net/ethernet/eth.c

   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 <net/flow_dissector.h>
  62 #include <linux/uaccess.h>
  63
  64 __setup("ether=", netdev_boot_setup);
  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  */
  79 int 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 = (struct ethhdr *)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 }
 114 EXPORT_SYMBOL(eth_header);
 115
 116 /**
 117  * eth_get_headlen - determine the length of header for an ethernet frame
 118  * @data: pointer to start of frame
 119  * @len: total length of frame
 120  *
 121  * Make a best effort attempt to pull the length for all of the headers for
 122  * a given frame in a linear buffer.
 123  */
 124 u32 eth_get_headlen(void *data, unsigned int len)
 125 {
 126         const struct ethhdr *eth = (const struct ethhdr *)data;
 127         struct flow_keys keys;
 128
 129         /* this should never happen, but better safe than sorry */
 130         if (unlikely(len < sizeof(*eth)))
 131                 return len;
 132
 133         /* parse any remaining L2/L3 headers, check for L4 */
 134         if (!skb_flow_dissect_flow_keys_buf(&keys, data, eth->h_proto,
 135                                             sizeof(*eth), len, 0))
 136                 return max_t(u32, keys.control.thoff, sizeof(*eth));
 137
 138         /* parse for any L4 headers */
 139         return min_t(u32, __skb_get_poff(NULL, data, &keys, len), len);
 140 }
 141 EXPORT_SYMBOL(eth_get_headlen);
 142
 143 /**
 144  * eth_type_trans - determine the packet's protocol ID.
 145  * @skb: received socket data
 146  * @dev: receiving network device
 147  *
 148  * The rule here is that we
 149  * assume 802.3 if the type field is short enough to be a length.
 150  * This is normal practice and works for any 'now in use' protocol.
 151  */
 152 __be16 eth_type_trans(struct sk_buff *skb, struct net_device *dev)
 153 {
 154         unsigned short _service_access_point;
 155         const unsigned short *sap;
 156         const struct ethhdr *eth;
 157
 158         skb->dev = dev;
 159         skb_reset_mac_header(skb);
 160
 161         eth = (struct ethhdr *)skb->data;
 162         skb_pull_inline(skb, ETH_HLEN);
 163
 164         if (unlikely(is_multicast_ether_addr_64bits(eth->h_dest))) {
 165                 if (ether_addr_equal_64bits(eth->h_dest, dev->broadcast))
 166                         skb->pkt_type = PACKET_BROADCAST;
 167                 else
 168                         skb->pkt_type = PACKET_MULTICAST;
 169         }
 170         else if (unlikely(!ether_addr_equal_64bits(eth->h_dest,
 171                                                    dev->dev_addr)))
 172                 skb->pkt_type = PACKET_OTHERHOST;
 173
 174         /*
 175          * Some variants of DSA tagging don't have an ethertype field
 176          * at all, so we check here whether one of those tagging
 177          * variants has been configured on the receiving interface,
 178          * and if so, set skb->protocol without looking at the packet.
 179          */
 180         if (unlikely(netdev_uses_dsa(dev)))
 181                 return htons(ETH_P_XDSA);
 182
 183         if (likely(eth_proto_is_802_3(eth->h_proto)))
 184                 return eth->h_proto;
 185
 186         /*
 187          *      This is a magic hack to spot IPX packets. Older Novell breaks
 188          *      the protocol design and runs IPX over 802.3 without an 802.2 LLC
 189          *      layer. We look for FFFF which isn't a used 802.2 SSAP/DSAP. This
 190          *      won't work for fault tolerant netware but does for the rest.
 191          */
 192         sap = skb_header_pointer(skb, 0, sizeof(*sap), &_service_access_point);
 193         if (sap && *sap == 0xFFFF)
 194                 return htons(ETH_P_802_3);
 195
 196         /*
 197          *      Real 802.2 LLC
 198          */
 199         return htons(ETH_P_802_2);
 200 }
 201 EXPORT_SYMBOL(eth_type_trans);
 202
 203 /**
 204  * eth_header_parse - extract hardware address from packet
 205  * @skb: packet to extract header from
 206  * @haddr: destination buffer
 207  */
 208 int eth_header_parse(const struct sk_buff *skb, unsigned char *haddr)
 209 {
 210         const struct ethhdr *eth = eth_hdr(skb);
 211         memcpy(haddr, eth->h_source, ETH_ALEN);
 212         return ETH_ALEN;
 213 }
 214 EXPORT_SYMBOL(eth_header_parse);
 215
 216 /**
 217  * eth_header_cache - fill cache entry from neighbour
 218  * @neigh: source neighbour
 219  * @hh: destination cache entry
 220  * @type: Ethernet type field
 221  *
 222  * Create an Ethernet header template from the neighbour.
