net/core/filter.c

   1 /*
   2  * Linux Socket Filter - Kernel level socket filtering
   3  *
   4  * Author:
   5  *     Jay Schulist <jschlst@samba.org>
   6  *
   7  * Based on the design of:
   8  *     - The Berkeley Packet Filter
   9  *
  10  * This program is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU General Public License
  12  * as published by the Free Software Foundation; either version
  13  * 2 of the License, or (at your option) any later version.
  14  *
  15  * Andi Kleen - Fix a few bad bugs and races.
  16  * Kris Katterjohn - Added many additional checks in sk_chk_filter()
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/types.h>
  21 #include <linux/mm.h>
  22 #include <linux/fcntl.h>
  23 #include <linux/socket.h>
  24 #include <linux/in.h>
  25 #include <linux/inet.h>
  26 #include <linux/netdevice.h>
  27 #include <linux/if_packet.h>
  28 #include <linux/gfp.h>
  29 #include <net/ip.h>
  30 #include <net/protocol.h>
  31 #include <net/netlink.h>
  32 #include <linux/skbuff.h>
  33 #include <net/sock.h>
  34 #include <linux/errno.h>
  35 #include <linux/timer.h>
  36 #include <asm/system.h>
  37 #include <asm/uaccess.h>
  38 #include <asm/unaligned.h>
  39 #include <linux/filter.h>
  40 #include <linux/reciprocal_div.h>
  41 #include <linux/ratelimit.h>
  42
  43 /* No hurry in this branch */
  44 static void *__load_pointer(const struct sk_buff *skb, int k, unsigned int size)
  45 {
  46         u8 *ptr = NULL;
  47
  48         if (k >= SKF_NET_OFF)
  49                 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
  50         else if (k >= SKF_LL_OFF)
  51                 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
  52
  53         if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
  54                 return ptr;
  55         return NULL;
  56 }
  57
  58 static inline void *load_pointer(const struct sk_buff *skb, int k,
  59                                  unsigned int size, void *buffer)
  60 {
  61         if (k >= 0)
  62                 return skb_header_pointer(skb, k, size, buffer);
  63         return __load_pointer(skb, k, size);
  64 }
  65
  66 /**
  67  *      sk_filter_trim_cap - run a packet through a socket filter
  68  *      @sk: sock associated with &sk_buff
  69  *      @skb: buffer to filter
  70  *      @cap: limit on how short the eBPF program may trim the packet
  71  *
  72  * Run the filter code and then cut skb->data to correct size returned by
  73  * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
  74  * than pkt_len we keep whole skb->data. This is the socket level
  75  * wrapper to sk_run_filter. It returns 0 if the packet should
  76  * be accepted or -EPERM if the packet should be tossed.
  77  *
  78  */
  79 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap)
  80 {
  81         int err;
  82         struct sk_filter *filter;
  83
  84         err = security_sock_rcv_skb(sk, skb);
  85         if (err)
  86                 return err;
  87
  88         rcu_read_lock();
  89         filter = rcu_dereference(sk->sk_filter);
  90         if (filter) {
  91                 unsigned int pkt_len = SK_RUN_FILTER(filter, skb);
  92                 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM;
  93         }
  94         rcu_read_unlock();
  95
  96         return err;
  97 }
  98 EXPORT_SYMBOL(sk_filter_trim_cap);
  99
 100 /**
 101  *      sk_run_filter - run a filter on a socket
 102  *      @skb: buffer to run the filter on
 103  *      @fentry: filter to apply
 104  *
 105  * Decode and apply filter instructions to the skb->data.
 106  * Return length to keep, 0 for none. @skb is the data we are
 107  * filtering, @filter is the array of filter instructions.
