1 /* 2 * net-sysfs.c - network device class and attributes 3 * 4 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org> 5 * 6 * This program is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU General Public License 8 * as published by the Free Software Foundation; either version 9 * 2 of the License, or (at your option) any later version. 10 */ 11 12 #include <linux/capability.h> 13 #include <linux/kernel.h> 14 #include <linux/netdevice.h> 15 #include <linux/if_arp.h> 16 #include <linux/slab.h> 17 #include <linux/nsproxy.h> 18 #include <net/sock.h> 19 #include <net/net_namespace.h> 20 #include <linux/rtnetlink.h> 21 #include <linux/wireless.h> 22 #include <linux/vmalloc.h> 23 #include <linux/export.h> 24 #include <net/wext.h> 25 26 #include "net-sysfs.h" 27 28 #ifdef CONFIG_SYSFS 29 static const char fmt_hex[] = "%#x\n"; 30 static const char fmt_long_hex[] = "%#lx\n"; 31 static const char fmt_dec[] = "%d\n"; 32 static const char fmt_udec[] = "%u\n"; 33 static const char fmt_ulong[] = "%lu\n"; 34 static const char fmt_u64[] = "%llu\n"; 35 36 static inline int dev_isalive(const struct net_device *dev) 37 { 38 return dev->reg_state <= NETREG_REGISTERED; 39 } 40 41 /* use same locking rules as GIF* ioctl's */ 42 static ssize_t netdev_show(const struct device *dev, 43 struct device_attribute *attr, char *buf, 44 ssize_t (*format)(const struct net_device *, char *)) 45 { 46 struct net_device *net = to_net_dev(dev); 47 ssize_t ret = -EINVAL; 48 49 read_lock(&dev_base_lock); 50 if (dev_isalive(net)) 51 ret = (*format)(net, buf); 52 read_unlock(&dev_base_lock); 53 54 return ret; 55 } 56 57 /* generate a show function for simple field */ 58 #define NETDEVICE_SHOW(field, format_string) \ 59 static ssize_t format_##field(const struct net_device *net, char *buf) \ 60 { \ 61 return sprintf(buf, format_string, net->field); \ 62 } \ 63 static ssize_t show_##field(struct device *dev, \ 64 struct device_attribute *attr, char *buf) \ 65 { \ 66 return netdev_show(dev, attr, buf, format_##field); \ 67 } 68 69 70 /* use same locking and permission rules as SIF* ioctl's */ 71 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr, 72 const char *buf, size_t len, 73 int (*set)(struct net_device *, unsigned long)) 74 { 75 struct net_device *net = to_net_dev(dev); 76 char *endp; 77 unsigned long new; 78 int ret = -EINVAL; 79 80 if (!capable(CAP_NET_ADMIN)) 81 return -EPERM; 82 83 new = simple_strtoul(buf, &endp, 0); 84 if (endp == buf) 85 goto err; 86 87 if (!rtnl_trylock()) 88 return restart_syscall(); 89 90 if (dev_isalive(net)) { 91 if ((ret = (*set)(net, new)) == 0) 92 ret = len; 93 } 94 rtnl_unlock(); 95 err: 96 return ret; 97 } 98 99 NETDEVICE_SHOW(dev_id, fmt_hex); 100 NETDEVICE_SHOW(addr_assign_type, fmt_dec); 101 NETDEVICE_SHOW(addr_len, fmt_dec); 102 NETDEVICE_SHOW(iflink, fmt_dec); 103 NETDEVICE_SHOW(ifindex, fmt_dec); 104 NETDEVICE_SHOW(type, fmt_dec); 105 NETDEVICE_SHOW(link_mode, fmt_dec); 106 107 /* use same locking rules as GIFHWADDR ioctl's */ 108 static ssize_t show_address(struct device *dev, struct device_attribute *attr, 109 char *buf) 110 { 111 struct net_device *net = to_net_dev(dev); 112 ssize_t ret = -EINVAL; 113 114 read_lock(&dev_base_lock); 115 if (dev_isalive(net)) 116 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len); 117 read_unlock(&dev_base_lock); 118 return ret; 119 } 120 121 static ssize_t show_broadcast(struct device *dev, 122 struct device_attribute *attr, char *buf) 123 { 124 struct net_device *net = to_net_dev(dev); 125 if (dev_isalive(net)) 126 return sysfs_format_mac(buf, net->broadcast, net->addr_len); 127 return -EINVAL; 128 } 129 130 static ssize_t show_carrier(struct device *dev, 131 struct device_attribute *attr, char *buf) 132 { 133 struct net_device *netdev = to_net_dev(dev); 134 if (netif_running(netdev)) { 135 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev)); 136 } 137 return -EINVAL; 138 } 139 140 static ssize_t show_speed(struct device *dev, 141 struct device_attribute *attr, char *buf) 142 { 143 struct net_device *netdev = to_net_dev(dev); 144 int ret = -EINVAL; 145 146 if (!rtnl_trylock()) 147 return restart_syscall(); 148 149 if (netif_running(netdev)) { 150 struct ethtool_cmd cmd; 151 if (!__ethtool_get_settings(netdev, &cmd)) 152 ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd)); 153 } 154 rtnl_unlock(); 155 return ret; 156 } 157 158 static ssize_t show_duplex(struct device *dev, 159 struct device_attribute *attr, char *buf) 160 { 161 struct net_device *netdev = to_net_dev(dev); 162 int ret = -EINVAL; 163 164 if (!rtnl_trylock()) 165 return restart_syscall(); 166 167 if (netif_running(netdev)) { 168 struct ethtool_cmd cmd; 169 if (!