 &dev_attr_rx_crc_errors.attr, 748 &dev_attr_rx_frame_errors.attr, 749 &dev_attr_rx_fifo_errors.attr, 750 &dev_attr_rx_missed_errors.attr, 751 &dev_attr_tx_aborted_errors.attr, 752 &dev_attr_tx_carrier_errors.attr, 753 &dev_attr_tx_fifo_errors.attr, 754 &dev_attr_tx_heartbeat_errors.attr, 755 &dev_attr_tx_window_errors.attr, 756 &dev_attr_rx_compressed.attr, 757 &dev_attr_tx_compressed.attr, 758 &dev_attr_rx_nohandler.attr, 759 NULL 760 }; 761 762 static const struct attribute_group netstat_group = { 763 .name = "statistics", 764 .attrs = netstat_attrs, 765 }; 766 767 static struct attribute *wireless_attrs[] = { 768 NULL 769 }; 770 771 static const struct attribute_group wireless_group = { 772 .name = "wireless", 773 .attrs = wireless_attrs, 774 }; 775 776 static bool wireless_group_needed(struct net_device *ndev) 777 { 778 #if IS_ENABLED(CONFIG_CFG80211) 779 if (ndev->ieee80211_ptr) 780 return true; 781 #endif 782 #if IS_ENABLED(CONFIG_WIRELESS_EXT) 783 if (ndev->wireless_handlers) 784 return true; 785 #endif 786 return false; 787 } 788 789 #else /* CONFIG_SYSFS */ 790 #define net_class_groups NULL 791 #endif /* CONFIG_SYSFS */ 792 793 #ifdef CONFIG_SYSFS 794 #define to_rx_queue_attr(_attr) \ 795 container_of(_attr, struct rx_queue_attribute, attr) 796 797 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj) 798 799 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr, 800 char *buf) 801 { 802 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr); 803 struct netdev_rx_queue *queue = to_rx_queue(kobj); 804 805 if (!attribute->show) 806 return -EIO; 807 808 return attribute->show(queue, buf); 809 } 810 811 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr, 812 const char *buf, size_t count) 813 { 814 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr); 815 struct netdev_rx_queue *queue = to_rx_queue(kobj); 816 817 if (!attribute->store) 818 return -EIO; 819 820 return attribute->store(queue, buf, count); 821 } 822 823 static const struct sysfs_ops rx_queue_sysfs_ops = { 824 .show = rx_queue_attr_show, 825 .store = rx_queue_attr_store, 826 }; 827 828 #ifdef CONFIG_RPS 829 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf) 830 { 831 struct rps_map *map; 832 cpumask_var_t mask; 833 int i, len; 834 835 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 836 return -ENOMEM; 837 838 rcu_read_lock(); 839 map = rcu_dereference(queue->rps_map); 840 if (map) 841 for (i = 0; i < map->len; i++) 842 cpumask_set_cpu(map->cpus[i], mask); 843 844 len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask)); 845 rcu_read_unlock(); 846 free_cpumask_var(mask); 847 848 return len < PAGE_SIZE ? len : -EINVAL; 849 } 850 851 static int netdev_rx_queue_set_rps_mask(struct netdev_rx_queue *queue, 852 cpumask_var_t mask) 853 { 854 static DEFINE_MUTEX(rps_map_mutex); 855 struct rps_map *old_map, *map; 856 int cpu, i; 857 858 map = kzalloc(max_t(unsigned int, 859 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES), 860 GFP_KERNEL); 861 if (!map) 862 return -ENOMEM; 863 864 i = 0; 865 for_each_cpu_and(cpu, mask, cpu_online_mask) 866 map->cpus[i++] = cpu; 867 868 if (i) { 869 map->len = i; 870 } else { 871 kfree(map); 872 map = NULL; 873 } 874 875 mutex_lock(&rps_map_mutex); 876 old_map = rcu_dereference_protected(queue->rps_map, 877 mutex_is_locked(&rps_map_mutex)); 878 rcu_assign_pointer(queue->rps_map, map); 879 880 if (map) 881 static_branch_inc(&rps_needed); 882 if (old_map) 883 static_branch_dec(&rps_needed); 884 885 mutex_unlock(&rps_map_mutex); 886 887 if (old_map) 888 kfree_rcu(old_map, rcu); 889 return 0; 890 } 891 892 int rps_cpumask_housekeeping(struct cpumask *mask) 893 { 894 if (!cpumask_empty(mask)) { 895 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN)); 896 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ)); 897 if (cpumask_empty(mask)) 898 return -EINVAL; 899 } 900 return 0; 901 } 902 903 static ssize_t store_rps_map(struct netdev_rx_queue *queue, 904 const char *buf, size_t len) 905 { 906 cpumask_var_t mask; 907 int err; 908 909 if (!capable(CAP_NET_ADMIN)) 910 return -EPERM; 911 912 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 913 return -ENOMEM; 914 915 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits); 916 if (err) 917 goto out; 918 919 err = rps_cpumask_housekeeping(mask); 920 if (err) 921 goto out; 922 923 err = netdev_rx_queue_set_rps_mask(queue, mask); 924 925 out: 926 free_cpumask_var(mask); 927 return err ? : len; 928 } 929 930 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue, 931 char *buf) 932 { 933 struct rps_dev_flow_table *flow_table; 934 unsigned long val = 0; 935 936 rcu_read_lock(); 937 flow_table = rcu_dereference(queue->rps_flow_table); 938 if (flow_table) 939 val = (unsigned long)flow_table->mask + 1; 940 rcu_read_unlock(); 941 942 return sysfs_emit(buf, "%lu\n", val); 943 } 944 945 static void rps_dev_flow_table_release(struct rcu_head *rcu) 946 { 947 struct rps_dev_flow_table *table = container_of(rcu, 948 struct rps_dev_flow_table, rcu); 949 vfree(table); 950 } 951 952 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue, 953 const char *buf, size_t len) 954 { 955 unsigned long mask, count; 956 struct rps_dev_flow_table *table, *old_table; 957 static DEFINE_SPINLOCK(rps_dev_flow_lock); 958 int rc; 959 960 if (!capable(CAP_NET_ADMIN)) 961 return -EPERM; 962 963 rc = kstrtoul(buf, 0, &count); 964 if (rc < 0) 965 return rc; 966 967 if (count) { 968 mask = count - 1; 969 /* mask = roundup_pow_of_two(count) - 1; 970 * without overflows... 971 */ 972 while ((mask | (mask >> 1)) != mask) 973 mask |= (mask >> 1); 974 /* On 64 bit arches, must check mask fits in table->mask (u32), 975 * and on 32bit arches, must check 976 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow. 977 */ 978 #if BITS_PER_LONG > 32 979 if (mask > (unsigned long)(u32)mask) 980 return -EINVAL; 981 #else 982 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1)) 983 / sizeof(struct rps_dev_flow)) { 984 /* Enforce a limit to prevent overflow */ 985 return -EINVAL; 986 } 987 #endif 988 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1)); 989 if (!table) 990 return -ENOMEM; 991 992 table->mask = mask; 993 for (count = 0; count <= mask; count++) 994 table->flows[count].cpu = RPS_NO_CPU; 995 } else { 996 table = NULL; 997 } 998 999 spin_lock(&rps_dev_flow_lock); 1000 old_table = rcu_dereference_protected(queue->rps_flow_table, 1001 lockdep_is_held(&rps_dev_flow_lock)); 1002 rcu_assign_pointer(queue->rps_flow_table, table); 1003 spin_unlock(&rps_dev_flow_lock); 1004 1005 if (old_table) 1006 call_rcu(&old_table->rcu, rps_dev_flow_table_release); 1007 1008 return len; 1009 } 1010 1011 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init 1012 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map); 1013 1014 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init 1015 = __ATTR(rps_flow_cnt, 0644, 1016 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt); 1017 #endif /* CONFIG_RPS */ 1018 1019 static struct attribute *rx_queue_default_attrs[] __ro_after_init = { 1020 #ifdef CONFIG_RPS 1021 &rps_cpus_attribute.attr, 1022 &rps_dev_flow_table_cnt_attribute.attr, 1023 #endif 1024 NULL 1025 }; 1026 ATTRIBUTE_GROUPS(rx_queue_default); 1027 1028 static void rx_queue_release(struct kobject *kobj) 1029 { 1030 struct netdev_rx_queue *queue = to_rx_queue(kobj); 1031 #ifdef CONFIG_RPS 1032 struct rps_map *map; 1033 struct rps_dev_flow_table *flow_table; 1034 1035 map = rcu_dereference_protected(queue->rps_map, 1); 1036 if (map) { 1037 RCU_INIT_POINTER(queue->rps_map, NULL); 1038 kfree_rcu(map, rcu); 1039 } 1040 1041 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1); 1042 if (flow_table) { 1043 RCU_INIT_POINTER(queue->rps_flow_table, NULL); 1044 call_rcu(&flow_table->rcu, rps_dev_flow_table_release); 1045 } 1046 #endif 1047 1048 memset(kobj, 0, sizeof(*kobj)); 1049 netdev_put(queue->dev, &queue->dev_tracker); 1050 } 1051 1052 static const void *rx_queue_namespace(const struct kobject *kobj) 1053 { 1054 struct netdev_rx_queue *queue = to_rx_queue(kobj); 1055 struct device *dev = &queue->dev->dev; 1056 const void *ns = NULL; 1057 1058 if (dev->class && dev->class->ns_type) 1059 ns = dev->class->namespace(dev); 1060 1061 return ns; 1062 } 1063 1064 static void rx_queue_get_ownership(const struct kobject *kobj, 1065 kuid_t *uid, kgid_t *gid) 1066 { 1067 const struct net *net = rx_queue_namespace(kobj); 1068 1069 net_ns_get_ownership(net, uid, gid); 1070 } 1071 1072 static const struct kobj_type rx_queue_ktype = { 1073 .sysfs_ops = &rx_queue_sysfs_ops, 1074 .release = rx_queue_release, 1075 .default_groups = rx_queue_default_groups, 1076 .namespace = rx_queue_namespace, 1077 .get_ownership = rx_queue_get_ownership, 1078 }; 1079 1080 static int rx_queue_default_mask(struct net_device *dev, 1081 struct netdev_rx_queue *queue) 1082 { 1083 #if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL) 1084 struct cpumask *rps_default_mask = READ_ONCE(dev_net(dev)->core.rps_default_mask); 1085 1086 if (rps_default_mask && !cpumask_empty(rps_default_mask)) 1087 return netdev_rx_queue_set_rps_mask(queue, rps_default_mask); 1088 #endif 1089 