 108  * Because all jumps are guaranteed to be before last instruction,
 109  * and last instruction guaranteed to be a RET, we dont need to check
 110  * flen. (We used to pass to this function the length of filter)
 111  */
 112 unsigned int sk_run_filter(const struct sk_buff *skb,
 113                            const struct sock_filter *fentry)
 114 {
 115         void *ptr;
 116         u32 A = 0;                      /* Accumulator */
 117         u32 X = 0;                      /* Index Register */
 118         u32 mem[BPF_MEMWORDS];          /* Scratch Memory Store */
 119         u32 tmp;
 120         int k;
 121
 122         /*
 123          * Process array of filter instructions.
 124          */
 125         for (;; fentry++) {
 126 #if defined(CONFIG_X86_32)
 127 #define K (fentry->k)
 128 #else
 129                 const u32 K = fentry->k;
 130 #endif
 131
 132                 switch (fentry->code) {
 133                 case BPF_S_ALU_ADD_X:
 134                         A += X;
 135                         continue;
 136                 case BPF_S_ALU_ADD_K:
 137                         A += K;
 138                         continue;
 139                 case BPF_S_ALU_SUB_X:
 140                         A -= X;
 141                         continue;
 142                 case BPF_S_ALU_SUB_K:
 143                         A -= K;
 144                         continue;
 145                 case BPF_S_ALU_MUL_X:
 146                         A *= X;
 147                         continue;
 148                 case BPF_S_ALU_MUL_K:
 149                         A *= K;
 150                         continue;
 151                 case BPF_S_ALU_DIV_X:
 152                         if (X == 0)
 153                                 return 0;
 154                         A /= X;
 155                         continue;
 156                 case BPF_S_ALU_DIV_K:
 157                         A = reciprocal_divide(A, K);
 158                         continue;
 159                 case BPF_S_ALU_AND_X:
 160                         A &= X;
 161                         continue;
 162                 case BPF_S_ALU_AND_K:
 163                         A &= K;
 164                         continue;
 165                 case BPF_S_ALU_OR_X:
 166                         A |= X;
 167                         continue;
 168                 case BPF_S_ALU_OR_K:
 169                         A |= K;
 170                         continue;
 171                 case BPF_S_ALU_LSH_X:
 172                         A <<= X;
 173                         continue;
 174                 case BPF_S_ALU_LSH_K:
 175                         A <<= K;
 176                         continue;
 177                 case BPF_S_ALU_RSH_X:
 178                         A >>= X;
 179                         continue;
 180                 case BPF_S_ALU_RSH_K:
 181                         A >>= K;
 182                         continue;
 183                 case BPF_S_ALU_NEG:
 184                         A = -A;
 185                         continue;
 186                 case BPF_S_JMP_JA:
 187                         fentry += K;
 188                         continue;
 189                 case BPF_S_JMP_JGT_K:
 190                         fentry += (A > K) ? fentry->jt : fentry->jf;
 191                         continue;
 192                 case BPF_S_JMP_JGE_K:
 193                         fentry += (A >= K) ? fentry->jt : fentry->jf;
 194                         continue;
 195                 case BPF_S_JMP_JEQ_K:
 196                         fentry += (A == K) ? fentry->jt : fentry->jf;
 197                         continue;
 198                 case BPF_S_JMP_JSET_K:
 199                         fentry += (A & K) ? fentry->jt : fentry->jf;
 200                         continue;
 201                 case BPF_S_JMP_JGT_X:
 202                         fentry += (A > X) ? fentry->jt : fentry->jf;
 203                         continue;
 204                 case BPF_S_JMP_JGE_X:
 205                         fentry += (A >= X) ? fentry->jt : fentry->jf;
 206                         continue;
 207                 case BPF_S_JMP_JEQ_X:
 208                         fentry += (A == X) ? fentry->jt : fentry->jf;
 209                         continue;
 210                 case BPF_S_JMP_JSET_X:
 211                         fentry += (A & X) ? fentry->jt : fentry->jf;
 212                         continue;
 213                 case BPF_S_LD_W_ABS:
 214                         k = K;
 215 load_w:
 216                         ptr = load_pointer(skb, k, 4, &tmp);
 217                         if (ptr != NULL) {