__ethtool_get_settings(netdev, &cmd)) 170 ret = sprintf(buf, "%s\n", 171 cmd.duplex ? "full" : "half"); 172 } 173 rtnl_unlock(); 174 return ret; 175 } 176 177 static ssize_t show_dormant(struct device *dev, 178 struct device_attribute *attr, char *buf) 179 { 180 struct net_device *netdev = to_net_dev(dev); 181 182 if (netif_running(netdev)) 183 return sprintf(buf, fmt_dec, !!netif_dormant(netdev)); 184 185 return -EINVAL; 186 } 187 188 static const char *const operstates[] = { 189 "unknown", 190 "notpresent", /* currently unused */ 191 "down", 192 "lowerlayerdown", 193 "testing", /* currently unused */ 194 "dormant", 195 "up" 196 }; 197 198 static ssize_t show_operstate(struct device *dev, 199 struct device_attribute *attr, char *buf) 200 { 201 const struct net_device *netdev = to_net_dev(dev); 202 unsigned char operstate; 203 204 read_lock(&dev_base_lock); 205 operstate = netdev->operstate; 206 if (!netif_running(netdev)) 207 operstate = IF_OPER_DOWN; 208 read_unlock(&dev_base_lock); 209 210 if (operstate >= ARRAY_SIZE(operstates)) 211 return -EINVAL; /* should not happen */ 212 213 return sprintf(buf, "%s\n", operstates[operstate]); 214 } 215 216 /* read-write attributes */ 217 NETDEVICE_SHOW(mtu, fmt_dec); 218 219 static int change_mtu(struct net_device *net, unsigned long new_mtu) 220 { 221 return dev_set_mtu(net, (int) new_mtu); 222 } 223 224 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr, 225 const char *buf, size_t len) 226 { 227 return netdev_store(dev, attr, buf, len, change_mtu); 228 } 229 230 NETDEVICE_SHOW(flags, fmt_hex); 231 232 static int change_flags(struct net_device *net, unsigned long new_flags) 233 { 234 return dev_change_flags(net, (unsigned) new_flags); 235 } 236 237 static ssize_t store_flags(struct device *dev, struct device_attribute *attr, 238 const char *buf, size_t len) 239 { 240 return netdev_store(dev, attr, buf, len, change_flags); 241 } 242 243 NETDEVICE_SHOW(tx_queue_len, fmt_ulong); 244 245 static int change_tx_queue_len(struct net_device *net, unsigned long new_len) 246 { 247 net->tx_queue_len = new_len; 248 return 0; 249 } 250 251 static ssize_t store_tx_queue_len(struct device *dev, 252 struct device_attribute *attr, 253 const char *buf, size_t len) 254 { 255 return netdev_store(dev, attr, buf, len, change_tx_queue_len); 256 } 257 258 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr, 259 const char *buf, size_t len) 260 { 261 struct net_device *netdev = to_net_dev(dev); 262 size_t count = len; 263 ssize_t ret; 264 265 if (!capable(CAP_NET_ADMIN)) 266 return -EPERM; 267 268 /* ignore trailing newline */ 269 if (len > 0 && buf[len - 1] == '\n') 270 --count; 271 272 if (!rtnl_trylock()) 273 return restart_syscall(); 274 ret = dev_set_alias(netdev, buf, count); 275 rtnl_unlock(); 276 277 return ret < 0 ? ret : len; 278 } 279 280 static ssize_t show_ifalias(struct device *dev, 281 struct device_attribute *attr, char *buf) 282 { 283 const struct net_device *netdev = to_net_dev(dev); 284 ssize_t ret = 0; 285 286 if (!rtnl_trylock()) 287 return restart_syscall(); 288 if (netdev->ifalias) 289 ret = sprintf(buf, "%s\n", netdev->ifalias); 290 rtnl_unlock(); 291 return ret; 292 } 293 294 NETDEVICE_SHOW(group, fmt_dec); 295 296 static int change_group(struct net_device *net, unsigned long new_group) 297 { 298 dev_set_group(net, (int) new_group); 299 return 0; 300 } 301 302 static ssize_t store_group(struct device *dev, struct device_attribute *attr, 303 const char *buf, size_t len) 304 { 305 return netdev_store(dev, attr, buf, len, change_group); 306 } 307 308 static struct device_attribute net_class_attributes[] = { 309 __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL), 310 __ATTR(addr_len, S_IRUGO, show_addr_len, NULL), 311 __ATTR(dev_id, S_IRUGO, show_dev_id, NULL), 312 __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias), 313 __ATTR(iflink, S_IRUGO, show_iflink, NULL), 314 __ATTR(ifindex, S_IRUGO, show_ifindex, NULL), 315 __ATTR(type, S_IRUGO, show_type, NULL), 316 __ATTR(link_mode, S_IRUGO, show_link_mode, NULL), 317 __ATTR(address, S_IRUGO, show_address, NULL), 318 __ATTR(broadcast, S_IRUGO, show_broadcast, NULL), 319 __ATTR(carrier, S_IRUGO, show_carrier, NULL), 320 __ATTR(speed, S_IRUGO, show_speed, NULL), 321 __ATTR(duplex, S_IRUGO, show_duplex, NULL), 322 __ATTR(dormant, S_IRUGO, show_dormant, NULL), 323 __ATTR(operstate, S_IRUGO, show_operstate, NULL), 324 __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu), 325 __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags), 326 __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len, 327 store_tx_queue_len), 328 __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group), 329 {} 330 }; 331 332 /* Show a given an attribute in the statistics group */ 333 static ssize_t netstat_show(const struct device *d, 334 struct device_attribute *attr, char *buf, 335 unsigned long offset) 336 { 337 struct net_device *dev = to_net_dev(d); 338 ssize_t ret = -EINVAL; 339 340 WARN_ON(offset > sizeof(struct rtnl_link_stats64) || 341 offset % sizeof(u64) != 0); 342 343 read_lock(&dev_base_lock); 344 if (dev_isalive(dev)) { 345 struct rtnl_link_stats64 temp; 346 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp); 347 348 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset)); 349 } 350 read_unlock(&dev_base_lock); 351 return ret; 352 } 353 354 /* generate a read-only statistics attribute */ 355 #define NETSTAT_ENTRY(name) \ 356 static ssize_t show_##name(struct device *d, \ 357 struct device_attribute *attr, char *buf) \ 358 { \ 359 return netstat_show(d, attr, buf, \ 360 offsetof(struct rtnl_link_stats64, name)); \ 361 } \ 362 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL) 363 364 NETSTAT_ENTRY(rx_packets); 365 NETSTAT_ENTRY(tx_packets); 366 NETSTAT_ENTRY(rx_bytes); 367 NETSTAT_ENTRY(tx_bytes); 368 NETSTAT_ENTRY(rx_errors); 369 NETSTAT_ENTRY(tx_errors); 370 NETSTAT_ENTRY(rx_dropped); 371 NETSTAT_ENTRY(tx_dropped); 372 NETSTAT_ENTRY(multicast); 373 NETSTAT_ENTRY(collisions); 374 NETSTAT_ENTRY(rx_length_errors); 375 NETSTAT_ENTRY(rx_over_errors); 376 NETSTAT_ENTRY(rx_crc_errors); 377 NETSTAT_ENTRY(rx_frame_errors); 378 NETSTAT_ENTRY(rx_fifo_errors); 379 NETSTAT_ENTRY(rx_missed_errors); 380 NETSTAT_ENTRY(tx_aborted_errors); 381 NETSTAT_ENTRY(tx_carrier_errors); 382 NETSTAT_ENTRY(tx_fifo_errors); 383 NETSTAT_ENTRY(tx_heartbeat_errors); 384 NETSTAT_ENTRY(tx_window_errors); 385 NETSTAT_ENTRY(rx_compressed); 386 NETSTAT_ENTRY(tx_compressed); 387 388 static struct attribute *netstat_attrs[] = { 389 &dev_attr_rx_packets.attr, 390 &dev_attr_tx_packets.attr, 391 &dev_attr_rx_bytes.attr, 392 &dev_attr_tx_bytes.attr, 393 &dev_attr_rx_errors.attr, 394 &dev_attr_tx_errors.attr, 395 &dev_attr_rx_dropped.attr, 396 &dev_attr_tx_dropped.attr, 397 &dev_attr_multicast.attr, 398 &dev_attr_collisions.attr, 399 &dev_attr_rx_length_errors.attr, 400 &dev_attr_rx_over_errors.attr, 401 &dev_attr_rx_crc_errors.attr, 402 &dev_attr_rx_frame_errors.attr, 403 &dev_attr_rx_fifo_errors.attr, 404 &dev_attr_rx_missed_errors.attr, 405 &dev_attr_tx_aborted_errors.attr, 406 &dev_attr_tx_carrier_errors.attr, 407 &dev_attr_tx_fifo_errors.attr, 408 &dev_attr_tx_heartbeat_errors.attr, 409 &dev_attr_tx_window_errors.attr, 410 &dev_attr_rx_compressed.attr, 411 &dev_attr_tx_compressed.attr, 412 NULL 413 }; 414 415 416 static struct attribute_group netstat_group = { 417 .name = "statistics", 418 .attrs = netstat_attrs, 419 }; 420 421 #ifdef CONFIG_WIRELESS_EXT_SYSFS 422 /* helper function that does all the locking etc for wireless stats */ 423 static ssize_t wireless_show(struct device *d, char *buf, 424 ssize_t (*format)(const struct iw_statistics *, 425 char *)) 426 { 427 struct net_device *dev = to_net_dev(d); 428 const struct iw_statistics *iw; 429 ssize_t ret = -EINVAL; 430 431 if (!rtnl_trylock()) 432 return restart_syscall(); 433 if (dev_isalive(dev)) { 434 iw = get_wireless_stats(dev); 435 if (iw) 436 ret = (*format)(iw, buf); 437 } 438 rtnl_unlock(); 439 440 return ret; 441 } 442 443 /* show function template for wireless fields */ 444 #define WIRELESS_SHOW(name, field, format_string) \ 445 static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \ 446 { \ 447 return sprintf(buf, format_string, iw->field); \ 448 } \ 449 static ssize_t show_iw_##name(struct device *d, \ 450 struct device_attribute *attr, char *buf) \ 451 { \ 452 return wireless_show(d, buf, format_iw_##name); \ 453 } \ 454 static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL) 455 456 WIRELESS_SHOW(status, status, fmt_hex); 457 WIRELESS_SHOW(link, qual.qual, fmt_dec); 458 WIRELESS_SHOW(level, qual.level, fmt_dec); 459 WIRELESS_SHOW(noise, qual.noise, fmt_dec); 460 WIRELESS_SHOW(nwid, discard.nwid, fmt_dec); 461 WIRELESS_SHOW(crypt, discard.code, fmt_dec); 462 