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