 218                                 A = get_unaligned_be32(ptr);
 219                                 continue;
 220                         }
 221                         return 0;
 222                 case BPF_S_LD_H_ABS:
 223                         k = K;
 224 load_h:
 225                         ptr = load_pointer(skb, k, 2, &tmp);
 226                         if (ptr != NULL) {
 227                                 A = get_unaligned_be16(ptr);
 228                                 continue;
 229                         }
 230                         return 0;
 231                 case BPF_S_LD_B_ABS:
 232                         k = K;
 233 load_b:
 234                         ptr = load_pointer(skb, k, 1, &tmp);
 235                         if (ptr != NULL) {
 236                                 A = *(u8 *)ptr;
 237                                 continue;
 238                         }
 239                         return 0;
 240                 case BPF_S_LD_W_LEN:
 241                         A = skb->len;
 242                         continue;
 243                 case BPF_S_LDX_W_LEN:
 244                         X = skb->len;
 245                         continue;
 246                 case BPF_S_LD_W_IND:
 247                         k = X + K;
 248                         goto load_w;
 249                 case BPF_S_LD_H_IND:
 250                         k = X + K;
 251                         goto load_h;
 252                 case BPF_S_LD_B_IND:
 253                         k = X + K;
 254                         goto load_b;
 255                 case BPF_S_LDX_B_MSH:
 256                         ptr = load_pointer(skb, K, 1, &tmp);
 257                         if (ptr != NULL) {
 258                                 X = (*(u8 *)ptr & 0xf) << 2;
 259                                 continue;
 260                         }
 261                         return 0;
 262                 case BPF_S_LD_IMM:
 263                         A = K;
 264                         continue;
 265                 case BPF_S_LDX_IMM:
 266                         X = K;
 267                         continue;
 268                 case BPF_S_LD_MEM:
 269                         A = mem[K];
 270                         continue;
 271                 case BPF_S_LDX_MEM:
 272                         X = mem[K];
 273                         continue;
 274                 case BPF_S_MISC_TAX:
 275                         X = A;
 276                         continue;
 277                 case BPF_S_MISC_TXA:
 278                         A = X;
 279                         continue;
 280                 case BPF_S_RET_K:
 281                         return K;
 282                 case BPF_S_RET_A:
 283                         return A;
 284                 case BPF_S_ST:
 285                         mem[K] = A;
 286                         continue;
 287                 case BPF_S_STX:
 288                         mem[K] = X;
 289                         continue;
 290                 case BPF_S_ANC_PROTOCOL:
 291                         A = ntohs(skb->protocol);
 292                         continue;
 293                 case BPF_S_ANC_PKTTYPE:
 294                         A = skb->pkt_type;
 295                         continue;
 296                 case BPF_S_ANC_IFINDEX:
 297                         if (!skb->dev)
 298                                 return 0;
 299                         A = skb->dev->ifindex;
 300                         continue;
 301                 case BPF_S_ANC_MARK:
 302                         A = skb->mark;
 303                         continue;
 304                 case BPF_S_ANC_QUEUE:
 305                         A = skb->queue_mapping;
 306                         continue;
 307                 case BPF_S_ANC_HATYPE:
 308                         if (!skb->dev)
 309                                 return 0;
 310                         A = skb->dev->type;
 311                         continue;
 312                 case BPF_S_ANC_RXHASH:
 313                         A = skb->rxhash;
 314                         continue;
 315                 case BPF_S_ANC_CPU:
 316                         A = raw_smp_processor_id();
 317                         continue;
 318                 case BPF_S_ANC_NLATTR: {
 319                         struct nlattr *nla;
 320
 321                         if (skb_is_nonlinear(skb))
 322                                 return 0;
 323                         if (skb->len < sizeof(struct nlattr))
 324                                 return 0;
 325                         if (A > skb->len - sizeof(struct nlattr))
 326                                 return 0;