WIRELESS_SHOW(fragment, discard.fragment, fmt_dec); 463 WIRELESS_SHOW(misc, discard.misc, fmt_dec); 464 WIRELESS_SHOW(retries, discard.retries, fmt_dec); 465 WIRELESS_SHOW(beacon, miss.beacon, fmt_dec); 466 467 static struct attribute *wireless_attrs[] = { 468 &dev_attr_status.attr, 469 &dev_attr_link.attr, 470 &dev_attr_level.attr, 471 &dev_attr_noise.attr, 472 &dev_attr_nwid.attr, 473 &dev_attr_crypt.attr, 474 &dev_attr_fragment.attr, 475 &dev_attr_retries.attr, 476 &dev_attr_misc.attr, 477 &dev_attr_beacon.attr, 478 NULL 479 }; 480 481 static struct attribute_group wireless_group = { 482 .name = "wireless", 483 .attrs = wireless_attrs, 484 }; 485 #endif 486 #endif /* CONFIG_SYSFS */ 487 488 #ifdef CONFIG_RPS 489 /* 490 * RX queue sysfs structures and functions. 491 */ 492 struct rx_queue_attribute { 493 struct attribute attr; 494 ssize_t (*show)(struct netdev_rx_queue *queue, 495 struct rx_queue_attribute *attr, char *buf); 496 ssize_t (*store)(struct netdev_rx_queue *queue, 497 struct rx_queue_attribute *attr, const char *buf, size_t len); 498 }; 499 #define to_rx_queue_attr(_attr) container_of(_attr, \ 500 struct rx_queue_attribute, attr) 501 502 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj) 503 504 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr, 505 char *buf) 506 { 507 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr); 508 struct netdev_rx_queue *queue = to_rx_queue(kobj); 509 510 if (!attribute->show) 511 return -EIO; 512 513 return attribute->show(queue, attribute, buf); 514 } 515 516 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr, 517 const char *buf, size_t count) 518 { 519 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr); 520 struct netdev_rx_queue *queue = to_rx_queue(kobj); 521 522 if (!attribute->store) 523 return -EIO; 524 525 return attribute->store(queue, attribute, buf, count); 526 } 527 528 static const struct sysfs_ops rx_queue_sysfs_ops = { 529 .show = rx_queue_attr_show, 530 .store = rx_queue_attr_store, 531 }; 532 533 static ssize_t show_rps_map(struct netdev_rx_queue *queue, 534 struct rx_queue_attribute *attribute, char *buf) 535 { 536 struct rps_map *map; 537 cpumask_var_t mask; 538 size_t len = 0; 539 int i; 540 541 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 542 return -ENOMEM; 543 544 rcu_read_lock(); 545 map = rcu_dereference(queue->rps_map); 546 if (map) 547 for (i = 0; i < map->len; i++) 548 cpumask_set_cpu(map->cpus[i], mask); 549 550 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask); 551 if (PAGE_SIZE - len < 3) { 552 rcu_read_unlock(); 553 free_cpumask_var(mask); 554 return -EINVAL; 555 } 556 rcu_read_unlock(); 557 558 free_cpumask_var(mask); 559 len += sprintf(buf + len, "\n"); 560 return len; 561 } 562 563 static ssize_t store_rps_map(struct netdev_rx_queue *queue, 564 struct rx_queue_attribute *attribute, 565 const char *buf, size_t len) 566 { 567 struct rps_map *old_map, *map; 568 cpumask_var_t mask; 569 int err, cpu, i; 570 static DEFINE_SPINLOCK(rps_map_lock); 571 572 if (!capable(CAP_NET_ADMIN)) 573 return -EPERM; 574 575 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 576 return -ENOMEM; 577 578 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits); 579 if (err) { 580 free_cpumask_var(mask); 581 return err; 582 } 583 584 map = kzalloc(max_t(unsigned, 585 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES), 586 GFP_KERNEL); 587 if (!map) { 588 free_cpumask_var(mask); 589 return -ENOMEM; 590 } 591 592 i = 0; 593 for_each_cpu_and(cpu, mask, cpu_online_mask) 594 map->cpus[i++] = cpu; 595 596 if (i) 597 map->len = i; 598 else { 599 kfree(map); 600 map = NULL; 601 } 602 603 spin_lock(&rps_map_lock); 604 old_map = rcu_dereference_protected(queue->rps_map, 605 lockdep_is_held(&rps_map_lock)); 606 rcu_assign_pointer(queue->rps_map, map); 607 spin_unlock(&rps_map_lock); 608 609 if (old_map) 610 kfree_rcu(old_map, rcu); 611 612 free_cpumask_var(mask); 613 return len; 614 } 615 616 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue, 617 struct rx_queue_attribute *attr, 618 char *buf) 619 { 620 struct rps_dev_flow_table *flow_table; 621 unsigned int val = 0; 622 623 rcu_read_lock(); 624 flow_table = rcu_dereference(queue->rps_flow_table); 625 if (flow_table) 626 val = flow_table->mask + 1; 627 rcu_read_unlock(); 628 629 return sprintf(buf, "%u\n", val); 630 } 631 632 static void rps_dev_flow_table_release_work(struct work_struct *work) 633 { 634 struct rps_dev_flow_table *table = container_of(work, 635 struct