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/uaccess.h>
  37 #include <asm/unaligned.h>
  38 #include <linux/filter.h>
  39 #include <linux/reciprocal_div.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/seccomp.h>
  42 #include <linux/if_vlan.h>
  43
  44 /* No hurry in this branch
  45  *
  46  * Exported for the bpf jit load helper.
  47  */
  48 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
  49 {
  50         u8 *ptr = NULL;
  51
  52         if (k >= SKF_NET_OFF)
  53                 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
  54         else if (k >= SKF_LL_OFF)
  55                 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
  56
  57         if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
  58                 return ptr;
  59         return NULL;
  60 }
  61
  62 static inline void *load_pointer(const struct sk_buff *skb, int k,
  63                                  unsigned int size, void *buffer)
  64 {
  65         if (k >= 0)
  66                 return skb_header_pointer(skb, k, size, buffer);
  67         return bpf_internal_load_pointer_neg_helper(skb, k, size);
  68 }
  69
  70 /**
  71  *      sk_filter - run a packet through a socket filter
  72  *      @sk: sock associated with &sk_buff
  73  *      @skb: buffer to filter
  74  *
  75  * Run the filter code and then cut skb->data to correct size returned by
  76  * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
  77  * than pkt_len we keep whole skb->data. This is the socket level
  78  * wrapper to sk_run_filter. It returns 0 if the packet should
  79  * be accepted or -EPERM if the packet should be tossed.
  80  *
  81  */
  82 int sk_filter(struct sock *sk, struct sk_buff *skb)
  83 {
  84         int err;
  85         struct sk_filter *filter;
  86
  87         /*
  88          * If the skb was allocated from pfmemalloc reserves, only
  89          * allow SOCK_MEMALLOC sockets to use it as this socket is
  90          * helping free memory
  91          */
  92         if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
  93                 return -ENOMEM;
  94
  95         err = security_sock_rcv_skb(sk, skb);
  96         if (err)
  97                 return err;
  98
  99         rcu_read_lock();
 100         filter = rcu_dereference(sk->sk_filter);
 101         if (filter) {
 102                 unsigned int pkt_len = SK_RUN_FILTER(filter, skb);
 103
 104                 err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
 105         }
 106         rcu_read_unlock();
 107
 108         return err;
 109 }
 110 EXPORT_SYMBOL(sk_filter);
 111
 112 /**
 113  *      sk_run_filter - run a filter on a socket
 114  *      @skb: buffer to run the filter on
 115  *      @fentry: filter to apply
 116  *
 117  * Decode and apply filter instructions to the skb->data.
 118  * Return length to keep, 0 for none. @skb is the data we are
 119  * filtering, @filter is the array of filter instructions.
 120  * Because all jumps are guaranteed to be before last instruction,
 121  * and last instruction guaranteed to be a RET, we dont need to check
 122  * flen. (We used to pass to this function the length of filter)
 123  */
 124 unsigned int sk_run_filter(const struct sk_buff *skb,
 125                            const struct sock_filter *fentry)
 126 {
 127         void *ptr;
 128         u32 A = 0;                      /* Accumulator */
 129         u32 X = 0;                      /* Index Register */
 130         u32 mem[BPF_MEMWORDS];          /* Scratch Memory Store */
 131         u32 tmp;
 132         int k;
 133
 134         /*
 135          * Process array of filter instructions.
 136          */
 137         for (;; fentry++) {
 138 #if defined(CONFIG_X86_32)
 139 #define K (fentry->k)
 140 #else
 141                 const u32 K = fentry->k;
 142 #endif
 143
 144                 switch (fentry->code) {
 145                 case BPF_S_ALU_ADD_X:
 146                         A += X;
 147                         continue;
 148                 case BPF_S_ALU_ADD_K:
 149                         A += K;
 150                         continue;
 151                 case BPF_S_ALU_SUB_X:
 152                         A -= X;
 153                         continue;
 154                 case BPF_S_ALU_SUB_K:
 155                         A -= K;
 156                         continue;
 157                 case BPF_S_ALU_MUL_X:
 158                         A *= X;
 159                         continue;
 160                 case BPF_S_ALU_MUL_K:
 161                         A *= K;
 162                         continue;
 163                 case BPF_S_ALU_DIV_X:
 164                         if (X == 0)
 165                                 return 0;
 166                         A /= X;
 167                         continue;
 168                 case BPF_S_ALU_DIV_K:
 169                         A = reciprocal_divide(A, K);
 170                         continue;
 171                 case BPF_S_ALU_MOD_X:
 172                         if (X == 0)
 173                                 return 0;
 174                         A %= X;
