kernel/bpf/core.c

   1 /*
   2  * Linux Socket Filter - Kernel level socket filtering
   3  *
   4  * Based on the design of the Berkeley Packet Filter. The new
   5  * internal format has been designed by PLUMgrid:
   6  *
   7  *      Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com
   8  *
   9  * Authors:
  10  *
  11  *      Jay Schulist <jschlst@samba.org>
  12  *      Alexei Starovoitov <ast@plumgrid.com>
  13  *      Daniel Borkmann <dborkman@redhat.com>
  14  *
  15  * This program is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU General Public License
  17  * as published by the Free Software Foundation; either version
  18  * 2 of the License, or (at your option) any later version.
  19  *
  20  * Andi Kleen - Fix a few bad bugs and races.
  21  * Kris Katterjohn - Added many additional checks in bpf_check_classic()
  22  */
  23
  24 #include <linux/filter.h>
  25 #include <linux/skbuff.h>
  26 #include <linux/vmalloc.h>
  27 #include <linux/random.h>
  28 #include <linux/moduleloader.h>
  29 #include <linux/bpf.h>
  30 #include <linux/frame.h>
  31
  32 #include <asm/unaligned.h>
  33
  34 /* Registers */
  35 #define BPF_R0  regs[BPF_REG_0]
  36 #define BPF_R1  regs[BPF_REG_1]
  37 #define BPF_R2  regs[BPF_REG_2]
  38 #define BPF_R3  regs[BPF_REG_3]
  39 #define BPF_R4  regs[BPF_REG_4]
  40 #define BPF_R5  regs[BPF_REG_5]
  41 #define BPF_R6  regs[BPF_REG_6]
  42 #define BPF_R7  regs[BPF_REG_7]
  43 #define BPF_R8  regs[BPF_REG_8]
  44 #define BPF_R9  regs[BPF_REG_9]
  45 #define BPF_R10 regs[BPF_REG_10]
  46
  47 /* Named registers */
  48 #define DST     regs[insn->dst_reg]
  49 #define SRC     regs[insn->src_reg]
  50 #define FP      regs[BPF_REG_FP]
  51 #define ARG1    regs[BPF_REG_ARG1]
  52 #define CTX     regs[BPF_REG_CTX]
  53 #define IMM     insn->imm
  54
  55 /* No hurry in this branch
  56  *
  57  * Exported for the bpf jit load helper.
  58  */
  59 void *bpf_internal_load_pointer_neg_helper(const struct sk_buff *skb, int k, unsigned int size)
  60 {
  61         u8 *ptr = NULL;
  62
  63         if (k >= SKF_NET_OFF)
  64                 ptr = skb_network_header(skb) + k - SKF_NET_OFF;
  65         else if (k >= SKF_LL_OFF)
  66                 ptr = skb_mac_header(skb) + k - SKF_LL_OFF;
  67
  68         if (ptr >= skb->head && ptr + size <= skb_tail_pointer(skb))
  69                 return ptr;
  70
  71         return NULL;
  72 }
  73
  74 struct bpf_prog *bpf_prog_alloc(unsigned int size, gfp_t gfp_extra_flags)
  75 {
  76         gfp_t gfp_flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO |
  77                           gfp_extra_flags;
  78         struct bpf_prog_aux *aux;
  79         struct bpf_prog *fp;
  80
  81         size = round_up(size, PAGE_SIZE);
  82         fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
  83         if (fp == NULL)
  84                 return NULL;
  85
  86         kmemcheck_annotate_bitfield(fp, meta);
  87
  88         aux = kzalloc(sizeof(*aux), GFP_KERNEL | gfp_extra_flags);
  89         if (aux == NULL) {
  90                 vfree(fp);
  91                 return NULL;
  92         }
  93
  94         fp->pages = size / PAGE_SIZE;
  95         fp->aux = aux;
  96         fp->aux->prog = fp;
  97
  98         return fp;
  99 }
 100 EXPORT_SYMBOL_GPL(bpf_prog_alloc);
 101
 102 struct bpf_prog *bpf_prog_realloc(struct bpf_prog *fp_old, unsigned int size,
 103                                   gfp_t gfp_extra_flags)
 104 {
 105         gfp_t gfp_flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO |
 106                           gfp_extra_flags;
 107         struct bpf_prog *fp;
 108
 109         BUG_ON(fp_old == NULL);
 110
 111         size = round_up(size, PAGE_SIZE);
 112         if (size <= fp_old->pages * PAGE_SIZE)
 113                 return fp_old;
 114
 115         fp = __vmalloc(size, gfp_flags, PAGE_KERNEL);
 116         if (fp != NULL) {
 117                 kmemcheck_annotate_bitfield(fp, meta);
 118
 119                 memcpy(fp, fp_old, fp_old->pages * PAGE_SIZE);
 120                 fp->pages = size / PAGE_SIZE;
 121                 fp->aux->prog = fp;
 122
 123                 /* We keep fp->aux from fp_old around in the new
 124                  * reallocated structure.
