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 /* KISS: just allocate small chunks and copy bits.
  68  *
  69  * So, in fact, this is documentation, explaining what we expect
  70  * of largesending device modulo TCP checksum, which is ignored for loopback.
  71  */
  72
  73 #ifdef LOOPBACK_TSO
  74 static void emulate_large_send_offload(struct sk_buff *skb)
  75 {
  76         struct iphdr *iph = ip_hdr(skb);
  77         struct tcphdr *th = (struct tcphdr *)(skb_network_header(skb) +
  78                                               (iph->ihl * 4));
  79         unsigned int doffset = (iph->ihl + th->doff) * 4;
  80         unsigned int mtu = skb_shinfo(skb)->gso_size + doffset;
  81         unsigned int offset = 0;
  82         u32 seq = ntohl(th->seq);
  83         u16 id  = ntohs(iph->id);
  84
  85         while (offset + doffset < skb->len) {
  86                 unsigned int frag_size = min(mtu, skb->len - offset) - doffset;
  87                 struct sk_buff *nskb = alloc_skb(mtu + 32, GFP_ATOMIC);
  88
  89                 if (!nskb)
  90                         break;
  91                 skb_reserve(nskb, 32);
  92                 skb_set_mac_header(nskb, -ETH_HLEN);
  93                 skb_reset_network_header(nskb);
  94                 iph = ip_hdr(nskb);
  95                 skb_copy_to_linear_data(nskb, skb_network_header(skb),
  96                                         doffset);
  97                 if (skb_copy_bits(skb,
  98                                   doffset + offset,
  99                                   nskb->data + doffset,
 100                                   frag_size))
 101                         BUG();
 102                 skb_put(nskb, doffset + frag_size);
 103                 nskb->ip_summed = CHECKSUM_UNNECESSARY;
 104                 nskb->dev = skb->dev;
 105                 nskb->priority = skb->priority;
 106                 nskb->protocol = skb->protocol;
 107                 nskb->dst = dst_clone(skb->dst);
 108                 memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
 109                 nskb->pkt_type = skb->pkt_type;
 110
 111                 th = (struct tcphdr *)(skb_network_header(nskb) + iph->ihl * 4);
 112                 iph->tot_len = htons(frag_size + doffset);
 113                 iph->id = htons(id);
 114                 iph->check = 0;
 115                 iph->check = ip_fast_csum((unsigned char *) iph, iph->ihl);
 116                 th->seq = htonl(seq);
 117                 if (offset + doffset + frag_size < skb->len)
 118                         th->fin = th->psh = 0;
 119                 netif_rx(nskb);
 120                 offset += frag_size;
 121                 seq += frag_size;
 122                 id++;
 123         }
 124
 125         dev_kfree_skb(skb);
 126 }
 127 #endif /* LOOPBACK_TSO */
 128
 129 /*
 130  * The higher levels take care of making this non-reentrant (it's
 131  * called with bh's disabled).
 132  */
 133 static int loopback_xmit(struct sk_buff *skb, struct net_device *dev)
 134 {
 135         struct pcpu_lstats *pcpu_lstats, *lb_stats;
 136
 137         skb_orphan(skb);
 138
 139         skb->protocol = eth_type_trans(skb,dev);
 140 #ifndef LOOPBACK_MUST_CHECKSUM
 141         skb->ip_summed = CHECKSUM_UNNECESSARY;
 142 #endif
 143
 144 #ifdef LOOPBACK_TSO
 145         if (skb_is_gso(skb)) {
 146                 BUG_ON(skb->protocol != htons(ETH_P_IP));
 147                 BUG_ON(ip_hdr(skb)->protocol != IPPROTO_TCP);
 148
 149                 emulate_large_send_offload(skb);
 150                 return 0;
 151         }
 152 #endif
 153         dev->last_rx = jiffies;
 154
 155         /* it's OK to use per_cpu_ptr() because BHs are off */
 156         pcpu_lstats = netdev_priv(dev);
 157         lb_stats = per_cpu_ptr(pcpu_lstats, smp_processor_id());
 158         lb_stats->bytes += skb->len;
 159         lb_stats->packets++;
 160
 161         netif_rx(skb);
 162
 163         return 0;
 164 }
 165
 166 static struct net_device_stats *get_stats(struct net_device *dev)
 167 {
 168         const struct pcpu_lstats *pcpu_lstats;
