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  */
  17
  18 #include <linux/module.h>
  19 #include <linux/types.h>
  20 #include <linux/sched.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 <linux/skbuff.h>
  31 #include <net/sock.h>
  32 #include <linux/errno.h>
  33 #include <linux/timer.h>
  34 #include <asm/system.h>
  35 #include <asm/uaccess.h>
  36 #include <linux/filter.h>
  37
  38 /* No hurry in this branch */
  39 static u8 *load_pointer(struct sk_buff *skb, int k)
  40 {
  41         u8 *ptr = NULL;
  42
  43         if (k >= SKF_NET_OFF)
  44                 ptr = skb->nh.raw + k - SKF_NET_OFF;
  45         else if (k >= SKF_LL_OFF)
  46                 ptr = skb->mac.raw + k - SKF_LL_OFF;
  47
  48         if (ptr >= skb->head && ptr < skb->tail)
  49                 return ptr;
  50         return NULL;
  51 }
  52
  53 /**
  54  *      sk_run_filter   -       run a filter on a socket
  55  *      @skb: buffer to run the filter on
  56  *      @filter: filter to apply
  57  *      @flen: length of filter
  58  *
  59  * Decode and apply filter instructions to the skb->data.
  60  * Return length to keep, 0 for none. skb is the data we are
  61  * filtering, filter is the array of filter instructions, and
  62  * len is the number of filter blocks in the array.
  63  */
  64
  65 int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int flen)
  66 {
  67         unsigned char *data = skb->data;
  68         /* len is UNSIGNED. Byte wide insns relies only on implicit
  69            type casts to prevent reading arbitrary memory locations.
  70          */
  71         unsigned int len = skb->len-skb->data_len;
  72         struct sock_filter *fentry;     /* We walk down these */
  73         u32 A = 0;                      /* Accumulator */
  74         u32 X = 0;                      /* Index Register */
  75         u32 mem[BPF_MEMWORDS];          /* Scratch Memory Store */
  76         int k;
  77         int pc;
  78
  79         /*
  80          * Process array of filter instructions.
  81          */
  82         for (pc = 0; pc < flen; pc++) {
  83                 fentry = &filter[pc];
  84
  85                 switch (fentry->code) {
  86                 case BPF_ALU|BPF_ADD|BPF_X:
  87                         A += X;
  88                         continue;
  89                 case BPF_ALU|BPF_ADD|BPF_K:
  90                         A += fentry->k;
  91                         continue;
  92                 case BPF_ALU|BPF_SUB|BPF_X:
  93                         A -= X;
  94                         continue;
  95                 case BPF_ALU|BPF_SUB|BPF_K:
  96                         A -= fentry->k;
  97                         continue;
  98                 case BPF_ALU|BPF_MUL|BPF_X:
  99                         A *= X;
 100                         continue;
 101                 case BPF_ALU|BPF_MUL|BPF_K:
 102                         A *= fentry->k;
 103                         continue;
 104                 case BPF_ALU|BPF_DIV|BPF_X:
 105                         if (X == 0)
 106                                 return 0;
 107                         A /= X;
 108                         continue;
 109                 case BPF_ALU|BPF_DIV|BPF_K:
 110                         if (fentry->k == 0)
 111                                 return 0;
 112                         A /= fentry->k;
 113                         continue;
 114                 case BPF_ALU|BPF_AND|BPF_X:
 115                         A &= X;
 116                         continue;
 117                 case BPF_ALU|BPF_AND|BPF_K:
 118                         A &= fentry->k;
 119                         continue;
 120                 case BPF_ALU|BPF_OR|BPF_X:
 121                         A |= X;
 122                         continue;
 123                 case BPF_ALU|BPF_OR|BPF_K:
 124                         A |= fentry->k;
 125                         continue;
 126                 case BPF_ALU|BPF_LSH|BPF_X:
 127                         A <<= X;
 128                         continue;
 129                 case BPF_ALU|BPF_LSH|BPF_K:
 130                         A <<= fentry->k;
 131                         continue;
 132                 case BPF_ALU|BPF_RSH|BPF_X:
 133                         A >>= X;
 134                         continue;
 135                 case BPF_ALU|BPF_RSH|BPF_K:
 136                         A >>= fentry->k;
 137                         continue;
 138                 case BPF_ALU|BPF_NEG:
 139                         A = -A;
 140                         continue;
 141                 case BPF_JMP|BPF_JA:
 142                         pc += fentry->k;
 143                         continue;
 144                 case BPF_JMP|BPF_JGT|BPF_K:
 145                         pc += (A > fentry->k) ? fentry->jt : fentry->jf;
 146                         continue;
 147                 case BPF_JMP|BPF_JGE|BPF_K:
 148                         pc += (A >= fentry->k) ? fentry->jt : fentry->jf;
 149                         continue;
 150                 case BPF_JMP|BPF_JEQ|BPF_K:
 151                         pc += (A == fentry->k) ? fentry->jt : fentry->jf;
 152                         continue;
 153                 case BPF_JMP|BPF_JSET|BPF_K:
 154                         pc += (A & fentry->k) ? fentry->jt : fentry->jf;
 155                         continue;
 156                 case BPF_JMP|BPF_JGT|BPF_X:
 157                         pc += (A > X) ? fentry->jt : fentry->jf;
 158                         continue;
 159                 case BPF_JMP|BPF_JGE|BPF_X:
 160                         pc += (A >= X) ? fentry->jt : fentry->jf;
 161                         continue;
 162                 case BPF_JMP|BPF_JEQ|BPF_X:
 163                         pc += (A == X) ? fentry->jt : fentry->jf;
 164                         continue;
 165                 case BPF_JMP|BPF_JSET|BPF_X:
 166                         pc += (A & X) ? fentry->jt : fentry->jf;
 167                         continue;
