net/core/filter.c

   1 /*
   2  * Linux Socket Filter - Kernel level socket filtering
   3  *
   4  * Author:
   5  *     Jay Schulist <jschlst@samba.org>
   6  *
   7  * Based on the design of:
   8  *     - The Berkeley Packet Filter
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public License
  12  * as published by the Free Software Foundation; either version
  13  * 2 of the License, or (at your option) any later version.
  14  *
  15  * Andi Kleen - Fix a few bad bugs and races.
  16  * Kris Katterjohn - Added many additional checks in sk_chk_filter()
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/types.h>
  21 #include <linux/mm.h>
  22 #include <linux/fcntl.h>
  23 #include <linux/socket.h>
  24 #include <linux/in.h>
  25 #include <linux/inet.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/if_packet.h>
  28 #include <net/ip.h>
  29 #include <net/protocol.h>
  30 #include <net/netlink.h>
  31 #include <linux/skbuff.h>
  32 #include <net/sock.h>
  33 #include <linux/errno.h>
  34 #include <linux/timer.h>
  35 #include <asm/system.h>
  36 #include <asm/uaccess.h>
  37 #include <asm/unaligned.h>
  38 #include <linux/filter.h>
  39
  40 /* No hurry in this branch */
  41 static void *__load_pointer(struct sk_buff *skb, int k)
  42 {
  43         u8 *ptr = NULL;
  44
  45         if (k >= SKF_NET_OFF)
  46                 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
  47         else if (k >= SKF_LL_OFF)
  48                 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
  49
  50         if (ptr >= skb->head && ptr < skb_tail_pointer(skb))
  51                 return ptr;
  52         return NULL;
  53 }
  54
  55 static inline void *load_pointer(struct sk_buff *skb, int k,
  56                                  unsigned int size, void *buffer)
  57 {
  58         if (k >= 0)
  59                 return skb_header_pointer(skb, k, size, buffer);
  60         else {
  61                 if (k >= SKF_AD_OFF)
  62                         return NULL;
  63                 return __load_pointer(skb, k);
  64         }
  65 }
  66
  67 /**
  68  *      sk_filter - run a packet through a socket filter
  69  *      @sk: sock associated with &sk_buff
  70  *      @skb: buffer to filter
  71  *
  72  * Run the filter code and then cut skb->data to correct size returned by
  73  * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
  74  * than pkt_len we keep whole skb->data. This is the socket level
  75  * wrapper to sk_run_filter. It returns 0 if the packet should
  76  * be accepted or -EPERM if the packet should be tossed.
  77  *
  78  */
  79 int sk_filter(struct sock *sk, struct sk_buff *skb)
  80 {
  81         int err;
  82         struct sk_filter *filter;
  83
  84         err = security_sock_rcv_skb(sk, skb);
  85         if (err)
  86                 return err;
  87
  88         rcu_read_lock_bh();
  89         filter = rcu_dereference(sk->sk_filter);
  90         if (filter) {
  91                 unsigned int pkt_len = sk_run_filter(skb, filter->insns,
  92                                 filter->len);
  93                 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
  94         }
  95         rcu_read_unlock_bh();
  96
  97         return err;
  98 }
  99 EXPORT_SYMBOL(sk_filter);
 100
 101 /**
 102  *      sk_run_filter - run a filter on a socket
 103  *      @skb: buffer to run the filter on
 104  *      @filter: filter to apply
 105  *      @flen: length of filter
 106  *
 107  * Decode and apply filter instructions to the skb->data.
 108  * Return length to keep, 0 for none. skb is the data we are
 109  * filtering, filter is the array of filter instructions, and
 110  * len is the number of filter blocks in the array.
 111  */
 112 unsigned int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int flen)
 113 {
 114         struct sock_filter *fentry;     /* We walk down these */
 115         void *ptr;
 116         u32 A = 0;                      /* Accumulator */
 117         u32 X = 0;                      /* Index Register */
 118         u32 mem[BPF_MEMWORDS];          /* Scratch Memory Store */
 119         u32 tmp;
 120         int k;
 121         int pc;
 122
 123         /*
 124          * Process array of filter instructions.