 125                  */
 126                 fp_old->aux = NULL;
 127                 __bpf_prog_free(fp_old);
 128         }
 129
 130         return fp;
 131 }
 132 EXPORT_SYMBOL_GPL(bpf_prog_realloc);
 133
 134 void __bpf_prog_free(struct bpf_prog *fp)
 135 {
 136         kfree(fp->aux);
 137         vfree(fp);
 138 }
 139 EXPORT_SYMBOL_GPL(__bpf_prog_free);
 140
 141 #ifdef CONFIG_BPF_JIT
 142 struct bpf_binary_header *
 143 bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
 144                      unsigned int alignment,
 145                      bpf_jit_fill_hole_t bpf_fill_ill_insns)
 146 {
 147         struct bpf_binary_header *hdr;
 148         unsigned int size, hole, start;
 149
 150         /* Most of BPF filters are really small, but if some of them
 151          * fill a page, allow at least 128 extra bytes to insert a
 152          * random section of illegal instructions.
 153          */
 154         size = round_up(proglen + sizeof(*hdr) + 128, PAGE_SIZE);
 155         hdr = module_alloc(size);
 156         if (hdr == NULL)
 157                 return NULL;
 158
 159         /* Fill space with illegal/arch-dep instructions. */
 160         bpf_fill_ill_insns(hdr, size);
 161
 162         hdr->pages = size / PAGE_SIZE;
 163         hole = min_t(unsigned int, size - (proglen + sizeof(*hdr)),
 164                      PAGE_SIZE - sizeof(*hdr));
 165         start = (prandom_u32() % hole) & ~(alignment - 1);
 166
 167         /* Leave a random number of instructions before BPF code. */
 168         *image_ptr = &hdr->image[start];
 169
 170         return hdr;
 171 }
 172
 173 void bpf_jit_binary_free(struct bpf_binary_header *hdr)
 174 {
 175         module_memfree(hdr);
 176 }
 177 #endif /* CONFIG_BPF_JIT */
 178
 179 /* Base function for offset calculation. Needs to go into .text section,
 180  * therefore keeping it non-static as well; will also be used by JITs
 181  * anyway later on, so do not let the compiler omit it.
 182  */
 183 noinline u64 __bpf_call_base(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
 184 {
 185         return 0;
 186 }
 187 EXPORT_SYMBOL_GPL(__bpf_call_base);
 188
 189 /**
 190  *      __bpf_prog_run - run eBPF program on a given context
 191  *      @ctx: is the data we are operating on
 192  *      @insn: is the array of eBPF instructions
 193  *
 194  * Decode and execute eBPF instructions.
 195  */
 196 static unsigned int __bpf_prog_run(void *ctx, const struct bpf_insn *insn)
 197 {
 198         u64 stack[MAX_BPF_STACK / sizeof(u64)];
 199         u64 regs[MAX_BPF_REG], tmp;
 200         static const void *jumptable[256] = {
 201                 [0 ... 255] = &&default_label,
 202                 /* Now overwrite non-defaults ... */
 203                 /* 32 bit ALU operations */
 204                 [BPF_ALU | BPF_ADD | BPF_X] = &&ALU_ADD_X,
 205                 [BPF_ALU | BPF_ADD | BPF_K] = &&ALU_ADD_K,
 206                 [BPF_ALU | BPF_SUB | BPF_X] = &&ALU_SUB_X,
 207                 [BPF_ALU | BPF_SUB | BPF_K] = &&ALU_SUB_K,
 208                 [BPF_ALU | BPF_AND | BPF_X] = &&ALU_AND_X,
 209                 [BPF_ALU | BPF_AND | BPF_K] = &&ALU_AND_K,
 210                 [BPF_ALU | BPF_OR | BPF_X]  = &&ALU_OR_X,
 211                 [BPF_ALU | BPF_OR | BPF_K]  = &&ALU_OR_K,
 212                 [BPF_ALU | BPF_LSH | BPF_X] = &&ALU_LSH_X,
 213                 [BPF_ALU | BPF_LSH | BPF_K] = &&ALU_LSH_K,
