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