LXR linux/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 <linux/gfp.h>
  29#include <net/ip.h>
  30#include <net/protocol.h>
  31#include <net/netlink.h>
  32#include <linux/skbuff.h>
  33#include <net/sock.h>
  34#include <linux/errno.h>
  35#include <linux/timer.h>
  36#include <asm/system.h>
  37#include <asm/uaccess.h>
  38#include <asm/unaligned.h>
  39#include <linux/filter.h>
  40
  41/* No hurry in this branch */
  42static void *__load_pointer(struct sk_buff *skb, int k)
  43{
  44        u8 *ptr = NULL;
  45
  46        if (k >= SKF_NET_OFF)
  47                ptr = skb_network_header(skb) + k - SKF_NET_OFF;
  48        else if (k >= SKF_LL_OFF)
  49                ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
  50
  51        if (ptr >= skb->head && ptr < skb_tail_pointer(skb))
  52                return ptr;
  53        return NULL;
  54}
  55
  56static inline void *load_pointer(struct sk_buff *skb, int k,
  57                                 unsigned int size, void *buffer)
  58{
  59        if (k >= 0)
  60                return skb_header_pointer(skb, k, size, buffer);
  61        else {
  62                if (k >= SKF_AD_OFF)
  63                        return NULL;
  64                return __load_pointer(skb, k);
  65        }
  66}
  67
  68/**
  69 *      sk_filter - run a packet through a socket filter
  70 *      @sk: sock associated with &sk_buff
  71 *      @skb: buffer to filter
  72 *
  73 * Run the filter code and then cut skb->data to correct size returned by
  74 * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
  75 * than pkt_len we keep whole skb->data. This is the socket level
  76 * wrapper to sk_run_filter. It returns 0 if the packet should
  77 * be accepted or -EPERM if the packet should be tossed.
  78 *
  79 */
  80int sk_filter(struct sock *sk, struct sk_buff *skb)
  81{
  82        int err;
  83        struct sk_filter *filter;
  84
  85        err = security_sock_rcv_skb(sk, skb);
  86        if (err)
  87                return err;
  88
  89        rcu_read_lock_bh();
  90        filter = rcu_dereference_bh(sk->sk_filter);
  91        if (filter) {
  92                unsigned int pkt_len = sk_run_filter(skb, filter->insns,
  93                                filter->len);
  94                err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
  95        }
  96        rcu_read_unlock_bh();
  97
  98        return err;
  99}
 100EXPORT_SYMBOL(sk_filter);
 101
 102/**
 103 *      sk_run_filter - run a filter on a socket
 104 *      @skb: buffer to run the filter on
 105 *      @filter: filter to apply
 106 *      @flen: length of filter
 107 *
 108 * Decode and apply filter instructions to the skb->data.
 109 * Return length to keep, 0 for none. skb is the data we are
 110 * filtering, filter is the array of filter instructions, and
 111 * len is the number of filter blocks in the array.
 112 */
 113unsigned int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int flen)
 114{
 115        struct sock_filter *fentry;     /* We walk down these */
 116        void *ptr;
 117        u32 A = 0;                      /* Accumulator */
 118        u32 X = 0;                      /* Index Register */
 119        u32 mem[BPF_MEMWORDS];          /* Scratch Memory Store */
 120        u32 tmp;
 121        int k;
 122        int pc;
 123
 124        /*
 125         * Process array of filter instructions.