 125          */
 126         for (pc = 0; pc < flen; pc++) {
 127                 fentry = &filter[pc];
 128
 129                 switch (fentry->code) {
 130                 case BPF_ALU|BPF_ADD|BPF_X:
 131                         A += X;
 132                         continue;
 133                 case BPF_ALU|BPF_ADD|BPF_K:
 134                         A += fentry->k;
 135                         continue;
 136                 case BPF_ALU|BPF_SUB|BPF_X:
 137                         A -= X;
 138                         continue;
 139                 case BPF_ALU|BPF_SUB|BPF_K:
 140                         A -= fentry->k;
 141                         continue;
 142                 case BPF_ALU|BPF_MUL|BPF_X:
 143                         A *= X;
 144                         continue;
 145                 case BPF_ALU|BPF_MUL|BPF_K:
 146                         A *= fentry->k;
 147                         continue;
 148                 case BPF_ALU|BPF_DIV|BPF_X:
 149                         if (X == 0)
 150                                 return 0;
 151                         A /= X;
 152                         continue;
 153                 case BPF_ALU|BPF_DIV|BPF_K:
 154                         A /= fentry->k;
 155                         continue;
 156                 case BPF_ALU|BPF_AND|BPF_X:
 157                         A &= X;
 158                         continue;
 159                 case BPF_ALU|BPF_AND|BPF_K:
 160                         A &= fentry->k;
 161                         continue;
 162                 case BPF_ALU|BPF_OR|BPF_X:
 163                         A |= X;
 164                         continue;
 165                 case BPF_ALU|BPF_OR|BPF_K:
 166                         A |= fentry->k;
 167                         continue;
 168                 case BPF_ALU|BPF_LSH|BPF_X:
 169                         A <<= X;
 170                         continue;
 171                 case BPF_ALU|BPF_LSH|BPF_K:
 172                         A <<= fentry->k;
 173                         continue;
 174                 case BPF_ALU|BPF_RSH|BPF_X:
 175                         A >>= X;
 176                         continue;
 177                 case BPF_ALU|BPF_RSH|BPF_K:
 178                         A >>= fentry->k;
 179                         continue;
 180                 case BPF_ALU|BPF_NEG:
 181                         A = -A;
 182                         continue;
 183                 case BPF_JMP|BPF_JA:
 184                         pc += fentry->k;
 185                         continue;
 186                 case BPF_JMP|BPF_JGT|BPF_K:
 187                         pc += (A > fentry->k) ? fentry->jt : fentry->jf;
 188                         continue;
 189                 case BPF_JMP|BPF_JGE|BPF_K:
 190                         pc += (A >= fentry->k) ? fentry->jt : fentry->jf;
 191                         continue;
 192                 case BPF_JMP|BPF_JEQ|BPF_K:
 193                         pc += (A == fentry->k) ? fentry->jt : fentry->jf;
 194                         continue;
 195                 case BPF_JMP|BPF_JSET|BPF_K:
 196                         pc += (A & fentry->k) ? fentry->jt : fentry->jf;
 197                         continue;
 198                 case BPF_JMP|BPF_JGT|BPF_X:
 199                         pc += (A > X) ? fentry->jt : fentry->jf;
 200                         continue;
 201                 case BPF_JMP|BPF_JGE|BPF_X:
 202                         pc += (A >= X) ? fentry->jt : fentry->jf;
 203                         continue;
 204                 case BPF_JMP|BPF_JEQ|BPF_X:
 205                         pc += (A == X) ? fentry->jt : fentry->jf;
 206                         continue;
 207                 case BPF_JMP|BPF_JSET|BPF_X:
 208                         pc += (A & X) ? fentry->jt : fentry->jf;
 209                         continue;
 210                 case BPF_LD|BPF_W|BPF_ABS:
 211                         k = fentry->k;
 212 load_w:
 213                         ptr = load_pointer(skb, k, 4, &tmp);
 214                         if (ptr != NULL) {
 215                                 A = get_unaligned_be32(ptr);
 216                                 continue;
 217                         }
 218                         break;
 219                 case BPF_LD|BPF_H|BPF_ABS:
 220                         k = fentry->k;
 221 load_h:
 222                         ptr = load_pointer(skb, k, 2, &tmp);
 223                         if (ptr != NULL) {
 224                                 A = get_unaligned_be16(ptr);
 225                                 continue;
 226                         }
 227                         break;
