1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * net-sysfs.c - network device class and attributes 4 * 5 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org> 6 */ 7 8 #include <linux/capability.h> 9 #include <linux/kernel.h> 10 #include <linux/netdevice.h> 11 #include <linux/if_arp.h> 12 #include <linux/slab.h> 13 #include <linux/sched/signal.h> 14 #include <linux/sched/isolation.h> 15 #include <linux/nsproxy.h> 16 #include <net/sock.h> 17 #include <net/net_namespace.h> 18 #include <linux/rtnetlink.h> 19 #include <linux/vmalloc.h> 20 #include <linux/export.h> 21 #include <linux/jiffies.h> 22 #include <linux/pm_runtime.h> 23 #include <linux/of.h> 24 #include <linux/of_net.h> 25 #include <linux/cpu.h> 26 27 #include "net-sysfs.h" 28 29 #ifdef CONFIG_SYSFS 30 static const char fmt_hex[] = "%#x\n"; 31 static const char fmt_dec[] = "%d\n"; 32 static const char fmt_ulong[] = "%lu\n"; 33 static const char fmt_u64[] = "%llu\n"; 34 35 static inline int dev_isalive(const struct net_device *dev) 36 { 37 return dev->reg_state <= NETREG_REGISTERED; 38 } 39 40 /* use same locking rules as GIF* ioctl's */ 41 static ssize_t netdev_show(const struct device *dev, 42 struct device_attribute *attr, char *buf, 43 ssize_t (*format)(const struct net_device *, char *)) 44 { 45 struct net_device *ndev = to_net_dev(dev); 46 ssize_t ret = -EINVAL; 47 48 read_lock(&dev_base_lock); 49 if (dev_isalive(ndev)) 50 ret = (*format)(ndev, buf); 51 read_unlock(&dev_base_lock); 52 53 return ret; 54 } 55 56 /* generate a show function for simple field */ 57 #define NETDEVICE_SHOW(field, format_string) \ 58 static ssize_t format_##field(const struct net_device *dev, char *buf) \ 59 { \ 60 return sprintf(buf, format_string, dev->field); \ 61 } \ 62 static ssize_t field##_show(struct device *dev, \ 63 struct device_attribute *attr, char *buf) \ 64 { \ 65 return netdev_show(dev, attr, buf, format_##field); \ 66 } \ 67 68 #define NETDEVICE_SHOW_RO(field, format_string) \ 69 NETDEVICE_SHOW(field, format_string); \ 70 static DEVICE_ATTR_RO(field) 71 72 #define NETDEVICE_SHOW_RW(field, format_string) \ 73 NETDEVICE_SHOW(field, format_string); \ 74 static DEVICE_ATTR_RW(field) 75 76 /* use same locking and permission rules as SIF* ioctl's */ 77 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr, 78 const char *buf, size_t len, 79 int (*set)(struct net_device *, unsigned long)) 80 { 81 struct net_device *netdev = to_net_dev(dev); 82 struct net *net = dev_net(netdev); 83 unsigned long new; 84 int ret; 85 86 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 87 return -EPERM; 88 89 ret = kstrtoul(buf, 0, &new); 90 if (ret) 91 goto err; 92 93 if (!rtnl_trylock()) 94 return restart_syscall(); 95 96 if (dev_isalive(netdev)) { 97 ret = (*set)(netdev, new); 98 if (ret == 0) 99 ret = len; 100 } 101 rtnl_unlock(); 102 err: 103 return ret; 104 } 105 106 NETDEVICE_SHOW_RO(dev_id, fmt_hex); 107 NETDEVICE_SHOW_RO(dev_port, fmt_dec); 108 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec); 109 NETDEVICE_SHOW_RO(addr_len, fmt_dec); 110 NETDEVICE_SHOW_RO(ifindex, fmt_dec); 111 NETDEVICE_SHOW_RO(type, fmt_dec); 112 NETDEVICE_SHOW_RO(link_mode, fmt_dec); 113 114 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr, 115 char *buf) 116 { 117 struct net_device *ndev = to_net_dev(dev); 118 119 return sprintf(buf, fmt_dec, dev_get_iflink(ndev)); 120 } 121 static DEVICE_ATTR_RO(iflink); 122 123 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf) 124 { 125 return sprintf(buf, fmt_dec, dev->name_assign_type); 126 } 127 128 static ssize_t name_assign_type_show(struct device *dev, 129 struct device_attribute *attr, 130 char *buf) 131 { 132 struct net_device *ndev = to_net_dev(dev); 133 ssize_t ret = -EINVAL; 134 135 if (ndev->name_assign_type != NET_NAME_UNKNOWN) 136 ret = netdev_show(dev, attr, buf, format_name_assign_type); 137 138 return ret; 139 } 140 static DEVICE_ATTR_RO(name_assign_type); 141 142 /* use same locking rules as GIFHWADDR ioctl's */ 143 static ssize_t address_show(struct device *dev, struct device_attribute *attr, 144 char *buf) 145 { 146 struct net_device *ndev = to_net_dev(dev); 147 ssize_t ret = -EINVAL; 148 149 read_lock(&dev_base_lock); 150 if (dev_isalive(ndev)) 151 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len); 152 read_unlock(&dev_base_lock); 153 return ret; 154 } 155 static DEVICE_ATTR_RO(address); 156 157 static ssize_t broadcast_show(struct device *dev, 158 struct device_attribute *attr, char *buf) 159 { 160 struct net_device *ndev = to_net_dev(dev); 161 162 if (dev_isalive(ndev)) 163 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len); 164 return -EINVAL; 165 } 166 static DEVICE_ATTR_RO(broadcast); 167 168 static int change_carrier(struct net_device *dev, unsigned long new_carrier) 169 { 170 if (!netif_running(dev)) 171 return -EINVAL; 172 return dev_change_carrier(dev, (bool)new_carrier); 173 } 174 175 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr, 176 const char *buf, size_t len) 177 { 178 return netdev_store(dev, attr, buf, len, change_carrier); 179 } 180 181 static ssize_t carrier_show(struct device *dev, 182 struct device_attribute *attr, char *buf) 183 { 184 struct net_device *netdev = to_net_dev(dev); 185 186 if (netif_running(netdev)) 187 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev)); 188 189 return -EINVAL; 190 } 191 static DEVICE_ATTR_RW(carrier); 192 193 static ssize_t speed_show(struct device *dev, 194 struct device_attribute *attr, char *buf) 195 { 196 struct net_device *netdev = to_net_dev(dev); 197 int ret = -EINVAL; 198 199 if (!rtnl_trylock()) 200 return restart_syscall(); 201 202 if (netif_running(netdev)) { 203 struct ethtool_link_ksettings cmd; 204 205 if (!__ethtool_get_link_ksettings(netdev, &cmd)) 206 ret = sprintf(buf, fmt_dec, cmd.base.speed); 207 } 208 rtnl_unlock(); 209 return ret; 210 } 211 static DEVICE_ATTR_RO(speed); 212 213 static ssize_t duplex_show(struct device *dev, 214 struct device_attribute *attr, char *buf) 215 { 216 struct net_device *netdev = to_net_dev(dev); 217 int ret = -EINVAL; 218 219 if (!rtnl_trylock()) 220 return restart_syscall(); 221 222 if (netif_running(netdev)) { 223 struct ethtool_link_ksettings cmd; 224 225 if (!