 126         */
 127        for (pc = 0; pc < flen; pc++) {
 128                fentry = &filter[pc];
 129
 130                switch (fentry->code) {
 131                case BPF_ALU|BPF_ADD|BPF_X:
 132                        A += X;
 133                        continue;
 134                case BPF_ALU|BPF_ADD|BPF_K:
 135                        A += fentry->k;
 136                        continue;
 137                case BPF_ALU|BPF_SUB|BPF_X:
 138                        A -= X;
 139                        continue;
 140                case BPF_ALU|BPF_SUB|BPF_K:
 141                        A -= fentry->k;
 142                        continue;
 143                case BPF_ALU|BPF_MUL|BPF_X:
 144                        A *= X;
 145                        continue;
 146                case BPF_ALU|BPF_MUL|BPF_K:
 147                        A *= fentry->k;
 148                        continue;
 149                case BPF_ALU|BPF_DIV|BPF_X:
 150                        if (X == 0)
 151                                return 0;
 152                        A /= X;
 153                        continue;
 154                case BPF_ALU|BPF_DIV|BPF_K:
 155                        A /= fentry->k;
 156                        continue;
 157                case BPF_ALU|BPF_AND|BPF_X:
 158                        A &= X;
 159                        continue;
 160                case BPF_ALU|BPF_AND|BPF_K:
 161                        A &= fentry->k;
 162                        continue;
 163                case BPF_ALU|BPF_OR|BPF_X:
 164                        A |= X;
 165                        continue;
 166                case BPF_ALU|BPF_OR|BPF_K:
 167                        A |= fentry->k;
 168                        continue;
 169                case BPF_ALU|BPF_LSH|BPF_X:
 170                        A <<= X;
 171                        continue;
 172                case BPF_ALU|BPF_LSH|BPF_K:
 173                        A <<= fentry->k;
 174                        continue;
 175                case BPF_ALU|BPF_RSH|BPF_X:
 176                        A >>= X;
 177                        continue;
 178                case BPF_ALU|BPF_RSH|BPF_K:
 179                        A >>= fentry->k;
 180                        continue;
 181                case BPF_ALU|BPF_NEG:
 182                        A = -A;
 183                        continue;
 184                case BPF_JMP|BPF_JA:
 185                        pc += fentry->k;
 186                        continue;
 187                case BPF_JMP|BPF_JGT|BPF_K:
 188                        pc += (A > fentry->k) ? fentry->jt : fentry->jf;
 189                        continue;
 190                case BPF_JMP|BPF_JGE|BPF_K:
 191                        pc += (A >= fentry->k) ? fentry->jt : fentry->jf;
 192                        continue;
 193                case BPF_JMP|BPF_JEQ|BPF_K:
 194                        pc += (A == fentry->k) ? fentry->jt : fentry->jf;
 195                        continue;
 196                case BPF_JMP|BPF_JSET|BPF_K:
 197                        pc += (A & fentry->k) ? fentry->jt : fentry->jf;
 198                        continue;
 199                case BPF_JMP|BPF_JGT|BPF_X:
 200                        pc += (A > X) ? fentry->jt : fentry->jf;
 201                        continue;
 202                case BPF_JMP|BPF_JGE|BPF_X:
 203                        pc += (A >= X) ? fentry->jt : fentry->jf;
 204                        continue;
 205                case BPF_JMP|BPF_JEQ|BPF_X:
 206                        pc += (A == X) ? fentry->jt : fentry->jf;
 207                        continue;
 208                case BPF_JMP|BPF_JSET|BPF_X:
 209                        pc += (A & X) ? fentry->jt : fentry->jf;
 210                        continue;
 211                case BPF_LD|BPF_W|BPF_ABS:
 212                        k = fentry->k;
 213load_w:
 214                        ptr = load_pointer(skb, k, 4, &tmp);
 215                        if (ptr != NULL) {
 216                                A = get_unaligned_be32(ptr);
 217                                continue;
 218                        }
 219                        break;
 220                case BPF_LD|BPF_H|BPF_ABS:
 221                        k = fentry->k;
 222load_h:
 223                        ptr = load_pointer(skb, k, 2, &tmp);
 224                        if (ptr != NULL) {
 225                                A = get_unaligned_be16(ptr);
 226                                continue;
 227                        }
 228                        break;
 229                case BPF_LD|BPF_B|BPF_ABS:
 230                        k = fentry->k;
 231load_b:
 232                        ptr = load_pointer(skb, k, 1, &tmp);
 233                        if (ptr != NULL) {
 234                                A = *(u8 *)ptr;
 235                                continue;
 236                        }
 237                        break;
 238                case BPF_LD|BPF_W|BPF_LEN:
 239                        A = skb->len;
 240                        continue;
