Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | * |
Jesper Juhl | 02c30a8 | 2005-05-05 16:16:16 -0700 | [diff] [blame] | 10 | * Authors: Ross Biro |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Eric Dumazet | 5175c37 | 2006-10-18 20:51:57 -0700 | [diff] [blame] | 61 | struct pcpu_lstats { |
| 62 | unsigned long packets; |
| 63 | unsigned long bytes; |
| 64 | }; |
| 65 | static DEFINE_PER_CPU(struct pcpu_lstats, pcpu_lstats); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Chuck Ebbert | d2ae1d2 | 2005-07-02 21:28:21 -0400 | [diff] [blame] | 75 | #ifdef LOOPBACK_TSO |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 76 | static void emulate_large_send_offload(struct sk_buff *skb) |
| 77 | { |
Arnaldo Carvalho de Melo | eddc9ec | 2007-04-20 22:47:35 -0700 | [diff] [blame] | 78 | struct iphdr *iph = ip_hdr(skb); |
Arnaldo Carvalho de Melo | d56f90a | 2007-04-10 20:50:43 -0700 | [diff] [blame] | 79 | struct tcphdr *th = (struct tcphdr *)(skb_network_header(skb) + |
| 80 | (iph->ihl * 4)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 81 | unsigned int doffset = (iph->ihl + th->doff) * 4; |
Herbert Xu | 7967168 | 2006-06-22 02:40:14 -0700 | [diff] [blame] | 82 | unsigned int mtu = skb_shinfo(skb)->gso_size + doffset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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); |
Arnaldo Carvalho de Melo | 48d49d0c | 2007-03-10 12:30:58 -0300 | [diff] [blame] | 94 | skb_set_mac_header(nskb, -ETH_HLEN); |
Arnaldo Carvalho de Melo | c1d2bbe | 2007-04-10 20:45:18 -0700 | [diff] [blame] | 95 | skb_reset_network_header(nskb); |
Arnaldo Carvalho de Melo | eddc9ec | 2007-04-20 22:47:35 -0700 | [diff] [blame] | 96 | iph = ip_hdr(nskb); |
Arnaldo Carvalho de Melo | 27d7ff4 | 2007-03-31 11:55:19 -0300 | [diff] [blame] | 97 | skb_copy_to_linear_data(nskb, skb_network_header(skb), |
| 98 | doffset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Arnaldo Carvalho de Melo | d56f90a | 2007-04-10 20:50:43 -0700 | [diff] [blame] | 113 | th = (struct tcphdr *)(skb_network_header(nskb) + iph->ihl * 4); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | } |
Chuck Ebbert | d2ae1d2 | 2005-07-02 21:28:21 -0400 | [diff] [blame] | 129 | #endif /* LOOPBACK_TSO */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | { |
Eric Dumazet | 5175c37 | 2006-10-18 20:51:57 -0700 | [diff] [blame] | 137 | struct pcpu_lstats *lb_stats; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 138 | |
| 139 | skb_orphan(skb); |
| 140 | |
Chuck Ebbert | 0e920bf | 2005-07-02 21:28:23 -0400 | [diff] [blame] | 141 | skb->protocol = eth_type_trans(skb,dev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | #ifndef LOOPBACK_MUST_CHECKSUM |
| 143 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
| 144 | #endif |
| 145 | |
Chuck Ebbert | d2ae1d2 | 2005-07-02 21:28:21 -0400 | [diff] [blame] | 146 | #ifdef LOOPBACK_TSO |
Herbert Xu | 89114af | 2006-07-08 13:34:32 -0700 | [diff] [blame] | 147 | if (skb_is_gso(skb)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 148 | BUG_ON(skb->protocol != htons(ETH_P_IP)); |
