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/err.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 ip_comp_hdr *ipch;
  79
  80         if (skb_linearize_cow(skb))
  81                 goto out;
  82
  83         skb->ip_summed = CHECKSUM_NONE;
  84
  85         /* Remove ipcomp header and decompress original payload */
  86         ipch = (void *)skb->data;
  87         skb->transport_header = skb->network_header + sizeof(*ipch);
  88         __skb_pull(skb, sizeof(*ipch));
  89         err = ipcomp_decompress(x, skb);
  90         if (err)
  91                 goto out;
  92
  93         err = ipch->nexthdr;
  94
  95 out:
  96         return err;
  97 }
  98
  99 static int ipcomp_compress(struct xfrm_state *x, struct sk_buff *skb)
 100 {
 101         struct ipcomp_data *ipcd = x->data;
 102         const int plen = skb->len;
 103         int dlen = IPCOMP_SCRATCH_SIZE;
 104         u8 *start = skb->data;
 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, 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
 135         if (skb->len < ipcd->threshold) {
 136                 /* Don't bother compressing */
 137                 goto out_ok;
 138         }
 139
 140         if (skb_linearize_cow(skb))
 141                 goto out_ok;
 142
 143         err = ipcomp_compress(x, skb);
 144
 145         if (err) {
 146                 goto out_ok;
 147         }
 148
 149         /* Install ipcomp header, convert into ipcomp datagram. */
 150         ipch = ip_comp_hdr(skb);
 151         ipch->nexthdr = *skb_mac_header(skb);
 152         ipch->flags = 0;
 153         ipch->cpi = htons((u16 )ntohl(x->id.spi));
 154         *skb_mac_header(skb) = IPPROTO_COMP;
 155 out_ok:
 156         skb_push(skb, -skb_network_offset(skb));
 157         return 0;
 158 }
 159
 160 static void ipcomp4_err(struct sk_buff *skb, u32 info)
 161 {
 162         __be32 spi;
 163         struct iphdr *iph = (struct iphdr *)skb->data;
 164         struct ip_comp_hdr *ipch = (struct ip_comp_hdr *)(skb->data+(iph->ihl<<2));
 165         struct xfrm_state *x;
 166
 167         if (icmp_hdr(skb)->type != ICMP_DEST_UNREACH ||
 168             icmp_hdr(skb)->code != ICMP_FRAG_NEEDED)
 169                 return;
 170
 171         spi = htonl(ntohs(ipch->cpi));
 172         x = xfrm_state_lookup((xfrm_address_t *)&iph->daddr,
 173                               spi, IPPROTO_COMP, AF_INET);
 174         if (!x)
 175                 return;
 176         NETDEBUG(KERN_DEBUG "pmtu discovery on SA IPCOMP/%08x/%u.%u.%u.%u\n",
 177                  spi, NIPQUAD(iph->daddr));
 178         xfrm_state_put(x);
 179 }
 180
 181 /* We always hold one tunnel user reference to indicate a tunnel */
 182 static struct xfrm_state *ipcomp_tunnel_create(struct xfrm_state *x)
 183 {
 184         struct xfrm_state *t;
 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         t->props.mode = x->props.mode;
 196         t->props.saddr.a4 = x->props.saddr.a4;
 197         t->props.flags = x->props.flags;
 198
 199         if (xfrm_init_state(t))
 200                 goto error;
 201
 202         atomic_set(&t->tunnel_users, 1);
 203 out:
 204         return t;
 205
 206 error:
 207         t->km.state = XFRM_STATE_DEAD;
 208         xfrm_state_put(t);
 209         t = NULL;
 210         goto out;
 211 }
 212
 213 /*
 214  * Must be protected by xfrm_cfg_mutex.  State and tunnel user references are
 215  * always incremented on success.