 214                 [BPF_ALU | BPF_RSH | BPF_X] = &&ALU_RSH_X,
 215                 [BPF_ALU | BPF_RSH | BPF_K] = &&ALU_RSH_K,
 216                 [BPF_ALU | BPF_XOR | BPF_X] = &&ALU_XOR_X,
 217                 [BPF_ALU | BPF_XOR | BPF_K] = &&ALU_XOR_K,
 218                 [BPF_ALU | BPF_MUL | BPF_X] = &&ALU_MUL_X,
 219                 [BPF_ALU | BPF_MUL | BPF_K] = &&ALU_MUL_K,
 220                 [BPF_ALU | BPF_MOV | BPF_X] = &&ALU_MOV_X,
 221                 [BPF_ALU | BPF_MOV | BPF_K] = &&ALU_MOV_K,
 222                 [BPF_ALU | BPF_DIV | BPF_X] = &&ALU_DIV_X,
 223                 [BPF_ALU | BPF_DIV | BPF_K] = &&ALU_DIV_K,
 224                 [BPF_ALU | BPF_MOD | BPF_X] = &&ALU_MOD_X,
 225                 [BPF_ALU | BPF_MOD | BPF_K] = &&ALU_MOD_K,
 226                 [BPF_ALU | BPF_NEG] = &&ALU_NEG,
 227                 [BPF_ALU | BPF_END | BPF_TO_BE] = &&ALU_END_TO_BE,
 228                 [BPF_ALU | BPF_END | BPF_TO_LE] = &&ALU_END_TO_LE,
 229                 /* 64 bit ALU operations */
 230                 [BPF_ALU64 | BPF_ADD | BPF_X] = &&ALU64_ADD_X,
 231                 [BPF_ALU64 | BPF_ADD | BPF_K] = &&ALU64_ADD_K,
 232                 [BPF_ALU64 | BPF_SUB | BPF_X] = &&ALU64_SUB_X,
 233                 [BPF_ALU64 | BPF_SUB | BPF_K] = &&ALU64_SUB_K,
 234                 [BPF_ALU64 | BPF_AND | BPF_X] = &&ALU64_AND_X,
 235                 [BPF_ALU64 | BPF_AND | BPF_K] = &&ALU64_AND_K,
 236                 [BPF_ALU64 | BPF_OR | BPF_X] = &&ALU64_OR_X,
 237                 [BPF_ALU64 | BPF_OR | BPF_K] = &&ALU64_OR_K,
 238                 [BPF_ALU64 | BPF_LSH | BPF_X] = &&ALU64_LSH_X,
 239                 [BPF_ALU64 | BPF_LSH | BPF_K] = &&ALU64_LSH_K,
 240                 [BPF_ALU64 | BPF_RSH | BPF_X] = &&ALU64_RSH_X,
 241                 [BPF_ALU64 | BPF_RSH | BPF_K] = &&ALU64_RSH_K,
 242                 [BPF_ALU64 | BPF_XOR | BPF_X] = &&ALU64_XOR_X,
 243                 [BPF_ALU64 | BPF_XOR | BPF_K] = &&ALU64_XOR_K,
 244                 [BPF_ALU64 | BPF_MUL | BPF_X] = &&ALU64_MUL_X,
 245                 [BPF_ALU64 | BPF_MUL | BPF_K] = &&ALU64_MUL_K,
 246                 [BPF_ALU64 | BPF_MOV | BPF_X] = &&ALU64_MOV_X,
 247                 [BPF_ALU64 | BPF_MOV | BPF_K] = &&ALU64_MOV_K,
 248                 [BPF_ALU64 | BPF_ARSH | BPF_X] = &&ALU64_ARSH_X,
 249                 [BPF_ALU64 | BPF_ARSH | BPF_K] = &&ALU64_ARSH_K,
 250                 [BPF_ALU64 | BPF_DIV | BPF_X] = &&ALU64_DIV_X,
 251                 [BPF_ALU64 | BPF_DIV | BPF_K] = &&ALU64_DIV_K,
 252                 [BPF_ALU64 | BPF_MOD | BPF_X] = &&ALU64_MOD_X,
 253                 [BPF_ALU64 | BPF_MOD | BPF_K] = &&ALU64_MOD_K,
 254                 [BPF_ALU64 | BPF_NEG] = &&ALU64_NEG,
 255                 /* Call instruction */
 256                 [BPF_JMP | BPF_CALL] = &&JMP_CALL,
 257                 [BPF_JMP | BPF_CALL | BPF_X] = &&JMP_TAIL_CALL,
 258                 /* Jumps */
 259                 [BPF_JMP | BPF_JA] = &&JMP_JA,
 260                 [BPF_JMP | BPF_JEQ | BPF_X] = &&JMP_JEQ_X,
 261                 [BPF_JMP | BPF_JEQ | BPF_K] = &&JMP_JEQ_K,
 262                 [BPF_JMP | BPF_JNE | BPF_X] = &&JMP_JNE_X,
 263                 [BPF_JMP | BPF_JNE | BPF_K] = &&JMP_JNE_K,
 264                 [BPF_JMP | BPF_JGT | BPF_X] = &&JMP_JGT_X,
 265                 [BPF_JMP | BPF_JGT | BPF_K] = &&JMP_JGT_K,
 266                 [BPF_JMP | BPF_JGE | BPF_X] = &&JMP_JGE_X,