__ethtool_get_link_ksettings(netdev, &cmd)) { 226 const char *duplex; 227 228 switch (cmd.base.duplex) { 229 case DUPLEX_HALF: 230 duplex = "half"; 231 break; 232 case DUPLEX_FULL: 233 duplex = "full"; 234 break; 235 default: 236 duplex = "unknown"; 237 break; 238 } 239 ret = sprintf(buf, "%s\n", duplex); 240 } 241 } 242 rtnl_unlock(); 243 return ret; 244 } 245 static DEVICE_ATTR_RO(duplex); 246 247 static ssize_t testing_show(struct device *dev, 248 struct device_attribute *attr, char *buf) 249 { 250 struct net_device *netdev = to_net_dev(dev); 251 252 if (netif_running(netdev)) 253 return sprintf(buf, fmt_dec, !!netif_testing(netdev)); 254 255 return -EINVAL; 256 } 257 static DEVICE_ATTR_RO(testing); 258 259 static ssize_t dormant_show(struct device *dev, 260 struct device_attribute *attr, char *buf) 261 { 262 struct net_device *netdev = to_net_dev(dev); 263 264 if (netif_running(netdev)) 265 return sprintf(buf, fmt_dec, !!netif_dormant(netdev)); 266 267 return -EINVAL; 268 } 269 static DEVICE_ATTR_RO(dormant); 270 271 static const char *const operstates[] = { 272 "unknown", 273 "notpresent", /* currently unused */ 274 "down", 275 "lowerlayerdown", 276 "testing", 277 "dormant", 278 "up" 279 }; 280 281 static ssize_t operstate_show(struct device *dev, 282 struct device_attribute *attr, char *buf) 283 { 284 const struct net_device *netdev = to_net_dev(dev); 285 unsigned char operstate; 286 287 read_lock(&dev_base_lock); 288 operstate = netdev->operstate; 289 if (!netif_running(netdev)) 290 operstate = IF_OPER_DOWN; 291 read_unlock(&dev_base_lock); 292 293 if (operstate >= ARRAY_SIZE(operstates)) 294 return -EINVAL; /* should not happen */ 295 296 return sprintf(buf, "%s\n", operstates[operstate]); 297 } 298 static DEVICE_ATTR_RO(operstate); 299 300 static ssize_t carrier_changes_show(struct device *dev, 301 struct device_attribute *attr, 302 char *buf) 303 { 304 struct net_device *netdev = to_net_dev(dev); 305 306 return sprintf(buf, fmt_dec, 307 atomic_read(&netdev->carrier_up_count) + 308 atomic_read(&netdev->carrier_down_count)); 309 } 310 static DEVICE_ATTR_RO(carrier_changes); 311 312 static ssize_t carrier_up_count_show(struct device *dev, 313 struct device_attribute *attr, 314 char *buf) 315 { 316 struct net_device *netdev = to_net_dev(dev); 317 318 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count)); 319 } 320 static DEVICE_ATTR_RO(carrier_up_count); 321 322 static ssize_t carrier_down_count_show(struct device *dev, 323 struct device_attribute *attr, 324 char *buf) 325 { 326 struct net_device *netdev = to_net_dev(dev); 327 328 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count)); 329 } 330 static DEVICE_ATTR_RO(carrier_down_count); 331 332 /* read-write attributes */ 333 334 static int change_mtu(struct net_device *dev, unsigned long new_mtu) 335 { 336 return dev_set_mtu(dev, (int)new_mtu); 337 } 338 339 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr, 340 const char *buf, size_t len) 341 { 342 return netdev_store(dev, attr, buf, len, change_mtu); 343 } 344 NETDEVICE_SHOW_RW(mtu, fmt_dec); 345 346 static int change_flags(struct net_device *dev, unsigned long new_flags) 347 { 348 return dev_change_flags(dev, (unsigned int)new_flags, NULL); 349 } 350 351 static ssize_t flags_store(struct device *dev, struct device_attribute *attr, 352 const char *buf, size_t len) 353 { 354 return netdev_store(dev, attr, buf, len, change_flags); 355 } 356 NETDEVICE_SHOW_RW(flags, fmt_hex); 357 358 static ssize_t tx_queue_len_store(struct device *dev, 359 struct device_attribute *attr, 360 const char *buf, size_t len) 361 { 362 if (!capable(CAP_NET_ADMIN)) 363 return -EPERM; 364 365 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len); 366 } 367 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec); 368 369 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val) 370 { 371 WRITE_ONCE(dev->gro_flush_timeout, val); 372 return 0; 373 } 374 375 static ssize_t gro_flush_timeout_store(struct device *dev, 376 struct device_attribute *attr, 377 const char *buf, size_t len) 378 { 379 if (!capable(CAP_NET_ADMIN)) 380 return -EPERM; 381 382 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout); 383 } 384 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong); 385 386 static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val) 387 { 388 WRITE_ONCE(dev->napi_defer_hard_irqs, val); 389 return 0; 390 } 391 392 static ssize_t napi_defer_hard_irqs_store(struct device *dev, 393 struct device_attribute *attr, 394 const char *buf, size_t len) 395 { 396 if (!capable(CAP_NET_ADMIN)) 397 return -EPERM; 398 399 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs); 400 } 401 NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec); 402 403 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr, 404 const char *buf, size_t len) 405 { 406 struct net_device *netdev = to_net_dev(dev); 407 struct net *net = dev_net(netdev); 408 size_t count = len; 409 ssize_t ret = 0; 410 411 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 412 return -EPERM; 413 414 /* ignore trailing newline */ 415 if (len > 0 && buf[len - 1] == '\n') 416 --count; 417 418 if (!rtnl_trylock()) 419 return restart_syscall(); 420 421 if (dev_isalive(netdev)) { 422 ret = dev_set_alias(netdev, buf, count); 423 if (ret < 0) 424 goto err; 425 ret = len; 426 netdev_state_change(netdev); 427 } 428 err: 429 rtnl_unlock(); 430 431 return ret; 432 } 433 434 static ssize_t ifalias_show(struct device *dev, 435 struct device_attribute *attr, char *buf) 436 { 437 const struct net_device *netdev = to_net_dev(dev); 438 char tmp[IFALIASZ]; 439 ssize_t ret = 0; 440 441 ret = dev_get_alias(netdev, tmp, sizeof(tmp)); 442 if (ret > 0) 443 ret = sprintf(buf, "%s\n", tmp); 444 return ret; 445 } 446 static DEVICE_ATTR_RW(ifalias); 447 448 static int change_group(struct net_device *dev, unsigned long new_group) 449 { 450 dev_set_group(dev, (int)new_group); 451 return 0; 452 } 453 454 static ssize_t group_store(struct device *dev, struct device_attribute *attr, 455 const char *buf, size_t len) 456 { 457 return netdev_store(dev, attr, buf, len, change_group); 458 } 459 NETDEVICE_SHOW(group, fmt_dec); 460 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store); 461 462 static int change_proto_down(struct net_device *dev, unsigned long proto_down) 463 { 464 return dev_change_proto_down(dev, (bool)proto_down); 465 } 466 467 static ssize_t proto_down_store(struct device *dev, 468 struct device_attribute *attr, 469 const char *buf, size_t len) 470 { 471 return netdev_store(dev, attr, buf, len, change_proto_down); 472 } 473 NETDEVICE_SHOW_RW(proto_down, fmt_dec); 474 475 static ssize_t phys_port_id_show(struct device *dev, 476 struct device_attribute *attr, char *buf) 477 { 478 struct net_device *netdev = to_net_dev(dev); 479 ssize_t ret = -EINVAL; 480 481 if (!rtnl_trylock()) 482 return restart_syscall(); 483 484 if (dev_isalive(netdev)) { 485 struct netdev_phys_item_id ppid; 486 487 ret = dev_get_phys_port_id(netdev, &ppid); 488 if (!ret) 489 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id); 490 } 491 rtnl_unlock(); 492 493 return ret; 494 } 495 static DEVICE_ATTR_RO(phys_port_id); 496 497 static ssize_t phys_port_name_show(struct device *dev, 498 struct device_attribute *attr, char *buf) 499 { 500 struct net_device *netdev = to_net_dev(dev); 501 ssize_t ret = -EINVAL; 502 503 if (!rtnl_trylock()) 504 return restart_syscall(); 505 506 if (dev_isalive(netdev)) { 507 char name[IFNAMSIZ]; 508 509 ret = dev_get_phys_port_name(netdev, name, sizeof(name)); 510 if (!ret) 511 ret = sprintf(buf, "%s\n", name); 512 } 513 rtnl_unlock(); 514 515 return ret; 516 } 517 static DEVICE_ATTR_RO(phys_port_name); 518 519 static ssize_t phys_switch_id_show(struct device *dev, 520 struct device_attribute *attr, char *buf) 521 { 522 struct net_device *netdev = to_net_dev(dev); 523 ssize_t ret = -EINVAL; 524 525 if (!rtnl_trylock()) 526 return restart_syscall(); 527 528 if (dev_isalive(netdev)) { 529 struct netdev_phys_item_id ppid = { }; 530 531 ret = dev_get_port_parent_id(netdev, &ppid, false); 532 if (!ret) 533 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id); 534 } 535 rtnl_unlock(); 536 537 return ret; 538 } 539 static DEVICE_ATTR_RO(phys_switch_id); 540 541 static struct attribute *net_class_attrs[] __ro_after_init = { 542 &dev_attr_netdev_group.attr, 543 &dev_attr_type.attr, 544 &dev_attr_dev_id.attr, 545 &dev_attr_dev_port.attr, 546 &dev_attr_iflink.attr, 547 &dev_attr_ifindex.attr, 548 &dev_attr_name_assign_type.attr, 549 &dev_attr_addr_assign_type.attr, 550 &dev_attr_addr_len.attr, 551 &dev_attr_link_mode.attr, 552 &dev_attr_address.attr, 553 &dev_attr_broadcast.attr, 554 &dev_attr_speed.attr, 555 &dev_attr_duplex.attr, 556 &dev_attr_dormant.attr, 557 &dev_attr_testing.attr, 558 &dev_attr_operstate.attr, 559 &dev_attr_carrier_changes.attr, 560 &dev_attr_ifalias.attr, 561 &dev_attr_carrier.attr, 562 &dev_attr_mtu.attr, 563 &dev_attr_flags.attr, 564 &dev_attr_tx_queue_len.attr, 565 &dev_attr_gro_flush_timeout.attr, 566 &dev_attr_napi_defer_hard_irqs.attr, 567 &dev_attr_phys_port_id.attr, 568 &dev_attr_phys_port_name.attr, 569 &dev_attr_phys_switch_id.attr, 570 &dev_attr_proto_down.attr, 571 &dev_attr_carrier_up_count.attr, 572 &dev_attr_carrier_down_count.attr, 573 NULL, 574 }; 575 ATTRIBUTE_GROUPS(net_class); 576 577 /* Show a given an attribute in the statistics group */ 578 static ssize_t netstat_show(const struct device *d, 579 struct device_attribute *attr, char *buf, 580 unsigned long offset) 581 { 582 struct net_device *dev = to_net_dev(d); 583 ssize_t ret = -EINVAL; 584 585 WARN_ON(offset > sizeof(struct rtnl_link_stats64) || 586 offset % sizeof(u64) != 0); 587 588 read_lock(&dev_base_lock); 589 if (dev_isalive(dev)) { 590 struct rtnl_link_stats64 temp; 591 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); 592 593 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset)); 594 } 595 read_unlock(&dev_base_lock); 596 return ret; 597 } 598 599 /* generate a read-only statistics attribute */ 600 #define NETSTAT_ENTRY(name) \ 601 static ssize_t name##_show(struct device *d, \ 602 struct device_attribute *attr, char *buf) \ 603 { \ 604 return netstat_show(d, attr, buf, \ 605 offsetof(struct rtnl_link_stats64, name)); \ 606 } \ 607 static DEVICE_ATTR_RO(name) 608 609 NETSTAT_ENTRY(rx_packets); 610 NETSTAT_ENTRY(tx_packets); 611 NETSTAT_ENTRY(rx_bytes); 612 NETSTAT_ENTRY(tx_bytes); 613 NETSTAT_ENTRY(rx_errors); 614 NETSTAT_ENTRY(tx_errors); 615 NETSTAT_ENTRY(rx_dropped); 616 NETSTAT_ENTRY(tx_dropped); 617 NETSTAT_ENTRY(multicast); 618 NETSTAT_ENTRY(collisions); 619 NETSTAT_ENTRY(rx_length_errors); 620 NETSTAT_ENTRY(rx_over_errors); 621 NETSTAT_ENTRY(rx_crc_errors); 622 NETSTAT_ENTRY(rx_frame_errors); 623 NETSTAT_ENTRY(rx_fifo_errors); 624 NETSTAT_ENTRY(rx_missed_errors); 625 NETSTAT_ENTRY(tx_aborted_errors); 626 NETSTAT_ENTRY(tx_carrier_errors); 627 NETSTAT_ENTRY(tx_fifo_errors); 628 NETSTAT_ENTRY(tx_heartbeat_errors); 629 NETSTAT_ENTRY(tx_window_errors); 630 NETSTAT_ENTRY(rx_compressed); 631 NETSTAT_ENTRY(tx_compressed); 632 NETSTAT_ENTRY(rx_nohandler); 633 634 static struct attribute *netstat_attrs[] __ro_after_init = { 635 &dev_attr_rx_packets.attr, 636 &dev_attr_tx_packets.attr, 637 &dev_attr_rx_bytes.attr, 638 &dev_attr_tx_bytes.attr, 639 &dev_attr_rx_errors.attr, 640 &dev_attr_tx_errors.attr, 641 &dev_attr_rx_dropped.attr, 642 &dev_attr_tx_dropped.attr, 643 &dev_attr_multicast.attr, 644 &dev_attr_collisions.attr, 645 &dev_attr_rx_length_errors.attr, 646 &dev_attr_rx_over_errors.attr, 647 &dev_attr_rx_crc_errors.attr, 648 &dev_attr_rx_frame_errors.attr, 649 &dev_attr_rx_fifo_errors.attr, 650 &dev_attr_rx_missed_errors.attr, 651 &dev_attr_tx_aborted_errors.attr, 652 &dev_attr_tx_carrier_errors.attr, 653 &dev_attr_tx_fifo_errors.attr, 654 &dev_attr_tx_heartbeat_errors.attr, 655 &dev_attr_tx_window_errors.attr, 656 &dev_attr_rx_compressed.attr, 657 &dev_attr_tx_compressed.attr, 658 &dev_attr_rx_nohandler.attr, 659 NULL 660 }; 661 662 static const struct attribute_group netstat_group = { 663 .name = "statistics", 664 .attrs = netstat_attrs, 665 }; 666 667 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211) 668 static struct attribute *wireless_attrs[] = { 669 NULL 670 }; 671 672 static const struct attribute_group wireless_group = { 673 .name = "wireless", 674 .attrs = wireless_attrs, 675 }; 676 #endif 677 678 #else /* CONFIG_SYSFS */ 679 #define net_class_groups NULL 680 #endif /* CONFIG_SYSFS */ 681 682 #ifdef CONFIG_SYSFS 683 #define to_rx_queue_attr(_attr) \ 684 container_of(_attr, struct rx_queue_attribute, attr) 685 686 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj) 687 688 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr, 689 char *buf) 690 { 691 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr); 692 struct netdev_rx_queue *queue = to_rx_queue(kobj); 693 694 if (!attribute->show) 695 return -EIO; 696 697 return attribute->show(queue, buf); 698 } 699 700 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr, 701 const char *buf, size_t count) 702 { 703 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr); 704 struct netdev_rx_queue *queue = to_rx_queue(kobj); 705 706 if (!attribute->store) 707 return -EIO; 708 709 return