1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Ethernet-type device handling. 7 * 8 * Version: @(#)eth.c 1.0.7 05/25/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Mark Evans, <evansmp@uhura.aston.ac.uk> 13 * Florian La Roche, <rzsfl@rz.uni-sb.de> 14 * Alan Cox, <gw4pts@gw4pts.ampr.org> 15 * 16 * Fixes: 17 * Mr Linux : Arp problems 18 * Alan Cox : Generic queue tidyup (very tiny here) 19 * Alan Cox : eth_header ntohs should be htons 20 * Alan Cox : eth_rebuild_header missing an htons and 21 * minor other things. 22 * Tegge : Arp bug fixes. 23 * Florian : Removed many unnecessary functions, code cleanup 24 * and changes for new arp and skbuff. 25 * Alan Cox : Redid header building to reflect new format. 26 * Alan Cox : ARP only when compiled with CONFIG_INET 27 * Greg Page : 802.2 and SNAP stuff. 28 * Alan Cox : MAC layer pointers/new format. 29 * Paul Gortmaker : eth_copy_and_sum shouldn't csum padding. 30 * Alan Cox : Protect against forwarding explosions with 31 * older network drivers and IFF_ALLMULTI. 32 * Christer Weinigel : Better rebuild header message. 33 * Andrew Morton : 26Feb01: kill ether_setup() - use netdev_boot_setup(). 34 * 35 * This program is free software; you can redistribute it and/or 36 * modify it under the terms of the GNU General Public License 37 * as published by the Free Software Foundation; either version 38 * 2 of the License, or (at your option) any later version. 39 */ 40 #include <linux/module.h> 41 #include <linux/types.h> 42 #include <linux/kernel.h> 43 #include <linux/string.h> 44 #include <linux/mm.h> 45 #include <linux/socket.h> 46 #include <linux/in.h> 47 #include <linux/inet.h> 48 #include <linux/ip.h> 49 #include <linux/netdevice.h> 50 #include <linux/etherdevice.h> 51 #include <linux/skbuff.h> 52 #include <linux/errno.h> 53 #include <linux/init.h> 54 #include <linux/if_ether.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 <linux/uaccess.h> 62 63 __setup("ether=", netdev_boot_setup); 64 65 /** 66 * eth_header - create the Ethernet header 67 * @skb: buffer to alter 68 * @dev: source device 69 * @type: Ethernet type field 70 * @daddr: destination address (NULL leave destination address) 71 * @saddr: source address (NULL use device source address) 72 * @len: packet length (<= skb->len) 73 * 74 * 75 * Set the protocol type. For a packet of type ETH_P_802_3/2 we put the length 76 * in here instead. 77 */ 78 int eth_header(struct sk_buff *skb, struct net_device *dev, 79 unsigned short type, 80 const void *daddr, const void *saddr, unsigned int len) 81 { 82 struct ethhdr *eth = (struct ethhdr *)skb_push(skb, ETH_HLEN); 83 84 if (type != ETH_P_802_3 && type != ETH_P_802_2) 85 eth->h_proto = htons(type); 86 else 87 eth->h_proto = htons(len); 88 89 /* 90 * Set the source hardware address. 91 */ 92 93 if (!saddr) 94 saddr = dev->dev_addr; 95 memcpy(eth->h_source, saddr, ETH_ALEN); 96 97 if (daddr) { 98 memcpy(eth->h_dest, daddr, ETH_ALEN); 99 return ETH_HLEN; 100 } 101 102 /* 103 * Anyway, the loopback-device should never use this function... 104 */ 105 106 if (dev->flags & (IFF_LOOPBACK | IFF_NOARP)) { 107 memset(eth->h_dest, 0, ETH_ALEN); 108 return ETH_HLEN; 109 } 110 111 return -ETH_HLEN; 112 } 113 EXPORT_SYMBOL(eth_header); 114 115 /** 116 * eth_rebuild_header- rebuild the Ethernet MAC header. 117 * @skb: socket buffer to update 118 * 119 * This is called after an ARP or IPV6 ndisc it's resolution on this 120 * sk_buff. We now let protocol (ARP) fill in the other fields. 121 * 122 * This routine CANNOT use cached dst->neigh! 123 * Really, it is used only when dst->neigh is wrong. 