 175                         continue;
 176                 case BPF_S_ALU_MOD_K:
 177                         A %= K;
 178                         continue;
 179                 case BPF_S_ALU_AND_X:
 180                         A &= X;
 181                         continue;
 182                 case BPF_S_ALU_AND_K:
 183                         A &= K;
 184                         continue;
 185                 case BPF_S_ALU_OR_X:
 186                         A |= X;
 187                         continue;
 188                 case BPF_S_ALU_OR_K:
 189                         A |= K;
 190                         continue;
 191                 case BPF_S_ANC_ALU_XOR_X:
 192                 case BPF_S_ALU_XOR_X:
 193                         A ^= X;
 194                         continue;
 195                 case BPF_S_ALU_XOR_K:
 196                         A ^= K;
 197                         continue;
 198                 case BPF_S_ALU_LSH_X:
 199                         A <<= X;
 200                         continue;
 201                 case BPF_S_ALU_LSH_K:
 202                         A <<= K;
 203                         continue;
 204                 case BPF_S_ALU_RSH_X:
 205                         A >>= X;
 206                         continue;
 207                 case BPF_S_ALU_RSH_K:
 208                         A >>= K;
 209                         continue;
 210                 case BPF_S_ALU_NEG:
 211                         A = -A;
 212                         continue;
 213                 case BPF_S_JMP_JA:
 214                         fentry += K;
 215                         continue;
 216                 case BPF_S_JMP_JGT_K:
 217                         fentry += (A > K) ? fentry->jt : fentry->jf;
 218                         continue;
 219                 case BPF_S_JMP_JGE_K:
 220                         fentry += (A >= K) ? fentry->jt : fentry->jf;
 221                         continue;
 222                 case BPF_S_JMP_JEQ_K:
 223                         fentry += (A == K) ? fentry->jt : fentry->jf;
 224                         continue;
 225                 case BPF_S_JMP_JSET_K:
 226                         fentry += (A & K) ? fentry->jt : fentry->jf;
 227                         continue;
 228                 case BPF_S_JMP_JGT_X:
 229                         fentry += (A > X) ? fentry->jt : fentry->jf;
 230                         continue;
 231                 case BPF_S_JMP_JGE_X:
 232                         fentry += (A >= X) ? fentry->jt : fentry->jf;
 233                         continue;
 234                 case BPF_S_JMP_JEQ_X:
 235                         fentry += (A == X) ? fentry->jt : fentry->jf;
 236                         continue;
 237                 case BPF_S_JMP_JSET_X:
 238                         fentry += (A & X) ? fentry->jt : fentry->jf;
 239                         continue;
 240                 case BPF_S_LD_W_ABS:
 241                         k = K;
 242 load_w:
 243                         ptr = load_pointer(skb, k, 4, &tmp);
 244                         if (ptr != NULL) {
 245                                 A = get_unaligned_be32(ptr);
 246                                 continue;
 247                         }
 248                         return 0;
 249                 case BPF_S_LD_H_ABS:
 250                         k = K;
 251 load_h:
 252                         ptr = load_pointer(skb, k, 2, &tmp);
 253                         if (ptr != NULL) {
 254                                 A = get_unaligned_be16(ptr);
 255                                 continue;
 256                         }
 257                         return 0;
 258                 case BPF_S_LD_B_ABS:
 259                         k = K;
 260 load_b:
 261                         ptr = load_pointer(skb, k, 1, &tmp);
 262                         if (ptr != NULL) {
 263                                 A = *(u8 *)ptr;
 264                                 continue;
 265                         }
 266                         return 0;
 267                 case BPF_S_LD_W_LEN:
 268                         A = skb->len;
 269                         continue;
 270                 case BPF_S_LDX_W_LEN:
 271                         X = skb->len;
 272                         continue;
 273                 case BPF_S_LD_W_IND:
 274                         k = X + K;
 275                         goto load_w;
 276                 case BPF_S_LD_H_IND:
 277                         k = X + K;
 278                         goto load_h;
 279                 case BPF_S_LD_B_IND:
 280                         k = X + K;
 281                         goto load_b;
 282                 case BPF_S_LDX_B_MSH:
 283                         ptr = load_pointer(skb, K, 1, &tmp);
 284                         if (ptr != NULL) {
 285                                 X = (*(u8 *)ptr & 0xf) << 2;
 286                                 continue;
 287                         }
 288                         return 0;
 289                 case BPF_S_LD_IMM:
 290                         A = K;
 291                         continue;