Arnaldo Carvalho de Melo | eddc9ec | 2007-04-20 22:47:35 -0700 | [diff] [blame] | 149 | BUG_ON(ip_hdr(skb)->protocol != IPPROTO_TCP); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 150 | |
| 151 | emulate_large_send_offload(skb); |
| 152 | return 0; |
| 153 | } |
Chuck Ebbert | d2ae1d2 | 2005-07-02 21:28:21 -0400 | [diff] [blame] | 154 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | dev->last_rx = jiffies; |
| 156 | |
Eric Dumazet | 58f5397 | 2006-10-20 00:32:41 -0700 | [diff] [blame] | 157 | /* it's OK to use __get_cpu_var() because BHs are off */ |
| 158 | lb_stats = &__get_cpu_var(pcpu_lstats); |
Eric Dumazet | 5175c37 | 2006-10-18 20:51:57 -0700 | [diff] [blame] | 159 | lb_stats->bytes += skb->len; |
| 160 | lb_stats->packets++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | |
| 162 | netif_rx(skb); |
| 163 | |
Eric Dumazet | 58f5397 | 2006-10-20 00:32:41 -0700 | [diff] [blame] | 164 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 165 | } |
| 166 | |
| 167 | static struct net_device_stats *get_stats(struct net_device *dev) |
| 168 | { |
Eric Dumazet | 3303680 | 2007-04-10 13:25:40 -0700 | [diff] [blame] | 169 | struct net_device_stats *stats = &dev->stats; |
Eric Dumazet | 5175c37 | 2006-10-18 20:51:57 -0700 | [diff] [blame] | 170 | unsigned long bytes = 0; |
| 171 | unsigned long packets = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | int i; |
| 173 | |
KAMEZAWA Hiroyuki | 0fed484 | 2006-03-28 01:56:37 -0800 | [diff] [blame] | 174 | for_each_possible_cpu(i) { |
Eric Dumazet | 5175c37 | 2006-10-18 20:51:57 -0700 | [diff] [blame] | 175 | const struct pcpu_lstats *lb_stats; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 176 | |
Eric Dumazet | 5175c37 | 2006-10-18 20:51:57 -0700 | [diff] [blame] | 177 | lb_stats = &per_cpu(pcpu_lstats, i); |
| 178 | bytes += lb_stats->bytes; |
| 179 | packets += lb_stats->packets; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | } |
Eric Dumazet | 5175c37 | 2006-10-18 20:51:57 -0700 | [diff] [blame] | 181 | stats->rx_packets = packets; |
| 182 | stats->tx_packets = packets; |
| 183 | stats->rx_bytes = bytes; |
| 184 | stats->tx_bytes = bytes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | return stats; |
| 186 | } |
| 187 | |
Stephen Hemminger | 7fa6b06 | 2006-09-27 20:33:34 -0700 | [diff] [blame] | 188 | static u32 always_on(struct net_device *dev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 189 | { |
| 190 | return 1; |
| 191 | } |
| 192 | |
Jeff Garzik | 7282d49 | 2006-09-13 14:30:00 -0400 | [diff] [blame] | 193 | static const struct ethtool_ops loopback_ethtool_ops = { |
Stephen Hemminger | 7fa6b06 | 2006-09-27 20:33:34 -0700 | [diff] [blame] | 194 | .get_link = always_on, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | .set_tso = ethtool_op_set_tso, |
Stephen Hemminger | 7fa6b06 | 2006-09-27 20:33:34 -0700 | [diff] [blame] | 196 | .get_tx_csum = always_on, |
| 197 | .get_sg = always_on, |
| 198 | .get_rx_csum = always_on, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | }; |
| 200 | |
Stephen Hemminger | 7fa6b06 | 2006-09-27 20:33:34 -0700 | [diff] [blame] | 201 | /* |