 267                 [BPF_JMP | BPF_JGE | BPF_K] = &&JMP_JGE_K,
 268                 [BPF_JMP | BPF_JSGT | BPF_X] = &&JMP_JSGT_X,
 269                 [BPF_JMP | BPF_JSGT | BPF_K] = &&JMP_JSGT_K,
 270                 [BPF_JMP | BPF_JSGE | BPF_X] = &&JMP_JSGE_X,
 271                 [BPF_JMP | BPF_JSGE | BPF_K] = &&JMP_JSGE_K,
 272                 [BPF_JMP | BPF_JSET | BPF_X] = &&JMP_JSET_X,
 273                 [BPF_JMP | BPF_JSET | BPF_K] = &&JMP_JSET_K,
 274                 /* Program return */
 275                 [BPF_JMP | BPF_EXIT] = &&JMP_EXIT,
 276                 /* Store instructions */
 277                 [BPF_STX | BPF_MEM | BPF_B] = &&STX_MEM_B,
 278                 [BPF_STX | BPF_MEM | BPF_H] = &&STX_MEM_H,
 279                 [BPF_STX | BPF_MEM | BPF_W] = &&STX_MEM_W,
 280                 [BPF_STX | BPF_MEM | BPF_DW] = &&STX_MEM_DW,
 281                 [BPF_STX | BPF_XADD | BPF_W] = &&STX_XADD_W,
 282                 [BPF_STX | BPF_XADD | BPF_DW] = &&STX_XADD_DW,
 283                 [BPF_ST | BPF_MEM | BPF_B] = &&ST_MEM_B,
 284                 [BPF_ST | BPF_MEM | BPF_H] = &&ST_MEM_H,
 285                 [BPF_ST | BPF_MEM | BPF_W] = &&ST_MEM_W,
 286                 [BPF_ST | BPF_MEM | BPF_DW] = &&ST_MEM_DW,
 287                 /* Load instructions */
 288                 [BPF_LDX | BPF_MEM | BPF_B] = &&LDX_MEM_B,
 289                 [BPF_LDX | BPF_MEM | BPF_H] = &&LDX_MEM_H,
 290                 [BPF_LDX | BPF_MEM | BPF_W] = &&LDX_MEM_W,
 291                 [BPF_LDX | BPF_MEM | BPF_DW] = &&LDX_MEM_DW,
 292                 [BPF_LD | BPF_ABS | BPF_W] = &&LD_ABS_W,
 293                 [BPF_LD | BPF_ABS | BPF_H] = &&LD_ABS_H,
 294                 [BPF_LD | BPF_ABS | BPF_B] = &&LD_ABS_B,
 295                 [BPF_LD | BPF_IND | BPF_W] = &&LD_IND_W,
 296                 [BPF_LD | BPF_IND | BPF_H] = &&LD_IND_H,
 297                 [BPF_LD | BPF_IND | BPF_B] = &&LD_IND_B,
 298                 [BPF_LD | BPF_IMM | BPF_DW] = &&LD_IMM_DW,
 299         };
 300         u32 tail_call_cnt = 0;
 301         void *ptr;
 302         int off;
 303
 304 #define CONT     ({ insn++; goto select_insn; })
 305 #define CONT_JMP ({ insn++; goto select_insn; })
 306
 307         FP = (u64) (unsigned long) &stack[ARRAY_SIZE(stack)];
 308         ARG1 = (u64) (unsigned long) ctx;
 309
 310 select_insn:
 311         goto *jumptable[insn->code];
 312
 313         /* ALU */
 314 #define ALU(OPCODE, OP)                 \
 315         ALU64_##OPCODE##_X:             \
 316                 DST = DST OP SRC;       \
 317                 CONT;                   \
 318         ALU_##OPCODE##_X:               \
 319                 DST = (u32) DST OP (u32) SRC;   \
 320                 CONT;                   \
 321         ALU64_##OPCODE##_K:             \
 322                 DST = DST OP IMM;               \
 323                 CONT;                   \
 324         ALU_##OPCODE##_K:               \
 325                 DST = (u32) DST OP (u32) IMM;   \
 326                 CONT;
 327
 328         ALU(ADD,  +)
 329         ALU(SUB,  -)
 330         ALU(AND,  &)
 331         ALU(OR,   |)
 332         ALU(LSH, <<)
 333         ALU(RSH, >>)
 334         ALU(XOR,  ^)
 335         ALU(MUL,  *)
 336 #undef ALU
 337         ALU_NEG:
 338                 DST = (u32) -DST;
 339                 CONT;
 340         ALU64_NEG:
 341                 DST = -DST;
 342                 CONT;