return 0; 1090 } 1091 1092 static int rx_queue_add_kobject(struct net_device *dev, int index) 1093 { 1094 struct netdev_rx_queue *queue = dev->_rx + index; 1095 struct kobject *kobj = &queue->kobj; 1096 int error = 0; 1097 1098 /* Kobject_put later will trigger rx_queue_release call which 1099 * decreases dev refcount: Take that reference here 1100 */ 1101 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL); 1102 1103 kobj->kset = dev->queues_kset; 1104 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL, 1105 "rx-%u", index); 1106 if (error) 1107 goto err; 1108 1109 if (dev->sysfs_rx_queue_group) { 1110 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group); 1111 if (error) 1112 goto err; 1113 } 1114 1115 error = rx_queue_default_mask(dev, queue); 1116 if (error) 1117 goto err; 1118 1119 kobject_uevent(kobj, KOBJ_ADD); 1120 1121 return error; 1122 1123 err: 1124 kobject_put(kobj); 1125 return error; 1126 } 1127 1128 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid, 1129 kgid_t kgid) 1130 { 1131 struct netdev_rx_queue *queue = dev->_rx + index; 1132 struct kobject *kobj = &queue->kobj; 1133 int error; 1134 1135 error = sysfs_change_owner(kobj, kuid, kgid); 1136 if (error) 1137 return error; 1138 1139 if (dev->sysfs_rx_queue_group) 1140 error = sysfs_group_change_owner( 1141 kobj, dev->sysfs_rx_queue_group, kuid, kgid); 1142 1143 return error; 1144 } 1145 #endif /* CONFIG_SYSFS */ 1146 1147 int 1148 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num) 1149 { 1150 #ifdef CONFIG_SYSFS 1151 int i; 1152 int error = 0; 1153 1154 #ifndef CONFIG_RPS 1155 if (!dev->sysfs_rx_queue_group) 1156 return 0; 1157 #endif 1158 for (i = old_num; i < new_num; i++) { 1159 error = rx_queue_add_kobject(dev, i); 1160 if (error) { 1161 new_num = old_num; 1162 break; 1163 } 1164 } 1165 1166 while (--i >= new_num) { 1167 struct kobject *kobj = &dev->_rx[i].kobj; 1168 1169 if (!refcount_read(&dev_net(dev)->ns.count)) 1170 kobj->uevent_suppress = 1; 1171 if (dev->sysfs_rx_queue_group) 1172 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group); 1173 kobject_put(kobj); 1174 } 1175 1176 return error; 1177 #else 1178 return 0; 1179 #endif 1180 } 1181 1182 static int net_rx_queue_change_owner(struct net_device *dev, int num, 1183 kuid_t kuid, kgid_t kgid) 1184 { 1185 #ifdef CONFIG_SYSFS 1186 int error = 0; 1187 int i; 1188 1189 #ifndef CONFIG_RPS 1190 if (!dev->sysfs_rx_queue_group) 1191 return 0; 1192 #endif 1193 for (i = 0; i < num; i++) { 1194 error = rx_queue_change_owner(dev, i, kuid, kgid); 1195 if (error) 1196 break; 1197 } 1198 1199 return error; 1200 #else 1201 return 0; 1202 #endif 1203 } 1204 1205 #ifdef CONFIG_SYSFS 1206 /* 1207 * netdev_queue sysfs structures and functions. 1208 */ 1209 struct netdev_queue_attribute { 1210 struct attribute attr; 1211 ssize_t (*show)(struct netdev_queue *queue, char *buf); 1212 ssize_t (*store)(struct netdev_queue *queue, 1213 const char *buf, size_t len); 1214 }; 1215 #define to_netdev_queue_attr(_attr) \ 1216 container_of(_attr, struct netdev_queue_attribute, attr) 1217 1218 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj) 1219 1220 static ssize_t netdev_queue_attr_show(struct kobject *kobj, 1221 struct attribute *attr, char *buf) 1222 { 1223 const struct netdev_queue_attribute *attribute 1224 = to_netdev_queue_attr(attr); 1225 struct netdev_queue *queue = to_netdev_queue(kobj); 1226 1227 if (!attribute->show) 1228 return -EIO; 1229 1230 return attribute->show(queue, buf); 1231 } 1232 1233 static ssize_t netdev_queue_attr_store(struct kobject *kobj, 1234 struct attribute *attr, 1235 const char *buf, size_t count) 1236 { 1237 const struct netdev_queue_attribute *attribute 1238 = to_netdev_queue_attr(attr); 1239 struct netdev_queue *queue = to_netdev_queue(kobj); 1240 1241 if (!attribute->store) 1242 return -EIO; 1243 1244 return attribute->store(queue, buf, count); 1245 } 1246 1247 static const struct sysfs_ops netdev_queue_sysfs_ops = { 1248 .show = netdev_queue_attr_show, 1249 .store = netdev_queue_attr_store, 1250 }; 1251 1252 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf) 1253 { 1254 unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout); 1255 1256 return sysfs_emit(buf, fmt_ulong, trans_timeout); 1257 } 1258 1259 static unsigned int get_netdev_queue_index(struct netdev_queue *queue) 1260 { 1261 struct net_device *dev = queue->dev; 1262 unsigned int i; 1263 1264 i = queue - dev->_tx; 1265 BUG_ON(i >= dev->num_tx_queues); 1266 1267 return i; 1268 } 1269 1270 static ssize_t traffic_class_show(struct netdev_queue *queue, 1271 