linux/net/ethernet/eth.c
<<
>>
Prefs
   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 <linux/of_net.h>
  56#include <linux/pci.h>
  57#include <net/dst.h>
  58#include <net/arp.h>
  59#include <net/sock.h>
  60#include <net/ipv6.h>
  61#include <net/ip.h>
  62#include <net/dsa.h>
  63#include <net/flow_dissector.h>
  64#include <linux/uaccess.h>
  65
  66__setup("ether=", netdev_boot_setup);
  67
  68/**
  69 * eth_header - create the Ethernet header
  70 * @skb:        buffer to alter
  71 * @dev:        source device
  72 * @type:       Ethernet type field
  73 * @daddr: destination address (NULL leave destination address)
  74 * @saddr: source address (NULL use device source address)
  75 * @len:   packet length (<= skb->len)
  76 *
  77 *
  78 * Set the protocol type. For a packet of type ETH_P_802_3/2 we put the length
  79 * in here instead.
  80 */
  81int eth_header(struct sk_buff *skb, struct net_device *dev,
  82               unsigned short type,
  83               const void *daddr, const void *saddr, unsigned int len)
  84{
  85        struct ethhdr *eth = (struct ethhdr *)skb_push(skb, ETH_HLEN);
  86
  87        if (type != ETH_P_802_3 && type != ETH_P_802_2)
  88                eth->h_proto = htons(type);
  89        else
  90                eth->h_proto = htons(len);
  91
  92        /*
  93         *      Set the source hardware address.
  94         */
  95
  96        if (!saddr)
  97                saddr = dev->dev_addr;
  98        memcpy(eth->h_source, saddr, ETH_ALEN);
  99
 100        if (daddr) {
 101                memcpy(eth->h_dest, daddr, ETH_ALEN);
 102                return ETH_HLEN;
 103        }
 104
 105        /*
 106         *      Anyway, the loopback-device should never use this function...
 107         */
 108
 109        if (dev->flags & (IFF_LOOPBACK | IFF_NOARP)) {
 110                eth_zero_addr(eth->h_dest);
 111                return ETH_HLEN;
 112        }
 113
 114        return -ETH_HLEN;
 115}
 116EXPORT_SYMBOL(eth_header);
 117
 118/**
 119 * eth_get_headlen - determine the length of header for an ethernet frame
 120 * @data: pointer to start of frame
 121 * @len: total length of frame
 122 *
 123 * Make a best effort attempt to pull the length for all of the headers for
 124 * a given frame in a linear buffer.
 125 */
 126u32 eth_get_headlen(void *data, unsigned int len)
 127{
 128        const unsigned int flags = FLOW_DISSECTOR_F_PARSE_1ST_FRAG;
 129        const struct ethhdr *eth = (const struct ethhdr *)data;
 130        struct flow_keys keys;
 131
 132        /* this should never happen, but better safe than sorry */
 133        if (unlikely(len < sizeof(*eth)))
 134                return len;
 135
 136        /* parse any remaining L2/L3 headers, check for L4 */
 137        if (!skb_flow_dissect_flow_keys_buf(&keys, data, eth->h_proto,
 138                                            sizeof(*eth), 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(is_multicast_ether_addr_64bits(eth->h_dest))) {
 168                if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
 169                        skb->pkt_type = PACKET_BROADCAST;
 170                else
 171                        skb->pkt_type = PACKET_MULTICAST;
 172        }
 173        else if (unlikely(!ether_addr_equal_64bits(eth->h_dest,
 174                                                   dev->dev_addr)))
 175                skb->pkt_type = PACKET_OTHERHOST;
 176
 177        /*
 178         * Some variants of DSA tagging don't have an ethertype field
 179         * at all, so we check here whether one of those tagging
 180         * variants has been configured on the receiving interface,
 181         * and if so, set skb->protocol without looking at the packet.