 228                 case BPF_LD|BPF_B|BPF_ABS:
 229                         k = fentry->k;
 230 load_b:
 231                         ptr = load_pointer(skb, k, 1, &tmp);
 232                         if (ptr != NULL) {
 233                                 A = *(u8 *)ptr;
 234                                 continue;
 235                         }
 236                         break;
 237                 case BPF_LD|BPF_W|BPF_LEN:
 238                         A = skb->len;
 239                         continue;
 240                 case BPF_LDX|BPF_W|BPF_LEN:
 241                         X = skb->len;
 242                         continue;
 243                 case BPF_LD|BPF_W|BPF_IND:
 244                         k = X + fentry->k;
 245                         goto load_w;
 246                 case BPF_LD|BPF_H|BPF_IND:
 247                         k = X + fentry->k;
 248                         goto load_h;
 249                 case BPF_LD|BPF_B|BPF_IND:
 250                         k = X + fentry->k;
 251                         goto load_b;
 252                 case BPF_LDX|BPF_B|BPF_MSH:
 253                         ptr = load_pointer(skb, fentry->k, 1, &tmp);
 254                         if (ptr != NULL) {
 255                                 X = (*(u8 *)ptr & 0xf) << 2;
 256                                 continue;
 257                         }
 258                         return 0;
 259                 case BPF_LD|BPF_IMM:
 260                         A = fentry->k;
 261                         continue;
 262                 case BPF_LDX|BPF_IMM:
 263                         X = fentry->k;
 264                         continue;
 265                 case BPF_LD|BPF_MEM:
 266                         A = mem[fentry->k];
 267                         continue;
 268                 case BPF_LDX|BPF_MEM:
 269                         X = mem[fentry->k];
 270                         continue;
 271                 case BPF_MISC|BPF_TAX:
 272                         X = A;
 273                         continue;
 274                 case BPF_MISC|BPF_TXA:
 275                         A = X;
 276                         continue;
 277                 case BPF_RET|BPF_K:
 278                         return fentry->k;
 279                 case BPF_RET|BPF_A:
 280                         return A;
 281                 case BPF_ST:
 282                         mem[fentry->k] = A;
 283                         continue;
 284                 case BPF_STX:
 285                         mem[fentry->k] = X;
 286                         continue;
 287                 default:
 288                         WARN_ON(1);
 289                         return 0;
 290                 }
 291
 292                 /*
 293                  * Handle ancillary data, which are impossible
 294                  * (or very difficult) to get parsing packet contents.
 295                  */
 296                 switch (k-SKF_AD_OFF) {
 297                 case SKF_AD_PROTOCOL:
 298                         A = ntohs(skb->protocol);
 299                         continue;
 300                 case SKF_AD_PKTTYPE:
 301                         A = skb->pkt_type;
 302                         continue;
 303                 case SKF_AD_IFINDEX:
 304                         A = skb->dev->ifindex;
 305                         continue;
 306                 case SKF_AD_NLATTR: {
 307                         struct nlattr *nla;
 308
 309                         if (skb_is_nonlinear(skb))
 310                                 return 0;
 311                         if (A > skb->len - sizeof(struct nlattr))
 312                                 return 0;
 313
 314                         nla = nla_find((struct nlattr *)&skb->data[A],
 315                                        skb->len - A, X);
 316                         if (nla)
 317                                 A = (void *)nla - (void *)skb->data;
 318                         else
 319                                 A = 0;
 320                         continue;
 321                 }
 322                 default:
 323                         return 0;
 324                 }
 325         }
 326
 327         return 0;
 328 }
 329 EXPORT_SYMBOL(sk_run_filter);
 330
 331 /**
 332  *      sk_chk_filter - verify socket filter code
 333  *      @filter: filter to verify
 334  *      @flen: length of filter
 335  *
 336  * Check the user's filter code. If we let some ugly
 337  * filter code slip through kaboom! The filter must contain
 338  * no references or jumps that are out of range, no illegal
 339  * instructions, and must end with a RET instruction.