 216  */
 217 static int ipcomp_tunnel_attach(struct xfrm_state *x)
 218 {
 219         int err = 0;
 220         struct xfrm_state *t;
 221
 222         t = xfrm_state_lookup((xfrm_address_t *)&x->id.daddr.a4,
 223                               x->props.saddr.a4, IPPROTO_IPIP, AF_INET);
 224         if (!t) {
 225                 t = ipcomp_tunnel_create(x);
 226                 if (!t) {
 227                         err = -EINVAL;
 228                         goto out;
 229                 }
 230                 xfrm_state_insert(t);
 231                 xfrm_state_hold(t);
 232         }
 233         x->tunnel = t;
 234         atomic_inc(&t->tunnel_users);
 235 out:
 236         return err;
 237 }
 238
 239 static void ipcomp_free_scratches(void)
 240 {
 241         int i;
 242         void **scratches;
 243
 244         if (--ipcomp_scratch_users)
 245                 return;
 246
 247         scratches = ipcomp_scratches;
 248         if (!scratches)
 249                 return;
 250
 251         for_each_possible_cpu(i)
 252                 vfree(*per_cpu_ptr(scratches, i));
 253
 254         free_percpu(scratches);
 255 }
 256
 257 static void **ipcomp_alloc_scratches(void)
 258 {
 259         int i;
 260         void **scratches;
 261
 262         if (ipcomp_scratch_users++)
 263                 return ipcomp_scratches;
 264
 265         scratches = alloc_percpu(void *);
 266         if (!scratches)
 267                 return NULL;
 268
 269         ipcomp_scratches = scratches;
 270
 271         for_each_possible_cpu(i) {
 272                 void *scratch = vmalloc(IPCOMP_SCRATCH_SIZE);
 273                 if (!scratch)
 274                         return NULL;
 275                 *per_cpu_ptr(scratches, i) = scratch;
 276         }
 277
 278         return scratches;
 279 }
 280
 281 static void ipcomp_free_tfms(struct crypto_comp **tfms)
 282 {
 283         struct ipcomp_tfms *pos;
 284         int cpu;
 285
 286         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
 287                 if (pos->tfms == tfms)
 288                         break;
 289         }
 290
 291         BUG_TRAP(pos);
 292
 293         if (--pos->users)
 294                 return;
 295
 296         list_del(&pos->list);
 297         kfree(pos);
 298
 299         if (!tfms)
 300                 return;
 301
 302         for_each_possible_cpu(cpu) {
 303                 struct crypto_comp *tfm = *per_cpu_ptr(tfms, cpu);
 304                 crypto_free_comp(tfm);
 305         }
 306         free_percpu(tfms);
 307 }
 308
 309 static struct crypto_comp **ipcomp_alloc_tfms(const char *alg_name)
 310 {
 311         struct ipcomp_tfms *pos;
 312         struct crypto_comp **tfms;
 313         int cpu;
 314
 315         /* This can be any valid CPU ID so we don't need locking. */
 316         cpu = raw_smp_processor_id();
 317
 318         list_for_each_entry(pos, &ipcomp_tfms_list, list) {
 319                 struct crypto_comp *tfm;
 320
 321                 tfms = pos->tfms;
 322                 tfm = *per_cpu_ptr(tfms, cpu);
 323
 324                 if (!strcmp(crypto_comp_name(tfm), alg_name)) {
 325                         pos->users++;
 326                         return tfms;
 327                 }
 328         }
 329
 330         pos = kmalloc(sizeof(*pos), GFP_KERNEL);
 331         if (!pos)
 332                 return NULL;
 333
 334         pos->users = 1;
 335         INIT_LIST_HEAD(&pos->list);
 336         list_add(&pos->list, &ipcomp_tfms_list);
 337
 338         pos->tfms = tfms = alloc_percpu(struct crypto_comp *);
 339         if (!tfms)
 340                 goto error;
 341
 342         for_each_possible_cpu(cpu) {
 343                 struct crypto_comp *tfm = crypto_alloc_comp(alg_name, 0,
 344                                                             CRYPTO_ALG_ASYNC);
 345                 if (IS_ERR(tfm))
 346                         goto error;
 347                 *per_cpu_ptr(tfms, cpu) = tfm;
 348         }
 349
 350         return tfms;
 351
 352 error:
 353         ipcomp_free_tfms(tfms);
 354         return NULL;
 355 }
 356
 357 static void ipcomp_free_data(struct ipcomp_data *ipcd)