 182         */
 183        if (unlikely(netdev_uses_dsa(dev)))
 184                return htons(ETH_P_XDSA);
 185
 186        if (likely(eth_proto_is_802_3(eth->h_proto)))
 187                return eth->h_proto;
 188
 189        /*
 190         *      This is a magic hack to spot IPX packets. Older Novell breaks
 191         *      the protocol design and runs IPX over 802.3 without an 802.2 LLC
 192         *      layer. We look for FFFF which isn't a used 802.2 SSAP/DSAP. This
 193         *      won't work for fault tolerant netware but does for the rest.
 194         */
 195        sap = skb_header_pointer(skb, 0, sizeof(*sap), &_service_access_point);
 196        if (sap && *sap == 0xFFFF)
 197                return htons(ETH_P_802_3);
 198
 199        /*
 200         *      Real 802.2 LLC
 201         */
 202        return htons(ETH_P_802_2);
 203}
 204EXPORT_SYMBOL(eth_type_trans);
 205
 206/**
 207 * eth_header_parse - extract hardware address from packet
 208 * @skb: packet to extract header from
 209 * @haddr: destination buffer
 210 */
 211int eth_header_parse(const struct sk_buff *skb, unsigned char *haddr)
 212{
 213        const struct ethhdr *eth = eth_hdr(skb);
 214        memcpy(haddr, eth->h_source, ETH_ALEN);
 215        return ETH_ALEN;
 216}
 217EXPORT_SYMBOL(eth_header_parse);
 218
 219/**
 220 * eth_header_cache - fill cache entry from neighbour
 221 * @neigh: source neighbour
 222 * @hh: destination cache entry
 223 * @type: Ethernet type field
 224 *
 225 * Create an Ethernet header template from the neighbour.
 226 */
 227int eth_header_cache(const struct neighbour *neigh, struct hh_cache *hh, __be16 type)
 228{
 229        struct ethhdr *eth;
 230        const struct net_device *dev = neigh->dev;
 231
 232        eth = (struct ethhdr *)
 233            (((u8 *) hh->hh_data) + (HH_DATA_OFF(sizeof(*eth))));
 234
 235        if (type == htons(ETH_P_802_3))
 236                return -1;
 237
 238        eth->h_proto = type;
 239        memcpy(eth->h_source, dev->dev_addr, ETH_ALEN);
 240        memcpy(eth->h_dest, neigh->ha, ETH_ALEN);
 241        hh->hh_len = ETH_HLEN;
 242        return 0;
 243}
 244EXPORT_SYMBOL(eth_header_cache);
 245
 246/**
 247 * eth_header_cache_update - update cache entry
 248 * @hh: destination cache entry
 249 * @dev: network device
 250 * @haddr: new hardware address
 251 *
 252 * Called by Address Resolution module to notify changes in address.
 253 */
 254void eth_header_cache_update(struct hh_cache *hh,
 255                             const struct net_device *dev,
 256                             const unsigned char *haddr)
 257{
 258        memcpy(((u8 *) hh->hh_data) + HH_DATA_OFF(sizeof(struct ethhdr)),
 259               haddr, ETH_ALEN);
 260}
 261EXPORT_SYMBOL(eth_header_cache_update);
 262
 263/**
 264 * eth_prepare_mac_addr_change - prepare for mac change
 265 * @dev: network device
 266 * @p: socket address
 267 */
 268int eth_prepare_mac_addr_change(struct net_device *dev, void *p)
 269{
 270        struct sockaddr *addr = p;
 271
 272        if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev))
 273                return -EBUSY;
 274        if (!is_valid_ether_addr(addr->sa_data))
 275                return -EADDRNOTAVAIL;
 276        return 0;
 277}
 278EXPORT_SYMBOL(eth_prepare_mac_addr_change);
 279
 280/**
 281 * eth_commit_mac_addr_change - commit mac change
 282 * @dev: network device
 283 * @p: socket address
 284 */
 285void eth_commit_mac_addr_change(struct net_device *dev, void *p)
 286{
 287        struct sockaddr *addr = p;
 288
 289        memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
 290}
 291EXPORT_SYMBOL(eth_commit_mac_addr_change);
 292
 293/**
 294 * eth_mac_addr - set new Ethernet hardware address
 295 * @dev: network device
 296 * @p: socket address
 297 *
 298 * Change hardware address of device.