 223  */
 224 int eth_header_cache(const struct neighbour *neigh, struct hh_cache *hh, __be16 type)
 225 {
 226         struct ethhdr *eth;
 227         const struct net_device *dev = neigh->dev;
 228
 229         eth = (struct ethhdr *)
 230             (((u8 *) hh->hh_data) + (HH_DATA_OFF(sizeof(*eth))));
 231
 232         if (type == htons(ETH_P_802_3))
 233                 return -1;
 234
 235         eth->h_proto = type;
 236         memcpy(eth->h_source, dev->dev_addr, ETH_ALEN);
 237         memcpy(eth->h_dest, neigh->ha, ETH_ALEN);
 238         hh->hh_len = ETH_HLEN;
 239         return 0;
 240 }
 241 EXPORT_SYMBOL(eth_header_cache);
 242
 243 /**
 244  * eth_header_cache_update - update cache entry
 245  * @hh: destination cache entry
 246  * @dev: network device
 247  * @haddr: new hardware address
 248  *
 249  * Called by Address Resolution module to notify changes in address.
 250  */
 251 void eth_header_cache_update(struct hh_cache *hh,
 252                              const struct net_device *dev,
 253                              const unsigned char *haddr)
 254 {
 255         memcpy(((u8 *) hh->hh_data) + HH_DATA_OFF(sizeof(struct ethhdr)),
 256                haddr, ETH_ALEN);
 257 }
 258 EXPORT_SYMBOL(eth_header_cache_update);
 259
 260 /**
 261  * eth_prepare_mac_addr_change - prepare for mac change
 262  * @dev: network device
 263  * @p: socket address
 264  */
 265 int eth_prepare_mac_addr_change(struct net_device *dev, void *p)
 266 {
 267         struct sockaddr *addr = p;
 268
 269         if (!(dev->priv_flags & IFF_LIVE_ADDR_CHANGE) && netif_running(dev))
 270                 return -EBUSY;
 271         if (!is_valid_ether_addr(addr->sa_data))
 272                 return -EADDRNOTAVAIL;
 273         return 0;
 274 }
 275 EXPORT_SYMBOL(eth_prepare_mac_addr_change);
 276
 277 /**
 278  * eth_commit_mac_addr_change - commit mac change
 279  * @dev: network device
 280  * @p: socket address
 281  */
 282 void eth_commit_mac_addr_change(struct net_device *dev, void *p)
 283 {
 284         struct sockaddr *addr = p;
 285
 286         memcpy(dev->dev_addr, addr->sa_data, ETH_ALEN);
 287 }
 288 EXPORT_SYMBOL(eth_commit_mac_addr_change);
 289
 290 /**
 291  * eth_mac_addr - set new Ethernet hardware address
 292  * @dev: network device
 293  * @p: socket address
 294  *
 295  * Change hardware address of device.
 296  *
 297  * This doesn't change hardware matching, so needs to be overridden
 298  * for most real devices.
 299  */
 300 int eth_mac_addr(struct net_device *dev, void *p)
 301 {
 302         int ret;
 303
 304         ret = eth_prepare_mac_addr_change(dev, p);
 305         if (ret < 0)
 306                 return ret;
 307         eth_commit_mac_addr_change(dev, p);
 308         return 0;
 309 }
 310 EXPORT_SYMBOL(eth_mac_addr);
 311
 312 /**
 313  * eth_change_mtu - set new MTU size
 314  * @dev: network device
 315  * @new_mtu: new Maximum Transfer Unit
 316  *
 317  * Allow changing MTU size. Needs to be overridden for devices
 318  * supporting jumbo frames.
 319  */
 320 int eth_change_mtu(struct net_device *dev, int new_mtu)
 321 {
 322         if (new_mtu < 68 || new_mtu > ETH_DATA_LEN)
 323                 return -EINVAL;
 324         dev->mtu = new_mtu;
 325         return 0;
 326 }
 327 EXPORT_SYMBOL(eth_change_mtu);
 328
 329 int eth_validate_addr(struct net_device *dev)
 330 {
 331         if (!is_valid_ether_addr(dev->dev_addr))
 332                 return -EADDRNOTAVAIL;
 333
 334         return 0;
 335 }
 336 EXPORT_SYMBOL(eth_validate_addr);
 337
 338 const struct header_ops eth_header_ops ____cacheline_aligned = {
 339         .create         = eth_header,
 340         .parse          = eth_header_parse,
 341         .cache          = eth_header_cache,
 342         .cache_update   = eth_header_cache_update,
 343 };
 344
 345 /**
 346  * ether_setup - setup Ethernet network device
 347  * @dev: network device
 348  *
 349  * Fill in the fields of the device structure with Ethernet-generic values.