rps_dev_flow_table, free_work); 636 637 vfree(table); 638 } 639 640 static void rps_dev_flow_table_release(struct rcu_head *rcu) 641 { 642 struct rps_dev_flow_table *table = container_of(rcu, 643 struct rps_dev_flow_table, rcu); 644 645 INIT_WORK(&table->free_work, rps_dev_flow_table_release_work); 646 schedule_work(&table->free_work); 647 } 648 649 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue, 650 struct rx_queue_attribute *attr, 651 const char *buf, size_t len) 652 { 653 unsigned int count; 654 char *endp; 655 struct rps_dev_flow_table *table, *old_table; 656 static DEFINE_SPINLOCK(rps_dev_flow_lock); 657 658 if (!capable(CAP_NET_ADMIN)) 659 return -EPERM; 660 661 count = simple_strtoul(buf, &endp, 0); 662 if (endp == buf) 663 return -EINVAL; 664 665 if (count) { 666 int i; 667 668 if (count > INT_MAX) 669 return -EINVAL; 670 count = roundup_pow_of_two(count); 671 if (count > (ULONG_MAX - sizeof(struct rps_dev_flow_table)) 672 / sizeof(struct rps_dev_flow)) { 673 /* Enforce a limit to prevent overflow */ 674 return -EINVAL; 675 } 676 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(count)); 677 if (!table) 678 return -ENOMEM; 679 680 table->mask = count - 1; 681 for (i = 0; i < count; i++) 682 table->flows[i].cpu = RPS_NO_CPU; 683 } else 684 table = NULL; 685 686 spin_lock(&rps_dev_flow_lock); 687 old_table = rcu_dereference_protected(queue->rps_flow_table, 688 lockdep_is_held(&rps_dev_flow_lock)); 689 rcu_assign_pointer(queue->rps_flow_table, table); 690 spin_unlock(&rps_dev_flow_lock); 691 692 if (old_table) 693 call_rcu(&old_table->rcu, rps_dev_flow_table_release); 694 695 return len; 696 } 697 698 static struct rx_queue_attribute rps_cpus_attribute = 699 __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map); 700 701 702 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute = 703 __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR, 704 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt); 705 706 static struct attribute *rx_queue_default_attrs[] = { 707 &rps_cpus_attribute.attr, 708 &rps_dev_flow_table_cnt_attribute.attr, 709 NULL 710 }; 711 712 static void rx_queue_release(struct kobject *kobj) 713 { 714 struct netdev_rx_queue *queue = to_rx_queue(kobj); 715 struct rps_map *map; 716 struct rps_dev_flow_table *flow_table; 717 718 719 map = rcu_dereference_protected(queue->rps_map, 1); 720 if (map) { 721 RCU_INIT_POINTER(queue->rps_map, NULL); 722 kfree_rcu(map, rcu); 723 } 724 725 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1); 726 if (flow_table) { 727 RCU_INIT_POINTER(queue->rps_flow_table, NULL); 728 call_rcu(&flow_table->rcu, rps_dev_flow_table_release); 729 } 730 731 memset(kobj, 0, sizeof(*kobj)); 732 dev_put(queue->dev); 733 } 734 735 static struct kobj_type rx_queue_ktype = { 736 .sysfs_ops = &rx_queue_sysfs_ops, 737 .release = rx_queue_release, 738 .default_attrs = rx_queue_default_attrs, 739 }; 740 741 static int rx_queue_add_kobject(struct net_device *net, int index) 742 { 743 struct netdev_rx_queue *queue = net->_rx + index; 744 struct kobject *kobj = &queue->kobj; 745 int error = 0; 746 747 kobj->kset = net->queues_kset; 748 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL, 749 "rx-%u", index); 750 if (error) { 751 kobject_put(kobj); 752 return error; 753 } 754 755 kobject_uevent(kobj, KOBJ_ADD); 756 dev_hold(queue->dev); 757 758 return error; 759 } 760 #endif /* CONFIG_RPS */ 761 762 int 763 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num) 764 { 765 #ifdef CONFIG_RPS 766 int i; 767 int error = 0; 768 769 for (i = old_num; i < new_num; i++) { 770 error = rx_queue_add_kobject(net, i); 771 if (error) { 772 new_num = old_num; 773 break; 774 } 775 } 776 777 while (--i >= new_num) 778 kobject_put(&net->_rx[i].kobj); 779 780 return error; 781 #else 782 return 0; 783 #endif 784 } 785 786 #ifdef CONFIG_XPS 787 /* 788 * netdev_queue sysfs structures and functions. 