 168                 case BPF_LD|BPF_W|BPF_ABS:
 169                         k = fentry->k;
 170  load_w:
 171                         if (k >= 0 && (unsigned int)(k+sizeof(u32)) <= len) {
 172                                 A = ntohl(*(u32*)&data[k]);
 173                                 continue;
 174                         }
 175                         if (k < 0) {
 176                                 u8 *ptr;
 177
 178                                 if (k >= SKF_AD_OFF)
 179                                         break;
 180                                 ptr = load_pointer(skb, k);
 181                                 if (ptr) {
 182                                         A = ntohl(*(u32*)ptr);
 183                                         continue;
 184                                 }
 185                         } else {
 186                                 u32 tmp;
 187                                 if (!skb_copy_bits(skb, k, &tmp, 4)) {
 188                                         A = ntohl(tmp);
 189                                         continue;
 190                                 }
 191                         }
 192                         return 0;
 193                 case BPF_LD|BPF_H|BPF_ABS:
 194                         k = fentry->k;
 195  load_h:
 196                         if (k >= 0 && (unsigned int)(k + sizeof(u16)) <= len) {
 197                                 A = ntohs(*(u16*)&data[k]);
 198                                 continue;
 199                         }
 200                         if (k < 0) {
 201                                 u8 *ptr;
 202
 203                                 if (k >= SKF_AD_OFF)
 204                                         break;
 205                                 ptr = load_pointer(skb, k);
 206                                 if (ptr) {
 207                                         A = ntohs(*(u16*)ptr);
 208                                         continue;
 209                                 }
 210                         } else {
 211                                 u16 tmp;
 212                                 if (!skb_copy_bits(skb, k, &tmp, 2)) {
 213                                         A = ntohs(tmp);
 214                                         continue;
 215                                 }
 216                         }
 217                         return 0;
 218                 case BPF_LD|BPF_B|BPF_ABS:
 219                         k = fentry->k;
 220 load_b:
 221                         if (k >= 0 && (unsigned int)k < len) {
 222                                 A = data[k];
 223                                 continue;
 224                         }
 225                         if (k < 0) {
 226                                 u8 *ptr;
 227
 228                                 if (k >= SKF_AD_OFF)
 229                                         break;
 230                                 ptr = load_pointer(skb, k);
 231                                 if (ptr) {
 232                                         A = *ptr;
 233                                         continue;
 234                                 }
 235                         } else {
 236                                 u8 tmp;
 237                                 if (!skb_copy_bits(skb, k, &tmp, 1)) {
 238                                         A = tmp;
 239                                         continue;
 240                                 }
 241                         }
 242                         return 0;
 243                 case BPF_LD|BPF_W|BPF_LEN:
 244                         A = len;
 245                         continue;
 246                 case BPF_LDX|BPF_W|BPF_LEN:
 247                         X = len;
 248                         continue;
 249                 case BPF_LD|BPF_W|BPF_IND:
 250                         k = X + fentry->k;
 251                         goto load_w;
 252                 case BPF_LD|BPF_H|BPF_IND:
 253                         k = X + fentry->k;
 254                         goto load_h;
 255                 case BPF_LD|BPF_B|BPF_IND:
 256                         k = X + fentry->k;
 257                         goto load_b;
 258                 case BPF_LDX|BPF_B|BPF_MSH:
 259                         if (fentry->k >= len)
 260                                 return 0;
 261                         X = (data[fentry->k] & 0xf) << 2;
 262                         continue;
 263                 case BPF_LD|BPF_IMM:
 264                         A = fentry->k;
 265                         continue;
 266                 case BPF_LDX|BPF_IMM:
 267                         X = fentry->k;
 268                         continue;
 269                 case BPF_LD|BPF_MEM:
 270                         A = mem[fentry->k];
 271                         continue;
 272                 case BPF_LDX|BPF_MEM:
 273                         X = mem[fentry->k];
 274                         continue;
 275                 case BPF_MISC|BPF_TAX:
 276                         X = A;
 277                         continue;
 278                 case BPF_MISC|BPF_TXA:
 279                         A = X;
 280                         continue;
 281                 case BPF_RET|BPF_K:
 282                         return ((unsigned int)fentry->k);
 283                 case BPF_RET|BPF_A:
 284                         return ((unsigned int)A);
 285                 case BPF_ST:
 286                         mem[fentry->k] = A;
 287                         continue;
 288                 case BPF_STX:
 289                         mem[fentry->k] = X;
 290                         continue;
 291                 default:
 292                         /* Invalid instruction counts as RET */
 293                         return 0;
 294                 }
 295
 296                 /*
 297                  * Handle ancillary data, which are impossible
 298                  * (or very difficult) to get parsing packet contents.