 241                case BPF_LDX|BPF_W|BPF_LEN:
 242                        X = skb->len;
 243                        continue;
 244                case BPF_LD|BPF_W|BPF_IND:
 245                        k = X + fentry->k;
 246                        goto load_w;
 247                case BPF_LD|BPF_H|BPF_IND:
 248                        k = X + fentry->k;
 249                        goto load_h;
 250                case BPF_LD|BPF_B|BPF_IND:
 251                        k = X + fentry->k;
 252                        goto load_b;
 253                case BPF_LDX|BPF_B|BPF_MSH:
 254                        ptr = load_pointer(skb, fentry->k, 1, &tmp);
 255                        if (ptr != NULL) {
 256                                X = (*(u8 *)ptr & 0xf) << 2;
 257                                continue;
 258                        }
 259                        return 0;
 260                case BPF_LD|BPF_IMM:
 261                        A = fentry->k;
 262                        continue;
 263                case BPF_LDX|BPF_IMM:
 264                        X = fentry->k;
 265                        continue;
 266                case BPF_LD|BPF_MEM:
 267                        A = mem[fentry->k];
 268                        continue;
 269                case BPF_LDX|BPF_MEM:
 270                        X = mem[fentry->k];
 271                        continue;
 272                case BPF_MISC|BPF_TAX:
 273                        X = A;
 274                        continue;
 275                case BPF_MISC|BPF_TXA:
 276                        A = X;
 277                        continue;
 278                case BPF_RET|BPF_K:
 279                        return fentry->k;
 280                case BPF_RET|BPF_A:
 281                        return A;
 282                case BPF_ST:
 283                        mem[fentry->k] = A;
 284                        continue;
 285                case BPF_STX:
 286                        mem[fentry->k] = X;
 287                        continue;
 288                default:
 289                        WARN_ON(1);
 290                        return 0;
 291                }
 292
 293                /*
 294                 * Handle ancillary data, which are impossible
 295                 * (or very difficult) to get parsing packet contents.
 296                 */
 297                switch (k-SKF_AD_OFF) {
 298                case SKF_AD_PROTOCOL:
 299                        A = ntohs(skb->protocol);
 300                        continue;
 301                case SKF_AD_PKTTYPE:
 302                        A = skb->pkt_type;
 303                        continue;
 304                case SKF_AD_IFINDEX:
 305                        A = skb->dev->ifindex;
 306                        continue;
 307                case SKF_AD_MARK:
 308                        A = skb->mark;
 309                        continue;
 310                case SKF_AD_QUEUE:
 311                        A = skb->queue_mapping;
 312                        continue;
 313                case SKF_AD_NLATTR: {
 314                        struct nlattr *nla;
 315
 316                        if (skb_is_nonlinear(skb))
 317                                return 0;
 318                        if (A > skb->len - sizeof(struct nlattr))
 319                                return 0;
 320
 321                        nla = nla_find((struct nlattr *)&skb->data[A],
 322                                       skb->len - A, X);
 323                        if (nla)
 324                                A = (void *)nla - (void *)skb->data;
 325                        else
 326                                A = 0;
 327                        continue;
 328                }
 329                case SKF_AD_NLATTR_NEST: {
 330                        struct nlattr *nla;
 331
 332                        if (skb_is_nonlinear(skb))
 333                                return 0;
 334                        if (A > skb->len - sizeof(struct nlattr))
 335                                return 0;
 336
 337                        nla = (struct nlattr *)&skb->data[A];
 338                        if (nla->nla_len > A - skb->len)
 339                                return 0;
 340
 341                        nla = nla_find_nested(nla, X);
 342                        if (nla)
 343                                A = (void *)nla - (void *)skb->data;
 344                        else
 345                                A = 0;
 346                        continue;
 347                }
 348                default:
 349                        return 0;
 350                }
 351        }
 352
 353        return 0;
 354}
 355EXPORT_SYMBOL(sk_run_filter);
 356
 357/**
 358 *      sk_chk_filter - verify socket filter code
 359 *      @filter: filter to verify
 360 *      @flen: length of filter
 361 *
 362 * Check the user's filter code. If we let some ugly
 363 * filter code slip through kaboom! The filter must contain
 364 * no references or jumps that are out of range, no illegal
 365 * instructions, and must end with a RET instruction.