 343         ALU_MOV_X:
 344                 DST = (u32) SRC;
 345                 CONT;
 346         ALU_MOV_K:
 347                 DST = (u32) IMM;
 348                 CONT;
 349         ALU64_MOV_X:
 350                 DST = SRC;
 351                 CONT;
 352         ALU64_MOV_K:
 353                 DST = IMM;
 354                 CONT;
 355         LD_IMM_DW:
 356                 DST = (u64) (u32) insn[0].imm | ((u64) (u32) insn[1].imm) << 32;
 357                 insn++;
 358                 CONT;
 359         ALU64_ARSH_X:
 360                 (*(s64 *) &DST) >>= SRC;
 361                 CONT;
 362         ALU64_ARSH_K:
 363                 (*(s64 *) &DST) >>= IMM;
 364                 CONT;
 365         ALU64_MOD_X:
 366                 if (unlikely(SRC == 0))
 367                         return 0;
 368                 div64_u64_rem(DST, SRC, &tmp);
 369                 DST = tmp;
 370                 CONT;
 371         ALU_MOD_X:
 372                 if (unlikely(SRC == 0))
 373                         return 0;
 374                 tmp = (u32) DST;
 375                 DST = do_div(tmp, (u32) SRC);
 376                 CONT;
 377         ALU64_MOD_K:
 378                 div64_u64_rem(DST, IMM, &tmp);
 379                 DST = tmp;
 380                 CONT;
 381         ALU_MOD_K:
 382                 tmp = (u32) DST;
 383                 DST = do_div(tmp, (u32) IMM);
 384                 CONT;
 385         ALU64_DIV_X:
 386                 if (unlikely(SRC == 0))
 387                         return 0;
 388                 DST = div64_u64(DST, SRC);
 389                 CONT;
 390         ALU_DIV_X:
 391                 if (unlikely(SRC == 0))
 392                         return 0;
 393                 tmp = (u32) DST;
 394                 do_div(tmp, (u32) SRC);
 395                 DST = (u32) tmp;
 396                 CONT;
 397         ALU64_DIV_K:
 398                 DST = div64_u64(DST, IMM);
 399                 CONT;
 400         ALU_DIV_K:
 401                 tmp = (u32) DST;
 402                 do_div(tmp, (u32) IMM);
 403                 DST = (u32) tmp;
 404                 CONT;
 405         ALU_END_TO_BE:
 406                 switch (IMM) {
 407                 case 16:
 408                         DST = (__force u16) cpu_to_be16(DST);
 409                         break;
 410                 case 32:
 411                         DST = (__force u32) cpu_to_be32(DST);
 412                         break;
 413                 case 64:
 414                         DST = (__force u64) cpu_to_be64(DST);
 415                         break;
 416                 }
 417                 CONT;
 418         ALU_END_TO_LE:
 419                 switch (IMM) {
 420                 case 16:
 421                         DST = (__force u16) cpu_to_le16(DST);
 422                         break;
 423                 case 32:
 424                         DST = (__force u32) cpu_to_le32(DST);
 425                         break;
 426                 case 64:
 427                         DST = (__force u64) cpu_to_le64(DST);
 428                         break;
 429                 }
 430                 CONT;
 431
 432         /* CALL */
 433         JMP_CALL:
 434                 /* Function call scratches BPF_R1-BPF_R5 registers,
 435                  * preserves BPF_R6-BPF_R9, and stores return value
 436                  * into BPF_R0.