 358 {
 359         if (ipcd->tfms)
 360                 ipcomp_free_tfms(ipcd->tfms);
 361         ipcomp_free_scratches();
 362 }
 363
 364 static void ipcomp_destroy(struct xfrm_state *x)
 365 {
 366         struct ipcomp_data *ipcd = x->data;
 367         if (!ipcd)
 368                 return;
 369         xfrm_state_delete_tunnel(x);
 370         mutex_lock(&ipcomp_resource_mutex);
 371         ipcomp_free_data(ipcd);
 372         mutex_unlock(&ipcomp_resource_mutex);
 373         kfree(ipcd);
 374 }
 375
 376 static int ipcomp_init_state(struct xfrm_state *x)
 377 {
 378         int err;
 379         struct ipcomp_data *ipcd;
 380         struct xfrm_algo_desc *calg_desc;
 381
 382         err = -EINVAL;
 383         if (!x->calg)
 384                 goto out;
 385
 386         if (x->encap)
 387                 goto out;
 388
 389         x->props.header_len = 0;
 390         switch (x->props.mode) {
 391         case XFRM_MODE_TRANSPORT:
 392                 break;
 393         case XFRM_MODE_TUNNEL:
 394                 x->props.header_len += sizeof(struct iphdr);
 395                 break;
 396         default:
 397                 goto out;
 398         }
 399
 400         err = -ENOMEM;
 401         ipcd = kzalloc(sizeof(*ipcd), GFP_KERNEL);
 402         if (!ipcd)
 403                 goto out;
 404
 405         mutex_lock(&ipcomp_resource_mutex);
 406         if (!ipcomp_alloc_scratches())
 407                 goto error;
 408
 409         ipcd->tfms = ipcomp_alloc_tfms(x->calg->alg_name);
 410         if (!ipcd->tfms)
 411                 goto error;
 412         mutex_unlock(&ipcomp_resource_mutex);
 413
 414         if (x->props.mode == XFRM_MODE_TUNNEL) {
 415                 err = ipcomp_tunnel_attach(x);
 416                 if (err)
 417                         goto error_tunnel;
 418         }
 419
 420         calg_desc = xfrm_calg_get_byname(x->calg->alg_name, 0);
 421         BUG_ON(!calg_desc);
 422         ipcd->threshold = calg_desc->uinfo.comp.threshold;
 423         x->data = ipcd;
 424         err = 0;
 425 out:
 426         return err;
 427
 428 error_tunnel:
 429         mutex_lock(&ipcomp_resource_mutex);
 430 error:
 431         ipcomp_free_data(ipcd);
 432         mutex_unlock(&ipcomp_resource_mutex);
 433         kfree(ipcd);
 434         goto out;
 435 }
 436
 437 static struct xfrm_type ipcomp_type = {
 438         .description    = "IPCOMP4",
 439         .owner          = THIS_MODULE,
 440         .proto          = IPPROTO_COMP,
 441         .init_state     = ipcomp_init_state,
 442         .destructor     = ipcomp_destroy,
 443         .input          = ipcomp_input,
 444         .output         = ipcomp_output
 445 };
 446
 447 static struct net_protocol ipcomp4_protocol = {
 448         .handler        =       xfrm4_rcv,
 449         .err_handler    =       ipcomp4_err,
 450         .no_policy      =       1,
 451 };
 452
 453 static int __init ipcomp4_init(void)
 454 {
 455         if (xfrm_register_type(&ipcomp_type, AF_INET) < 0) {
 456                 printk(KERN_INFO "ipcomp init: can't add xfrm type\n");
 457                 return -EAGAIN;
 458         }
 459         if (inet_add_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0) {
 460                 printk(KERN_INFO "ipcomp init: can't add protocol\n");
 461                 xfrm_unregister_type(&ipcomp_type, AF_INET);
 462                 return -EAGAIN;
 463         }
 464         return 0;
 465 }
 466
 467 static void __exit ipcomp4_fini(void)
 468 {
 469         if (inet_del_protocol(&ipcomp4_protocol, IPPROTO_COMP) < 0)
 470                 printk(KERN_INFO "ip ipcomp close: can't remove protocol\n");
 471         if (xfrm_unregister_type(&ipcomp_type, AF_INET) < 0)
 472                 printk(KERN_INFO "ip ipcomp close: can't remove xfrm type\n");
 473 }
 474
 475 module_init(ipcomp4_init);
 476 module_exit(ipcomp4_fini);
 477
 478 MODULE_LICENSE("GPL");
 479 MODULE_DESCRIPTION("IP Payload Compression Protocol (IPComp) - RFC3173");
 480 MODULE_AUTHOR("James Morris <jmorris@intercode.com.au>");
 481
 482 MODULE_ALIAS_XFRM_TYPE(AF_INET, XFRM_PROTO_COMP);