eth.c (17569B)
1// SPDX-License-Identifier: GPL-2.0-or-later 2/* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Ethernet-type device handling. 8 * 9 * Version: @(#)eth.c 1.0.7 05/25/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Mark Evans, <evansmp@uhura.aston.ac.uk> 14 * Florian La Roche, <rzsfl@rz.uni-sb.de> 15 * Alan Cox, <gw4pts@gw4pts.ampr.org> 16 * 17 * Fixes: 18 * Mr Linux : Arp problems 19 * Alan Cox : Generic queue tidyup (very tiny here) 20 * Alan Cox : eth_header ntohs should be htons 21 * Alan Cox : eth_rebuild_header missing an htons and 22 * minor other things. 23 * Tegge : Arp bug fixes. 24 * Florian : Removed many unnecessary functions, code cleanup 25 * and changes for new arp and skbuff. 26 * Alan Cox : Redid header building to reflect new format. 27 * Alan Cox : ARP only when compiled with CONFIG_INET 28 * Greg Page : 802.2 and SNAP stuff. 29 * Alan Cox : MAC layer pointers/new format. 30 * Paul Gortmaker : eth_copy_and_sum shouldn't csum padding. 31 * Alan Cox : Protect against forwarding explosions with 32 * older network drivers and IFF_ALLMULTI. 33 * Christer Weinigel : Better rebuild header message. 34 * Andrew Morton : 26Feb01: kill ether_setup() - use netdev_boot_setup(). 35 */ 36#include <linux/module.h> 37#include <linux/types.h> 38#include <linux/kernel.h> 39#include <linux/string.h> 40#include <linux/mm.h> 41#include <linux/socket.h> 42#include <linux/in.h> 43#include <linux/inet.h> 44#include <linux/ip.h> 45#include <linux/netdevice.h> 46#include <linux/nvmem-consumer.h> 47#include <linux/etherdevice.h> 48#include <linux/skbuff.h> 49#include <linux/errno.h> 50#include <linux/init.h> 51#include <linux/if_ether.h> 52#include <linux/of_net.h> 53#include <linux/pci.h> 54#include <linux/property.h> 55#include <net/dst.h> 56#include <net/arp.h> 57#include <net/sock.h> 58#include <net/ipv6.h> 59#include <net/ip.h> 60#include <net/dsa.h> 61#include <net/flow_dissector.h> 62#include <net/gro.h> 63#include <linux/uaccess.h> 64#include <net/pkt_sched.h> 65 66/** 67 * eth_header - create the Ethernet header 68 * @skb: buffer to alter 69 * @dev: source device 70 * @type: Ethernet type field 71 * @daddr: destination address (NULL leave destination address) 72 * @saddr: source address (NULL use device source address) 73 * @len: packet length (<= skb->len) 74 * 75 * 76 * Set the protocol type. For a packet of type ETH_P_802_3/2 we put the length 77 * in here instead. 78 */ 79int eth_header(struct sk_buff *skb, struct net_device *dev, 80 unsigned short type, 81 const void *daddr, const void *saddr, unsigned int len) 82{ 83 struct ethhdr *eth = skb_push(skb, ETH_HLEN); 84 85 if (type != ETH_P_802_3 && type != ETH_P_802_2) 86 eth->h_proto = htons(type); 87 else 88 eth->h_proto = htons(len); 89 90 /* 91 * Set the source hardware address. 92 */ 93 94 if (!saddr) 95 saddr = dev->dev_addr; 96 memcpy(eth->h_source, saddr, ETH_ALEN); 97 98 if (daddr) { 99 memcpy(eth->h_dest, daddr, ETH_ALEN); 100 return ETH_HLEN; 101 } 102 103 /* 104 * Anyway, the loopback-device should never use this function... 105 */ 106 107 if (dev->flags & (IFF_LOOPBACK | IFF_NOARP)) { 108 eth_zero_addr(eth->h_dest); 109 return ETH_HLEN; 110 } 111 112 return -ETH_HLEN; 113} 114EXPORT_SYMBOL(eth_header); 115 116/** 117 * eth_get_headlen - determine the length of header for an ethernet frame 118 * @dev: pointer to network device 119 * @data: pointer to start of frame 120 * @len: total length of frame 121 * 122 * Make a best effort attempt to pull the length for all of the headers for 123 * a given frame in a linear buffer. 