| 202 | * The loopback device is special. There is only one instance and |
| 203 | * it is statically allocated. Don't do this for other devices. |
| 204 | */ |
Daniel Lezcano | 854d836 | 2007-09-25 19:18:04 -0700 | [diff] [blame^] | 205 | static void loopback_setup(struct net_device *dev) |
| 206 | { |
| 207 | dev->get_stats = &get_stats; |
| 208 | dev->mtu = (16 * 1024) + 20 + 20 + 12; |
| 209 | dev->hard_start_xmit = loopback_xmit; |
| 210 | dev->hard_header = eth_header; |
| 211 | dev->hard_header_cache = eth_header_cache; |
| 212 | dev->header_cache_update = eth_header_cache_update; |
| 213 | dev->hard_header_len = ETH_HLEN; /* 14 */ |
| 214 | dev->addr_len = ETH_ALEN; /* 6 */ |
| 215 | dev->tx_queue_len = 0; |
| 216 | dev->type = ARPHRD_LOOPBACK; /* 0x0001*/ |
| 217 | dev->rebuild_header = eth_rebuild_header; |
| 218 | dev->flags = IFF_LOOPBACK; |
| 219 | dev->features = NETIF_F_SG | NETIF_F_FRAGLIST |
Chuck Ebbert | d2ae1d2 | 2005-07-02 21:28:21 -0400 | [diff] [blame] | 220 | #ifdef LOOPBACK_TSO |
Daniel Lezcano | 854d836 | 2007-09-25 19:18:04 -0700 | [diff] [blame^] | 221 | | NETIF_F_TSO |
Chuck Ebbert | d2ae1d2 | 2005-07-02 21:28:21 -0400 | [diff] [blame] | 222 | #endif |
Daniel Lezcano | 854d836 | 2007-09-25 19:18:04 -0700 | [diff] [blame^] | 223 | | NETIF_F_NO_CSUM |
| 224 | | NETIF_F_HIGHDMA |
| 225 | | NETIF_F_LLTX |
| 226 | | NETIF_F_NETNS_LOCAL, |
| 227 | dev->ethtool_ops = &loopback_ethtool_ops; |
| 228 | } |
Daniel Lezcano | de3cb74 | 2007-09-25 19:16:28 -0700 | [diff] [blame] | 229 | |
Ralf Baechle | 2278364 | 2005-08-18 14:05:18 -0700 | [diff] [blame] | 230 | /* Setup and register the loopback device. */ |
Adrian Bunk | 60903f2 | 2007-01-02 00:35:48 -0800 | [diff] [blame] | 231 | static int __init loopback_init(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | { |
Daniel Lezcano | 854d836 | 2007-09-25 19:18:04 -0700 | [diff] [blame^] | 233 | struct net_device *dev; |
| 234 | int err; |
Herbert Xu | aeed9e8 | 2007-07-30 16:37:19 -0700 | [diff] [blame] | 235 | |
Daniel Lezcano | 854d836 | 2007-09-25 19:18:04 -0700 | [diff] [blame^] | 236 | err = -ENOMEM; |
| 237 | dev = alloc_netdev(0, "lo", loopback_setup); |
| 238 | if (!dev) |
| 239 | goto out; |
| 240 | |
| 241 | err = register_netdev(dev); |
| 242 | if (err) |
| 243 | goto out_free_netdev; |
| 244 | |
| 245 | err = 0; |
| 246 | loopback_dev = dev; |
| 247 | |
| 248 | out: |
Herbert Xu | aeed9e8 | 2007-07-30 16:37:19 -0700 | [diff] [blame] | 249 | if (err) |
| 250 | panic("loopback: Failed to register netdevice: %d\n", err); |
Herbert Xu | aeed9e8 | 2007-07-30 16:37:19 -0700 | [diff] [blame] | 251 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 252 | |
Daniel Lezcano | 854d836 | 2007-09-25 19:18:04 -0700 | [diff] [blame^] | 253 | out_free_netdev: |
| 254 | free_netdev(dev); |
| 255 | goto out; |
| 256 | } |
Adrian Bunk | 60903f2 | 2007-01-02 00:35:48 -0800 | [diff] [blame] | 257 | |
Daniel Lezcano | 854d836 | 2007-09-25 19:18:04 -0700 | [diff] [blame^] | 258 | fs_initcall(loopback_init); |
| 259 | |
| 260 | struct net_device *loopback_dev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 261 | EXPORT_SYMBOL(loopback_dev); |