Pileus Git - ~andy/linux/blob - net/ipv4/ipcomp.c

   1 /*
   2  * IP Payload Compression Protocol (IPComp) - RFC3173.
   3  *
   4  * Copyright (c) 2003 James Morris <jmorris@intercode.com.au>
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the Free
   8  * Software Foundation; either version 2 of the License, or (at your option)
   9  * any later version.
  10  *
  11  * Todo:
  12  *   - Tunable compression parameters.
  13  *   - Compression stats.
  14  *   - Adaptive compression.
  15  */
  16 #include <linux/module.h>
  17 #include <asm/scatterlist.h>
  18 #include <asm/semaphore.h>
  19 #include <linux/crypto.h>
  20 #include <linux/pfkeyv2.h>
  21 #include <linux/percpu.h>
  22 #include <linux/smp.h>
  23 #include <linux/list.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/rtnetlink.h>
  26 #include <linux/mutex.h>
  27 #include <net/ip.h>
  28 #include <net/xfrm.h>
  29 #include <net/icmp.h>
  30 #include <net/ipcomp.h>
  31 #include <net/protocol.h>
  32
  33 struct ipcomp_tfms {
  34         struct list_head list;
  35         struct crypto_comp **tfms;
  36         int users;
  37 };
  38
  39 static DEFINE_MUTEX(ipcomp_resource_mutex);
  40 static void **ipcomp_scratches;
  41 static int ipcomp_scratch_users;
  42 static LIST_HEAD(ipcomp_tfms_list);
  43
  44 static int ipcomp_decompress(struct xfrm_state *x, struct sk_buff *skb)
  45 {
  46         struct ipcomp_data *ipcd = x->data;
  47         const int plen = skb->len;
  48         int dlen = IPCOMP_SCRATCH_SIZE;
  49         const u8 *start = skb->data;
  50         const int cpu = get_cpu();
  51         u8 *scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
  52         struct crypto_comp *tfm = *per_cpu_ptr(ipcd->tfms, cpu);
  53         int err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
  54
  55         if (err)
  56                 goto out;
  57
  58         if (dlen < (plen + sizeof(struct ip_comp_hdr))) {
  59                 err = -EINVAL;
  60                 goto out;
  61         }
  62
  63         err = pskb_expand_head(skb, 0, dlen - plen, GFP_ATOMIC);
  64         if (err)
  65                 goto out;
  66
  67         skb->truesize += dlen - plen;
  68         __skb_put(skb, dlen - plen);
  69         skb_copy_to_linear_data(skb, scratch, dlen);
  70 out:
  71         put_cpu();
  72         return err;
  73 }
  74
  75 static int ipcomp_input(struct xfrm_state *x, struct sk_buff *skb)
  76 {
  77         int err = -ENOMEM;
  78         struct iphdr *iph;
  79         struct ip_comp_hdr *ipch;
  80
  81         if (skb_linearize_cow(skb))
  82                 goto out;
  83
  84         skb->ip_summed = CHECKSUM_NONE;
  85
  86         /* Remove ipcomp header and decompress original payload */
  87         iph = ip_hdr(skb);
  88         ipch = (void *)skb->data;
  89         iph->protocol = ipch->nexthdr;
  90         skb->transport_header = skb->network_header + sizeof(*ipch);
  91         __skb_pull(skb, sizeof(*ipch));
  92         err = ipcomp_decompress(x, skb);
  93
  94 out:
  95         return err;
  96 }
  97
  98 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
  99 {
 100         struct ipcomp_data *ipcd = x->data;
 101         const int ihlen = skb_transport_offset(skb);
 102         const int plen = skb->len - ihlen;
 103         int dlen = IPCOMP_SCRATCH_SIZE;
 104         u8 *start = skb_transport_header(skb);
 105         const int cpu = get_cpu();
 106         u8 *scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
 107         struct crypto_comp *tfm = *per_cpu_ptr(ipcd->tfms, cpu);
 108         int err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
 109
 110         if (err)
 111                 goto out;
 112