attribute->store(queue, buf, count); 710 } 711 712 static const struct sysfs_ops rx_queue_sysfs_ops = { 713 .show = rx_queue_attr_show, 714 .store = rx_queue_attr_store, 715 }; 716 717 #ifdef CONFIG_RPS 718 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf) 719 { 720 struct rps_map *map; 721 cpumask_var_t mask; 722 int i, len; 723 724 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 725 return -ENOMEM; 726 727 rcu_read_lock(); 728 map = rcu_dereference(queue->rps_map); 729 if (map) 730 for (i = 0; i < map->len; i++) 731 cpumask_set_cpu(map->cpus[i], mask); 732 733 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask)); 734 rcu_read_unlock(); 735 free_cpumask_var(mask); 736 737 return len < PAGE_SIZE ? len : -EINVAL; 738 } 739 740 static ssize_t store_rps_map(struct netdev_rx_queue *queue, 741 const char *buf, size_t len) 742 { 743 struct rps_map *old_map, *map; 744 cpumask_var_t mask; 745 int err, cpu, i, hk_flags; 746 static DEFINE_MUTEX(rps_map_mutex); 747 748 if (!capable(CAP_NET_ADMIN)) 749 return -EPERM; 750 751 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 752 return -ENOMEM; 753 754 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits); 755 if (err) { 756 free_cpumask_var(mask); 757 return err; 758 } 759 760 if (!cpumask_empty(mask)) { 761 hk_flags = HK_FLAG_DOMAIN | HK_FLAG_WQ; 762 cpumask_and(mask, mask, housekeeping_cpumask(hk_flags)); 763 if (cpumask_empty(mask)) { 764 free_cpumask_var(mask); 765 return -EINVAL; 766 } 767 } 768 769 map = kzalloc(max_t(unsigned int, 770 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES), 771 GFP_KERNEL); 772 if (!map) { 773 free_cpumask_var(mask); 774 return -ENOMEM; 775 } 776 777 i = 0; 778 for_each_cpu_and(cpu, mask, cpu_online_mask) 779 map->cpus[i++] = cpu; 780 781 if (i) { 782 map->len = i; 783 } else { 784 kfree(map); 785 map = NULL; 786 } 787 788 mutex_lock(&rps_map_mutex); 789 old_map = rcu_dereference_protected(queue->rps_map, 790 mutex_is_locked(&rps_map_mutex)); 791 rcu_assign_pointer(queue->rps_map, map); 792 793 if (map) 794 static_branch_inc(&rps_needed); 795 if (old_map) 796 static_branch_dec(&rps_needed); 797 798 mutex_unlock(&rps_map_mutex); 799 800 if (old_map) 801 kfree_rcu(old_map, rcu); 802 803 free_cpumask_var(mask); 804 return len; 805 } 806 807 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue, 808 char *buf) 809 { 810 struct rps_dev_flow_table *flow_table; 811 unsigned long val = 0; 812 813 rcu_read_lock(); 814 flow_table = rcu_dereference(queue->rps_flow_table); 815 if (flow_table) 816 val = (unsigned long)flow_table->mask + 1; 817 rcu_read_unlock(); 818 819 return sprintf(buf, "%lu\n", val); 820 } 821 822 static void rps_dev_flow_table_release(struct rcu_head *rcu) 823 { 824 struct rps_dev_flow_table *table = container_of(rcu, 825 struct rps_dev_flow_table, rcu); 826 vfree(table); 827 } 828 829 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue, 830 const char *buf, size_t len) 831 { 832 unsigned long mask, count; 833 struct rps_dev_flow_table *table, *old_table; 834 static DEFINE_SPINLOCK(rps_dev_flow_lock); 835 int rc; 836 837 if (!capable(CAP_NET_ADMIN)) 838 return -EPERM; 839 840 rc = kstrtoul(buf, 0, &count); 841 if (rc < 0) 842 return rc; 843 844 if (count) { 845 mask = count - 1; 846 /* mask = roundup_pow_of_two(count) - 1; 847 * without overflows... 848 */ 849 while ((mask | (mask >> 1)) != mask) 850 mask |= (mask >> 1); 851 /* On 64 bit arches, must check mask fits in table->mask (u32), 852 * and on 32bit arches, must check 853 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow. 854 */ 855 #if BITS_PER_LONG > 32 856 if (mask > (unsigned long)(u32)mask) 857 return -EINVAL; 858 #else 859 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1)) 860 / sizeof(struct rps_dev_flow)) { 861 /* Enforce a limit to prevent overflow */ 862 return -EINVAL; 863 } 864 #endif 865 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1)); 866 if (!table) 867 return -ENOMEM; 868 869 table->mask = mask; 870 for (count = 0; count <= mask; count++) 871 table->flows[count].cpu = RPS_NO_CPU; 872 } else { 873 table = NULL; 874 } 875 876 spin_lock(&rps_dev_flow_lock); 877 old_table = rcu_dereference_protected(queue->rps_flow_table, 878 lockdep_is_held(&rps_dev_flow_lock)); 879 rcu_assign_pointer(queue->rps_flow_table, table); 880 spin_unlock(&rps_dev_flow_lock); 881 882 if (old_table) 883 call_rcu(&old_table->rcu, rps_dev_flow_table_release); 884 885 return len; 886 } 887 888 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init 889 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map); 890 891 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init 892 = __ATTR(rps_flow_cnt, 0644, 893 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt); 894 #endif /* CONFIG_RPS */ 895 896 static struct attribute *rx_queue_default_attrs[] __ro_after_init = { 897 #ifdef CONFIG_RPS 898 &rps_cpus_attribute.attr, 899 &rps_dev_flow_table_cnt_attribute.attr, 900 #endif 901 NULL 902 }; 903 ATTRIBUTE_GROUPS(rx_queue_default); 904 905 static void rx_queue_release(struct kobject *kobj) 906 { 907 struct netdev_rx_queue *queue = to_rx_queue(kobj); 908 #ifdef CONFIG_RPS 909 struct rps_map *map; 910 struct rps_dev_flow_table *flow_table; 911 912 map = rcu_dereference_protected(queue->rps_map, 1); 913 if (map) { 914 RCU_INIT_POINTER(queue->rps_map, NULL); 915 kfree_rcu(map, rcu); 916 } 917 918 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1); 919 if (flow_table) { 920 RCU_INIT_POINTER(queue->rps_flow_table, NULL); 921 call_rcu(&flow_table->rcu, rps_dev_flow_table_release); 922 } 923 #endif 924 925 memset(kobj, 0, sizeof(*kobj)); 926 dev_put(queue->dev); 927 } 928 929 static const void *rx_queue_namespace(struct kobject *kobj) 930 { 931 struct netdev_rx_queue *queue = to_rx_queue(kobj); 932 struct device *dev = &queue->dev->dev; 933 const void *ns = NULL; 934 935 if (dev->class && dev->class->ns_type) 936 ns = dev->class->namespace(dev); 937 938 return ns; 939 } 940 941 static void rx_queue_get_ownership(struct kobject *kobj, 942 kuid_t *uid, kgid_t *gid) 943 { 944 const struct net *net = rx_queue_namespace(kobj); 945 946 net_ns_get_ownership(net, uid, gid); 947 } 948 949 static struct kobj_type rx_queue_ktype __ro_after_init = { 950 .sysfs_ops = &rx_queue_sysfs_ops, 951 .release = rx_queue_release, 952 .default_groups = rx_queue_default_groups, 953 .namespace = rx_queue_namespace, 954 .get_ownership = rx_queue_get_ownership, 955 }; 956 957 static int rx_queue_add_kobject(struct net_device *dev, int