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/config.h>
  17 #include <linux/module.h>
  18 #include <asm/scatterlist.h>
  19 #include <asm/semaphore.h>
  20 #include <linux/crypto.h>
  21 #include <linux/pfkeyv2.h>
  22 #include <linux/percpu.h>
  23 #include <linux/smp.h>
  24 #include <linux/list.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/rtnetlink.h>
  27 #include <net/ip.h>
  28 #include <net/xfrm.h>
  29 #include <net/icmp.h>
  30 #include <net/ipcomp.h>
  31
  32 struct ipcomp_tfms {
  33         struct list_head list;
  34         struct crypto_tfm **tfms;
  35         int users;
  36 };
  37
  38 static DECLARE_MUTEX(ipcomp_resource_sem);
  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         int err, plen, dlen;
  46         struct iphdr *iph;
  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         iph = skb->nh.iph;
  76         iph->tot_len = htons(dlen + iph->ihl * 4);
  77 out:
  78         put_cpu();
  79         return err;
  80 }
  81
  82 static int ipcomp_input(struct xfrm_state *x,
  83                         struct xfrm_decap_state *decap, struct sk_buff *skb)
  84 {
  85         u8 nexthdr;
  86         int err = 0;
  87         struct iphdr *iph;
  88         union {
  89                 struct iphdr    iph;
  90                 char            buf[60];
  91         } tmp_iph;
  92
  93
  94         if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
  95             skb_linearize(skb, GFP_ATOMIC) != 0) {
  96                 err = -ENOMEM;
  97                 goto out;
  98         }
  99
 100         skb->ip_summed = CHECKSUM_NONE;
 101
 102         /* Remove ipcomp header and decompress original payload */
 103         iph = skb->nh.iph;
 104         memcpy(&tmp_iph, iph, iph->ihl * 4);
 105         nexthdr = *(u8 *)skb->data;
 106         skb_pull(skb, sizeof(struct ip_comp_hdr));
 107         skb->nh.raw += sizeof(struct ip_comp_hdr);
 108         memcpy(skb->nh.raw, &tmp_iph, tmp_iph.iph.ihl * 4);
 109         iph = skb->nh.iph;
 110         iph->tot_len = htons(ntohs(iph->tot_len) - sizeof(struct ip_comp_hdr));
 111         iph->protocol = nexthdr;
 112         skb->h.raw = skb->data;
 113         err = ipcomp_decompress(x, skb);
 114
 115 out:
 116         return err;
 117 }
 118
 119 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
 120 {
 121         int err, plen, dlen, ihlen;
 122         struct iphdr *iph = skb->nh.iph;
 123         struct ipcomp_data *ipcd = x->data;
 124         u8 *start, *scratch;
 125         struct crypto_tfm *tfm;
 126         int cpu;
 127
 128         ihlen = iph->ihl * 4;
 129         plen = skb->len - ihlen;
 130         dlen = IPCOMP_SCRATCH_SIZE;
 131         start = skb->data + ihlen;
 132
 133         cpu = get_cpu();
 134         scratch = *per_cpu_ptr(ipcomp_scratches, cpu);
 135         tfm = *per_cpu_ptr(ipcd->tfms, cpu);
 136
 137         err = crypto_comp_compress(tfm, start, plen, scratch, &dlen);
 138         if (err)
 139                 goto out;
 140
 141         if ((dlen + sizeof(struct ip_comp_hdr)) >= plen) {
 142                 err = -EMSGSIZE;
 143                 goto out;
 144         }
 145
 146         memcpy(start + sizeof(struct ip_comp_hdr), scratch, dlen);
 147         put_cpu();
 148
 149         pskb_trim(skb, ihlen + dlen + sizeof(struct ip_comp_hdr));
 150         return 0;
 151
 152 out:
 153         put_cpu();
 154         return err;
 155 }
 156
 157 static int ipcomp_output(struct sk_buff *skb)
 158 {
 159         int err;
 160         struct dst_entry *dst = skb->dst;
 161         struct xfrm_state *x = dst->xfrm;
 162         struct iphdr *iph;
 163         struct ip_comp_hdr *ipch;
