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_tfm **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         int err, plen, dlen;
  47         struct ipcomp_data *ipcd = x->data;
  48         u8 *start, *scratch;
  49         struct crypto_tfm *tfm;
  50         int cpu;
  51
  52         plen = skb->len;
  53         dlen = IPCOMP_SCRATCH_SIZE;
  54         start = skb->data;
  55
  56         cpu = get_cpu();
  57         scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
  58         tfm = *per_cpu_ptr(ipcd->tfms, cpu);
  59
  60         err = crypto_comp_decompress(tfm, start, plen, scratch, &dlen);
  61         if (err)
  62                 goto out;
  63
  64         if (dlen < (plen + sizeof(struct ip_comp_hdr))) {
  65                 err = -EINVAL;
  66                 goto out;
  67         }
  68
  69         err = pskb_expand_head(skb, 0, dlen - plen, GFP_ATOMIC);
  70         if (err)
  71                 goto out;
  72
  73         skb_put(skb, dlen - plen);
  74         memcpy(skb->data, scratch, dlen);
  75 out:
  76         put_cpu();
  77         return err;
  78 }
  79
  80 static int ipcomp_input(struct xfrm_state *x, struct sk_buff *skb)
  81 {
  82         int err = -ENOMEM;
  83         struct iphdr *iph;
  84         struct ip_comp_hdr *ipch;
  85
  86         if (skb_linearize_cow(skb))
  87                 goto out;
  88
  89         skb->ip_summed = CHECKSUM_NONE;
  90
  91         /* Remove ipcomp header and decompress original payload */
  92         iph = skb->nh.iph;
  93         ipch = (void *)skb->data;
  94         iph->protocol = ipch->nexthdr;
  95         skb->h.raw = skb->nh.raw + sizeof(*ipch);
  96         __skb_pull(skb, sizeof(*ipch));
  97         err = ipcomp_decompress(x, skb);
  98
  99 out:
 100         return err;
 101 }
 102
 103 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
 104 {
 105         int err, plen, dlen, ihlen;
 106         struct iphdr *iph = skb->nh.iph;
 107         struct ipcomp_data *ipcd = x->data;
 108         u8 *start, *scratch;
 109         struct crypto_tfm *tfm;
 110         int cpu;
 111
 112         ihlen = iph->ihl * 4;
 113         plen = skb->len - ihlen;
 114         dlen = IPCOMP_SCRATCH_SIZE;
 115         start = skb->data + ihlen;
 116
 117         cpu = get_cpu();
 118         scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
 119         tfm = *per_cpu_ptr(ipcd->tfms, cpu);
 120
 121         err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
 122         if (err)
 123                 goto out;
 124
 125         if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
 126                 err = -EMSGSIZE;
 127                 goto out;
 128         }
 129
 130         memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
 131         put_cpu();
 132
 133         pskb_trim(skb, ihlen + dlen + sizeof(struct ip_comp_hdr));
 134         return 0;
 135
 136 out:
 137         put_cpu();
 138         return err;
 139 }
 140
 141 static int ipcomp_output(struct xfrm_state *x, struct sk_buff *skb)
 142 {
 143         int err;
 144         struct iphdr *iph;
 145         struct ip_comp_hdr *ipch;
 146         struct ipcomp_data *ipcd = x->data;
 147         int hdr_len = 0;
 148
 149         iph = skb->nh.iph;
 150         iph->tot_len = htons(skb->len);
 151         hdr_len = iph->ihl * 4;
 152         if ((skb->len - hdr_len) < ipcd->threshold) {
 153                 /* Don't bother compressing */
 154                 goto out_ok;
 155         }
 156
 157         if (skb_linearize_cow(skb))
 158                 goto out_ok;
 159
 160         err = ipcomp_compress(x, skb);
 161         iph = skb->nh.iph;
 162
 163         if (err) {
 164                 goto out_ok;
 165         }
 166
 167         /* Install ipcomp header, convert into ipcomp datagram. */
 168         iph->tot_len = htons(skb->len);
 169         ipch = (struct ip_comp_hdr *)((char *)iph + iph->ihl * 4);
 170         ipch->nexthdr = iph->protocol;
 171         ipch->flags = 0;
 172         ipch->cpi = htons((u16 )ntohl(x->id.spi));
 173         iph->protocol = IPPROTO_COMP;
 174         ip_send_check(iph);
 175         return 0;
 176
 177 out_ok:
 178         if (x->props.mode)
 179                 ip_send_check(iph);
 180         return 0;
 181 }
 182
 183 static void ipcomp4_err(struct sk_buff *skb, u32 info)
 184 {
 185         u32 spi;
 186         struct iphdr *iph = (struct iphdr *)skb->data;
 187         struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
 188         struct xfrm_state *x;
 189
 190         if (skb->h.icmph->type != ICMP_DEST_UNREACH ||
 191             skb->h.icmph->code != ICMP_FRAG_NEEDED)
 192                 return;