index) 958 { 959 struct netdev_rx_queue *queue = dev->_rx + index; 960 struct kobject *kobj = &queue->kobj; 961 int error = 0; 962 963 /* Kobject_put later will trigger rx_queue_release call which 964 * decreases dev refcount: Take that reference here 965 */ 966 dev_hold(queue->dev); 967 968 kobj->kset = dev->queues_kset; 969 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL, 970 "rx-%u", index); 971 if (error) 972 goto err; 973 974 if (dev->sysfs_rx_queue_group) { 975 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group); 976 if (error) 977 goto err; 978 } 979 980 kobject_uevent(kobj, KOBJ_ADD); 981 982 return error; 983 984 err: 985 kobject_put(kobj); 986 return error; 987 } 988 989 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid, 990 kgid_t kgid) 991 { 992 struct netdev_rx_queue *queue = dev->_rx + index; 993 struct kobject *kobj = &queue->kobj; 994 int error; 995 996 error = sysfs_change_owner(kobj, kuid, kgid); 997 if (error) 998 return error; 999 1000 if (dev->sysfs_rx_queue_group) 1001 error = sysfs_group_change_owner( 1002 kobj, dev->sysfs_rx_queue_group, kuid, kgid); 1003 1004 return error; 1005 } 1006 #endif /* CONFIG_SYSFS */ 1007 1008 int 1009 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num) 1010 { 1011 #ifdef CONFIG_SYSFS 1012 int i; 1013 int error = 0; 1014 1015 #ifndef CONFIG_RPS 1016 if (!dev->sysfs_rx_queue_group) 1017 return 0; 1018 #endif 1019 for (i = old_num; i < new_num; i++) { 1020 error = rx_queue_add_kobject(dev, i); 1021 if (error) { 1022 new_num = old_num; 1023 break; 1024 } 1025 } 1026 1027 while (--i >= new_num) { 1028 struct kobject *kobj = &dev->_rx[i].kobj; 1029 1030 if (!refcount_read(&dev_net(dev)->count)) 1031 kobj->uevent_suppress = 1; 1032 if (dev->sysfs_rx_queue_group) 1033 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group); 1034 kobject_put(kobj); 1035 } 1036 1037 return error; 1038 #else 1039 return 0; 1040 #endif 1041 } 1042 1043 static int net_rx_queue_change_owner(struct net_device *dev, int num, 1044 kuid_t kuid, kgid_t kgid) 1045 { 1046 #ifdef CONFIG_SYSFS 1047 int error = 0; 1048 int i; 1049 1050 #ifndef CONFIG_RPS 1051 if (!dev->sysfs_rx_queue_group) 1052 return 0; 1053 #endif 1054 for (i = 0; i < num; i++) { 1055 error = rx_queue_change_owner(dev, i, kuid, kgid); 1056 if (error) 1057 break; 1058 } 1059 1060 return error; 1061 #else 1062 return 0; 1063 #endif 1064 } 1065 1066 #ifdef CONFIG_SYSFS 1067 /* 1068 * netdev_queue sysfs structures and functions. 1069 */ 1070 struct netdev_queue_attribute { 1071 struct attribute attr; 1072 ssize_t (*show)(struct netdev_queue *queue, char *buf); 1073 ssize_t (*store)(struct netdev_queue *queue, 1074 const char *buf, size_t len); 1075 }; 1076 #define to_netdev_queue_attr(_attr) \ 1077 container_of(_attr, struct netdev_queue_attribute, attr) 1078 1079 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj) 1080 1081 static ssize_t netdev_queue_attr_show(struct kobject *kobj, 1082 struct attribute *attr, char *buf) 1083 { 1084 const struct netdev_queue_attribute *attribute 1085 = to_netdev_queue_attr(attr); 1086 struct netdev_queue *queue = to_netdev_queue(kobj); 1087 1088 if (!attribute->show) 1089 return -EIO; 1090 1091 return attribute->show(queue, buf); 1092 } 1093 1094 static ssize_t netdev_queue_attr_store(struct kobject *kobj, 1095 struct attribute *attr, 1096 const char *buf, size_t count) 1097 { 1098 const struct netdev_queue_attribute *attribute 1099 = to_netdev_queue_attr(attr); 1100 struct netdev_queue *queue = to_netdev_queue(kobj); 1101 1102 if (!attribute->store) 1103 return -EIO; 1104 1105 return attribute->store(queue, buf, count); 1106 } 1107 1108 static const struct sysfs_ops netdev_queue_sysfs_ops = { 1109 .show = netdev_queue_attr_show, 1110 .store = netdev_queue_attr_store, 1111 }; 1112 1113 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf) 1114 { 1115 unsigned long trans_timeout; 1116 1117 spin_lock_irq(&queue->_xmit_lock); 1118 trans_timeout = queue->trans_timeout; 1119 spin_unlock_irq(&queue->_xmit_lock); 1120 1121 return sprintf(buf, fmt_ulong, trans_timeout); 1122 } 1123 1124 static unsigned int get_netdev_queue_index(struct netdev_queue *queue) 1125 { 1126 struct net_device *dev = queue->dev; 1127 unsigned int i; 1128 1129 i = queue - dev->_tx; 1130 BUG_ON(i >= dev->num_tx_queues); 1131 1132 return i; 1133 } 1134 1135 static ssize_t traffic_class_show(struct netdev_queue *queue, 1136 char *buf) 1137 { 1138 struct net_device *dev = queue->dev; 1139 int index; 1140 int tc; 1141 1142 if (!netif_is_multiqueue(dev)) 1143 return -ENOENT; 1144 1145 index = get_netdev_queue_index(queue); 1146 1147 /* If queue belongs to subordinate dev use its TC mapping */ 1148 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 1149 1150 tc = netdev_txq_to_tc(dev, index); 1151 if (tc < 0) 1152 return -EINVAL; 1153 1154 /* We can report the traffic class one of two ways: 1155 * Subordinate device traffic classes are reported with the traffic 1156 * class first, and then the subordinate class so for example TC0 on 1157 * subordinate device 2 will be reported as "0-2". If the queue 1158 * belongs to the root device it will be reported with just the 1159 * traffic class, so just "0" for TC 0 for example. 1160 */ 1161 return dev->num_tc < 0 ? sprintf(buf, "%u%d\n", tc, dev->num_tc) : 1162 sprintf(buf, "%u\n", tc); 1163 } 1164 1165 #ifdef CONFIG_XPS 1166 static ssize_t tx_maxrate_show(struct netdev_queue *queue, 1167 char *buf) 1168 { 1169 return sprintf(buf, "%lu\n", queue->tx_maxrate); 1170 } 1171 1172 static ssize_t tx_maxrate_store(struct netdev_queue *queue, 1173 const char *buf, size_t len) 1174 { 1175 struct net_device *dev = queue->dev; 1176 int err, index = get_netdev_queue_index(queue); 1177 u32 rate = 0; 1178 1179 if (!capable(CAP_NET_ADMIN)) 1180 return -EPERM; 1181 1182 err = kstrtou32(buf, 10, &rate); 1183 if (err < 0) 1184 return err; 1185 1186 if (!rtnl_trylock()) 1187 return restart_syscall(); 1188 1189 err = -EOPNOTSUPP; 1190 if (dev->netdev_ops->ndo_set_tx_maxrate) 1191 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate); 1192 1193 rtnl_unlock(); 1194 if (!err) { 1195 queue->tx_maxrate = rate; 1196 return len; 1197 } 1198 return err; 1199 } 1200 1201 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init 1202 = __ATTR_RW(tx_maxrate); 1203 #endif 1204 1205 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init 1206 = __ATTR_RO(tx_timeout); 1207 1208 static struct netdev_queue_attribute queue_traffic_class __ro_after_init 1209 = __ATTR_RO(traffic_class); 1210 1211 #ifdef CONFIG_BQL 1212 /* 1213 * Byte queue limits sysfs structures and functions. 