789 */ 790 struct netdev_queue_attribute { 791 struct attribute attr; 792 ssize_t (*show)(struct netdev_queue *queue, 793 struct netdev_queue_attribute *attr, char *buf); 794 ssize_t (*store)(struct netdev_queue *queue, 795 struct netdev_queue_attribute *attr, const char *buf, size_t len); 796 }; 797 #define to_netdev_queue_attr(_attr) container_of(_attr, \ 798 struct netdev_queue_attribute, attr) 799 800 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj) 801 802 static ssize_t netdev_queue_attr_show(struct kobject *kobj, 803 struct attribute *attr, char *buf) 804 { 805 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr); 806 struct netdev_queue *queue = to_netdev_queue(kobj); 807 808 if (!attribute->show) 809 return -EIO; 810 811 return attribute->show(queue, attribute, buf); 812 } 813 814 static ssize_t netdev_queue_attr_store(struct kobject *kobj, 815 struct attribute *attr, 816 const char *buf, size_t count) 817 { 818 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr); 819 struct netdev_queue *queue = to_netdev_queue(kobj); 820 821 if (!attribute->store) 822 return -EIO; 823 824 return attribute->store(queue, attribute, buf, count); 825 } 826 827 static const struct sysfs_ops netdev_queue_sysfs_ops = { 828 .show = netdev_queue_attr_show, 829 .store = netdev_queue_attr_store, 830 }; 831 832 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue) 833 { 834 struct net_device *dev = queue->dev; 835 int i; 836 837 for (i = 0; i < dev->num_tx_queues; i++) 838 if (queue == &dev->_tx[i]) 839 break; 840 841 BUG_ON(i >= dev->num_tx_queues); 842 843 return i; 844 } 845 846 847 static ssize_t show_xps_map(struct netdev_queue *queue, 848 struct netdev_queue_attribute *attribute, char *buf) 849 { 850 struct net_device *dev = queue->dev; 851 struct xps_dev_maps *dev_maps; 852 cpumask_var_t mask; 853 unsigned long index; 854 size_t len = 0; 855 int i; 856 857 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) 858 return -ENOMEM; 859 860 index = get_netdev_queue_index(queue); 861 862 rcu_read_lock(); 863 dev_maps = rcu_dereference(dev->xps_maps); 864 if (dev_maps) { 865 for_each_possible_cpu(i) { 866 struct xps_map *map = 867 rcu_dereference(dev_maps->cpu_map[i]); 868 if (map) { 869 int j; 870 for (j = 0; j < map->len; j++) { 871 if (map->queues[j] == index) { 872 cpumask_set_cpu(i, mask); 873 break; 874 } 875 } 876 } 877 } 878 } 879 rcu_read_unlock(); 880 881 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask); 882 if (PAGE_SIZE - len < 3) { 883 free_cpumask_var(mask); 884 return -EINVAL; 885 } 886 887 free_cpumask_var(mask); 888 len += sprintf(buf + len, "\n"); 889 return len; 890 } 891 892 static DEFINE_MUTEX(xps_map_mutex); 893 #define xmap_dereference(P) \ 894 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) 895 896 static ssize_t store_xps_map(struct netdev_queue *queue, 897 struct netdev_queue_attribute *attribute, 898 const char *buf, size_t len) 899 { 900 struct net_device *dev = queue->dev; 901 cpumask_var_t mask; 902 int err, i, cpu, pos, map_len, alloc_len, need_set; 903 unsigned long index; 904 struct xps_map *map, *new_map; 905 struct xps_dev_maps *dev_maps, *new_dev_maps; 906 int nonempty = 0; 907 int numa_node = -2; 908 909 if (!capable(CAP_NET_ADMIN)) 910 return -EPERM; 911 912 if (!alloc_cpumask_var(&mask, GFP_KERNEL)) 913 return -ENOMEM; 914 915 index = get_netdev_queue_index(queue); 916 917 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits); 918 if (err) { 919 free_cpumask_var(mask); 920 return err; 921 } 922 923 new_dev_maps = kzalloc(max_t(unsigned, 924 XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL); 925 if (!new_dev_maps) { 926 free_cpumask_var(mask); 927 return -ENOMEM; 928 } 929 930 mutex_lock(&xps_map_mutex); 931 932 dev_maps = xmap_dereference(dev->xps_maps); 933 934 for_each_possible_cpu(cpu) { 935 map = dev_maps ? 936 xmap_dereference(dev_maps->cpu_map[cpu]) : NULL; 937 new_map = map; 938 if (map) { 939 for (pos = 0; pos < map->len; pos++) 940 if (map->queues[pos] == index) 941 break; 942 map_len = map->len; 943 alloc_len = map->alloc_len; 944 } else 945 pos = map_len = alloc_len = 0; 946 947 need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu); 948 #ifdef CONFIG_NUMA 949 if (need_set) { 950 if (numa_node == -2) 951 numa_node = cpu_to_node(cpu); 952 else if (numa_node != cpu_to_node(cpu)) 953 numa_node = -1; 954 } 955 #endif 956 if (need_set && pos >= map_len) { 957 /* Need to add queue to this CPU's map */ 958 if (map_len >= alloc_len) { 959 alloc_len = alloc_len ? 960 2 * alloc_len : XPS_MIN_MAP_ALLOC; 961 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), 962 GFP_KERNEL, 963 cpu_to_node(cpu)); 964 if (!new_map) 965 goto error; 966 new_map->alloc_len = alloc_len; 967 for (i = 0; i < map_len; i++) 968 new_map->queues[i] = map->queues[i]; 969 new_map->len = map_len; 970 } 971 new_map->queues[new_map->len++] = index; 972 } else if (!need_set && pos < map_len) { 973 /* Need to remove queue from this CPU's map */ 974 if (map_len > 1) 975 new_map->queues[pos] = 976 new_map->queues[--new_map->len]; 977 else 978 new_map = NULL; 979 } 980 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map); 981 } 982 983 /* Cleanup old maps */ 984 for_each_possible_cpu(cpu) { 985 map = dev_maps ? 