 164         struct ipcomp_data *ipcd = x->data;
 165         int hdr_len = 0;
 166
 167         iph = skb->nh.iph;
 168         iph->tot_len = htons(skb->len);
 169         hdr_len = iph->ihl * 4;
 170         if ((skb->len - hdr_len) < ipcd->threshold) {
 171                 /* Don't bother compressing */
 172                 if (x->props.mode) {
 173                         ip_send_check(iph);
 174                 }
 175                 goto out_ok;
 176         }
 177
 178         if ((skb_is_nonlinear(skb) || skb_cloned(skb)) &&
 179             skb_linearize(skb, GFP_ATOMIC) != 0) {
 180                 err = -ENOMEM;
 181                 goto error;
 182         }
 183
 184         err = ipcomp_compress(x, skb);
 185         if (err) {
 186                 if (err == -EMSGSIZE) {
 187                         if (x->props.mode) {
 188                                 iph = skb->nh.iph;
 189                                 ip_send_check(iph);
 190                         }
 191                         goto out_ok;
 192                 }
 193                 goto error;
 194         }
 195
 196         /* Install ipcomp header, convert into ipcomp datagram. */
 197         iph = skb->nh.iph;
 198         iph->tot_len = htons(skb->len);
 199         ipch = (struct ip_comp_hdr *)((char *)iph + iph->ihl * 4);
 200         ipch->nexthdr = iph->protocol;
 201         ipch->flags = 0;
 202         ipch->cpi = htons((u16 )ntohl(x->id.spi));
 203         iph->protocol = IPPROTO_COMP;
 204         ip_send_check(iph);
 205
 206 out_ok:
 207         err = 0;
 208
 209 error:
 210         return err;
 211 }
 212
 213 static void ipcomp4_err(struct sk_buff *skb, u32 info)
 214 {
 215         u32 spi;
 216         struct iphdr *iph = (struct iphdr *)skb->data;
 217         struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
 218         struct xfrm_state *x;
 219
 220         if (skb->h.icmph->type != ICMP_DEST_UNREACH ||
 221             skb->h.icmph->code != ICMP_FRAG_NEEDED)
 222                 return;
 223
 224         spi = ntohl(ntohs(ipch->cpi));
 225         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
 226                               spi, IPPROTO_COMP, AF_INET);
 227         if (!x)
 228                 return;
 229         NETDEBUG(printk(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
 230                spi, NIPQUAD(iph->daddr)));
 231         xfrm_state_put(x);
 232 }
 233
 234 /* We always hold one tunnel user reference to indicate a tunnel */
 235 static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
 236 {
 237         struct xfrm_state *t;
 238
 239         t = xfrm_state_alloc();
 240         if (t == NULL)
 241                 goto out;
 242
 243         t->id.proto = IPPROTO_IPIP;
 244         t->id.spi = x->props.saddr.a4;
 245         t->id.daddr.a4 = x->id.daddr.a4;
 246         memcpy(&t->sel, &x->sel, sizeof(t->sel));
 247         t->props.family = AF_INET;
 248         t->props.mode = 1;
 249         t->props.saddr.a4 = x->props.saddr.a4;
 250         t->props.flags = x->props.flags;
 251
 252         t->type = xfrm_get_type(IPPROTO_IPIP, t->props.family);
 253         if (t->type == NULL)
 254                 goto error;
 255
 256         if (t->type->init_state(t, NULL))
 257                 goto error;
 258
 259         t->km.state = XFRM_STATE_VALID;
 260         atomic_set(&t->tunnel_users, 1);
 261 out:
 262         return t;
 263
 264 error:
 265         t->km.state = XFRM_STATE_DEAD;
 266         xfrm_state_put(t);
 267         t = NULL;
 268         goto out;
 269 }
 270
 271 /*
 272  * Must be protected by xfrm_cfg_sem.  State and tunnel user references are
 273  * always incremented on success.
