drivers/net/loopback.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  *              Pseudo-driver for the loopback interface.
   7  *
   8  * Version:     @(#)loopback.c  1.0.4b  08/16/93
   9  *
  10  * Authors:     Ross Biro
  11  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12  *              Donald Becker, <becker@scyld.com>
  13  *
  14  *              Alan Cox        :       Fixed oddments for NET3.014
  15  *              Alan Cox        :       Rejig for NET3.029 snap #3
  16  *              Alan Cox        :       Fixed NET3.029 bugs and sped up
  17  *              Larry McVoy     :       Tiny tweak to double performance
  18  *              Alan Cox        :       Backed out LMV's tweak - the linux mm
  19  *                                      can't take it...
  20  *              Michael Griffith:       Don't bother computing the checksums
  21  *                                      on packets received on the loopback
  22  *                                      interface.
  23  *              Alexey Kuznetsov:       Potential hang under some extreme
  24  *                                      cases removed.
  25  *
  26  *              This program is free software; you can redistribute it and/or
  27  *              modify it under the terms of the GNU General Public License
  28  *              as published by the Free Software Foundation; either version
  29  *              2 of the License, or (at your option) any later version.
  30  */
  31 #include <linux/kernel.h>
  32 #include <linux/jiffies.h>
  33 #include <linux/module.h>
  34 #include <linux/interrupt.h>
  35 #include <linux/fs.h>
  36 #include <linux/types.h>
  37 #include <linux/string.h>
  38 #include <linux/socket.h>
  39 #include <linux/errno.h>
  40 #include <linux/fcntl.h>
  41 #include <linux/in.h>
  42 #include <linux/init.h>
  43
  44 #include <asm/system.h>
  45 #include <asm/uaccess.h>
  46 #include <asm/io.h>
  47
  48 #include <linux/inet.h>
  49 #include <linux/netdevice.h>
  50 #include <linux/etherdevice.h>
  51 #include <linux/skbuff.h>
  52 #include <linux/ethtool.h>
  53 #include <net/sock.h>
  54 #include <net/checksum.h>
  55 #include <linux/if_ether.h>     /* For the statistics structure. */
  56 #include <linux/if_arp.h>       /* For ARPHRD_ETHER */
  57 #include <linux/ip.h>
  58 #include <linux/tcp.h>
  59 #include <linux/percpu.h>
  60 #include <net/net_namespace.h>
  61
  62 struct pcpu_lstats {
  63         unsigned long packets;
  64         unsigned long bytes;
  65 };
  66
  67 #define LOOPBACK_OVERHEAD (128 + MAX_HEADER + 16 + 16)
  68
  69 /* KISS: just allocate small chunks and copy bits.
  70  *
  71  * So, in fact, this is documentation, explaining what we expect
  72  * of largesending device modulo TCP checksum, which is ignored for loopback.
  73  */
  74
  75 #ifdef LOOPBACK_TSO
  76 static void emulate_large_send_offload(struct sk_buff *skb)
  77 {
  78         struct iphdr *iph = ip_hdr(skb);
  79         struct tcphdr *th = (struct tcphdr *)(skb_network_header(skb) +
  80                                               (iph->ihl * 4));
  81         unsigned int doffset = (iph->ihl + th->doff) * 4;
  82         unsigned int mtu = skb_shinfo(skb)->gso_size + doffset;
  83         unsigned int offset = 0;
  84         u32 seq = ntohl(th->seq);
  85         u16 id  = ntohs(iph->id);
  86
  87         while (offset + doffset < skb->len) {
  88                 unsigned int frag_size = min(mtu, skb->len - offset) - doffset;
  89                 struct sk_buff *nskb = alloc_skb(mtu + 32, GFP_ATOMIC);
  90
  91                 if (!nskb)
  92                         break;
  93                 skb_reserve(nskb, 32);
  94                 skb_set_mac_header(nskb, -ETH_HLEN);
  95                 skb_reset_network_header(nskb);
  96                 iph = ip_hdr(nskb);
  97                 skb_copy_to_linear_data(nskb, skb_network_header(skb),
  98                                         doffset);
  99                 if (skb_copy_bits(skb,
 100                                   doffset + offset,
 101                                   nskb->data + doffset,
 102                                   frag_size))
 103                         BUG();
 104                 skb_put(nskb, doffset + frag_size);
 105                 nskb->ip_summed = CHECKSUM_UNNECESSARY;
 106                 nskb->dev = skb->dev;
 107                 nskb->priority = skb->priority;
 108                 nskb->protocol = skb->protocol;
 109                 nskb->dst = dst_clone(skb->dst);
 110                 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
 111                 nskb->pkt_type = skb->pkt_type;
 112
 113                 th = (struct tcphdr *)(skb_network_header(nskb) + iph->ihl * 4);
 114                 iph->tot_len = htons(frag_size + doffset);
 115                 iph->id = htons(id);
 116                 iph->check = 0;
 117                 iph->check = ip_fast_csum((unsigned char *) iph, iph->ihl);
 118                 th->seq = htonl(seq);
 119                 if (offset + doffset + frag_size < skb->len)
 120                         th->fin = th->psh = 0;
 121                 netif_rx(nskb);
 122                 offset += frag_size;
 123                 seq += frag_size;
 124                 id++;
 125         }
 126
 127         dev_kfree_skb(skb);
 128 }
 129 #endif /* LOOPBACK_TSO */
 130
 131 /*
 132  * The higher levels take care of making this non-reentrant (it's
 133  * called with bh's disabled).