 366 *
 367 * All jumps are forward as they are not signed.
 368 *
 369 * Returns 0 if the rule set is legal or -EINVAL if not.
 370 */
 371int sk_chk_filter(struct sock_filter *filter, int flen)
 372{
 373        struct sock_filter *ftest;
 374        int pc;
 375
 376        if (flen == 0 || flen > BPF_MAXINSNS)
 377                return -EINVAL;
 378
 379        /* check the filter code now */
 380        for (pc = 0; pc < flen; pc++) {
 381                ftest = &filter[pc];
 382
 383                /* Only allow valid instructions */
 384                switch (ftest->code) {
 385                case BPF_ALU|BPF_ADD|BPF_K:
 386                case BPF_ALU|BPF_ADD|BPF_X:
 387                case BPF_ALU|BPF_SUB|BPF_K:
 388                case BPF_ALU|BPF_SUB|BPF_X:
 389                case BPF_ALU|BPF_MUL|BPF_K:
 390                case BPF_ALU|BPF_MUL|BPF_X:
 391                case BPF_ALU|BPF_DIV|BPF_X:
 392                case BPF_ALU|BPF_AND|BPF_K:
 393                case BPF_ALU|BPF_AND|BPF_X:
 394                case BPF_ALU|BPF_OR|BPF_K:
 395                case BPF_ALU|BPF_OR|BPF_X:
 396                case BPF_ALU|BPF_LSH|BPF_K:
 397                case BPF_ALU|BPF_LSH|BPF_X:
 398                case BPF_ALU|BPF_RSH|BPF_K:
 399                case BPF_ALU|BPF_RSH|BPF_X:
 400                case BPF_ALU|BPF_NEG:
 401                case BPF_LD|BPF_W|BPF_ABS:
 402                case BPF_LD|BPF_H|BPF_ABS:
 403                case BPF_LD|BPF_B|BPF_ABS:
 404                case BPF_LD|BPF_W|BPF_LEN:
 405                case BPF_LD|BPF_W|BPF_IND:
 406                case BPF_LD|BPF_H|BPF_IND:
 407                case BPF_LD|BPF_B|BPF_IND:
 408                case BPF_LD|BPF_IMM:
 409                case BPF_LDX|BPF_W|BPF_LEN:
 410                case BPF_LDX|BPF_B|BPF_MSH:
 411                case BPF_LDX|BPF_IMM:
 412                case BPF_MISC|BPF_TAX:
 413                case BPF_MISC|BPF_TXA:
 414                case BPF_RET|BPF_K:
 415                case BPF_RET|BPF_A:
 416                        break;
 417
 418                /* Some instructions need special checks */
 419
 420                case BPF_ALU|BPF_DIV|BPF_K:
 421                        /* check for division by zero */
 422                        if (ftest->k == 0)
 423                                return -EINVAL;
 424                        break;
 425
 426                case BPF_LD|BPF_MEM:
 427                case BPF_LDX|BPF_MEM:
 428                case BPF_ST:
 429                case BPF_STX:
 430                        /* check for invalid memory addresses */
 431                        if (ftest->k >= BPF_MEMWORDS)
 432                                return -EINVAL;
 433                        break;
 434
 435                case BPF_JMP|BPF_JA:
 436                        /*
 437                         * Note, the large ftest->k might cause loops.