986 xmap_dereference(dev_maps->cpu_map[cpu]) : NULL; 987 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map) 988 kfree_rcu(map, rcu); 989 if (new_dev_maps->cpu_map[cpu]) 990 nonempty = 1; 991 } 992 993 if (nonempty) 994 RCU_INIT_POINTER(dev->xps_maps, new_dev_maps); 995 else { 996 kfree(new_dev_maps); 997 RCU_INIT_POINTER(dev->xps_maps, NULL); 998 } 999 1000 if (dev_maps) 1001 kfree_rcu(dev_maps, rcu); 1002 1003 netdev_queue_numa_node_write(queue, (numa_node >= 0) ? numa_node : 1004 NUMA_NO_NODE); 1005 1006 mutex_unlock(&xps_map_mutex); 1007 1008 free_cpumask_var(mask); 1009 return len; 1010 1011 error: 1012 mutex_unlock(&xps_map_mutex); 1013 1014 if (new_dev_maps) 1015 for_each_possible_cpu(i) 1016 kfree(rcu_dereference_protected( 1017 new_dev_maps->cpu_map[i], 1018 1)); 1019 kfree(new_dev_maps); 1020 free_cpumask_var(mask); 1021 return -ENOMEM; 1022 } 1023 1024 static struct netdev_queue_attribute xps_cpus_attribute = 1025 __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map); 1026 1027 static struct attribute *netdev_queue_default_attrs[] = { 1028 &xps_cpus_attribute.attr, 1029 NULL 1030 }; 1031 1032 static void netdev_queue_release(struct kobject *kobj) 1033 { 1034 struct netdev_queue *queue = to_netdev_queue(kobj); 1035 struct net_device *dev = queue->dev; 1036 struct xps_dev_maps *dev_maps; 1037 struct xps_map *map; 1038 unsigned long index; 1039 int i, pos, nonempty = 0; 1040 1041 index = get_netdev_queue_index(queue); 1042 1043 mutex_lock(&xps_map_mutex); 1044 dev_maps = xmap_dereference(dev->xps_maps); 1045 1046 if (dev_maps) { 1047 for_each_possible_cpu(i) { 1048 map = xmap_dereference(dev_maps->cpu_map[i]); 1049 if (!map) 1050 continue; 1051 1052 for (pos = 0; pos < map->len; pos++) 1053 if (map->queues[pos] == index) 1054 break; 1055 1056 if (pos < map->len) { 1057 if (map->len > 1) 1058 map->queues[pos] = 1059 map->queues[--map->len]; 1060 else { 1061 RCU_INIT_POINTER(dev_maps->cpu_map[i], 1062 NULL); 1063 kfree_rcu(map, rcu); 1064 map = NULL; 1065 } 1066 } 1067 if (map) 1068 nonempty = 1; 1069 } 1070 1071 if (!nonempty) { 1072 RCU_INIT_POINTER(dev->xps_maps, NULL); 1073 kfree_rcu(dev_maps, rcu); 1074 } 1075 } 1076 1077 mutex_unlock(&xps_map_mutex); 1078 1079 memset(kobj, 0, sizeof(*kobj)); 1080 dev_put(queue->dev); 1081 } 1082 1083 static struct kobj_type netdev_queue_ktype = { 1084 .sysfs_ops = &netdev_queue_sysfs_ops, 1085 .release = netdev_queue_release, 1086 .default_attrs = netdev_queue_default_attrs, 1087 }; 1088 1089 static int netdev_queue_add_kobject(struct net_device *net, int index) 1090 { 1091 struct netdev_queue *queue = net->_tx + index; 1092 struct kobject *kobj = &queue->kobj; 1093 int error = 0; 1094 1095 kobj->kset = net->queues_kset; 1096 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL, 1097 "tx-%u", index); 1098 if (error) { 1099 kobject_put(kobj); 1100 return error; 1101 } 1102 1103 kobject_uevent(kobj, KOBJ_ADD); 1104 dev_hold(queue->dev); 1105 1106 return error; 1107 } 1108 #endif /* CONFIG_XPS */ 1109 1110 int 1111 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num) 1112 { 1113 #ifdef CONFIG_XPS 1114 int i; 1115 int error = 0; 1116 1117 for (i = old_num; i < new_num; i++) { 1118 error = netdev_queue_add_kobject(net, i); 1119 if (error) { 1120 new_num = old_num; 1121 break; 1122 } 1123 } 1124 1125 while (--i >= new_num) 1126 kobject_put(&net->_tx[i].kobj); 1127 1128 return error; 1129 #else 1130 return 0; 1131 #endif 1132 } 1133 1134 static int register_queue_kobjects(struct net_device *net) 1135 { 1136 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0; 1137 1138 #if defined(CONFIG_RPS) || defined(CONFIG_XPS) 1139 net->queues_kset = kset_create_and_add("queues", 1140 NULL, &net->dev.kobj); 1141 if (!net->queues_kset) 1142 return -ENOMEM; 1143 #endif 1144 1145 #ifdef CONFIG_RPS 1146 real_rx = net->real_num_rx_queues; 1147 #endif 1148 real_tx = net->real_num_tx_queues; 1149 1150 error = net_rx_queue_update_kobjects(net, 0, real_rx); 1151 if (error) 1152 goto error; 1153 rxq = real_rx; 1154 1155 error = netdev_queue_update_kobjects(net, 0, real_tx); 1156 if (error) 1157 goto error; 1158 txq = real_tx; 1159 1160 return 0; 1161 1162 error: 1163 netdev_queue_update_kobjects(net, txq, 0); 1164 net_rx_queue_update_kobjects(net, rxq, 0); 1165 return error; 1166 } 1167 1168 static void