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