124 */ 125 int eth_rebuild_header(struct sk_buff *skb) 126 { 127 struct ethhdr *eth = (struct ethhdr *)skb->data; 128 struct net_device *dev = skb->dev; 129 130 switch (eth->h_proto) { 131 #ifdef CONFIG_INET 132 case htons(ETH_P_IP): 133 return arp_find(eth->h_dest, skb); 134 #endif 135 default: 136 netdev_dbg(dev, 137 "%s: unable to resolve type %X addresses.\n", 138 dev->name, ntohs(eth->h_proto)); 139 140 memcpy(eth->h_source, dev->dev_addr, ETH_ALEN); 141 break; 142 } 143 144 return 0; 145 } 146 EXPORT_SYMBOL(eth_rebuild_header); 147 148 /** 149 * eth_get_headlen - determine the the length of header for an ethernet frame 150 * @data: pointer to start of frame 151 * @len: total length of frame 152 * 153 * Make a best effort attempt to pull the length for all of the headers for 154 * a given frame in a linear buffer. 155 */ 156 u32 eth_get_headlen(void *data, unsigned int len) 157 { 158 const struct ethhdr *eth = (const struct ethhdr *)data; 159 struct flow_keys keys; 160 161 /* this should never happen, but better safe than sorry */ 162 if (len < sizeof(*eth)) 163 return len; 164 165 /* parse any remaining L2/L3 headers, check for L4 */ 166 if (!__skb_flow_dissect(NULL, &keys, data, 167 eth->h_proto, sizeof(*eth), len)) 168 return max_t(u32, keys.thoff, sizeof(*eth)); 169 170 /* parse for any L4 headers */ 171 return min_t(u32, __skb_get_poff(NULL, data, &keys, len), len); 172 } 173 EXPORT_SYMBOL(eth_get_headlen); 174 175 /** 176 * eth_type_trans - determine the packet's protocol ID. 177 * @skb: received socket data 178 * @dev: receiving network device 179 * 180 * The rule here is that we 181 * assume 802.3 if the type field is short enough to be a length. 182 * This is normal practice and works for any 'now in use' protocol. 183 */ 184 __be16 eth_type_trans(struct sk_buff *skb, struct net_device *dev) 185 { 186 unsigned short _service_access_point; 187 const unsigned short *sap; 188 const struct ethhdr *eth; 189 190 skb->dev = dev; 191 skb_reset_mac_header(skb); 192 skb_pull_inline(skb, ETH_HLEN); 193 eth = eth_hdr(skb); 194 195 if (unlikely(is_multicast_ether_addr(eth->h_dest))) { 196 if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast)) 197 skb->pkt_type = PACKET_BROADCAST; 198 else 199 skb->pkt_type = PACKET_MULTICAST; 200 } 201 else if (unlikely(!ether_addr_equal_64bits(eth->h_dest, 202 dev->dev_addr))) 203 skb->pkt_type = PACKET_OTHERHOST; 204 205 /* 206 * Some variants of DSA tagging don't have an ethertype field 207 * at all, so we check here whether one of those tagging 208 * variants has been configured on the receiving interface, 209 * and if so, set skb->protocol without looking at the packet. 210 */ 211 if (unlikely(netdev_uses_dsa(dev))) 212 return htons(ETH_P_XDSA); 213 214 if (likely(ntohs(eth->h_proto) >= ETH_P_802_3_MIN)) 215 return eth->h_proto; 216 217 /* 218 * This is a magic hack to spot IPX packets. Older Novell breaks 219 * the protocol design and runs IPX over 802.3 without an 802.2 LLC 220 * layer. We look for FFFF which isn't a used 802.2 SSAP/DSAP. This 221 * won't work for fault tolerant netware but does for the rest. 222 */ 223 sap = skb_header_pointer(skb, 0, sizeof(*sap), &_service_access_point); 224 if (sap && *sap == 0xFFFF) 225 return htons(ETH_P_802_3); 226 227 /* 228 * Real 802.2 LLC 229 */ 230 return htons(ETH_P_802_2); 231 } 232 EXPORT_SYMBOL(eth_type_trans); 233 234 /** 235 * eth_header_parse - extract hardware address from packet 236 * @skb: packet to extract header from 237 * @haddr: destination buffer 238 */ 239 int eth_header_parse(const struct sk_buff *skb, unsigned char *haddr) 240 { 241 const struct ethhdr *eth = eth_hdr(skb); 242 memcpy(haddr, eth->h_source, ETH_ALEN); 243 return ETH_ALEN; 244 } 245 EXPORT_SYMBOL(eth_header_parse); 246 247 /** 248 * eth_header_cache - fill cache entry from neighbour 249 * @neigh: source neighbour 250 * @hh: destination cache entry 251 * @type: Ethernet type field 252 * 253 * Create an Ethernet header template from the neighbour. 