 274  */
 275 static int ipcomp_tunnel_attach(struct xfrm_state *x)
 276 {
 277         int err = 0;
 278         struct xfrm_state *t;
 279
 280         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
 281                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
 282         if (!t) {
 283                 t = ipcomp_tunnel_create(x);
 284                 if (!t) {
 285                         err = -EINVAL;
 286                         goto out;
 287                 }
 288                 xfrm_state_insert(t);
 289                 xfrm_state_hold(t);
 290         }
 291         x->tunnel = t;
 292         atomic_inc(&t->tunnel_users);
 293 out:
 294         return err;
 295 }
 296
 297 static void ipcomp_free_scratches(void)
 298 {
 299         int i;
 300         void **scratches;
 301
 302         if (--ipcomp_scratch_users)
 303                 return;
 304
 305         scratches = ipcomp_scratches;
 306         if (!scratches)
 307                 return;
 308
 309         for_each_cpu(i) {
 310                 void *scratch = *per_cpu_ptr(scratches, i);
 311                 if (scratch)
 312                         vfree(scratch);
 313         }
 314
 315         free_percpu(scratches);
 316 }
 317
 318 static void **ipcomp_alloc_scratches(void)
 319 {
 320         int i;
 321         void **scratches;
 322
 323         if (ipcomp_scratch_users++)
 324                 return ipcomp_scratches;
 325
 326         scratches = alloc_percpu(void *);
 327         if (!scratches)
 328                 return NULL;
 329
 330         ipcomp_scratches = scratches;
 331
 332         for_each_cpu(i) {
 333                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
 334                 if (!scratch)
 335                         return NULL;
 336                 *per_cpu_ptr(scratches, i) = scratch;
 337         }
 338
 339         return scratches;
 340 }
 341
 342 static void ipcomp_free_tfms(struct crypto_tfm **tfms)
 343 {
 344         struct ipcomp_tfms *pos;
 345         int cpu;
 346
 347         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
 348                 if (pos->tfms == tfms)
 349                         break;
 350         }
 351
 352         BUG_TRAP(pos);
 353
 354         if (--pos->users)
 355                 return;
 356
 357         list_del(&pos->list);
 358         kfree(pos);
 359
 360         if (!tfms)
 361                 return;
 362
 363         for_each_cpu(cpu) {
 364                 struct crypto_tfm *tfm = *per_cpu_ptr(tfms, cpu);
 365                 if (tfm)
 366                         crypto_free_tfm(tfm);
 367         }
 368         free_percpu(tfms);
 369 }
 370
 371 static struct crypto_tfm **ipcomp_alloc_tfms(const char *alg_name)
 372 {
 373         struct ipcomp_tfms *pos;
 374         struct crypto_tfm **tfms;
 375         int cpu;
 376
 377         /* This can be any valid CPU ID so we don't need locking. */
 378         cpu = smp_processor_id();
 379
 380         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
 381                 struct crypto_tfm *tfm;
 382
 383                 tfms = pos->tfms;
 384                 tfm = *per_cpu_ptr(tfms, cpu);
 385
 386                 if (!strcmp(crypto_tfm_alg_name(tfm), alg_name)) {
 387                         pos->users++;
 388                         return tfms;
 389                 }
 390         }
 391
 392         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
 393         if (!pos)
 394                 return NULL;
 395
 396         pos->users = 1;
 397         INIT_LIST_HEAD(&pos->list);
 398         list_add(&pos->list, &ipcomp_tfms_list);
 399
 400         pos->tfms = tfms = alloc_percpu(struct crypto_tfm *);
 401         if (!tfms)
 402                 goto error;
 403
 404         for_each_cpu(cpu) {
 405                 struct crypto_tfm *tfm = crypto_alloc_tfm(alg_name, 0);
 406                 if (!tfm)
 407                         goto error;
 408                 *per_cpu_ptr(tfms, cpu) = tfm;