 292                 case BPF_S_LDX_IMM:
 293                         X = K;
 294                         continue;
 295                 case BPF_S_LD_MEM:
 296                         A = mem[K];
 297                         continue;
 298                 case BPF_S_LDX_MEM:
 299                         X = mem[K];
 300                         continue;
 301                 case BPF_S_MISC_TAX:
 302                         X = A;
 303                         continue;
 304                 case BPF_S_MISC_TXA:
 305                         A = X;
 306                         continue;
 307                 case BPF_S_RET_K:
 308                         return K;
 309                 case BPF_S_RET_A:
 310                         return A;
 311                 case BPF_S_ST:
 312                         mem[K] = A;
 313                         continue;
 314                 case BPF_S_STX:
 315                         mem[K] = X;
 316                         continue;
 317                 case BPF_S_ANC_PROTOCOL:
 318                         A = ntohs(skb->protocol);
 319                         continue;
 320                 case BPF_S_ANC_PKTTYPE:
 321                         A = skb->pkt_type;
 322                         continue;
 323                 case BPF_S_ANC_IFINDEX:
 324                         if (!skb->dev)
 325                                 return 0;
 326                         A = skb->dev->ifindex;
 327                         continue;
 328                 case BPF_S_ANC_MARK:
 329                         A = skb->mark;
 330                         continue;
 331                 case BPF_S_ANC_QUEUE:
 332                         A = skb->queue_mapping;
 333                         continue;
 334                 case BPF_S_ANC_HATYPE:
 335                         if (!skb->dev)
 336                                 return 0;
 337                         A = skb->dev->type;
 338                         continue;
 339                 case BPF_S_ANC_RXHASH:
 340                         A = skb->rxhash;
 341                         continue;
 342                 case BPF_S_ANC_CPU:
 343                         A = raw_smp_processor_id();
 344                         continue;
 345                 case BPF_S_ANC_VLAN_TAG:
 346                         A = vlan_tx_tag_get(skb);
 347                         continue;
 348                 case BPF_S_ANC_VLAN_TAG_PRESENT:
 349                         A = !!vlan_tx_tag_present(skb);
 350                         continue;
 351                 case BPF_S_ANC_NLATTR: {
 352                         struct nlattr *nla;
 353
 354                         if (skb_is_nonlinear(skb))
 355                                 return 0;
 356                         if (A > skb->len - sizeof(struct nlattr))
 357                                 return 0;
 358
 359                         nla = nla_find((struct nlattr *)&skb->data[A],
 360                                        skb->len - A, X);
 361                         if (nla)
 362                                 A = (void *)nla - (void *)skb->data;
 363                         else
 364                                 A = 0;
 365                         continue;
 366                 }
 367                 case BPF_S_ANC_NLATTR_NEST: {
 368                         struct nlattr *nla;
 369
 370                         if (skb_is_nonlinear(skb))
 371                                 return 0;
 372                         if (A > skb->len - sizeof(struct nlattr))
 373                                 return 0;
 374
 375                         nla = (struct nlattr *)&skb->data[A];
 376                         if (nla->nla_len > A - skb->len)
 377                                 return 0;
 378
 379                         nla = nla_find_nested(nla, X);
 380                         if (nla)
 381                                 A = (void *)nla - (void *)skb->data;
 382                         else
 383                                 A = 0;
 384                         continue;
 385                 }
 386 #ifdef CONFIG_SECCOMP_FILTER
 387                 case BPF_S_ANC_SECCOMP_LD_W:
 388                         A = seccomp_bpf_load(fentry->k);
 389                         continue;
 390 #endif
 391                 default:
 392                         WARN_RATELIMIT(1, "Unknown code:%u jt:%u tf:%u k:%u\n",
 393                                        fentry->code, fentry->jt,
 394                                        fentry->jf, fentry->k);
 395                         return 0;
 396                 }
 397         }
 398
 399         return 0;
 400 }
 401 EXPORT_SYMBOL(sk_run_filter);
 402
 403 /*
 404  * Security :
 405  * A BPF program is able to use 16 cells of memory to store intermediate
 406  * values (check u32 mem[BPF_MEMWORDS] in sk_run_filter())
 407  * As we dont want to clear mem[] array for each packet going through
 408  * sk_run_filter(), we check that filter loaded by user never try to read
 409  * a cell if not previously written, and we check all branches to be sure
 410  * a malicious user doesn't try to abuse us.