 134  */
 135 static int loopback_xmit(struct sk_buff *skb, struct net_device *dev)
 136 {
 137         struct pcpu_lstats *pcpu_lstats, *lb_stats;
 138
 139         skb_orphan(skb);
 140
 141         skb->protocol = eth_type_trans(skb,dev);
 142 #ifndef LOOPBACK_MUST_CHECKSUM
 143         skb->ip_summed = CHECKSUM_UNNECESSARY;
 144 #endif
 145
 146 #ifdef LOOPBACK_TSO
 147         if (skb_is_gso(skb)) {
 148                 BUG_ON(skb->protocol != htons(ETH_P_IP));
 149                 BUG_ON(ip_hdr(skb)->protocol != IPPROTO_TCP);
 150
 151                 emulate_large_send_offload(skb);
 152                 return 0;
 153         }
 154 #endif
 155         dev->last_rx = jiffies;
 156
 157         /* it's OK to use per_cpu_ptr() because BHs are off */
 158         pcpu_lstats = netdev_priv(dev);
 159         lb_stats = per_cpu_ptr(pcpu_lstats, smp_processor_id());
 160         lb_stats->bytes += skb->len;
 161         lb_stats->packets++;
 162
 163         netif_rx(skb);
 164
 165         return 0;
 166 }
 167
 168 static struct net_device_stats *get_stats(struct net_device *dev)
 169 {
 170         const struct pcpu_lstats *pcpu_lstats;
 171         struct net_device_stats *stats = &dev->stats;
 172         unsigned long bytes = 0;
 173         unsigned long packets = 0;
 174         int i;
 175
 176         pcpu_lstats = netdev_priv(dev);
 177         for_each_possible_cpu(i) {
 178                 const struct pcpu_lstats *lb_stats;
 179
 180                 lb_stats = per_cpu_ptr(pcpu_lstats, i);
 181                 bytes   += lb_stats->bytes;
 182                 packets += lb_stats->packets;
 183         }
 184         stats->rx_packets = packets;
 185         stats->tx_packets = packets;
 186         stats->rx_bytes = bytes;
 187         stats->tx_bytes = bytes;
 188         return stats;
 189 }
 190
 191 static u32 always_on(struct net_device *dev)
 192 {
 193         return 1;
 194 }
 195
 196 static const struct ethtool_ops loopback_ethtool_ops = {
 197         .get_link               = always_on,
 198         .set_tso                = ethtool_op_set_tso,
 199         .get_tx_csum            = always_on,
 200         .get_sg                 = always_on,
 201         .get_rx_csum            = always_on,
 202 };
 203
 204 static int loopback_dev_init(struct net_device *dev)
 205 {
 206         struct pcpu_lstats *lstats;
 207
 208         lstats = alloc_percpu(struct pcpu_lstats);
 209         if (!lstats)
 210                 return -ENOMEM;
 211
 212         dev->priv = lstats;
 213         return 0;
 214 }
 215
 216 static void loopback_dev_free(struct net_device *dev)
 217 {
 218         struct pcpu_lstats *lstats = netdev_priv(dev);
 219
 220         free_percpu(lstats);
 221         free_netdev(dev);
 222 }
 223
 224 /*
 225  * The loopback device is special. There is only one instance
 226  * per network namespace.
 227  */
 228 static void loopback_setup(struct net_device *dev)
 229 {
 230         dev->get_stats          = &get_stats;
 231         dev->mtu                = (16 * 1024) + 20 + 20 + 12;
 232         dev->hard_start_xmit    = loopback_xmit;
 233         dev->hard_header_len    = ETH_HLEN;     /* 14   */
 234         dev->addr_len           = ETH_ALEN;     /* 6    */
 235         dev->tx_queue_len       = 0;
 236         dev->type               = ARPHRD_LOOPBACK;      /* 0x0001*/
 237         dev->flags              = IFF_LOOPBACK;
 238         dev->features           = NETIF_F_SG | NETIF_F_FRAGLIST
 239 #ifdef LOOPBACK_TSO
 240                 | NETIF_F_TSO
 241 #endif
 242                 | NETIF_F_NO_CSUM
 243                 | NETIF_F_HIGHDMA
 244                 | NETIF_F_LLTX
 245                 | NETIF_F_NETNS_LOCAL,
 246         dev->ethtool_ops        = &loopback_ethtool_ops;
 247         dev->header_ops         = &eth_header_ops;
 248         dev->init = loopback_dev_init;
 249         dev->destructor = loopback_dev_free;
 250 }
 251
 252 /* Setup and register the loopback device. */
 253 static __net_init int loopback_net_init(struct net *net)
 254 {
 255         struct net_device *dev;
 256         int err;
 257
 258         err = -ENOMEM;
 259         dev = alloc_netdev(0, "lo", loopback_setup);
 260         if (!dev)
 261                 goto out;
 262
 263         dev->nd_net = net;
 264         err = register_netdev(dev);
 265         if (err)
 266                 goto out_free_netdev;
 267
 268         net->loopback_dev = dev;
 269         return 0;
 270
 271
 272 out_free_netdev:
 273         free_netdev(dev);
 274 out:
 275         if (net == &init_net)
 276                 panic("loopback: Failed to register netdevice: %d\n", err);
 277         return err;
 278 }
 279
 280 static __net_exit void loopback_net_exit(struct net *net)
 281 {
 282         struct net_device *dev = net->loopback_dev;
 283
 284         unregister_netdev(dev);
 285 }
 286
 287 static struct pernet_operations __net_initdata loopback_net_ops = {
 288        .init = loopback_net_init,
 289        .exit = loopback_net_exit,
 290 };
 291
 292 static int __init loopback_init(void)
 293 {
 294         return register_pernet_device(&loopback_net_ops);
 295 }
 296
 297 /* Loopback is special. It should be initialized before any other network
 298  * device and network subsystem.
 299  */
 300 fs_initcall(loopback_init);