 327
 328                         nla = nla_find((struct nlattr *)&skb->data[A],
 329                                        skb->len - A, X);
 330                         if (nla)
 331                                 A = (void *)nla - (void *)skb->data;
 332                         else
 333                                 A = 0;
 334                         continue;
 335                 }
 336                 case BPF_S_ANC_NLATTR_NEST: {
 337                         struct nlattr *nla;
 338
 339                         if (skb_is_nonlinear(skb))
 340                                 return 0;
 341                         if (skb->len < sizeof(struct nlattr))
 342                                 return 0;
 343                         if (A > skb->len - sizeof(struct nlattr))
 344                                 return 0;
 345
 346                         nla = (struct nlattr *)&skb->data[A];
 347                         if (nla->nla_len > skb->len - A)
 348                                 return 0;
 349
 350                         nla = nla_find_nested(nla, X);
 351                         if (nla)
 352                                 A = (void *)nla - (void *)skb->data;
 353                         else
 354                                 A = 0;
 355                         continue;
 356                 }
 357                 default:
 358                         WARN_RATELIMIT(1, "Unknown code:%u jt:%u tf:%u k:%u\n",
 359                                        fentry->code, fentry->jt,
 360                                        fentry->jf, fentry->k);
 361                         return 0;
 362                 }
 363         }
 364
 365         return 0;
 366 }
 367 EXPORT_SYMBOL(sk_run_filter);
 368
 369 /*
 370  * Security :
 371  * A BPF program is able to use 16 cells of memory to store intermediate
 372  * values (check u32 mem[BPF_MEMWORDS] in sk_run_filter())
 373  * As we dont want to clear mem[] array for each packet going through
 374  * sk_run_filter(), we check that filter loaded by user never try to read
 375  * a cell if not previously written, and we check all branches to be sure
 376  * a malicious user doesn't try to abuse us.
 377  */
 378 static int check_load_and_stores(struct sock_filter *filter, int flen)
 379 {
 380         u16 *masks, memvalid = 0; /* one bit per cell, 16 cells */
 381         int pc, ret = 0;
 382
 383         BUILD_BUG_ON(BPF_MEMWORDS > 16);
 384         masks = kmalloc(flen * sizeof(*masks), GFP_KERNEL);
 385         if (!masks)
 386                 return -ENOMEM;
 387         memset(masks, 0xff, flen * sizeof(*masks));
 388
 389         for (pc = 0; pc < flen; pc++) {
 390                 memvalid &= masks[pc];
 391
 392                 switch (filter[pc].code) {
 393                 case BPF_S_ST:
 394                 case BPF_S_STX:
 395                         memvalid |= (1 << filter[pc].k);
 396                         break;
 397                 case BPF_S_LD_MEM:
 398                 case BPF_S_LDX_MEM:
 399                         if (!(memvalid & (1 << filter[pc].k))) {
 400                                 ret = -EINVAL;
 401                                 goto error;
 402                         }
 403                         break;
 404                 case BPF_S_JMP_JA:
 405                         /* a jump must set masks on target */
 406                         masks[pc + 1 + filter[pc].k] &= memvalid;
 407                         memvalid = ~0;
 408                         break;
 409                 case BPF_S_JMP_JEQ_K:
 410                 case BPF_S_JMP_JEQ_X:
 411                 case BPF_S_JMP_JGE_K:
 412                 case BPF_S_JMP_JGE_X:
 413                 case BPF_S_JMP_JGT_K:
 414                 case BPF_S_JMP_JGT_X:
 415                 case BPF_S_JMP_JSET_X:
 416                 case BPF_S_JMP_JSET_K:
 417                         /* a jump must set masks on targets */
 418                         masks[pc + 1 + filter[pc].jt] &= memvalid;
 419                         masks[pc + 1 + filter[pc].jf] &= memvalid;
 420                         memvalid = ~0;
 421                         break;
 422                 }
 423         }
 424 error:
 425         kfree(masks);
 426         return ret;
 427 }
 428
 429 /**
 430  *      sk_chk_filter - verify socket filter code
 431  *      @filter: filter to verify
 432  *      @flen: length of filter
 433  *
 434  * Check the user's filter code. If we let some ugly
 435  * filter code slip through kaboom! The filter must contain
 436  * no references or jumps that are out of range, no illegal
 437  * instructions, and must end with a RET instruction.