char *buf) 1272 { 1273 struct net_device *dev = queue->dev; 1274 int num_tc, tc; 1275 int index; 1276 1277 if (!netif_is_multiqueue(dev)) 1278 return -ENOENT; 1279 1280 if (!rtnl_trylock()) 1281 return restart_syscall(); 1282 1283 index = get_netdev_queue_index(queue); 1284 1285 /* If queue belongs to subordinate dev use its TC mapping */ 1286 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 1287 1288 num_tc = dev->num_tc; 1289 tc = netdev_txq_to_tc(dev, index); 1290 1291 rtnl_unlock(); 1292 1293 if (tc < 0) 1294 return -EINVAL; 1295 1296 /* We can report the traffic class one of two ways: 1297 * Subordinate device traffic classes are reported with the traffic 1298 * class first, and then the subordinate class so for example TC0 on 1299 * subordinate device 2 will be reported as "0-2". If the queue 1300 * belongs to the root device it will be reported with just the 1301 * traffic class, so just "0" for TC 0 for example. 1302 */ 1303 return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) : 1304 sysfs_emit(buf, "%d\n", tc); 1305 } 1306 1307 #ifdef CONFIG_XPS 1308 static ssize_t tx_maxrate_show(struct netdev_queue *queue, 1309 char *buf) 1310 { 1311 return sysfs_emit(buf, "%lu\n", queue->tx_maxrate); 1312 } 1313 1314 static ssize_t tx_maxrate_store(struct netdev_queue *queue, 1315 const char *buf, size_t len) 1316 { 1317 struct net_device *dev = queue->dev; 1318 int err, index = get_netdev_queue_index(queue); 1319 u32 rate = 0; 1320 1321 if (!capable(CAP_NET_ADMIN)) 1322 return -EPERM; 1323 1324 /* The check is also done later; this helps returning early without 1325 * hitting the trylock/restart below. 1326 */ 1327 if (!dev->netdev_ops->ndo_set_tx_maxrate) 1328 return -EOPNOTSUPP; 1329 1330 err = kstrtou32(buf, 10, &rate); 1331 if (err < 0) 1332 return err; 1333 1334 if (!rtnl_trylock()) 1335 return restart_syscall(); 1336 1337 err = -EOPNOTSUPP; 1338 if (dev->netdev_ops->ndo_set_tx_maxrate) 1339 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate); 1340 1341 rtnl_unlock(); 1342 if (!err) { 1343 queue->tx_maxrate = rate; 1344 return len; 1345 } 1346 return err; 1347 } 1348 1349 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init 1350 = __ATTR_RW(tx_maxrate); 1351 #endif 1352 1353 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init 1354 = __ATTR_RO(tx_timeout); 1355 1356 static struct netdev_queue_attribute queue_traffic_class __ro_after_init 1357 = __ATTR_RO(traffic_class); 1358 1359 #ifdef CONFIG_BQL 1360 /* 1361 * Byte queue limits sysfs structures and functions. 1362 */ 1363 static ssize_t bql_show(char *buf, unsigned int value) 1364 { 1365 return sysfs_emit(buf, "%u\n", value); 1366 } 1367 1368 static ssize_t bql_set(const char *buf, const size_t count, 1369 unsigned int *pvalue) 1370 { 1371 unsigned int value; 1372 int err; 1373 1374 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) { 1375 value = DQL_MAX_LIMIT; 1376 } else { 1377 err = kstrtouint(buf, 10, &value); 1378 if (err < 0) 1379 return err; 1380 if (value > DQL_MAX_LIMIT) 1381 return -EINVAL; 1382 } 1383 1384 *pvalue = value; 1385 1386 return count; 1387 } 1388 1389 static ssize_t bql_show_hold_time(struct netdev_queue *queue, 1390 char *buf) 1391 { 1392 struct dql *dql = &queue->dql; 1393 1394 return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time)); 1395 } 1396 1397 static ssize_t bql_set_hold_time(struct netdev_queue *queue, 1398 const char *buf, size_t len) 1399 { 1400 struct dql *dql = &queue->dql; 1401 unsigned int value; 1402 int err; 1403 1404 err = kstrtouint(buf, 10, &value); 1405 if (err < 0) 1406 return err; 1407 1408 dql->slack_hold_time = msecs_to_jiffies(value); 1409 1410 return len; 1411 } 1412 1413 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init 1414 = __ATTR(hold_time, 0644, 1415 bql_show_hold_time, bql_set_hold_time); 1416 1417 static ssize_t bql_show_inflight(struct netdev_queue *queue, 1418 char *buf) 1419 { 1420 struct dql *dql = &queue->dql; 1421 1422 return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed); 1423 } 1424 1425 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init = 1426 __ATTR(inflight, 0444, bql_show_inflight, NULL); 1427 1428 #define BQL_ATTR(NAME, FIELD) \ 1429 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \ 1430 char *buf) \ 1431 { \ 1432 return bql_show(buf, queue->dql.FIELD); \ 1433 } \ 1434 \ 1435 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \ 1436 const char *buf, size_t len) \ 1437 { \ 1438 return bql_set(buf, len, &queue->dql.FIELD); \ 1439 } \ 1440 \ 1441 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \ 1442 = __ATTR(NAME, 0644, \ 1443 bql_show_ ## NAME, bql_set_ ## NAME) 1444 1445 BQL_ATTR(limit, limit); 1446 BQL_ATTR(limit_max, max_limit); 1447 BQL_ATTR(limit_min, min_limit); 1448 1449 static struct attribute *dql_attrs[] __ro_after_init = { 1450 &bql_limit_attribute.attr, 1451 &bql_limit_max_attribute.attr, 1452 &bql_limit_min_attribute.attr, 1453 &bql_hold_time_attribute.attr, 1454 &bql_inflight_attribute.attr, 1455 NULL 1456 }; 1457 1458 static const struct attribute_group dql_group = { 1459 .name = "byte_queue_limits", 1460 .attrs = dql_attrs, 1461 }; 1462 #else 1463 /* Fake declaration, all the code using it should be dead */ 1464 extern const struct attribute_group dql_group; 1465 #endif /* CONFIG_BQL */ 1466 1467 #ifdef CONFIG_XPS 1468 static ssize_t xps_queue_show(struct net_device *dev, unsigned int index, 1469 int tc, char *buf, enum xps_map_type type) 1470 { 1471 struct xps_dev_maps *dev_maps; 1472 unsigned long *mask; 1473 unsigned int nr_ids; 1474 int j, len; 1475 1476 rcu_read_lock(); 1477 dev_maps = rcu_dereference(dev->xps_maps[type]); 1478 1479 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0 1480 * when dev_maps hasn't been allocated yet, to be backward compatible. 1481 */ 1482 nr_ids = dev_maps ? dev_maps->nr_ids : 1483 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues); 1484 1485 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT); 1486 if (!mask) { 1487 rcu_read_unlock(); 1488 return -ENOMEM; 1489 } 1490 1491 if (!dev_maps || tc >= dev_maps->num_tc) 1492 goto out_no_maps; 1493 1494 for (j = 0; j < nr_ids; j++) { 1495 int i, tci = j * dev_maps->num_tc + tc; 1496 struct xps_map *map; 1497 1498 map = rcu_dereference(dev_maps->attr_map[tci]); 1499 if (!map) 1500 continue; 1501 1502 for (i = map->len; i--;) { 1503 if (map->queues[i] == index) { 1504 __set_bit(j, mask); 1505 break; 1506 } 1507 } 1508 } 1509 out_no_maps: 1510 rcu_read_unlock(); 1511 1512 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids); 1513 bitmap_free(mask); 1514 1515 return len < PAGE_SIZE ? len : -EINVAL; 1516 } 1517 1518 static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf) 1519 { 1520 struct net_device *dev = queue->dev; 1521 unsigned int index; 1522 int len, tc; 1523 1524 if (!netif_is_multiqueue(dev)) 1525 return -ENOENT; 1526 1527 index = get_netdev_queue_index(queue); 1528 1529 if (!rtnl_trylock()) 1530 return restart_syscall(); 1531 1532 /* If queue belongs to subordinate dev use its map */ 1533 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 1534 1535 tc = netdev_txq_to_tc(dev, index); 1536 if (tc < 0) { 1537 rtnl_unlock(); 1538 return -EINVAL; 1539 } 1540 1541 /* Make sure the subordinate device can't be freed */ 1542 get_device(&dev->dev); 1543 rtnl_unlock(); 1544 1545 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS); 1546 1547 put_device(&dev->dev); 1548 return len; 1549 } 1550 1551 static ssize_t xps_cpus_store(struct netdev_queue *queue, 1552 const char *buf, size_t len) 1553 { 1554 struct net_device *dev = queue->dev; 1555 unsigned int index; 1556 cpumask_var_t mask; 1557 int err; 1558 1559 if (!netif_is_multiqueue(dev)) 1560 return -ENOENT; 1561 1562 if (!capable(CAP_NET_ADMIN)) 1563 return -EPERM; 1564 1565 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 1566 return -ENOMEM; 1567 1568 index = get_netdev_queue_index(queue); 1569 1570 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits); 1571 if (err) { 1572 free_cpumask_var(mask); 1573 return err; 1574 } 1575 1576 if (!rtnl_trylock()) { 1577 free_cpumask_var(mask); 1578 return restart_syscall(); 1579 } 1580 1581 err = netif_set_xps_queue(dev, mask, index); 1582 rtnl_unlock(); 1583 1584 free_cpumask_var(mask); 1585 1586 return err ? : len; 1587 } 1588 1589 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init 1590 = __ATTR_RW(xps_cpus); 1591 1592 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf) 1593 { 1594 struct net_device *dev = queue->dev; 1595 unsigned int index; 1596 int tc; 1597 1598 index = get_netdev_queue_index(queue); 1599 1600 if (!rtnl_trylock()) 1601 return restart_syscall(); 1602 1603 tc = netdev_txq_to_tc(dev, index); 1604 rtnl_unlock(); 1605 if (tc < 0) 1606 return -EINVAL; 1607 1608 return xps_queue_show(dev, index, tc, buf, XPS_RXQS); 1609 } 1610 1611 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf, 1612 size_t len) 1613 { 1614 struct net_device *dev = queue->dev; 1615 struct net *net = dev_net(dev); 1616 unsigned long *mask; 1617 unsigned int index; 1618 int err; 1619 1620 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 1621 return -EPERM; 1622 1623 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL); 1624 if (!mask) 1625 return -ENOMEM; 1626 1627 index = get_netdev_queue_index(queue); 1628 1629 err = bitmap_parse(buf, len, mask, dev->num_rx_queues); 1630 if (err) { 1631 bitmap_free(mask); 1632 return err; 1633 } 1634 1635 if (!rtnl_trylock()) { 1636 bitmap_free(mask); 1637 return restart_syscall(); 1638 } 1639 1640 cpus_read_lock(); 1641 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS); 1642 cpus_read_unlock(); 1643 1644 rtnl_unlock(); 1645 1646 bitmap_free(mask); 1647 return err ? : len; 1648 } 1649 1650 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init 1651 = __ATTR_RW(xps_rxqs); 1652 #endif /* CONFIG_XPS */ 1653 1654 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = { 1655 &queue_trans_timeout.attr, 1656 &queue_traffic_class.attr, 1657 #ifdef CONFIG_XPS 1658 &xps_cpus_attribute.attr, 1659 &xps_rxqs_attribute.attr, 1660 &queue_tx_maxrate.attr, 1661 #endif 1662 NULL 1663 }; 1664 ATTRIBUTE_GROUPS(netdev_queue_default); 1665 1666 static void netdev_queue_release(struct kobject *kobj) 1667 { 1668 struct netdev_queue *queue = to_netdev_queue(kobj); 1669 1670 memset(kobj, 0, sizeof(*kobj)); 1671 netdev_put(queue->dev, &queue->dev_tracker); 1672 } 1673 1674 static const void *netdev_queue_namespace(const struct kobject *kobj) 1675 { 1676 struct netdev_queue *queue = to_netdev_queue(kobj); 1677 struct device *dev = &queue->dev->dev; 1678 const void *ns = NULL; 1679 1680 if (dev->class && dev->class->ns_type) 1681 ns = dev->class->namespace(dev); 1682 1683 return ns; 1684 } 1685 1686 static void netdev_queue_get_ownership(const struct kobject *kobj, 1687 kuid_t *uid, kgid_t *gid) 1688 { 1689 const struct net *net = netdev_queue_namespace(kobj); 1690 1691 net_ns_get_ownership(net, uid, gid); 1692 } 1693 1694 static const struct kobj_type netdev_queue_ktype = { 1695 .sysfs_ops = &netdev_queue_sysfs_ops, 1696 .release = netdev_queue_release, 1697 .default_groups = netdev_queue_default_groups, 1698 .namespace = netdev_queue_namespace, 1699 .get_ownership = netdev_queue_get_ownership, 1700 }; 1701 1702 static bool netdev_uses_bql(const struct net_device *dev) 1703 { 1704 if (dev->features & NETIF_F_LLTX || 1705 dev->priv_flags & IFF_NO_QUEUE) 1706 return false; 1707 1708 return IS_ENABLED(CONFIG_BQL); 1709 } 1710 1711 static int netdev_queue_add_kobject(struct net_device *dev, int index) 1712 { 1713 struct netdev_queue *queue = dev->_tx + index; 1714 struct kobject *kobj = &queue->kobj; 1715 int error = 0; 1716 1717 /* Kobject_put later will trigger netdev_queue_release call 1718 * which decreases dev refcount: Take that reference here 1719 */ 1720 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL); 1721 1722 kobj->kset = dev->queues_kset; 1723 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL, 1724 "tx-%u", index); 1725 if (error) 1726 goto err; 1727 1728 if (netdev_uses_bql(dev)) { 1729 error = sysfs_create_group(kobj, &dql_group); 1730 if (error) 1731 goto err; 1732 } 1733 1734 kobject_uevent(kobj, KOBJ_ADD); 1735 return 0; 1736 1737 err: 1738 kobject_put(kobj); 1739 return error; 1740 } 1741 1742 static int tx_queue_change_owner(struct net_device *ndev, int index, 1743 kuid_t kuid, kgid_t kgid) 1744 { 1745 struct netdev_queue *queue = ndev->_tx + index; 1746 struct kobject *kobj = &queue->kobj; 1747 int error; 1748 1749 error = sysfs_change_owner(kobj, kuid, kgid); 1750 if (error) 1751 return error; 1752 1753 if (netdev_uses_bql(ndev)) 1754 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid); 1755 1756 return error; 1757 } 1758 #endif /* CONFIG_SYSFS */ 1759 1760 int 1761 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num) 1762 { 1763 #ifdef CONFIG_SYSFS 1764 int i; 1765 int error = 0; 1766 1767 /* Tx queue kobjects are allowed to be updated when a device is being 1768 * unregistered, but solely to remove queues from qdiscs. Any path 1769 * adding queues should be fixed. 