 411  */
 412 static int check_load_and_stores(struct sock_filter *filter, int flen)
 413 {
 414         u16 *masks, memvalid = 0; /* one bit per cell, 16 cells */
 415         int pc, ret = 0;
 416
 417         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 418         masks = kmalloc(flen * sizeof(*masks), GFP_KERNEL);
 419         if (!masks)
 420                 return -ENOMEM;
 421         memset(masks, 0xff, flen * sizeof(*masks));
 422
 423         for (pc = 0; pc < flen; pc++) {
 424                 memvalid &= masks[pc];
 425
 426                 switch (filter[pc].code) {
 427                 case BPF_S_ST:
 428                 case BPF_S_STX:
 429                         memvalid |= (1 << filter[pc].k);
 430                         break;
 431                 case BPF_S_LD_MEM:
 432                 case BPF_S_LDX_MEM:
 433                         if (!(memvalid & (1 << filter[pc].k))) {
 434                                 ret = -EINVAL;
 435                                 goto error;
 436                         }
 437                         break;
 438                 case BPF_S_JMP_JA:
 439                         /* a jump must set masks on target */
 440                         masks[pc + 1 + filter[pc].k] &= memvalid;
 441                         memvalid = ~0;
 442                         break;
 443                 case BPF_S_JMP_JEQ_K:
 444                 case BPF_S_JMP_JEQ_X:
 445                 case BPF_S_JMP_JGE_K:
 446                 case BPF_S_JMP_JGE_X:
 447                 case BPF_S_JMP_JGT_K:
 448                 case BPF_S_JMP_JGT_X:
 449                 case BPF_S_JMP_JSET_X:
 450                 case BPF_S_JMP_JSET_K:
 451                         /* a jump must set masks on targets */
 452                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 453                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 454                         memvalid = ~0;
 455                         break;
 456                 }
 457         }
 458 error:
 459         kfree(masks);
 460         return ret;
 461 }
 462
 463 /**
 464  *      sk_chk_filter - verify socket filter code
 465  *      @filter: filter to verify
 466  *      @flen: length of filter
 467  *
 468  * Check the user's filter code. If we let some ugly
 469  * filter code slip through kaboom! The filter must contain
 470  * no references or jumps that are out of range, no illegal
 471  * instructions, and must end with a RET instruction.
 472  *
 473  * All jumps are forward as they are not signed.
 474  *
 475  * Returns 0 if the rule set is legal or -EINVAL if not.
 476  */
 477 int sk_chk_filter(struct sock_filter *filter, unsigned int flen)
 478 {
 479         /*
 480          * Valid instructions are initialized to non-0.
 481          * Invalid instructions are initialized to 0.
 482          */
 483         static const u8 codes[] = {
 484                 [BPF_ALU|BPF_ADD|BPF_K]  = BPF_S_ALU_ADD_K,
 485                 [BPF_ALU|BPF_ADD|BPF_X]  = BPF_S_ALU_ADD_X,
 486                 [BPF_ALU|BPF_SUB|BPF_K]  = BPF_S_ALU_SUB_K,
 487                 [BPF_ALU|BPF_SUB|BPF_X]  = BPF_S_ALU_SUB_X,
 488                 [BPF_ALU|BPF_MUL|BPF_K]  = BPF_S_ALU_MUL_K,
 489                 [BPF_ALU|BPF_MUL|BPF_X]  = BPF_S_ALU_MUL_X,
 490                 [BPF_ALU|BPF_DIV|BPF_X]  = BPF_S_ALU_DIV_X,
 491                 [BPF_ALU|BPF_MOD|BPF_K]  = BPF_S_ALU_MOD_K,
 492                 [BPF_ALU|BPF_MOD|BPF_X]  = BPF_S_ALU_MOD_X,
 493                 [BPF_ALU|BPF_AND|BPF_K]  = BPF_S_ALU_AND_K,
 494                 [BPF_ALU|BPF_AND|BPF_X]  = BPF_S_ALU_AND_X,
 495                 [BPF_ALU|BPF_OR|BPF_K]   = BPF_S_ALU_OR_K,
 496                 [BPF_ALU|BPF_OR|BPF_X]   = BPF_S_ALU_OR_X,
 497                 [BPF_ALU|BPF_XOR|BPF_K]  = BPF_S_ALU_XOR_K,
 498                 [BPF_ALU|BPF_XOR|BPF_X]  = BPF_S_ALU_XOR_X,