 438  *
 439  * All jumps are forward as they are not signed.
 440  *
 441  * Returns 0 if the rule set is legal or -EINVAL if not.
 442  */
 443 int sk_chk_filter(struct sock_filter *filter, unsigned int flen)
 444 {
 445         /*
 446          * Valid instructions are initialized to non-0.
 447          * Invalid instructions are initialized to 0.
 448          */
 449         static const u8 codes[] = {
 450                 [BPF_ALU|BPF_ADD|BPF_K]  = BPF_S_ALU_ADD_K,
 451                 [BPF_ALU|BPF_ADD|BPF_X]  = BPF_S_ALU_ADD_X,
 452                 [BPF_ALU|BPF_SUB|BPF_K]  = BPF_S_ALU_SUB_K,
 453                 [BPF_ALU|BPF_SUB|BPF_X]  = BPF_S_ALU_SUB_X,
 454                 [BPF_ALU|BPF_MUL|BPF_K]  = BPF_S_ALU_MUL_K,
 455                 [BPF_ALU|BPF_MUL|BPF_X]  = BPF_S_ALU_MUL_X,
 456                 [BPF_ALU|BPF_DIV|BPF_X]  = BPF_S_ALU_DIV_X,
 457                 [BPF_ALU|BPF_AND|BPF_K]  = BPF_S_ALU_AND_K,
 458                 [BPF_ALU|BPF_AND|BPF_X]  = BPF_S_ALU_AND_X,
 459                 [BPF_ALU|BPF_OR|BPF_K]   = BPF_S_ALU_OR_K,
 460                 [BPF_ALU|BPF_OR|BPF_X]   = BPF_S_ALU_OR_X,
 461                 [BPF_ALU|BPF_LSH|BPF_K]  = BPF_S_ALU_LSH_K,
 462                 [BPF_ALU|BPF_LSH|BPF_X]  = BPF_S_ALU_LSH_X,
 463                 [BPF_ALU|BPF_RSH|BPF_K]  = BPF_S_ALU_RSH_K,
 464                 [BPF_ALU|BPF_RSH|BPF_X]  = BPF_S_ALU_RSH_X,
 465                 [BPF_ALU|BPF_NEG]        = BPF_S_ALU_NEG,
 466                 [BPF_LD|BPF_W|BPF_ABS]   = BPF_S_LD_W_ABS,
 467                 [BPF_LD|BPF_H|BPF_ABS]   = BPF_S_LD_H_ABS,
 468                 [BPF_LD|BPF_B|BPF_ABS]   = BPF_S_LD_B_ABS,
 469                 [BPF_LD|BPF_W|BPF_LEN]   = BPF_S_LD_W_LEN,
 470                 [BPF_LD|BPF_W|BPF_IND]   = BPF_S_LD_W_IND,
 471                 [BPF_LD|BPF_H|BPF_IND]   = BPF_S_LD_H_IND,
 472                 [BPF_LD|BPF_B|BPF_IND]   = BPF_S_LD_B_IND,
 473                 [BPF_LD|BPF_IMM]         = BPF_S_LD_IMM,
 474                 [BPF_LDX|BPF_W|BPF_LEN]  = BPF_S_LDX_W_LEN,
 475                 [BPF_LDX|BPF_B|BPF_MSH]  = BPF_S_LDX_B_MSH,
 476                 [BPF_LDX|BPF_IMM]        = BPF_S_LDX_IMM,
 477                 [BPF_MISC|BPF_TAX]       = BPF_S_MISC_TAX,
 478                 [BPF_MISC|BPF_TXA]       = BPF_S_MISC_TXA,
 479                 [BPF_RET|BPF_K]          = BPF_S_RET_K,
 480                 [BPF_RET|BPF_A]          = BPF_S_RET_A,