 409         }
 410
 411         return tfms;
 412
 413 error:
 414         ipcomp_free_tfms(tfms);
 415         return NULL;
 416 }
 417
 418 static void ipcomp_free_data(struct ipcomp_data *ipcd)
 419 {
 420         if (ipcd->tfms)
 421                 ipcomp_free_tfms(ipcd->tfms);
 422         ipcomp_free_scratches();
 423 }
 424
 425 static void ipcomp_destroy(struct xfrm_state *x)
 426 {
 427         struct ipcomp_data *ipcd = x->data;
 428         if (!ipcd)
 429                 return;
 430         xfrm_state_delete_tunnel(x);
 431         down(&ipcomp_resource_sem);
 432         ipcomp_free_data(ipcd);
 433         up(&ipcomp_resource_sem);
 434         kfree(ipcd);
 435 }
 436
 437 static int ipcomp_init_state(struct xfrm_state *x, void *args)
 438 {
 439         int err;
 440         struct ipcomp_data *ipcd;
 441         struct xfrm_algo_desc *calg_desc;
 442
 443         err = -EINVAL;
 444         if (!x->calg)
 445                 goto out;
 446
 447         if (x->encap)
 448                 goto out;
 449
 450         err = -ENOMEM;
 451         ipcd = kmalloc(sizeof(*ipcd), GFP_KERNEL);
 452         if (!ipcd)
 453                 goto out;
 454
 455         memset(ipcd, 0, sizeof(*ipcd));
 456         x->props.header_len = 0;
 457         if (x->props.mode)
 458                 x->props.header_len += sizeof(struct iphdr);
 459
 460         down(&ipcomp_resource_sem);
 461         if (!ipcomp_alloc_scratches())
 462                 goto error;
 463
 464         ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
 465         if (!ipcd->tfms)
 466                 goto error;
 467         up(&ipcomp_resource_sem);
 468
 469         if (x->props.mode) {
 470                 err = ipcomp_tunnel_attach(x);
 471                 if (err)
 472                         goto error_tunnel;
 473         }
 474
 475         calg_desc = xfrm_calg_get_byname(x->calg->alg_name);
 476         BUG_ON(!calg_desc);
 477         ipcd->threshold = calg_desc->uinfo.comp.threshold;
 478         x->data = ipcd;
 479         err = 0;
 480 out:
 481         return err;
 482
 483 error_tunnel:
 484         down(&ipcomp_resource_sem);
 485 error:
 486         ipcomp_free_data(ipcd);
 487         up(&ipcomp_resource_sem);
 488         kfree(ipcd);
 489         goto out;
 490 }
 491
 492 static struct xfrm_type ipcomp_type = {
 493         .description    = "IPCOMP4",
 494         .owner          = THIS_MODULE,
 495         .proto          = IPPROTO_COMP,
 496         .init_state     = ipcomp_init_state,
 497         .destructor     = ipcomp_destroy,
 498         .input          = ipcomp_input,
 499         .output         = ipcomp_output
 500 };
 501
 502 static struct net_protocol ipcomp4_protocol = {
 503         .handler        =       xfrm4_rcv,
 504         .err_handler    =       ipcomp4_err,
 505         .no_policy      =       1,
 506 };
 507
 508 static int __init ipcomp4_init(void)
 509 {
 510         if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
 511                 printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
 512                 return -EAGAIN;
 513         }
 514         if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
 515                 printk(KERN_INFO "ipcomp init: can't add protocol\n");
 516                 xfrm_unregister_type(&ipcomp_type, AF_INET);
 517                 return -EAGAIN;
 518         }
 519         return 0;
 520 }
 521
 522 static void __exit ipcomp4_fini(void)
 523 {
 524         if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
 525                 printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
 526         if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
 527                 printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
 528 }
 529
 530 module_init(ipcomp4_init);
 531 module_exit(ipcomp4_fini);
 532
 533 MODULE_LICENSE("GPL");
 534 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
 535 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
 536