 499                 [BPF_ALU|BPF_LSH|BPF_K]  = BPF_S_ALU_LSH_K,
 500                 [BPF_ALU|BPF_LSH|BPF_X]  = BPF_S_ALU_LSH_X,
 501                 [BPF_ALU|BPF_RSH|BPF_K]  = BPF_S_ALU_RSH_K,
 502                 [BPF_ALU|BPF_RSH|BPF_X]  = BPF_S_ALU_RSH_X,
 503                 [BPF_ALU|BPF_NEG]        = BPF_S_ALU_NEG,
 504                 [BPF_LD|BPF_W|BPF_ABS]   = BPF_S_LD_W_ABS,
 505                 [BPF_LD|BPF_H|BPF_ABS]   = BPF_S_LD_H_ABS,
 506                 [BPF_LD|BPF_B|BPF_ABS]   = BPF_S_LD_B_ABS,
 507                 [BPF_LD|BPF_W|BPF_LEN]   = BPF_S_LD_W_LEN,
 508                 [BPF_LD|BPF_W|BPF_IND]   = BPF_S_LD_W_IND,
 509                 [BPF_LD|BPF_H|BPF_IND]   = BPF_S_LD_H_IND,
 510                 [BPF_LD|BPF_B|BPF_IND]   = BPF_S_LD_B_IND,
 511                 [BPF_LD|BPF_IMM]         = BPF_S_LD_IMM,
 512                 [BPF_LDX|BPF_W|BPF_LEN]  = BPF_S_LDX_W_LEN,
 513                 [BPF_LDX|BPF_B|BPF_MSH]  = BPF_S_LDX_B_MSH,
 514                 [BPF_LDX|BPF_IMM]        = BPF_S_LDX_IMM,
 515                 [BPF_MISC|BPF_TAX]       = BPF_S_MISC_TAX,
 516                 [BPF_MISC|BPF_TXA]       = BPF_S_MISC_TXA,
 517                 [BPF_RET|BPF_K]          = BPF_S_RET_K,
 518                 [BPF_RET|BPF_A]          = BPF_S_RET_A,
 519                 [BPF_ALU|BPF_DIV|BPF_K]  = BPF_S_ALU_DIV_K,
 520                 [BPF_LD|BPF_MEM]         = BPF_S_LD_MEM,
 521                 [BPF_LDX|BPF_MEM]        = BPF_S_LDX_MEM,
 522                 [BPF_ST]                 = BPF_S_ST,
 523                 [BPF_STX]                = BPF_S_STX,
 524                 [BPF_JMP|BPF_JA]         = BPF_S_JMP_JA,
 525                 [BPF_JMP|BPF_JEQ|BPF_K]  = BPF_S_JMP_JEQ_K,
 526                 [BPF_JMP|BPF_JEQ|BPF_X]  = BPF_S_JMP_JEQ_X,
 527                 [BPF_JMP|BPF_JGE|BPF_K]  = BPF_S_JMP_JGE_K,
 528                 [BPF_JMP|BPF_JGE|BPF_X]  = BPF_S_JMP_JGE_X,
 529                 [BPF_JMP|BPF_JGT|BPF_K]  = BPF_S_JMP_JGT_K,
 530                 [BPF_JMP|BPF_JGT|BPF_X]  = BPF_S_JMP_JGT_X,
 531                 [BPF_JMP|BPF_JSET|BPF_K] = BPF_S_JMP_JSET_K,
 532                 [BPF_JMP|BPF_JSET|BPF_X] = BPF_S_JMP_JSET_X,
 533         };
 534         int pc;
 535
 536         if (flen == 0 || flen > BPF_MAXINSNS)
 537                 return -EINVAL;
 538
 539         /* check the filter code now */
 540         for (pc = 0; pc < flen; pc++) {
 541                 struct sock_filter *ftest = &filter[pc];
 542                 u16 code = ftest->code;
 543
 544                 if (code >= ARRAY_SIZE(codes))
 545                         return -EINVAL;
 546                 code = codes[code];
 547                 if (!code)
 548                         return -EINVAL;
 549                 /* Some instructions need special checks */
 550                 switch (code) {
 551                 case BPF_S_ALU_DIV_K:
 552                         /* check for division by zero */
 553                         if (ftest->k == 0)
 554                                 return -EINVAL;
 555                         ftest->k = reciprocal_value(ftest->k);
 556                         break;
 557                 case BPF_S_ALU_MOD_K:
 558                         /* check for division by zero */
 559                         if (ftest->k == 0)
 560                                 return -EINVAL;
 561                         break;
 562                 case BPF_S_LD_MEM:
 563                 case BPF_S_LDX_MEM:
 564                 case BPF_S_ST:
 565                 case BPF_S_STX:
 566                         /* check for invalid memory addresses */
 567                         if (ftest->k >= BPF_MEMWORDS)
 568                                 return -EINVAL;
 569                         break;
 570                 case BPF_S_JMP_JA:
 571                         /*
 572                          * Note, the large ftest->k might cause loops.