 481                 [BPF_ALU|BPF_DIV|BPF_K]  = BPF_S_ALU_DIV_K,
 482                 [BPF_LD|BPF_MEM]         = BPF_S_LD_MEM,
 483                 [BPF_LDX|BPF_MEM]        = BPF_S_LDX_MEM,
 484                 [BPF_ST]                 = BPF_S_ST,
 485                 [BPF_STX]                = BPF_S_STX,
 486                 [BPF_JMP|BPF_JA]         = BPF_S_JMP_JA,
 487                 [BPF_JMP|BPF_JEQ|BPF_K]  = BPF_S_JMP_JEQ_K,
 488                 [BPF_JMP|BPF_JEQ|BPF_X]  = BPF_S_JMP_JEQ_X,
 489                 [BPF_JMP|BPF_JGE|BPF_K]  = BPF_S_JMP_JGE_K,
 490                 [BPF_JMP|BPF_JGE|BPF_X]  = BPF_S_JMP_JGE_X,
 491                 [BPF_JMP|BPF_JGT|BPF_K]  = BPF_S_JMP_JGT_K,
 492                 [BPF_JMP|BPF_JGT|BPF_X]  = BPF_S_JMP_JGT_X,
 493                 [BPF_JMP|BPF_JSET|BPF_K] = BPF_S_JMP_JSET_K,
 494                 [BPF_JMP|BPF_JSET|BPF_X] = BPF_S_JMP_JSET_X,
 495         };
 496         int pc;
 497
 498         if (flen == 0 || flen > BPF_MAXINSNS)
 499                 return -EINVAL;
 500
 501         /* check the filter code now */
 502         for (pc = 0; pc < flen; pc++) {
 503                 struct sock_filter *ftest = &filter[pc];
 504                 u16 code = ftest->code;
 505
 506                 if (code >= ARRAY_SIZE(codes))
 507                         return -EINVAL;
 508                 code = codes[code];
 509                 if (!code)
 510                         return -EINVAL;
 511                 /* Some instructions need special checks */
 512                 switch (code) {
 513                 case BPF_S_ALU_DIV_K:
 514                         /* check for division by zero */
 515                         if (ftest->k == 0)
 516                                 return -EINVAL;
 517                         ftest->k = reciprocal_value(ftest->k);
 518                         break;
 519                 case BPF_S_LD_MEM:
 520                 case BPF_S_LDX_MEM:
 521                 case BPF_S_ST:
 522                 case BPF_S_STX:
 523                         /* check for invalid memory addresses */
 524                         if (ftest->k >= BPF_MEMWORDS)
 525                                 return -EINVAL;
 526                         break;
 527                 case BPF_S_JMP_JA:
 528                         /*
 529                          * Note, the large ftest->k might cause loops.