 437                  */
 438                 BPF_R0 = (__bpf_call_base + insn->imm)(BPF_R1, BPF_R2, BPF_R3,
 439                                                        BPF_R4, BPF_R5);
 440                 CONT;
 441
 442         JMP_TAIL_CALL: {
 443                 struct bpf_map *map = (struct bpf_map *) (unsigned long) BPF_R2;
 444                 struct bpf_array *array = container_of(map, struct bpf_array, map);
 445                 struct bpf_prog *prog;
 446                 u64 index = BPF_R3;
 447
 448                 if (unlikely(index >= array->map.max_entries))
 449                         goto out;
 450
 451                 if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT))
 452                         goto out;
 453
 454                 tail_call_cnt++;
 455
 456                 prog = READ_ONCE(array->ptrs[index]);
 457                 if (unlikely(!prog))
 458                         goto out;
 459
 460                 /* ARG1 at this point is guaranteed to point to CTX from
 461                  * the verifier side due to the fact that the tail call is
 462                  * handeled like a helper, that is, bpf_tail_call_proto,
 463                  * where arg1_type is ARG_PTR_TO_CTX.
 464                  */
 465                 insn = prog->insnsi;
 466                 goto select_insn;
 467 out:
 468                 CONT;
 469         }
 470         /* JMP */
 471         JMP_JA:
 472                 insn += insn->off;
 473                 CONT;
 474         JMP_JEQ_X:
 475                 if (DST == SRC) {
 476                         insn += insn->off;
 477                         CONT_JMP;
 478                 }
 479                 CONT;
 480         JMP_JEQ_K:
 481                 if (DST == IMM) {
 482                         insn += insn->off;
 483                         CONT_JMP;
 484                 }
 485                 CONT;
 486         JMP_JNE_X:
 487                 if (DST != SRC) {
 488                         insn += insn->off;
 489                         CONT_JMP;
 490                 }
 491                 CONT;
 492         JMP_JNE_K:
 493                 if (DST != IMM) {
 494                         insn += insn->off;
 495                         CONT_JMP;
 496                 }
 497                 CONT;
 498         JMP_JGT_X:
 499                 if (DST > SRC) {
 500                         insn += insn->off;
 501                         CONT_JMP;
 502                 }
 503                 CONT;
 504         JMP_JGT_K:
 505                 if (DST > IMM) {
 506                         insn += insn->off;
 507                         CONT_JMP;
 508                 }
 509                 CONT;
 510         JMP_JGE_X:
 511                 if (DST >= SRC) {
 512                         insn += insn->off;
 513                         CONT_JMP;
 514                 }
 515                 CONT;
 516         JMP_JGE_K:
 517                 if (DST >= IMM) {
 518                         insn += insn->off;
 519                         CONT_JMP;
 520                 }
 521                 CONT;
 522         JMP_JSGT_X:
 523                 if (((s64) DST) > ((s64) SRC)) {
 524                         insn += insn->off;
 525                         CONT_JMP;
 526                 }
 527                 CONT;
 528         JMP_JSGT_K:
 529                 if (((s64) DST) > ((s64) IMM)) {
 530                         insn += insn->off;
 531                         CONT_JMP;
 532                 }
 533                 CONT;
 534         JMP_JSGE_X:
 535                 if (((s64) DST) >= ((s64) SRC)) {
 536                         insn += insn->off;
 537                         CONT_JMP;
 538                 }
 539                 CONT;
 540         JMP_JSGE_K:
 541                 if (((s64) DST) >= ((s64) IMM)) {
 542                         insn += insn->off;
 543                         CONT_JMP;
 544                 }
 545                 CONT;
 546         JMP_JSET_X:
 547                 if (DST & SRC) {
 548                         insn += insn->off;
 549                         CONT_JMP;
 550                 }
 551                 CONT;
 552         JMP_JSET_K:
 553                 if (DST & IMM) {
 554                         insn += insn->off;
 555                         CONT_JMP;
 556                 }
 557                 CONT;
 558         JMP_EXIT:
 559                 return BPF_R0;
 560
 561         /* STX and ST and LDX*/
 562 #define LDST(SIZEOP, SIZE)                                              \
 563         STX_MEM_##SIZEOP:                                               \
 564                 *(SIZE *)(unsigned long) (DST + insn->off) = SRC;       \
 565                 CONT;                                                   \
 566         ST_MEM_##SIZEOP:                                                \
 567                 *(SIZE *)(unsigned long) (DST + insn->off) = IMM;       \
 568                 CONT;                                                   \
 569         LDX_MEM_##SIZEOP:                                               \
 570                 DST = *(SIZE *)(unsigned long) (SRC + insn->off);       \
 571                 CONT;
 572
 573         LDST(B,   u8)
 574         LDST(H,  u16)
 575         LDST(W,  u32)
 576         LDST(DW, u64)
 577 #undef LDST
 578         STX_XADD_W: /* lock xadd *(u32 *)(dst_reg + off16) += src_reg */
 579                 atomic_add((u32) SRC, (atomic_t *)(unsigned long)
 580                            (DST + insn->off));
 581                 CONT;
 582         STX_XADD_DW: /* lock xadd *(u64 *)(dst_reg + off16) += src_reg */
 583                 atomic64_add((u64) SRC, (atomic64_t *)(unsigned long)
 584                              (DST + insn->off));
 585                 CONT;
 586         LD_ABS_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + imm32)) */
 587                 off = IMM;
 588 load_word:
 589                 /* BPF_LD + BPD_ABS and BPF_LD + BPF_IND insns are
 590                  * only appearing in the programs where ctx ==
 591                  * skb. All programs keep 'ctx' in regs[BPF_REG_CTX]
 592                  * == BPF_R6, bpf_convert_filter() saves it in BPF_R6,
 593                  * internal BPF verifier will check that BPF_R6 ==
 594                  * ctx.