 573                          * Compare this with conditional jumps below,
 574                          * where offsets are limited. --ANK (981016)
 575                          */
 576                         if (ftest->k >= (unsigned int)(flen-pc-1))
 577                                 return -EINVAL;
 578                         break;
 579                 case BPF_S_JMP_JEQ_K:
 580                 case BPF_S_JMP_JEQ_X:
 581                 case BPF_S_JMP_JGE_K:
 582                 case BPF_S_JMP_JGE_X:
 583                 case BPF_S_JMP_JGT_K:
 584                 case BPF_S_JMP_JGT_X:
 585                 case BPF_S_JMP_JSET_X:
 586                 case BPF_S_JMP_JSET_K:
 587                         /* for conditionals both must be safe */
 588                         if (pc + ftest->jt + 1 >= flen ||
 589                             pc + ftest->jf + 1 >= flen)
 590                                 return -EINVAL;
 591                         break;
 592                 case BPF_S_LD_W_ABS:
 593                 case BPF_S_LD_H_ABS:
 594                 case BPF_S_LD_B_ABS:
 595 #define ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE:        \
 596                                 code = BPF_S_ANC_##CODE;        \
 597                                 break
 598                         switch (ftest->k) {
 599                         ANCILLARY(PROTOCOL);
 600                         ANCILLARY(PKTTYPE);
 601                         ANCILLARY(IFINDEX);
 602                         ANCILLARY(NLATTR);
 603                         ANCILLARY(NLATTR_NEST);
 604                         ANCILLARY(MARK);
 605                         ANCILLARY(QUEUE);
 606                         ANCILLARY(HATYPE);
 607                         ANCILLARY(RXHASH);
 608                         ANCILLARY(CPU);
 609                         ANCILLARY(ALU_XOR_X);
 610                         ANCILLARY(VLAN_TAG);
 611                         ANCILLARY(VLAN_TAG_PRESENT);
 612                         }
 613                 }
 614                 ftest->code = code;
 615         }
 616
 617         /* last instruction must be a RET code */
 618         switch (filter[flen - 1].code) {
 619         case BPF_S_RET_K:
 620         case BPF_S_RET_A:
 621                 return check_load_and_stores(filter, flen);
 622         }
 623         return -EINVAL;
 624 }
 625 EXPORT_SYMBOL(sk_chk_filter);
 626
 627 /**
 628  *      sk_filter_release_rcu - Release a socket filter by rcu_head
 629  *      @rcu: rcu_head that contains the sk_filter to free
 630  */
 631 void sk_filter_release_rcu(struct rcu_head *rcu)
 632 {
 633         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
 634
 635         bpf_jit_free(fp);
 636         kfree(fp);
 637 }
 638 EXPORT_SYMBOL(sk_filter_release_rcu);
 639
 640 static int __sk_prepare_filter(struct sk_filter *fp)
 641 {
 642         int err;
 643
 644         fp->bpf_func = sk_run_filter;
 645
 646         err = sk_chk_filter(fp->insns, fp->len);
 647         if (err)
 648                 return err;
 649
 650         bpf_jit_compile(fp);
 651         return 0;
 652 }
 653
 654 /**
 655  *      sk_unattached_filter_create - create an unattached filter
 656  *      @fprog: the filter program
 657  *      @pfp: the unattached filter that is created
 658  *
 659  * Create a filter independent of any socket. We first run some
 660  * sanity checks on it to make sure it does not explode on us later.
 661  * If an error occurs or there is insufficient memory for the filter
 662  * a negative errno code is returned. On success the return is zero.
 663  */
 664 int sk_unattached_filter_create(struct sk_filter **pfp,
 665                                 struct sock_fprog *fprog)
 666 {
 667         struct sk_filter *fp;
 668         unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
 669         int err;
 670
 671         /* Make sure new filter is there and in the right amounts. */
 672         if (fprog->filter == NULL)
 673                 return -EINVAL;
 674
 675         fp = kmalloc(fsize + sizeof(*fp), GFP_KERNEL);
 676         if (!fp)
 677                 return -ENOMEM;
 678         memcpy(fp->insns, fprog->filter, fsize);
 679
 680         atomic_set(&fp->refcnt, 1);
 681         fp->len = fprog->len;
 682
 683         err = __sk_prepare_filter(fp);
 684         if (err)
 685                 goto free_mem;
 686
 687         *pfp = fp;
 688         return 0;
 689 free_mem:
 690         kfree(fp);
 691         return err;
 692 }
 693 EXPORT_SYMBOL_GPL(sk_unattached_filter_create);
 694
 695 void sk_unattached_filter_destroy(struct sk_filter *fp)
 696 {
 697         sk_filter_release(fp);
 698 }
 699 EXPORT_SYMBOL_GPL(sk_unattached_filter_destroy);
 700
 701 /**
 702  *      sk_attach_filter - attach a socket filter
 703  *      @fprog: the filter program
 704  *      @sk: the socket to use
 705  *
 706  * Attach the user's filter code. We first run some sanity checks on
 707  * it to make sure it does not explode on us later. If an error
 708  * occurs or there is insufficient memory for the filter a negative
 709  * errno code is returned. On success the return is zero.