124 */ 125u32 eth_get_headlen(const struct net_device *dev, const void *data, u32 len) 126{ 127 const unsigned int flags = FLOW_DISSECTOR_F_PARSE_1ST_FRAG; 128 const struct ethhdr *eth = (const struct ethhdr *)data; 129 struct flow_keys_basic keys; 130 131 /* this should never happen, but better safe than sorry */ 132 if (unlikely(len < sizeof(*eth))) 133 return len; 134 135 /* parse any remaining L2/L3 headers, check for L4 */ 136 if (!skb_flow_dissect_flow_keys_basic(dev_net(dev), NULL, &keys, data, 137 eth->h_proto, sizeof(*eth), 138 len, flags)) 139 return max_t(u32, keys.control.thoff, sizeof(*eth)); 140 141 /* parse for any L4 headers */ 142 return min_t(u32, __skb_get_poff(NULL, data, &keys, len), len); 143} 144EXPORT_SYMBOL(eth_get_headlen); 145 146/** 147 * eth_type_trans - determine the packet's protocol ID. 148 * @skb: received socket data 149 * @dev: receiving network device 150 * 151 * The rule here is that we 152 * assume 802.3 if the type field is short enough to be a length. 153 * This is normal practice and works for any 'now in use' protocol. 154 */ 155__be16 eth_type_trans(struct sk_buff *skb, struct net_device *dev) 156{ 157 unsigned short _service_access_point; 158 const unsigned short *sap; 159 const struct ethhdr *eth; 160 161 skb->dev = dev; 162 skb_reset_mac_header(skb); 163 164 eth = (struct ethhdr *)skb->data; 165 skb_pull_inline(skb, ETH_HLEN); 166 167 if (unlikely(!ether_addr_equal_64bits(eth->h_dest, 168 dev->dev_addr))) { 169 if (unlikely(is_multicast_ether_addr_64bits(eth->h_dest))) { 170 if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast)) 171 skb->pkt_type = PACKET_BROADCAST; 172 else 173 skb->pkt_type = PACKET_MULTICAST; 174 } else { 175 skb->pkt_type = PACKET_OTHERHOST; 176 } 177 } 178 179 /* 180 * Some variants of DSA tagging don't have an ethertype field 181 * at all, so we check here whether one of those tagging 182 * variants has been configured on the receiving interface, 183 * and if so, set skb->protocol without looking at the packet. 184 */ 185 if (unlikely(netdev_uses_dsa(dev))) 186 return htons(ETH_P_XDSA); 187 188 if (likely(eth_proto_is_802_3(eth->h_proto))) 189 return eth->h_proto; 190 191 /* 192 * This is a magic hack to spot IPX packets. Older Novell breaks 193 * the protocol design and runs IPX over 802.3 without an 802.2 LLC 194 * layer. We look for FFFF which isn't a used 802.2 SSAP/DSAP. This 195 * won't work for fault tolerant netware but does for the rest. 196 */ 197 sap = skb_header_pointer(skb, 0, sizeof(*sap), &_service_access_point); 198 if (sap && *sap == 0xFFFF) 199 return htons(ETH_P_802_3); 200 201 /* 202 * Real 802.2 LLC 203 */ 204 return htons(ETH_P_802_2); 205} 206EXPORT_SYMBOL(eth_type_trans); 207 208/** 209 * eth_header_parse - extract hardware address from packet 210 * @skb: packet to extract header from 211 * @haddr: destination buffer 212 */ 213int eth_header_parse(const struct sk_buff *skb, unsigned char *haddr) 214{ 215 const struct ethhdr *eth = eth_hdr(skb); 216 memcpy(haddr, eth->h_source, ETH_ALEN); 217 return ETH_ALEN; 218} 219EXPORT_SYMBOL(eth_header_parse); 220 221/** 222 * eth_header_cache - fill cache entry from neighbour 223 * @neigh: source neighbour 224 * @hh: destination cache entry 225 * @type: Ethernet type field 226 * 227 * Create an Ethernet header template from the neighbour. 228 */ 229int eth_header_cache(const struct neighbour *neigh, struct hh_cache *hh, __be16 type) 230{ 231 struct ethhdr *eth; 232 const struct net_device *dev = neigh->dev; 233 234 eth = (struct ethhdr *) 235 (((u8 *) hh->hh_data) + (HH_DATA_OFF(sizeof(*eth)))); 236 237 if (type == htons(ETH_P_802_3)) 238 return -1; 239 240 eth->h_proto = type; 241 memcpy(eth->h_source, dev->dev_addr, ETH_ALEN); 242 memcpy(eth->h_dest, neigh->ha, ETH_ALEN); 243 244 /* Pairs with READ_ONCE() in neigh_resolve_output(), 245 * neigh_hh_output() and neigh_update_hhs(). 