1770 */ 1771 WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num, 1772 "New queues can't be registered after device unregistration."); 1773 1774 for (i = old_num; i < new_num; i++) { 1775 error = netdev_queue_add_kobject(dev, i); 1776 if (error) { 1777 new_num = old_num; 1778 break; 1779 } 1780 } 1781 1782 while (--i >= new_num) { 1783 struct netdev_queue *queue = dev->_tx + i; 1784 1785 if (!refcount_read(&dev_net(dev)->ns.count)) 1786 queue->kobj.uevent_suppress = 1; 1787 1788 if (netdev_uses_bql(dev)) 1789 sysfs_remove_group(&queue->kobj, &dql_group); 1790 1791 kobject_put(&queue->kobj); 1792 } 1793 1794 return error; 1795 #else 1796 return 0; 1797 #endif /* CONFIG_SYSFS */ 1798 } 1799 1800 static int net_tx_queue_change_owner(struct net_device *dev, int num, 1801 kuid_t kuid, kgid_t kgid) 1802 { 1803 #ifdef CONFIG_SYSFS 1804 int error = 0; 1805 int i; 1806 1807 for (i = 0; i < num; i++) { 1808 error = tx_queue_change_owner(dev, i, kuid, kgid); 1809 if (error) 1810 break; 1811 } 1812 1813 return error; 1814 #else 1815 return 0; 1816 #endif /* CONFIG_SYSFS */ 1817 } 1818 1819 static int register_queue_kobjects(struct net_device *dev) 1820 { 1821 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0; 1822 1823 #ifdef CONFIG_SYSFS 1824 dev->queues_kset = kset_create_and_add("queues", 1825 NULL, &dev->dev.kobj); 1826 if (!dev->queues_kset) 1827 return -ENOMEM; 1828 real_rx = dev->real_num_rx_queues; 1829 #endif 1830 real_tx = dev->real_num_tx_queues; 1831 1832 error = net_rx_queue_update_kobjects(dev, 0, real_rx); 1833 if (error) 1834 goto error; 1835 rxq = real_rx; 1836 1837 error = netdev_queue_update_kobjects(dev, 0, real_tx); 1838 if (error) 1839 goto error; 1840 txq = real_tx; 1841 1842 return 0; 1843 1844 error: 1845 netdev_queue_update_kobjects(dev, txq, 0); 1846 net_rx_queue_update_kobjects(dev, rxq, 0); 1847 #ifdef CONFIG_SYSFS 1848 kset_unregister(dev->queues_kset); 1849 #endif 1850 return error; 1851 } 1852 1853 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid) 1854 { 1855 int error = 0, real_rx = 0, real_tx = 0; 1856 1857 #ifdef CONFIG_SYSFS 1858 if (ndev->queues_kset) { 1859 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid); 1860 if (error) 1861 return error; 1862 } 1863 real_rx = ndev->real_num_rx_queues; 1864 #endif 1865 real_tx = ndev->real_num_tx_queues; 1866 1867 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid); 1868 if (error) 1869 return error; 1870 1871 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid); 1872 if (error) 1873 return error; 1874 1875 return 0; 1876 } 1877 1878 static void remove_queue_kobjects(struct net_device *dev) 1879 { 1880 int real_rx = 0, real_tx = 0; 1881 1882 #ifdef CONFIG_SYSFS 1883 real_rx = dev->real_num_rx_queues; 1884 #endif 1885 real_tx = dev->real_num_tx_queues; 1886 1887 net_rx_queue_update_kobjects(dev, real_rx, 0); 1888 netdev_queue_update_kobjects(dev, real_tx, 0); 1889 1890 dev->real_num_rx_queues = 0; 1891 dev->real_num_tx_queues = 0; 1892 #ifdef CONFIG_SYSFS 1893 kset_unregister(dev->queues_kset); 1894 #endif 1895 } 1896 1897 static bool net_current_may_mount(void) 1898 { 1899 struct net *net = current->nsproxy->net_ns; 1900 1901 return ns_capable(net->user_ns, CAP_SYS_ADMIN); 1902 } 1903 1904 static void *net_grab_current_ns(void) 1905 { 1906 struct net *ns = current->nsproxy->net_ns; 1907 #ifdef CONFIG_NET_NS 1908 if (ns) 1909 refcount_inc(&ns->passive); 1910 #endif 1911 return ns; 1912 } 1913 1914 static const void *net_initial_ns(void) 1915 { 1916 return &init_net; 1917 } 1918 1919 static const void *net_netlink_ns(struct sock *sk) 1920 { 1921 return sock_net(sk); 1922 } 1923 1924 const struct kobj_ns_type_operations net_ns_type_operations = { 1925 .type = KOBJ_NS_TYPE_NET, 1926 .current_may_mount = net_current_may_mount, 1927 .grab_current_ns = net_grab_current_ns, 1928 .netlink_ns = net_netlink_ns, 1929 .initial_ns = net_initial_ns, 1930 .drop_ns = net_drop_ns, 1931 }; 1932 EXPORT_SYMBOL_GPL(net_ns_type_operations); 1933 1934 static int netdev_uevent(const struct device *d, struct kobj_uevent_env *env) 1935 { 1936 const struct net_device *dev = to_net_dev(d); 1937 int retval; 1938 1939 /* pass interface to uevent. */ 1940 retval = add_uevent_var(env, "INTERFACE=%s", dev->name); 1941 if (retval) 1942 goto exit; 1943 1944 /* pass ifindex to uevent. 1945 * ifindex is useful as it won't change (interface name may change) 1946 * and is what RtNetlink uses natively. 1947 */ 1948 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex); 1949 1950 exit: 1951 return retval; 1952 } 1953 1954 /* 1955 * netdev_release -- destroy and free a dead device. 1956 * Called when last reference to device kobject is gone. 1957 */ 1958 static void netdev_release(struct device *d) 1959 { 1960 struct net_device *dev = to_net_dev(d); 1961 1962 BUG_ON(dev->reg_state != NETREG_RELEASED); 1963 1964 /* no need to wait for rcu grace period: 1965 * device is dead and about to be freed. 