 340  *
 341  * All jumps are forward as they are not signed.
 342  *
 343  * Returns 0 if the rule set is legal or -EINVAL if not.
 344  */
 345 int sk_chk_filter(struct sock_filter *filter, int flen)
 346 {
 347         struct sock_filter *ftest;
 348         int pc;
 349
 350         if (flen == 0 || flen > BPF_MAXINSNS)
 351                 return -EINVAL;
 352
 353         /* check the filter code now */
 354         for (pc = 0; pc < flen; pc++) {
 355                 ftest = &filter[pc];
 356
 357                 /* Only allow valid instructions */
 358                 switch (ftest->code) {
 359                 case BPF_ALU|BPF_ADD|BPF_K:
 360                 case BPF_ALU|BPF_ADD|BPF_X:
 361                 case BPF_ALU|BPF_SUB|BPF_K:
 362                 case BPF_ALU|BPF_SUB|BPF_X:
 363                 case BPF_ALU|BPF_MUL|BPF_K:
 364                 case BPF_ALU|BPF_MUL|BPF_X:
 365                 case BPF_ALU|BPF_DIV|BPF_X:
 366                 case BPF_ALU|BPF_AND|BPF_K:
 367                 case BPF_ALU|BPF_AND|BPF_X:
 368                 case BPF_ALU|BPF_OR|BPF_K:
 369                 case BPF_ALU|BPF_OR|BPF_X:
 370                 case BPF_ALU|BPF_LSH|BPF_K:
 371                 case BPF_ALU|BPF_LSH|BPF_X:
 372                 case BPF_ALU|BPF_RSH|BPF_K:
 373                 case BPF_ALU|BPF_RSH|BPF_X:
 374                 case BPF_ALU|BPF_NEG:
 375                 case BPF_LD|BPF_W|BPF_ABS:
 376                 case BPF_LD|BPF_H|BPF_ABS:
 377                 case BPF_LD|BPF_B|BPF_ABS:
 378                 case BPF_LD|BPF_W|BPF_LEN:
 379                 case BPF_LD|BPF_W|BPF_IND:
 380                 case BPF_LD|BPF_H|BPF_IND:
 381                 case BPF_LD|BPF_B|BPF_IND:
 382                 case BPF_LD|BPF_IMM:
 383                 case BPF_LDX|BPF_W|BPF_LEN:
 384                 case BPF_LDX|BPF_B|BPF_MSH:
 385                 case BPF_LDX|BPF_IMM:
 386                 case BPF_MISC|BPF_TAX:
 387                 case BPF_MISC|BPF_TXA:
 388                 case BPF_RET|BPF_K:
 389                 case BPF_RET|BPF_A:
 390                         break;
 391
 392                 /* Some instructions need special checks */
 393
 394                 case BPF_ALU|BPF_DIV|BPF_K:
 395                         /* check for division by zero */
 396                         if (ftest->k == 0)
 397                                 return -EINVAL;
 398                         break;
 399
 400                 case BPF_LD|BPF_MEM:
 401                 case BPF_LDX|BPF_MEM:
 402                 case BPF_ST:
 403                 case BPF_STX:
 404                         /* check for invalid memory addresses */
 405                         if (ftest->k >= BPF_MEMWORDS)
 406                                 return -EINVAL;
 407                         break;
 408
 409                 case BPF_JMP|BPF_JA:
 410                         /*
 411                          * Note, the large ftest->k might cause loops.