246 */ 247 smp_store_release(&hh->hh_len, ETH_HLEN); 248 249 return 0; 250} 251EXPORT_SYMBOL(eth_header_cache); 252 253/** 254 * eth_header_cache_update - update cache entry 255 * @hh: destination cache entry 256 * @dev: network device 257 * @haddr: new hardware address 258 * 259 * Called by Address Resolution module to notify changes in address. 260 */ 261void eth_header_cache_update(struct hh_cache *hh, 262 const struct net_device *dev, 263 const unsigned char *haddr) 264{ 265 memcpy(((u8 *) hh->hh_data) + HH_DATA_OFF(sizeof(struct ethhdr)), 266 haddr, ETH_ALEN); 267} 268EXPORT_SYMBOL(eth_header_cache_update); 269 270/** 271 * eth_header_parse_protocol - extract protocol from L2 header 272 * @skb: packet to extract protocol from 273 */ 274__be16 eth_header_parse_protocol(const struct sk_buff *skb) 275{ 276 const struct ethhdr *eth = eth_hdr(skb); 277 278 return eth->h_proto; 279} 280EXPORT_SYMBOL(eth_header_parse_protocol); 281 282/** 283 * eth_prepare_mac_addr_change - prepare for mac change 284 * @dev: network device 285 * @p: socket address 286 */ 287int eth_prepare_mac_addr_change(struct net_device *dev, void *p) 288{ 289 struct sockaddr *addr = p; 290 291 if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev)) 292 return -EBUSY; 293 if (!is_valid_ether_addr(addr->sa_data)) 294 return -EADDRNOTAVAIL; 295 return 0; 296} 297EXPORT_SYMBOL(eth_prepare_mac_addr_change); 298 299/** 300 * eth_commit_mac_addr_change - commit mac change 301 * @dev: network device 302 * @p: socket address 303 */ 304void eth_commit_mac_addr_change(struct net_device *dev, void *p) 305{ 306 struct sockaddr *addr = p; 307 308 eth_hw_addr_set(dev, addr->sa_data); 309} 310EXPORT_SYMBOL(eth_commit_mac_addr_change); 311 312/** 313 * eth_mac_addr - set new Ethernet hardware address 314 * @dev: network device 315 * @p: socket address 316 * 317 * Change hardware address of device. 318 * 319 * This doesn't change hardware matching, so needs to be overridden 320 * for most real devices. 321 */ 322int eth_mac_addr(struct net_device *dev, void *p) 323{ 324 int ret; 325 326 ret = eth_prepare_mac_addr_change(dev, p); 327 if (ret < 0) 328 return ret; 329 eth_commit_mac_addr_change(dev, p); 330 return 0; 331} 332EXPORT_SYMBOL(eth_mac_addr); 333 334int eth_validate_addr(struct net_device *dev) 335{ 336 if (!is_valid_ether_addr(dev->dev_addr)) 337 return -EADDRNOTAVAIL; 338 339 return 0; 340} 341EXPORT_SYMBOL(eth_validate_addr); 342 343const struct header_ops eth_header_ops ____cacheline_aligned = { 344 .create = eth_header, 345 .parse = eth_header_parse, 346 .cache = eth_header_cache, 347 .cache_update = eth_header_cache_update, 348 .parse_protocol = eth_header_parse_protocol, 349}; 350 351/** 352 * ether_setup - setup Ethernet network device 353 * @dev: network device 354 * 355 * Fill in the fields of the device structure with Ethernet-generic values. 356 */ 357void ether_setup(struct net_device *dev) 358{ 359 dev->header_ops = ð_header_ops; 360 dev->type = ARPHRD_ETHER; 361 dev->hard_header_len = ETH_HLEN; 362 dev->min_header_len = ETH_HLEN; 363 dev->mtu = ETH_DATA_LEN; 364 dev->min_mtu = ETH_MIN_MTU; 365 dev->max_mtu = ETH_DATA_LEN; 366 dev->addr_len = ETH_ALEN; 367 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; 368 dev->flags = IFF_BROADCAST|IFF_MULTICAST; 369 dev->priv_flags |= IFF_TX_SKB_SHARING; 370 371 eth_broadcast_addr(dev->broadcast); 372 373} 374EXPORT_SYMBOL(ether_setup); 375 376/** 377 * alloc_etherdev_mqs - Allocates and sets up an Ethernet device 378 * @sizeof_priv: Size of additional driver-private structure to be allocated 379 * for this Ethernet device 380 * @txqs: The number of TX queues this device has. 381 * @rxqs: The number of RX queues this device has. 382 * 383 * Fill in the fields of the device structure with Ethernet-generic 384 * values. Basically does everything except registering the device. 