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