 595                  *
 596                  * BPF_ABS and BPF_IND are wrappers of function calls,
 597                  * so they scratch BPF_R1-BPF_R5 registers, preserve
 598                  * BPF_R6-BPF_R9, and store return value into BPF_R0.
 599                  *
 600                  * Implicit input:
 601                  *   ctx == skb == BPF_R6 == CTX
 602                  *
 603                  * Explicit input:
 604                  *   SRC == any register
 605                  *   IMM == 32-bit immediate
 606                  *
 607                  * Output:
 608                  *   BPF_R0 - 8/16/32-bit skb data converted to cpu endianness
 609                  */
 610
 611                 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 4, &tmp);
 612                 if (likely(ptr != NULL)) {
 613                         BPF_R0 = get_unaligned_be32(ptr);
 614                         CONT;
 615                 }
 616
 617                 return 0;
 618         LD_ABS_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + imm32)) */
 619                 off = IMM;
 620 load_half:
 621                 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 2, &tmp);
 622                 if (likely(ptr != NULL)) {
 623                         BPF_R0 = get_unaligned_be16(ptr);
 624                         CONT;
 625                 }
 626
 627                 return 0;
 628         LD_ABS_B: /* BPF_R0 = *(u8 *) (skb->data + imm32) */
 629                 off = IMM;
 630 load_byte:
 631                 ptr = bpf_load_pointer((struct sk_buff *) (unsigned long) CTX, off, 1, &tmp);
 632                 if (likely(ptr != NULL)) {
 633                         BPF_R0 = *(u8 *)ptr;
 634                         CONT;
 635                 }
 636
 637                 return 0;
 638         LD_IND_W: /* BPF_R0 = ntohl(*(u32 *) (skb->data + src_reg + imm32)) */
 639                 off = IMM + SRC;
 640                 goto load_word;
 641         LD_IND_H: /* BPF_R0 = ntohs(*(u16 *) (skb->data + src_reg + imm32)) */
 642                 off = IMM + SRC;
 643                 goto load_half;
 644         LD_IND_B: /* BPF_R0 = *(u8 *) (skb->data + src_reg + imm32) */
 645                 off = IMM + SRC;
 646                 goto load_byte;
 647
 648         default_label:
 649                 /* If we ever reach this, we have a bug somewhere. */
 650                 WARN_RATELIMIT(1, "unknown opcode %02x\n", insn->code);
 651                 return 0;
 652 }
 653 STACK_FRAME_NON_STANDARD(__bpf_prog_run); /* jump table */
 654
 655 bool bpf_prog_array_compatible(struct bpf_array *array,
 656                                const struct bpf_prog *fp)
 657 {
 658         if (!array->owner_prog_type) {
 659                 /* There's no owner yet where we could check for
 660                  * compatibility.
 661                  */
 662                 array->owner_prog_type = fp->type;
 663                 array->owner_jited = fp->jited;
 664
 665                 return true;
 666         }
 667
 668         return array->owner_prog_type == fp->type &&
 669                array->owner_jited == fp->jited;
 670 }
 671
 672 static int bpf_check_tail_call(const struct bpf_prog *fp)
 673 {
 674         struct bpf_prog_aux *aux = fp->aux;
 675         int i;
 676
 677         for (i = 0; i < aux->used_map_cnt; i++) {
 678                 struct bpf_map *map = aux->used_maps[i];
 679                 struct bpf_array *array;
 680
 681                 if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY)
 682                         continue;
 683
 684                 array = container_of(map, struct bpf_array, map);
 685                 if (!bpf_prog_array_compatible(array, fp))
 686                         return -EINVAL;
 687         }
 688
 689         return 0;
 690 }
 691
 692 /**
 693  *      bpf_prog_select_runtime - select exec runtime for BPF program
 694  *      @fp: bpf_prog populated with internal BPF program
 695  *
 696  * Try to JIT eBPF program, if JIT is not available, use interpreter.