385 * 386 * Constructs a new net device, complete with a private data area of 387 * size (sizeof_priv). A 32-byte (not bit) alignment is enforced for 388 * this private data area. 389 */ 390 391struct net_device *alloc_etherdev_mqs(int sizeof_priv, unsigned int txqs, 392 unsigned int rxqs) 393{ 394 return alloc_netdev_mqs(sizeof_priv, "eth%d", NET_NAME_ENUM, 395 ether_setup, txqs, rxqs); 396} 397EXPORT_SYMBOL(alloc_etherdev_mqs); 398 399ssize_t sysfs_format_mac(char *buf, const unsigned char *addr, int len) 400{ 401 return scnprintf(buf, PAGE_SIZE, "%*phC\n", len, addr); 402} 403EXPORT_SYMBOL(sysfs_format_mac); 404 405struct sk_buff *eth_gro_receive(struct list_head *head, struct sk_buff *skb) 406{ 407 const struct packet_offload *ptype; 408 unsigned int hlen, off_eth; 409 struct sk_buff *pp = NULL; 410 struct ethhdr *eh, *eh2; 411 struct sk_buff *p; 412 __be16 type; 413 int flush = 1; 414 415 off_eth = skb_gro_offset(skb); 416 hlen = off_eth + sizeof(*eh); 417 eh = skb_gro_header_fast(skb, off_eth); 418 if (skb_gro_header_hard(skb, hlen)) { 419 eh = skb_gro_header_slow(skb, hlen, off_eth); 420 if (unlikely(!eh)) 421 goto out; 422 } 423 424 flush = 0; 425 426 list_for_each_entry(p, head, list) { 427 if (!NAPI_GRO_CB(p)->same_flow) 428 continue; 429 430 eh2 = (struct ethhdr *)(p->data + off_eth); 431 if (compare_ether_header(eh, eh2)) { 432 NAPI_GRO_CB(p)->same_flow = 0; 433 continue; 434 } 435 } 436 437 type = eh->h_proto; 438 439 ptype = gro_find_receive_by_type(type); 440 if (ptype == NULL) { 441 flush = 1; 442 goto out; 443 } 444 445 skb_gro_pull(skb, sizeof(*eh)); 446 skb_gro_postpull_rcsum(skb, eh, sizeof(*eh)); 447 448 pp = indirect_call_gro_receive_inet(ptype->callbacks.gro_receive, 449 ipv6_gro_receive, inet_gro_receive, 450 head, skb); 451 452out: 453 skb_gro_flush_final(skb, pp, flush); 454 455 return pp; 456} 457EXPORT_SYMBOL(eth_gro_receive); 458 459int eth_gro_complete(struct sk_buff *skb, int nhoff) 460{ 461 struct ethhdr *eh = (struct ethhdr *)(skb->data + nhoff); 462 __be16 type = eh->h_proto; 463 struct packet_offload *ptype; 464 int err = -ENOSYS; 465 466 if (skb->encapsulation) 467 skb_set_inner_mac_header(skb, nhoff); 468 469 ptype = gro_find_complete_by_type(type); 470 if (ptype != NULL) 471 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, 472 ipv6_gro_complete, inet_gro_complete, 473 skb, nhoff + sizeof(*eh)); 474 475 return err; 476} 477EXPORT_SYMBOL(eth_gro_complete); 478 479static struct packet_offload eth_packet_offload __read_mostly = { 480 .type = cpu_to_be16(ETH_P_TEB), 481 .priority = 10, 482 .callbacks = { 483 .gro_receive = eth_gro_receive, 484 .gro_complete = eth_gro_complete, 485 }, 486}; 487 488static int __init eth_offload_init(void) 489{ 490 dev_add_offload(ð_packet_offload); 491 492 return 0; 493} 494 495fs_initcall(eth_offload_init); 496 497unsigned char * __weak arch_get_platform_mac_address(void) 498{ 499 return NULL; 500} 501 502int eth_platform_get_mac_address(struct device *dev, u8 *mac_addr) 503{ 504 unsigned char *addr; 505 int ret; 506 507 ret = of_get_mac_address(dev->of_node, mac_addr); 508 if (!ret) 509 return 0; 510 511 addr = arch_get_platform_mac_address(); 512 if (!addr) 513 return -ENODEV; 514 515 ether_addr_copy(mac_addr, addr); 516 517 return 0; 518} 519EXPORT_SYMBOL(eth_platform_get_mac_address); 520 521/** 522 * platform_get_ethdev_address - Set netdev's MAC address from a given device 523 * @dev: Pointer to the device 524 * @netdev: Pointer to netdev to write the address to 525 * 526 * Wrapper around eth_platform_get_mac_address() which writes the address 527 * directly to netdev->dev_addr. 528 */ 529int platform_get_ethdev_address(struct device *dev, struct net_device *netdev) 530{ 531 u8 addr[ETH_ALEN] __aligned(2); 532 int ret; 533 534 ret = eth_platform_get_mac_address(dev, addr); 535 if (!ret) 536 eth_hw_addr_set(netdev, addr); 537 return ret; 538} 539EXPORT_SYMBOL(platform_get_ethdev_address); 540 541/** 542 * nvmem_get_mac_address - Obtain the MAC address from an nvmem cell named 543 * 'mac-address' associated with given device. 