remove_queue_kobjects(struct net_device *net) 1169 { 1170 int real_rx = 0, real_tx = 0; 1171 1172 #ifdef CONFIG_RPS 1173 real_rx = net->real_num_rx_queues; 1174 #endif 1175 real_tx = net->real_num_tx_queues; 1176 1177 net_rx_queue_update_kobjects(net, real_rx, 0); 1178 netdev_queue_update_kobjects(net, real_tx, 0); 1179 #if defined(CONFIG_RPS) || defined(CONFIG_XPS) 1180 kset_unregister(net->queues_kset); 1181 #endif 1182 } 1183 1184 static void *net_grab_current_ns(void) 1185 { 1186 struct net *ns = current->nsproxy->net_ns; 1187 #ifdef CONFIG_NET_NS 1188 if (ns) 1189 atomic_inc(&ns->passive); 1190 #endif 1191 return ns; 1192 } 1193 1194 static const void *net_initial_ns(void) 1195 { 1196 return &init_net; 1197 } 1198 1199 static const void *net_netlink_ns(struct sock *sk) 1200 { 1201 return sock_net(sk); 1202 } 1203 1204 struct kobj_ns_type_operations net_ns_type_operations = { 1205 .type = KOBJ_NS_TYPE_NET, 1206 .grab_current_ns = net_grab_current_ns, 1207 .netlink_ns = net_netlink_ns, 1208 .initial_ns = net_initial_ns, 1209 .drop_ns = net_drop_ns, 1210 }; 1211 EXPORT_SYMBOL_GPL(net_ns_type_operations); 1212 1213 #ifdef CONFIG_HOTPLUG 1214 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env) 1215 { 1216 struct net_device *dev = to_net_dev(d); 1217 int retval; 1218 1219 /* pass interface to uevent. */ 1220 retval = add_uevent_var(env, "INTERFACE=%s", dev->name); 1221 if (retval) 1222 goto exit; 1223 1224 /* pass ifindex to uevent. 1225 * ifindex is useful as it won't change (interface name may change) 1226 * and is what RtNetlink uses natively. */ 1227 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex); 1228 1229 exit: 1230 return retval; 1231 } 1232 #endif 1233 1234 /* 1235 * netdev_release -- destroy and free a dead device. 1236 * Called when last reference to device kobject is gone. 1237 */ 1238 static void netdev_release(struct device *d) 1239 { 1240 struct net_device *dev = to_net_dev(d); 1241 1242 BUG_ON(dev->reg_state != NETREG_RELEASED); 1243 1244 kfree(dev->ifalias); 1245 kfree((char *)dev - dev->padded); 1246 } 1247 1248 static const void *net_namespace(struct device *d) 1249 { 1250 struct net_device *dev; 1251 dev = container_of(d, struct net_device, dev); 1252 return dev_net(dev); 1253 } 1254 1255 static struct class net_class = { 1256 .name = "net", 1257 .dev_release = netdev_release, 1258 #ifdef CONFIG_SYSFS 1259 .dev_attrs = net_class_attributes, 1260 #endif /* CONFIG_SYSFS */ 1261 #ifdef CONFIG_HOTPLUG 1262 .dev_uevent = netdev_uevent, 1263 #endif 1264 .ns_type = &net_ns_type_operations, 1265 .namespace = net_namespace, 1266 }; 1267 1268 /* Delete sysfs entries but hold kobject reference until after all 1269 * netdev references are gone. 1270 */ 1271 void netdev_unregister_kobject(struct net_device * net) 1272 { 1273 struct device *dev = &(net->dev); 1274 1275 kobject_get(&dev->kobj); 1276 1277 remove_queue_kobjects(net); 1278 1279 device_del(dev); 1280 } 1281 1282 /* Create sysfs entries for network device. */ 1283 int netdev_register_kobject(struct net_device *net) 1284 { 1285 struct device *dev = &(net->dev); 1286 const struct attribute_group **groups = net->sysfs_groups; 1287 int error = 0; 1288 1289 device_initialize(dev); 1290 dev->class = &net_class; 1291 dev->platform_data = net; 1292 dev->groups = groups; 1293 1294 dev_set_name(dev, "%s", net->name); 1295 1296 #ifdef CONFIG_SYSFS 1297 /* Allow for a device specific group */ 1298 if (*groups) 1299 groups++; 1300 1301 *groups++ = &netstat_group; 1302 #ifdef CONFIG_WIRELESS_EXT_SYSFS 1303 if (net->ieee80211_ptr) 1304 *groups++ = &wireless_group; 1305 #ifdef CONFIG_WIRELESS_EXT 1306 else if (net->wireless_handlers) 1307 *groups++ = &wireless_group; 1308 #endif 1309 #endif 1310 #endif /* CONFIG_SYSFS */ 1311 1312 error = device_add(dev); 1313 if (error) 1314 return error; 1315 1316 error = register_queue_kobjects(net); 1317 if (error) { 1318 device_del(dev); 1319 return error; 1320 } 1321 1322 return error; 1323 } 1324 1325 int netdev_class_create_file(struct class_attribute *class_attr) 1326 { 1327 return class_create_file(&net_class, class_attr); 1328 } 1329 EXPORT_SYMBOL(netdev_class_create_file); 1330 1331 void netdev_class_remove_file(struct class_attribute *class_attr) 1332 { 1333 class_remove_file(&net_class, class_attr); 1334 } 1335 EXPORT_SYMBOL(netdev_class_remove_file); 1336 1337 int netdev_kobject_init(void) 1338 { 1339 kobj_ns_type_register(&net_ns_type_operations); 1340 return class_register(&net_class); 1341 } 1342