 299 *
 300 * This doesn't change hardware matching, so needs to be overridden
 301 * for most real devices.
 302 */
 303int eth_mac_addr(struct net_device *dev, void *p)
 304{
 305        int ret;
 306
 307        ret = eth_prepare_mac_addr_change(dev, p);
 308        if (ret < 0)
 309                return ret;
 310        eth_commit_mac_addr_change(dev, p);
 311        return 0;
 312}
 313EXPORT_SYMBOL(eth_mac_addr);
 314
 315/**
 316 * eth_change_mtu - set new MTU size
 317 * @dev: network device
 318 * @new_mtu: new Maximum Transfer Unit
 319 *
 320 * Allow changing MTU size. Needs to be overridden for devices
 321 * supporting jumbo frames.
 322 */
 323int eth_change_mtu(struct net_device *dev, int new_mtu)
 324{
 325        if (new_mtu < 68 || new_mtu > ETH_DATA_LEN)
 326                return -EINVAL;
 327        dev->mtu = new_mtu;
 328        return 0;
 329}
 330EXPORT_SYMBOL(eth_change_mtu);
 331
 332int eth_validate_addr(struct net_device *dev)
 333{
 334        if (!is_valid_ether_addr(dev->dev_addr))
 335                return -EADDRNOTAVAIL;
 336
 337        return 0;
 338}
 339EXPORT_SYMBOL(eth_validate_addr);
 340
 341const struct header_ops eth_header_ops ____cacheline_aligned = {
 342        .create         = eth_header,
 343        .parse          = eth_header_parse,
 344        .cache          = eth_header_cache,
 345        .cache_update   = eth_header_cache_update,
 346};
 347
 348/**
 349 * ether_setup - setup Ethernet network device
 350 * @dev: network device
 351 *
 352 * Fill in the fields of the device structure with Ethernet-generic values.
 353 */
 354void ether_setup(struct net_device *dev)
 355{
 356        dev->header_ops         = &eth_header_ops;
 357        dev->type               = ARPHRD_ETHER;
 358        dev->hard_header_len    = ETH_HLEN;
 359        dev->mtu                = ETH_DATA_LEN;
 360        dev->addr_len           = ETH_ALEN;
 361        dev->tx_queue_len       = 1000; /* Ethernet wants good queues */
 362        dev->flags              = IFF_BROADCAST|IFF_MULTICAST;
 363        dev->priv_flags         |= IFF_TX_SKB_SHARING;
 364
 365        eth_broadcast_addr(dev->broadcast);
 366
 367}
 368EXPORT_SYMBOL(ether_setup);
 369
 370/**
 371 * alloc_etherdev_mqs - Allocates and sets up an Ethernet device
 372 * @sizeof_priv: Size of additional driver-private structure to be allocated
 373 *      for this Ethernet device
 374 * @txqs: The number of TX queues this device has.
 375 * @rxqs: The number of RX queues this device has.
 376 *
 377 * Fill in the fields of the device structure with Ethernet-generic
 378 * values. Basically does everything except registering the device.
 379 *
 380 * Constructs a new net device, complete with a private data area of
 381 * size (sizeof_priv).  A 32-byte (not bit) alignment is enforced for
 382 * this private data area.