 710  */
 711 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
 712 {
 713         struct sk_filter *fp, *old_fp;
 714         unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
 715         int err;
 716
 717         /* Make sure new filter is there and in the right amounts. */
 718         if (fprog->filter == NULL)
 719                 return -EINVAL;
 720
 721         fp = sock_kmalloc(sk, fsize+sizeof(*fp), GFP_KERNEL);
 722         if (!fp)
 723                 return -ENOMEM;
 724         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
 725                 sock_kfree_s(sk, fp, fsize+sizeof(*fp));
 726                 return -EFAULT;
 727         }
 728
 729         atomic_set(&fp->refcnt, 1);
 730         fp->len = fprog->len;
 731
 732         err = __sk_prepare_filter(fp);
 733         if (err) {
 734                 sk_filter_uncharge(sk, fp);
 735                 return err;
 736         }
 737
 738         old_fp = rcu_dereference_protected(sk->sk_filter,
 739                                            sock_owned_by_user(sk));
 740         rcu_assign_pointer(sk->sk_filter, fp);
 741
 742         if (old_fp)
 743                 sk_filter_uncharge(sk, old_fp);
 744         return 0;
 745 }
 746 EXPORT_SYMBOL_GPL(sk_attach_filter);
 747
 748 int sk_detach_filter(struct sock *sk)
 749 {
 750         int ret = -ENOENT;
 751         struct sk_filter *filter;
 752
 753         filter = rcu_dereference_protected(sk->sk_filter,
 754                                            sock_owned_by_user(sk));
 755         if (filter) {
 756                 RCU_INIT_POINTER(sk->sk_filter, NULL);
 757                 sk_filter_uncharge(sk, filter);
 758                 ret = 0;
 759         }
 760         return ret;
 761 }
 762 EXPORT_SYMBOL_GPL(sk_detach_filter);
 763
 764 static void sk_decode_filter(struct sock_filter *filt, struct sock_filter *to)
 765 {
 766         static const u16 decodes[] = {
 767                 [BPF_S_ALU_ADD_K]       = BPF_ALU|BPF_ADD|BPF_K,
 768                 [BPF_S_ALU_ADD_X]       = BPF_ALU|BPF_ADD|BPF_X,
 769                 [BPF_S_ALU_SUB_K]       = BPF_ALU|BPF_SUB|BPF_K,
 770                 [BPF_S_ALU_SUB_X]       = BPF_ALU|BPF_SUB|BPF_X,
 771                 [BPF_S_ALU_MUL_K]       = BPF_ALU|BPF_MUL|BPF_K,
 772                 [BPF_S_ALU_MUL_X]       = BPF_ALU|BPF_MUL|BPF_X,
 773                 [BPF_S_ALU_DIV_X]       = BPF_ALU|BPF_DIV|BPF_X,
 774                 [BPF_S_ALU_MOD_K]       = BPF_ALU|BPF_MOD|BPF_K,
 775                 [BPF_S_ALU_MOD_X]       = BPF_ALU|BPF_MOD|BPF_X,
 776                 [BPF_S_ALU_AND_K]       = BPF_ALU|BPF_AND|BPF_K,
 777                 [BPF_S_ALU_AND_X]       = BPF_ALU|BPF_AND|BPF_X,
 778                 [BPF_S_ALU_OR_K]        = BPF_ALU|BPF_OR|BPF_K,
 779                 [BPF_S_ALU_OR_X]        = BPF_ALU|BPF_OR|BPF_X,
 780                 [BPF_S_ALU_XOR_K]       = BPF_ALU|BPF_XOR|BPF_K,
 781                 [BPF_S_ALU_XOR_X]       = BPF_ALU|BPF_XOR|BPF_X,
 782                 [BPF_S_ALU_LSH_K]       = BPF_ALU|BPF_LSH|BPF_K,
 783                 [BPF_S_ALU_LSH_X]       = BPF_ALU|BPF_LSH|BPF_X,
 784                 [BPF_S_ALU_RSH_K]       = BPF_ALU|BPF_RSH|BPF_K,
 785                 [BPF_S_ALU_RSH_X]       = BPF_ALU|BPF_RSH|BPF_X,
 786                 [BPF_S_ALU_NEG]         = BPF_ALU|BPF_NEG,
 787                 [BPF_S_LD_W_ABS]        = BPF_LD|BPF_W|BPF_ABS,
 788                 [BPF_S_LD_H_ABS]        = BPF_LD|BPF_H|BPF_ABS,
 789                 [BPF_S_LD_B_ABS]        = BPF_LD|BPF_B|BPF_ABS,
 790                 [BPF_S_ANC_PROTOCOL]    = BPF_LD|BPF_B|BPF_ABS,
 791                 [BPF_S_ANC_PKTTYPE]     = BPF_LD|BPF_B|BPF_ABS,
 792                 [BPF_S_ANC_IFINDEX]     = BPF_LD|BPF_B|BPF_ABS,
 793                 [BPF_S_ANC_NLATTR]      = BPF_LD|BPF_B|BPF_ABS,
 794                 [BPF_S_ANC_NLATTR_NEST] = BPF_LD|BPF_B|BPF_ABS,
 795                 [BPF_S_ANC_MARK]        = BPF_LD|BPF_B|BPF_ABS,
 796                 [BPF_S_ANC_QUEUE]       = BPF_LD|BPF_B|BPF_ABS,
 797                 [BPF_S_ANC_HATYPE]      = BPF_LD|BPF_B|BPF_ABS,
 798                 [BPF_S_ANC_RXHASH]      = BPF_LD|BPF_B|BPF_ABS,