254 */ 255 int eth_header_cache(const struct neighbour *neigh, struct hh_cache *hh, __be16 type) 256 { 257 struct ethhdr *eth; 258 const struct net_device *dev = neigh->dev; 259 260 eth = (struct ethhdr *) 261 (((u8 *) hh->hh_data) + (HH_DATA_OFF(sizeof(*eth)))); 262 263 if (type == htons(ETH_P_802_3)) 264 return -1; 265 266 eth->h_proto = type; 267 memcpy(eth->h_source, dev->dev_addr, ETH_ALEN); 268 memcpy(eth->h_dest, neigh->ha, ETH_ALEN); 269 hh->hh_len = ETH_HLEN; 270 return 0; 271 } 272 EXPORT_SYMBOL(eth_header_cache); 273 274 /** 275 * eth_header_cache_update - update cache entry 276 * @hh: destination cache entry 277 * @dev: network device 278 * @haddr: new hardware address 279 * 280 * Called by Address Resolution module to notify changes in address. 281 */ 282 void eth_header_cache_update(struct hh_cache *hh, 283 const struct net_device *dev, 284 const unsigned char *haddr) 285 { 286 memcpy(((u8 *) hh->hh_data) + HH_DATA_OFF(sizeof(struct ethhdr)), 287 haddr, ETH_ALEN); 288 } 289 EXPORT_SYMBOL(eth_header_cache_update); 290 291 /** 292 * eth_prepare_mac_addr_change - prepare for mac change 293 * @dev: network device 294 * @p: socket address 295 */ 296 int eth_prepare_mac_addr_change(struct net_device *dev, void *p) 297 { 298 struct sockaddr *addr = p; 299 300 if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev)) 301 return -EBUSY; 302 if (!is_valid_ether_addr(addr->sa_data)) 303 return -EADDRNOTAVAIL; 304 return 0; 305 } 306 EXPORT_SYMBOL(eth_prepare_mac_addr_change); 307 308 /** 309 * eth_commit_mac_addr_change - commit mac change 310 * @dev: network device 311 * @p: socket address 312 */ 313 void eth_commit_mac_addr_change(struct net_device *dev, void *p) 314 { 315 struct sockaddr *addr = p; 316 317 memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN); 318 } 319 EXPORT_SYMBOL(eth_commit_mac_addr_change); 320 321 /** 322 * eth_mac_addr - set new Ethernet hardware address 323 * @dev: network device 324 * @p: socket address 325 * 326 * Change hardware address of device. 327 * 328 * This doesn't change hardware matching, so needs to be overridden 329 * for most real devices. 330 */ 331 int eth_mac_addr(struct net_device *dev, void *p) 332 { 333 int ret; 334 335 ret = eth_prepare_mac_addr_change(dev, p); 336 if (ret < 0) 337 return ret; 338 eth_commit_mac_addr_change(dev, p); 339 return 0; 340 } 341 EXPORT_SYMBOL(eth_mac_addr); 342 343 /** 344 * eth_change_mtu - set new MTU size 345 * @dev: network device 346 * @new_mtu: new Maximum Transfer Unit 347 * 348 * Allow changing MTU size. Needs to be overridden for devices 349 * supporting jumbo frames. 350 */ 351 int eth_change_mtu(struct net_device *dev, int new_mtu) 352 { 353 if (new_mtu < 68 || new_mtu > ETH_DATA_LEN) 354 return -EINVAL; 355 dev->mtu = new_mtu; 356 return 0; 357 } 358 EXPORT_SYMBOL(eth_change_mtu); 359 360 int eth_validate_addr(struct net_device *dev) 361 { 362 if (!is_valid_ether_addr(dev->dev_addr)) 363 return -EADDRNOTAVAIL; 364 365 return 0; 366 } 367 EXPORT_SYMBOL(eth_validate_addr); 368 369 const struct header_ops eth_header_ops ____cacheline_aligned = { 370 .create = eth_header, 371 .parse = eth_header_parse, 372 .rebuild = eth_rebuild_header, 373 .cache = eth_header_cache, 374 .cache_update = eth_header_cache_update, 375 }; 376 377 /** 378 * ether_setup - setup Ethernet network device 379 * @dev: network device 380 * 381 * Fill in the fields of the device structure with Ethernet-generic values. 