 299                  */
 300                 switch (k-SKF_AD_OFF) {
 301                 case SKF_AD_PROTOCOL:
 302                         A = htons(skb->protocol);
 303                         continue;
 304                 case SKF_AD_PKTTYPE:
 305                         A = skb->pkt_type;
 306                         continue;
 307                 case SKF_AD_IFINDEX:
 308                         A = skb->dev->ifindex;
 309                         continue;
 310                 default:
 311                         return 0;
 312                 }
 313         }
 314
 315         return 0;
 316 }
 317
 318 /**
 319  *      sk_chk_filter - verify socket filter code
 320  *      @filter: filter to verify
 321  *      @flen: length of filter
 322  *
 323  * Check the user's filter code. If we let some ugly
 324  * filter code slip through kaboom! The filter must contain
 325  * no references or jumps that are out of range, no illegal instructions
 326  * and no backward jumps. It must end with a RET instruction
 327  *
 328  * Returns 0 if the rule set is legal or a negative errno code if not.
 329  */
 330 int sk_chk_filter(struct sock_filter *filter, int flen)
 331 {
 332         struct sock_filter *ftest;
 333         int pc;
 334
 335         if (((unsigned int)flen >= (~0U / sizeof(struct sock_filter))) || flen == 0)
 336                 return -EINVAL;
 337
 338         /* check the filter code now */
 339         for (pc = 0; pc < flen; pc++) {
 340                 /* all jumps are forward as they are not signed */
 341                 ftest = &filter[pc];
 342                 if (BPF_CLASS(ftest->code) == BPF_JMP) {
 343                         /* but they mustn't jump off the end */
 344                         if (BPF_OP(ftest->code) == BPF_JA) {
 345                                 /*
 346                                  * Note, the large ftest->k might cause loops.
 347                                  * Compare this with conditional jumps below,
 348                                  * where offsets are limited. --ANK (981016)
 349                                  */
 350                                 if (ftest->k >= (unsigned)(flen-pc-1))
 351                                         return -EINVAL;
 352                         } else {
 353                                 /* for conditionals both must be safe */
 354                                 if (pc + ftest->jt +1 >= flen ||
 355                                     pc + ftest->jf +1 >= flen)
 356                                         return -EINVAL;
 357                         }
 358                 }
 359
 360                 /* check that memory operations use valid addresses. */
 361                 if (ftest->k >= BPF_MEMWORDS) {
 362                         /* but it might not be a memory operation... */
 363                         switch (ftest->code) {
 364                         case BPF_ST:
 365                         case BPF_STX:
 366                         case BPF_LD|BPF_MEM:
 367                         case BPF_LDX|BPF_MEM:
 368                                 return -EINVAL;
 369                         }
 370                 }
 371         }
 372
 373         /*
 374          * The program must end with a return. We don't care where they
 375          * jumped within the script (its always forwards) but in the end
 376          * they _will_ hit this.
 377          */
 378         return (BPF_CLASS(filter[flen - 1].code) == BPF_RET) ? 0 : -EINVAL;
 379 }
 380
 381 /**
 382  *      sk_attach_filter - attach a socket filter
 383  *      @fprog: the filter program
 384  *      @sk: the socket to use
 385  *
 386  * Attach the user's filter code. We first run some sanity checks on
 387  * it to make sure it does not explode on us later. If an error
 388  * occurs or there is insufficient memory for the filter a negative
 389  * errno code is returned. On success the return is zero.
 390  */
 391 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
 392 {
 393         struct sk_filter *fp;
 394         unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
 395         int err;
 396
 397         /* Make sure new filter is there and in the right amounts. */
 398         if (fprog->filter == NULL || fprog->len > BPF_MAXINSNS)
 399                 return -EINVAL;
 400
 401         fp = sock_kmalloc(sk, fsize+sizeof(*fp), GFP_KERNEL);
 402         if (!fp)
 403                 return -ENOMEM;
 404         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
 405                 sock_kfree_s(sk, fp, fsize+sizeof(*fp));
 406                 return -EFAULT;
 407         }
 408
 409         atomic_set(&fp->refcnt, 1);
 410         fp->len = fprog->len;
 411
 412         err = sk_chk_filter(fp->insns, fp->len);
 413         if (!err) {
 414                 struct sk_filter *old_fp;
 415
 416                 spin_lock_bh(&sk->sk_lock.slock);
 417                 old_fp = sk->sk_filter;
 418                 sk->sk_filter = fp;
 419                 spin_unlock_bh(&sk->sk_lock.slock);
 420                 fp = old_fp;
 421         }
 422
 423         if (fp)
 424                 sk_filter_release(sk, fp);
 425         return err;
 426 }
 427
 428 EXPORT_SYMBOL(sk_chk_filter);
 429 EXPORT_SYMBOL(sk_run_filter);