 412                          * Compare this with conditional jumps below,
 413                          * where offsets are limited. --ANK (981016)
 414                          */
 415                         if (ftest->k >= (unsigned)(flen-pc-1))
 416                                 return -EINVAL;
 417                         break;
 418
 419                 case BPF_JMP|BPF_JEQ|BPF_K:
 420                 case BPF_JMP|BPF_JEQ|BPF_X:
 421                 case BPF_JMP|BPF_JGE|BPF_K:
 422                 case BPF_JMP|BPF_JGE|BPF_X:
 423                 case BPF_JMP|BPF_JGT|BPF_K:
 424                 case BPF_JMP|BPF_JGT|BPF_X:
 425                 case BPF_JMP|BPF_JSET|BPF_K:
 426                 case BPF_JMP|BPF_JSET|BPF_X:
 427                         /* for conditionals both must be safe */
 428                         if (pc + ftest->jt + 1 >= flen ||
 429                             pc + ftest->jf + 1 >= flen)
 430                                 return -EINVAL;
 431                         break;
 432
 433                 default:
 434                         return -EINVAL;
 435                 }
 436         }
 437
 438         return (BPF_CLASS(filter[flen - 1].code) == BPF_RET) ? 0 : -EINVAL;
 439 }
 440 EXPORT_SYMBOL(sk_chk_filter);
 441
 442 /**
 443  *      sk_filter_rcu_release: Release a socket filter by rcu_head
 444  *      @rcu: rcu_head that contains the sk_filter to free
 445  */
 446 static void sk_filter_rcu_release(struct rcu_head *rcu)
 447 {
 448         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
 449
 450         sk_filter_release(fp);
 451 }
 452
 453 static void sk_filter_delayed_uncharge(struct sock *sk, struct sk_filter *fp)
 454 {
 455         unsigned int size = sk_filter_len(fp);
 456
 457         atomic_sub(size, &sk->sk_omem_alloc);
 458         call_rcu_bh(&fp->rcu, sk_filter_rcu_release);
 459 }
 460
 461 /**
 462  *      sk_attach_filter - attach a socket filter
 463  *      @fprog: the filter program
 464  *      @sk: the socket to use
 465  *
 466  * Attach the user's filter code. We first run some sanity checks on
 467  * it to make sure it does not explode on us later. If an error
 468  * occurs or there is insufficient memory for the filter a negative
 469  * errno code is returned. On success the return is zero.
 470  */
 471 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
 472 {
 473         struct sk_filter *fp, *old_fp;
 474         unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
 475         int err;
 476
 477         /* Make sure new filter is there and in the right amounts. */
 478         if (fprog->filter == NULL)
 479                 return -EINVAL;
 480
 481         fp = sock_kmalloc(sk, fsize+sizeof(*fp), GFP_KERNEL);
 482         if (!fp)
 483                 return -ENOMEM;
 484         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
 485                 sock_kfree_s(sk, fp, fsize+sizeof(*fp));
 486                 return -EFAULT;
 487         }
 488
 489         atomic_set(&fp->refcnt, 1);
 490         fp->len = fprog->len;
 491
 492         err = sk_chk_filter(fp->insns, fp->len);
 493         if (err) {
 494                 sk_filter_uncharge(sk, fp);
 495                 return err;
 496         }
 497
 498         rcu_read_lock_bh();
 499         old_fp = rcu_dereference(sk->sk_filter);
 500         rcu_assign_pointer(sk->sk_filter, fp);
 501         rcu_read_unlock_bh();
 502
 503         if (old_fp)
 504                 sk_filter_delayed_uncharge(sk, old_fp);
 505         return 0;
 506 }
 507
 508 int sk_detach_filter(struct sock *sk)
 509 {
 510         int ret = -ENOENT;
 511         struct sk_filter *filter;
 512
 513         rcu_read_lock_bh();
 514         filter = rcu_dereference(sk->sk_filter);
 515         if (filter) {
 516                 rcu_assign_pointer(sk->sk_filter, NULL);
 517                 sk_filter_delayed_uncharge(sk, filter);
 518                 ret = 0;
 519         }
 520         rcu_read_unlock_bh();
 521         return ret;
 522 }