 193
 194         spi = htonl(ntohs(ipch->cpi));
 195         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
 196                               spi, IPPROTO_COMP, AF_INET);
 197         if (!x)
 198                 return;
 199         NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
 200                  spi, NIPQUAD(iph->daddr));
 201         xfrm_state_put(x);
 202 }
 203
 204 /* We always hold one tunnel user reference to indicate a tunnel */
 205 static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
 206 {
 207         struct xfrm_state *t;
 208
 209         t = xfrm_state_alloc();
 210         if (t == NULL)
 211                 goto out;
 212
 213         t->id.proto = IPPROTO_IPIP;
 214         t->id.spi = x->props.saddr.a4;
 215         t->id.daddr.a4 = x->id.daddr.a4;
 216         memcpy(&t->sel, &x->sel, sizeof(t->sel));
 217         t->props.family = AF_INET;
 218         t->props.mode = 1;
 219         t->props.saddr.a4 = x->props.saddr.a4;
 220         t->props.flags = x->props.flags;
 221
 222         if (xfrm_init_state(t))
 223                 goto error;
 224
 225         atomic_set(&t->tunnel_users, 1);
 226 out:
 227         return t;
 228
 229 error:
 230         t->km.state = XFRM_STATE_DEAD;
 231         xfrm_state_put(t);
 232         t = NULL;
 233         goto out;
 234 }
 235
 236 /*
 237  * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
 238  * always incremented on success.
 239  */
 240 static int ipcomp_tunnel_attach(struct xfrm_state *x)
 241 {
 242         int err = 0;
 243         struct xfrm_state *t;
 244
 245         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
 246                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
 247         if (!t) {
 248                 t = ipcomp_tunnel_create(x);
 249                 if (!t) {
 250                         err = -EINVAL;
 251                         goto out;
 252                 }
 253                 xfrm_state_insert(t);
 254                 xfrm_state_hold(t);
 255         }
 256         x->tunnel = t;
 257         atomic_inc(&t->tunnel_users);
 258 out:
 259         return err;
 260 }
 261
 262 static void ipcomp_free_scratches(void)
 263 {
 264         int i;
 265         void **scratches;
 266
 267         if (--ipcomp_scratch_users)
 268                 return;
 269
 270         scratches = ipcomp_scratches;
 271         if (!scratches)
 272                 return;
 273
 274         for_each_possible_cpu(i)
 275                 vfree(*per_cpu_ptr(scratches, i));
 276
 277         free_percpu(scratches);
 278 }
 279
 280 static void **ipcomp_alloc_scratches(void)
 281 {
 282         int i;
 283         void **scratches;
 284
 285         if (ipcomp_scratch_users++)
 286                 return ipcomp_scratches;
 287
 288         scratches = alloc_percpu(void *);
 289         if (!scratches)
 290                 return NULL;
 291
 292         ipcomp_scratches = scratches;
 293
 294         for_each_possible_cpu(i) {
 295                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
 296                 if (!scratch)
 297                         return NULL;
 298                 *per_cpu_ptr(scratches, i) = scratch;
 299         }
 300
 301         return scratches;
 302 }
 303
 304 static void ipcomp_free_tfms(struct crypto_tfm **tfms)
 305 {
 306         struct ipcomp_tfms *pos;
 307         int cpu;
 308
 309         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
 310                 if (pos->tfms == tfms)
 311                         break;
 312         }
 313
 314         BUG_TRAP(pos);
 315
 316         if (--pos->users)
 317                 return;
 318
 319         list_del(&pos->list);
 320         kfree(pos);
 321
 322         if (!tfms)
 323                 return;
 324
 325         for_each_possible_cpu(cpu) {
 326                 struct crypto_tfm *tfm = *per_cpu_ptr(tfms, cpu);
 327                 crypto_free_tfm(tfm);
 328         }
 329         free_percpu(tfms);
 330 }
 331
 332 static struct crypto_tfm **ipcomp_alloc_tfms(const char *alg_name)
 333 {
 334         struct ipcomp_tfms *pos;
 335         struct crypto_tfm **tfms;
 336         int cpu;
 337
 338         /* This can be any valid CPU ID so we don't need locking. */
 339         cpu = raw_smp_processor_id();
 340
 341         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
 342                 struct crypto_tfm *tfm;
 343
 344                 tfms = pos->tfms;
 345                 tfm = *per_cpu_ptr(tfms, cpu);
 346
 347                 if (!strcmp(crypto_tfm_alg_name(tfm), alg_name)) {
 348                         pos->users++;
 349                         return tfms;
 350                 }