1966 */ 1967 kfree(rcu_access_pointer(dev->ifalias)); 1968 netdev_freemem(dev); 1969 } 1970 1971 static const void *net_namespace(const struct device *d) 1972 { 1973 const struct net_device *dev = to_net_dev(d); 1974 1975 return dev_net(dev); 1976 } 1977 1978 static void net_get_ownership(const struct device *d, kuid_t *uid, kgid_t *gid) 1979 { 1980 const struct net_device *dev = to_net_dev(d); 1981 const struct net *net = dev_net(dev); 1982 1983 net_ns_get_ownership(net, uid, gid); 1984 } 1985 1986 static struct class net_class __ro_after_init = { 1987 .name = "net", 1988 .dev_release = netdev_release, 1989 .dev_groups = net_class_groups, 1990 .dev_uevent = netdev_uevent, 1991 .ns_type = &net_ns_type_operations, 1992 .namespace = net_namespace, 1993 .get_ownership = net_get_ownership, 1994 }; 1995 1996 #ifdef CONFIG_OF 1997 static int of_dev_node_match(struct device *dev, const void *data) 1998 { 1999 for (; dev; dev = dev->parent) { 2000 if (dev->of_node == data) 2001 return 1; 2002 } 2003 2004 return 0; 2005 } 2006 2007 /* 2008 * of_find_net_device_by_node - lookup the net device for the device node 2009 * @np: OF device node 2010 * 2011 * Looks up the net_device structure corresponding with the device node. 2012 * If successful, returns a pointer to the net_device with the embedded 2013 * struct device refcount incremented by one, or NULL on failure. The 2014 * refcount must be dropped when done with the net_device. 2015 */ 2016 struct net_device *of_find_net_device_by_node(struct device_node *np) 2017 { 2018 struct device *dev; 2019 2020 dev = class_find_device(&net_class, NULL, np, of_dev_node_match); 2021 if (!dev) 2022 return NULL; 2023 2024 return to_net_dev(dev); 2025 } 2026 EXPORT_SYMBOL(of_find_net_device_by_node); 2027 #endif 2028 2029 /* Delete sysfs entries but hold kobject reference until after all 2030 * netdev references are gone. 2031 */ 2032 void netdev_unregister_kobject(struct net_device *ndev) 2033 { 2034 struct device *dev = &ndev->dev; 2035 2036 if (!refcount_read(&dev_net(ndev)->ns.count)) 2037 dev_set_uevent_suppress(dev, 1); 2038 2039 kobject_get(&dev->kobj); 2040 2041 remove_queue_kobjects(ndev); 2042 2043 pm_runtime_set_memalloc_noio(dev, false); 2044 2045 device_del(dev); 2046 } 2047 2048 /* Create sysfs entries for network device. */ 2049 int netdev_register_kobject(struct net_device *ndev) 2050 { 2051 struct device *dev = &ndev->dev; 2052 const struct attribute_group **groups = ndev->sysfs_groups; 2053 int error = 0; 2054 2055 device_initialize(dev); 2056 dev->class = &net_class; 2057 dev->platform_data = ndev; 2058 dev->groups = groups; 2059 2060 dev_set_name(dev, "%s", ndev->name); 2061 2062 #ifdef CONFIG_SYSFS 2063 /* Allow for a device specific group */ 2064 if (*groups) 2065 groups++; 2066 2067 *groups++ = &netstat_group; 2068 2069 if (wireless_group_needed(ndev)) 2070 *groups++ = &wireless_group; 2071 #endif /* CONFIG_SYSFS */ 2072 2073 error = device_add(dev); 2074 if (error) 2075 return error; 2076 2077 error = register_queue_kobjects(ndev); 2078 if (error) { 2079 device_del(dev); 2080 return error; 2081 } 2082 2083 pm_runtime_set_memalloc_noio(dev, true); 2084 2085 return error; 2086 } 2087 2088 /* Change owner for sysfs entries when moving network devices across network 2089 * namespaces owned by different user namespaces. 2090 */ 2091 int netdev_change_owner(struct net_device *ndev, const struct net *net_old, 2092 const struct net *net_new) 2093 { 2094 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID; 2095 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID; 2096 struct device *dev = &ndev->dev; 2097 int error; 2098 2099 net_ns_get_ownership(net_old, &old_uid, &old_gid); 2100 net_ns_get_ownership(net_new, &new_uid, &new_gid); 2101 2102 /* The network namespace was changed but the owning user namespace is 2103 * identical so there's no need to change the owner of sysfs entries. 2104 */ 2105 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid)) 2106 return 0; 2107 2108 error = device_change_owner(dev, new_uid, new_gid); 2109 if (error) 2110 return error; 2111 2112 error = queue_change_owner(ndev, new_uid, new_gid); 2113 if (error) 2114 return error; 2115 2116 return 0; 2117 } 2118 2119 int netdev_class_create_file_ns(const struct class_attribute *class_attr, 2120 const void *ns) 2121 { 2122 return class_create_file_ns(&net_class, class_attr, ns); 2123 } 2124 EXPORT_SYMBOL(netdev_class_create_file_ns); 2125 2126 void netdev_class_remove_file_ns(const struct class_attribute *class_attr, 2127 const void *ns) 2128 { 2129 class_remove_file_ns(&net_class, class_attr, ns); 2130 } 2131 EXPORT_SYMBOL(netdev_class_remove_file_ns); 2132 2133 int __init netdev_kobject_init(void) 2134 { 2135 kobj_ns_type_register(&net_ns_type_operations); 2136 return class_register(&net_class); 2137 } 2138