 438                         * Compare this with conditional jumps below,
 439                         * where offsets are limited. --ANK (981016)
 440                         */
 441                        if (ftest->k >= (unsigned)(flen-pc-1))
 442                                return -EINVAL;
 443                        break;
 444
 445                case BPF_JMP|BPF_JEQ|BPF_K:
 446                case BPF_JMP|BPF_JEQ|BPF_X:
 447                case BPF_JMP|BPF_JGE|BPF_K:
 448                case BPF_JMP|BPF_JGE|BPF_X:
 449                case BPF_JMP|BPF_JGT|BPF_K:
 450                case BPF_JMP|BPF_JGT|BPF_X:
 451                case BPF_JMP|BPF_JSET|BPF_K:
 452                case BPF_JMP|BPF_JSET|BPF_X:
 453                        /* for conditionals both must be safe */
 454                        if (pc + ftest->jt + 1 >= flen ||
 455                            pc + ftest->jf + 1 >= flen)
 456                                return -EINVAL;
 457                        break;
 458
 459                default:
 460                        return -EINVAL;
 461                }
 462        }
 463
 464        return (BPF_CLASS(filter[flen - 1].code) == BPF_RET) ? 0 : -EINVAL;
 465}
 466EXPORT_SYMBOL(sk_chk_filter);
 467
 468/**
 469 *      sk_filter_rcu_release: Release a socket filter by rcu_head
 470 *      @rcu: rcu_head that contains the sk_filter to free
 471 */
 472static void sk_filter_rcu_release(struct rcu_head *rcu)
 473{
 474        struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
 475
 476        sk_filter_release(fp);
 477}
 478
 479static void sk_filter_delayed_uncharge(struct sock *sk, struct sk_filter *fp)
 480{
 481        unsigned int size = sk_filter_len(fp);
 482
 483        atomic_sub(size, &sk->sk_omem_alloc);
 484        call_rcu_bh(&fp->rcu, sk_filter_rcu_release);
 485}
 486
 487/**
 488 *      sk_attach_filter - attach a socket filter
 489 *      @fprog: the filter program
 490 *      @sk: the socket to use
 491 *
 492 * Attach the user's filter code. We first run some sanity checks on
 493 * it to make sure it does not explode on us later. If an error
 494 * occurs or there is insufficient memory for the filter a negative
 495 * errno code is returned. On success the return is zero.
 496 */
 497int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
 498{
 499        struct sk_filter *fp, *old_fp;
 500        unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
 501        int err;
 502
 503        /* Make sure new filter is there and in the right amounts. */
 504        if (fprog->filter == NULL)
 505                return -EINVAL;
 506
 507        fp = sock_kmalloc(sk, fsize+sizeof(*fp), GFP_KERNEL);
 508        if (!fp)
 509                return -ENOMEM;
 510        if (copy_from_user(fp->insns, fprog->filter, fsize)) {
 511                sock_kfree_s(sk, fp, fsize+sizeof(*fp));
 512                return -EFAULT;
 513        }
 514
 515        atomic_set(&fp->refcnt, 1);
 516        fp->len = fprog->len;
 517
 518        err = sk_chk_filter(fp->insns, fp->len);
 519        if (err) {
 520                sk_filter_uncharge(sk, fp);
 521                return err;
 522        }
 523
 524        rcu_read_lock_bh();
 525        old_fp = rcu_dereference_bh(sk->sk_filter);
 526        rcu_assign_pointer(sk->sk_filter, fp);
 527        rcu_read_unlock_bh();
 528
 529        if (old_fp)
 530                sk_filter_delayed_uncharge(sk, old_fp);
 531        return 0;
 532}
 533EXPORT_SYMBOL_GPL(sk_attach_filter);
 534
 535int sk_detach_filter(struct sock *sk)
 536{
 537        int ret = -ENOENT;
 538        struct sk_filter *filter;
 539
 540        rcu_read_lock_bh();
 541        filter = rcu_dereference_bh(sk->sk_filter);
 542        if (filter) {
 543                rcu_assign_pointer(sk->sk_filter, NULL);
 544                sk_filter_delayed_uncharge(sk, filter);
 545                ret = 0;
 546        }
 547        rcu_read_unlock_bh();
 548        return ret;
 549}
 550EXPORT_SYMBOL_GPL(sk_detach_filter);
 551