382 */ 383 void ether_setup(struct net_device *dev) 384 { 385 dev->header_ops = ð_header_ops; 386 dev->type = ARPHRD_ETHER; 387 dev->hard_header_len = ETH_HLEN; 388 dev->mtu = ETH_DATA_LEN; 389 dev->addr_len = ETH_ALEN; 390 dev->tx_queue_len = 1000; /* Ethernet wants good queues */ 391 dev->flags = IFF_BROADCAST|IFF_MULTICAST; 392 dev->priv_flags |= IFF_TX_SKB_SHARING; 393 394 memset(dev->broadcast, 0xFF, ETH_ALEN); 395 396 } 397 EXPORT_SYMBOL(ether_setup); 398 399 /** 400 * alloc_etherdev_mqs - Allocates and sets up an Ethernet device 401 * @sizeof_priv: Size of additional driver-private structure to be allocated 402 * for this Ethernet device 403 * @txqs: The number of TX queues this device has. 404 * @rxqs: The number of RX queues this device has. 405 * 406 * Fill in the fields of the device structure with Ethernet-generic 407 * values. Basically does everything except registering the device. 408 * 409 * Constructs a new net device, complete with a private data area of 410 * size (sizeof_priv). A 32-byte (not bit) alignment is enforced for 411 * this private data area. 412 */ 413 414 struct net_device *alloc_etherdev_mqs(int sizeof_priv, unsigned int txqs, 415 unsigned int rxqs) 416 { 417 return alloc_netdev_mqs(sizeof_priv, "eth%d", NET_NAME_UNKNOWN, 418 ether_setup, txqs, rxqs); 419 } 420 EXPORT_SYMBOL(alloc_etherdev_mqs); 421 422 ssize_t sysfs_format_mac(char *buf, const unsigned char *addr, int len) 423 { 424 return scnprintf(buf, PAGE_SIZE, "%*phC\n", len, addr); 425 } 426 EXPORT_SYMBOL(sysfs_format_mac); 427 428 struct sk_buff **eth_gro_receive(struct sk_buff **head, 429 struct sk_buff *skb) 430 { 431 struct sk_buff *p, **pp = NULL; 432 struct ethhdr *eh, *eh2; 433 unsigned int hlen, off_eth; 434 const struct packet_offload *ptype; 435 __be16 type; 436 int flush = 1; 437 438 off_eth = skb_gro_offset(skb); 439 hlen = off_eth + sizeof(*eh); 440 eh = skb_gro_header_fast(skb, off_eth); 441 if (skb_gro_header_hard(skb, hlen)) { 442 eh = skb_gro_header_slow(skb, hlen, off_eth); 443 if (unlikely(!eh)) 444 goto out; 445 } 446 447 flush = 0; 448 449 for (p = *head; p; p = p->next) { 450 if (!NAPI_GRO_CB(p)->same_flow) 451 continue; 452 453 eh2 = (struct ethhdr *)(p->data + off_eth); 454 if (compare_ether_header(eh, eh2)) { 455 NAPI_GRO_CB(p)->same_flow = 0; 456 continue; 457 } 458 } 459 460 type = eh->h_proto; 461 462 rcu_read_lock(); 463 ptype = gro_find_receive_by_type(type); 464 if (ptype == NULL) { 465 flush = 1; 466 goto out_unlock; 467 } 468 469 skb_gro_pull(skb, sizeof(*eh)); 470 skb_gro_postpull_rcsum(skb, eh, sizeof(*eh)); 471 pp = ptype->callbacks.gro_receive(head, skb); 472 473 out_unlock: 474 rcu_read_unlock(); 475 out: 476 NAPI_GRO_CB(skb)->flush |= flush; 477 478 return pp; 479 } 480 EXPORT_SYMBOL(eth_gro_receive); 481 482 int eth_gro_complete(struct sk_buff *skb, int nhoff) 483 { 484 struct ethhdr *eh = (struct ethhdr *)(skb->data + nhoff); 485 __be16 type = eh->h_proto; 486 struct packet_offload *ptype; 487 int err = -ENOSYS; 488 489 if (skb->encapsulation) 490 skb_set_inner_mac_header(skb, nhoff); 491 492 rcu_read_lock(); 493 ptype = gro_find_complete_by_type(type); 494 if (ptype != NULL) 495 err = ptype->callbacks.gro_complete(skb, nhoff + 496 sizeof(struct ethhdr)); 497 498 rcu_read_unlock(); 499 return err; 500 } 501 EXPORT_SYMBOL(eth_gro_complete); 502 503 static struct packet_offload eth_packet_offload __read_mostly = { 504 .type = cpu_to_be16(ETH_P_TEB), 505 .callbacks = { 506 .gro_receive = eth_gro_receive, 507 .gro_complete = eth_gro_complete, 508 }, 509 }; 510 511 static int __init eth_offload_init(void) 512 { 513 dev_add_offload(ð_packet_offload); 514 515 return 0; 516 } 517 518 fs_initcall(eth_offload_init); 519