 113         if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
 114                 err = -EMSGSIZE;
 115                 goto out;
 116         }
 117
 118         memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
 119         put_cpu();
 120
 121         pskb_trim(skb, ihlen + dlen + sizeof(struct ip_comp_hdr));
 122         return 0;
 123
 124 out:
 125         put_cpu();
 126         return err;
 127 }
 128
 129 static int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
 130 {
 131         int err;
 132         struct ip_comp_hdr *ipch;
 133         struct ipcomp_data *ipcd = x->data;
 134         int hdr_len = 0;
 135         struct iphdr *iph = ip_hdr(skb);
 136
 137         skb_push(skb, -skb_network_offset(skb));
 138         iph->tot_len = htons(skb->len);
 139         hdr_len = iph->ihl * 4;
 140         if ((skb->len - hdr_len) < ipcd->threshold) {
 141                 /* Don't bother compressing */
 142                 goto out_ok;
 143         }
 144
 145         if (skb_linearize_cow(skb))
 146                 goto out_ok;
 147
 148         err = ipcomp_compress(x, skb);
 149         iph = ip_hdr(skb);
 150
 151         if (err) {
 152                 goto out_ok;
 153         }
 154
 155         /* Install ipcomp header, convert into ipcomp datagram. */
 156         iph->tot_len = htons(skb->len);
 157         ipch = ip_comp_hdr(skb);
 158         ipch->nexthdr = *skb_mac_header(skb);
 159         ipch->flags = 0;
 160         ipch->cpi = htons((u16 )ntohl(x->id.spi));
 161         *skb_mac_header(skb) = IPPROTO_COMP;
 162         ip_send_check(iph);
 163         return 0;
 164
 165 out_ok:
 166         if (x->props.mode == XFRM_MODE_TUNNEL)
 167                 ip_send_check(iph);
 168         return 0;
 169 }
 170
 171 static void ipcomp4_err(struct sk_buff *skb, u32 info)
 172 {
 173         __be32 spi;
 174         struct iphdr *iph = (struct iphdr *)skb->data;
 175         struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
 176         struct xfrm_state *x;
 177
 178         if (icmp_hdr(skb)->type != ICMP_DEST_UNREACH ||
 179             icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
 180                 return;
 181
 182         spi = htonl(ntohs(ipch->cpi));
 183         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
 184                               spi, IPPROTO_COMP, AF_INET);
 185         if (!x)
 186                 return;
 187         NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
 188                  spi, NIPQUAD(iph->daddr));
 189         xfrm_state_put(x);
 190 }
 191
 192 /* We always hold one tunnel user reference to indicate a tunnel */
 193 static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
 194 {
 195         struct xfrm_state *t;
 196         u8 mode = XFRM_MODE_TUNNEL;
 197
 198         t = xfrm_state_alloc();
 199         if (t == NULL)
 200                 goto out;
 201
 202         t->id.proto = IPPROTO_IPIP;
 203         t->id.spi = x->props.saddr.a4;
 204         t->id.daddr.a4 = x->id.daddr.a4;
 205         memcpy(&t->sel, &x->sel, sizeof(t->sel));
 206         t->props.family = AF_INET;
 207         if (x->props.mode == XFRM_MODE_BEET)
 208                 mode = x->props.mode;
 209         t->props.mode = mode;
 210         t->props.saddr.a4 = x->props.saddr.a4;
 211         t->props.flags = x->props.flags;
 212
 213         if (xfrm_init_state(t))
 214                 goto error;
 215
 216         atomic_set(&t->tunnel_users, 1);
 217 out:
 218         return t;
 219
 220 error:
 221         t->km.state = XFRM_STATE_DEAD;
 222         xfrm_state_put(t);
 223         t = NULL;
 224         goto out;
 225 }
 226
 227 /*
 228  * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
 229  * always incremented on success.