 351         }
 352
 353         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
 354         if (!pos)
 355                 return NULL;
 356
 357         pos->users = 1;
 358         INIT_LIST_HEAD(&pos->list);
 359         list_add(&pos->list, &ipcomp_tfms_list);
 360
 361         pos->tfms = tfms = alloc_percpu(struct crypto_tfm *);
 362         if (!tfms)
 363                 goto error;
 364
 365         for_each_possible_cpu(cpu) {
 366                 struct crypto_tfm *tfm = crypto_alloc_tfm(alg_name, 0);
 367                 if (!tfm)
 368                         goto error;
 369                 *per_cpu_ptr(tfms, cpu) = tfm;
 370         }
 371
 372         return tfms;
 373
 374 error:
 375         ipcomp_free_tfms(tfms);
 376         return NULL;
 377 }
 378
 379 static void ipcomp_free_data(struct ipcomp_data *ipcd)
 380 {
 381         if (ipcd->tfms)
 382                 ipcomp_free_tfms(ipcd->tfms);
 383         ipcomp_free_scratches();
 384 }
 385
 386 static void ipcomp_destroy(struct xfrm_state *x)
 387 {
 388         struct ipcomp_data *ipcd = x->data;
 389         if (!ipcd)
 390                 return;
 391         xfrm_state_delete_tunnel(x);
 392         mutex_lock(&ipcomp_resource_mutex);
 393         ipcomp_free_data(ipcd);
 394         mutex_unlock(&ipcomp_resource_mutex);
 395         kfree(ipcd);
 396 }
 397
 398 static int ipcomp_init_state(struct xfrm_state *x)
 399 {
 400         int err;
 401         struct ipcomp_data *ipcd;
 402         struct xfrm_algo_desc *calg_desc;
 403
 404         err = -EINVAL;
 405         if (!x->calg)
 406                 goto out;
 407
 408         if (x->encap)
 409                 goto out;
 410
 411         err = -ENOMEM;
 412         ipcd = kmalloc(sizeof(*ipcd), GFP_KERNEL);
 413         if (!ipcd)
 414                 goto out;
 415
 416         memset(ipcd, 0, sizeof(*ipcd));
 417         x->props.header_len = 0;
 418         if (x->props.mode)
 419                 x->props.header_len += sizeof(struct iphdr);
 420
 421         mutex_lock(&ipcomp_resource_mutex);
 422         if (!ipcomp_alloc_scratches())
 423                 goto error;
 424
 425         ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
 426         if (!ipcd->tfms)
 427                 goto error;
 428         mutex_unlock(&ipcomp_resource_mutex);
 429
 430         if (x->props.mode) {
 431                 err = ipcomp_tunnel_attach(x);
 432                 if (err)
 433                         goto error_tunnel;
 434         }
 435
 436         calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
 437         BUG_ON(!calg_desc);
 438         ipcd->threshold = calg_desc->uinfo.comp.threshold;
 439         x->data = ipcd;
 440         err = 0;
 441 out:
 442         return err;
 443
 444 error_tunnel:
 445         mutex_lock(&ipcomp_resource_mutex);
 446 error:
 447         ipcomp_free_data(ipcd);
 448         mutex_unlock(&ipcomp_resource_mutex);
 449         kfree(ipcd);
 450         goto out;
 451 }
 452
 453 static struct xfrm_type ipcomp_type = {
 454         .description    = "IPCOMP4",
 455         .owner          = THIS_MODULE,
 456         .proto          = IPPROTO_COMP,
 457         .init_state     = ipcomp_init_state,
 458         .destructor     = ipcomp_destroy,
 459         .input          = ipcomp_input,
 460         .output         = ipcomp_output
 461 };
 462
 463 static struct net_protocol ipcomp4_protocol = {
 464         .handler        =       xfrm4_rcv,
 465         .err_handler    =       ipcomp4_err,
 466         .no_policy      =       1,
 467 };
 468
 469 static int __init ipcomp4_init(void)
 470 {
 471         if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
 472                 printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
 473                 return -EAGAIN;
 474         }
 475         if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
 476                 printk(KERN_INFO "ipcomp init: can't add protocol\n");
 477                 xfrm_unregister_type(&ipcomp_type, AF_INET);
 478                 return -EAGAIN;
 479         }
 480         return 0;
 481 }
 482
 483 static void __exit ipcomp4_fini(void)
 484 {
 485         if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
 486                 printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
 487         if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
 488                 printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
 489 }
 490
 491 module_init(ipcomp4_init);
 492 module_exit(ipcomp4_fini);
 493
 494 MODULE_LICENSE("GPL");
 495 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
 496 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
 497