 799                 [BPF_S_ANC_CPU]         = BPF_LD|BPF_B|BPF_ABS,
 800                 [BPF_S_ANC_ALU_XOR_X]   = BPF_LD|BPF_B|BPF_ABS,
 801                 [BPF_S_ANC_SECCOMP_LD_W] = BPF_LD|BPF_B|BPF_ABS,
 802                 [BPF_S_ANC_VLAN_TAG]    = BPF_LD|BPF_B|BPF_ABS,
 803                 [BPF_S_ANC_VLAN_TAG_PRESENT] = BPF_LD|BPF_B|BPF_ABS,
 804                 [BPF_S_LD_W_LEN]        = BPF_LD|BPF_W|BPF_LEN,
 805                 [BPF_S_LD_W_IND]        = BPF_LD|BPF_W|BPF_IND,
 806                 [BPF_S_LD_H_IND]        = BPF_LD|BPF_H|BPF_IND,
 807                 [BPF_S_LD_B_IND]        = BPF_LD|BPF_B|BPF_IND,
 808                 [BPF_S_LD_IMM]          = BPF_LD|BPF_IMM,
 809                 [BPF_S_LDX_W_LEN]       = BPF_LDX|BPF_W|BPF_LEN,
 810                 [BPF_S_LDX_B_MSH]       = BPF_LDX|BPF_B|BPF_MSH,
 811                 [BPF_S_LDX_IMM]         = BPF_LDX|BPF_IMM,
 812                 [BPF_S_MISC_TAX]        = BPF_MISC|BPF_TAX,
 813                 [BPF_S_MISC_TXA]        = BPF_MISC|BPF_TXA,
 814                 [BPF_S_RET_K]           = BPF_RET|BPF_K,
 815                 [BPF_S_RET_A]           = BPF_RET|BPF_A,
 816                 [BPF_S_ALU_DIV_K]       = BPF_ALU|BPF_DIV|BPF_K,
 817                 [BPF_S_LD_MEM]          = BPF_LD|BPF_MEM,
 818                 [BPF_S_LDX_MEM]         = BPF_LDX|BPF_MEM,
 819                 [BPF_S_ST]              = BPF_ST,
 820                 [BPF_S_STX]             = BPF_STX,
 821                 [BPF_S_JMP_JA]          = BPF_JMP|BPF_JA,
 822                 [BPF_S_JMP_JEQ_K]       = BPF_JMP|BPF_JEQ|BPF_K,
 823                 [BPF_S_JMP_JEQ_X]       = BPF_JMP|BPF_JEQ|BPF_X,
 824                 [BPF_S_JMP_JGE_K]       = BPF_JMP|BPF_JGE|BPF_K,
 825                 [BPF_S_JMP_JGE_X]       = BPF_JMP|BPF_JGE|BPF_X,
 826                 [BPF_S_JMP_JGT_K]       = BPF_JMP|BPF_JGT|BPF_K,
 827                 [BPF_S_JMP_JGT_X]       = BPF_JMP|BPF_JGT|BPF_X,
 828                 [BPF_S_JMP_JSET_K]      = BPF_JMP|BPF_JSET|BPF_K,
 829                 [BPF_S_JMP_JSET_X]      = BPF_JMP|BPF_JSET|BPF_X,
 830         };
 831         u16 code;
 832
 833         code = filt->code;
 834
 835         to->code = decodes[code];
 836         to->jt = filt->jt;
 837         to->jf = filt->jf;
 838
 839         if (code == BPF_S_ALU_DIV_K) {
 840                 /*
 841                  * When loaded this rule user gave us X, which was
 842                  * translated into R = r(X). Now we calculate the
 843                  * RR = r(R) and report it back. If next time this
 844                  * value is loaded and RRR = r(RR) is calculated
 845                  * then the R == RRR will be true.
 846                  *
 847                  * One exception. X == 1 translates into R == 0 and
 848                  * we can't calculate RR out of it with r().
 849                  */
 850
 851                 if (filt->k == 0)
 852                         to->k = 1;
 853                 else
 854                         to->k = reciprocal_value(filt->k);
 855
 856                 BUG_ON(reciprocal_value(to->k) != filt->k);
 857         } else
 858                 to->k = filt->k;
 859 }
 860
 861 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, unsigned int len)
 862 {
 863         struct sk_filter *filter;
 864         int i, ret;
 865
 866         lock_sock(sk);
 867         filter = rcu_dereference_protected(sk->sk_filter,
 868                         sock_owned_by_user(sk));
 869         ret = 0;
 870         if (!filter)
 871                 goto out;
 872         ret = filter->len;
 873         if (!len)
 874                 goto out;
 875         ret = -EINVAL;
 876         if (len < filter->len)
 877                 goto out;
 878
 879         ret = -EFAULT;
 880         for (i = 0; i < filter->len; i++) {
 881                 struct sock_filter fb;
 882
 883                 sk_decode_filter(&filter->insns[i], &fb);
 884                 if (copy_to_user(&ubuf[i], &fb, sizeof(fb)))
 885                         goto out;
 886         }
 887
 888         ret = filter->len;
 889 out:
 890         release_sock(sk);
 891         return ret;
 892 }