 697  * The BPF program will be executed via BPF_PROG_RUN() macro.
 698  */
 699 int bpf_prog_select_runtime(struct bpf_prog *fp)
 700 {
 701         fp->bpf_func = (void *) __bpf_prog_run;
 702
 703         bpf_int_jit_compile(fp);
 704         bpf_prog_lock_ro(fp);
 705
 706         /* The tail call compatibility check can only be done at
 707          * this late stage as we need to determine, if we deal
 708          * with JITed or non JITed program concatenations and not
 709          * all eBPF JITs might immediately support all features.
 710          */
 711         return bpf_check_tail_call(fp);
 712 }
 713 EXPORT_SYMBOL_GPL(bpf_prog_select_runtime);
 714
 715 static void bpf_prog_free_deferred(struct work_struct *work)
 716 {
 717         struct bpf_prog_aux *aux;
 718
 719         aux = container_of(work, struct bpf_prog_aux, work);
 720         bpf_jit_free(aux->prog);
 721 }
 722
 723 /* Free internal BPF program */
 724 void bpf_prog_free(struct bpf_prog *fp)
 725 {
 726         struct bpf_prog_aux *aux = fp->aux;
 727
 728         INIT_WORK(&aux->work, bpf_prog_free_deferred);
 729         schedule_work(&aux->work);
 730 }
 731 EXPORT_SYMBOL_GPL(bpf_prog_free);
 732
 733 /* RNG for unpriviledged user space with separated state from prandom_u32(). */
 734 static DEFINE_PER_CPU(struct rnd_state, bpf_user_rnd_state);
 735
 736 void bpf_user_rnd_init_once(void)
 737 {
 738         prandom_init_once(&bpf_user_rnd_state);
 739 }
 740
 741 u64 bpf_user_rnd_u32(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
 742 {
 743         /* Should someone ever have the rather unwise idea to use some
 744          * of the registers passed into this function, then note that
 745          * this function is called from native eBPF and classic-to-eBPF
 746          * transformations. Register assignments from both sides are
 747          * different, f.e. classic always sets fn(ctx, A, X) here.
 748          */
 749         struct rnd_state *state;
 750         u32 res;
 751
 752         state = &get_cpu_var(bpf_user_rnd_state);
 753         res = prandom_u32_state(state);
 754         put_cpu_var(state);
 755
 756         return res;
 757 }
 758
 759 /* Weak definitions of helper functions in case we don't have bpf syscall. */
 760 const struct bpf_func_proto bpf_map_lookup_elem_proto __weak;
 761 const struct bpf_func_proto bpf_map_update_elem_proto __weak;
 762 const struct bpf_func_proto bpf_map_delete_elem_proto __weak;
 763
 764 const struct bpf_func_proto bpf_get_prandom_u32_proto __weak;
 765 const struct bpf_func_proto bpf_get_smp_processor_id_proto __weak;
 766 const struct bpf_func_proto bpf_ktime_get_ns_proto __weak;
 767 const struct bpf_func_proto bpf_get_current_pid_tgid_proto __weak;
 768 const struct bpf_func_proto bpf_get_current_uid_gid_proto __weak;
 769 const struct bpf_func_proto bpf_get_current_comm_proto __weak;
 770 const struct bpf_func_proto * __weak bpf_get_trace_printk_proto(void)
 771 {
 772         return NULL;
 773 }
 774
 775 /* Always built-in helper functions. */
 776 const struct bpf_func_proto bpf_tail_call_proto = {
 777         .func           = NULL,
 778         .gpl_only       = false,
 779         .ret_type       = RET_VOID,
 780         .arg1_type      = ARG_PTR_TO_CTX,
 781         .arg2_type      = ARG_CONST_MAP_PTR,
 782         .arg3_type      = ARG_ANYTHING,
 783 };
 784
 785 /* For classic BPF JITs that don't implement bpf_int_jit_compile(). */
 786 void __weak bpf_int_jit_compile(struct bpf_prog *prog)
 787 {
 788 }
 789
 790 /* To execute LD_ABS/LD_IND instructions __bpf_prog_run() may call
 791  * skb_copy_bits(), so provide a weak definition of it for NET-less config.
 792  */
 793 int __weak skb_copy_bits(const struct sk_buff *skb, int offset, void *to,
 794                          int len)
 795 {
 796         return -EFAULT;
 797 }