 530                          * Compare this with conditional jumps below,
 531                          * where offsets are limited. --ANK (981016)
 532                          */
 533                         if (ftest->k >= (unsigned)(flen-pc-1))
 534                                 return -EINVAL;
 535                         break;
 536                 case BPF_S_JMP_JEQ_K:
 537                 case BPF_S_JMP_JEQ_X:
 538                 case BPF_S_JMP_JGE_K:
 539                 case BPF_S_JMP_JGE_X:
 540                 case BPF_S_JMP_JGT_K:
 541                 case BPF_S_JMP_JGT_X:
 542                 case BPF_S_JMP_JSET_X:
 543                 case BPF_S_JMP_JSET_K:
 544                         /* for conditionals both must be safe */
 545                         if (pc + ftest->jt + 1 >= flen ||
 546                             pc + ftest->jf + 1 >= flen)
 547                                 return -EINVAL;
 548                         break;
 549                 case BPF_S_LD_W_ABS:
 550                 case BPF_S_LD_H_ABS:
 551                 case BPF_S_LD_B_ABS:
 552 #define ANCILLARY(CODE) case SKF_AD_OFF + SKF_AD_##CODE:        \
 553                                 code = BPF_S_ANC_##CODE;        \
 554                                 break
 555                         switch (ftest->k) {
 556                         ANCILLARY(PROTOCOL);
 557                         ANCILLARY(PKTTYPE);
 558                         ANCILLARY(IFINDEX);
 559                         ANCILLARY(NLATTR);
 560                         ANCILLARY(NLATTR_NEST);
 561                         ANCILLARY(MARK);
 562                         ANCILLARY(QUEUE);
 563                         ANCILLARY(HATYPE);
 564                         ANCILLARY(RXHASH);
 565                         ANCILLARY(CPU);
 566                         }
 567                 }
 568                 ftest->code = code;
 569         }
 570
 571         /* last instruction must be a RET code */
 572         switch (filter[flen - 1].code) {
 573         case BPF_S_RET_K:
 574         case BPF_S_RET_A:
 575                 return check_load_and_stores(filter, flen);
 576         }
 577         return -EINVAL;
 578 }
 579 EXPORT_SYMBOL(sk_chk_filter);
 580
 581 /**
 582  *      sk_filter_release_rcu - Release a socket filter by rcu_head
 583  *      @rcu: rcu_head that contains the sk_filter to free
 584  */
 585 void sk_filter_release_rcu(struct rcu_head *rcu)
 586 {
 587         struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu);
 588
 589         bpf_jit_free(fp);
 590         kfree(fp);
 591 }
 592 EXPORT_SYMBOL(sk_filter_release_rcu);
 593
 594 /**
 595  *      sk_attach_filter - attach a socket filter
 596  *      @fprog: the filter program
 597  *      @sk: the socket to use
 598  *
 599  * Attach the user's filter code. We first run some sanity checks on
 600  * it to make sure it does not explode on us later. If an error
 601  * occurs or there is insufficient memory for the filter a negative
 602  * errno code is returned. On success the return is zero.
 603  */
 604 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk)
 605 {
 606         struct sk_filter *fp, *old_fp;
 607         unsigned int fsize = sizeof(struct sock_filter) * fprog->len;
 608         int err;
 609
 610         /* Make sure new filter is there and in the right amounts. */
 611         if (fprog->filter == NULL)
 612                 return -EINVAL;
 613
 614         fp = sock_kmalloc(sk, fsize+sizeof(*fp), GFP_KERNEL);
 615         if (!fp)
 616                 return -ENOMEM;
 617         if (copy_from_user(fp->insns, fprog->filter, fsize)) {
 618                 sock_kfree_s(sk, fp, fsize+sizeof(*fp));
 619                 return -EFAULT;
 620         }
 621
 622         atomic_set(&fp->refcnt, 1);
 623         fp->len = fprog->len;
 624         fp->bpf_func = sk_run_filter;
 625
 626         err = sk_chk_filter(fp->insns, fp->len);
 627         if (err) {
 628                 sk_filter_uncharge(sk, fp);
 629                 return err;
 630         }
 631
 632         bpf_jit_compile(fp);
 633
 634         old_fp = rcu_dereference_protected(sk->sk_filter,
 635                                            sock_owned_by_user(sk));
 636         rcu_assign_pointer(sk->sk_filter, fp);
 637
 638         if (old_fp)
 639                 sk_filter_uncharge(sk, old_fp);
 640         return 0;
 641 }
 642 EXPORT_SYMBOL_GPL(sk_attach_filter);
 643
 644 int sk_detach_filter(struct sock *sk)
 645 {
 646         int ret = -ENOENT;
 647         struct sk_filter *filter;
 648
 649         filter = rcu_dereference_protected(sk->sk_filter,
 650                                            sock_owned_by_user(sk));
 651         if (filter) {
 652                 RCU_INIT_POINTER(sk->sk_filter, NULL);
 653                 sk_filter_uncharge(sk, filter);
 654                 ret = 0;
 655         }
 656         return ret;
 657 }
 658 EXPORT_SYMBOL_GPL(sk_detach_filter);