 230  */
 231 static int ipcomp_tunnel_attach(struct xfrm_state *x)
 232 {
 233         int err = 0;
 234         struct xfrm_state *t;
 235
 236         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
 237                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
 238         if (!t) {
 239                 t = ipcomp_tunnel_create(x);
 240                 if (!t) {
 241                         err = -EINVAL;
 242                         goto out;
 243                 }
 244                 xfrm_state_insert(t);
 245                 xfrm_state_hold(t);
 246         }
 247         x->tunnel = t;
 248         atomic_inc(&t->tunnel_users);
 249 out:
 250         return err;
 251 }
 252
 253 static void ipcomp_free_scratches(void)
 254 {
 255         int i;
 256         void **scratches;
 257
 258         if (--ipcomp_scratch_users)
 259                 return;
 260
 261         scratches = ipcomp_scratches;
 262         if (!scratches)
 263                 return;
 264
 265         for_each_possible_cpu(i)
 266                 vfree(*per_cpu_ptr(scratches, i));
 267
 268         free_percpu(scratches);
 269 }
 270
 271 static void **ipcomp_alloc_scratches(void)
 272 {
 273         int i;
 274         void **scratches;
 275
 276         if (ipcomp_scratch_users++)
 277                 return ipcomp_scratches;
 278
 279         scratches = alloc_percpu(void *);
 280         if (!scratches)
 281                 return NULL;
 282
 283         ipcomp_scratches = scratches;
 284
 285         for_each_possible_cpu(i) {
 286                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
 287                 if (!scratch)
 288                         return NULL;
 289                 *per_cpu_ptr(scratches, i) = scratch;
 290         }
 291
 292         return scratches;
 293 }
 294
 295 static void ipcomp_free_tfms(struct crypto_comp **tfms)
 296 {
 297         struct ipcomp_tfms *pos;
 298         int cpu;
 299
 300         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
 301                 if (pos->tfms == tfms)
 302                         break;
 303         }
 304
 305         BUG_TRAP(pos);
 306
 307         if (--pos->users)
 308                 return;
 309
 310         list_del(&pos->list);
 311         kfree(pos);
 312
 313         if (!tfms)
 314                 return;
 315
 316         for_each_possible_cpu(cpu) {
 317                 struct crypto_comp *tfm = *per_cpu_ptr(tfms, cpu);
 318                 crypto_free_comp(tfm);
 319         }
 320         free_percpu(tfms);
 321 }
 322
 323 static struct crypto_comp **ipcomp_alloc_tfms(const char *alg_name)
 324 {
 325         struct ipcomp_tfms *pos;
 326         struct crypto_comp **tfms;
 327         int cpu;
 328
 329         /* This can be any valid CPU ID so we don't need locking. */
 330         cpu = raw_smp_processor_id();
 331
 332         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
 333                 struct crypto_comp *tfm;
 334
 335                 tfms = pos->tfms;
 336                 tfm = *per_cpu_ptr(tfms, cpu);
 337
 338                 if (!strcmp(crypto_comp_name(tfm), alg_name)) {
 339                         pos->users++;
 340                         return tfms;
 341                 }
 342         }
 343
 344         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
 345         if (!pos)
 346                 return NULL;
 347
 348         pos->users = 1;
 349         INIT_LIST_HEAD(&pos->list);
 350         list_add(&pos->list, &ipcomp_tfms_list);
 351
 352         pos->tfms = tfms = alloc_percpu(struct crypto_comp *);
 353         if (!tfms)
 354                 goto error;
 355
 356         for_each_possible_cpu(cpu) {
 357                 struct crypto_comp *tfm = crypto_alloc_comp(alg_name, 0,
 358                                                             CRYPTO_ALG_ASYNC);
 359                 if (!tfm)
 360                         goto error;
 361                 *per_cpu_ptr(tfms, cpu) = tfm;
 362         }
 363
 364         return tfms;
 365
 366 error:
 367         ipcomp_free_tfms(tfms);
 368         return NULL;
 369 }
 370
 371 static void ipcomp_free_data(struct ipcomp_data *ipcd)
 372 {
 373         if (ipcd->tfms)
 374                 ipcomp_free_tfms(ipcd->tfms);
 375         ipcomp_free_scratches();
 376 }
 377
 378 static void ipcomp_destroy(struct xfrm_state *x)
 379 {
 380         struct ipcomp_data *ipcd = x->data;
 381         if (!ipcd)
 382                 return;
 383         xfrm_state_delete_tunnel(x);
 384         mutex_lock(&ipcomp_resource_mutex);
 385         ipcomp_free_data(ipcd);
 386         mutex_unlock(&ipcomp_resource_mutex);
 387         kfree(ipcd);
 388 }
 389
 390 static int ipcomp_init_state(struct xfrm_state *x)
 391 {
 392         int err;
 393         struct ipcomp_data *ipcd;
 394         struct xfrm_algo_desc *calg_desc;
 395
 396         err = -EINVAL;
 397         if (!x->calg)
 398                 goto out;
 399
 400         if (x->encap)
 401                 goto out;
 402
 403         err = -ENOMEM;
 404         ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
 405         if (!ipcd)
 406                 goto out;
 407
 408         x->props.header_len = 0;
 409         if (x->props.mode == XFRM_MODE_TUNNEL)
 410                 x->props.header_len += sizeof(struct iphdr);
 411
 412         mutex_lock(&ipcomp_resource_mutex);
 413         if (!ipcomp_alloc_scratches())
 414                 goto error;
 415
 416         ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
 417         if (!ipcd->tfms)
 418                 goto error;
 419         mutex_unlock(&ipcomp_resource_mutex);
 420
 421         if (x->props.mode == XFRM_MODE_TUNNEL) {
 422                 err = ipcomp_tunnel_attach(x);
 423                 if (err)
 424                         goto error_tunnel;
 425         }
 426
 427         calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
 428         BUG_ON(!calg_desc);
 429         ipcd->threshold = calg_desc->uinfo.comp.threshold;
 430         x->data = ipcd;
 431         err = 0;
 432 out:
 433         return err;
 434
 435 error_tunnel:
 436         mutex_lock(&ipcomp_resource_mutex);
 437 error:
 438         ipcomp_free_data(ipcd);
 439         mutex_unlock(&ipcomp_resource_mutex);
 440         kfree(ipcd);
 441         goto out;
 442 }
 443
 444 static struct xfrm_type ipcomp_type = {
 445         .description    = "IPCOMP4",
 446         .owner          = THIS_MODULE,
 447         .proto          = IPPROTO_COMP,
 448         .init_state     = ipcomp_init_state,
 449         .destructor     = ipcomp_destroy,
 450         .input          = ipcomp_input,
 451         .output         = ipcomp_output
 452 };
 453
 454 static struct net_protocol ipcomp4_protocol = {
 455         .handler        =       xfrm4_rcv,
 456         .err_handler    =       ipcomp4_err,
 457         .no_policy      =       1,
 458 };
 459
 460 static int __init ipcomp4_init(void)
 461 {
 462         if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
 463                 printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
 464                 return -EAGAIN;
 465         }
 466         if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
 467                 printk(KERN_INFO "ipcomp init: can't add protocol\n");
 468                 xfrm_unregister_type(&ipcomp_type, AF_INET);
 469                 return -EAGAIN;
 470         }
 471         return 0;
 472 }
 473
 474 static void __exit ipcomp4_fini(void)
 475 {
 476         if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
 477                 printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
 478         if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
 479                 printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
 480 }
 481
 482 module_init(ipcomp4_init);
 483 module_exit(ipcomp4_fini);
 484
 485 MODULE_LICENSE("GPL");
 486 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
 487 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
 488
 489 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_COMP);