544 * 545 * @dev: Device with which the mac-address cell is associated. 546 * @addrbuf: Buffer to which the MAC address will be copied on success. 547 * 548 * Returns 0 on success or a negative error number on failure. 549 */ 550int nvmem_get_mac_address(struct device *dev, void *addrbuf) 551{ 552 struct nvmem_cell *cell; 553 const void *mac; 554 size_t len; 555 556 cell = nvmem_cell_get(dev, "mac-address"); 557 if (IS_ERR(cell)) 558 return PTR_ERR(cell); 559 560 mac = nvmem_cell_read(cell, &len); 561 nvmem_cell_put(cell); 562 563 if (IS_ERR(mac)) 564 return PTR_ERR(mac); 565 566 if (len != ETH_ALEN || !is_valid_ether_addr(mac)) { 567 kfree(mac); 568 return -EINVAL; 569 } 570 571 ether_addr_copy(addrbuf, mac); 572 kfree(mac); 573 574 return 0; 575} 576 577static int fwnode_get_mac_addr(struct fwnode_handle *fwnode, 578 const char *name, char *addr) 579{ 580 int ret; 581 582 ret = fwnode_property_read_u8_array(fwnode, name, addr, ETH_ALEN); 583 if (ret) 584 return ret; 585 586 if (!is_valid_ether_addr(addr)) 587 return -EINVAL; 588 return 0; 589} 590 591/** 592 * fwnode_get_mac_address - Get the MAC from the firmware node 593 * @fwnode: Pointer to the firmware node 594 * @addr: Address of buffer to store the MAC in 595 * 596 * Search the firmware node for the best MAC address to use. 'mac-address' is 597 * checked first, because that is supposed to contain to "most recent" MAC 598 * address. If that isn't set, then 'local-mac-address' is checked next, 599 * because that is the default address. If that isn't set, then the obsolete 600 * 'address' is checked, just in case we're using an old device tree. 601 * 602 * Note that the 'address' property is supposed to contain a virtual address of 603 * the register set, but some DTS files have redefined that property to be the 604 * MAC address. 605 * 606 * All-zero MAC addresses are rejected, because those could be properties that 607 * exist in the firmware tables, but were not updated by the firmware. For 608 * example, the DTS could define 'mac-address' and 'local-mac-address', with 609 * zero MAC addresses. Some older U-Boots only initialized 'local-mac-address'. 610 * In this case, the real MAC is in 'local-mac-address', and 'mac-address' 611 * exists but is all zeros. 612 */ 613int fwnode_get_mac_address(struct fwnode_handle *fwnode, char *addr) 614{ 615 if (!fwnode_get_mac_addr(fwnode, "mac-address", addr) || 616 !fwnode_get_mac_addr(fwnode, "local-mac-address", addr) || 617 !fwnode_get_mac_addr(fwnode, "address", addr)) 618 return 0; 619 620 return -ENOENT; 621} 622EXPORT_SYMBOL(fwnode_get_mac_address); 623 624/** 625 * device_get_mac_address - Get the MAC for a given device 626 * @dev: Pointer to the device 627 * @addr: Address of buffer to store the MAC in 628 */ 629int device_get_mac_address(struct device *dev, char *addr) 630{ 631 return fwnode_get_mac_address(dev_fwnode(dev), addr); 632} 633EXPORT_SYMBOL(device_get_mac_address); 634 635/** 636 * device_get_ethdev_address - Set netdev's MAC address from a given device 637 * @dev: Pointer to the device 638 * @netdev: Pointer to netdev to write the address to 639 * 640 * Wrapper around device_get_mac_address() which writes the address 641 * directly to netdev->dev_addr. 642 */ 643int device_get_ethdev_address(struct device *dev, struct net_device *netdev) 644{ 645 u8 addr[ETH_ALEN]; 646 int ret; 647 648 ret = device_get_mac_address(dev, addr); 649 if (!ret) 650 eth_hw_addr_set(netdev, addr); 651 return ret; 652} 653EXPORT_SYMBOL(device_get_ethdev_address);