1214 */ 1215 static ssize_t bql_show(char *buf, unsigned int value) 1216 { 1217 return sprintf(buf, "%u\n", value); 1218 } 1219 1220 static ssize_t bql_set(const char *buf, const size_t count, 1221 unsigned int *pvalue) 1222 { 1223 unsigned int value; 1224 int err; 1225 1226 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) { 1227 value = DQL_MAX_LIMIT; 1228 } else { 1229 err = kstrtouint(buf, 10, &value); 1230 if (err < 0) 1231 return err; 1232 if (value > DQL_MAX_LIMIT) 1233 return -EINVAL; 1234 } 1235 1236 *pvalue = value; 1237 1238 return count; 1239 } 1240 1241 static ssize_t bql_show_hold_time(struct netdev_queue *queue, 1242 char *buf) 1243 { 1244 struct dql *dql = &queue->dql; 1245 1246 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time)); 1247 } 1248 1249 static ssize_t bql_set_hold_time(struct netdev_queue *queue, 1250 const char *buf, size_t len) 1251 { 1252 struct dql *dql = &queue->dql; 1253 unsigned int value; 1254 int err; 1255 1256 err = kstrtouint(buf, 10, &value); 1257 if (err < 0) 1258 return err; 1259 1260 dql->slack_hold_time = msecs_to_jiffies(value); 1261 1262 return len; 1263 } 1264 1265 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init 1266 = __ATTR(hold_time, 0644, 1267 bql_show_hold_time, bql_set_hold_time); 1268 1269 static ssize_t bql_show_inflight(struct netdev_queue *queue, 1270 char *buf) 1271 { 1272 struct dql *dql = &queue->dql; 1273 1274 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed); 1275 } 1276 1277 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init = 1278 __ATTR(inflight, 0444, bql_show_inflight, NULL); 1279 1280 #define BQL_ATTR(NAME, FIELD) \ 1281 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \ 1282 char *buf) \ 1283 { \ 1284 return bql_show(buf, queue->dql.FIELD); \ 1285 } \ 1286 \ 1287 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \ 1288 const char *buf, size_t len) \ 1289 { \ 1290 return bql_set(buf, len, &queue->dql.FIELD); \ 1291 } \ 1292 \ 1293 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \ 1294 = __ATTR(NAME, 0644, \ 1295 bql_show_ ## NAME, bql_set_ ## NAME) 1296 1297 BQL_ATTR(limit, limit); 1298 BQL_ATTR(limit_max, max_limit); 1299 BQL_ATTR(limit_min, min_limit); 1300 1301 static struct attribute *dql_attrs[] __ro_after_init = { 1302 &bql_limit_attribute.attr, 1303 &bql_limit_max_attribute.attr, 1304 &bql_limit_min_attribute.attr, 1305 &bql_hold_time_attribute.attr, 1306 &bql_inflight_attribute.attr, 1307 NULL 1308 }; 1309 1310 static const struct attribute_group dql_group = { 1311 .name = "byte_queue_limits", 1312 .attrs = dql_attrs, 1313 }; 1314 #endif /* CONFIG_BQL */ 1315 1316 #ifdef CONFIG_XPS 1317 static ssize_t xps_cpus_show(struct netdev_queue *queue, 1318 char *buf) 1319 { 1320 struct net_device *dev = queue->dev; 1321 int cpu, len, num_tc = 1, tc = 0; 1322 struct xps_dev_maps *dev_maps; 1323 cpumask_var_t mask; 1324 unsigned long index; 1325 1326 if (!netif_is_multiqueue(dev)) 1327 return -ENOENT; 1328 1329 index = get_netdev_queue_index(queue); 1330 1331 if (dev->num_tc) { 1332 /* Do not allow XPS on subordinate device directly */ 1333 num_tc = dev->num_tc; 1334 if (num_tc < 0) 1335 return -EINVAL; 1336 1337 /* If queue belongs to subordinate dev use its map */ 1338 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; 1339 1340 tc = netdev_txq_to_tc(dev, index); 1341 if (tc < 0) 1342 return -EINVAL; 1343 } 1344 1345 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 1346 return -ENOMEM; 1347 1348 rcu_read_lock(); 1349 dev_maps = rcu_dereference(dev->xps_cpus_map); 1350 if (dev_maps) { 1351 for_each_possible_cpu(cpu) { 1352 int i, tci = cpu * num_tc + tc; 1353 struct xps_map *map; 1354 1355 map = rcu_dereference(dev_maps->attr_map[tci]); 1356 if (!map) 1357 continue; 1358 1359 for (i = map->len; i--;) { 1360 if (map->queues[i] == index) { 1361 cpumask_set_cpu(cpu, mask); 1362 break; 1363 } 1364 } 1365 } 1366 } 1367 rcu_read_unlock(); 1368 1369 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask)); 1370 free_cpumask_var(mask); 1371 return len < PAGE_SIZE ? len : -EINVAL; 1372 } 1373 1374 static ssize_t xps_cpus_store(struct netdev_queue *queue, 1375 const char *buf, size_t len) 1376 { 1377 struct net_device *dev = queue->dev; 1378 unsigned long index; 1379 cpumask_var_t mask; 1380 int err; 1381 1382 if (!netif_is_multiqueue(dev)) 1383 return -ENOENT; 1384 1385 if (!capable(CAP_NET_ADMIN)) 1386 return -EPERM; 1387 1388 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 1389 return -ENOMEM; 1390 1391 index = get_netdev_queue_index(queue); 1392 1393 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits); 1394 if (err) { 1395 free_cpumask_var(mask); 1396 return err; 1397 } 1398 1399 err = netif_set_xps_queue(dev, mask, index); 1400 1401 free_cpumask_var(mask); 1402 1403 return err ? : len; 1404 } 1405 1406 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init 1407 = __ATTR_RW(xps_cpus); 1408 1409 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf) 1410 { 1411 struct net_device *dev = queue->dev; 1412 struct xps_dev_maps *dev_maps; 1413 unsigned long *mask, index; 1414 int j, len, num_tc = 1, tc = 0; 1415 1416 index = get_netdev_queue_index(queue); 1417 1418 if (dev->num_tc) { 1419 num_tc = dev->num_tc; 1420 tc = netdev_txq_to_tc(dev, index); 1421 if (tc < 0) 1422 return -EINVAL; 1423 } 1424 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL); 1425 if (!mask) 1426 return -ENOMEM; 1427 1428 rcu_read_lock(); 1429 dev_maps = rcu_dereference(dev->xps_rxqs_map); 1430 if (!dev_maps) 1431 goto out_no_maps; 1432 1433 for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues), 1434 j < dev->num_rx_queues;) { 1435 int i, tci = j * num_tc + tc; 1436 struct xps_map *map; 1437 1438 map = rcu_dereference(dev_maps->attr_map[tci]); 1439 if (!map) 1440 continue; 1441 1442 for (i = map->len; i--;) { 1443 if (map->queues[i] == index) { 1444 set_bit(j, mask); 1445 break; 1446 } 1447 } 1448 } 1449 out_no_maps: 1450 rcu_read_unlock(); 1451 1452 len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues); 1453 bitmap_free(mask); 1454 1455 return len < PAGE_SIZE ? len : -EINVAL; 1456 } 1457 1458 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf, 1459 size_t len) 1460 { 1461 struct net_device *dev = queue->dev; 1462 struct net *net = dev_net(dev); 1463 unsigned long *mask, index; 1464 int err; 1465 1466 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 