 350  */
 351 void ether_setup(struct net_device *dev)
 352 {
 353         dev->header_ops         = &eth_header_ops;
 354         dev->type               = ARPHRD_ETHER;
 355         dev->hard_header_len    = ETH_HLEN;
 356         dev->mtu                = ETH_DATA_LEN;
 357         dev->addr_len           = ETH_ALEN;
 358         dev->tx_queue_len       = 1000; /* Ethernet wants good queues */
 359         dev->flags              = IFF_BROADCAST|IFF_MULTICAST;
 360         dev->priv_flags         |= IFF_TX_SKB_SHARING;
 361
 362         eth_broadcast_addr(dev->broadcast);
 363
 364 }
 365 EXPORT_SYMBOL(ether_setup);
 366
 367 /**
 368  * alloc_etherdev_mqs - Allocates and sets up an Ethernet device
 369  * @sizeof_priv: Size of additional driver-private structure to be allocated
 370  *      for this Ethernet device
 371  * @txqs: The number of TX queues this device has.
 372  * @rxqs: The number of RX queues this device has.
 373  *
 374  * Fill in the fields of the device structure with Ethernet-generic
 375  * values. Basically does everything except registering the device.
 376  *
 377  * Constructs a new net device, complete with a private data area of
 378  * size (sizeof_priv).  A 32-byte (not bit) alignment is enforced for
 379  * this private data area.
 380  */
 381
 382 struct net_device *alloc_etherdev_mqs(int sizeof_priv, unsigned int txqs,
 383                                       unsigned int rxqs)
 384 {
 385         return alloc_netdev_mqs(sizeof_priv, "eth%d", NET_NAME_UNKNOWN,
 386                                 ether_setup, txqs, rxqs);
 387 }
 388 EXPORT_SYMBOL(alloc_etherdev_mqs);
 389
 390 ssize_t sysfs_format_mac(char *buf, const unsigned char *addr, int len)
 391 {
 392         return scnprintf(buf, PAGE_SIZE, "%*phC\n", len, addr);
 393 }
 394 EXPORT_SYMBOL(sysfs_format_mac);
 395
 396 struct sk_buff **eth_gro_receive(struct sk_buff **head,
 397                                  struct sk_buff *skb)
 398 {
 399         struct sk_buff *p, **pp = NULL;
 400         struct ethhdr *eh, *eh2;
 401         unsigned int hlen, off_eth;
 402         const struct packet_offload *ptype;
 403         __be16 type;
 404         int flush = 1;
 405
 406         off_eth = skb_gro_offset(skb);
 407         hlen = off_eth + sizeof(*eh);
 408         eh = skb_gro_header_fast(skb, off_eth);
 409         if (skb_gro_header_hard(skb, hlen)) {
 410                 eh = skb_gro_header_slow(skb, hlen, off_eth);
 411                 if (unlikely(!eh))
 412                         goto out;
 413         }
 414
 415         flush = 0;
 416
 417         for (p = *head; p; p = p->next) {
 418                 if (!NAPI_GRO_CB(p)->same_flow)
 419                         continue;
 420
 421                 eh2 = (struct ethhdr *)(p->data + off_eth);
 422                 if (compare_ether_header(eh, eh2)) {
 423                         NAPI_GRO_CB(p)->same_flow = 0;
 424                         continue;
 425                 }
 426         }
 427
 428         type = eh->h_proto;
 429
 430         rcu_read_lock();
 431         ptype = gro_find_receive_by_type(type);
 432         if (ptype == NULL) {
 433                 flush = 1;
 434                 goto out_unlock;
 435         }
 436
 437         skb_gro_pull(skb, sizeof(*eh));
 438         skb_gro_postpull_rcsum(skb, eh, sizeof(*eh));
 439         pp = ptype->callbacks.gro_receive(head, skb);
 440
 441 out_unlock:
 442         rcu_read_unlock();
 443 out:
 444         NAPI_GRO_CB(skb)->flush |= flush;
 445
 446         return pp;
 447 }
 448 EXPORT_SYMBOL(eth_gro_receive);
 449
 450 int eth_gro_complete(struct sk_buff *skb, int nhoff)
 451 {
 452         struct ethhdr *eh = (struct ethhdr *)(skb->data + nhoff);
 453         __be16 type = eh->h_proto;
 454         struct packet_offload *ptype;
 455         int err = -ENOSYS;
 456
 457         if (skb->encapsulation)
 458                 skb_set_inner_mac_header(skb, nhoff);
 459
 460         rcu_read_lock();
 461         ptype = gro_find_complete_by_type(type);
 462         if (ptype != NULL)
 463                 err = ptype->callbacks.gro_complete(skb, nhoff +
 464                                                     sizeof(struct ethhdr));
 465
 466         rcu_read_unlock();
 467         return err;
 468 }
 469 EXPORT_SYMBOL(eth_gro_complete);
 470
 471 static struct packet_offload eth_packet_offload __read_mostly = {
 472         .type = cpu_to_be16(ETH_P_TEB),
 473         .priority = 10,
 474         .callbacks = {
 475                 .gro_receive = eth_gro_receive,
 476                 .gro_complete = eth_gro_complete,
 477         },
 478 };
 479
 480 static int __init eth_offload_init(void)
 481 {
 482         dev_add_offload(&eth_packet_offload);
 483
 484         return 0;
 485 }
 486
 487 fs_initcall(eth_offload_init);