 169         struct net_device_stats *stats = &dev->stats;
 170         unsigned long bytes = 0;
 171         unsigned long packets = 0;
 172         int i;
 173
 174         pcpu_lstats = netdev_priv(dev);
 175         for_each_possible_cpu(i) {
 176                 const struct pcpu_lstats *lb_stats;
 177
 178                 lb_stats = per_cpu_ptr(pcpu_lstats, i);
 179                 bytes   += lb_stats->bytes;
 180                 packets += lb_stats->packets;
 181         }
 182         stats->rx_packets = packets;
 183         stats->tx_packets = packets;
 184         stats->rx_bytes = bytes;
 185         stats->tx_bytes = bytes;
 186         return stats;
 187 }
 188
 189 static u32 always_on(struct net_device *dev)
 190 {
 191         return 1;
 192 }
 193
 194 static const struct ethtool_ops loopback_ethtool_ops = {
 195         .get_link               = always_on,
 196         .set_tso                = ethtool_op_set_tso,
 197         .get_tx_csum            = always_on,
 198         .get_sg                 = always_on,
 199         .get_rx_csum            = always_on,
 200 };
 201
 202 static int loopback_dev_init(struct net_device *dev)
 203 {
 204         struct pcpu_lstats *lstats;
 205
 206         lstats = alloc_percpu(struct pcpu_lstats);
 207         if (!lstats)
 208                 return -ENOMEM;
 209
 210         dev->priv = lstats;
 211         return 0;
 212 }
 213
 214 static void loopback_dev_free(struct net_device *dev)
 215 {
 216         struct pcpu_lstats *lstats = netdev_priv(dev);
 217
 218         free_percpu(lstats);
 219         free_netdev(dev);
 220 }
 221
 222 /*
 223  * The loopback device is special. There is only one instance
 224  * per network namespace.
 225  */
 226 static void loopback_setup(struct net_device *dev)
 227 {
 228         dev->get_stats          = &get_stats;
 229         dev->mtu                = (16 * 1024) + 20 + 20 + 12;
 230         dev->hard_start_xmit    = loopback_xmit;
 231         dev->hard_header_len    = ETH_HLEN;     /* 14   */
 232         dev->addr_len           = ETH_ALEN;     /* 6    */
 233         dev->tx_queue_len       = 0;
 234         dev->type               = ARPHRD_LOOPBACK;      /* 0x0001*/
 235         dev->flags              = IFF_LOOPBACK;
 236         dev->features           = NETIF_F_SG | NETIF_F_FRAGLIST
 237 #ifdef LOOPBACK_TSO
 238                 | NETIF_F_TSO
 239 #endif
 240                 | NETIF_F_NO_CSUM
 241                 | NETIF_F_HIGHDMA
 242                 | NETIF_F_LLTX
 243                 | NETIF_F_NETNS_LOCAL;
 244         dev->ethtool_ops        = &loopback_ethtool_ops;
 245         dev->header_ops         = &eth_header_ops;
 246         dev->init = loopback_dev_init;
 247         dev->destructor = loopback_dev_free;
 248 }
 249
 250 /* Setup and register the loopback device. */
 251 static __net_init int loopback_net_init(struct net *net)
 252 {
 253         struct net_device *dev;
 254         int err;
 255
 256         err = -ENOMEM;
 257         dev = alloc_netdev(0, "lo", loopback_setup);
 258         if (!dev)
 259                 goto out;
 260
 261         dev->nd_net = net;
 262         err = register_netdev(dev);
 263         if (err)
 264                 goto out_free_netdev;
 265
 266         net->loopback_dev = dev;
 267         return 0;
 268
 269
 270 out_free_netdev:
 271         free_netdev(dev);
 272 out:
 273         if (net == &init_net)
 274                 panic("loopback: Failed to register netdevice: %d\n", err);
 275         return err;
 276 }
 277
 278 static __net_exit void loopback_net_exit(struct net *net)
 279 {
 280         struct net_device *dev = net->loopback_dev;
 281
 282         unregister_netdev(dev);
 283 }
 284
 285 static struct pernet_operations __net_initdata loopback_net_ops = {
 286        .init = loopback_net_init,
 287        .exit = loopback_net_exit,
 288 };
 289
 290 static int __init loopback_init(void)
 291 {
 292         return register_pernet_device(&loopback_net_ops);
 293 }
 294
 295 /* Loopback is special. It should be initialized before any other network
 296  * device and network subsystem.
 297  */
 298 fs_initcall(loopback_init);