1467 return -EPERM; 1468 1469 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL); 1470 if (!mask) 1471 return -ENOMEM; 1472 1473 index = get_netdev_queue_index(queue); 1474 1475 err = bitmap_parse(buf, len, mask, dev->num_rx_queues); 1476 if (err) { 1477 bitmap_free(mask); 1478 return err; 1479 } 1480 1481 cpus_read_lock(); 1482 err = __netif_set_xps_queue(dev, mask, index, true); 1483 cpus_read_unlock(); 1484 1485 bitmap_free(mask); 1486 return err ? : len; 1487 } 1488 1489 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init 1490 = __ATTR_RW(xps_rxqs); 1491 #endif /* CONFIG_XPS */ 1492 1493 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = { 1494 &queue_trans_timeout.attr, 1495 &queue_traffic_class.attr, 1496 #ifdef CONFIG_XPS 1497 &xps_cpus_attribute.attr, 1498 &xps_rxqs_attribute.attr, 1499 &queue_tx_maxrate.attr, 1500 #endif 1501 NULL 1502 }; 1503 ATTRIBUTE_GROUPS(netdev_queue_default); 1504 1505 static void netdev_queue_release(struct kobject *kobj) 1506 { 1507 struct netdev_queue *queue = to_netdev_queue(kobj); 1508 1509 memset(kobj, 0, sizeof(*kobj)); 1510 dev_put(queue->dev); 1511 } 1512 1513 static const void *netdev_queue_namespace(struct kobject *kobj) 1514 { 1515 struct netdev_queue *queue = to_netdev_queue(kobj); 1516 struct device *dev = &queue->dev->dev; 1517 const void *ns = NULL; 1518 1519 if (dev->class && dev->class->ns_type) 1520 ns = dev->class->namespace(dev); 1521 1522 return ns; 1523 } 1524 1525 static void netdev_queue_get_ownership(struct kobject *kobj, 1526 kuid_t *uid, kgid_t *gid) 1527 { 1528 const struct net *net = netdev_queue_namespace(kobj); 1529 1530 net_ns_get_ownership(net, uid, gid); 1531 } 1532 1533 static struct kobj_type netdev_queue_ktype __ro_after_init = { 1534 .sysfs_ops = &netdev_queue_sysfs_ops, 1535 .release = netdev_queue_release, 1536 .default_groups = netdev_queue_default_groups, 1537 .namespace = netdev_queue_namespace, 1538 .get_ownership = netdev_queue_get_ownership, 1539 }; 1540 1541 static int netdev_queue_add_kobject(struct net_device *dev, int index) 1542 { 1543 struct netdev_queue *queue = dev->_tx + index; 1544 struct kobject *kobj = &queue->kobj; 1545 int error = 0; 1546 1547 /* Kobject_put later will trigger netdev_queue_release call 1548 * which decreases dev refcount: Take that reference here 1549 */ 1550 dev_hold(queue->dev); 1551 1552 kobj->kset = dev->queues_kset; 1553 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL, 1554 "tx-%u", index); 1555 if (error) 1556 goto err; 1557 1558 #ifdef CONFIG_BQL 1559 error = sysfs_create_group(kobj, &dql_group); 1560 if (error) 1561 goto err; 1562 #endif 1563 1564 kobject_uevent(kobj, KOBJ_ADD); 1565 return 0; 1566 1567 err: 1568 kobject_put(kobj); 1569 return error; 1570 } 1571 1572 static int tx_queue_change_owner(struct net_device *ndev, int index, 1573 kuid_t kuid, kgid_t kgid) 1574 { 1575 struct netdev_queue *queue = ndev->_tx + index; 1576 struct kobject *kobj = &queue->kobj; 1577 int error; 1578 1579 error = sysfs_change_owner(kobj, kuid, kgid); 1580 if (error) 1581 return error; 1582 1583 #ifdef CONFIG_BQL 1584 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid); 1585 #endif 1586 return error; 1587 } 1588 #endif /* CONFIG_SYSFS */ 1589 1590 int 1591 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num) 1592 { 1593 #ifdef CONFIG_SYSFS 1594 int i; 1595 int error = 0; 1596 1597 for (i = old_num; i < new_num; i++) { 1598 error = netdev_queue_add_kobject(dev, i); 1599 if (error) { 1600 new_num = old_num; 1601 break; 1602 } 1603 } 1604 1605 while (--i >= new_num) { 1606 struct netdev_queue *queue = dev->_tx + i; 1607 1608 if (!refcount_read(&dev_net(dev)->count)) 1609 queue->kobj.uevent_suppress = 1; 1610 #ifdef CONFIG_BQL 1611 sysfs_remove_group(&queue->kobj, &dql_group); 1612 #endif 1613 kobject_put(&queue->kobj); 1614 } 1615 1616 return error; 1617 #else 1618 return 0; 1619 #endif /* CONFIG_SYSFS */ 1620 } 1621 1622 static int net_tx_queue_change_owner(struct net_device *dev, int num, 1623 kuid_t kuid, kgid_t kgid) 1624 { 1625 #ifdef CONFIG_SYSFS 1626 int error = 0; 1627 int i; 1628 1629 for (i = 0; i < num; i++) { 1630 error = tx_queue_change_owner(dev, i, kuid, kgid); 1631 if (error) 1632 break; 1633 } 1634 1635 return error; 1636 #else 1637 return 0; 1638 #endif /* CONFIG_SYSFS */ 1639 } 1640 1641 static int register_queue_kobjects(struct net_device *dev) 1642 { 1643 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0; 1644 1645 #ifdef CONFIG_SYSFS 1646 dev->queues_kset = kset_create_and_add("queues", 1647 NULL, &dev->dev.kobj); 1648 if (!dev->queues_kset) 1649 return -ENOMEM; 1650 real_rx = dev->real_num_rx_queues; 1651 #endif 1652 real_tx = dev->real_num_tx_queues; 1653 1654 error = net_rx_queue_update_kobjects(dev, 0, real_rx); 1655 if (error) 1656 goto error; 1657 rxq = real_rx; 1658 1659 error = netdev_queue_update_kobjects(dev, 0, real_tx); 1660 if (error) 1661 goto error; 1662 txq = real_tx; 1663 1664 return 0; 1665 1666 error: 1667 netdev_queue_update_kobjects(dev, txq, 0); 1668 net_rx_queue_update_kobjects(dev, rxq, 0); 1669 #ifdef CONFIG_SYSFS 1670 kset_unregister(dev->queues_kset); 1671 #endif 1672 return error; 1673 } 1674 1675 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid) 1676 { 1677 int error = 0, real_rx = 0, real_tx = 0; 1678 1679 #ifdef CONFIG_SYSFS 1680 if (ndev->queues_kset) { 1681 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid); 1682 if (error) 1683 return error; 1684 } 1685 real_rx = ndev->real_num_rx_queues; 1686 #endif 1687 real_tx = ndev->real_num_tx_queues; 1688 1689 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid); 1690 if (error) 1691 return error; 1692 1693 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid); 1694 if (error) 1695 return error; 1696 1697 return 0; 1698 } 1699 1700 static void remove_queue_kobjects(struct net_device *dev) 1701 { 1702 int real_rx = 0, real_tx = 0; 1703 1704 #ifdef CONFIG_SYSFS 1705 real_rx = dev->real_num_rx_queues; 1706 #endif 1707 real_tx = dev->real_num_tx_queues; 1708 1709 net_rx_queue_update_kobjects(dev, real_rx, 0); 1710 netdev_queue_update_kobjects(dev, real_tx, 0); 1711 #ifdef CONFIG_SYSFS 1712 kset_unregister(dev->queues_kset); 1713 #endif 1714 } 1715 1716 static bool net_current_may_mount(void) 1717 { 1718 struct net *net = current->nsproxy->net_ns; 1719 1720 return ns_capable(net->user_ns, CAP_SYS_ADMIN); 1721 } 1722 1723 static void *net_grab_current_ns(void) 1724 { 1725 struct net *ns = current->nsproxy->net_ns; 1726 #ifdef CONFIG_NET_NS 1727 if (ns) 1728 refcount_inc(&ns->passive); 1729 #endif 1730 return ns; 1731 } 1732 1733 static const void *net_initial_ns(void) 1734 { 1735 return &init_net; 1736 } 1737 1738 static const void *net_netlink_ns(struct sock *sk) 1739 { 1740 return sock_net(sk); 1741 } 1742 1743 const struct kobj_ns_type_operations net_ns_type_operations = { 1744 .type = KOBJ_NS_TYPE_NET, 1745 .current_may_mount = net_current_may_mount, 1746 .grab_current_ns = net_grab_current_ns, 1747 .netlink_ns = net_netlink_ns, 1748 .initial_ns = net_initial_ns, 1749 .drop_ns = net_drop_ns, 1750 }; 1751 EXPORT_SYMBOL_GPL(net_ns_type_operations); 1752 1753 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env) 1754 { 1755 struct net_device *dev = to_net_dev(d); 1756 int retval; 1757 1758 /* pass interface to uevent. */ 1759 retval = add_uevent_var(env, "INTERFACE=%s", dev->name); 1760 if (retval) 1761 goto exit; 1762 1763 /* pass ifindex to uevent. 