 383 */
 384
 385struct net_device *alloc_etherdev_mqs(int sizeof_priv, unsigned int txqs,
 386                                      unsigned int rxqs)
 387{
 388        return alloc_netdev_mqs(sizeof_priv, "eth%d", NET_NAME_UNKNOWN,
 389                                ether_setup, txqs, rxqs);
 390}
 391EXPORT_SYMBOL(alloc_etherdev_mqs);
 392
 393ssize_t sysfs_format_mac(char *buf, const unsigned char *addr, int len)
 394{
 395        return scnprintf(buf, PAGE_SIZE, "%*phC\n", len, addr);
 396}
 397EXPORT_SYMBOL(sysfs_format_mac);
 398
 399struct sk_buff **eth_gro_receive(struct sk_buff **head,
 400                                 struct sk_buff *skb)
 401{
 402        struct sk_buff *p, **pp = NULL;
 403        struct ethhdr *eh, *eh2;
 404        unsigned int hlen, off_eth;
 405        const struct packet_offload *ptype;
 406        __be16 type;
 407        int flush = 1;
 408
 409        off_eth = skb_gro_offset(skb);
 410        hlen = off_eth + sizeof(*eh);
 411        eh = skb_gro_header_fast(skb, off_eth);
 412        if (skb_gro_header_hard(skb, hlen)) {
 413                eh = skb_gro_header_slow(skb, hlen, off_eth);
 414                if (unlikely(!eh))
 415                        goto out;
 416        }
 417
 418        flush = 0;
 419
 420        for (p = *head; p; p = p->next) {
 421                if (!NAPI_GRO_CB(p)->same_flow)
 422                        continue;
 423
 424                eh2 = (struct ethhdr *)(p->data + off_eth);
 425                if (compare_ether_header(eh, eh2)) {
 426                        NAPI_GRO_CB(p)->same_flow = 0;
 427                        continue;
 428                }
 429        }
 430
 431        type = eh->h_proto;
 432
 433        rcu_read_lock();
 434        ptype = gro_find_receive_by_type(type);
 435        if (ptype == NULL) {
 436                flush = 1;
 437                goto out_unlock;
 438        }
 439
 440        skb_gro_pull(skb, sizeof(*eh));
 441        skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
 442        pp = ptype->callbacks.gro_receive(head, skb);
 443
 444out_unlock:
 445        rcu_read_unlock();
 446out:
 447        NAPI_GRO_CB(skb)->flush |= flush;
 448
 449        return pp;
 450}
 451EXPORT_SYMBOL(eth_gro_receive);
 452
 453int eth_gro_complete(struct sk_buff *skb, int nhoff)
 454{
 455        struct ethhdr *eh = (struct ethhdr *)(skb->data + nhoff);
 456        __be16 type = eh->h_proto;
 457        struct packet_offload *ptype;
 458        int err = -ENOSYS;
 459
 460        if (skb->encapsulation)
 461                skb_set_inner_mac_header(skb, nhoff);
 462
 463        rcu_read_lock();
 464        ptype = gro_find_complete_by_type(type);
 465        if (ptype != NULL)
 466                err = ptype->callbacks.gro_complete(skb, nhoff +
 467                                                    sizeof(struct ethhdr));
 468
 469        rcu_read_unlock();
 470        return err;
 471}
 472EXPORT_SYMBOL(eth_gro_complete);
 473
 474static struct packet_offload eth_packet_offload __read_mostly = {
 475        .type = cpu_to_be16(ETH_P_TEB),
 476        .priority = 10,
 477        .callbacks = {
 478                .gro_receive = eth_gro_receive,
 479                .gro_complete = eth_gro_complete,
 480        },
 481};
 482
 483static int __init eth_offload_init(void)
 484{
 485        dev_add_offload(&eth_packet_offload);
 486
 487        return 0;
 488}
 489
 490fs_initcall(eth_offload_init);
 491
 492unsigned char * __weak arch_get_platform_mac_address(void)
 493{
 494        return NULL;
 495}
 496
 497int eth_platform_get_mac_address(struct device *dev, u8 *mac_addr)
 498{
 499        const unsigned char *addr;
 500        struct device_node *dp;
 501
 502        if (dev_is_pci(dev))
 503                dp = pci_device_to_OF_node(to_pci_dev(dev));
 504        else
 505                dp = dev->of_node;
 506
 507        addr = NULL;
 508        if (dp)
 509                addr = of_get_mac_address(dp);
 510        if (!addr)
 511                addr = arch_get_platform_mac_address();
 512
 513        if (!addr)
 514                return -ENODEV;
 515
 516        ether_addr_copy(mac_addr, addr);
 517        return 0;
 518}
 519EXPORT_SYMBOL(eth_platform_get_mac_address);
 520