1764 * ifindex is useful as it won't change (interface name may change) 1765 * and is what RtNetlink uses natively. 1766 */ 1767 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex); 1768 1769 exit: 1770 return retval; 1771 } 1772 1773 /* 1774 * netdev_release -- destroy and free a dead device. 1775 * Called when last reference to device kobject is gone. 1776 */ 1777 static void netdev_release(struct device *d) 1778 { 1779 struct net_device *dev = to_net_dev(d); 1780 1781 BUG_ON(dev->reg_state != NETREG_RELEASED); 1782 1783 /* no need to wait for rcu grace period: 1784 * device is dead and about to be freed. 1785 */ 1786 kfree(rcu_access_pointer(dev->ifalias)); 1787 netdev_freemem(dev); 1788 } 1789 1790 static const void *net_namespace(struct device *d) 1791 { 1792 struct net_device *dev = to_net_dev(d); 1793 1794 return dev_net(dev); 1795 } 1796 1797 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid) 1798 { 1799 struct net_device *dev = to_net_dev(d); 1800 const struct net *net = dev_net(dev); 1801 1802 net_ns_get_ownership(net, uid, gid); 1803 } 1804 1805 static struct class net_class __ro_after_init = { 1806 .name = "net", 1807 .dev_release = netdev_release, 1808 .dev_groups = net_class_groups, 1809 .dev_uevent = netdev_uevent, 1810 .ns_type = &net_ns_type_operations, 1811 .namespace = net_namespace, 1812 .get_ownership = net_get_ownership, 1813 }; 1814 1815 #ifdef CONFIG_OF_NET 1816 static int of_dev_node_match(struct device *dev, const void *data) 1817 { 1818 for (; dev; dev = dev->parent) { 1819 if (dev->of_node == data) 1820 return 1; 1821 } 1822 1823 return 0; 1824 } 1825 1826 /* 1827 * of_find_net_device_by_node - lookup the net device for the device node 1828 * @np: OF device node 1829 * 1830 * Looks up the net_device structure corresponding with the device node. 1831 * If successful, returns a pointer to the net_device with the embedded 1832 * struct device refcount incremented by one, or NULL on failure. The 1833 * refcount must be dropped when done with the net_device. 1834 */ 1835 struct net_device *of_find_net_device_by_node(struct device_node *np) 1836 { 1837 struct device *dev; 1838 1839 dev = class_find_device(&net_class, NULL, np, of_dev_node_match); 1840 if (!dev) 1841 return NULL; 1842 1843 return to_net_dev(dev); 1844 } 1845 EXPORT_SYMBOL(of_find_net_device_by_node); 1846 #endif 1847 1848 /* Delete sysfs entries but hold kobject reference until after all 1849 * netdev references are gone. 1850 */ 1851 void netdev_unregister_kobject(struct net_device *ndev) 1852 { 1853 struct device *dev = &ndev->dev; 1854 1855 if (!refcount_read(&dev_net(ndev)->count)) 1856 dev_set_uevent_suppress(dev, 1); 1857 1858 kobject_get(&dev->kobj); 1859 1860 remove_queue_kobjects(ndev); 1861 1862 pm_runtime_set_memalloc_noio(dev, false); 1863 1864 device_del(dev); 1865 } 1866 1867 /* Create sysfs entries for network device. */ 1868 int netdev_register_kobject(struct net_device *ndev) 1869 { 1870 struct device *dev = &ndev->dev; 1871 const struct attribute_group **groups = ndev->sysfs_groups; 1872 int error = 0; 1873 1874 device_initialize(dev); 1875 dev->class = &net_class; 1876 dev->platform_data = ndev; 1877 dev->groups = groups; 1878 1879 dev_set_name(dev, "%s", ndev->name); 1880 1881 #ifdef CONFIG_SYSFS 1882 /* Allow for a device specific group */ 1883 if (*groups) 1884 groups++; 1885 1886 *groups++ = &netstat_group; 1887 1888 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211) 1889 if (ndev->ieee80211_ptr) 1890 *groups++ = &wireless_group; 1891 #if IS_ENABLED(CONFIG_WIRELESS_EXT) 1892 else if (ndev->wireless_handlers) 1893 *groups++ = &wireless_group; 1894 #endif 1895 #endif 1896 #endif /* CONFIG_SYSFS */ 1897 1898 error = device_add(dev); 1899 if (error) 1900 return error; 1901 1902 error = register_queue_kobjects(ndev); 1903 if (error) { 1904 device_del(dev); 1905 return error; 1906 } 1907 1908 pm_runtime_set_memalloc_noio(dev, true); 1909 1910 return error; 1911 } 1912 1913 /* Change owner for sysfs entries when moving network devices across network 1914 * namespaces owned by different user namespaces. 1915 */ 1916 int netdev_change_owner(struct net_device *ndev, const struct net *net_old, 1917 const struct net *net_new) 1918 { 1919 struct device *dev = &ndev->dev; 1920 kuid_t old_uid, new_uid; 1921 kgid_t old_gid, new_gid; 1922 int error; 1923 1924 net_ns_get_ownership(net_old, &old_uid, &old_gid); 1925 net_ns_get_ownership(net_new, &new_uid, &new_gid); 1926 1927 /* The network namespace was changed but the owning user namespace is 1928 * identical so there's no need to change the owner of sysfs entries. 1929 */ 1930 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid)) 1931 return 0; 1932 1933 error = device_change_owner(dev, new_uid, new_gid); 1934 if (error) 1935 return error; 1936 1937 error = queue_change_owner(ndev, new_uid, new_gid); 1938 if (error) 1939 return error; 1940 1941 return 0; 1942 } 1943 1944 int netdev_class_create_file_ns(const struct class_attribute *class_attr, 1945 const void *ns) 1946 { 1947 return class_create_file_ns(&net_class, class_attr, ns); 1948 } 1949 EXPORT_SYMBOL(netdev_class_create_file_ns); 1950 1951 void netdev_class_remove_file_ns(const struct class_attribute *class_attr, 1952 const void *ns) 1953 { 1954 class_remove_file_ns(&net_class, class_attr, ns); 1955 } 1956 EXPORT_SYMBOL(netdev_class_remove_file_ns); 1957 1958 int __init netdev_kobject_init(void) 1959 { 1960 kobj_ns_type_register(&net_ns_type_operations); 1961 return class_register(&net_class); 1962 } 1963